A Simple Explanation of the

Finite Element Method for

Stress Analysis
Dr. Lonny L. Thompson
March 23 2025

Finite Element Stress Analysis of a Spreader Bar showing Inter-

polated Von-Mises Stress Contours.

Image: Innovex Engineering

https://www.innovexeng.com/engineering-services/fea-stress-analysis-services

Swipe for the next page

Contents

1 A Simple Explanation of the Finite Element Method

for Stress Analysis 2
1.1 External Virtual Work, Wext . . . . . . . . . 3
1.2 Internal Virtual Work, Wint . . . . . . . . . . 4
1.3 Structure and Form of the Element Stiffness
Matrix . . . . . . . . . . . . . . . . . . . . . 5
1.4 Properties of Element Stiffness Matrix . . . 12
1.5 Linear Displacement Triangle Element . . 13
1.6 4-Node Quadrilateral (QUAD4) Element 16
1.7 Linear Displacement Tetrahedral Element . 23
1.8 Application of the Principle of Virtual Work
in the Finite Element Method . . . . . . . . 27
1.9 Solving the Nodal Force Balance Equations 28
1.10 Assembly of Global Stiffness Matrix from El-
ement Contributions . . . . . . . . . . . . . 30
1.11 Application of Displacement Constraints with
Matrix Partitioning . . . . . . . . . . . . . . 31
1.12 Global Nodal Basis for Mesh Nodes . . . . 33
1.13 Sparsity of the Assembled Finite Element
Stiffness Matrix . . . . . . . . . . . . . . . . 35
1.14 Calculation of Element Stresses . . . . . . . 38
1.15 Summary of the PVW for Deriving the Ele-
ment Stiffness Matrix and Solution for Nodal
Displacements . . . . . . . . . . . . . . . . 42

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 1 of 46

1 A Simple Explanation of the Finite El-
ement Method for Stress Analysis
• The Principle of Virtual Work (PVW) states that, for a
system in equilibrium, the virtual work done by ex-
ternal forces must equal the internal virtual work.
• We derive the form of the stiffness matrix and nodal
force equilibrium equations for stress analysis by
defining both external and internal virtual work ex-
pressions.
• This is not a theory that needs to be disproven;
this is the basis for the finite element method (FEM)
and its application in finite element analysis (FEA)
in stress analysis.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 2 of 46

1.1 External Virtual Work, Wext
• The external virtual work is the work done by exter-
nal nodal forces {Fext} acting on the nodes of the
element, resulting from virtual nodal displacements
{δunod}:
Wext = {δunod}T {Fext}
 

 


 


 


 


 


 


 

 
F

 


ext 


δuTnod
 
 
Wext = 






 


 


 


 


 


 


 


 

 

• Recall work is the product of displacements with

forces.
• The superscript T represents the transpose opera-
tion of a column vector, resulting in a row vector.
• Multiplying the row vector of virtual nodal displace-
ments with the column vector of external nodal forces
gives the dot-product (inner-product) of the two vec-
tors, resulting in a scalar, as it should, since work is
a scalar.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 3 of 46

• If the units of displacement are meters and force in
Newtons, then the product is J = N-m, the unit of
work and energy.

1.2 Internal Virtual Work, Wint

• The internal virtual work within the element is due
to virtual strains {δε} from virtual displacements.
• Using the stress-strain relationship {σ} = [E]{ε},
where [E] is the elasticity matrix, we write:
T T
Z Z
Wint = V {δε} {σ} dV = V {δε} [E]{ε} dV

 

 


 

 

δεT
Z   
σ  dV
 
Wint = V 



 


 

  

 

  

 

δεT
Z    
E ε  dV
 
=

 
V 




 

 

 

• Note the elasticity matrix [E] has units of Young’s

modulus, same as stress; for example, Pa = N/m2.
• After integrating over the volume, the result is units
of work (internal energy) of J = N-m.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 4 of 46

1.3 Structure and Form of the Element
Stiffness Matrix
• Since virtual strain {δε} relates to virtual displace-
ments {δunod} through the strain-displacement ma-
trix [B] with units of one over length unit, we have:

{δε} = [B]{δunod}
 

 


 


 


 


 

    
 

  
 

   
δunod

 
   
 


 
   
 

ε  = B
     
 
    

   
 


 
   
 


 
 
 


 


 


 


 


 


 

 

• The matrix [B] is defined with derivatives of the di-

mensionless shape functions [N ] used to approxi-
mate the displacement field at points within an el-
ement in terms of nodal displacements.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 5 of 46

• For a 3-node triangle element, the displacement
approximation is expressed as a linear interpolation
of unknown nodal displacements,

u(x, y) = N1e(x, y) u1 + N2e(x, y) u2 + N3e(x, y) u3

u3y

u3x

u1y

u1x u2y

u2x
Triangle element with two displacement DOF (uix, uiy ) at
each of the three nodes.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 6 of 46

• In component-matrix form with nodal DOF (uix, uiy ):
 
N1u1x + N2u2x + N3u
 
ux
 
3x 
  
 
   
{u} =   = 
   

uy  N1 u1y + N2 u2y + N3 u3y 






 







u1x







u

 

1y

 


 

  

  u2x 

 


 

= = [N ]{unod}

 
u
 
2y 

 


 

 

u

 

 
3x

 


 


 

 u

 

3y 
 

• With nodal position coordinates at each of the three

nodes, (xi, yi), the local element shape functions
are linear in x and y:

Ni(x) = ai + bix + ciy, i = 1, 2, 3.

• The coefficients ai, bi, ci are computed using the stan-

dard area coordinate interpolation formulas:
x k y ` − x` y k yk − y` x ` − xk
ai = , bi = , ci = ,
2A 2A 2A
where:
(xk , yk ) and (x`, y`) are the coordinates of the other
two nodes in element, excluding node i,
LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 7 of 46
 A is the area of triange element:
1
A = |x1(y2 − y3) + x2(y3 − y1) + x3(y1 − y2)| .
2
• The local element nodal shape functions Ni(x) sat-
isfy the interpolation property that they evaluate to
one at their node and are zero at the other two
nodes (and zero on the back edge between these
two other nodes). This property is summarized by
the Kronecker delta function:

Ni(xj ) = δij .

where 

1, if i = j,





δij = 
0,
 if i 6= j.
This ensures that at node i, the shape function as-
sociated with node j evaluates to 1 if i = j and 0
otherwise, providing the essential partition of unity
property in finite element interpolation.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 8 of 46

Example of Linear shape function N1(x, y) with unit value
at node 1, used to interpolate nodal displacements.

• Substituting this together with the stress

{σ} = [E]{ε} = [E][B]{unod}

into the internal virtual work due to the product of

stress and virtual strain gives:

Wint = V {δunod}T [B]T [E][B] {unod} dV

Z

= {δunod} V [B]T [E][B] {unod} dV

T
Z

T T
Z 
= {δunod} V [B] [E][B] dV {unod}
| {z }
[Ke ]
= {δunod}T [Ke] {unod}
= {δunod}T {Fint}

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 9 of 46

 where we have identified the element stiffness ma-
trix
[Ke] = V [B]T [E][B] dV
Z

and

{Fint} , V [B]T {σ} dV

Z

= V [B]T [E]{ε} dV
Z

= V [B]T [E][B] {unod} dV

Z

= ( V [B]T [E][B] dV ) {unod}

Z

= [Ke] {unod}

represents the nodal internal forces as the product

of the element stiffness [Ke] with nodal displace-
ments {unod}.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 10 of 46

 Element Internal Nodal Forces and Stiffness Ma-
trix
T
Z
{Fint} = [Ke]{unod}, [Ke] = V [B] [E][B] dV
    

 
 
 


 
   


 
   


 
   


 
   


 
   


 
   

   
F Ke unod

 
   
 
int 

 
   
   







= 











 
   


 
   


 
   


 
   


 
   


 
   


 
   


 
 
 

   

• Thus, we have shown that the internal virtual work

for the element can be expressed in alternative forms:
T T
Z
Wint = V {δε} {σ} dV = {δu nod } {Fint}

or
T T
Z
Wint = V {δε} [E]{ε} dV = {δu nod } [Ke] {unod}

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 11 of 46

1.4 Properties of Element Stiffness Ma-
trix
Dimensions of the element stiffness matrix
• The dimensions of the element stiffness matrix [Ke]
depend on the number of nodes of the element
and the number of nodal displacement components,
also called degrees of freedom (DOF) per node.

Dimension Element Nodes dof/node degrees-of-freedom

1D Bar 2 1 2 ×1 = 2
2D Triangle 3 2 3 ×2 = 6
2D Quadrilateral 4 2 4 ×2 = 8
3D Tetrahedral 4 3 4 ×3 = 12
3D Hexahedral 8 3 8 ×3 = 24

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 12 of 46

1.5 Linear Displacement Triangle Element

• For the Linear Displacement, Constant Strain Trian-

gle (CST) Element, the number of nodes is three,
and there are two perpendicular displacement com-
ponents per node, resulting in 3 × 2 = 6 nodal dis-
placement components for the element.

u3y

u3x

u1y

u1x u2y

u2x
Constant Strain Triangle (CST) element showing two plane
displacement components (uix, uiy ) at each of the three
nodes

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 13 of 46

• For the plane stress CST element, the strain-displacement
matrix [B] has dimensions of three rows for the three
independent strain components (two perpendicu-
lar normal strains and in-plane shear strain) and six
columns corresponding to the six nodal displace-
ment components (DOF) for the element.
 




u1x 




∂ux
  
 

u1y
  
 

   
∂x
       
εxx

 
 
 


 
 
 
 
 

u2x

 
 
 
   
 

∂uy

 
 
 
   
 

{ε} =  εyy  =  B
     
=
 
 
∂y   
u 
2y

   
 
2ε

 
 
 
  
 

xy ∂ux ∂uy 
 
 
  
 

+ u

 
 
 

   
3x
   

∂y ∂x  









 u

 

3y 
 

• For plane stress, the elasticity matrix [E] in {σ} =

[E]{ε} is a symmetric square matrix, [E]T = [E], with
three rows and three columns.
• For isotropic materials:
    
σ 
 
xx  1 ν 0 ε 
 
xx 
E

 
  


 
   

{σ} =  σyy  = ν 1 εyy 
   
0

 
(1 − ν 2 ) 


1
−

σ 0 0 (1 ν) γ

 
 
 

xy xy
 
 
 

2

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 14 of 46

• Multiplying the transpose of the matrix [B]T with
dimensions of 6 rows and three columns after the
transpose operation, with the three-by-three elas-
ticity matrix [E], and the strain-displacement matrix
[B], the product of the three matrices [B]T [E][B] re-
sults in a square matrix with six rows and six columns.
• After integrating over the volume (area times thick-
ness for plain stress), the stiffness matrix [Ke] has
dimensions of 6 × 6 corresponding to the 6 DOF
for the triangle stress element.
[B]T [E] [B] dV
Z
[Ke] = V
6×6 (6×3) (3×3)(3×6)
 
 
 
 
    
 
 
BT
     
Z      
E B
 
= dV
     
   
V
     
     
     
 
 
 
 
 
 
 

 
 
 
 
 
 
 
Ke
 
 
 
= 







 
 
 
 
 
 
 

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 15 of 46

1.6 4-Node Quadrilateral (QUAD4) Ele-
ment
• For the planar 4-node quadrilateral element, the
number of nodes is four. Since there are two per-
pendicular displacement components, also called
Degrees of Freedom (DOF) per node, this element
has 4 × 2 = 8 nodal DOF.

u3y
u4y
u3x
u4x

u1y
u2y
y
u1x
u2x
x
Quadrilateral 4-node element (QUAD4) showing two
plane displacement components at each of the nodes.

• For plane stress Finite Element Analysis (FEA) using

this element, the matrix [B] relating the strain ap-
proximation to nodal DOF has dimensions of three

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 16 of 46

 rows and eight columns:
 
u1x












u1y

 


 


 

 
∂ux
  
 

u2x
  
 

   
∂x
       
εxx

 
 
 


 
 
 
 
 

u2y

 
 
 
   
 

∂uy

 
 
 
   
 

{ε} =  εyy  =  B
     
=
 
 
∂y   
u 
3x

   
 
2ε

 
 
 
  
 

xy ∂ux ∂uy 
 
 
  
 

+ u

 
 
 

   
 
∂y ∂x

 



 3y 




 
u

 

4x

 


 


 

 u

 

 
4y 

• For a 4-node quadrilateral element, the bilinear shape

functions are defined in terms of local dimension-
less coordinates ξ and η to facilitate numerical in-
tegration of the stiffness matrix:
1 1
N1 = (1 − ξ)(1 − η), N2 = (1 + ξ)(1 − η)
4 4
1 1
N4 = (1 − ξ)(1 + η), N3 = (1 + ξ)(1 + η)
4 4

• The stiffness matrix [Ke] for this element has dimen-

sions of 8 × 8 corresponding to the 8 DOF for the
quadrilateral stress element.

[B]T [E] [B] dV

Z
[Ke] = V
8×8 (8×3) (3×3)(3×8)

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 17 of 46

Bilinear shape (interpolation) function N1(ξ, η) with unit
value at node 1, for 4-node Quadrilateral (QUAD4) ele-
ment

• The double integration of the complicated quadri-

lateral shape is performed using mapping and chang-
ing variables.
Z
1 Z1 T
Ke = −1 −1 B EB det(J ) dξdη.

where det(J ) is the determinant of a Jacobian ma-

trix transformation of derivatives.
• The determinant of the Jacobian mapping matrix
has units of square of length [L]2 because it scales
the physical differential area dA = dxdy from a di-
mensionless dξdη, with dA = det(J )dξdη.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 18 of 46

 η
3 (x3 , y3 )

4 (x, y) ← (ξ , η )
(x4 , y4 )

ξ η
y
4 3
(−1,1) (1,1)
1
(x1 , y1 )
2
(x2 , y2 )
ξ
x
1 2
(−1,−1) (1,−1)

Dimensionless points (ξ, η) in the reference bilinear

square, are mapped to physical points (x, y) in the quadri-
lateral .

• For the special quadrilateral geometry of a rectan-

gle, the determinant is a constant equal to the ratio
of the rectangle area A divided by the area of a ref-
erence biunit square equal to 2 × 2 = 4.
• For quadrilateral-shaped finite elements, numerical
integration with Gaussian (Gauss-Legendre) quadra-
ture is used to evaluate the stiffness matrix.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 19 of 46

• For stress analysis, 4-node quadrilateral elements
are less stiff than 3-node triangle elements and are
preferred.
• The sweet spot between accuracy and computa-
tional efficiency for finite element stress analysis is
often the use of quadrilateral and triangle elements
with quadratic (not linear) displacement approxi-
mation.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 20 of 46

Symmetry of Element Stiffness Matrix
• The stiffness matrix is symmetric because the elas-
ticity matrix [E] is symmetric.
• Since [E]T = [E], then

[Ke] = V ([B]T [E][B])T dV

T
Z

= V [B]T [E]T [B] dV

Z

= V [B]T [E][B] dV = [Ke]

Z

Since the stiffness matrix is symmetric, interchang-

ing the rows and columns does not change the ma-
trix.
Kij = Kji
 
 
 
 
 
 
 
 
 

[Ke]T = [Ke] =
 
 
 
 
 
 
 
 
 
 
 
 

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 21 of 46

Units of Stiffness Matrix
• The units of the stiffness matrix are the units of the
matrix products [B]T [E][B] integrated over the vol-
ume.
The units of [B] and its transpose [B]T are one
over length, for example m−1.
The units of [E] are N/m2, multiplying, the units
of the products [B]T [E][B] we have N/m4.
After integrating over the volume with units m3,
the values in the stiffness matrix [Ke] have units:

(N/m4)(m3) = N/m,

- the units of a linear spring as expected.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 22 of 46

1.7 Linear Displacement Tetrahedral El-
ement
• For the Linear Displacement Tetrahedral Stress Ele-
ment (pyramid-shaped volume with four triangular
faces) for 3D applications, there are four nodes at
the vertices, with three perpendicular displacement
components per node, resulting in 4 × 3 = 12 nodal
displacement components (DOF) for the element.

u4z

u4y
u4x
u3z
u1z
u3y
u2z u3x
u1y
u1x
u2y
u2x
Tetrahedon 3D solid element showing three perpendicu-
lar displacement components at each of the four nodes

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 23 of 46

• The strain-displacement matrix [B] has dimensions
of six rows for the six independent strain compo-
nents (three perpendicular normal strains and three
shear strains) and twelve columns corresponding to
the twelve nodal displacement components (DOF)
for the element.
 
u1x 











u1y

 


 


 

∂u
  
 

x  
u1z

 
 
 

∂x

 
 
 


 
 
 

  
 
 
 

ε u2x
   
∂uy
  
 
 
 

xx

 
 
 
 
 


 
 
 
 
 

∂y

 
 
 
 
 

ε u2y

 
 
 
 
 

yy

 
 
 
 
 

     








 ∂u z









 εzz  u2z

 
 
 
 
 

∂z

  
 
 
 

{ε} = 
    
= ∂uy
= [B]
 2ε yz








+ ∂uz 

 6×12



 u3x




∂z ∂y

 
 
 
 
 

     
2ε u

 
 
 
 
 

     

 xz 
 

∂u ∂u

 
 3y 

+
     








 x z 








2ε ∂z ∂x u

 
 
 
 
 

xy 3z
 
 
 
 
 


 
 
 

   
∂u
   

∂u y
 
u

x
+

 
 
 




∂y ∂x








4x 




u

 

 
4y

 


 


 

 u

 

4z 
 

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 24 of 46

• The elasticity matrix [E] in {σ} = [E]{ε} is a square
symmetric matrix with six rows and six columns.
• For isotropic materials:
    
σ 







xx 





λ + 2G

λ λ 0 0 0







ε 
xx 





σ λ λ + 2G λ 0 0 0 ε

 
  
 
yy  yy 

  
 

 
  

 
 




σ λ λ λ + 2G 0 0 0 ε

 
  
   

zz  zz 
   
{σ} =   = 
  






σyz   0 0 0 G 0 0







γyz 
σ 0 0 0 0 G 0 γxz 

 
   
  





xz 












σ G γxy 
   
0 0 0 0 0

 
  
 
xy 
  


where
Eν E
λ= , G=
(1 − 2ν)(1 + ν) 2(1 + ν)

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 25 of 46

• Multiplying the transpose of the matrix [B]T with di-
mensions of 12 rows and 6 columns after the trans-
pose operation, with the six-by-six elasticity ma-
trix [E], and the strain-displacement matrix [B], the
product of the three matrices [B]T [E][B] results in a
square matrix with 12 rows and 12 columns.

[B]T [E] [B] dV

Z
[Ke] = V
12×12 (12×6)(6×6)(6×12)

• After integrating over the volume, the stiffness ma-

trix [Ke] has dimensions of 12 × 12 corresponding to
the 12 DOF for the tetrahedron stress element.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 26 of 46

1.8 Application of the Principle of Vir-
tual Work in the Finite Element Method
• According to the PVW, we equate internal and ex-
ternal virtual works:

Wext = Wint

• Equating, we obtain:

{δunod}T {Fext} = {δunod}T {Fint}

that is,

{δunod}T {Fext} = {δunod}T [Ke] {unod}

• Since this holds for any arbitrary virtual displace-

ment {δunod}, subject to any displacement constraints,
we conclude:
{Fext} = {Fint}
that is,
{Fext} = [Ke] {unod}

• These equations represent force equilibrium at each

node, balancing external with internal forces.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 27 of 46

1.9 Solving the Nodal Force Balance Equa-
tions
• After applying nodal displacement constraints to
reduce the number of equations, we solve for the
remaining unknown nodal displacements {ūnod}, in
terms of applied external nodal forces combined
with forces due to stiffness multiplied by any spec-
ified nodal displacements, denoted {F̄ext}.
• After doing this, the reduced system of equations
can be expressed as,

[K̄e] {ūnod} = {F̄ext}

• In abstract form, we can solve these equations for

nodal displacements using the matrix inverse,

{ūnod} = [K̄e]−1 {F̄ext}

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 28 of 46

 w classes of elements

Practical
The lowest Considerations
order element on parallelograms
Mesh of Connected Finite Elements
imes dividing a domain into triangles is not the best idea. Some domains,
gles, are•much better subdivided
In practice, into smaller
to obtain more accuraterectangles. Alsomore
solutions, sometimes t
become really
than messy. For instance
one element Figure
is used in a3.1 shows
finite a typical
element triangular g
mesh
gle as produced by the PDE
of connected Toolbox of Matlab. Because of the way these t
elements.
are produced, working from the boundary to the interior and avoiding ver
, they display a very disorganized
• In general, the promiseand non–symmetric
is that the accuracypattern.
of theIf your p
directions,solution
maybe this
willisimprove
not the best
by way to begin your
subdividing the discretization.
part vol-
ume into many smaller elements.

Figure 3.1: A typical triangular grid of a rectangle.

are going to introduce here finite elements on rectangles and parallelograms

nts share many features with finite elements on triangles, but there are pl
es. To learn
LinkedIn about finite elements on arbitrary quadrilaterals (trapezes
� - Dr. Lonny Thompson, Clemson University, March 23, 2025. 29 of 46
and
1.10 Assembly of Global Stiffness Ma-
trix from Element Contributions
• In practice, the global stiffness matrix [K] for all
nodes in the finite element mesh is determined by
assembling local element matrices [Ke] in appro-
priate rows and columns, accounting for nodal dis-
placement compatibility between connected ele-
ments, using information from node and displace-
ment component numbers connected with each el-
ement.

⎡ ⎤
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
[K ] = ⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ ! ⎥
⎢ ⎥
⎣ ⎦

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 30 of 46

 1.11 Application of Displacement Con-
straints with Matrix Partitioning

• After applying essential nodal displacement con-

straints, for example, u1 = 0 for fixed or nonzero
values like 1 mm for prescribed, we solve the re-
maining unknown nodal displacements {ūnod}, in
terms of applied external nodal forces combined
with forces due to stiffness multiplied by the spec-
ified nodal displacements.
    








u1 











R1 





   
u2 F2
    
  
 
 

  
 
 

  
 
 

   
K u3 F3
    
  
 
 

  
 
 












=  


   
.. ..
    
  
 
 

  
 
 

  
 
 

 










    
 











 
 
 


LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 31 of 46

 [K̄] {ūnod} = {F̄ext}
      








u 
2










F 
2
















     
u F
      
  
 
 
 
 






 3






 3











     
− 
      

 K̄ 
 =  u1
.. ..
    
  
 
 

  
 
 
 
 

  
 
 
 
 

  
 
 
 
 

  
 
 
 
 

 

















 
 
 
 
 


• Since u1 is prescribed, the corresponding force R1

is a reaction.
• For solutions to exist and be unique, the degrees-
of-freedom in the nodal constraints u1 must be spec-
ified such that there are no rigid-body (zero non-
deformation) motions, both rigid-body translation
and rotation.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 32 of 46

1.12 Global Nodal Basis for Mesh Nodes
• When describing the finite element method (FEM),
we often emphasize the role of elements and their
connections in a mesh; however, the big-picture is
driven by the nodes and degrees of freedom con-
necting these elements.
• Each row in the global stiffness matrix represents
the equilibrium equation and stiffness relation asso-
ciated with that degree of freedom (DOF), coupling
all other nodes that are influenced by this node
through the support of the basis (shape) functions
associated with this DOF.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 33 of 46

• The global basis (shape) function ϕi of a node in the
mesh is constructed from the local shape functions
Ni associated with the elements that connect at this
node.
• The global basis (shape) functions used in the fi-
nite element method (FEM) have compact support,
meaning they influence only a local patch of ele-
ments and their connected nodes and DOF.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 34 of 46

1.13 Sparsity of the Assembled Finite El-
ement Stiffness Matrix
• The assembled finite element global stiffness ma-
trix (or mass matrix) is often filled with many ze-
ros—hence, it is called a sparse matrix.
• The global stiffness matrix [K] is sparse, meaning
most of its entries are zero.

Sparsity Pattern of Nonzero values for the Stiffness Ma-

trix for 2D Plane Elasticity with 2-DOF/node, (ux, uy ) with
a 4 × 4 square mesh of 4-node Quadrilateral (QUAD4) El-
ements.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 35 of 46

• This sparsity arises because each finite element only
interacts with a small subset of the total degrees of
freedom in the system.
• The number of nonzero entries in any row corre-
sponds to the nodes that interact with a given node
in the mesh. This interaction is dictated by the local
support of the nodal basis function and the element
connectivity.
• Knowing the sparsity counts of nonzero values and
patterns in advance helps optimize storage and solver
performance.
• In practice, this can make the difference between
waiting 24 hours or 1 hour for a solver job to com-
plete and whether you can run this job on your lap-
top or need a more powerful computer.
• Sparse direct solvers, iterative equation-solving meth-
ods, and preconditioners all benefit from this infor-
mation.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 36 of 46

Equation Solving Methods
• In practice, the inverse [K̄]−1 is never taken to solve
for unknown nodal DOF since it would be a very
computationally expensive operation for a large num-
ber of equations in a finite element mesh.
Instead, the equations are solved using direct al-
gorithms such as factorizations or other general-
izations of Gaussian elimination learned in linear
algebra or for very large systems of equations,
for example, one-million equations, to reduce the
computer memory requirements, iterative solvers
such as the Conjugate-Gradient method are more
efficient.

Illustration of convergence to a solution x from starting

estimate x0 using iterative equation solving process.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 37 of 46

1.14 Calculation of Element Stresses
• After the nodal displacements are solved from the
nodal force balance (stiffness equations), we can
distribute nodal displacements to each element and
calculate stress within each element using the elas-
ticity matrix [E] and the linear elastic relationship
between stress and strain:

{σ} = [E]{ε} = [E][B]{unod}

 

 


 


 


 


 

     
 

  
 

   
δunod

 
     
 


 
     
 

σ  = E B
       
   
      

     
 


 
     
 


 
 
 


 


 


 


 


 


 

 

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 38 of 46

1.14.1 Stress Recovery
• The displacement approximation is piecewise lin-
ear and continuous across shared edges of adja-
cent elements in the finite element mesh.
• However, since the strains are defined in terms of
derivatives of displacements, these are discontinu-
ous between element boundaries.
• Since the stress is computed from strains through
the material constitutive equations, the stresses are
also discontinuous between elements and at nodes.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 39 of 46

• Once the stress components are determined within
each element, additional calculations to determine
Principal Normal Stresses, Von-Mises stress, and other
criteria are used for yield and failure analysis.

y ωx = 0.08424
ωy = 0.16328
εxy = 0.11424
(ωy , →εxy ) ωavg = 0.12376
εmax = 0.12086
ωp1 = 0.24462
ωp2 = 0.00288
(ωp2 , 0) (ωavg , 0) (ωp1 , 0) 2ϑp = 109.08→
+ω 2ϑs = 19.08→
O
2ϑp
R 2ϑs

(ωx , εxy )
(εmax , ωavg )
x
+ε

Mohr’s circle graphically illustrates the state of plane

stress, principle normal stresses, and maximum in-plane
shear stress.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 40 of 46

• The stresses within each element are then interpo-
lated/extrapolated or smoothed to recover more
accurate nodal stresses.
• A widely used technique is patch recovery, taking
the stress values at Gauss integration points with
each element, interpolating them across neighbor-
ing elements in a patch, and then smooths or aver-
aging these values at the nodes.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 41 of 46

1.15 Summary of the PVW for Deriving
the Element Stiffness Matrix and So-
lution for Nodal Displacements
• In deriving the stiffness matrix for finite elements,
we apply the Principle of Virtual Work (PVW).
The PVW states that at equilibrium, the virtual work
done by external forces must balance the virtual
work of internal forces.
• We proceed by defining the internal nodal forces
from the matrix product of the element stiffness
matrix with nodal displacements and show the in-
ternal virtual work for the stress element is
T T
Z
Wint = V {δε} {σ} dV = {δu nod } {Fint}

from which we deduce the element stiffness matrix

[Ke]:
T
[E]{ε} dV = {δunod}T [Ke] {unod}
Z
Wint = V {δε}

• We then apply the PVW to determine the nodal

equilibrium equations, which balance external nodal
forces with internal nodal forces defined by the prod-
uct of the stiffness matrix and nodal displacements.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 42 of 46

1.15.1 Internal Nodal Forces, {Fint}
The internal nodal forces {Fint} are defined by the
product of the element stiffness matrix [Ke] and the
nodal displacements {unod}:
T
Z
{Fint} = [Ke]{unod}, [Ke] = V [B] [E][B] dV

1.15.2 External Virtual Work, Wext

The virtual work done by the external forces {Fext}
acting on the nodes of the element due to virtual dis-
placements {δunod} is:

Wext = {δunod}T {Fext}

1.15.3 Internal Virtual Work, Wint

The virtual work done by the internal forces is simi-
larly given by:

Wint = {δunod}T {Fint} = {δunod}T [Ke]{unod}

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 43 of 46

1.15.4 Applying the Principle of Virtual Work
• According to the Principle of Virtual Work, the ex-
ternal virtual work equals the internal virtual work:

Wext = Wint

• Expanding both sides, we have:

{δunod}T {Fext} = {δunod}T [Ke]{unod}

• Since this equality holds for any arbitrary virtual dis-

placement {δunod}, it follows that:

{Fext} = [Ke]{unod}

subject to constraints on nodal virtual displacements.

• These equations represent nodal force equilibrium,
balancing external with internal forces.

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 44 of 46

Solution of Nodal Force Balance Equations
• This relationship confirms that after assembly of el-
ement stiffness matrix contributions in a finite el-
ement mesh of the part, and nodal displacement
constraints are applied,
[K̄]{ūnod} = {F̄ext}
the reduced stiffness matrix [K̄] represents the map-
ping between nodal displacements and external nodal
forces in equilibrium adjusted by the product of
stiffness and nodal constraints, represented by {F̄ext}.
• By expressing the internal virtual work for the ele-
ment as the product of virtual displacements with
the internal nodal forces equating to {Fint} = [Ke]{unod},
applying the Principle of Virtual Work with nodal
displacement constraints, accounting for element-
node connectivity in an assembly of finite elements,
we obtain the system of nodal equilibrium equa-
tions with the solution for nodal displacement
{ūnod} = [K̄]−1 {F̄ext}
After distributing nodal displacements to each con-
nected element in the part mesh, stress is calcu-
lated from the linear elastic equation:
{σ} = [E]{ε} = [E][B]{unod}
LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 45 of 46
We Love the
Finite Element Method (FEM)
and its application in Finite Element Anal-
ysis (FEA)!

Surface models

Part Geometry Simplify (Feature Suppression) Thin features

Wireframe models

Fillets

Assemblies

Young’s modulus

Material Properties Poisson’s ratio

Mass density

Concentrated Forces

Distributed
Loads
FE Model Pressure

Gravity

Dr. Lonny fixed

Thompson pin
Boundary conditions at supports
hinge

roller

Rigid Constraints

Contact Surfaces
Requirements Mates, Joints
Constraints
Analysis Type Interfaces
Objectives
Problem Statement Fastener, Bolt
Assumptions Connections
Weld
Limitations Bonded
Document
Recommendations Subdivision
Design of Experiments
Node coordinates
Mesh Generation
Sensitivity Analysis Element-node connectivity
Changes in key parameters Bar, Truss
Changes in supports Optimization 1D Line (Wireframe)
Discretization Beam, Frame
Geometry
Linear FEA for CAE Cross-Sections, Profile Shapes
Materials
Engineers Triangle
Displacement, Distortion Quadrilateral
2D Surface
Elements
Deformed Shape Visualization Plane Stress/Strain
Stress Contours Applications Axisymmetric (Revolve)
Stress concentrations Thin Shell Surface models
Interpret Results
Rapid changes in stress
3D Solid Tetrahedron
Orders of magnitude with scale Hexahedron
formulas
Linear
Simple models Shape functions
Validation Quadratic
Experiments
Post-Process Element
Prototypes FEM Numerical integration
Stiffness Matrix Assemble global
Coincident nodes

Element connectivity gaps Discrete Equilibrium Equations

Element distortion limits Checks Apply boundary conditions

Remeshing Enforce constraints

Solve
Reaction Forces Nodal Degrees-of-Freedom

Experiments Verification Yield Criteria

Check Strength
Failure Criteria

Strain
Recovery
Stress

Reaction Forces

LinkedIn � - Dr. Lonny Thompson, Clemson University, March 23, 2025. 46 of 46