MAE4700/5700

Finite Element Analysis for

Mechanical and Aerospace Design
Cornell University, Fall 2009

Nicholas Zabaras
Materials Process Design and Control Laboratory
Sibley School of Mechanical and Aerospace Engineering
101 Rhodes Hall
Cornell University
Ithaca, NY 14853-3801

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
1
 Weak Form
• One of the key elements of the FE analysis is the
derivation of the weak (integral) form of the
governing differential equations (strong form).
• This is the most critical step in the FEM.
• The process is common in all FEM procedures.
• Introduction of interpolation functions in the weak
form will lead to the discretized FE equations.
• In this lecture, we work with 1D boundary value
problems governed by ordinary differential eqs.
(ODEs).

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
2
Strong form: Axial loading of an elastic bar

x=0 x=L
• Consider an elastic bar loaded as shown above. We
introduce the following notation:
u( x ) Displacement at po int x
σ ( x ) Stress ( force / area ) at po int x
du
ε ( x) = Strain at po int x
dx
F ( x ) Internal force at po int x
A( x ) Cross sec tional area at po int x
b( x ) Distributed load ( force / length ) at po int x
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
3
Strong form: Axial loading of an elastic bar
F ( x) b( x ) F ( x + dx )

dx
− F ( x ) + b( x + )dx + F ( x + dx) = 0
x=0 x=L 2
F ( x + dx) − F ( x) dx
+ b( x + ) = 0, take dx → 0,
dx 2
dF
+b = 0
dx
• We apply force balance on a differential element.
• We will use the displacement u( x ) as the main unknown.
• Using σ = Eε ( Hooke ' s law − constitutive equation), we can write:
du
F = σ A = Eε A = EA , and the differential equation becomes:
dx d du
( EA ) + b = 0, 0 < x < L
dx dx
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
4
Strong form: Axial loading of an elastic bar
Sign convention
¾ The prescribed traction is taken as
t positive when is acting in the
positive direction regardless on
u( L) = u which face (left or right) is acting.
x=L ¾ The stress is positive in tension &
x=0 negative in compression.
• To complete our problem definition, we need boundary
conditions – in each end we need to prescribe either the
displacement or the traction (force/area).
• For demonstration, let us take prescribed displacement u
at x = L and prescribed traction t > 0 at x = 0 .
• So the complete strong form of our BVP is the following:
d du
( EA ) + b = 0, 0 < x < L
dx dx
du
u ( L) = u σ (0) = E (0) = −t
dx
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
5
 Heat conduction in one-dimension

x=0 x=L

• Consider heat conduction through a bar as shown above.

We introduce the following notation:
T ( x ) Temperature at po int x
q( x ) Heat flux ( energy /( area × time)) at po int x

A( x ) Cross sec tional area at po int x

b( x ) Heat source ( energy / length ) at po int x

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
6
 Strong form: Heat conduction in a bar
q( x ) A( x ) b( x ) q( x + dx ) A( x + dx )

dx
− q ( x) A( x) − b( x + ) dx + q( x + dx) A( x + dx) = 0
 
2
  

 
x=L
heat flow in the CV heat flow out of the CV
x=0 heat generated in t he CV

q ( x + dx) A( x + dx) − q ( x) A( x) dx
− b( x + ) = 0, take dx → 0,
dx 2
d ( qA )
=b
dx
• We apply energy balance on a differential element.
• We will use the temperature T ( x ) as the main unknown.
dT
Substituting, q = − k ( Fourier law − constitutive equation), leads to:
dx
d dT
− ( kA ) = b, 0 < x < L
dx dx

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
7
 Strong form: Heat conduction in a bar
Sign convention
¾ The prescribed flux is taken as
q
positive if heat flows out of the bar.
¾ The heat flux q(0) as defined enters
T ( L) = T the system (i.e. is negative)
x=0 x=L

• To complete our problem definition, we need boundary

conditions – in each end we need to prescribe either the
temperature or the heat flux.
• For demonstration, let us take prescribed temperature T
at x = L and prescribed heat flux q > 0 at x = 0.
• So the complete strong form of our BVP is the following:
d dT
( kA ) + b = 0, 0 < x < L
dx dx
dT
T ( L) = T q (0) = − k (0) = − q
dx
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
8
 Weak form: Axial loading of an elastic bar

u( L) = u
x=0 x =L L
⎛d du ⎞
∫0 ⎜⎝ dx dx ⎟⎠w( x )dx = 0, ∀w( x ) in 0 < x < L
( EA ) + b
d du
( EA ) + b = 0, 0 < x < L u ( L) = u
dx dx
⎛ ⎛ ⎞⎞
⎜ ⎜ du ⎟⎟
u ( L) = u
du
σ (0) = E (0) = −t wA
⎜ ⎜ dx E (0) + t ⎟ ⎟ = 0 ∀w on x = 0
dx ⎜ ⎜  
⎟⎟
⎝ ⎝ σ ⎠ ⎠ x =0
• To derive the weak form, multiply the ODE by an arbitrary weight
function w(x) and integrate on the domain.
• Similarly for the natural BC.
• These two equs. to be equivalent with those in the strong form, need to
be true for ALL w. We assume (we will come to this shortly) that w(L)=0.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
9
 Weak form: Axial loading of an elastic bar
L
⎛d du ⎞
Compute u(x): ∫0 ⎜⎝ dx dx ⎟⎠w( x )dx = 0, ∀w( x ) in 0 < x < L
( EA ) + b

u( L) = u
⎛ ⎛ du ⎞⎞
⎜ ⎜ dx
wA E (0) + t ⎟ ⎟ = 0 ∀w in 0 ≤ x ≤ L
⎝ ⎝ ⎠ ⎠ x =0

• Integration by parts of the 1st equation gives:

du x = L L du dw L
EA w |x =0 − ∫ EA dx + ∫ bwdx = 0, ∀w( x ) in 0 < x < L, with w( L) = 0
dx 0 dx dx 0

• Using the weak form of the traction BC and that w( L) = 0 , we

can Lsimplify: L
du dw
∫ EA
0
dx dx
dx = wAt |x =0 + ∫ bwdx, ∀w( x) in 0 < x < L, with w( L) = 0
0

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
10
 Weak problem statement
• Find u(x) among the smooth functions with u( L) = u such that
L L
du dw
∫0 dx dx dx = wAt |x=0 + ∫0 bwdx, ∀w( x) in 0 < x < L, with w( L) = 0
EA

• The integration by parts leads to a symmetric form in u and

w – this leads to a symmetric stiffness after finite element
discretization is introduced.
• The weak statement is EQUIVALENT to the strong form. In
other words, not only if u(x) satisfies the strong form the
above is true, but also if u(x) satisfies the weak form then it
is also the solution of the strong form of our problem.
• Smoothness in u and w is needed for the integrals in the
weak form to exist (we need H1 functions u and w that
have 1st derivative that is squared integrable)!
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
11
1st derivative squared integrable functions
• What smoothness conditions on u and w we need for the
integrals in the weak form to exist?
L L
du dw
∫0 dx dx dx = wAt |x=0 + ∫0 bwdx, ∀w( x) in 0 < x < L, with w( L) = 0
EA

• In the first integral, we have product of 1st order derivatives

of u and w. We need to have functions for which this integral
exists.
• Thinking for a moment (because of the symmetry in u and
w) that u=w, we want trial u and weight w functions that are
1st derivative square integrable, i.e. functions v( x) ∈ H 1 ( I ),
I = [0, L] such that:
2
⎛ dv ⎞
L

∫0 ⎜⎝ dx ⎟⎠ dx < ∞, EA > 0
EA

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
12
Continuous functions with piecewise continuous derivatives
• Let us select as a function space to define the weak form
the following:
V = {v : v is continuous on I = [0, L], v ' is piecewi se c ontinuous and bounded on I }
• Can you verify that this type of functions will work for our
weak form?
• A function in V is piecewise continuously differentiable, i.e.
its first derivative is continuous except at selected points.
• Find u ∈ V , with u ( L) = u, L L
du dw
such that ∀w ∈ V , with w( L) = 0, ∫0 EA dx dx dx = wAt |x=0 + ∫0 bwdx
the following holds:
• A function with V-continuity, can have discontinuity (kinks) in
its slope.

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
13
 C-1 continuity
• Consider functions (C-1 continuity) with discontinuities
(jumps) and discontinuities in slope (kinks)
• Such functions that are not infinite have 1st order derivatives
that are integrable (you can integrate delta functions!)
• However, note that in our weak form we have a product of
L

the 2 derivatives: ∫ EA du dw
dx dx
dx
0

• If both u and w are in C-1 and have a discontinuity at the

same point a of magnitudes α and β, respectively, then the
1st integral in the eq. above takes the form:
L
This integral
∫ αβδ ( x − a )dx
2
does not exist!
0
• So the C-1 functions are not good candidates for our weight
and trial functions.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
14
 C0 functions
• Let us define the C0-continuity functions as follows:
{
C 0 ( I ) = v : v is a continuous function defined on I = [0, L] }
• These functions in general are not appropriate as weight
and trial functions in our weak form.
1 1
− ( )λ x x≤
2 2
• Indeed, consider the following function: u ( x) =
1 1 1
( x − )λ − ( )λ x x>
2 2 2

• It is C0 continuous for all λ>0, however an H1 function (e.g.

1st derivative square integrable) only for λ>1/2. Verify this by
computing 1 ⎛ du ⎞2 dx analytically.
∫ ⎜⎝ dx ⎟⎠
0

C0 continuous but
not H1 function

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
15
Continuous functions with piecewise continuous derivatives versus H1

• We will see in an upcoming lecture, that in finite

element analysis we consider piecewise
polynomial (linear, quadratic, etc.) functions on
subdivisions (triangulations) of a bounded domain
I = [0, L] on elements.

• Our selection of finite element approximation

space Vh ⊂ H 1 ( I ) , is satisfied in this case by
selecting Vh ⊂ C 0 ( I )
• This is the case since v ∈ C 0 ( I ) will be shown to be
continuous between the boundaries of adjacent
elements and that its first derivative exists and is
piecewise continuous so that v ∈ H 1 ( I )
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
16
 Weak formulation in H1
• Find u ∈ V , with u ( L) = u, L L
du dw
such that ∀w ∈V , with w( L) = 0, ∫0 dx dx dx = wAt |x=0 + ∫0 bwdx
EA

the following holds:

V = {v : v is continuous on I = [0, L], v ' is piecewi se c ontinuous and bounded on I }

• When producing weak (variational) formulations of boundary

value problems, it is (from mathematical point of view)
natural and very useful to work with function spaces that are
slighlty larger (i.e. spaces that contain more functions) than
the spaces of continuous functions with piecewise
continuous derivatives introduced earlier.
• It is also useful to endow V with scalar products and
corresponding norms related to the actual boundary value
problem.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
17
 The weak form in H1
• We modify the weak problem statement as follows:

Find u ∈ H , with u ( L) = u,
1
L L
du dw
such that ∀w ∈ H 1 , with w( L) = 0, ∫0 dx dx dx = wAt |x=0 + ∫0 bwdx
EA
the following holds:
• The space H1, is the space of 1st derivative square
integrable functions.
2
1
L
⎛ du ⎞ we will later introduce
Wint (u ) = ∫ EA ⎜ ⎟ dx < ∞, EA > 0 as the energy norm. Wint (u )

2 ⎝ dx ⎠
Internal 0
strain
energy

• H1 is the largest space we can write the above weak form

and it naturally arises from the specific BVP. We will also
see that the basic error estimate for the FE is in the H1
norm.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
18
 Square integrable functions L2(I)
• Consider in 1D the interval I=(0,L). We define the space of
square integrable functons on I as:
⎧ ⎫
L2 ( I ) = ⎨v : v is defined on I and ∫ v dx < ∞ ⎬
2

⎩ I ⎭
• The space L2(I) is a Hilbert space with the (L2-) scalar
(inner) product: (v, w) L ( I ) = ∫ vwdx
2
1/2
I ⎛ 2 ⎞
and corresponding (L2-) norm: || v ||L2 ( I ) = ⎜ ∫ v dx ⎟ = (v, v)1/2
⎝I ⎠
• By Cauchy’s inequality:
| (v, w) L2 ( I ) |≤|| v ||L2 ( I ) || w ||L2 ( I )

• For this class, think of v ∈ L2 ( I ) as a piecewise continuous

function, possibly unbounded, such that: ∫ v 2 dx < ∞.
I

• As an example, for x ∈ I = (0,1), v( x) = x − β ∈ L2 ( I ) if β < 1 / 2.

MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
19
First derivative square integrable functions H1(I)
• We formally introduce the space H1(I), I=(0,L) as follows:
⎧ dv ⎫
H 1 ( I ) = ⎨v : v and v ' ≡ belong to L2 ( I ) ⎬
⎩ dx ⎭

• The space H1(I) consists of the functions defined on I which

together with their 1st derivatives are square integrable.
• This space is equipped with the following inner (scalar)
product:
(v, w) H 1 ( I ) = ∫ (vw + v ' w ')dx
1/2
⎛ ⎞
I
• The corresponding norm is: || v ||H 1 ( I ) = ⎜ ∫ ( v + (v ') )dx ⎟
2 2

⎝I ⎠
• When v(0)=0 and/or v(L)=0, in literature, we often denote
the corresponding space as
H 10 ( I ) = {v ∈ H 1 ( I ) : v( L) = 0}
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
20
 Recall smoothness conditions for beams
• Recall that beam shape functions were C1 continuous –
both displacements and slopes were continuous (4th other
ODEs).
dV d 2 ⎛ d u y ⎞
2

q=− = 2 ⎜ EI ⎟⎟
dx dx ⎜⎝ dx 2 ⎠

e e
E I 2 E I e e d 2u ey
Ue = ∫e 2 κ = ∫e 2 dx 2 dx
2
dx ( )
Ω Ω

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
21
 Weak formÆ Strong form
• We will here show that if u satisfies the weak form, it is also
a solution of the strong form. Let u:
L
L du dw
∫
0
EA
dx dx
dx = wAt |x =0 + ∫ bwdx, ∀w( x ) in 0 < x < L, with w( L) = 0
0

• Integrate backwards the first term as follows:

L
du L L d ⎛ du ⎞
EA w |0 − ∫ ⎜ EA ⎟ wdx = wAt |x =0 + ∫ bwdx, ∀w( x ) in 0 < x < L, with w( L) = 0 ⇒
dx 0 dx ⎝ dx ⎠ 0
L⎡d ⎛ du ⎞ ⎤
∫0 ⎢⎣ dx ⎜⎝ dx ⎟⎠ + b ⎥⎦ wdx + wA(t + σ ) |x=0 = 0, ∀w( x) in 0 < x < L, with w( L) = 0
EA

⎛d ⎛ du ⎞ ⎞
• Since w(x) is arbitrary, select w = ψ ( x ) ⎜ dx ⎜ EA ⎟ + b⎟
⎝ ⎝ dx ⎠ ⎠
withψ ( x ) > 0 a smooth function withψ (0) = ψ ( L) = 0 e.g.ψ = x ( L − x )
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
22
 Weak formÆ Strong form
L ⎡d ⎛ du ⎞ ⎤
∫0 ⎢ dx ⎜
⎣ ⎝
EA ⎟ + b ⎥ wdx + wA(t + σ ) |x =0 , ∀w( x ) in 0 < x < L, with w( L) = 0
dx ⎠ ⎦

• With this selection of w, the Eq. above becomes:

2
L ⎡d ⎛ du ⎞ ⎤
∫0 ψ ( x ) ⎢ dx ⎜
⎣ ⎝
EA ⎟ + b ⎥ dx = 0
dx ⎠ ⎦

• Withψ ( x ) > 0 , the above can be true only if:

d ⎛ du ⎞
⎜ EA ⎟ + b = 0, 0 < x < L
dx ⎝ dx ⎠

• This proves that if u is a solution of the weak form, it also

satisfies the ODE strong form. It remains to show that it
satisfies the natural BC.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
23
 Weak formÆ Strong form
L ⎡d ⎛ du ⎞ ⎤
∫0 ⎢ dx ⎜
⎣ ⎝
EA ⎟ + b ⎥ wdx + wA(t + σ ) |x =0 = 0, ∀w( x) in 0 < x < L, with w( L) = 0
dx ⎠ ⎦

• Returning to the above equation, we have:

wA(t + σ ) |x =0 = 0, ∀w( x) with w( L) = 0

• Since w is arbitrary, select it such that w(0)=1, w(L)=0, then

σ = −t at x = 0

• In conclusion, if u is a solution of the weak form, it also

satisfies the complete strong form of our BVP.

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
24
Strong form: Axial loading of an elastic bar
• Let us generalize the earlier examined problem by allowing
arbitrary part of the boundary Γt with prescribed
displacement and the remaining part Γu with prescribed
traction:
d du
( EA ) + b = 0, 0 < x < L
dx dx
Γu ∩ Γt = 0
u = u on Γu
Γu ∪ Γt = Γ
du
σ n = E n = t on Γt
dx
• The nornal vector n is introduced to allow arbitrary
application of the traction BC (recall t > 0). If n=1 (right
boundary), then σ = t on Γt . If a positive force is applied at
the left boundary (n=-1), then σ = −t on Γt .
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
25
Weak form: Axial loading of an elastic bar
• Multiplying the ODE with w and integrating by
parts gives:
du du dw
EA wn |Γ − ∫ EA dx + ∫ bwdx = 0, ∀w( x) in 0 < x < L, with w = 0 on Γu ⇒
dx Ω
dx dx Ω
du dw
∫Ω dx dx dx = Awt |Γ t + Ω∫ bwdx, ∀w( x) in 0 < x < L, with w = 0 on Γu
EA

• We require that this holds for u( x ) ∈ H 1,

where
u = u on Γ and for all w( x ) ∈ H with w = 0 on Γ . H 1 is the
1
u u

space of 1st derivative square integrable

functions.
• As before, the weak form is symmetric in u and
w (first derivatives for both u and w). This will
lead to a symmetric stiffness matrix.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
26
Summary of weak forms for the two problems
• Axial deformation problem: Find u( x ) ∈U, U = {u ( x) ∈ H 1 , u = u in Γu }
du dw
∫ EA
Ω
dx dx
dx = Awt |Γ t + ∫ bwdx, ∀w( x) ∈ U 0 ,
Ω

U 0 = {w( x ) ∈ H 1 , w = 0 in Γu }

• Heat conduction problem: Find u( x ) ∈U , U = {u ( x) ∈ H 1 , u = u in Γu }

du dw
∫Ω dx dx dx = − Awq |Γ q +Ω∫ bwdx, ∀w( x) ∈U 0 ,
Ak

U 0 = {w( x ) ∈ H 1 , w = 0 in Γ u }

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
27
 Minimum potential energy
• We used earlier the principle of mimimum potential
energy to derive the FEM eqs.
• How is this method related to the weak form? Let
us revisit the elastic bar problem.

u( L) = u
x=0 x=L ⎛ ⎞
1 du 2 ⎜ ⎟
Find u ( x) ∈ U ,
 
N ∫ AE ( ) dx − ⎜ ∫ ubdx + (uAt ) |Γt ⎟
such that min
 
⎜⎜ Ω 
⎟⎟
u( x) 2 Ω
Kinematically admissible
dx
displacements ,
satisfy kinematic
Wint ⎝ Wext ⎠
( displacement ) boundary
conditions u ( L ) =u

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
28
 Minimum potential energy
Find u ( x) ∈ U , such that min Ρ, where
u( x)

⎛ ⎞
1 du 2 ⎜ ⎟
P = ∫ AE ( ) dx − ⎜ ∫ ubdx + (uAt ) |Γt ⎟
 
⎜⎜ Ω 
⎟⎟
2Ω dx

Wint ⎝ Wext ⎠

• We will show that the minimizer of the potential energy

satisfies the weak form (and thus the strong form).
• P(u(x)) is a functional – takes a function as an input and
returns a scalar! We need to use calculus of variations for
its mimimization.
• A variation of the function u is denoted by δu(x) ≡ζw(x)
where w(x) is an arbitrary function, and 0<ζ<1 is a very
small positive number.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
29
 Variation of a functional
• The corresponding change in the functional is called
the variation in the functional and denoted by δW ,
which is defined by

δ P =P(u(x)+ζ w(x))-P(u(x))=P(u(x)+δ u(x))-P(u(x))

• From the principle of minimum potential energy, it is

clear that the function u(x) +ζ w(x) must still be in U. To
meet this condition, w(x) must be smooth and vanish
on the essential boundaries, i.e. w ∈ U 0
• So w(x) vanishes at the parts of the boundary with
essential boundary conditions!

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
30
 Variation of the internal energy
• Let us evaluate the variation of the first term in Wint .
2 2
1 ⎛ du dw ⎞ 1 ⎛ du ⎞
δ Wint = ∫ AE ⎜ + ς ⎟ dx − ∫ AE ⎜ ⎟ dx
2Ω ⎝ dx dx ⎠ 2Ω ⎝ dx ⎠
1 ⎛ ⎛ du ⎞2 du dw ⎛ dw ⎞
2
⎞ 1 ⎛ ⎞
du
2

= ∫ AE ⎜ ⎜ ⎟ + 2ς +ς 2 ⎜ ⎟ ⎟⎟ dx − ∫ AE ⎜ ⎟ dx
⎜
2Ω ⎝ ⎝ dx ⎠ dx dx ⎝ dx ⎠ ⎠ 2Ω ⎝ dx ⎠

• Dropping the higher order terms in ζ, leads to:

du dw
δ Wint = ς ∫ AE dx
Ω
dx dx

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
31
 Variation of the external work
• Let us evaluate the variation of the 2nd term, Wext .
δ Wext = ∫ (u + ς w)bdx + (u + ς w) At |Γ − ∫ ubdx − uAt |Γ = ∫ ς wbdx + ς wt |Γ
t t t
Ω Ω Ω

• Combining the last 2 terms, the variation of the

potential energy is given as:
du dw
δ P = ς ∫ AE dx − ∫ ς wbdx − ς wAt |Γt
Ω
dx dx Ω

• Minimization of the potential energy requires: δ P = 0,

for any δ u = ς w, i.e. for any ς :
du dw
∫ AE
Ω
dx dx
dx = ∫ wbdx + wAt |Γt
Ω

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
32
 Variational principle
du dw
∫ AE
Ω
dx dx
dx = ∫ wbdx + wAt |Γt
Ω

• This is identical with our weak form. So minimization of

the potential energy leads to the same solution as the
weak problem, thus
Strong p r oblem ⇔ Weak problem ⇔ Variational problem
• Note that now we understand why in the weak form we
considered w to vanish on the boundary with essential
conditions: δ u = ς w( x) and the miminization is wrt
functions u(x) that satisfy the essential BCs, thus

δ u |Γ = ς w( x) |Γ = 0 ⇒ w( x) |Γ = 0
u u u

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
33
 Principle of virtual work
• The minimization of P, δ P = 0,
du dw
δ P = ς ∫ AE dx − ∫ ς wbdx − ς wAt |Γt = 0
Ω
dx dx Ω

• This can also be written as (δ u ≡ ζ w) :

du d δ u
δP = ∫E Adx − ∫ δ ubdx − δ uAt |Γt = 0
N
dx dx
Ω N N dV Ω
σ δε
• This is simplified as: ⎛ ⎞
⎜ ⎟
δ P = ∫ σδε dV − ⎜ ∫ δ ubdx + δ uAt |Γt ⎟ = 0

Ω
 
⎜⎜ Ω 
⎟⎟
δ Wint ⎝ δ Wext ⎠

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
34
 Principle of virtual work
⎛ ⎞
⎜ ⎟
δ P = ∫ σδε dV − ⎜ ∫ δ ubdx + δ uAt |Γt ⎟ = 0

Ω
 
⎜⎜ Ω 
⎟⎟
δ Wint ⎝ δ Wext ⎠
• The internal work done by the actual stresses on the
arbitrary strains δε ≡ d δ u induced by the arbitrary
dx
displacements δ u ∈U 0 is equal to the external work done
by the body forces b and applied tractions t on the
arbitrary displacement field δ u .

• This statement is of course the same as the weak form

but presented in ` mechanics’ language.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
35
 Limitations of the variational principle
• There are many problems to which the variational
principle is not applicable. For example, variational
principles cannot be developed for the advection-
diffusion equation.

• Variational principles can only be developed for

systems that are self-adjoint.

• The weak form for the advection-diffusion equation

is not symmetric, and it is not a self-adjoint
system.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
36
 Example problem 1
• Derive the weak form for the following BVP:
d du du
(10 ) + 2 x = 0, 1 < x < 3, (1) = 0.1, u (3) = 0.001
dx dx dx

• We multiply by w and integrate from 1 to 3, for

every w, w(3)=0.
⎛d ⎞
3
du
∫1 ⎜⎝ dx dx ) + 2 x ⎟⎠wdx = 0, ∀w, w(3) = 0
(10

• Integration by parts of the 1st term gives:

3 3
du du dw
10 w |xx==13 − ∫ 10 dx + ∫ 2 xwdx = 0, ∀w, w(3) = 0 ⇒
dx 1
dx dx 1
3 3
du du du dw
w(3) − 10 (1) w(1) − ∫ 10
10 (3) N dx + ∫ 2 xwdx = 0, ∀w, w(3) = 0 ⇒
dx 0
dx
N 1
dx dx 1
0.1
3 3
du dw
10∫ dx = − w(1) + 2 ∫ xwdx, ∀w, w(3) = 0
1
dx dx 1

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
37
 Example problem 2
3 3
du dw
∫1 dx dx dx = − w(1) + 2∫1 xwdx, ∀w, w(3) = 0
10

• For the problem discussed earlier, consider a trial

(candidate) solution of the form u( x) = α + α ( x − 3) and a 0 1

weight function of the same form, w( x) = β + β ( x − 3) . 0 1

Obtain a solution to the weak form shown above.

– Check if the equilibrium equation in the strong form is
satisfied?
– Check if the natural boundary condition is satisfied?

dw
• Since w(3)=0, w( x) = β 0 + β1 ( x − 3) ⇒ w( x) = β1 ( x − 3) and
dx
= β1

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
38
 Example problem 2
3 3
du dw
∫1 dx dx dx = −w(1) + 2∫1 xwdx, ∀w, w(3) = 0
10

du dw
u ( x) = α 0 + α1 ( x − 3), = α1 , w( x) = β1 ( x − 3), = β1
dx dx
• Substitution in the weak form gives:
3 3
20
∫10α β dx = 2β
1
1 1 1 + 2∫ xβ1 ( x − 3)dx, ∀β1 ⇒
1
20α1β1 = 2 β1 −
3
β1 , ∀β1 ⇒
20 7
20α1 = 2 − ⇒ α1 = −
3 30

7
• So the solution is: u ( x) = 0.001 − ( x − 3) (here we used u (3) = 0.001)
30

• Is the strong form satisfied?

d du d 7
(10 ) + 2 x = − (10 ) + 2 x = 2 x ≠ 0, not satisfied
dx dx dx 30
du 7
• Is the natural BC satisfied? dx
(1) = − ≠ 0.1, not satisfied
30
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
39
 Example problem 3
• Find the weak form of the following BVP:
d 2u
k 2 − λu + 2 x 2 = 0, 0 < x < 1, u (0) = 1, u (1) = −2, k , λ cons tan ts
dx

• Multiplying by w with w(0)=w(1)=0 and integration from 0 to

1 gives: 1 2
⎛ d u 2⎞
∫0 ⎜⎝ dx 2
k − λ u + 2 x ⎟wdx = 0, ∀w, w(0) = w(1) = 0 ⇒
⎠
1 1 1
du du du dw
k (1) w(1) − k (0) w(0) − k ∫ dx − λ ∫ uwdx + 2∫ x 2 wdx = 0, ∀w, w(0) = w(1) = 0 ⇒
dx dx 0
dx dx 0 0

1 1 1
du dw
k∫ dx + λ ∫ uwdx =2∫ x 2 wdx, ∀w, w(0) = w(1) = 0
0
dx dx 0 0

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
40
 Example problem 4
• Consider a bar subjected to linear
body force b(x)= Cx . The bar has a
constant cross-sectional area A and
Young’s modulus E. Assume
quadratic trial solution and weight
functions:
u(x)=α1+α2x+α3 x2
w(x)= β1+β2x+β3 x2
• For what value of αi is u(x) kinematically admissible?

• Using the weak form, set up the equations for αi and solve.

• Solve the problem using two 2-node elements of equal size like the
elements we used in the lecture on truss structures (direct method).
Approximate the external load at node 2 by integrating the body force
from x=L/4 to x=3L/4; Likewise compute the external at node 3 by
integrating the body force from x=3L/4 to x=L.

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
41
 Example problem 4
• On Γu : u (0) = 0 ⇒ α1 = 0
• The constraint on the weight
functions is then: w(0) = 0 ⇒ β1 = 0

du
u ( x) = α 2 x + α 3 x 2 , = α 2 + 2α 3 x
dx
dw
w( x) = β 2 x + β3 x 2 , = β 2 + 2β3 x
dx
• The weak for this problem is the following:
L L
du dw
∫0 dx dx dx − ∫0 wbdx − wAt |Γt = 0 ⇒
AE
L L

∫ ( α + α )( β + β ) = ∫ 2 3 ) xdx ⇒
( β + β 2
AE 2 2 3 x 2 2 3 x dx C x x
0 0
L L
AE
C 0∫ ( β α
2 2 + 2 β α
2 3 x + 2 β α
3 2 x + 4 β α
3 3 x 2
) dx = ∫ ( β 2 x 2
+ β 3 x 3
) dx ⇒
0
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
42
 Example problem 4
L
AE
C 0∫ ( β 2α 2 + 2β 2α 3 x + 2 β3α 2 x + 4β3α 3 x 2 ) dx
L
= ∫ ( β 2 x 2 + β3 x3 ) dx ∀β 2 , β3 ⇒
0
L L
AE
C ∫0
( β α
2 2 + (2 β α
2 3 + 2 β α
3 2 ) x + 4 β α
3 3 x 2
) dx = (
∫0 2β x 2
+ β 3 x 3
) dx ∀β2 , β3 ⇒
AE ⎛ L2 L3 ⎞ L3 L4
⎜ β 2α 2 L + (2 β 2α 3 + 2 β3α 2 ) + 4β3α 3 ⎟ = β 2 + β3 ∀β 2 , β3 ⇒
C ⎝ 2 3⎠ 3 4

⎧ AE L3 ⎫ ⎧ AE ⎛ L3 ⎞ L4 ⎫
β2 ⎨ ( α 2 L + α 3 L ) − ⎬ + β3 ⎨
2
⎜ α 2 L + 4α 3 ⎟ − ⎬ = 0 ∀β 2 , β3 ⇒
2

⎩C 3⎭ ⎩C ⎝ 3⎠ 4⎭

⎡ AEL AEL2 ⎤ ⎧ L3 ⎫ ⎧ 7CL2 ⎫

⎢ C C ⎥⎥ ⎧α 2 ⎫ ⎪⎪ 3 ⎪⎪ ⎧α 2 ⎫ ⎪⎪12 AE ⎪⎪ 7CL2 CL 2
⎢ 3 ⎨ ⎬=⎨ 4⎬⇒⎨ ⎬=⎨ ⎬⇒ u ( x) = x− x
⎢ AEL
2
4 AEL ⎥ ⎩α 3 ⎭ ⎪ L ⎪ ⎩α 3 ⎭ ⎪ −CL ⎪ 12 AE 4 AE
⎢⎣ C 3 C ⎥⎦ ⎪⎩ 4 ⎪⎭ ⎪⎩ 4 AE ⎪⎭

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
43
 Example problem 4
3 L /4
CL2
f 2 = ∫ Cxdx =
L /4
4
L
7CL2 f2 f3
f3 = ∫ Cxdx =
3 L /4
32

⎡ 1 −1 0 ⎤
2 AE ⎡ 1 −1⎤ (2) 2 AE ⎡ 1 −1⎤ 2 AE ⎢
K (1) = ⎢ ⎥ ,K = ⎢ ⎥ ,K = −1 2 −1⎥
L ⎣ −1 1 ⎦ L ⎣ −1 1 ⎦ L ⎢ ⎥
⎢⎣ 0 −1 1 ⎥⎦

Our final system of assembled equations is the following:

⎧ ⎫
⎪ r1 ⎪ ⎧ CL2 ⎫ ⎧15CL3 ⎫
⎡ 1 −1 0 ⎤ ⎧ 0 ⎫ ⎪ 2 ⎪ ⎪⎪ 4 ⎪⎪ ⎧u2 ⎫ ⎪⎪ 64 AE ⎪⎪
⎪ ⎪ ⎪ ⎪ ⎡ 2 −1⎤ ⎧u ⎫
2 AE ⎢
−1 2 −1⎥ ⎨u2 ⎬ = ⎨ ⇒ ⎨u ⎬ = ⎨
CL 2 AE
=⎨ 3 ⎬
2
⎬ ⇒ ⎢ ⎥ ⎨ ⎬ ⎬
L ⎢ ⎥ 4 ⎪ L ⎣ −1 1 ⎦ ⎩ 3 ⎭ ⎪ 7CL ⎪
u 2 ⎩ ⎭ ⎪ 11CL ⎪
⎢⎣ 0 −1 1 ⎥⎦ ⎪⎩u3 ⎪⎭ ⎪
3

⎪ 7CL2 ⎪ ⎪⎩ 32 ⎪⎭ ⎩⎪ 32 AE ⎭⎪
⎪ ⎪
⎩ 32 ⎭ 2 AE 2 AE 15CL3 15
r1 = (−u2 ) = (− ) = − CL2
L L 64 AE 32

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
44
 Discretized weak form with finite elements
• Consider the following BVP: −u’’(x) = f(x), 0 < x < 1
u(0) = u(1) = 0
• We multiply the equation −u’’ = f by an arbitrary function w ∈V

• Integrate over (0, 1) which gives: 1

−(u’’, w) = (f, w) [ We use the following notation: ( f , g ) ≡ ∫ fgdx ]

0

• We now integrate the left-hand side by parts using the fact that w(0) =
w(1) = 0 to get:
−(u’’, w) = −u’(1)w(1) + u’(0)w(0) + (u’, w’) = (u’, w’)

du dw
1 1
• We conclude that: (u ', w ') = ( f , w), ∀w ∈V or equivalently : ∫ dx = ∫ fwdx ∀w ∈V
0 dx dx 0

• We are next going to look for approximate solutions using piece-wise

linear C0 (continuous) functions.

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
45
Finite element discretization: Piecewise linear functions
• Construct a finite-dimensional subspace Vh ‫ ؿ‬V as follows:
• Define intervals Ij = (xj−1, xj) of length hj = xj − xj−1, j = 1, 2, . .
. , (M + 1) and set h = maxj hj.
• Let Vh be the set of functions v such that:
– v is linear on each subinterval Ij
– v is continuous on [0, 1] and
– v(0) = v(1) = 0.
• The quantity h = maxj hj is a measure of how fine the
partition is.

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
46
 Piecewise linear basis functions Nj(x), j = 1, 2, . . .,M

∈ h These are global basis functions each one

• A function v V has the defined for each global node
representation: N1
M
v( x) = ∑ v( xi ) N i ( x), 0 ≤ x ≤ 1
i =1
N
Nodal values
of v ( x )

where for i,j=1,…,M N2

Nj (xi) = δij

• The space Vh is a linear N3

space of dimension M with
basis {Ni}, i=1,..,M N i ( x) ∈ H 01 (0,1)
This is not much different from a Fourier expansion – the difference is that we
use piecewise linear functions instead of sin and cos functions
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
47
Linear system of equations: Stiffness and load
M

• We approximate the solution as uh ( x) = ∑ j =1

uh ( x j ) N j ( x), 0 ≤ x ≤ 1

with uh(xj) the nodal values of uh(x).

• Recall our weak formulation: (u ', w ') = ( f , w), ∀w ∈V
• It now takes the form:
Find uh : (uh' , wh' ) = ( f , wh ), ∀wh ∈V h
• The space Vh is spanned by M functions Ni, thus the weak
form takes the form:

Find uh : (uh' , Ni' ) = ( f , Ni ), i = 1, 2,..., M

M

• We now left with substituting in this equation: uh ( x) = ∑

j =1
uh ( x j ) N j ( x).
M
Find uh ( x j ) : ∑ ( N , N ) uh ( x j ) = ( f , N i ), i = 1, 2,..., M
' '

j =1 
 



j i

Kij dj Fi

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
48
 Properties of the stiffness matrix
M
Find uh : ∑ ( N , N ) uh ( x j ) = ( f , N i ), i = 1, 2,..., M
1 1
where : K ij = ∫ N N dx, Fi = ∫ f ( x) Ni dx
' ' ' '

j =1 
 


j i
0
i j
0
Kij dj Fi

• K is symmetric: Kij = Kji, ( N 'j , Ni' ) = ( Ni' , N 'j ), i, j = 1, 2,..., M

• K is sparse (i.e. only a few elements of K are
nonzero) N
x x 1 1 1 1
( N 'j , N 'j ) = ∫ ∫
j −1
j +1
+ = + , j = 1, 2,..., M Nj
j
dx dx
x j −1 h 2 x j h2 h j h j +1
j j +1

1 xj 1
( N , N ) = ( N , N ) = −∫
'
j
'
j −1
'
j −1
'
j dx = − , j = 2,..., M
x j −1 h 2 hj
j
( N i' , N 'j ) = 0, | i − j |> 1
• K is positive definite: Indeed for ∀η ∈ \ M we obtain:
M M M M M M
η .Kη = ∑∑ηi Kijη j = ∑∑ηi ( N , N )η j = (∑ηi N , ∑η j N 'j ) = (η ,η ) ≥ 0
i
' '
j i
'

i =1 j =1 i =1 j =1

 
i =1

j =1

η .Kη = 0, only for η = 0. η

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
49
 Properties of the stiffness matrix
• Since K is a positive definite matrix, we conclude that K is
non-singular (note that the boundary conditions are already
considered).
• It follows that the system Kd = F has a unique solution.
• For the particular case of hj = h = 1/(M+1), the system Kd= F
becomes:
⎡ 2 −1 0 0 . . . 0 ⎤ ⎧ d1 ⎫ ⎧ F1 ⎫
⎢ −1 2 −1 0 . . . 0 ⎥ ⎪ d 2 ⎪ ⎪ F2 ⎪
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎢ 0 −1 2 −1 0 . . 0 ⎥⎪ . ⎪ ⎪ . ⎪
⎢ ⎥⎪ ⎪ ⎪ ⎪
1⎢ . 0 −1 2 −1 . . . ⎥⎪ . ⎪ ⎪ . ⎪
⎨ ⎬=⎨ ⎬
h⎢ . . . . . . . . ⎥⎪ . ⎪ ⎪ . ⎪
⎢ ⎥⎪
⎢ . . . . . . . . ⎥ ⎪ . ⎪⎪ ⎪⎪ . ⎪⎪
⎢ −1 2 −1⎥ ⎪d M −1 ⎪ ⎪ FM −1 ⎪
⎢ ⎥
⎢⎣ 0 −1 2 ⎥⎦ ⎩⎪ d M ⎭⎪ ⎩⎪ FM ⎭⎪

MAE 4700 – FE Analysis for Mechanical & Aerospace Design

N. Zabaras (9/23/2009)
50
 Final solution
• Let us assume that
the solution
obtained with 3 uh ( xi ) N i ( x)
0.7 N 2 ( x)
0.9 N1 ( x)
nodes (excluding
the boundary nodes
with u=0) is the 0.2 N 3 ( x)
following:
uh ( x) = uh ( x1 ) N1 ( x) + uh ( x2 ) N 2 ( x) + uh ( x2 ) N3 ( x) =
0.9 N1 ( x) + 0.7 N 2 ( x) + 0.2 N 3 ( x)
uh ( x )

• The piece-wise
linear
representation of
the FE solution is
shown here.
MAE 4700 – FE Analysis for Mechanical & Aerospace Design
N. Zabaras (9/23/2009)
51