Variational Principle

9/12/15
CE502: Finite Element

Analysis
Variational Principle
By
Dr. A. Chakraborty
Department of Civil Engineering
Indian Institute of Technology Guwahati,
India
9/12/15
Optimization of a
f ( x)
amax)
Let,
has extremum (min xor
function
f ( x)
at
. Expandx a
in Taylor series at
around
i.e.
of a
df in
(a) the neighborhood
1 d 2 f (a)
2
f ( x) f (a )
( x a)
( x a ) L
2
dx
2 dx
For minimum value
f ( x ) f (a ) 0
x a x
x
Here,
is either +ve or -ve i.e. nonzero
9/12/15
Let us examine equal to zero first: 1st term

( x) f ( a) 0 and here
will be fdominant
for
minimum, therefore
df (a )
0
dx x a
This would be same for both max and min.
Next, considering second term
f ( x) f (a) 0
1 d 2 f (a )
2
(
x
a
)
0
2
2 dx x a
df (a)
0
Note:
dx x a
d 2 f (a)
0
2
dx x a
min
d 2 f (a)
0
2
dx x a
max
3
9/12/15
Observation:
( x)
Max or minf of
depends on the values
of
Only differentiationa,is
b enough to identify
extremum
Extremum are points (like
) on the space
that defines independent variable
9/12/15
Optimum of a
dv Pu
functional
T
i i
of &
( x, y, z ) of
is a function
is a ufunction
.
is
a functional
Note:
We
are looking foruoptimal , so that
is minimum. Individually speaking, one can
f1 , f 2 , out
f3 , f 4 maximum
x
find
or minimum of
and get . However, our objective
is to optimize
the following functional
x2
T F ( f ( x), x)dx given f ( x1 ) f1 & f ( x2 ) f 2
x1
9/12/15
x1 x x2
( )
f ( x)a function
Delta operator
: Let us consider
in
[ f ( x)] f%( x) f ( x)
f%( x)
Here,
is an admissible function that
satisfies boundary
f%( x1 ) f ( xcondition
and f%( x2 ) f ( x2 )
1)
[ f ( x )] f%( x ) f ( x ) 0
1
[ f ( x2 )] f%( x2 ) f ( x2 ) 0
f%( x)
is any admissible candidate function

6
9/12/15
Strong & Weak

Strong
Variation: if we impose no restriction
Variation
%
%
f (of
x)
f ( x)
off the
and
then
( x) derivative
is a strong variation.
f ( x) restriction
f%( x) impose
f%( x )
Weak Variation: if we
that
not much difference in derivative of
and
then
is a weak
variation
9/12/15
Properties of Delta Operator:

Differentiation of variation
d ( f ) d %
d % d
df
(f f)
f
f
dx
dx
dx
dx
dx
Variation of integration
% fdx ( f% f )dx fdx
fdx fdx
g ( x) f1 ( x ) f 2 ( x )
%
g ( x) g% g f%
1 f 2 f1 f 2 [ f1 ] [ f 2 ]
f1 f 2 [ f1 ] f 2 f1 [ f 2 ]
[ f1 ] f 2 f1 [ f 2 ]
f1 f
2
f 22
f f
8
9/12/15
Functional: Taylor series expansion

F
F
%%
F (u, v , x) F (u, v, x)
u v h.o.t.
u
v
u u% u
v and
v% v
Where,
.
Neglecting the effect of the higher order
F
F
terms
F (u%%
, v , x) F (u, v, x)
u v
u
v
F
F
F
u v
u
v
Note: 1st variation has similar structure like
differentiation but there is significant
difference
9/12/15
Let,
I (u, u) F (u, u , x)dx
I (u, u) F (u, u, x)dx

I F (u, u , x)dx
F
F

u
u dx
u
u
Example:
I (u, v) sin xu 2 ( x) u ( x) e v ( x )
Find
I sin xu 2 ( x) u ( x) ev ( x )
sin x 2u u u ev ( x ) u ev ( x ) v
2u sin x ev ( x ) u u ev ( x ) v
10
9/12/15
Theorem of Functional: Let,

x2
I (u, v, u , v) F (u , v, u , v)dx
x1
x2
F
F
F
F
%%%%
I (u , v , u , v ) I (u, v, u , v )
u v u v
u
v
u
v
x1
dx h.o.t.
x2
F
F
F
F
I
u v u v dx
u
v
u
v
x1
d
Note:
by property of Delta
u
operatordx
x2
x2
F d F
F
F
F d F
I
u u
u
v dx
u x1 v x1
v dx
v
u dx u
11
9/12/15
Therefore, for minimize

) I (u, v, u , v) 0
I (u%%%
, v , u, v%
x
x
u
v
u x admissible
0 v x 0
equate it to zero.
Also
&
means F d F & F d F
I

v dx
u
v dx v
u dx u
Apply localization theorem, the governing
equations are F d F 0
u dx u
F d F
0
v dx v
2
12
9/12/15
1 du 2
(u ) EA qu dx PuL
Example:
2 dx
0
. Given
0
L
u (0) u0 0
du du
q u dx P u L 0
0 2 EA2 dx dx
du d u
0 EA dx dx q u dx P uL 0
L
du
d
du
EA u
EA q udx P u L 0
dx 0 0 dx
dx
du
du
d
du
EA
EA
EA
udx 0
dx
dx 0 0 dx
dx
0
13
9/12/15
Governing equation:
d
du
EA
q 0 0 x L
dx
dx
Boundary condition:
du
EA P 0 x L
dx
Essential Boundary
Condition
u (0)
Natural Boundary
Condition
du
EA P x L
dx
Dirichlet BVP
Neuman BVP
Displacement BVP
Force BVP
14
9/12/15
Example: The total potential energy of a

linear elastic body in 3D subjected to body
1
and traction
by
(is
u ) given
f
u
V 2 ij ij i i dv S tiui ds
2
Son
ui
ui is
Given,
. Using variational
1
principle find governing equation and force
boundary condition. Material is isotropic.
: Differential equation of equilibrium &
Caucheys stress-strain relation
Note: Variational principle provides
a) Homogenous form of the governing
equation using essential boundary
condition
b) Natural boundary condition needs to be
satisfied
15
9/12/15
Integral Formulation:
Let us consider the problem d
dv
a( x) f ( x) 0 x L
dx
dx
du
u
(0)
u
&
a
QL
Given,
0
dx a L
Note: A large class of problem may be
N
classified in this format
u ( x) ; u N ( x) C j j ( x) 0 ( x)
j 1
uN ( x)
Where,
are unknown coefficients,
Cj
satisfies the differential equation
du N
d
a( x) f ( x)
0 xL
dx
dx
du N
d
R ( x, C j ) a ( x ) f ( x ) 0
0xL
dx
dx
16
9/12/15
u N ( x) zero as
The above equation is not exactly
R( xis
, C japproximate.
)
R
N
Cj
is residue. But we need
to be
for solve
x xi ,K & i 1, 2, , N
zero at R( x, Cpoints
to
j) 0
17
9/12/15
To achieve this we apply Weighted Residual

wi
Method
L
Let,
are
the weights at i th points wi ( x)R( x, C j )dx 0 i 1, 2,K , N
0
wi
Note: For different
methods
Name
Petrov-Galerkin
Galerkin
Least square
Collocation
*
we have different
wi
wi i i
wi i
di
d
wi a( x)
dx
dx
wi ( x xi )*
() is Dirac delta function; not

operator
18
9/12/15
Dirac delta function:

( x x0 ) 0 x x0
= 1 x x0
f ( x) ( x x0 )dx f ( x0 )
19
9/12/15
Development of Weak
Form w d a dv f 0
L
dx
dw du
dv
a
wf
dx
wa
0 dx dx
dx
L
dx
0
0
dw du
wf
0 dx dx dx wQ 0 wQ L 0
The above eq. for the weak form of the

differential eq. in the given weighted-integral
u
statement. Weak
refers to reduced order
w
differential
of
(i.e. 1st order and not in 2nd
0
order). Hence, w(0) must
satisfies the
homogeneous form of the essential boundary
condition:
20
9/12/15
Weak form:
L
dw du
wf
0 dx dx dx wQ L 0
Let,
L
dw du
B( w, u ) a
dx
bilinear
dx dx
0
L
l ( w) wfdx wQ L
linear
Hence, weak form equation cane rewritten

B( w, u ) l ( w)
21
9/12/15
Note: ifw u
2
d u du
a du
1 du du
B( u , u ) a
dx dx a
dx
dx dx
2 dx
2 0 dx dx
0
0
1
B( u , u ) B(u, u )
2
L
L
and
L
l ( u ) ufdx u (t ) Qt ufdx u Q L
0
0
l ( u ) l (u )
1
I (u ) B (u, u ) l (u )
2
I B( u , u ) l ( u )
22
9/12/15
2 I B( u, u ) 0
For minima
for
u 0
L
1 du 2
(u ) EA qu dx P u L
2 dx
0
L
du
du
d

EA q udx EA P u L
dx
dx
dx
0
0
u w
In above eq.
Check the Standard Form

d du
a f 0
dx dx
du
u
(0)
u
a
QL
Subjected to 0
and
dx x L
23
9/12/15
a EA, q f , w u and
L
dw du
B a
dx
dx dx
0
L
dw du
EA
dx
dx dx
0
L
du
l EA P u L w(q )dx
dx
0
B of in
using relation
, one can get
B ( u , u ) l ( u )
1
B (u , u ) l (u )
2
1
B(u, u ) l (u )
2
24
9/12/15
Then it can be represented by

1
I B(u, u ) l (u )
2
L
1 du 2
(u ) EA qu dx P u L
2 dx
0
u (0) 0
(u ) 0
here,
is homogenous essential
condition. Applying,
d
du
EA
q 0
dx
dx
du
EA P
at x L
dx
The above two equations are represents
governing equation and natural boundary
condition respectively.
25
9/12/15
Example: Solve the above governing equation

and fixed weak form. Also solve using Ritzs
method with N=2
Weighted Residual Method:
L
d
du
0 w dx EA dx q dx 0
L
d
du
w
EA
0 dx dx dx 0 wqdx 0
L
du
dw du
wEA
EA dx wqdx 0
dx 0 0 dx
dx
0
L
dw du
du
EA
dx wEA
wqdx 0
dx dx
dx 0 0
0
26
9/12/15
here
L
dw du
B( w, u ) EA
dx
dx dx
0
L
du
du
l (u ) w( L) EA
w(0) EA
wqdx
dx L
dx 0 0
hence
B( w, u ) l ( w)
Weak Form
Ritz Solution:
N
Let,
u ( x) ; u%
( x ) C j j 0
j 1
u (0) 0
Note:
can set
already zero. Sowe

0 0
u%
( x) C11 C22
N 2
u (0) 0; 1 (0) 2 (0) 0
Now satisfying the above condition
2
(
x
)
x
and
(
x
)
x
Let, 1
2
27
9/12/15
Note: You may assume any other

C2
C1
polynomial that satisfy the essential
boundary condition. Here
and
are
unknown 2
u%
( x) C1 x C2 x
wi i w1 x, w2 x 2
du%
C1 x 2C2 x
dx
L
dw du
dw1 dw2
&
B
(
w
,
u
)
EA
dx Ky&
2x
dx dx
dx
dx
0
L
EA(C 2C x)dx EAC dx 2 EAC xdx

1
2
EA
2
x
(
C
2
C
x
)
dx
EAC
2
xdx
4
EAC
x
1
2
1
2 dx
28
9/12/15
2 EAC2 xdx
EAC1dx
0
0
B ( w, u ) L
L
2
2 EAC2 xdx 4 EAC2 x dx
0
0
L
L
2 xdx
dx
C 1
0
0
EA L
L
C 2
2 xdx 4 x 2 dx
0
here
Note: The above matrix is

symmetric.
29
9/12/15
L
du
du
l (u ) wqdx w( L) EA
w(0) EA
dx L
dx 0
0
L
wqdx w( L) EA
0
du
dx L
wqdx w( L) P
0
du
EA
In above equation
dx L
From here, one
assume
L
w1 xqdx PL
P
is replaced
by
w2 Lx 2 qdx PL2
0
30
9/12/15
Hence,
xqdx PL
l (u )
Lx 2 qdx PL 2

In above equation, the integral part is

component due to body force and second part
L
L
L
due to point load.

dx
2 xdx
xqdx PL
Matrix Form: 0
C 1 0
EA L
C 2 L
Lx 2 qdx PL 2
2 xdx 4 x 2 dx

qL
2
L L
PL
C
1 2
EA 2 4 3
3
L
L C 2
qL PL 2
3
3
31
9/12/15
C1
Now solve
Cand
2
Let,P 100kN , L 1m, d 12.7mm A 126.68mm 2 , E 2.1E 08kN / m 2

g 9.81m / s 2
P
3
If /only
applied
and not body
P / A 7.89 10.5 kN
m 2 , 3.76is
10
force
.
C1 0.0038
Use these values
above
7
C in1.8219
10equations
2
u ( x) 0.0038L 1.8219 10 7 L2
u ( L) 3.8mm
Note: Variational formulation tell that impact

of gravitational pull is negligible.
32
9/12/15
Note:
b
G( x) ( x)dx 0
a xb
( x)
G ( x) 0 for arbitrary
Lemma:
G ( x) ( x)
b
2
G
(
x
)
(
x
)
dx
( x)dx 0
G ( x) 0
( a, b)
33
9/12/15
General Statement:
(a)
G (a)
a x in
b
is arbitrary
a x in
b
is arbitrary
G( x) ( x)dx G(a) (a) 0

a
2
2
G
(
x
)
dx
G
(a) 0 Sum of two +ve quantity
G ( a ) 0 & G ( x) 0
34
9/12/15
Thank You!!!
35

Variational Principle

Uploaded by

Copyright:

Available Formats

Variational Principle

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Variational Principle

Uploaded by

Copyright:

Available Formats

9/12/15

CE502: Finite Element

Let us examine equal to zero first: 1st term

is any admissible candidate function

Strong & Weak

Properties of Delta Operator:

Functional: Taylor series expansion

I (u, u) F (u, u , x)dx

I (u, u) F (u, u, x)dx

Theorem of Functional: Let,

Therefore, for minimize

Example: The total potential energy of a

To achieve this we apply Weighted Residual

() is Dirac delta function; not

Dirac delta function:

The above eq. for the weak form of the

Hence, weak form equation cane rewritten

Check the Standard Form

Then it can be represented by

Example: Solve the above governing equation

already zero. Sowe

Note: You may assume any other

EA(C 2C x)dx EAC dx 2 EAC xdx

Note: The above matrix is

In above equation, the integral part is

due to point load.

Let,P 100kN , L 1m, d 12.7mm A 126.68mm 2 , E 2.1E 08kN / m 2

Note: Variational formulation tell that impact

G( x) ( x)dx G(a) (a) 0

You might also like