Lecture 2

Variational Principles
in
Computational Solid
Mechanics
Somenath Mukherjee Gangan Prathap

Scientist, Director,
Structural Technologies Division, National Institute of Science and
National Aerospace Laboratories (NAL), Information Resources (NISCAIR),
Bangalore, New Delhi,
Karnataka, India India

1
 Lecture 2
Variational Principles in Computational
Solid Mechanics
Chapters

1. Principle of virtual work. Reciprocal Theorem.

2. The Least Action Principle in dynamics.

(Hamilton’s principle and the Euler-Lagrange equation)

3. Weighted Residual (Galerkin’s) Method and the principle

of minimum potential energy.

4. Element formulations using variational principles.

2
 Lecture 2

Chapter 1

Principle of virtual work.

Reciprocal Theorem.

3
Archimedes was the first person to use the
concept of virtual work 4
 1.1 Virtual work (over virtual displacements)
What is virtual displacement?
Virtual displacement is an imaginary displacement of any system
in frozen (fixed) time and space, over and above the actual
displacement.
In particular, virtual displacement can be imposed upon an
actual displacement at equilibrium.

What is virtual work?

Virtual work is the work done by the forces (external and internal)
traveling through virtual displacements.

5
 1.2 Principle of Virtual work

For any arbitrary yet admissible set of virtual

displacements, superposed over the equilibrium
configuration, the net virtual work done by the
forces (external and internal) is zero.

Equilibrium Zero virtual work

δWext − δWint = 0 (1.1)

6
Archimedes was the first person to use the concept
of Virtual Work in his calculations for the Lever.

δ2
L1
δ1 L2

F1 F2
Equilibrium: Virtual Work Principle
F δ −F δ =0
11 2 2
δ δ
"Give me a place to Geometric Compatibility 1= 2
L L
stand on, and I will 1 2
Together
move the Earth."
F1 L1= F2 L2
7
 Example 1

Derive the equations of equilibrium for the bar system.

Suppose that actual orthogonal displacement components
of load point at equilibrium are u and v.

α1 α2 α3 α4 Actual strain (at equilibrium)

in the i-th bar is
u. cos α i v. sin α i
L2 εi = +
L3 L4 Li Li
L1
Actual stress (at equilibrium) in the
u i-th bar is
β u. cos α i v. sin α i
σ i = Ei ε i = E i +
P Li Li
v 8
α1 α2 α3 α4 Apply a virtual displacement
u along direction u.
L2
L3 L4
L1 By Principle of virtual work,

δWext = δWint
β
u P
4
δu.P cos β = σ iδε i .( Ai Li )
i =1
4
u. cos α i v. sin α i δu. cos α i
= Ei + ( Ai Li )
i =1 Li Li Li

( )
4
Ei Ai
P cos β = u cos 2 α i + v sin α i cos α i
i =1 Li 9
α1 α2 α3 α4 Apply a virtual displacement
v along direction v.
L2
L3 L4
L1 By Principle of virtual work,

δWext = δWint
v β
4
δv.P sin β = σ iδε i .( Ai Li )
i =1
P 4
u. cos α i v. sin α i δv. sin α i
= Ei + ( Ai Li )
i =1 Li Li Li

( )
4
Ei Ai
P sin β = u sin α i cos α i + v sin 2 α i
i =1 Li
10
 α1 α2 α3 α4

L2
L3 L4
L1

u
v β
P
The equation of equilibrium for the bar system is

( ) P cos β
4 4 u
Ei Ai Ei Ai
cos 2 α i (sin α i . cos α i )
i =1 Li i =1 Li =
( )
4 4
Ei Ai Ei Ai
(sin α i . cos α i ) sin 2 α i
i =1 Li i =1 Li v P sin β

Solving this equation gives displacements u and v, from which

the stresses in the bars can be determined. 11
 1.3 Conservative Systems
In a conservative system, the net work done by
the forces in any closed path is zero.

In a conservative system, the net work from one

equilibrium state to another is path independent.
2 1
dW = 0 dW = − dW (1.2)
1 2 2
2

1
1 2
dW = 0 dW is path independent
1
12
1.4 Symmetry in Linear Elastic Conservative Systems
In a linear elastic conservative system, the stiffness
and flexibility matrices are symmetric.

kij = k ji fij = f ji (1.3)

Example 2

Demonstrate the invariance of work done in a bar with

changing order of loading.

1 2
P Q
l
L
13
 1 2
P Q
l L

Case 1. First load point 1 by P and then load 2 by Q.

Work done by loads:
1 1 1 Pl Ql 1 QL
WPQ = P.u1,1 + P.u1, 2 + Q.u2,2 = P +P + Q
2 2 2 EA EA 2 EA
Case 2. First load point 2 by Q and then load 1 by P.
Work done by loads:
1 1 1 QL Pl 1 Pl
WQP = Q.u2,2 + Q.u2,1 + P.u1,1 = Q +Q + P
2 2 2 EA EA 2 EA
WPQ = WQP
This demonstrates the invariance of work done with changing order of loading.
u1, 2 l
f12 = =

Symmetry in flexibility
Q
u2,1
EA
l
f12 = f 21
f 21 = = 14
P EA
 1.5 Maxwell Betti’s Reciprocal Theorem
On a given structure (linear and conservative
system), the virtual work done by load system A
over virtual displacement for load system B is
equal to the virtual work done by load system B
over virtual displacement for load system A.

{δ B } {FA } = {δ A } {FB }
T T (1.4)
Proof {δ B }T {FA } = {δ B }T [ K ]{δ A } = {δ A }T [ K ]{δ B }
= {δ A }T {FB }
…because stiffness [K] is symmetric
(linear conservative system) 15
Example 3
Determine the support reactions of the portal
frame under the loading shown.
P
L/2 L/2

C
VA VB L
A B
MA
HA MB HB

16
 P
L/2 L/2

C {δ2} {F1} = {δ1} {F2}

T T

VA System 1 VB
HA A B y A, 2 .V A − yC , 2 .P = 0
MB HB
MA
yC,2 P. yC , 2
VA =
y A, 2

yA,2 System 2

17
 P
L/2 L/2

C {δ3} {F1} = {δ1} {F3}

T T

VA System 1 VB
HA A B x A,3 .H A − yC ,3 .P = 0
MB HB
MA

yC,3 P. yC ,3
HA =
x A, 3

System 3
xA,3
18
 P
L/2 L/2

C {δ4} {F1} = {δ1} {F4}

T T

VA System 1 VB
HA A B θ A, 4 .M A − yC , 4 .P = 0

MB HB
MA
yC,4 P. yC , 4
MA =
θ A, 4
System 4

A,4
19
 P
L/2 L/2 yC,2

C
System 1 System 2
VA VB yA,2
HA A B V A = VB =
P. yC , 2
y A, 2

MB HB
MA
yC,3 yC,4

System 3 System 4
P. yC , 4
P. yC ,3 M A = −M B =
H A = −H B = θ A, 4
xA,3 x A, 3
A,4
20
1.6 Consequence of the Reciprocal Theorem
- The unit load method
P 1
L L

∆ δ

BM1 BM2
M=Px x m=x x

System 1 System 2
By Reciprocal Theorem:
L
∆.1= δ.P
But δ.P=(slope2).(BM1) =
m( x )
dx M ( x)
0
EI

L L
M ( x).m( x) Px.x PL3
∆= dx = dx = (1.5)
0
EI 0
EI 3EI 21
 Lecture 2
Variational Principles
in
Computational Solid
Mechanics
Somenath Mukherjee Gangan Prathap

Scientist, Director,
Structural Technologies Division, National Institute of Science and
National Aerospace Laboratories (NAL), Information Resources (NISCAIR),
Bangalore, New Delhi,
Karnataka, India India

1
 Lecture 2

Chapter 2
The Least Action Principle in dynamics

(Hamilton’s principle
and the Euler-Lagrange equation)

2
“Nature always acts by the shortest paths”.

. - Pierre de Fermat

3
 The pioneers of Analytical Mechanics
&
Variational Calculus

Maupertuis
Fermat

Galileo Newton

Lagrange Hamilton
4
Euler
 2.1 The Least Action Principle
Early History of Variational Calculus
Archimedes (287-212 BC)
For the first time in history the principle of virtual work (from virtual displacements)
was discovered and employed by Archimedes to derive the Lever Rule.
F1L1 = F2 L2 [F1δ1 + F2 (−δ 2 ) = 0 δ1 / L1 = δ 2 / L2 ]

Pierre de Fermat (1601-1665 AD)

Discovered the Principle of Least time for optical path. Derived Snell’s Law of
Refraction (and reflection) of light rays using this principle.
B
“Nature always acts by the shortest paths”. dl
δ =0 µ1 sin θ1 = µ 2 sin θ 2
A
v
Pierre Louis Moreau de Maupertuis (1698-1759AD)
Defined a parameter called the Action: (Action=Path Integral of the momentum)
He postulated that Nature always behaves in a way so as to optimize (minimize) this
action. B B
(Euler presented the mathematical proof of this postulate.) I = mv.dl ; δ mv.dl = 0
A A

Johann Bernoulli (1667-1748)

Presented the Brachistochrone problem (finding the path of fastest descent between
two fixed points) in Acta Eruditorum and its solution is 1682. Solutions were provided
independently by
Isaac Newton, Jacob Bernoulli (elder brother of Johann) and L’ Hospital.
5
 Archimedes’ Lever
Equilibrium of any system under given forces
conforms to the vanishing of the Net Virtual work
done by these forces traveling over the
geometrically compatible virtual displacements: L1 2
1 L2
F1L1 = F2 L2 [F1δ1 + F2 (−δ 2 ) = 0 δ1 / L1 = δ 2 / L2 ]

F1 F2
Fermat’s Optical Path
From one fixed point to another, Light travels
along the path of least time joining these points.
Length a 2 + (d − x) 2 b2 + x2
T AB = = +
Speed v1 v2 A Medium 1
dT AB (x − d ) x
= +
dx v1 a 2 + ( d − x) 2 v 2 b 2 + x 2 a 1

For the optimized path of least time

dT AB ( d − x) x
=0 = Medium 2
dx 2
v1 a + (d − x) 2 2
v2 b + x 2
b
sin α 1 sin α 2
= sin α1 sin α 2 2
v1 v2 c
v1
=c
v2
d B
c =speed of light in vacuum µ1 sin α1 = µ 2 sin α 2 µ=
c x
v =speed of light in medium v
6
µ sin α = Consant
The Brachistochrone
The Path of Fastest Descent under gravity from one point to another.
• Given two fixed points A and B.
• What is the path for shortest time of fall from A to B?
2 R
A x
θ
2R
B
v= (2gy)
y
For the path for shortest time of fall from A to B

B
dl
δ =0
A
v θ-sin θ)
x=R(θ
sin α
= k (Consant )
θ)
y=R(1-cosθ
v
Speed : v = 2 gy The Brachistochrone is a Cycloid
between points A and B.
Radius of generating wheel R=0.25/(k*k*g)
7
 Hamilton’s Principle and the Euler-Lagrange Equations
Hamilton’s Principle of Least Action:
Nature determines the path of any particle from point 1 to point 2 in a way
so as to minimize the Action I
L=Lagrangian, T=Kinetic Energy, V=Potential Energy
2
I = L(q,q, t )dt L = T −V q = q (t )
1
∆I ( for virtual changes about any path) = δI + δ 2 I + ..

For the Action to be a minimum, any

virtual changes about the minimum point
should lead to a net increase in the Action,
i.e.
∆I ( for virtual changes about min) ≥ 0 .

∆I > 0 δI = 0 δ 2I > 0

i.e. The first variation of the Action ( I)

should vanish for small perturbations
(virtual displacements) about the minimum point. 8
 First variation for the Lagrangian L (for virtual displacement in frozen time is)
n
∂L ∂L subject to δqi (t = t1 ) = δqi (t = t 2 ) = 0
δL = δqi + δqi
i =1 ∂qi ∂q
2

Action I = L(qi ,qi , t ) dt L = T −V qi = qi (t ) : generalised co − ordinates

1
2 n 2
First variation for the Action for any δI = δL(q ,q , t )dt = ∂L ∂L
i i δqi + δqi dt
path 1-2 is 1 i =1 1 ∂qi ∂q
For the critical path, I=0
n 2 n 2 2 2
∂L ∂L d ∂L Path A
0 = δqi dt + δqi − δqi dt
i =1 1 ∂qi i =1 ∂q 1 1
dt ∂q q
2 2 1 Critical Path
n n
∂L d ∂L ∂L Path B
0 = − δqi dt + δqi
i =1 1 ∂qi dt ∂qi i =1 ∂qi 1
2
∂L t
δqi =0 δqi (t = t1 ) = δqi (t = t2 ) = 0 t1 t2
∂qi 1

d ∂L ∂L
Euler-Lagrange Equation (Conservative Systems) − =0
dt ∂q i ∂qi
9
Euler-Lagrange Equation (Non-Conservative Systems)

d ∂L ∂L
− = Qi
dt ∂q i ∂qi

Qi = Non − conservative generalised force

What is a generalized force?

Virtual work=(Virtual physical displacement) × Physical force

=(Virtual generalized displacement) × Generalized force

= (δqi ).Qi

10
 2.2 Applications of the Least Action Principle
Free Vibration of Single Degree of freedom systems

The Pendulum
Kinetic Energy: T = mx 2 = m(lθ )
1 1 2
Potential Energy: V = mgl (1 − cos θ )
2 2

d ∂L ∂L
Euler-Lagrange Equation (Conservative Systems) − == 0 L = T −V
dt ∂θ ∂θ
g
Equation of Motion ml 2θ + mgl. sin θ = 0 θ + . sin θ = 0
l

Linear Analysis (small oscillations)

g θ l
θ + θ =0 θ = A. sin(ωnt + ε )
l
Natural frequency and Time Period
g 1 g 1 l
ωn = (rad / sec) fn = ( Hz ) τ= = 2π (sec)
l 2π l fn g
11
 2.3 Some simple applications of the Least Action Principle
Free Vibration of Single Degree of freedom systems
Example 1 The Spring-Mass model

1 1 2
Kinetic Energy: T = mx 2 Potential Energy: V= kx
2 2
∂L d ∂L
Euler-Lagrange Equation (Conservative Systems) − =0 L = T −V
∂x dt ∂x

Equation of Motion (free vibration) mx + kx = 0 k

m x(t)
Solution (free vibration) x = A. sin(ωnt + ε )
Natural frequency and Time Period

k 1 k 1 m
ωn = (rad / sec) fn = ( Hz ) τ= = 2π (sec)
m 2π m fn k

12
 Free Vibration of Multi-degree of freedom systems

Example 2 The Spring-Mass model for a 2 DOF system

1 2 1
Kinetic Energy: mT=x
1 1 + m2 x2 2
2 2
d ∂L ∂L
1 1 − =0 L = T −V
Potential Energy: V = k1 x12 + k 2( x2 − x1 ) 2 dt ∂x1 ∂x1
2 2
Euler-Lagrange Equation (Conservative Systems) d ∂L − ∂L = 0
dt ∂x2 ∂x2
Equation of Motion (free vibration)
m1 0 x1 k1 + k 2 − k2 x1 0 k1 k2
+ =
0 m2 x2 − k2 k2 x2 0 m1 m2
x1 A1
Solution = . sin(ωnt + ε ) x2(t)
x2 A2 x1(t)

2 m1 0 A1 k1 + k 2 − k2 A1 0 k1 + k 2 − k2 2 m1 0
− ωn + = det − ωn =0
0 m2 A2 − k2 k2 A2 0 − k2 k2 0 m2
A1,1 A1, 2
Eigenvalues and Eigen-modes ω12 : {φ1} = ω2 2 : {φ2 } =
A2,1 A2, 2
13
 The Spring-Mass model for a 2 DOF system (Continued)
Orthogonality of the Eigen-modes (Normal modes)
{φ1} [ K ]{φ2 } = {φ2 }T [ K ]{φ1} = 0
T
{φ1}T [ K ]{φ1} = k11 {φ2 }T [ K ]{φ2 } = k22
{φ1}T [ M ]{φ2 } = {φ2 }T [ M ]{φ1} = 0 {φ1}T [ M ]{φ1} = m11 {φ2 }T [ M ]{φ2 } = m22
Natural Frequencies (rad/sec)
kii : generalized modal stiffness for mode i
k k 22
ω1 = 11 ω2 = mii:: generalized modal mass for mode i
m11 m22
x1 x1 F1 (t )
Estimation of dynamic response for [M ] + [K ] =
x2 x2 F2 (t )
x1 q1
Modal superposition = q1 (t ){φ1}+ q2 (t ){φ2 } = [{φ1}, {φ2 }] = [φ ]{q}
x2 q2
Pre-multiply equation of motion by modal matrix φ []
x1 x1 T F1 (t ) q1 q1 Q1 (t )
[φ ] [M ]
T
+ [φ ] [K ]
T
= [φ ] [φ ] [M ][φ ]
T
+ [φ ] [K ][φ ]
T
=
x2 x2 F2 (t ) q2 q2 Q2 (t )

m11 0 q1 k11 0 q1 Q1 (t ) mii qi + kii qi = Qi (t )

+ =
0 m22 q2 0 k 22 q2 Q2 (t ) F1 (t )
Qi (t ) = {φi }T
F2 (t )
qi(t)
kii mii
/\/\/\/\/\ Qi(t) 14
Modal Decoupling:
 The Spring-Mass model: equal stiffnesses and masses
m 0 x1 2k −k x1 0
+ =
Equation of Motion (free vibration): 0 m x2 −k k x2 0

x1 A1
Solution: = . sin(ωnt + ε )
x2 A2
2k −k 2 m 0
Characteristic Equation:
det − ωn =0
−k k 0 m

Natural frequencies (rad/sec) and Eigen-modes

k k
k 1
ω1 = (0.618) : {φ1 } = m m
m 1.618
k 1
ω 2 = (1.618) : {φ 2 } = x1(t) x2(t)
m − 0.618

Observation: The natural frequencies and mode shapes for this particular
system are related to the Golden Numbers :
(1+ 5)/2=1.618 and (1- 5)/2=-0.618 (!!!)
15
Stationary Property of the Rayleigh Quotient at the Normal Modes of Vibration
The Spring-Mass model: equal stiffnesses and masses
x1 X1 x1 X
Solution: = . sin(ωt ) = ω 1 . cos(ωt )
x2 X2 Velocity Vector : x2 X2

Conservation of Energy: KE max = PE max

1 1 1 1 2
mω 2 X 12 + mω 2 X 2 2 = kX 1 2 + k ( X 2 − X 1 ) 2 X2
2 2 2 2 −1 +1
k X1
ω2 = 2
Rayleigh Quotient m X2
+1
X1
k k

m m

x1(t) x2(t)
2
dω 2 X2 X2 X2
=0 : − −1 = 0 = 1.618, − 0.618
X2 X1 X1 X1
d
X1

Observation: The Rayleigh Quotient is extremum (stationary) at the normal modes.

16
 Example 3 A rigid car chassis as a 2 DOF system

a0 b0

CM α ∆2

∆1 hCM
k1 k2

Case 1: Point of reference for displacements : Center of mass, CM

Expressions for kinetic energy T, potential energy V and the Lagrangian L are

1 1
T= mhCM 2 + I CM α 2
2 2
1 1 1 1
V = k1∆12 + k 2 ∆ 2 2 = k1{a0α − hCM }2 + k 2 {b0α + hCM }2
2 2 2 2
1 1 1 1
L = T - V = mhCM 2 + I CM α 2 − k1{a0α − hCM }2 − k 2 {b0α + hCM }2
2 2 2 2
17
 Example 3 A rigid car chassis as a 2 DOF system (Continued)
a0 b0

CM α ∆2

∆1 hCM
k1 k2

Case 1: Point of reference for displacements : Center of mass, CM

d ∂L ∂L d ∂L ∂L
Lagrange’s Equations − =0 − =0
dt ∂h ∂h dt ∂α ∂α
Equations of free vibration in matrix form
m 0 hCM k1 + k 2 − (k1a0 − k 2b0 ) hCM 0
+ =
0 I CM α − (k1a0 − k 2b0 ) k1a0 2 + k 2b0 2 α 0

Note:
For displacements at the center of mass, the system equations are inertially decoupled.
But the system equations are in general stiffness coupled.
k b
If however, the spring stiffness values be in the ratio 1 = 0 k1a0 − k 2 b0 = 0
then the system is also stiffness decoupled. k 2 a0
18
 Example 3 A rigid car chassis as a 2 DOF system (Continued)

as e bs
α ∆2

∆1 S CM hs
k1 k2
Case 2: A special point of reference S for displacements so that

bs k1
= k1as − k 2bs = 0 (k1∆ )as − (k 2 ∆ )bs = 0
as k 2
This means that under an external force F acting the point S, one can have
pure translation (no rotation) of the rigid chassis without any rotation
(equal and opposite moments are induced by the springs about point S).
F
as bs

k1∆ k2∆ ∆

S CM
k1 k2 19
 Example 3 A rigid car chassis as a 2 DOF system (Continued)

as e bs
α ∆2

∆1 S CM hs
k1 k2
Case 2: A special point of moment balance S for displacements

T=
1
2
1
2
1
2
(
mhCM 2 + I CM α 2 = m hS + eα )2 + 12 I 2
CM α =
1 1
mhs 2 + mehsα + I Sα 2
2 2
I s = I CM + me 2
1 1 1 1
V= k1∆12 + k 2 ∆ 2 2 = k1{asα − hs }2 + k2 {bsα + hs }2
2 2 2 2
1 1 2 1 2 1
L = T - V = mhs + mehsα + I sα − k1{asα − hs } − k 2 {bsα + hs }2
2
2 2 2 2

20
 Example 3 A rigid car chassis as a 2 DOF system (Continued)

as e bs
α ∆2

∆1 S CM hs
k1 k2
Case 2: A special point of moment balance S for displacements
d ∂L ∂L d ∂L ∂L
− =0 & − =0
Lagrange’s Equations dt ∂h ∂h dt ∂α ∂α

bs k1
Apply the condition = k1as − k 2bs = 0
as k 2
Equations of free vibration in matrix form

m me hs k1 + k2 0 hs 0
+ =
me I CM α 0 k1as 2 + k 2bs 2 α 0
Note:
For displacements at the special point S for balance of moments from spring forces
the system equations are stiffness decoupled.
But the system equations are in general inertially coupled.
If CM and S coincide then system is both stiffness wise and inertially decoupled. 21
 2.4 Constrained Systems; Lagrange’s Multipliers
A system of ‘n’ degrees of freedom (DOFs) is holonomic if all its DOFs are mutually independent.

A system is non-holonomic when some of its ‘n’ degrees of freedom (DOFs) are dependent on the
others. This means that it has some ‘r’ constraints so that the actual number of its independent
degrees of freedom is (n-r).
t2
We need to find the stationary points of the functional
I = L(qi , qi , t )dt i = 1,2,...n
Subject to r constraints t1
t2
f j (qi ) = constant or f j (qi , qi , t )dt = constant j = 1,2...r
t1

Procedure:
Modify the Lagrangian r
L* = L + λj f j
j =1
t2
Equation of motion can be had from δI = 0 where I = L * ( q i , q i , t ) dt
t1
d ∂L ∂L r ∂f j
Modified Lagrange’s Equations: − + λj =0 or = Qi : Non − conservative force
dt ∂qi ∂qi j =1 ∂qi 22
 Example 4 The Atwood Machine
Find the expressions for accelerations of the two unequal masses suspended
from the machine by an in-extensible string.
T
Method 1: Using Newton’s Law
Light pulley
m1a = T − m1g .............(i )
T a m1 m2 a = m2 g − T ...........(ii )
(i ) + (ii )
a m2 m1g (m1 + m2 )a = (m2 − m1 ) g
x
y ( m2 − m1 )
a= g m2 ≥ m1
m1 ( m2 + m1 )
m2g
m2 x = −a ↑ y=a↓

Method 2: Using Lagrange’s Equations

1 1 1 1 1
KE = m1x 2 + m2 y 2 = m1x 2 + m2 (− x 2 ) = (m1 + m2 ) x 2
2 2 2 2 2
V = −m1gx − m2 gy = −m1 gx − m2 g {− x − (πR + )} = −(m1 − m2 ).gx + (constant)1
x + y + πR = = fixed length of string L = KE - V
∴ y = − x − (πR + ) = − x − constant 1
L= (m1 + m2 ) x 2 − [− (m1 − m2 ).gx + (constant)1 ]
and y = − x and y = −x 2
d ∂L ∂L
− = 0 → (m1 + m2 ) x − (m1 − m2 ) g = 0
dt ∂x ∂x
(m − m2 ) (m − m1 )
x= 1 g y = −x = 2 g
(m1 + m2 ) (m1 + m2 ) 23
 Example 4 The Atwood Machine (continued)
Method 3: Using Lagrange’s Equations with Lagrange’s
Multipliers
Light pulley
Constraint Equation
x + y + πR = = fixed length of string
x f ( x, y ) = x + y = constant y = −x
y
m1 Lagrangian L
1 1
m2 KE = m1x 2 + m2 y 2 V = − m1gx − m2 gy
2 2 One equation from
1 2 1 2 constraint
L = KE - V = m1x + m2 y + (m1gx + m2 gy )
Lagrange’s Equations 2 2
d ∂L ∂L ∂f Two more equations
− +λ =0 → m1 x − m1 g + λ .1 = 0
dt ∂x ∂x dx from Lagrange
Equations
d ∂L ∂L ∂f
− +λ =0 → m2 y − m2 g + λ .1 = 0
dt ∂y ∂y ∂y
(m − m ) (m2 − m1 )
Solving x= 1 2
g y = −x = g The Lagrange Multiplier is the
all three (m1 + m2 ) (m1 + m2 )
constraint force,
equations: λ = m1 ( g − x) = T viz. tension T is string 24
 z
Example 5 A simple constrained minimal problem y
y=2x
Minimize the function
z (x, y) = x2 + y2 - 2xy - 4x x
Subject to the constraint
-4
y – 2x = 0
Using Direct Using Lagrange’s Using Penalty Method
Substitution Multiplier (This type is used to
describe many FE
formulations)
y = 2x z1 = x 2 + y 2 − 2 xy − 4 x + λ ( y − 2 x) z 2 = x 2 + y 2 − 2 xy − 4 x + κ ( y − 2 x) 2

∴ z = x 2 + y 2 − 2 xy − 4 x = x 2 − 4 x At stationary point (minima ) At stationary point (minima )

∂z1 ∂z 2
∂z = 2 x − 2 y − 4 − 2λ = 0 = 2 x − 2 y − 4 − 4κ ( y − 2 x) = 0
= 2x − 4 ∂x ∂x
∂x ∂z 2
∂z1 = −2 x + 2 y + 2κ ( y − 2 x) = 0
At stationary point = −2 x + 2 y + λ = 0 ∂y
∂y
∂z Solving ;
=0 ( xm , ym ) = (2,4) ∂z1
∂x = −2 x + y = 0 1 1
∂λ xm = 2 1 + , ym = 2 2 +
κ κ
zmin = xm 2 − 4 xm = −4 Solving ;
As κ → ∞,
xm = 2 , ym = 4 ( z1 ) m = −4
x m → 2, y m → 4 ( z 2 ) m → −4

25
 Lecture 2
Variational Principles
in
Computational Solid
Mechanics
Somenath Mukherjee Gangan Prathap

Scientist, Director,
Structural Technologies Division, National Institute of Science and
National Aerospace Laboratories (NAL), Information Resources (NISCAIR),
Bangalore, New Delhi,
Karnataka, India India

1
 Lecture 2

Chapter 3

Weighted Residual (Galerkin’s) method and

the principle of minimum potential energy.

2
Lord Rayleigh James Clerk Maxwell

3
 3.1 The self adjoint operator
Consider a differential equation: (u) = f
Force (kinetic)
Displacement (kinematic) Γ f Specified
Specified Boundary Boundary
Γd

Use a trial function v as virtual displacement

Ω Domain

v. (u)..dΩ = u . (v).dΩ + {d (v) f (u ) − d (u ) f * (v)}dΓ

Ω Ω Γ =Γd + Γ f

If v. (u)..dΩ = u . (v).dΩ is self − adjoint operator (3.1)

Ω Ω
This is the consequence of the Reciprocal Theorem (valid for conservative
systems only).
A conservative system is characterized by a differential equation of self- adjoint
operator. A conservative system has a symmetric stiffness matrix
(or compliance matrix). The energy of a conservative system is independent 4 of
changing order of loads.
 Example 1. Show that the linear differential operator for the axially
loaded bar is self-adjoint.
x,u(x)
d du L
− EA =q d d
=− EA q
dx dx dx dx
L x=L L
d du du dv du
v. (u)..dΩ = v − EA dx = v − EA + EA dx
Ω 0
dx dx dx 0
dx dx
x =0
x=L L x= L x=L L
du du dv du dv d dv
= v − EA + EA dx = v − EA + EA .u + u − EA dx
dx x =0 0
dx dx dx x =0 dx x =0 0
dx dx
L x=L x=L
d dv du dv
= u − EA dx + v − EA − u − EA
0
dx dx dx x =0 dx x =0 0

v. (u)..dΩ = u . (v).dΩ + {vf (u ) − uf * (v)}Γ=Γd +Γ f

Ω Ω
L L
d du d dv
v − EA dx = u − EA dx Reciprocal Theorem
0
dx dx 0
dx dx 5
 3.2 Weighted Residual Method
for a system of differential equation of equilibrium with
self-adjoint operator
δu h .f .dΩ
Consider a differential equation: (u) = f
Ω
Use an admissible trial function δuh as virtual displacement, and
u is the analytical solution
0
δu .{ (u) - f }.dΩ = a(δu , u ) − (δu , f ) − δu .R B
h h h
[ h
]
Ω
B is the domain boundary. Boundary terms vanish when either
δuh =0 (specified kinematic boundary condition) or reaction R=0
at boundary B.
0 = a(δu h , u ) − (δu h , f )
h h (3.2)
a(δu , u ) = (δu , f )
For a self-adjoint operator, a(u,v) is symmetric bilinear functional

δu h . (u)..dΩ = u . ( δu h ).dΩ a(δu h , u ) = a(u, δu h ) (3.3)

6
Ω Ω
 Case 1
If the trial function is a variation of the exact displacement
δuh = δu
a (δu , u ) = (δu , f )
1
δ a (u , u ) = δW
2
1
Here a (u, u ) is the analytical strain energy U of the system
2

and δW is the virtual work done by the external forces.

If we define a system potential energy (for a conservative system)
1
Π = a(u, u ) − (u, f ) = U − W
2 (3.4)
Then equilibrium condition : =0
Principle of stationary potential energy 7
 Case 2
If the exact function (of differential equation) is replaced by
N N
the trial function h h
u ( x ) = aiφi ( x ) δu ( x) = (δai ).φi ( x)
i =1 i =1

Thus (u h ) - f = e e=residual
Weighted residual statement is
{ }
N
h h δai φ j . aiφi ( x) - f .dΩ = 0
δu . (u ) - f .dΩ = 0 j = 1,2...N
Ω Ω i =1
(3.5)
a(δu h , u h ) = (δu h , f ) Galerkin’s Method
1
δ a(u h , u h ) = δW Rayleigh-Ritz Method
2

If we define a system potential energy (for a conservative system)

1
Π = a(u h , u h ) − (u h , f ) = U − W
2 (3.6)
Then equilibrium condition : =0 8
 Rayleigh Ritz Method (RRM) or Galerkin’s Method (GM);
Which should we use?
For conservative systems, the equations of equilibrium (with any
given approximate function for displacement) determined by both
Rayleigh- Ritz and Galerkin’s Methods are identical.
For non-conservative systems, we cannot adopt the Rayleigh-Ritz Method
(RRM), since the reciprocal theorem (symmetry; equation (3.1)) does not
hold, and consequently we cannot define the potential energy functional
for the system.

Such non-conservative systems are described by differential equations with

Non-self-adjoint operators.

However, the virtual work principle remains valid even for non-conservative
systems. Thus Galerkin’s method (GM) of weighted residuals can be applied
even to non-conservative systems to establish the finite element equations
(force-displacement relation) of equilibrium from the corresponding
differential equations of non-self-adjoint differential operators .
Message: Use RRM or GM for conservative systems
Use only GM for non-conservative systems 9
 Example 2. Weighted residual statements for the axially loaded bar.
x,u
d du L
− EA =q d d
=− EA q
dx dx dx dx
L L
d du dδu h
δu .{ (u) - q}.dΩ = δu
h h
− EA − q dx h
a (δu , u ) = EA
du
dx
Ω 0
dx dx dx dx
0
x=L L L
du dδu h du L
0 = δu h
− EA + EA dx − δu h q.dx 1 a (u , u ) = 1 du EA du dx
dx dx dx
0 x =0 0 0 2 2 0 dx dx
0 = a (δu h , u ) − (δu h , q ) L
1 h h 1 du h du h
a (u , u ) = EA dx
Weighted residual statement is 2 2 0 dx dx
N
δai φ j . aiφi ( x) - q .dΩ = 0 j = 1,2...N
Ω i =1 Galerkin’s Method
=0 Rayleigh-Ritz Method
L L
1 h h h 1 du h du h
Π = a (u , u ) − (u , q) = EA dx − u h q.dx Π = U −W
2 2 0 dx dx 0 10
3.3 The principle of minimum potential energy and
equations of equilibrium
For a conservative (self-adjoint) system, we can identify
a potential energy functional , so that at equilibrium,
any admissible virtual displacement leads to
=0

=0

Virtual work Equations of Equilibrium

11
Incompatible elements Compatible elements
in harmony not in harmony
(equilibrium)
12
 The principle of minimum potential energy
(for searching equilibrium)

“Of all the possible admissible displacements (satisfying

the kinematic boundary conditions) the one that
corresponds to equilibrium (of forces) is the one that makes
the potential energy stationary.”

(i.e. any virtual displacement about equilibrium (of forces)

brings about a vanishing first variation of potential
Energy)
=0 for u (3.7)

A stable equilibrium makes the potential energy a minimum

13
 The principle of stationary complementary energy
(for searching compatibility)

“Of all the possible admissible forces (in equilibrium) the

one that corresponds to displacement compatibility is the
one that makes the complementary energy stationary.”

(i.e. any virtual change of forces in equilibrium brings about

a vanishing first variation of complementary energy).
*=0 for F (3.8)

Complementary energy and its variation is

*=U*-W *= U*- W(for F)

14
 3.4 Castigliano’s Theorems
The principle of minimum potential energy
(for searching equilibrium)
n
δΠ = δU − Piδui = 0
i=
∂U
Pi = (3.9)
∂ui
The principle of stationary complementary energy
(for searching compatibility) Stress

δΠ* = δU * − uiδPi = 0
U*
∂U * (3.10)
ui =
∂Pi
U
For a linear structure, U* = U Strain

Complementary energy/volume equals strain energy/volume 15

 Example 3. Derive the equations of equilibrium and boundary
conditions for the axially loaded bar with spring supported ends.
x,u
L
/\/\/\ q /\/\/\
k0 kL
x=0 x=L
Potential energy functional =U-W
L L
1 du du 1 2 1 2
Π= EA dx + k0u0 + k Lu L − uq.dx
2 0 dx dx 2 2 0
L L
du du
δΠ = δ .EA dx + + k0u0 .δu0 + k Lu L .δu L − δu.q.dx
0
dx dx 0
x= L L L
du d du
δΠ = EA .δu + k0u0 .δu0 + k Lu L .δu L − δu EA − δu.q.dx
dx x =0 0
dx dx 0
L
du du d du
δΠ = − EA + k0u0 δu0 + EA + k Lu L δu L + δu − EA − q dx
dx x =0 dx x= L 0
dx dx
16
 Equilibrium =0
L
du du d du
0= − EA + k0u0 δu0 + EA + k Lu L δu L + δu − EA − q dx
dx x =0 dx x= L 0
dx dx

For non-trivial arbitrary, yet admissible virtual displacements

d du Differential equation for axially
− EA −q =0
dx dx loaded bar
(3.11)
du
Either δu 0 = 0 or − EA + k 0u0 =0
dx x =0 Boundary conditions
du at the ends x=0 & x=L
Either δu L = 0 or EA + k LuL =0
dx x=L (3.12)

For specified kinematic Force boundary conditions with spring ends

boundary conditions
du du
− EA = 0, EA =0
Force boundary conditions with free ends dx x =0
dx =L
x17
 Example 4. Derive the equations of equilibrium and boundary
conditions for the Euler beam with spring supported ends.

r0 rL
q
x=0
L x=L
k0 kL

Potential energy functional =U-W

L 2 2 L
1 d 2w d 2w 1 2 1 2 1 dw 1 dw
Π= EI dx + k w
0 0 + k w
L L + r0 + rL − wq.dx
2 0 dx 2 dx 2 2 2 2 dx x =0 2 dx x= L 0
L L
d 2w d 2w
δΠ = δ 2
.EI 2
dx − δwq.dx
0 dx dx 0
dw dw dw dw
+ k0 w0 .δw0 + k L wL .δwL + r0 δ + rL δ
dx x =0 dx x =0 dx x=L dx x= L

18
 L
d2 d 2w
δΠ = δw. 2
EI 2
− q dx
0 dx dx

d 2w
dw dw d 2w dw dw
+ − EI 2 + r0 δ + EI + rL δ
dx dx
x =0
dx x =0 dx 2 dx
x= L
dx x =0

d d 2w d d 2w
+ EI 2 + k0 w δwx =0 + − EI 2 + k L w δwx = L
dx dx dx dx
x =0 x= L
For non-trivial, arbitrary, yet admissible virtual displacements,
Equilibrium =0. dw d 2w dw
Either δ =0 or − EI 2
+ r0 =0
dx x =0 dx dx
x =0
2 2
d d w d 2w
EI −q =0 Either δ
dw
=0 or EI + rL
dw
=0
dx 2 dx 2 dx x=L dx 2 dx
x=L

Differential equation d d 2w
Either δw0 = 0 or EI 2
+ k0 w =0
for the Euler beam dx dx x =0

d d 2w
(3.13) Either δw L = 0 or −
dx
EI 2
+ kLw =0
dx x=L

Boundary conditions at the ends x=0 & x=L (3.14)

19
 3.5 Constrained media problems
(use of penalty methods)
Elementary beam theory as constrained media problem
Principal assumption in Euler’s simple beam theory
(valid for thin beams):
Plane sections normal to the neutral axis before bending
remain normal to it even after bending.
This means that
(a) bending rotation is equal to the slope of the neutral axis.
(b) shear deformation zx is (assumed) zero.

Euler beam model Shear flexible beam model

x
dw/dx
dw/dx
z,w x
=dw/dx NA zx= - dw/dx 20
 Elementary beam theory as constrained media problem (continued)
Euler beam model 1
L
dθ
2 L
1
L
dw
2
δΠ = δ EI dx − qwdx + κ θ− =0
2 dx 2 dx
x 0
Equilibrium Equations
0 0

d 2θ dw
EI 2
−κ θ − =0 ....(i )
dx dx
dw/dx
dθ d 2 w
z,w x κ −
dx dx 2
=q i.e.
d
dx
κ θ−
dw
dx
=q ....(ii )

NA Combining (i ) & (ii )

=dw/dx d d 2θ
EI 2 − q = 0 ...(iii )
dx dx
The Euler beam has Boundary conditions at x = 0 & x=L
infinite shear rigidity dθ
Either = 0 or δθ = 0
EI
dx
But the practice of using a dw
Either κ θ− = 0 or δw = 0
large shear rigidity for dx
thin beams creates a As κ → ∞ θ →
dw
problem called Shear dx
Locking in shear-flexible d 4w
beam elements. Equation (iii ) reduces to EI 4
− q → 0 21
dx
 Lecture 2
Variational Principles
in
Computational Solid
Mechanics
Somenath Mukherjee Gangan Prathap

Scientist, Director,
Structural Technologies Division, National Institute of Science and
National Aerospace Laboratories (NAL), Information Resources (NISCAIR),
Bangalore, New Delhi,
Karnataka, India India

1
 Lecture 2

Chapter 4

Element formulations
using variational principles

2
 4.1 Applications of the Rayleigh-Ritz method
x
q
EI
L
Example 1. Determine the deflection curves and bending curvature
distributions of a simply supported uniform beam under u.d.l. q using
the Rayleigh-Ritz method.
Use the following admissible displacement functions for approximations,
satisfying the essential (kinematic) boundary conditions:

w(x=0)=0 & w(x=L)=0

(i ) w( x) = a.x( L − x)
πx
(ii ) w( x) = a. sin
L 3
 x
q
Le=L/2 Le=L/2
qL 2 qL2
12 EI 8 EI
Exact Curvature
d 2w Approximate
(i ) w( x) = a.x( L − x) ε =− 2
= 2a
Curvature=2a
dx
2 Areas under the
L 2 L
1 d w 2 qL3 curvature curves are
Π= EI dx − q.w.dx = 2a ( EIL) − a
20 dx 2 6 Equal. WHY?
0
2 L qL2 qL3
Aexact = × =
qL3 3 8EI 12 EI
Equilibrium : δΠ = 4a (δa).( EIL) − (δa). =0
6 qL2 qL3
2 2
Aapprox = L× =
qL qL 12 EI 12 EI
a= w= x( L − x)
24 EI 24 EI Aexact = Aapprox4
 x
q
Le=L/2 Le=L/2
qL2
Exact Curvature
8 EI

2 Areas under the

πx d 2w π πx
(ii ) w( x) = a. sin
L
ε =− 2
=a
L
sin
L
curvature curves
dx
L 2
2 L 4 2 L qL2 qL3
1 d w 2 π qL Aexact = × =
Π= EI dx − q.w.dx = a ( EIL ) − 2 a 3 8 EI 12 EI
20 dx 2 0 4 L4 π 2L
4qL4 π πx
4 Aapprox = sin dx
π qL π 5 EI L L
Equilibrium : δΠ = 0 2a(δa). ( EIL) − ( 2δa) =0 0
4L 4 π
qL3
Aapprox =
4qL4 4qL4 πx 12 EI
a= 5
w= 5
sin
π EI π EI L
Aexact = Aapprox
5
 2
x q 2
L 2
d w
ε = EI − 2
dx
0 dx
EI L 2 h
2
L h 2 d w
ε = EI − 2
dx
0 dx
Displacement function Deflection Error of the Energy of
and curvature (strain) at beam Energy the Error
center x=L/2
h h 2 h 2
(superscript signifies 2
ε −ε ε −ε
approximated function)
wh ( x) = a.x( L − x) qL4 q 2 L5 q 2 L5
(0.01041)
ε h = 2a EI (1440) EI (1440) EI
wh ( x) = a. sin(πx / L) qL4
(0.01307) 1
−
4 q 2 L5 1
−
4 q 2 L5
ε h = a.(π / L) 2 sin(πx / L) EI 240 π 6 EI 240 π 6 EI

Exact
qL4
w( x) =
q
{ }
x( L3 − 2 x 2 L + x 3 ) (0.01302)
24 EI EI 0 0
q
ε= x( L − x) 6
2 EI
 4.2 Element formulations using the principle of
minimum potential energy
4.2.1 The bar element (2 nodes, 2 DOF) N1 (ξ ) =
1− ξ
2
R Le
R
1 x 2 1
u1,F1 u2,F2 2
1
=-1 EAe =1 N 2 (ξ ) =
1+ ξ
2
1
Displacement function 2x
ξ= −1 1
(C0 continuity)
Le 2
2 u1 1− ξ 1+ ξ
h
u = N i (ξ ).ui =[N1 N2 ] = [ N ]{δ e } N1 (ξ ) = N 2 (ξ ) =
i =1 u2 2 2
[N]: Shape function matrix (4.1)

h u 2 − u1 1 1 u1
Strain ε = e
= − = [ B ]{δ e } (4.2)
L Le Le u2
[B]: Strain-displacement matrix 7
 Le
The bar element R1 R2
u1,F1 u2,F2
Stress =-1 EAe =1
σ h = EAeε h = [ D][ B]{δ e } [ D] = EAe
Strain Energy of element
x= L ξ =1
Ue =
1
2 x =0
εh { } {σ }dx ={δ
T h e T
}
1
2 ξ =−1
L 1
[ B]T [ D][ B] dξ {δ e } = {δ e }T [ K e ]{δ e }
2 2
ξ =1 e 1
L EA −1
where e
[K ] = T
[ B] [ D][ B] dξ = (4.3)
ξ =−1
2 L −1 1
Work done by external loading q(x)
L 1
e h T
e T T L
W = {u } q ( x).dx ={δ } [ N ] q(ξ ). dξ ={δ e }T {F e }
x =0 ξ = −1
2
1 1
e T L F1e e L
where {F } = [ N ] q (ξ ). dξ = e
, Fi = N i (ξ ).q(ξ ). dξ
2 F2 2
ξ = −1 ξ = −1
(4.4)
8
 Le
The bar element R1 R2
u1,F1 u2,F2
Potential Energy of element is =-1 EAe =1
Πe = U e −W e { R h ,e } =
R1
R2
1
= {δ e }T [ K e ]{δ e } − {δ e }T {F e } − {δ e }T {R h,e } Reaction vector
2
e
Equilibrium of element demands that δΠ = 0
[ ] [ ] [
0 = δ {δ e }T [ K e ]{δ e } − δ {δ e }T {F e } − δ {δ e }T {R h,e } ]
[ K e ]{δ e } = {F e } + {R h,e } (4.5)
ξ =1
e T L EAe 1 − 1
[K ] = [ B ] [ D ][ B ] dξ =
Element Stiffness ξ = −1
2 L −1 1
1
e L
Generalized force vector Fi = N i (ξ ).q (ξ ). dξ
ξ = −1
2 9
4.2.2 The Euler beam element (2 nodes, 4 DOF)
w1,F1 w2,F2
Le
x EI
1,M1 2,M2

=-1 2x =1
ξ= e
−1
L
Displacement function (C1 continuity) w1
4 θ1
w = h
N i (ξ ).δ i =[N1 N2 N3 N4 ] = [ N ]{δ e } (4.6)
i =1 w2
θ2
1
{
N1 = 2 + ξ (ξ 2 − 3)
4
} Le
N 2 = (ξ + 1)(ξ − 1)2
8
1
{ }
N 3 = 2 − ξ (ξ 2 − 3)
4
Le
N 2 = (ξ + 1)2 (ξ − 1)
8

[N]: Shape function matrix

10
 The Euler beam element
Strain and stress resultant vectors [B]: Strain-displacement matrix

h
ε =
d 2 wh
=
4 d 2 wh
=
1
[6ξ Le (3ξ − 1) − 6ξ ]
Le (3ξ + 1) {δ e } = [ B]{δ e }
dx 2
(L )
e 2 dξ 2
(L )e 2

σ h = M h = EI eε h = [ D][ B]{δ e } [ D] = EI e (4.7)

Strain Energy of element
x= L ξ =1
Ue =
1
2 x =0
εh { } {σ }dx ={δ
T h e T
}
1
2 ξ =−1
L 1
[ B]T [ D][ B] dξ {δ e } = {δ e }T [ K e ]{δ e }
2 2
ξ =1 Stiffness
e L T
where [K ] = [ B] [ D][ B] dξ matrix
ξ = −1
2
3 2 3 2
12 ( EI e / Le ) 6 ( EI e / Le ) − 12 ( EI e / Le ) 6 ( EI e / Le )
2 2
[K ] =
e 6 ( EI e / Le ) 4 ( EI e / Le ) − 6 ( EI e / Le ) 2 ( EI e / Le )
− 12 ( EI e / Le )
3
− 6 ( EI e / Le )
2 3
12 ( EI e / Le ) − 6 ( EI e / Le )
2
(4.8)
2 2
6 ( EI e / Le ) 2 ( EI e / Le ) − 6 ( EI e / Le ) 4 ( EI e / Le ) 11
 The Euler beam element
Work done by external loading q(x)
L 1
e h T
e T L T
W = {w } q( x).dx ={δ } [ N ] q(ξ ). dξ ={δ e }T {F e }
x =0 ξ =−1
2
F1
1
e T L M1
where {F } = [ N ] q(ξ ). dξ =
ξ =−1
2 F2 (4.9)
M2
Potential Energy of element is
R1
Πe = U e −W e R2
{ R h ,e } =
1 e T e R3
= {δ } [ K ]{δ e } − {δ e }T {F e } − {δ e }T {R h,e }
2 R4
e
Equilibrium of element demands that δΠ = 0 Reaction vector

[ K e ]{δ e } = {F e } + {R h,e } (4.10)

12
 4.3 The general methods of element formulation
The Rayleigh Ritz Method The Galerkin Method
Steps
1. Use a displacement interpolation {u h } = [ N ]{δ e }
2. Develop strain and stress resultants.
Differential eqn.
ε h = [ B]{δ e }
σ h = [ D]ε h = [ D][ B]{δ e }
(u) = q

3. Generate potential function Virtual work

Πe = U e −W e
1
= {δ e }T [ K e ]{δ e } − {δ e }T {F e } − {δ e }T {R h,e }
2 { ({u })- q}.dΩ = 0
[ N ]T . h

Ω
e
4. Use PMPE δΠ = 0 (4.12)
[ K e ] = [ B]T [ D][ B]dx = [ B]T [ D][ B] det[ J ]dξ [ K e ]{δ e } = {F e } + {R h,e }
e e

e T T
{F } = [ N ] q ( x).dx = [ N ] q(ξ ). det[ J ]dξ (4.11) 13
e e
 4.4 Some specimens from the Lagrangian family of C0 elements

Displacement approximation in an element through Lagrangian interpolation

n
h
u = N1 ( x)u1 + N 2 ( x)u2 + .....N n ( x)un = N i ( x).ui =[ N ]{δ e }
i =1
n n
∏ ( x j − x) ∏ ( x j − x)
( x2 − x)( x3 − x)...( xn − x) j =2 j ≠i
N1 ( x) = = n
N i ( x) = n
( x2 − x1 )( x3 − x1 )...( xn − x1 )
∏ ( x j − x1 ) ∏ ( x j − xi )
j =2 j ≠i
n
N i (ξ ) = 1
i =1
(4.13)
Such elements are of C0 continuity because across the element
boundary, only the function interpolated is continuous, but its
derivatives are not continuous.
n
N i (ξ ) = 1 Ensures rigid body displacement of element
i =1 14
 (a) For the two noded bar element, a linear Lagrangian is used for
displacement, hence a constant strain over the element results:

1− ξ
2x N1 (ξ ) =
Le ξ= e
−1 2
R1 x R2 L 1
u1,F1 u2,F2 2
1
=-1 EAe =1 N 2 (ξ ) =
1+ ξ
2
1
Displacement function
1 2
2 u1 1− ξ 1+ ξ
h
u = N i (ξ ).ui =[N1 N2 ] = [ N ]{δ e } N1 (ξ ) = N 2 (ξ ) =
i =1 u2 2 2
2
N i (ξ ) = 1 [N]: Shape function matrix (4.14)
i =1
u 2 − u1 1 1 u1
Strain h
ε = e
= − e e
= [ B ]{δ e } (4.15)
L L L u2
15
[B]: Strain-displacement matrix
(b) For the three noded bar element, a quadratic Lagrangian is used for
displacement, hence a linear strain over the element results:
Le
x Element Stiffness Matrix
=-1 =1
7 1 −8
1 3 =0 2 e
K ]=
1
1
T e
[ B] EA [ B ]
dx EAe
dξ = e 1 7 −8
2 ξ = −1 dξ 3L
− 8 − 8 16

Element force vector

2x 1/ 6
ξ= −1 1
dx
Le {F e Applied } = [ N ]T q. dξ = qLe 1 / 6
−1
dξ
2/3
(4.17)
q=force p.u.length

3
u1
1 1
uh = N i (ξ ).ui = [N1 N2 N 3 ] u2 = [ N ]{δ e } N1 = − ξ (1 − ξ ), N 2 = ξ (1 + ξ ), N 3 = (1 − ξ 2 )
i =1 2 2
u3
3 (4.16)
N i (ξ ) = 1
i =1

h du h du h dξ 2 du h 2 dN1 dN 2 dN 3 2 1 − 2ξ 2 1 + 2ξ 2
ε = = = e = e {δ e } = − , , ( −2ξ ) {δ e }
dx dξ dx L dξ L dξ dξ dξ e
L 2 e
L 2 L e

ε h = [ B]{δ e } 16
 Example 2. A robotic arm is rotating
at a constant angular velocity
=30 rad/sec. Determine the axial stress
distribution from centrifugal forces (using
two quadratic elements).

Assembly of stiffness matrices

7 −8 1 0 0 (1)
− 8 16 − 8 0 0 ( 2)
2
107 × 0.6
[ KG ] = [K e ] = 1 − 8 14 − 8 1 (3)
e =1 3 × 21
0 0 − 8 16 − 8 (4)
0 0 1 −8 7 (5)

Element stiffness matrices

Element force vectors
1
7 1 − 8 (1) e dx T
[K 1] =
10 7
× 0.6
1 7 − 8 (3) {F applied } = [ N ] q. dξ
3 × 21
−1
dξ
− 8 − 8 16 (2)
dx Le
7 1 − 8 (3) =
10 7
× 0. 6 dξ 2 2x
[K 2 ] = 1 7 − 8 (5) ξ= e
−1
3 × 21 L
− 8 − 8 16 (4) 17
 Computation of element force vectors
with averaged body force
e ρ ( x e mean )ω 2
f Average = lb/in3
g

( (1)
f Average = )0.2836 × (10.5) × 30 2
32.2 ×12
= 6.94

1 / 6 (1)
( )
{F 1applied } = f (1) Average Ae Le 1 / 6 (2)
2 / 3 (3)

( ( 2)
f Average = ) 0.2836 × (31.5) × 30 2
32.2 ×12
= 20.81

1 / 6 (3)
( )
{F 2 applied } = f ( 2) Average Ae Le 1 / 6 (5)
2 / 3 ( 4)

The global (applied) force vector after force assembly

{F G Applied } = [14.57 58.26 58.26 174.79 43.70]T

18
 Solving for displacements
(u1=0). Eliminating u1
16 −8 0 0 u2 58.26
107 × 0.6 − 8 14 − 8 1 u3 58.26
=
63 0 − 8 16 − 8 u4 174.79
0 1 − 8 7 u5 43.7
[
{δ G } = 10 −3 × 0 0.5735 1.0706 1.4147 1.5294]T ]
Element strains and stresses
2 1 − 2ξ 2 1 + 2ξ 2
ε h = [ B]{δ e } = e
− , e
, e
(−2ξ ) {δ e }
L 2 L 2 L
2 1 − 2ξ 2 1 + 2ξ 2
σ h = Eε h = 107 × e
− , e
, e
(−2ξ ) {δ e
L 2 L 2 L

2
ρω 2 L2 x
Exact σ = 1−
2g L

Question
Why does the stress obtained from
the present FE solution does not
19
coincide with the best-fit ?
 Question
Why does the stress obtained from the present
solution does not coincide with the best-fit ?

From the differential equation

d du
− EA = q (x )
dx dx

Exact stress (quadratic) is

du 1 1
σ =E = − e (q)dx = − e ( fAe )dx
dx A e A e
2
ρω 2 L2 x
Exact σ= 1−
With the consistent force vector, we 2g L
can get the FE stress as the linear best-fit
to the analytical stress in each element. This implies that the true body force f should
be linearly varying (and not constant average)
within the element
This is the consistent force vector ρω 2 x ρω 2 x e
e
for each element. f = q = fA = A
g g
Le
e T T
1
Le Le
{F applied } = [ N ] q( x).dx = [ N ] q(ξ ). dξ
−1
2 xe L
2
0
2 e e
1 3 2
ρω x(ξ ) e ρω [ x L + L (ξ + 1) / 2] e =-1 =0 =1
q(ξ ) = A = A 20
g g
 Question
Centrifugal body force, f Why does the stress obtained from the present
solution does not coincide with the best-fit ?

x From the differential equation

0
Exact d du
FE (averaged body force)
− EA = q (x )
dx dx
FE (consistent force
Axial with actual linearly Exact stress (quadratic) is
Stress varying body force)
du 1 1
σ =E = − e (q)dx = − e ( fAe )dx
dx A e A e
2
ρω 2 L2 x
Exact σ= 1−
With the consistent force vector, we 2g L
can get the FE stress as the linear best-fit
to the analytical stress in each element. This implies that the true body force f should
be linearly varying (and not constant average)
within the element
This is the consistent force vector ρω 2 x ρω 2 x e
e
for each element. f = q = fA = A
g g
Le
e T T
1
Le Le
{F applied } = [ N ] q( x).dx = [ N ] q(ξ ). dξ
−1
2 xe L
2
0
2 e e
1 3 2
ρω x(ξ ) e ρω [ x L + L (ξ + 1) / 2] e =-1 =0 =1
q(ξ ) = A = A 21
g g
 4.5 Some specimens from the Isoparametric family of elements
Displacement approximation and Geometry of element described by the
same set of Lagrangian interpolation functions
{u h } = [ N ]{δ e }
{x} = [ N ]{x e }
(4.18)
Advantages:
Curved/inclined outlines/surfaces of elements can be suitably modeled.

(a) The aforementioned two-noded bar element is a linear

isoparametric element. Le
x
1 2
=-1 EAe =1
2 u1 1− ξ 1+ ξ
h
u = N i (ξ ).ui =[N1 N2 ] = [ N ]{δ e } N1 (ξ ) = N 2 (ξ ) =
i =1 u2 2 2
2 x1
x= N i (ξ ).xi =[N1 N2 ] = [ N ]{x e } 22
i =1 x2
(b) The three-noded (quadratic) bar element with quadratic geometry
description
3
u1
uh = N i (ξ ).ui = [N1 N2 N3 ] u2 = [ N ]{δ e }
i =1 u3
1 1
N1 = − ξ (1 − ξ ), N 2 = ξ (1 + ξ ), N 3 = (1 − ξ 2 )
2 2
3
x1
x = N i (ξ ).xi = [N1 N 2 N 3 ] x2 = [ N ]{x e }
i =1 x3

h du h dξ du h (du h / dξ )
ε = = = = [ B]{δ e }
dx dx dξ dx / dξ (4.19)
Le
x
=-1 =1
1 3 =0 2

1 3 2 23
 (c) The four-noded (linear) quadrilataral element
y,v for plane stress problem
3 Node,i i i

4 1 -1 -1
2 +1 -1
3 +1 +1
2
1 x,u 4 -1 +1

Shape functions
1 1
N1 = (1 − ξ )(1 − η ), N2 = (1 + ξ )(1 − η )
4 4
1 1
N3 = (1 + ξ )(1 + η ), N 2 = (1 − ξ )(1 + η )
4 4

u1 Geometry
Displacement v1 4
u2 N i xi
x
uh N1 0 N2 0 N3 0 N4 0 v2 = i =1 = [ N ]{ X e }
= = [ N ]{δ e } y 4
vh 0 N1 0 N2 0 N3 0 N4 u3 N i yi
v3 i =1
u4
v4 (4.20) (4.21)
24
 (c) The four-noded (linear) quadrilataral element
3 for plane stress problem
Element stress
4
σ xh 1 µ 0 ε xh
E
{σ h } = σ y h = µ 1 0 ε y h = [ D][ B ]{δ e }
(1 − µ 2 )
2 τ xy h 0 0 (1 − µ ) / 2 γ xy h

1
u1
Element strain
h v1
∂u
∂x u2
ε xh ∂N1 / ∂x 0 ∂N 2 / ∂x 0 ∂N 3 / ∂x 0 ∂N 4 / ∂x 0
∂v h v2
{ε h } = ε y h = = 0 ∂N1 / ∂y 0 ∂N 2 / ∂y 0 ∂N 3 / ∂y 0 ∂N 4 / ∂y = [ B]{δ e }
∂y u
γ xy h h h ∂N1 / ∂y ∂N1 / ∂x ∂N 2 / ∂y ∂N 2 / ∂x ∂N 3 / ∂y ∂N 3 / ∂x ∂N 4 / ∂y ∂N 4 / ∂x 3
∂u ∂v v3
+
∂y ∂x u4
v4

Jacobian 4 4 (4.22)
(∂Ni / ∂ξ )xi (∂Ni / ∂ξ ) yi
∂ ( x, y ) ∂x / ∂ξ ∂y / ∂ξ i =1 i =1
[J ] = = =
∂ (ξ ,η ) ∂x / ∂η ∂y / ∂η 4 4
(∂Ni / ∂η )xi (∂Ni / ∂η ) yi
i =1 i =1 (4.23)
25
 (c) The four-noded (linear) quadrilataral element
for plane stress problem

y,v
3
4 ∂Ni / ∂x −1 ∂N i / ∂ξ
= [J ]
∂N i / ∂y ∂Ni / ∂η
2
1 x,u
Element stiffness matrix:

1 1
[K e ] = [ B]T [ D][ B].t.dxdy = [ B ]T [ D][ B].t det[ J ].dξdη (4.24)
e −1−1

t=element thickness

26
Example 3. Plane stress analysis of a plate with a hole using quad4 elements

x y xy
27
 4.6 Numerical Integration by Gauss Quadrature
If f( ) is a polynomial of degree (2n-1) or less, then it is exactly integrated
within the domain − 1 ≤ ξ ≤ 1 by the n-point Gaussian quadrature rule.

f (ξ ) = p2 n (ξ ) = a1 + a2ξ + a3ξ 3 ..... + a2nξ 2 n−1 −1 ≤ ξ ≤ 1

1 n
f (ξ )dξ = wi f (ξ i )
−1 i =1
(4.25)

28