FINITE ELEMENT METHOD

VARIATIONAL METHODS
Variational Principle
Is a scientific principle used in mathematical analysis that uses variations, which has small
changes in functions and functional, which develops general methods for finding functions which
extremize the value of variable quantities that depend upon those functions.

Functional:
A functional, in simple terms, is a function of functions. In many areas of applied mathematics,
functional in the integral form involving first or higher – order derivatives. More specifically a
functional is defined as an expression that takes on a particular value which is dependent on the
function used in the functional. A functional takes a function as input and returns a scalar number
as output.
In most of the engineering field, problems are formulated in terms of differential equations.
These differential equations are solved by using variational methods by formulating the differential
equation in variational form and then arrive at an approximate solution.
A system in solid/structural mechanics is defined as the physical structure, its support and
loads applied on it. The system is said to be conservative system if there is no loss of energy i.e.
the internal strain energy stored due the load is equal to the work done by the external forces
applied.
In Solid/Structural mechanics the most commonly used functional is that of Potential
Energy. Potential Energy P has two components one the Strain Energy Ue due to internal
forces/stresses and the other is the work done/potential We due to the applied loads.

Potential Energy of an Elastic Body:

Consider a three dimensional linear elastic body occupying volume ‘V’, with body forces and
surfaces subjected to traction force. Then the strain energy in the element due of volume ‘dv’ is
the area under stress strain diagram.
Thus dUe = ½  dv
Which is written in matrix form as dUe = ½{}T{} dv

1
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD
Then total strain energy in three dimensional body will be
Ue = ½ ∫v{}T{} dv
Strain energy can also be expressed in terms of stiffness and displacement as
Ue = ½ KQ2 Where K is stiffness and Q is displacement

In finite element method these expression is written in the matrix form as

Ue = ½ {Q}T[K]{Q}

Work Potential due to Externally Applied Forces

Work done due to physical load acting on the system can be expressed as
T T n T
We = ∫v {U} {b}dv + ∫s {U} {t}ds + ∑i=1 {U}iT{Pc} i+ ∑i=1nd{U}i {M} i

Work done can also be expressed in terms of displacement and force as We = QF

Where Q is displacement and F is applied Force
In finite element method these expression is written in the matrix form as We = {Q}T{F}
Thus the total potential  = Ue - We
 = ½ KQ2 - QF
In finite element method these expression is written in the matrix form as
 = ½ {Q}T[K]{Q} - {Q}T{F}

Principle of Minimum Potential Energy

Principle of Minimum Potential Energy states that out of all the possible displacements
configuration of a body which satisfy the kinematic loading and geometric boundary conditions,
the displacement which can satisfy the equilibrium condition is the one having minimum potential
energy
or
For all kinematically admissible configuration of a conservative system, those that satisfy the
equilibrium state extremize the total potential energy. When the extremum is a minimum, the
equilibrium state is stable.
or

2
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD
Among all admissible configuration of a conservative system satisfying the kinematic,
geometric loading and boundary conditions the system will be in equilibrium when the total
potential is at its minimum.
or
For Conservative system, off all the kinematically admissible displacement fields, those
corresponding to equilibrium extremize the total potential energy. If the extremum is minimum,
the equilibrium state is stable
Then by the Principle of minimum potential energy for the system to be equilibrium the total
potential should be at its minimum.
Mathematically Maximum is derived by differentiating the total potential with respect to the
displacement and equating to zero
i.e ∂/∂Q = 0 Since  = ½ {Q}T[K]{Q} - {Q}T{F}
Then by Principle of Minimum Potential Energy ∂/∂Q = 0
We get [K]{Q} - {F} = 0 or [K]{Q} = {F}

NUMERICAL EXAMPLES
Write the equation of equilibrium for the spring system shown in the figure using the principle of
minimum potential energy and solve for the displacements in the springs when the spring stiffness
are k1 = 30 N/mm, k2 = 20 N/mm, k3 = 100 N/mm and k4 = 80 N/mm and
k1 Forces are F1 = 45 N

and F2 = 300 N k3 1 F1
k2 q1
Let 1, 2, 3 and 4 be the displacements at the springs F2
2
q2
1, 2, 3 and 4 and q1 and q2 be the displacements at k4
the point 1and 2.
The total potential energy for the given system is
 = Ue - We
 = 1/2 k1q12 + 1/2 k2(q1 - q2)2 + 1/2 k3q22 + 1/2 k4q22 - F1q1 - F2q2
 = 1/2 k1q12 + 1/2 k2(q12 + q22 - 2q1q2) + 1/2 k3q22 + 1/2 k4q22 - F1q1 - F2q2
 = 1/2 k1q12 + 1/2 k2q12 + 1/2 k2q22 - k2q1q2 + 1/2 k3q22 + 1/2 k4q22 - F1q1 - F2q2
By the principle of minimum potential energy we have ∂/∂qi = 0 for i = 1 and 2. Thus we
have ∂/∂q1 = k1q1 + k2q1 - k2q2 - F1 = 0
∂/∂q2 = k2q2 - k2q1 + k3q2 + k4q2- F2 = 0

3
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD

or
(k1 + k2)q1 - k2q2 = F1
- k2q1 + (k2 + k3 + k4)q2 = F2 which can be written in the matrix form as

(k1 + k2) -k2 q1 = F1

-k2 (k2 + k3 + k4) q2 F2

Substituting the numerical values we have

(30+20)q1 - 20q2 = 45
-20q1 + (20 + 100 + 80)q2 = 300
50q1 - 20q2 = 45 or 5q1 - 2q2 = 45
-20q1 + 200q2 = 300 -2q1 + 20q2 = 30
Multiply the first equation by 10 and then add them we get
50q1 - 20q2 = 450
-2q1 + 20q2 = 30
-------------------------------
48q1 = 480

Thus q1 = 10 mm

Substituting q1 = 10 mm into the first or second equation

5x10 - 2q2 = 45 or

q2 = (50 - 45)/2 = 2.5mm

q2 = 2.5mm
Thus the solution for the displacement is q2 = 10 mm and q2 = 2.5 mm

4
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD

RAYLEIGH - RITZ METHOD

Rayleigh-Ritz Method is a classical variational method for determining approximate solution
for the unknown field variable of a given problem. This method was first developed by Lord
Rayleigh (1877) and later in the year 1908 Walter Ritz refined/modified for better results.

Since the behavior of the continuum is described by partial differential equation the need to
solve differential equations can be avoided by applying Rayleigh-Ritz method to a Functional P.
Solution determined by Rayleigh-Ritz method is nearly exact but becomes more accurate as more
degrees of freedom are incorporated.

This method consists of three steps

1) Approximate the solution by assuming the displacement fields
u = f(x) + ∑ciNi for i = 1,2,3…n Where Ni is the admissible function and ci is the undetermined
coefficient.
2) With the help of the assumed admissible function evaluate the expression for total potential.
3) By the virtue of Principle of Minimum Potential Energy which will yield necessary system of
equations which are solved for the unknown field variable coefficient ci which is substituted into
the assumed displacement function.

NUMERICAL EXAMPLES
Using Rayleigh-Ritz's method determine the displacement, Slope, Bending Moment and Shear
Force in a Cantilever beam subjected to uniformly distributed load of 0.5 N/m. Given L = 0.5m,
Breath = 0.03 m, Depth = 0.04 m and Young's Modulus = 200 GPa.
Given Data:
Length : L = 0.5 m
Breath : b = 0.03 m
Depth : d = 0.04 m
Uniformly distributed load: px = 0.5 N/m
Modulus of Elasticity: E = 200 GPa = 2x1011 N/m2
Solution: Moment of Inertia of the cross section I = bd3/12 = (0.03*0.043)/12 = 1.6x10-7 m4

5
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD
Assume the displacement as v(x) = a0 + a1x + a2x2 + a3x3 At x = 0, v = 0 then a0 = 0
At x = 0, dv/dx = 0 then a1 = 0
Then v(x) =a2x2 + a3x3 is the admissible function. Then dv/dx = 2a2x + 3a3x2
dv2/dx2 = 2a2 + 6a3x
Strain Energy Ue = 1/2 ∫L [ EI (dv2/dx2)2dx = 1
/2 EI ∫L(2a2 + 6a3x)2dx

Ue = 1/2EI ∫L (4a22 + 36a32x2 + 24a2a3x)dx = 2EI ∫L (a22 + 9a32x2 + 6a2a3x)dx

L
Ue = 2EI [ a22x + 9a32(x3/3) + 6a2a3(x2/2) ]0 = 2EI [ a22L + 3a32L3 + 3a2a3L2 ]

Thus Ue = 2EI L [ a22 + 3a2a3L + 3a32L2 ] ………. (1)

L
We = ∫L Pv(x)dx = ∫LP[a2x2+a3x3]dx = P[ a2(x3/3) + a3(x4/4)]0

We = P[a2(L3/3) + a3(L4/4)] = PL3(4a2 + 3a3L)/12 ………. (1)

 = Ue - We = 2EI L [ a22 + 3a2a3L + 3a32L2 ] - PL3(4a2 + 3a3L)/12

By the principle of minimum potential energy we have ∂/∂ai = 0 for i = 2 and 3
∂/∂a2 = 2EIL [ 2a2 + 3a3L ] - PL3/3 = 0 or 2EIL [ 2a2 + 3a3L ] =
PL3/3
∂/∂a3 = 2EIL [ 3a2L + 6a3L2 ] - PL4/4 = 0 2EIL [ 3a2L + 6a3L2] =
Pl4/4
2a2 + 3a3L = PL2/6EI
a2 + 2a3L = PL2/24EI

Solving for a2 and a3 we get a2 = 5PL2/24EI and a3 = -PL/12EI

Substituting a2 and a3 into equation for the displacement i.e. v(x) = a2x2 + a3x3
v(x) = [(5PL2/24EI)x2 + (-PL/12EI)x3]

Thus the expression for the displacement is v(x) = PL(5Lx2 - 2x3)/24EI

Then the displacement at the fixed end where x = 0 we have v(x=0) = PL(5L*0 - 2*0)/24EI = 0
and at the free end x = L i.e. v(x=L) = PL4/8EI
Thus the displacement at the free end is v(x=L=0.5 m) = 0.5*0.54/(8*2*1011*1.6*10-7)
i.e. v(x=L=0.5 m) = 1.22x10-7m or 1.22x10-4 mm

6
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD

Slope = dv/dx = PL(10Lx - 6x2)/24EI

Slope at the fixed end where x = 0 we have (dv/dx)x=0 = PL(10L*0 - 6*0)/24EI = 0
and at the free end x = L i.e. (dv/dx)x=L = PL(10L2 - 6L2)/24EI = PL3/6EI
Thus the slope at the free end (dv/dx)x=L = 0.5*0.53/(6*2**1011*1.6*10-7) = 3.26*10-7rad
i.e. (dv/dx)x=L = 3.26*10-7rad
Bending Moment = Mb = EI(d2v/dx2) = EI (d2/d2x)[ PL(5Lx2 - 2x3)/24EI] = PL(10L -
12x)/24
Thus Bending Moment = Mb = EI(d2v/dx2) = PL(10L - 12x)/24
Thus the bending moment at the fixed end where x = 0 we have (Mb)x=0 = 5PL2/12
and at the free end x = L i.e. (Mb)x=L = PL(10L - 12L)/24 = -PL2/12
Thus the bending moment at the fixed end (Mb)x=0 = 5*0.5*0.52/12 = 1.67*10-8 N/m
i.e. (Mb)x=0 = 1.67*10-8 N/m
and at the free end (Mb)x=L = -0.5*0.52/12 = -1.04*10-2 N/m
i.e. (Mb)x=L = 1.04*10-2 N/m

Shear Force = V = EI(d3v/dx3) = EI (d3/d3x)[ PL(5Lx2 - 2x3)/24EI] = PL(- 12)/24 = -PL/2

Thus the shear force at both the fixed end and the free end will be
V = -PL/2 = -0.5*0.5/12 = -0.02 N
It is observed that the solution for the displacement and slope exactly matches with the analytical
solution from the basic mechanics of material. Where as the bending moment and shear force has
some error using the Rayleigh-Ritz's method because of the approximation of the admissible
function which was considered was satisfying only the displacement and slope continuity and the
bending moment and shear force conditions were not taken into account.

WEAK AND STRONG FORM:

The objective of most of the engineering analyses is to determine the functions, called
dependent variable which are governed by a set of differential equation valid with in a particular
domain  while satisfying the prescribed boundary conditions on the boundary 

7
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD
An integral expression such as Functional that implicitly contains the differential equations
is called the weak form. Weak form states conditions that must be met only in an average or
integral sense
A weak form of the system equations is usually created using one of the following widely
used variational methods:
Whatever functions that define these field variables have to be differentiable up to the order
of the partial differential equations that exist in the strong form of the system equations
Governing differential equations plus the boundary conditions are said to state a problem in
Strong Form. Strong form states that the conditions must be met at every material Point, whereas
the weak form states the condition in an average sense. Obtaining the exact solution for a strong
form of the system equation is usually very difficult for practical engineering problems.
Unknown Field Variables are of two types 1) Primary variable are the quantities which has to
be Continuous in the domain and 2) Secondary variables need not be Continuous
In solid or structural mechanics Bar problem, the primary variable is displacement ‘u’ which
has to be continuous. Otherwise, there will be opening or overlap in the member. The secondary
variable is strain i.e. du/dx. It can be discontinuous. In the stepped bar shown, stress is
discontinuous at change of cross section and hence strain also. Where as in case of beam problems
primary variables are the deflection and slope and the secondary variables are bending moment
and shear force.

METHOD OF WEIGTED RESIDUALS

Method of weighted residuals (MWR) is an approximate technique for solving boundary value
problems that utilizes the assumed trial functions that is satisfying the specified boundary
conditions and an integral formulation to minimize the error, in an average sense, over the domain
of the problem under study.
Given the governing differential equation in the form
D[y(x), x] = 0 for a<x<b ….. (1)
subject to the boundary condition
y(a) = y(b) = 0 …… (2)
The method of weighted residuals requires an approximate solution in the form
n
y*(x) =∑i=1 ciNi(x) ……(3)

8
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD
where y*(x) is the approximate solution expressed in terms of the product of unknown constant
parameter ci to be determined and Ni the trial function.
The trial function should be an admissible function i.e. the trail function should be continuous
over the entire domain and satisfy the specified boundary conditions. The trial function should
mitigate the physics of the problem thus the trail function may not lead to exact solution. Thus
substituting the assumed trial function into the governing differential equation as given in equation
(1) a residual error will occur such that
R(x) = D[y*(x), x] ≠ 0 …..(4)
Where R(x) is the residual which is also a function of unknown parameter ci thus the method
of weighted residual requires that the unknown parameter ci should be evaluated such that
b
∫a wi(x) R(x) dx = 0 for i = 1,2,3,…n ……(5)

Equation (5) represents the integral sum of the weighted residual error over the entire domain
equal to zero. In which wi(x) represents ‘n’ arbitrary weighing functions thus on integrating we get
‘n’ algebraic equations in terms ci which can be solved for ‘n’ ci’s.
Several variations of method of weighted residual exist and the techniques depends on how
the weighting factors are selected. The most common methods adopted are point collocation, sub
domain collocation, least squares and Galerkin’s method. Among all these Galerkin’s method is
found to readily adoptable in finite element method.

Galerkin’s Method
In Galerkin’s method, weighting functions are chosen to be same as the trial function
i.e. wi(x) = Ni(x) for i = 1,2,3,…n ……(6)
Then the unknown parameters are determined by
b b
∫a wi(x) R(x) dx = ∫a Ni(x) R(x) dx = 0 for i = 1,2,3,…n

which results in ‘n’ algebraic equation in terms of unknown parameters ‘ci’s are solved.

Steps in Galerkin’s Method

1. Formulate the differential equation of equilibrium for the given problem.

2. Assume a trial function for the given problem, the function should be admissible function
which should satisfy the boundary conditions.
9
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD

3. Substitute the trial function into differential equation of equilibrium and calculate
residual 𝑅(𝑥).

4. Using the Galerkin’s formula, minimize the residual by multiplying with a weight
𝑤𝑖 (𝑥) and Integrate over the domain
𝑏 𝑏
∫𝑎 𝑤𝑖 (𝑥)𝑅 (𝑥)𝑑𝑥 = ∫𝑎 𝑁𝑖 (𝑥)𝑅 (𝑥)𝑑𝑥 = 0
Where 𝑤𝑖 (𝑥) are the weighting functions and 𝑁𝑖 (𝑥) is the trail functions.

5. This will results in 𝑛 algebraic equation from which unknown parameters can be
determined and then by substituting into the function the field variable and derived
quantity can be determined.

BOUNDARY CONDITIONS
Boundary conditions can be classified into two types based on variational approach
1. Essential Boundary Conditions
2. Forced/Natural/Non-Essential Boundary Conditions
Essential boundary conditions are those which are associated with the primary variable and
Natural/Non-Essential boundary conditions are those associated with secondary variable. For a
Bar, essential boundary conditions are associated with displacement ‘u’ (Prescribed Displacement)
and natural/nonessential boundary conditions are associated with derivative of the displacement
‘du/dx’ [Prescribed strain du/dx = P/(AE) for given A and E] at the boundaries. For Beam, the
unknown primary field variables (which has to be continuous) are transverse displacement ‘v’ and
slope or rotation displacement ‘dv/dx’ and the secondary field variables (Which can be
discontinuous) are bending moment and shear force. Hence essential boundary conditions are
associated with transverse displacement ‘v’ and slope or rotation displacement ‘dv/dx’ and
Natural/nonessential boundary conditions are associated with bending moments (d2v/dx2 = M/EI)
and shear force (d3v/dx3 = V/EI)
In Solid/Structural mechanics, Essential boundary conditions are also known as geometric
boundary conditions (Which are essential to be satisfied) which are also known Dirichlet boundary
conditions
10
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru
 FINITE ELEMENT METHOD
Non Essential boundary conditions are known as Force or dynamic boundary conditions
(Automatically satisfied in Functional approach) which are also known as Neumann boundary
conditions
BOUNDARY AND INITIAL VALUE PROBLEMS
A domain is a collection of points in space with the property that if a point ‘P’ is in the domain,
then all points adjacent to ‘P’ belong to the domain. This property implies that a domain consists
only of internal points. The boundary of a domain is the set of points such that in any neighborhood
of each of these points there are points that belong to the domain as well as points that do not.
The main objective of most analyses is to determine the functions, called dependent variable,
which satisfy a given set of differential equation in a given region or domain and satisfy some
boundary conditions on the boundary of the domain.
A differential equation is said to describe a Boundary Value Problem if the unknown
dependent field variable and possibly its derivatives are required to have/take on the specified
values on the boundary of the domain.
An Initial Value Problem is that in which unknown dependent field variable and possibly its
derivatives are specified initially (i.e. at time t = 0). Generally Initial Value Problems are time
dependent.
Given the governing differential equation in the form
D[cu(x), x] =  for a < x < b in the domain Ω≡ (a, b)
The set of specified values for the conditions (boundary or initial) are called data of the
problem. When the specified values of the conditions (boundary or initial) are non zero, then the
conditions are said to be non-homogeneous otherwise they are said to homogeneous.
Differential equations in which the right hand side ‘’ is zero then the differential equations
are called as homogeneous differential equation. The problem of determining the values of the
constant ‘l’ such that
D[cu(x), x] = u for 0 < x <  and y(0) = 0 and [C(dy/dx)]x=l = 0

is called the Eigen value problem associated with the differential equation. The value of ‘’ are
called Eigen values and the corresponding unknown field variable ‘u’ is known as Eigen vector.

11
Dr. Jagadish. T, Former Professor and Head, Department of Mechanical Engineering, BIT, Bengaluru
Presently Director, Research and Development, DHIO Research and Engineering Pvt. Ltd., Bengaluru