Structural Finite Element

Challenge

Are you up for a challenge? Let’s test our knowledge

of structural mechanics and finite element analysis!
Consider a cantilevered slender beam with:

• Uniform bending rigidity EI ,

• Length-to-height ratio of 10,

• Subjected to a distributed load,

• Modeled using Euler-Bernoulli finite elements.

q(x)
Ry

Mz

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 1 of 33

The Question:
Can Euler-Bernoulli finite elements provide exact an-
alytical solutions for the Euler-Bernoulli beam equa-
tions for the following quantities?
Your Task: Answer Yes or No, and explain why.

1. Nodal Displacements
2. Nodal Slopes (cross-section rotations)
3. Curvature
4. Moment
5. Shear

Hint: Think about the governing assumptions of the

Euler-Bernoulli beam theory, the polynomial degree
of the displacement approximation, and the relation-
ship between higher-order derivatives of the deflec-
tion curve.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 2 of 33

Answer to Challenge
1. Nodal Displacements: Yes, exact at nodes. (Ap-
proximate between nodes)
2. Nodal Slopes (cross-section rotations): Yes, exact
at nodes. (Approximate between nodes)
3. Curvature: No, and curvature is discontinuous at
nodes.
4. Moment: Since moment is proportional to Curva-
ture for uniform beams, then No, and discontinu-
ous (with jumps) at nodes across elements.
5. Shear: No. Since the shear is the derivative of the
moment, it also approximates and is discontinuous
at nodes.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 3 of 33

Explanations:
• The Euler-Bernoulli element with uniform flexural
rigidity EI properties gives better accuracy than
one would expect.
• The Euler-Bernoulli element uses Cubic Hermite poly-
nomials for shape function approximation.
• It can be shown mathematically that even in the
case of a distributed load, the Euler-Bernoulli finite
element gives Exact analytical displacements and
slopes at the nodes for the uniform Euler-Bernoulli
beam equilibrium equations, including the max de-
flection at the last node. This is true for any bound-
ary conditions, not just a Cantilever beam.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 4 of 33

• For example, for a uniform cantilever beam with
uniform distributed load q over the beam length (60
in. long by 2 in. high I -beam) and modeled with a
single Euler-Bernoulli beam element, the finite ele-
ment solution gives the exact analytical solution at
the last node:
3 2
   
L qL L qL
vmax = , θmax =
EI 8 EI 6

In this case, the cantilever beam is very lender (length-

to-height ratio = 60/2 = 30), making shear deforma-
tion negligible.
• Between the nodes, the element displacement and
slope solutions are approximate but will converge
to exact solutions with mesh refinement (increasing
the number of elements and nodes).

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 5 of 33

• For example, for a uniform distributed load q , the
exact deflection at points x along the beam is a 4th-
order polynomial:

q 4L 4 x
v(x) = (s − 4s3 + 6s2), s=
24EI L
Since the finite element approximation is only a cu-
bic 3rd order polynomial, a single element cannot
match the exact solution.
• However, at node points in a finite element mesh,
for example for two beams; at the node points:
start point 0, midpoint L/2, and end point L, (s =
0, 1/2, 1), in the case of a uniform beam with con-
stant EI , the finite element solution for displace-
ment and slope are exact:

v(0) = 0
!
4 3 2
qL4 17qL4
     
1 1 1
v(L/2) = · −4 +6 =
24EI 2 2 2 384EI
qL4 4 3 2

qL4
v(L) = · 1 −4·1 +6·1 =
24EI 8EI

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 6 of 33

• The curvature, moment, and shear within the el-
ement are only approximate when there is a dis-
tributed load.
• For example, for a uniform load q , the exact analyt-
ical solution for bending moment and shear force
at points x along the beam is quadratic and linear
in x, respectively

q dM
M (x) = − (L − x)2, V (x) = = q(L − x)
2 dx
Since the finite element approximation for displace-
ment is cubic, the moment is linear, and the shear is
constant. It can only approximate the exact parabolic
moment and linear shear force functions.
• Comparing the curvature, bending moment, and
shear force for the finite element approximation at
the fixed support will not be very accurate with just
one element with a distributed load. However, the
finite element solution will converge to the exact
solution everywhere for all variables, with mesh re-
finement.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 7 of 33

Proof displacements and slopes are exact
at nodes for a uniform beam
with any distributed load
• To prove the finite element solution is exact at the
finite element mesh nodes for both kinematic vari-
ables, displacement and slope (beam section rota-
tion), for a uniform beam with any distributed load
q(x), we are going to have to do (that is, enjoy do-
ing!) some mathematical analysis.
• Having a mathematical proof of the finite element
approximation for beam elements, we can avoid
running numerous numerical experiments to help
justify this conclusion.
• Numerical simulations can be used to verify these
results.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 8 of 33

Governing equilibrium equation
for a uniform beam
• The governing combined force and moment equi-
librium equation for the deflection of the centroidal
axis v(x) for Euler-Bernoulli beams is
2 2
 
d dv
EI 2 + q = 0
dx2 dx

• For a uniform beam with constant materials and

cross-section properties, then the bending rigid-
ity EI is constant, and the governing equilibrium
equation can be simplified:

d4v
EI 4 = q
dx
This is a 4th-order differential equation in the de-
pendent variable v .
• For the cantilevered beam, this equation must hold
for all positions x between the fixed end at x = 0
and the free end x = L.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 9 of 33

• The analytical solution depends on the nature of
the distributed load q(x); here, it is assumed to be
positive in the opposite direction as v .
• In the case of uniformly distributed load, q = q0
is constant. However, other distributed loads may
also be applied, for example, a linearly increasing
distributed load q(x) = q0x, or even a trigonometric
loading, such as q(x) = sin(π/2)(x/L)).

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 10 of 33

Solution for a point force load applied at
a node
• To prove the finite element solution is exact for nodal
displacement v(xi) and slope dv/dx degrees of free-
dom (DOF) evaluated at node point coordinates xi,
we consider a concentrated unit force at a node
point x = xi.
• For analysis, the concentrated unit force is modeled
as a Dirac delta function δ(x − x0) representing a
point force with unit magnitude applied at point xi.
• The Dirac function is not a function in the classical
sense but rather an operator defined by its action
on (continuous) functions. Let w(x) be continuous
on the interval between 0 and L, then we have the
sifting property:
Z
w(x)δ(x − xi) dx = w(xi)
L
resulting in the value of w at xi.
• For this special point force loading, we must solve
the governing equilibrium equation for deflection
(displacement) u(x) such that:
d4u
EI 4 = δ(x − xi)
dx
LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 11 of 33
Properties of Dirac delta functions
1. The integral of a Delta function δ(x − xi) is the unit
Step function H(x − xi).
2. The integral of the Step (Heaviside) function H(x−
xi) = hx−xii0 is equal to Macaulay bracket hx−xii.

3. The integral of bracket hx − xii is half the square

of Macaulay bracket hx − xii2.
4. The integral of the square of the power bracket is
one-third of the cube of the bracket hx − xii3.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 12 of 33

• To solve for the displacement everywhere u(x) due
to a point load at a node, we integrate the 4th-
order differential equation four times and solve for
integration constants using the boundary conditions
for the Cantilevered beam.
d3u 1
= (H(x − xi) + C1)
dx3 EI
d2u 1
= (hx − xii + C1x + C2)
dx2 EI  
du 1 1 1
= hx − xii2 + C1x2 + C2x + C3
dx EI 2 2
 
1 1 1 1
u= hx − xii3 + C1x3 + C2x2 + C3x + C4
EI 6 6 2

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 13 of 33

Boundary Conditions
• The boundary conditions for a Cantilevered beam
are:
1. At the fixed end, both the displacement and slope
(equal to beam cross-section rotation) are zero:

du
u(0) = 0, =0
dx x=0

2. At the other free end, there is no applied shear

force or moment; thus
d2u dM d3u
M (L) = EI 2 = 0, V (L) = = EI 3 =
dx x=L dx x=L dx x=L

3. u(0) = 0 means the function value is zero at x =

0, representing the fixed boundary (the clamped
left end for the Cantilevered beam).
du
4. dx = 0 means the slope (or derivative) of the
x=0
function is also zero at x = 0. This means the
clamped boundary prevents not only displace-
ment but also cross-section rotation.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 14 of 33

• Applying these four boundary conditions, we find

C1 = −1
C2 = −1 + xi
1
C3 = x2i
2
1 3
C 4 = xi
6
and the displacement u(x) ∈ V h due to the unit
point load at a node xi is a piecewise cubic function
in the kinematic set of displacement functions:
h
 
h h 2 h dv
V = v ∈ H (0, L), v (0) = 0, =0
dx x=0

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 15 of 33

• The kinematic space V h of admissible displacement
functions is a collection of smooth deflection shapes
for a beam that is clamped at x = 0.
• These deflection shapes are smooth enough to cal-
culate curvature (bending) and satisfy the fixed phys-
ical constraints of no movement and no rotation at
the boundary. It’s a mathematically structured way
to describe potential solutions to this problem.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 16 of 33

• Formally, the space V h represents a set of displace-
ment functions uh(x) that satisfy smoothness require-
ments and boundary conditions over the interval
(0, L).

− uh(x) ∈ H 2(0, L) means that uh(x) is a twice differ-

entiable function (in a generalized sense), and its
derivatives (up to the second order) are square-
integrable over the domain (0, L).
− The superscript h indicates a piecewise function
across the nodes with spacing between nodes h.
− The displacement function u due to a point load
at a node is smooth except at xi, where the ’jerk’
(derivative of curvature)

d d d2u d3u
κ(x) = 2
= 3
dx dx dx dx
is discontinuous (it suddenly changes (jumps) from
one side of the node to the other). Despite this,
at the nodes xi, we have u(x) ∈ V h.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 17 of 33

Principle of Virtual Work
• The principle of virtual work (the weak form of the
boundary value problem) for the Cantilever beam
problem with unit point load at node xi can be shown
to be: Find u such that:
Z L 2 h Z L
d δu d2u h
2
EI 2
dx = δu δ(x − xi) dx
0 dx dx 0

for all admissible virtual displacements (displace-

ment variations or weighting or test functions) δuh ∈
V h.

• Using the sifting property of the Dirac delta func-

tion, the virtual work of the unit point force is the
product of the virtual displacement at the point of
the load multiplied by the unit amplitude:
Z L
δuh δ(x − xi) dx = δuh(xi) · 1
0

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 18 of 33

Orthogonality of finite element curvature error
with respect to any admissible virtual displace-
ment
• The finite element solution to this problem is the
same with displacement u replaced by the piece-
wise cubic approximation uh ∈ V h ⊂ V and the
same loading term. Subtracting these,
Z L 2 h  2 2 h

d δu du du
2
EI 2
− 2
dx = 0
0 dx dx dx

Similarly, for the problem of finding the deflection

v(x) due to any distributed load q(x), the principle
of virtual work is,
Z L 2 h 2 Z L
d δv dv h
2
EI 2
dx = δv q dx
0 dx dx 0

for all admissible virtual displacements δv h ∈ V h.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 19 of 33

• The finite element solution to this problem with any
distributed load is the same with the exact v re-
placed by the piecewise cubic approximation v h ∈
V h ⊂ V and the same loading term. Subtracting
these,
Z L 2 h  2 2 h

d δv dv dv
2
EI 2
− 2
dx = 0
0 dx dx dx

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 20 of 33

Final Proof that the nodal displacements
are exact
• Let’s now (finally, with these preliminary results taken
care of) consider the difference between the exact
and finite element solutions at the nodes for the
problem of any nonzero (or zero) distributed loads
and represent this by the sifting property of the
Diract delta function representing a unit point load:
Z L
v(xi) − v h(xi) = (v − v h) · δ(x − xi) dx
0

• Using the principle of virtual work for the unit point

force problem, we can replace the right-hand-side
integral with,
Z L Z L 2 h
d2u

h d δv δv
(v−v )·δ(x−xi) dx = 2
− 2 EI 2 dx
0 0 dx dx dx

• Now we can use our previous result that the piece-

wise cubic solution u to the nodal point load prob-
lem implies that u ∈ V h, so that:
Z L 2  2 2 h

du d δv d δv
2
EI 2
− 2
dx = 0
0 dx dx dx

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 21 of 33

• Combining these results, we have, for a uniform
beam:
v(xi) − v h(xi) = 0
proving without a shadow of a doubt that the nodal
values of the finite element solution displacement
v h(xi) = v(xi) are exact for any distributed load case.

• Readers with access to finite element analysis (FEA)

software are encouraged to verify this conclusion
numerically with different distributed load cases and
boundary conditions.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 22 of 33

Proof that the nodal slopes are exact
• Now, to prove slopes are exact, we need to con-
sider unit (concentrated) point moments at nodes
represented by a Dipole function, which is a gener-
alized derivative of the Dirac delta function. Wow,
this is getting intense. It is a derivative of the Dirac
delta function in some generalized sense.
• The governing differential equilibrium problem for
the solution u(x) to this problem is:
d4u
EI 4 = δ,x(x − xi)
dx
where the comma subscript x refers to the gener-
alized derivative of the Dirac delta function.
• Integrating four times to find a solution and ap-
plying the boundary conditions for the Cantilever
beam leads to a solution that is smooth except at
xi, where its curvature
1 d2u
κ= = 2
ρ dx
is discontinuous (it suddenly changes (jumps) from
one side of the node to the other). Despite this, at
the nodes xi, we have u(x) ∈ V h.
LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 23 of 33
• The principle of virtual work (weak form) for this
problem of a concentrated unit point moment ap-
plied at a node is:
L L
d2δuh d2u
Z Z
EI 2 dx = δuh δ,x(x−xi) dx = δθh(xi)·1
0 dx2 dx 0

dδuh
for all δuh ∈ V h, where δθh(xi) = dx .
x=xi

• Following the same steps as for displacements,

L 2 2 h
d2u
Z 
dv dv
0= − EI 2 dx
dx2 dx2 dx
Z0 L
= (v − v h) · δ,x(x − xi) dx
0
h
= θ (xi) − θ(xi)

Giving the result that for any distributed load on the

uniform beam, the finite element nodal rotations
(slopes of deflection curve) are exact,

θh(xi) = θ(xi)

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 24 of 33

Results for Exact Nodal Displacement and
Slope for Uniform Beams
Generalizing, it can be shown that these results show-
ing nodal displacements and nodal slope (rotations)
are exact; they also apply to any boundary conditions
and loads, not just a cantilever beam.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 25 of 33

Approximation for Displacement, Slope,
Curvature, Moment, and Shear
within beam elements
• From the virtual work expression and for Hermite
cubic polynomial finite element approximation, the
curvature defined by second derivatives of displace-
ment is piecewise linear but discontinuous at nodes
across elements in a mesh of finite elements.
• Since bending moment M is directly proportional
to the curvature, the same applies.
• Furthermore, for taking a third derivative of the fi-
nite element approximation, the internal shear force
V is constant within each element and discontinu-
ous across elements.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 26 of 33

• For nonzero distributed loads, the solution for a
uniform beam is not a cubic polynomial. In this
case, the finite element solution for the kinematic
variables, displacement, and rotation (slope) within
the element (not at the nodes) for the uniform beam
is approximate, not exact.
• However, as we have shown, they will be exact at
the nodes. For a nonzero distributed load, the bend-
ing moment and shear force for the finite element
solution will be an approximation of the exact mo-
ment and shear.
• In the very special case of no distributed load, for
the cantilever beam with no other loads, the solu-
tion is trivially zero everywhere.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 27 of 33

Problem with end loads but no distributed
load
Ry M0

Mz P

• If we add end shear force and bending moment

loading to the free end of the cantilevered beam,
in the case of null distributed load, q = 0, the finite
element solution displacements, rotations (slope),
curvature, bending moment, and shear force will
be exact everywhere since the cubic Hermite shape
functions for the displacement/slope interpolation
are capable of representing exactly the analytical
solution to the uniform beam solution with no dis-
tributed loads.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 28 of 33

• For example, for a uniform cantilever beam (60 in.
long by 2 in. high I -beam) with applied end trans-
verse force P and moment M0 modeled with a sin-
gle Euler-Bernoulli beam element, the finite ele-
ment solution gives the exact analytical solution at
the last node:
3
 
L P M0/L
vmax = +
EI 3 2

and
2
 
L P
θmax = + M0/L
EI 2

• This is the reason that for beams and frames with

concentrated point moments or forces applied at
connections (not between spans), only one Euler-
Bernoulli element between spans is needed to ob-
tain exact solutions everywhere.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 29 of 33

• Generalizing, for a uniform beam with no distributed
loads, only a single Euler-Bernoulli element is needed
between nodal point forces or moments to give ex-
act solutions. We have shown this will not be true
with distributed loads. In that case, mesh refine-
ment is needed to represent kinematic quantities
of displacement and slope between nodes accu-
rately (but not at nodes where they are exact) and
internal bending moment and shear forces.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 30 of 33

Message to the Reader
• While you might not clearly understand all the steps
in this proof, I hope you will have gained some ap-
preciation for how learning the finite element method
(FEM) and its analysis is powerful and gives insights
into finite element solutions for applications using
software for finite element analysis (FEA).
• Unlike some other design procedures, the Finite El-
ement Method (FEM) is not just about formulas.
The FEM is a mathematical procedure based on
the principle of virtual work (equivalent to the weak
form for structures) formulated with consistent piece-
wise approximation over a mesh of non-overlapping
connected elements to provide the best approxi-
mation possible and to guarantee convergence to
exact solutions with mesh refinement.

Thanks for reading to the end!

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 31 of 33

Reference
The Finite Element Method: Linear Static and Dy-
namic Finite Element Analysis, by Thomas J.R. Hughes,
1987, Reprinted by Dover in 2000. Section 1.10, Ex-
ercise 1.16.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 32 of 33

We love the FEM and its
application in FEA.

LinkedIn � - Dr. Lonny Thompson, Clemson University, Dec. 11, 2024. 33 of 33