AN

INTRODUCTION
TO THE
FINITE
ELEMENT
METHOD
FOR
YOUNG
ENGINEERS
PART TWO: 2D BEAM FORMULATIONS
By: Eduardo DeSantiago, PhD, PE, SE
Table of Contents

SECTION I INTRODUCTION............................................................................................2

SECTION II 2-D EXAMPLE...............................................................................................2

SECTION III DISCUSSION...............................................................................................27

SECTION IV REFERENCES.............................................................................................28

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 1
I. Introduction

This article is the second in a series that will attempt to introduce some of the rich and complex
theory that forms the foundation of the finite element method of analysis (FEM). The focus for
this article is on beam formulations which in the author’s opinion constitute the vast majority of
FEM analysis conducted by practicing structural engineers. Although the current discussions
will be limited to 2D Bernoulli Euler beam formulations, several topics that play a role in more
advanced applications will be introduced. With a much clearer understanding of the limitations
and key assumptions underlying beam formulations, it is the author’s hope that practitioners
will become more sophisticated users of commercial FEM software in regards to beam and
frame analysis.

II. 2-D Example

To help introduce basic mathematical concepts we will limit current discussion to a two-
dimensional problem involving flexure of a beam under the action of a transverse distributed
load. For current discussions shear deformations will be neglected, i.e. a Bernoulli-Euler
formulation will be employed, and axial loading/deformation will not be included. Subsequent
articles will introduce the effects of employing shear deformations, i.e. Timoshenko beam
theory, and including axial forces/deformations before extending the formulations to 2D frames
and eventually 3D frames. The article will conclude with a numerical example to tie together
some of the key concepts introduced.

A. Model Problem and Sign Convention

For the model beam problem we will consider a beam spanning along the x direction from x=0
to x=L. The beam is acted upon by a transverse distributed load f(x) along its axis in addition to a
concentrated moment M0 and transverse point load Q at the left end. The loading results in a
transverse deflection in the y direction given by v(x) and a rotation of the beam cross section at
any point x given by derivative of v(x), i.e. dv/dx, due to the Bernoulli-Euler formulation in which
the cross sections will remain perpendicular to the neutral axis of the beam. The beam consists
of an isotropic linear elastic material with a Young’s Modulus of E and a geometric cross section
with a moment of inertia I. At this point the flexural stiffness of the beam given by EI will be
assumed to vary with location along the beam, i.e. EI=EI(x).

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 2
For all subsequent discussions we will adopt the following sign convention for the internal
shear force V(x) and internal bending moment M(x) that develops in the beam:

Additionally, the transverse loading f(x) and transverse deflection v(x) are both positive if acting
in the positive y direction. Concentrated loads are considered positive if acting in the positive y
direction while concentrated moments and rotations are positive if counterclockwise.

For a well-defined 2-D Bernoulli-Euler beam problem with a unique solution, a total of four
boundary conditions are required; two at each end. For the model problem the two boundary
conditions at x=0 are called displacement type boundary conditions and are given by:

𝑣𝑣(0) = 𝑔𝑔
𝑑𝑑𝑑𝑑
= 𝛽𝛽
𝑑𝑑𝑑𝑑
in which g and b are prescribed constants.

At x=L, the two boundary conditions for the model problem are called traction or force type
boundary conditions and are given by:

𝑉𝑉(𝐿𝐿) = −𝑄𝑄
𝑀𝑀(𝐿𝐿) = 𝑀𝑀0

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 3
Commonly encountered boundary conditions for Bernoulli-Euler beams include:

• Fixed ends: v=0 and dv/dx=0, i.e. transverse displacements and rotations are precluded
• Simply supported ends: v=0 and M=0, transverse displacement is prevented and the
internal bending moment is zero
• Free ends: V=0, M=0, both the internal shear force and bending moments are zero

Enforcing equilibrium and the Bernoulli-Euler assumption that cross sections will remain
perpendicular to the neutral axis of the beam results in the following equations 1

𝑑𝑑𝑑𝑑(𝑥𝑥)
= 𝑓𝑓(𝑥𝑥) ∀ 𝑥𝑥 ∈ 0, 𝐿𝐿
𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑(𝑥𝑥)
= 𝑉𝑉(𝑥𝑥) ∀ 𝑥𝑥 ∈ 0, 𝐿𝐿
𝑑𝑑𝑑𝑑
𝑑𝑑2 𝑣𝑣(𝑥𝑥)
𝑀𝑀(𝑥𝑥) = 𝐸𝐸𝐸𝐸(𝑥𝑥) ∀ 𝑥𝑥 ∈ 0, 𝐿𝐿
𝑑𝑑𝑑𝑑 2
Combining the above equations results in the differential equation of equilibrium that must be
satisfied at every single point along the beam

𝑑𝑑 𝑑𝑑2 𝑣𝑣(𝑥𝑥)
�𝐸𝐸𝐸𝐸(𝑥𝑥) � = 𝑓𝑓(𝑥𝑥) ∀ 𝑥𝑥 ∈ 0, 𝐿𝐿
𝑑𝑑𝑑𝑑 2 𝑑𝑑𝑑𝑑 2

As discussed in Article 1 of the series, if a finite difference approach where to be undertaken to

approximate the solution to the model problem; approximations for the fourth order derivatives
present in the above equation would be required.

B. Weak (Variational) Form of Model Problem

Prior to introducing finite element approximations for solving the model problem, the
differential equation of equilibrium along with the prescribed boundary conditions need to be
recast into a variational form. Derivation of the variational form (or weak form) will not be
presented but the interested reader can consult the references listed at the end of the article.
For the model problem, the variational form of the equation of equilibrium takes the form

1The mathematical symbol ∀ is shorthand notation denoting “for all and any” and ∈ is shorthand notation
for “element in”.

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 4
 𝐿𝐿
𝑑𝑑 2 𝜔𝜔(𝑥𝑥) 𝑑𝑑2 𝑣𝑣(𝑥𝑥) 𝑑𝑑𝑑𝑑(𝑥𝑥)
� 𝐸𝐸𝐸𝐸(𝑥𝑥) 𝑑𝑑𝑑𝑑 = 𝑀𝑀0 � + 𝑄𝑄𝑄𝑄(𝐿𝐿)
0 𝑑𝑑𝑥𝑥 2 𝑑𝑑𝑑𝑑 2 𝑑𝑑𝑑𝑑 𝑥𝑥=𝐿𝐿
𝐿𝐿
+ � 𝑓𝑓(𝑥𝑥)𝜔𝜔(𝑥𝑥)𝑑𝑑𝑑𝑑 ∀ 𝜔𝜔(𝑥𝑥) ∈ 𝑊𝑊
0

To complete the variational form of the model problem, a definition for the space of admissible
transverse displacements v(x) and admissible weighting functions (x) must be defined, T and W
respectively.

𝑑𝑑𝑑𝑑(𝑥𝑥)
𝑇𝑇 = �𝑣𝑣(𝑥𝑥)|𝑣𝑣: (0, 𝐿𝐿) → ℝ, 𝑣𝑣(𝑥𝑥) ∈ 𝐶𝐶𝑏𝑏1 (0, 𝐿𝐿), 𝑣𝑣(0) = 𝑔𝑔, � = 𝛽𝛽�
𝑑𝑑𝑑𝑑 𝑥𝑥=0

𝑑𝑑𝑑𝑑(𝑥𝑥)
𝑊𝑊 = �𝑤𝑤(𝑥𝑥)|𝑤𝑤: (0, 𝐿𝐿) → ℝ, 𝑤𝑤(𝑥𝑥) ∈ 𝐶𝐶𝑏𝑏1 (0, 𝐿𝐿), 𝑤𝑤(𝑥𝑥) = 0, � = 0�
𝑑𝑑𝑑𝑑 𝑥𝑥=0

Before proceeding any further an explanation of the definitions for both the space of admissible
displacements, T, and weighting functions, W, would be beneficial. The term v:(0,L) → ℝ simply
states that the transverse displacement function v(x) accepts values of x between 0 and L and
returns a real number. That real number physically signifies the transverse displacement at
location x along the length of the beam. The term 𝑣𝑣(𝑥𝑥) ∈ 𝐶𝐶𝑏𝑏1 signifies that the transverse
displacements are to be continuous and bounded. Furthermore, the first derivatives of the
𝑑𝑑𝑑𝑑
transverse displacements, i.e. , are also required to be continuous and bounded. In general, a
𝑑𝑑𝑑𝑑
function f is said to be of class 𝐶𝐶𝑏𝑏𝑘𝑘 continuity if its first k derivatives (k=0,1,…) are continuous and
𝑑𝑑 𝑘𝑘 𝑓𝑓
bounded, i.e. � � < ∞. Because of the Bernoulli-Euler approximation, requiring the first
𝑑𝑑𝑥𝑥 𝑘𝑘
derivative to be continuous means that the beam will not experience sudden jumps in the
rotation of the cross sections along the axis of the beam. Finally, the last two terms in the
definition of T require that the admissible transverse displacement functions must meet the
displacement type boundary conditions a-priori. Similar descriptions apply to the definition of
the space of admissible weighting functions, W, with the exception that the weighting functions
are required to be equal to zero at any point where the transverse displacement is prescribed
and the derivative of the weighting functions must be zero at any points where the slope of the
beam is prescribed.

A few more observations on the variational form for the model problem:

• Consider the result of using linear piecewise continuous functions for both the
transverse displacements and the weighting functions as was done with the 1-D
model bar problem of article 1.

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 5
For the above piecewise linear approximations, the first and second derivatives of v(x) would
have the following form:

Here 𝛿𝛿(𝑥𝑥 − 𝑥𝑥𝑎𝑎 ) is the Dirac delta function given by

1
𝑥𝑥𝑎𝑎 − 𝜀𝜀 ≤ 𝑥𝑥 ≤ 𝑥𝑥𝑎𝑎 + 𝜀𝜀
𝛿𝛿(𝑥𝑥 − 𝑥𝑥𝑎𝑎 ) = lim � 2𝜀𝜀
𝜀𝜀→0
0 𝑥𝑥 < 𝑥𝑥𝑎𝑎 − 𝜀𝜀; 𝑥𝑥 > 𝑥𝑥𝑎𝑎 + 𝜀𝜀

Additionally, the Dirac delta function has the following property

+∞
� [𝛿𝛿(𝑥𝑥 − 𝑥𝑥𝑎𝑎 )]2 𝑑𝑑𝑑𝑑 = ∞
−∞

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 6
 𝑑𝑑𝑑𝑑 𝑑𝑑𝑑𝑑
Therefore if both and experience discontinuities (jumps) at the same location x then the
𝑑𝑑𝑑𝑑 𝑑𝑑𝑑𝑑
integral
𝐿𝐿
𝑑𝑑2 𝜔𝜔(𝑥𝑥) 𝑑𝑑2 𝑣𝑣(𝑥𝑥)
� 𝐸𝐸𝐸𝐸(𝑥𝑥) 𝑑𝑑𝑑𝑑
0 𝑑𝑑𝑥𝑥 2 𝑑𝑑𝑥𝑥 2

becomes unbounded and thus undefined. The above once again illustrates why functions that
are continuous and have continuous first derivatives will need to be employed for our
approximations.

• The weak form is just a restatement of the principle of virtual work where the internal
virtual work (neglecting shear and axial strain contributions) due to a virtual displacement
ω(x) is given by
𝐿𝐿
𝑑𝑑2 𝜔𝜔(𝑥𝑥) 𝑑𝑑2 𝑣𝑣(𝑥𝑥)
� 𝐸𝐸𝐸𝐸(𝑥𝑥) 𝑑𝑑𝑑𝑑
0 𝑑𝑑𝑥𝑥 2 𝑑𝑑𝑥𝑥 2

𝑑𝑑 2 𝑣𝑣(𝑥𝑥) 𝑑𝑑 2 𝜔𝜔(𝑥𝑥)
noting that 𝐸𝐸𝐸𝐸(𝑥𝑥) 𝑑𝑑𝑥𝑥 2
corresponds to the internal bending moment at location x and that
𝑑𝑑𝑑𝑑 2
is the virtual curvature of the beam at location x. The external virtual work is given by
𝐿𝐿
𝑑𝑑𝑑𝑑(𝑥𝑥)
𝑀𝑀0 � + 𝑄𝑄𝑄𝑄(𝑙𝑙) + � 𝑓𝑓(𝑥𝑥)𝜔𝜔(𝑥𝑥)𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑 𝑥𝑥=𝐿𝐿 0

where the first term is the virtual work done by applied moment M0, the second term is the
virtual work done by the applied transverse load Q, and the last term accounts for the virtual
work performed by the applied distributed transverse load f(x).

C. FEM Formulation
Using the above variational form of the model problem under consideration, a finite element
formulation can now be presented. The FEM formulation results from discretizing (breaking up)
the beam into n finite elements as shown below:

Figure 1: FEM discretization of model problem

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 7
The total length of the beam is broken into n, n=1, 2, 3, … non overlapping elements labeled with
an integer number enclosed in a box. At each element end there exists a node denoted by a
number enclosed by a circle. For future discussions we will take an element by element view of
the model problem. From the element point of view the right node is denoted by local node 1
and is enclosed in a triangle while the left node is denoted by local node 2 and is also enclosed
in a triangle. Thus, global node 2 is the same as local node 2 for element 1 and local node 1 for
element 2. At each node i we will have an unknown transverse displacement denoted by di along
𝑑𝑑𝑑𝑑
with a corresponding unknown rotation θi= � . Similarly, the weighting function (virtual
𝑑𝑑𝑑𝑑 𝑥𝑥=𝑥𝑥𝑖𝑖
transverse displacement) will have a value at node i which is denoted by ωi along with a
corresponding unknown derivative (i.e. virtual beam rotation) at node i which is denoted by
𝑑𝑑𝑑𝑑(𝑥𝑥𝑖𝑖 )
ϕi= � . Therefore at each node two degrees of freedom (dof) exist, a transverse
𝑑𝑑𝑑𝑑 𝑥𝑥=𝑥𝑥𝑖𝑖

displacement and a rotation. Since each element has two nodes, one at each end, each element
has a total of 4 dofs (2 per node).

Figure 2: Element Description

1. Notation
Before proceeding any further a definition of the notation that will be employed from this point
on will be useful. At times we will take a global view of the model problem, that is the entire set
of elements and nodes will be taken as a whole. In this point of view we will denote global nodes
by the labels enclosed in the circles and denote particular elements using the labels enclosed in
the boxes as shown on Figure 1. Global displacements at a particular node i will be denoted by
di and the corresponding rotation by θi. Similarly, the virtual global displacements at the same
node i will be denoted by ωi and the corresponding virtual rotation by ϕi.

Another important point of view to be used in the following discussion is within an element e.
Figure 2 shows the element point of view for the nodal displacements/rotations and the nodal
virtual displacements/rotations. When discussing quantities at the element level, a superscript
e will be employed to denote any element quantity associated with a general element e, while
the element integer label will be used as a superscript to denote quantities associated with that
particular element.

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 8
Finally, as will be seen shortly approximations to the transverse displacement function v(x) and
the virtual displacement function ω(x) will be constructed. The approximations will be tied
directly to the underlying finite element mesh (system of nodes and elements) that discretizes
the problem. That is; change the underlying fem mesh by increasing/decreasing the number of
elements employed and/or moving the node locations around and the approximations employed
will be altered. We will make use of the notation that he represents the length of an element e.
Therefore to underscore the relationship between our approximations and the fem mesh we
employ, our approximations will be denoted as follows:

𝑣𝑣(𝑥𝑥) ≈ 𝑣𝑣 ℎ (𝑥𝑥)

𝜔𝜔(𝑥𝑥) ≈ 𝜔𝜔ℎ (𝑥𝑥)

2. Element Shape Functions

We have already shown that simple piecewise linear approximations will not work for the model
beam problem. Instead approximations that are continuous and have continuous first order
derivatives are required. To derive the Cubic Hermitian Shape Functions commonly employed in
FEM beam formulations, let us consider a counterpart to the differential equation of equilibrium
for the model problem in which EI(x) is replaced with a constant average value within each
element and the transverse distributed load f(x) is taken as zero. With these simplifications the
differential equation of equilibrium takes the following form within each element

𝑑𝑑4 𝑣𝑣 ℎ (𝑥𝑥)
𝐸𝐸𝐸𝐸 = 0 ∀ 𝑥𝑥 ∈ 𝑥𝑥1𝑒𝑒 ≤ 𝑥𝑥 ≤ 𝑥𝑥2𝑒𝑒
𝑑𝑑𝑥𝑥 4
Here x1e is the x coordinate of local node 1 for the element and x2e is the x coordinate of local
node 2 for the element. A general solution for the above equation within each element takes the
form

𝑣𝑣 ℎ (𝑥𝑥) = 𝐶𝐶0 𝑥𝑥 3 + 𝐶𝐶1 𝑥𝑥 2 + 𝐶𝐶2 𝑥𝑥 + 𝐶𝐶3

where vh(x) is an approximation to the actual transverse displacement v(x) and C0, C1, C2, and C3
𝑑𝑑𝑣𝑣 ℎ
are constants of integration that can be solved for by enforcing values of vh and at specific
𝑑𝑑𝑥𝑥
points along the element. To force our approximation to match the displacements and rotations
at each end of the element we require the following equations to hold:

𝐶𝐶0 (𝑥𝑥1𝑒𝑒 )3 + 𝐶𝐶1 (𝑥𝑥1𝑒𝑒 )2 + 𝐶𝐶2 𝑥𝑥1𝑒𝑒 + 𝐶𝐶3 = 𝑑𝑑1𝑒𝑒

𝐶𝐶0 (𝑥𝑥2𝑒𝑒 )3 + 𝐶𝐶1 (𝑥𝑥2𝑒𝑒 )2 + 𝐶𝐶2 𝑥𝑥2𝑒𝑒 + 𝐶𝐶3 = 𝑑𝑑2𝑒𝑒

3𝐶𝐶0 (𝑥𝑥1𝑒𝑒 )2 + 2𝐶𝐶1 𝑥𝑥1𝑒𝑒 + 𝐶𝐶2 = 𝜃𝜃1𝑒𝑒

3𝐶𝐶0 (𝑥𝑥2𝑒𝑒 )2 + 2𝐶𝐶1 𝑥𝑥2𝑒𝑒 + 𝐶𝐶2 = 𝜃𝜃2𝑒𝑒

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 9
Solving the above system of linear equations for the constants of integration and rearranging
terms leads to the following approximation for the transverse displacement within each
element

𝑣𝑣 ℎ (𝑥𝑥) = 𝑁𝑁1𝑒𝑒 (𝑥𝑥)𝑑𝑑1𝑒𝑒 + 𝑁𝑁2𝑒𝑒 (𝑥𝑥)𝜃𝜃1𝑒𝑒 + 𝑁𝑁3𝑒𝑒 (𝑥𝑥)𝑑𝑑2𝑒𝑒 + 𝑁𝑁4𝑒𝑒 (𝑥𝑥)𝜃𝜃2𝑒𝑒

where the element shape functions are given by

1
𝑁𝑁1𝑒𝑒 = (𝑥𝑥 − 𝑥𝑥2𝑒𝑒 )�2𝑥𝑥 2 − 𝑥𝑥2𝑒𝑒 𝑥𝑥 − 3𝑥𝑥1𝑒𝑒 𝑥𝑥 − 𝑥𝑥2𝑒𝑒 (𝑥𝑥2𝑒𝑒 − 3𝑥𝑥1𝑒𝑒 )�
(ℎ𝑒𝑒 )3
1
𝑁𝑁2𝑒𝑒 = (𝑥𝑥 − 𝑥𝑥1𝑒𝑒 )(𝑥𝑥 − 𝑥𝑥2𝑒𝑒 )2
(ℎ𝑒𝑒 )2
1
𝑁𝑁3𝑒𝑒 = (𝑥𝑥 − 𝑥𝑥1𝑒𝑒 )�−2𝑥𝑥 2 + 𝑥𝑥1𝑒𝑒 𝑥𝑥 + 3𝑥𝑥2𝑒𝑒 𝑥𝑥 + 𝑥𝑥1𝑒𝑒 (𝑥𝑥1𝑒𝑒 − 3𝑥𝑥2𝑒𝑒 )�
(ℎ𝑒𝑒 )3
1
𝑁𝑁4𝑒𝑒 = (𝑥𝑥 − 𝑥𝑥2𝑒𝑒 )(𝑥𝑥 − 𝑥𝑥1𝑒𝑒 )2
(ℎ𝑒𝑒 )2
ℎ𝑒𝑒 = 𝑥𝑥2𝑒𝑒 − 𝑥𝑥1𝑒𝑒 is the element length

Upon more careful observation, the reader will note that all shape functions are zero at 𝑥𝑥1𝑒𝑒
except for 𝑁𝑁1𝑒𝑒 which is exactly equal to one at that location. Similarly, all shape functions are

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 10
equal to zero at 𝑥𝑥2𝑒𝑒 except for 𝑁𝑁3𝑒𝑒 which is exactly equal to one at that location. Likewise, all
shape functions have slopes that are equal to zero at 𝑥𝑥1𝑒𝑒 except for 𝑁𝑁2𝑒𝑒 which has a slope that is
exactly equal to one at that location. All shape functions have slopes that are equal to zero at 𝑥𝑥2𝑒𝑒
except for 𝑁𝑁4𝑒𝑒 which has a slope that is exactly equal to one at that location. In this way our
approximation will result in displacements and rotations at the ends of the element that will
match the nodal displacements and nodal rotations at each end.

3. Nodal Shape Functions

We have already seen that an internal node will be shared between two adjacent elements such
that the node will be identical to local node 2 for the element on the right and to local node 1 for
the element on the left. Combining the element shape functions for adjacent elements that
share a node, we can obtain the shape functions associated with each node. These functions are
called the nodal shape functions. If we denote the element on the right by the label e and the
element on the left by e+1 we obtain for the nodal shape functions at node A (the shared node),
the following:

𝑁𝑁3𝑒𝑒 (𝑥𝑥) 𝑥𝑥𝐴𝐴−1 ≤ 𝑥𝑥 < 𝑥𝑥𝐴𝐴

𝑁𝑁𝐴𝐴 (𝑥𝑥) = �𝑁𝑁1𝑒𝑒+1 (𝑥𝑥) 𝑥𝑥𝐴𝐴 ≤ 𝑥𝑥 ≤ 𝑥𝑥𝐴𝐴+1
0 otherwise

𝑁𝑁4𝑒𝑒 (𝑥𝑥) 𝑥𝑥𝐴𝐴−1 ≤ 𝑥𝑥 < 𝑥𝑥𝐴𝐴

�𝐴𝐴 (𝑥𝑥) = �𝑁𝑁 𝑒𝑒+1 (𝑥𝑥)
𝑁𝑁 𝑥𝑥𝐴𝐴 ≤ 𝑥𝑥 ≤ 𝑥𝑥𝐴𝐴+1
2
0 otherwise

Here there are two shape functions at each node A. Shape function NA is used to interpolate the
transverse displacements at each node while shape function 𝑁𝑁�𝐴𝐴 is used to interpolate the
rotations at each node. Taking a global view of our approximations (for the entire beam) we then
obtain the following approximation to our transverse displacement function and weighting
function, respectively
𝑛𝑛+1 𝑛𝑛+1
𝑣𝑣 ℎ (𝑥𝑥) = � 𝑁𝑁𝐴𝐴 (𝑥𝑥)𝑑𝑑𝐴𝐴 + � �𝐴𝐴 (𝑥𝑥)𝜃𝜃𝐴𝐴
𝑁𝑁
𝐴𝐴=1 𝐴𝐴=1
𝑛𝑛+1 𝑛𝑛+1
𝜔𝜔ℎ (𝑥𝑥) = � 𝑁𝑁𝐴𝐴 (𝑥𝑥)𝜔𝜔𝐴𝐴 + � �𝐴𝐴 (𝑥𝑥)𝜙𝜙𝐴𝐴
𝑁𝑁
𝐴𝐴=1 𝐴𝐴=1

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 11
Plotting the nodal shape functions we obtain

A few observations on the nodal shape functions:

• Nodal shape function NA(x) is exactly equal to 1 at node A but is zero at all other
nodes. Furthermore, the nodal shape function 𝑁𝑁 �𝐴𝐴 (𝑥𝑥) is equal to zero at all nodes
including node A. Thus the transverse displacement at x=xA will be exactly equal to
dA, the nodal displacement at node A.
• The slopes of nodal shape function NA(x) is exactly equal to 0 at all nodes including
node A. Furthermore, the nodal shape function 𝑁𝑁 �𝐴𝐴 (𝑥𝑥) has slope equal to 1 at x=xA
and zero slope at all other nodes. Thus the rotation at x=xA will be exactly equal to θA,
the nodal rotation at node A.
• The shape functions are continuous and have continuous first derivatives as required
to obtain solutions that are bounded.
• The shape functions are local approximations in that each shape function is non zero
only over a specific region. The locality of the approximations used will lead to a

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 12
 specific structure to the stiffness matrix that can be exploited within an algorithm to
obtain the fem solution.

4. Element Stiffness Matrices and Force Vectors

Now with an approximation in hand for the transverse displacement and weighting functions,
we can return our attention to the variational form of the model problem. Consider the internal
virtual work term given by
𝐿𝐿
𝑑𝑑2 𝜔𝜔ℎ (𝑥𝑥) 𝑑𝑑2 𝑣𝑣 ℎ (𝑥𝑥)
� 𝐸𝐸𝐸𝐸(𝑥𝑥) 𝑑𝑑𝑑𝑑
0 𝑑𝑑𝑥𝑥 2 𝑑𝑑𝑥𝑥 2

Since we have constructed our approximations on an element by element basis, it would be

convenient to split the integrand into a summation of integrands calculated within each element
domain, i.e.
𝑛𝑛 𝑥𝑥𝑒𝑒+1
𝑑𝑑2 𝜔𝜔ℎ (𝑥𝑥) 𝑑𝑑2 𝑣𝑣 ℎ (𝑥𝑥)
�� 𝐸𝐸𝐸𝐸(𝑥𝑥) 𝑑𝑑𝑑𝑑
𝑑𝑑𝑥𝑥 2 𝑑𝑑𝑥𝑥 2
𝑒𝑒=1 𝑥𝑥𝑒𝑒

Within each element we rewrite our approximations in matrix form as follows

𝑑𝑑1𝑒𝑒
⎛𝜃𝜃 𝑒𝑒 ⎞
𝑣𝑣 ℎ (𝑥𝑥) = [𝑁𝑁1𝑒𝑒 , 𝑁𝑁2𝑒𝑒 , 𝑁𝑁3𝑒𝑒 , 𝑁𝑁4𝑒𝑒 ]1𝑥𝑥4 ⎜ 1𝑒𝑒 ⎟ 𝑒𝑒
= ℕ1𝑥𝑥4 𝑑𝑑⃗4𝑥𝑥1
𝑒𝑒
𝑑𝑑2
𝜃𝜃 𝑒𝑒
⎝ 2 ⎠4𝑥𝑥1

𝑤𝑤1𝑒𝑒
⎛ 𝜙𝜙 𝑒𝑒 ⎞
𝑤𝑤 ℎ (𝑥𝑥) = [𝑁𝑁1𝑒𝑒 , 𝑁𝑁2𝑒𝑒 , 𝑁𝑁3𝑒𝑒 , 𝑁𝑁4𝑒𝑒 ]1𝑥𝑥4 ⎜ 1𝑒𝑒 ⎟ 𝑒𝑒
= ℕ1𝑥𝑥4 𝜔𝜔 𝑒𝑒
�⃗4𝑥𝑥1
𝑤𝑤2
𝜙𝜙 𝑒𝑒
⎝ 2⎠ 4𝑥𝑥1

Here the subscripts denote the size of the individual matrices, shown for clarity, with the
notation that mxn indicates a matrix with m rows and n columns. 𝑑𝑑⃗4𝑥𝑥1
𝑒𝑒
and 𝜔𝜔 𝑒𝑒
�⃗4𝑥𝑥1 are the 4x1
element displacement vector and element virtual displacement vector, respectively. We also
introduce an element strain displacement matrix 𝔹𝔹𝑒𝑒 that relates the curvature in the beam to
the element displacement vector such that

𝑑𝑑2 𝑑𝑑2 𝑑𝑑2 𝑒𝑒 𝑑𝑑2 𝑒𝑒

𝔹𝔹𝑒𝑒 (𝑥𝑥) = � 2 𝑁𝑁1𝑒𝑒 , 2 𝑁𝑁2𝑒𝑒 , 𝑁𝑁 , 𝑁𝑁 �
𝑑𝑑𝑑𝑑 𝑑𝑑𝑑𝑑 𝑑𝑑𝑑𝑑 2 3 𝑑𝑑𝑑𝑑 2 4 1𝑥𝑥4

and

𝑑𝑑2 𝑣𝑣 ℎ 𝑒𝑒
= 𝔹𝔹1𝑥𝑥4 𝑑𝑑⃗4𝑥𝑥1
𝑒𝑒
𝑑𝑑𝑑𝑑 2

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 13
 𝑑𝑑2 𝜔𝜔ℎ 𝑒𝑒 𝑒𝑒 𝑇𝑇
= 𝔹𝔹1𝑥𝑥4 𝜔𝜔
�⃗4𝑥𝑥1 �⃗𝑒𝑒 )1𝑥𝑥4
= (𝜔𝜔 𝔹𝔹𝑒𝑒𝑒𝑒
4𝑥𝑥1
𝑑𝑑𝑑𝑑 2
The internal virtual work term now becomes
𝑛𝑛 𝑥𝑥𝑒𝑒+1
�� 𝑇𝑇
�⃗𝑒𝑒 )1𝑥𝑥4
𝐸𝐸𝐸𝐸(𝑥𝑥)(𝜔𝜔 𝔹𝔹𝑒𝑒𝑒𝑒 𝑒𝑒 ⃗𝑒𝑒
4𝑥𝑥1 𝔹𝔹1𝑥𝑥4 𝑑𝑑4𝑥𝑥1 𝑑𝑑𝑑𝑑
𝑒𝑒=1 𝑥𝑥𝑒𝑒

Or, since the element displacement vector and element virtual displacement vector do not
depend on location x
𝑛𝑛 𝑥𝑥𝑒𝑒+1
�(𝜔𝜔 𝑒𝑒 )𝑇𝑇
�⃗ 1𝑥𝑥4 �� 𝐸𝐸𝐸𝐸(𝑥𝑥)𝔹𝔹𝑒𝑒𝑒𝑒 𝑒𝑒 ⃗𝑒𝑒
4𝑥𝑥1 𝔹𝔹1𝑥𝑥4 𝑑𝑑𝑑𝑑 � 𝑑𝑑4𝑥𝑥1
𝑒𝑒=1 𝑥𝑥𝑒𝑒

Introducing the element stiffness matrix 𝕜𝕜𝑒𝑒 that is a 4x4 matrix as

𝑥𝑥𝑒𝑒+1
𝕜𝕜𝑒𝑒4𝑥𝑥4 = � 𝐸𝐸𝐸𝐸(𝑥𝑥)𝔹𝔹𝑒𝑒𝑇𝑇 𝑒𝑒
4𝑥𝑥1 𝔹𝔹1𝑥𝑥4 𝑑𝑑𝑑𝑑
𝑥𝑥𝑒𝑒

The internal virtual work term now simplifies to

𝑛𝑛
𝑇𝑇 {𝕜𝕜𝑒𝑒 }𝑑𝑑⃗𝑒𝑒
�⃗𝑒𝑒 )1𝑥𝑥4
�(𝜔𝜔 4𝑥𝑥4 4𝑥𝑥1
𝑒𝑒=1

For the common case of EI(x) being equal to a constant value, i.e. a prismatic beam the element
stiffness matrix takes the following form
12 6 −12 6
⎡ 𝑒𝑒 2 ⎤
⎢ (ℎ ) ℎ𝑒𝑒 (ℎ𝑒𝑒 )2 ℎ𝑒𝑒 ⎥
⎢ 6 −6 ⎥
𝐸𝐸𝐸𝐸 ⎢ ℎ𝑒𝑒 4 𝑒𝑒
2 ⎥
𝕜𝕜𝑒𝑒4𝑥𝑥4 = 𝑒𝑒 ⎢ ℎ
ℎ −12 −6 12 −6⎥
⎢ 𝑒𝑒 2 ⎥
⎢(ℎ ) ℎ𝑒𝑒 (ℎ𝑒𝑒 )2 ℎ𝑒𝑒 ⎥
⎢ 6 −6 ⎥
⎣ ℎ𝑒𝑒 2 𝑒𝑒
4 ⎦
ℎ
The external virtual work due to the applied distributed transverse load f(x) is then given by
𝑛𝑛 𝑥𝑥𝑒𝑒+1 𝑛𝑛
ℎ 𝑇𝑇 ⃗𝑒𝑒
�� �⃗𝑒𝑒 )1𝑥𝑥4
𝑓𝑓(𝑥𝑥)𝜔𝜔 (𝑥𝑥)𝑑𝑑𝑑𝑑 = �(𝜔𝜔 𝑓𝑓4𝑥𝑥1
𝑒𝑒=1 𝑥𝑥𝑒𝑒 𝑒𝑒=1

where the element force vector 𝑓𝑓⃗4𝑥𝑥1

𝑒𝑒
is given by
𝑥𝑥𝑒𝑒+1
𝑓𝑓⃗4𝑥𝑥1
𝑒𝑒
=� 𝑓𝑓(𝑥𝑥)(ℕ𝑒𝑒 )𝑇𝑇4𝑥𝑥1 𝑑𝑑𝑑𝑑
𝑥𝑥𝑒𝑒

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 14
Before proceeding further let’s consider the format for the element force vector for specific
cases of transverse loading.

• Uniform Distributed load f(x)=q0

1
⎡ ℎ𝑒𝑒 ⎤
𝑞𝑞0 ℎ𝑒𝑒 ⎢ 6 ⎥
𝑓𝑓⃗4𝑥𝑥1
𝑒𝑒
= ⎢ ⎥
2 ⎢ 1 ⎥
𝑒𝑒
⎢ ℎ ⎥
−
⎣ 6⎦

𝑞𝑞0
• Triangular distributed load 𝑓𝑓(𝑥𝑥) = (𝑥𝑥 − 𝑥𝑥1𝑒𝑒 )
ℎ 𝑒𝑒

3
⎡ ⎤
⎢ 10𝑒𝑒 ⎥
⎢ ℎ ⎥
𝑞𝑞0 ℎ𝑒𝑒 ⎢ 15 ⎥
𝑓𝑓⃗4𝑥𝑥1
𝑒𝑒
=
2 ⎢ 7 ⎥
⎢ ⎥
⎢ 10𝑒𝑒 ⎥
⎢ ℎ ⎥
⎣− 10⎦

On closer inspection of the results for the above two cases, one can see that the applied
distributed load is in essence replaced with a statically equivalent set of nodal forces acting at
the ends of the element.

The external virtual work also consists of two additional terms involving the applied
concentrated moment M0 and concentrated transverse load Q acting at the end of the beam
located at x=L. It will be best to treat these terms on a global basis once the element equations
are assembled as discussed shortly for the full beam. Before leaving the element point of view
for our model problem we need to consider how to handle the effects of non-zero prescribed

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 15
displacements and/or rotations at specific locations (i.e. nodes) along the beam. For the model
problem both the transverse displacement and rotation have specified prescribed values at x=0
equal to g and b; respectively. Most elements will not have any prescribed displacements or
rotations at either end with only a few elements being affected by prescribed non-zero
displacements/rotations. For the model problem currently under consideration, only element 1
is affected by non-zero prescribed displacements/rotations.

Consider an element e in which some of the displacements and/or rotations at either end are
prescribed to have a non-zero value. This situation arises when a support point along the beam
experiences a known settlement or rotation. The displacement vector for the element can be
split into a sum of two vectors, a vector containing only the unknown degrees of freedom for the
element, 𝑑𝑑̂4𝑥𝑥1
𝑒𝑒
, and a vector containing the non-zero prescribed values, 𝑑𝑑𝑔𝑔𝑒𝑒
4𝑥𝑥1

𝑑𝑑⃗4𝑥𝑥1
𝑒𝑒
= 𝑑𝑑̂4𝑥𝑥1
𝑒𝑒
+ 𝑑𝑑𝑔𝑔𝑒𝑒
4𝑥𝑥1

With this notation in place we can rewrite the internal virtual work term as
𝑛𝑛 𝑛𝑛 𝑛𝑛
𝑒𝑒
�(𝜔𝜔 𝑇𝑇
�⃗𝑒𝑒 )1𝑥𝑥4 �{𝕜𝕜𝑒𝑒4𝑥𝑥4 }𝑑𝑑̂4𝑥𝑥1
𝑒𝑒
+ {𝕜𝕜𝑒𝑒4𝑥𝑥4 }𝑑𝑑𝑔𝑔𝑒𝑒 � = �(𝜔𝜔 𝑇𝑇 {𝕜𝕜𝑒𝑒 }𝑑𝑑̂ 𝑒𝑒
�⃗𝑒𝑒 )1𝑥𝑥4 4𝑥𝑥4 4𝑥𝑥1
𝑇𝑇
�⃗𝑒𝑒 )1𝑥𝑥4
+ �(𝜔𝜔 �−𝑓𝑓⃗𝑔𝑔 �
4𝑥𝑥1 4𝑥𝑥1
𝑒𝑒=1 𝑒𝑒=1 𝑒𝑒=1

Here 𝑓𝑓⃗𝑔𝑔𝑒𝑒 = −{𝕜𝕜𝑒𝑒4𝑥𝑥4 }𝑑𝑑𝑔𝑔𝑒𝑒 is an element force vector emanating from the non-zero prescribed
4𝑥𝑥1
displacements. The negative sign has been added for convenience as this term will be lumped
together with the external virtual work terms upon assembling the global system of equations
for the model problem.

5. Assembly of the Global System of Equations

To obtain an approximation to our model problem using the fem formulation presented, a global
system of equations must be assembled that account for all the degrees of freedom. Recall that
we have made use of a system of finite elements and global nodes for the entire beam as
follows:

The direct stiffness method is an efficient method for assembling the global system of
equations making use of the element by element contributions to both the internal and external
virtual work terms. Furthermore, it is a method that can easily be implemented as a computer

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 16
algorithm. The interested reader is referred to the first two references for more details. In this
article only the main ideas will be presented.

As a first step to the direct stiffness method, the displacements and rotations are renumbered
such that the lowest numbered global node will have its displacement and rotation numbered
sequentially followed by the next lowest numbered global node and so on. If a displacement or a
rotation at a node is prescribed then that dof is not numbered but is instead assigned a label of
0. Once all the displacements and rotations have been numbered by the preceding procedure
the largest integer number will represent the total number of degrees of freedom for the entire
problem, i.e. the total number of unknown displacements/rotations for the problem, denoted by
𝑒𝑒
neq. Furthermore, at each element e a mapping array, 𝑙𝑙𝑚𝑚 , will be used to map (link) the
elements dofs to the globally numbered degrees of freedom.

Figure 3: Global dofs labelling

For the current model problem the displacement and rotation at global node 1 are not
numbered since both have been prescribed values. The displacement at global node 2 is
numbered 1 while the rotation at global node 2 is numbered 2 and so on. When all the nodal
displacements and rotations have been numbered there will be a total of neq=2*n equations
(dofs) where n represents the integer number of elements used. The mapping arrays for the 1st,
2nd, nth-1, and nth elements are given by
0 1 2𝑛𝑛 − 5 2𝑛𝑛 − 3
1
𝑙𝑙𝑚𝑚 = �0� 2
𝑙𝑙𝑚𝑚 = �2� 𝑛𝑛−1
𝑙𝑙𝑚𝑚 = �2𝑛𝑛 − 4� 𝑛𝑛
𝑙𝑙𝑚𝑚 = �2𝑛𝑛 − 2�
1 3 2𝑛𝑛 − 3 2𝑛𝑛 − 1
2 4 2𝑛𝑛 − 2 2𝑛𝑛
Once the element contributions to the global system of equations is completed the variational
form of the model problem will take the following shape

𝜔𝜔𝑇𝑇 1𝑥𝑥𝑥𝑥𝑥𝑥𝑥𝑥 �𝕂𝕂𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑑𝑑⃗𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1 − 𝐹𝐹⃗𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1 � = 0 ∀ 𝜔𝜔𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1

Where 𝜔𝜔𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1 is the global virtual displacement vector that contains one row for each global
dof, 𝕂𝕂𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 is the global stiffness matrix, and 𝐹𝐹⃗𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1 is the global force vector that has neq
rows. Since the above equation must hold for any global virtual displacement vector ω, the
following equation must also hold

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 17
 𝕂𝕂𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑒𝑒𝑒𝑒 𝑑𝑑⃗𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1 = 𝐹𝐹⃗𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1

The global displacement vector for the model problem takes the following form where it should
be noted that only displacements and rotations that are not prescribed are included:
𝑑𝑑2
⎧ 𝜃𝜃 ⎫
⎪ 2 ⎪
⎪ 𝑑𝑑3 ⎪
𝜃𝜃3
𝑑𝑑⃗𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1 =
⎨ ⋮ ⎬
⎪ 𝜃𝜃𝑛𝑛 ⎪
⎪𝑑𝑑𝑛𝑛+1 ⎪
⎩𝜃𝜃𝑛𝑛+1 ⎭

The global stiffness matrix 𝕂𝕂 is assembled by creating a zero matrix of size neq x neq with all
entries equal to zero. The individual element stiffness matrices are then added to the global
stiffness matrix employing the following algorithm:

For each element e from 1 to n:

For each row i from 1 to 4:
𝑒𝑒
I=𝑙𝑙𝑚𝑚 (i) (global index for local row i)
If (I ≠ 0) then
For each column j from 1 to 4:
𝑒𝑒
J=𝑙𝑙𝑚𝑚 (j) (global index for local column j)
If (J ≠ 0) then
𝕂𝕂(I,J)= 𝕂𝕂 (I,J)+𝕜𝕜𝑒𝑒 (𝑖𝑖, 𝑗𝑗)
End if
End loop over j
End if
End loop over i
End loop over elements

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 18
The global force vector 𝐹𝐹⃗𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1 is assembled by creating a zero vector of size neq x 1 with all
entries equal to zero. The individual element force vectors due to the applied transverse load,
𝑒𝑒
𝑓𝑓4𝑥𝑥1 , and non-zero prescribed displacements, 𝑓𝑓𝑔𝑔 𝑒𝑒 , are then added to the global force vector
4𝑥𝑥1
employing the following algorithm:

For each element e from 1 to n:

For each row i from 1 to 4:
I=lme(i) (global index for row i)
If (I≠0) then
𝐹𝐹⃗ (𝐼𝐼) = 𝐹𝐹⃗ (𝐼𝐼) + 𝑓𝑓 𝑒𝑒 (𝑖𝑖) + 𝑓𝑓𝑔𝑔𝑒𝑒 (𝑖𝑖)
End if
End loop over i
End loop over elements

As a final step to assembling the global force vector, any applied concentrated loads and
moments are added to the appropriate rows in the global force vector. For the model problem
under consideration the value of Q, the applied transverse concentrated load at x=L, is added to
the 2n-1 row of the global force vector; while the value of M0 is added to the 2nth row of the
global force vector.
𝐹𝐹1
⎧ 𝐹𝐹2 ⎫
⎪ ⎪
𝐹𝐹⃗𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛1 = ⋮
⎨𝐹𝐹2𝑛𝑛−1 + 𝑄𝑄 ⎬
⎪ ⎪
⎩ 𝐹𝐹2𝑛𝑛 + 𝑀𝑀0 ⎭

Once the global matrices are assembled, the system of equations can be solved using a
numerical method such as Gaussian Elimination.

D. Assembly of the Global System of Equations

To highlight the previous topics covered we will consider an example problem of a simply
supported prismatic beam acted upon by an exponential transverse load as shown below.

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 19
 𝐸𝐸𝐸𝐸 = 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 = 4 × 106 𝑘𝑘𝑖𝑖𝑖𝑖 ∙ 𝑓𝑓𝑓𝑓 2
𝑞𝑞0 𝑥𝑥
� �
𝑓𝑓(𝑥𝑥) = �𝑒𝑒 𝐿𝐿 − 1�
(𝑒𝑒 1 − 1)
𝑘𝑘𝑘𝑘𝑘𝑘
𝑞𝑞0 = 10 𝐿𝐿 = 15 𝑓𝑓𝑓𝑓
𝑓𝑓𝑓𝑓

The differential equation of equilibrium has a known solution for the transverse displacement
given by
𝑥𝑥
� � 1 𝑥𝑥 4 3 𝑒𝑒 1 𝑥𝑥 3 1 𝑥𝑥 2 31 5𝑒𝑒 1 𝑥𝑥
𝑣𝑣(𝑥𝑥) = 𝛼𝛼 �𝑒𝑒 𝐿𝐿 − � � + � − �� � − � � + � − � � � − 1�
24 𝐿𝐿 12 6 𝐿𝐿 2 𝐿𝐿 24 6 𝐿𝐿

where
𝑞𝑞0 𝐿𝐿4
𝛼𝛼 = = 0.07366 ft
𝐸𝐸𝐸𝐸(𝑒𝑒 1 −1)

The internal bending moment in the beam is given by

𝑑𝑑2 𝑣𝑣(𝑥𝑥) 𝑞𝑞0 𝐿𝐿2 𝑥𝑥

� � 1 𝑥𝑥 2 3 𝑥𝑥 𝑥𝑥
𝑀𝑀(𝑥𝑥) = 𝐸𝐸𝐸𝐸 = �𝑒𝑒 𝐿𝐿 − � � + � � − 𝑒𝑒 1 � � − 1�
𝑑𝑑𝑑𝑑 2 (𝑒𝑒 1 − 1) 2 𝐿𝐿 2 𝐿𝐿 𝐿𝐿

while the internal shear force in the beam is given by

𝑑𝑑𝑑𝑑(𝑥𝑥) 𝑞𝑞0 𝐿𝐿 𝑥𝑥
� � 𝑥𝑥 3
𝑉𝑉(𝑥𝑥) = = 1 �𝑒𝑒 𝐿𝐿 − � � + − 𝑒𝑒 1 �
𝑑𝑑𝑑𝑑 (𝑒𝑒 − 1) 𝐿𝐿 2

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 20
 Vertical displacement vs. x
0.0007

0.0006

0.0005
Displacement (ft)

0.0004

0.0003

0.0002

0.0001

0
-1 1 3 5 7 9 11 13 15
-0.0001
x (ft)

Moment vs x
0
0 2 4 6 8 10 12 14
-20

-40
Moment (k-ft)

-60

-80

-100

-120

-140
x (ft)

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 21
 Shear Force vs x
50

40

30
Shear Force (kips)

20

10

0
0 2 4 6 8 10 12 14
-10

-20

-30
x (ft)

The example problem will be discretized with 3 elements of equal length. Due to the prescribed
displacements at x=0 and x=L; the total number of global dofs are equal to neq=6 and are
labeled in red in Figure 4.

Figure 4: FE mesh with global dofs labeled in red.

The three element stiffness matrices are identical and equal to:

3.84 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 −3.84 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘
⎡ ⎤
⎢ 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 3.2 × 106 𝑘𝑘𝑘𝑘𝑝𝑝 ∙ 𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓⎥
𝕜𝕜1 = 𝕜𝕜2 = 𝕜𝕜3 = ⎢ 5
⎢−3.84 × 10 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 3.84 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 ⎥⎥
⎣ 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 3.2 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓⎦

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 22
The element force vectors due to the applied transverse load are given by:
1.570 𝑘𝑘𝑘𝑘𝑘𝑘 7.946 𝑘𝑘𝑘𝑘𝑘𝑘 16.846 𝑘𝑘𝑘𝑘𝑘𝑘
1.760 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 7.253 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 14.92 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
𝑓𝑓 1 = � � 𝑓𝑓 2 = � � 𝑓𝑓 3 = � �
3.867 𝑘𝑘𝑘𝑘𝑘𝑘 11.15 𝑘𝑘𝑘𝑘𝑘𝑘 21.32 𝑘𝑘𝑘𝑘𝑘𝑘
−2.717 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −8.589 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −16.783 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓

Employing the direct stiffness matrix to assemble the global system of equations we obtain
after assembling element 1 the following:

3.2 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 0 0
⎡ ⎤
5
⎢ −9.6 × 10 𝑘𝑘𝑘𝑘𝑘𝑘 3.84 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑡𝑡 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 0 0 0⎥
6
𝕂𝕂 = ⎢1.6 × 10 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 3.2 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 0 0⎥ 𝐹𝐹 =
⎢ 0 0 0 0 0 0⎥
⎢ 0 0 0 0 0 0⎥
⎣ 0 0 0 0 0 0⎦
1.76 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
⎧ 3.867 𝑘𝑘𝑘𝑘𝑘𝑘 ⎫
⎪ ⎪
−2.717 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
⎨ 0 ⎬
⎪ 0 ⎪
⎩ 0 ⎭

After assembling element 2 the global matrices become:

𝕂𝕂 =
3.2 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 0 0
⎡ 5
⎤
⎢ −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 5
7.68 × 10 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 0 −3.84 × 10 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 0⎥
⎢1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 6.4 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0⎥
⎢ −3.84 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 0⎥
⎢ 0 3.84 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘
0 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 6
1.6 × 10 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 ⎥⎥
⎢ −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 3.2 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
⎣ 0 0 0 0⎦
0 0

1.76 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓

⎧ 11.814 𝑘𝑘𝑘𝑘𝑘𝑘 ⎫
⎪ ⎪
4.536 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
𝐹𝐹 =
⎨ 11.153 𝑘𝑘𝑘𝑘𝑘𝑘 ⎬
⎪−8.589 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓⎪
⎩ 0 ⎭

After assembling element 3 we obtain final version of the global matrices due to the fact that
there are no non-zero prescribed displacements or concentrated forces/moments applied:

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 23
𝕂𝕂 =
3.2 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 0 0
⎡ ⎤
⎢ −9.6 × 10 5 𝑘𝑘𝑘𝑘𝑝𝑝 5
7.68 × 10 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 0 −3.84 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 0 ⎥
⎢1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 6.4 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 ⎥
⎢ −3.84 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 −9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 7.68 × 105 𝑘𝑘𝑘𝑘𝑘𝑘/𝑓𝑓𝑓𝑓 0 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 ⎥
⎢ 0 ⎥
⎢ 0 5
9.6 × 10 𝑘𝑘𝑘𝑘𝑘𝑘 6
1.6 × 10 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 0 6.4 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓⎥
⎣ 0 0 0 9.6 × 105 𝑘𝑘𝑘𝑘𝑘𝑘 1.6 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 3.2 × 106 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 ⎦

1.76 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓

⎧
11.814 𝑘𝑘𝑘𝑘𝑘𝑘 ⎫
⎪ ⎪
4.536 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
𝐹𝐹 =
⎨ 28 𝑘𝑘𝑘𝑘𝑘𝑘 ⎬
⎪ 6.331 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓 ⎪
⎩−16.783 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓⎭

Solving for the global displacements we obtain:

𝜃𝜃1 1.298 × 10−4 𝑟𝑟𝑟𝑟𝑟𝑟
⎧𝑑𝑑 ⎫ ⎧ 5.524 × 10−4 𝑓𝑓𝑓𝑓 ⎫
⎪ 2⎪ ⎪ ⎪
𝜃𝜃2 −5
𝑑𝑑⃗ = 𝕂𝕂 𝐹𝐹⃗ =
−1
= 7.295 × 10 −4𝑟𝑟𝑟𝑟𝑟𝑟
⎨𝑑𝑑3 ⎬ ⎨ 5.947 × 10 𝑓𝑓𝑓𝑓 ⎬
⎪𝜃𝜃3 ⎪ ⎪−6.193 × 10−5 𝑟𝑟𝑟𝑟𝑟𝑟⎪
⎩ 𝜃𝜃4 ⎭ ⎩ −1.527 × 10−4 𝑟𝑟𝑟𝑟𝑟𝑟 ⎭

Due to the boundary conditions we also have 𝑑𝑑1 = 𝑑𝑑4 = 0. The resulting approximation to the
transverse displacement vs the exact solution is shown below.

Vertical displacement vs. x

7.00E-04

6.00E-04

5.00E-04
Displacement (ft)

4.00E-04 approximate disp e1

3.00E-04 Exact displacement
approximate displacement e2
2.00E-04
approximate displacement e3
1.00E-04

0.00E+00
0 5 10 15
-1.00E-04
x (ft)

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 24
 1. Assembly of the Global System of Equations
Having obtained a solution for the global displacements the element displacements vectors are
now used to determine the resultant forces and moments at each end of the element via the
equation
𝐹𝐹 𝑒𝑒
⎧ 1𝑒𝑒 ⎫
𝑀𝑀1
𝐹𝐹⃗ 𝑒𝑒 = 𝕜𝕜𝑒𝑒4𝑥𝑥4 𝑑𝑑⃗4𝑥𝑥1
𝑒𝑒
− 𝑓𝑓⃗4𝑥𝑥1
𝑒𝑒
= 𝑒𝑒
⎨ 𝐹𝐹2 ⎬
⎩𝑀𝑀2𝑒𝑒 ⎭

𝑑𝑑𝑑𝑑(𝑥𝑥) 𝑑𝑑𝑑𝑑(𝑥𝑥)
Since we have the relationships = 𝑉𝑉(𝑥𝑥) and = 𝑓𝑓(𝑥𝑥) we can make use of the element
𝑑𝑑𝑑𝑑 𝑑𝑑𝑑𝑑
shape functions to obtain approximations to the shear force and bending moment within each
element by making use of the nodal values. The approximation to the internal shear force within
the element is given by:

𝐹𝐹1𝑒𝑒
⎛𝑓𝑓(𝑥𝑥 𝑒𝑒 )⎞
𝑉𝑉 ℎ (𝑥𝑥) = [𝑁𝑁1𝑒𝑒 , 𝑁𝑁2𝑒𝑒 , 𝑁𝑁3𝑒𝑒 , 𝑁𝑁4𝑒𝑒 ]1𝑥𝑥4 ⎜ 1𝑒𝑒 ⎟
−𝐹𝐹2
𝑒𝑒
⎝𝑓𝑓(𝑥𝑥 2 )⎠
4𝑥𝑥1

The approximation to the internal bending moment within the element is given by:

−𝑀𝑀1𝑒𝑒
⎛ 𝐹𝐹 𝑒𝑒 ⎞
𝑀𝑀ℎ (𝑥𝑥) = [𝑁𝑁1𝑒𝑒 , 𝑁𝑁2𝑒𝑒 , 𝑁𝑁3𝑒𝑒 , 𝑁𝑁4𝑒𝑒 ]1𝑥𝑥4 ⎜ 1𝑒𝑒 ⎟
𝑀𝑀2
𝑒𝑒
−𝐹𝐹
⎝ 2⎠ 4𝑥𝑥1

For the current numerical example the element resultant forces are given by:

a) Element 1
0
1.298 × 10−4 𝑟𝑟𝑟𝑟𝑟𝑟
𝑑𝑑⃗4𝑥𝑥1
1
=� �
5.524 × 10−4 𝑓𝑓𝑓𝑓
7.295 × 10−5 𝑟𝑟𝑟𝑟𝑟𝑟

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 25
 −19.05 𝑘𝑘𝑘𝑘𝑘𝑘
1 ⃗1 0.016 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
𝐹𝐹⃗1 = 𝑘𝑘4𝑥𝑥4 𝑑𝑑4𝑥𝑥1 − 𝑓𝑓⃗4𝑥𝑥1
1
=� �
13.61 𝑘𝑘𝑘𝑘𝑘𝑘
−86.467 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓

b) Element 2
5.524 × 10−4 𝑓𝑓𝑓𝑓
−5
𝑑𝑑⃗4𝑥𝑥1
2
= � 7.295 × 10 −4𝑟𝑟𝑟𝑟𝑑𝑑 �
5.947 × 10 𝑓𝑓𝑓𝑓
−6.193 × 10−5 𝑟𝑟𝑟𝑟𝑟𝑟
−13.61 𝑘𝑘𝑘𝑘𝑘𝑘
2 ⃗2 86.491 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
𝐹𝐹⃗ 2 = 𝑘𝑘4𝑥𝑥4 𝑑𝑑4𝑥𝑥1 − 𝑓𝑓⃗4𝑥𝑥1
2
=� �
−5.489 𝑘𝑘𝑘𝑘𝑘𝑘
−113.475 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓

c) Element 3
5.947 × 10−4 𝑓𝑓𝑓𝑓
−5
𝑑𝑑⃗4𝑥𝑥1
3
= �−6.193 × 10 𝑟𝑟𝑟𝑟𝑟𝑟 �
0
−1.527 × 10−4 𝑟𝑟𝑎𝑎𝑑𝑑
5.474 𝑘𝑘𝑘𝑘𝑘𝑘
3 ⃗3 113.5 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
𝐹𝐹⃗ 3 = 𝑘𝑘4𝑥𝑥4 𝑑𝑑4𝑥𝑥1 − 𝑓𝑓⃗4𝑥𝑥1
3
=� �
−43.64 𝑘𝑘𝑘𝑘𝑘𝑘
−0.033 𝑘𝑘𝑘𝑘𝑘𝑘 ∙ 𝑓𝑓𝑓𝑓
Plotting the approximations to the internal shear and bending moment vs the exact solution we
obtain:

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 26
 Shear Force vs x
50
40
Shear Force (kips)

30
20
10
0
-10 0 2 4 6 8 10 12 14
-20
-30
x (ft)

exact shear force approx shear element 1

approx shear element 2 approx shear element 3

Moment vs x
20
0
-20 0 2 4 6 8 10 12 14
Moment (k-ft)

-40
-60
-80
-100
-120
-140
x (ft)

Exact moment Approx Moment element 1

Approx moment element 2 approx moment element 3

III. Discussion
The astute reader will note that we have employed cubic polynomials to approximate the
transverse displacements. Such approximations should lead to internal bending moments that
are at best linear within each element and internal shear forces that are constant within each
element. The process of using the nodal values for the resultant forces to obtain
approximations to the bending moment and internal shear force within each element is a form
of “stress averaging” that is used as a post processing step in most FEM programs.

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 27
Unfortunately such “stress averaging” can make the results of a FEM analysis appear to be
more precise than is the case since it hides the sudden jump in values for the internal forces
from element to element.

It is also important to note that in general a FEM approximation will typically overestimate the
stiffness of the structure. In the numerical example illustrated above, the maximum
displacement from the exact solution is underestimated by the FEM approximation. Although
not investigated in the current article, it should be expected that as the number of elements are
increased better approximations to the displacements, moments, and shears should be
expected. The downside to increasing the number of elements is that the number of equations
to be solved increases.

Finally, to extend the current discussion to 2D frames and 3D frames one must include the
effects of axial forces. For 3D frames, torsion is an additional force resultant that needs to be
considered. Future articles in the series will extend the current formulation to include the
effects of shear deformations, i.e. Timoshenko beams, and the effects of axial forces before
considering the application of column-beam frames.

IV. References

1. Hughes, T.J.R., “The Finite Element Method-Linear Static and Dynamic Finite Element
Analysis,” Prentice-Hall, Inc., copyright 1987.

2. Bathe, K-J, “Finite Element Procedures,” Prentice-Hall, Inc., copyright 1996.

Primera An Introduction to the Finite Element Method for Young Engineers // Part 2: 2D Beam Formulations 28