Approximation Methods

Governing Differential Equation

L2 m [φ ] = r
linear, symmetric, positive definite differential operator
L2m with up to (2m)th order

φ ( x) State variable

r ( x) Forcing function

Constraint Equation or Boundary Condition

Bi [φ ] = qi on Si , i = 1,
1 22,..., 2m
The Weighted Residual Method
(WRM)
 linearly independent
Trial function Base function 2m times differentiable

n
φ = ∑aj f j fj satisfy all B.C’s
j =1

n f0 satisfy nonhomogeneous B.C’s

φ = f0 + ∑ a j f j
j =1 fj satisfy homogeneous B.C’s

Differential equation Residual or Error

r − L2 m [φ ] = 0 R = r − L2 m [φ ]
n
R = r ( xi ) − ∑ a j L2 m [ f j ( xi )]
j =1
 n
1) Collocation Method φ = ∑ aj f j
j =1
n
φ = f0 + ∑ a j f j
R ( xi ) = 0, i = 1,2,..., n
j =1

R = r − L2 m [φ ]
n
r ( xi ) − L2 m [∑ a j f j ( xi )] = 0, i = 1, 2,..., n
j =1

∑a L
j =1
j=
j 2m [ f j ( xi )] = r ( xi ), i = 1, 2,..., n

Non-symmetric matrix

Collocation Points
 n
φ = ∑ aj f j
j =1
2) Least Squares Method n
φ = f0 + ∑ a j f j
Minimize
∫
D
R 2 dD for ai j =1

R = r − L2 m [φ ]
∂
∂ai ∫D
R 2 dD = 0, i = 1, 2,..., n
∂R ∂ ⎧ n ⎫
∂R = ⎨r − ∑ a j L2 m [ f j ]⎬ = L2 m [ f i ]
∂ai ∂ai ⎩
∫D
R
∂ai
dD = 0, i = 1, 2,..., n j =1 ⎭

⎧ n ⎫
∫D ⎨⎩r − ∑
j =1
a j L2 m [ f j ]⎬ L2 m [ f i ] dD = 0,, i = 1,, 2,...,
⎭
, ,n

∫ L2 m [ fi ]∑ a j L2 m [ f j ] dD = ∫ r L2 m [ f i ] dD, i = 11, 22,..., n

D D
j =1

Symmetric matrix
 n
φ = ∑ aj f j
3) Petrov
Petrov-Galerkin
Galerkin Method j =1
n
φ = f0 + ∑ a j f j
∫D
gi R dD = 0, i = 1, 2,.., n j =1

R = r − L2 m [φ ]
Non-symmetric matrix
 n
φ = ∑ aj f j
j =1

4) Bubnov-Galerkin Method n
φ = f0 + ∑ a j f j
gi = fi j =1

R = r − L2 m [φ ]
∫D
fi R dD = 0, i = 1, 2,.., n
⎧ n ⎫
∫ fi ⎨r − ∑ a j L2 m [ f j ]⎬ dD = 0
⎩ ⎭
D
j=1

Symmetric matrix
Example. Bar with Variable Cross Section
A( x) = 4 − x
L=2
E =1

Geometric B.C.
Governing differential equation
u (0) = 0
d ⎛ d ⎞
du
⎜ AE ⎟=0 Force B.C
dx ⎝ dx ⎠
du(2)
du (2)
AE =P
dx
Analytical Solution

u ( x ) = − P ln(1 − x / 4)
u ( 2) = 0.6932 P
T i l function
Trial f ti
d ⎛ du ⎞
u ≈ u = a1 x + a2 x 2 ⎜ AE ⎟=0
dx ⎝ dx ⎠
u ((0)) = 0
du du (2)
AE = 2(a1 + 4a2 ) = P AE =P
dx x=2
dx

P
a1 = −4a2 +
2
Px
u ( x) = + a2 ( x − 4 x )
2

2
1) Collocation Method
d ⎛ du ⎞
⎜ AE ⎟=0 d ⎛ du ⎞
dx ⎝ dx ⎠
⎜ AE ⎟=0
P
dx ⎝ dx ⎠
−4a2 − 4a2 x − = 0
2 d ⎛ du ⎞
R= ⎜ AE ⎟
Collocation Point dx ⎝ dx ⎠
x =1 x = 1.5 Px
u= + a2 ( x 2 − 4 x )
2
P P
a2 = − a2 = −
16 20
Px P 2 Px P 2 Analytical Solution
u= − ( x − 4 x) u= − ( x − 4 x)
2 16 2 20 u ( x ) = − P ln(1
l (1 − x / 4)
u (2) = 0.750 P u (2) = 1.2 P u ( 2) = 0.6932 P
2) Bubnov-Galerkin Method

∫D
fi R dD = 0, i = 1, 2,.., n d ⎛
⎜ AE
dx ⎝
du ⎞
⎟=0
dx ⎠
⎛ P⎞
( )
2 2
∫ f1 Rdx = ∫ x − 4 x ⎜ 12a2 − 4a2 x − ⎟ dx = 0 d ⎛ du ⎞
2
0 0
⎝ 2⎠ R= ⎜ AE ⎟
dx ⎝ dx ⎠
Px
a2 = −0.071P u= + a2 ( x 2 − 4 x )
2
Px
u= − 0.071
0 071P ( x 2 − 4 x )
2 Analytical Solution

u (2) = 1.286
1 286PP u ( x ) = − P ln(1
l (1 − x / 4)
u ( 2) = 0.6932 P
Ritz Method
 linearly independent
Trial function Base function
2m times differentiable
n
φ = ∑aj f j fj satisfy geometric all B.C’s only
j =1

n
φ = f0 + ∑ a j f j f0 satisfy nonhomogeneous geometric B.C’s only
j =1
fj satisfy homogeneous geometric B.C’s only

Variational Statement
∂Π ∂Π
Define Π (φ ) δΠ = δ a1 + ... + δ an = 0 ∀δ ai
∂a1 ∂an
Minimize Π (a1 , a2 , ...) ∀δ ai ∂Π
i = 1,2,..., n = 0, i = 1,2,..., n
∂ai
What about the force B.C.’s?
 Example. Bar with Variable Cross Section
Geometric B.C.

u (0) = 0

A = 2c − ( x / L ) c Force B.C.

fb = const. du ( L)
AE =P
L=2 dx
Exact Solution
fb L 2 fb L2 x ( fb L − P) L x
u ( x) = x+ ( − )+
ln(1 ( − )
ln(1
Ec Ec 2L Ec 2L
 Method #1
Total Potential

L ⎡ EA( x) ⎛ du ⎞ 2 ⎤ L
Π (u ) = ∫ ⎢ ⎜ ⎟ ⎥ dx − ∫0 f budx − Pu ( L)
0
⎢⎣ 2 ⎝ dx ⎠ ⎥⎦

Trial Function
2
x ⎛x⎞
u ≈ u = a1 + a2 ⎜ ⎟
L ⎝L⎠
⎧⎪ EA( x) ⎡ a 2a ⎛ x ⎞ ⎤ 2 ⎫⎪ ⎡ x ⎛ x ⎞
2
⎤
⎜ ⎟ ⎥ ⎬ dx − ∫0 fb ⎢ a1 + a2 ⎜ ⎟ ⎥ dx − P [ a1 + a2 ]
L L
Π (u ) = ∫ ⎨ ⎢
1
+ 2
0
⎩⎪ 2 ⎣ L L ⎝ L ⎠ ⎦ ⎭⎪ ⎣⎢ L ⎝ L ⎠ ⎦⎥
Variation of a1, a2
L ⎧ ⎡ a1 2a2 ⎛ x ⎞ ⎤ ⎡ δ a1 2δ a2 ⎛ x ⎞ ⎤ ⎫
δΠ (u ) = ∫ ⎨ EA( x) ⎢ + ⎜ ⎟⎥ ⎢ + ⎜ ⎟ ⎥ ⎬ dx
0
⎩ ⎣L L ⎝ L ⎠⎦ ⎣ L L ⎝ L ⎠⎦ ⎭
⎡ x ⎛ ⎞ ⎤
x
2

d − P [δ a1 + δ a2 ] = 0
L
−∫ fb ⎢δ a1 + δ a2 ⎜ ⎟ ⎥ dx
0
⎢⎣ L ⎝ L ⎠ ⎥⎦
 Method #2

Total Potential
L ⎡ EA( x) ⎛ du ⎞ 2 ⎤ L 2
Π (u ) = ∫ ⎢ ⎜ ⎟ ⎥ dx − ∫0 fb udx − Pu ( L) x ⎛x⎞
u ≈ u = a1 + a2 ⎜ ⎟
0
⎢⎣ 2 ⎝ dx d ⎠ ⎥⎦ L ⎝L⎠
Variation of u
L ⎡ du ⎛ du ⎞ ⎤ L
δΠ (u ) = ∫ ⎢ EA( x) dx δ ⎜ dx ⎟ ⎥ dx − ∫0 f bδ udx − Pδ u ( L) = 0
0
⎣ ⎝ ⎠⎦
2
du a1 2a2 ⎛ x ⎞ ⎛ du ⎞ δ a1 2δ a2 ⎛ x ⎞ x ⎛x⎞
= + ⎜ ⎟ δ⎜ ⎟= + ⎜ ⎟ δ u = δ a1 + δ a2 ⎜ ⎟
dx L L ⎝L⎠ ⎝ dx ⎠ L L ⎝L⎠ L ⎝L⎠

L ⎧ ⎡ a1 2a2 ⎛ x ⎞ ⎤ ⎡ δ a1 2δ a2 ⎛ x ⎞ ⎤ ⎫
δΠ (u ) = ∫ ⎨ EA( x) ⎢ + ⎜ ⎟⎥ ⎢ + ⎜ ⎟ ⎥ ⎬ dx
0
⎩ ⎣L L ⎝ L ⎠⎦ ⎣ L L ⎝ L ⎠⎦ ⎭
⎡ x ⎛x⎞ ⎤
2

f b ⎢δ a1 + δ a2 ⎜ ⎟ ⎥ dx − P [δ a1 + δ a2 ] = 0
L
−∫
0
⎢⎣ L ⎝ L ⎠ ⎥⎦
 L ⎧ ⎡ a1 2a2 ⎛ x ⎞ ⎤ ⎡ δ a1 2δ a2 ⎛ x ⎞ ⎤ ⎫
δΠ (u ) = ∫ ⎨ EA( x) ⎢ + ⎜ ⎟⎥ ⎢ + ⎜ ⎟ ⎥ ⎬ dx
0
⎩ ⎣L L ⎝ L ⎠⎦ ⎣ L L ⎝ L ⎠⎦ ⎭
⎡ x ⎛ ⎞ ⎤
x
2

d − P [δ a1 + δ a2 ] = 0
L
−∫ fb ⎢δ a1 + δ a2 ⎜ ⎟ ⎥ dx
0
⎢⎣ L ⎝ L ⎠ ⎥⎦
Rearranging
⎧ L EA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤ L x ⎫
⎨ ∫0 ⎢ + ⎜ ⎟ ⎥ dx − ∫0 f b dx − P ⎬ δ a1 +
⎩ L ⎣L L ⎝ L ⎠⎦ L ⎭
⎧⎪ L 2 xEA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤ ⎛ ⎞
x
2
⎪⎫
⎨ ∫0 ⎢ + ⎜ ⎟ ⎥ dx − f b ⎜ ⎟ dx − P ⎬ δ a2 = 0
⎪⎩ L2
⎣L L ⎝ L ⎠⎦ ⎝L⎠ ⎪⎭
Simultaneous Equations

EA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤
L L x
( ∂Π ∂a1 = 0 ) ∫0 L ⎢⎣ L L ⎜⎝ L ⎟⎠ ⎦⎥ ∫0 b L ddx − P = 0
+ d
dx − f

L 2 xEA( x ) ⎡ a ⎤
2
( ∂Π ∂a2 = 0 ) 2a2 ⎛ x ⎞ ⎛x⎞
∫0 L2 ⎢⎣ L L ⎜⎝ L ⎟⎠ ⎥⎦
1
+ dx − f b⎜
⎝L⎠
⎟ dx − P
 L EA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤ L x
∫0 L ⎣⎢ L + L ⎜⎝ L ⎟⎠⎦⎥ dx d − ∫ fb ddx − P = 0
0 L
L 2 xEA( x ) ⎡ a 2a2 ⎛ x ⎞ ⎤
2
⎛x⎞
∫0 L2 ⎢⎣ L + L ⎜⎝ L ⎟⎠ ⎥⎦ dx − fb ⎜⎝ L ⎟⎠ dx − P
1

Determination of coefficients

7 fb L2 + 6 PL 3( PL − fb L2 )
a1 = , a2 =
13Ec 13Ec

Approximate solution

7 f b L2 + 6 PL x 3( PL − fb L2 ) ⎛ x ⎞
2

u≈ + ⎜ ⎟
13Ec L 13Ec ⎝L⎠
 EA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤
L L x
∫0 L ⎢⎣ L + L ⎜⎝ L ⎟⎠⎥⎦ dx − ∫0 fb L dx − P = 0
L 2 xEA( x ) ⎡ a 2a2 ⎛ x ⎞ ⎤
2
⎛x⎞
∫0 L2 ⎢⎣ L + L ⎜⎝ L ⎟⎠⎥⎦ dx − fb ⎜⎝ L ⎟⎠ dx − P
1

fb = 0
L EA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤
∫0 L ⎣L ⎢ + ⎜ ⎟ ⎥ dx − P = 0
L ⎝ L ⎠⎦ a1 , a2
L 2 xEA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤
∫0 L2 ⎢
⎣L
+ ⎜ ⎟ ⎥ dx − P = 0
L ⎝ L ⎠⎦

Galerkin Method
 Comparison
7 f b L + 6 PL x 3( PL − f b L ) ⎛ x ⎞
2 2 2
Approximate u≈ + ⎜ ⎟
13Ec L 13Ec ⎝L⎠
fb L 2 fb L2 x ( f L − P) L x
Exact u ( x) = x+ ln(1 − ) + b ln(1 − )
Ec Ec 2L Ec 2L
fb = 0
− PL x
Exact Solution u ( x) = ln(1 − )
Ec 2L

3PL 2 x ⎛ x ⎞ ⎤
⎡ 2
Approximate Solution u≈ ⎢ +⎜ ⎟ ⎥
13Ec ⎢⎣ L ⎝ L ⎠ ⎥⎦
13Ec
 Displacement at the Tip

fb = 0
A = 2c − ( x / L ) c
L=2
E d Deflection
End D fl ti
Example. Natural Frequency of
a Vibration
Vib ti P
Problem
bl

Governing Equation

∂ ⎛ ∂u ⎞ ∂ 2u
⎜ EA ⎟ − ρ A 2 = 0 0< x< L
∂x ⎝ ∂x ⎠ ∂t
Body force
Solution

u ( x , t ) = u0 ( x ) e iω t b = ρ Aω 2u0

d ⎛ du0 ⎞ ∂ 2u0
⎜ EA ⎟ + ρ Aω 2
u0 = 0 E 2 +b = 0
dx ⎝ dx ⎠ ∂x
∂x
 WRM(Bubnov-Galerkin)
Trial Function ρ L2 kL
2 L =1 =1 =1
u0 = ∑ ci x
EA
i
satisfy geometric B.C. u0 (0) = 0 E
i=1

du ( L) 3c2 L
F
Force B
B.C.
C AE 0 = − ku0 ( L) c1 = −
dx 2
c2 c
u0 = − (3Lx − 2 x 2 ) = a1 (3Lx − 2 x 2 ) a1 = − 2 f1 = ((3Lx − 2 x 2 )
2 2
d 2 f1
∫
L
fi R dD = 0, i = 1, 2,.., n 0=∫ f1 ( EAa1 2 + ρ Aω 2 a1 f1 )dx
D 0 dx
⎛ 4 2 10 ⎞
⎜ ω − ⎟ a1 = 0 ω1 = 4.167
⎝5 3⎠

Exact Solution ω1 = 2.029, ω2 = 4.913

 Variational statement
L ⎡ EA ⎛ du0 ⎞ 2 ⎤ k
Π=∫ ⎢ ⎜ ⎟ − bu 0 ⎥ dx + u 0 ( L ) 2
0
⎢⎣ 2 ⎝ dx ⎠ ⎥⎦ 2

δΠ = 0 ∀u0 ( x)

L ⎧ du0 ⎛ du0 ⎞ ⎫
δΠ = ∫ ⎨ EA δ⎜ ⎟ − b δ u0 ⎬ dx + ku0 ( L) δ u0 ( L) = 0
0
⎩ dx ⎝ dx ⎠ ⎭
Trial Function
2
u0 ≈ u0 = ∑ ai x i
satisfy geometric B.C. only
i =1
∫{ }
L
EA( a1 + 2a2 x)(δ a1 + 2 xδ a2 ) − ρ Aω 2 ( a1 x + a2 x 2 )( xδ a1 + x 2δ a2 ) dx
0

+ k ( a1 L + a2 L2 )( Lδ a1 + L2δ a2 ) = 0

{ 1 2 1 2 }
⎡ L EA(a + 2a x) − ρ Aω 2 (a x + a x 2 ) x dx +k (a L + a L2 ) L ⎤ δ a
⎢⎣ ∫0 1 2 ⎥⎦ 1

{ }
+ ∫ EA(a1 + 2a2 x)2 x − ρ Aω 2 (a1 x + a2 x 2 ) x 2 dx +k (a1 L + a2 L2 ) L2 ⎤ δ a2 = 0
⎡ L

⎢⎣ 0 ⎥⎦

⎧ ρ Aω 2 L3 ⎫ ⎧ ρ Aω 2 4
L 3⎫
⎨ EAL − + kL ⎬ a1 + ⎨ EAL − + kL ⎬ a2 = 0
2 2

⎩ 3 ⎭ ⎩ 4 ⎭

⎧ ρ Aω 2 L4 3⎫ ⎧ 4 EAL3 ρ Aω 2 L5 4⎫
⎨ EAL − + kL ⎬ a1 + ⎨ − + kL ⎬ a2 = 0
2

⎩ 4 ⎭ ⎩ 3 5 ⎭
 ⎧ ρ Aω 2 L3 ⎫ ⎧ ρ Aω 2 4
L 3⎫
⎨ EAL − + kL ⎬ a1 + ⎨ EAL − + kL ⎬ a2 = 0
2 2

⎩ 3 ⎭ ⎩ 4 ⎭
⎧ ρ Aω 2 4
L ⎫ ⎧ 4 EAL 3
ρ A ω 2 5
L 4⎫
⎨ EAL2
− + kL3
+
⎬ 1 ⎨
a − + kL ⎬ a2 = 0
⎩ 4 ⎭ ⎩ 3 5 ⎭

Eigenvalue Problem L =1
ρ L2
⎡ 2 2 ⎤ ⎧ a1 ⎫ ⎡1 1 ⎤ =1
2 ⎢ 3 4 ⎥ ⎧ a1 ⎫ E
⎢ ⎥ ⎨ ⎬ = ω ⎨ ⎬
⎢⎣ 2 7 ⎢1 ⎥
3 ⎥⎦ ⎩ 2 ⎭ 1 ⎩a2 ⎭
a kL
⎣ 4 5⎦ =1
EA

Approximate Solution ω1 = 2.038,, ω2 = 6.026

Exact Solution ω1 = 2.029, ω2 = 4.913

Galerkin Method
 d ⎛ du ⎞
⎜ AE ⎟=0
dx ⎝ dx ⎠
Bubnov-Galerkin
d ⎛ du ⎞
R= ⎜ AE ⎟
dx ⎝ dx ⎠
∫
D
fi R dD = 0, i = 1, 2,.., n
Px
u= + a2 ( x 2 − 4 x )
2
 Galerkin

d ⎛ du ⎞
2 2 ⎜ AE ⎟=0
∫0
fi Rdx = ∫ f i ( AEu ′)′dx = 0
0
dx ⎝ dx ⎠

Integration by parts
d ⎛ du ⎞
R= ⎜ AE ⎟
dx ⎝ dx ⎠
[ fi AEu′]0 − ∫0
2
fi ′AEu′dx = 0
2
2
x ⎛x⎞
u ≈ u = a1 + a2 ⎜ ⎟
L ⎝L⎠

2 a1 ⎡ a1 2a2 x ⎤
a1 P − ∫ AE ⎢ + ⎥ dx = 0
0 L ⎣L L L⎦

⎛ 2a x ⎞
2 ⎡ a 2a x ⎤
a2 P − ∫ ⎜ 2 ⎟ AE ⎢ 1 + 2 ⎥ dx = 0
0
⎝ L L⎠ ⎣L L L⎦
 Galerkin

L EA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤
∫ 0 L ⎣L ⎢ + ⎜ ⎟ ⎥ dx − P = 0
L ⎝ L ⎠⎦
a1 , a2
L EA( x) ⎡ a1 2a2 ⎛ x ⎞ ⎤
2 xEA
∫
0 L2 ⎢
⎣L
+ ⎜ ⎟ ⎥ dx − P = 0
L ⎝ L ⎠⎦

Ritz Method
 Small function set Trial function
L2 m [φ ] = r n

Geometric B
B.C.
C φ = ∑aj f j
Geometric B.C. j =1
Force B.C.
Force B.C. 2 m −1
C
WRM Galerkin
Bubnov-Galerkin
Integration by Parts

∫D
fi R (φ ) dD = 0 FEM Inclusion of Force B.C.

i = 1, 2,.., n
Large
g function set

geometric B.C.

Ritz Method m −1
C
Minimize Π (φ ) Hamilton’s principle

δΠ (φ ) = 0 Principle of virtual work

 ( v ") dx − ⎢⎣ ∫0 qv dx + v( L) P + v '( L) M ⎥⎦⎤
⎡
L EI L
Π (v ) = ∫
2
0 2

v
Continuous
v′