MKM321 Continuum and computational structural

mechanics notes
Prof. Schalk Kok
October 2013
1 Introduction
In these notes Ill review some basic concepts from continuum mechanics, such as the
traction vector, stress and strain tensors and the equilibrium conditions. Then Ill pose
the basic form of the mixed boundary value problem of solid mechanics. Only at this
point is it appropriate to start learning about the nite element method.
2 The traction vector
Consider a general solid body B, depicted in Figure 1. The surface of the body B is
denoted B. Imagine a smooth surface area S on B, with outward unit normal vector
n through point P. If the force transmitted from one side of S to the other is f, the
traction vector at P is dened as
t
(n)
= lim
S0
f
S
(1)
Note that the traction vector denition implicitly requires two distinct directions: the
orientation of the surface S, dened through the outward unit normal n, and the direc-
tion of the force f. The notation employed here directly indicates the dependency on
the direction n by using it as a superscript.
You rst introduction to stress, somewhere in your second year strength of materials
course, probably relied on the above concept of the traction vector to explain stress. You
learned that normal stress is a stress that exists in a structure if the force acting upon
it acts perpendicular to the exposed surface. Later on you learned about shear stresses,
which exist in a structure if the force acting on the structure acts parallel to the exposed
surface. These concepts about stress was necessarily vague (and imprecise), because you
lacked the required mathematical background to treat the concept of stress rigorously. I
will now illustrate the subtle but important dierence between a traction vector and the
stress tensor.
1
P
B
B
S
n
f
Figure 1: The traction vector
3 The stress tensor
We can make particular choices for n, such as the unit vectors e
1
, e
2
and e
3
along the
Cartesian axes. In this case, we can write the resulting traction vectors as
t
( e
1
)
= t
( e
1
)
1
e
1
+ t
( e
1
)
2
e
2
+ t
( e
1
)
3
e
3
(2)
t
( e
2
)
= t
( e
2
)
1
e
1
+ t
( e
2
)
2
e
2
+ t
( e
2
)
3
e
3
(3)
t
( e
3
)
= t
( e
3
)
1
e
1
+ t
( e
3
)
2
e
2
+ t
( e
3
)
3
e
3
(4)
Eqs.(2)(4) can be expressed much simpler by introducing the second order stress tensor
, with components
ij
. Note the use of two distinct subscripts: the subscript i indicates
which plane is perpendicular to this stress component, while the subscript j indicates in
which direction this stress component acts.
Any vector a can be expressed as the sum of its components i.e.
a = a
1
e
1
+ a
2
e
3
+ a
3
e
3
=
3

i=1
a
i
e
i
. (5)
Similarly, any second order tensor A can be expressed as
A =A
11
e
1
e
1
+ A
12
e
1
e
2
+ A
13
e
1
e
3
+
A
21
e
2
e
1
+ A
22
e
2
e
2
+ A
23
e
2
e
3
+
A
31
e
3
e
1
+ A
32
e
3
e
2
+ A
33
e
3
e
3
=
3

i=1
3

j=1
A
ij
e
i
e
j
,
(6)
where the mathematical operator is known as the dyadic product. The dyadic product
operator is dened as follows: Given three vectors a, b and c,
(a b)c = a(b c). (7)
2
At this point it is convenient to introduce Einsteins summation convention. This con-
vention implies summation over any repeated index e.g.
c = a b = a
T
b = c =
3

i=1
a
i
b
i
= c = a
i
b
i
(8)
c = Ab = c
i
=
3

j=1
A
ij
b
j
= c
i
= A
ij
b
j
(9)
Using the denition of the stress tensor components, Eqs.(2)(4) are rewritten as
t
( e
1
)
=
11
e
1
+
12
e
2
+
13
e
3
(10)
t
( e
2
)
=
21
e
1
+
22
e
2
+
23
e
3
(11)
t
( e
3
)
=
31
e
1
+
32
e
2
+
33
e
3
. (12)
The stress components are graphically depicted in Figure 2.
e
1
e
2
e
3

33

31

31

32

11

13

22

23

21

22

21

13

11

12

33

23

12

32
Figure 2: Stress components in a Cartesian reference frame
3.1 Relationship between the traction vector and stress tensor
The relationship between the traction vector t
(n)
and the stress tensor is now explained.
Consider the tetrahedral volume element depicted in Figure 3. The orientation of the
slanted surface with area dA is dened by the outward pointing unit normal n. The three
3
dA
A
B
C
D

33
x
3
n =
_
_
_
n
1
n
2
n
3
_
_
_
n
3
dA
n
1
dA
n
2
dA

11

12

31

32

13

22

21

23
x
1
x
2
t
(n)
1
t
(n)
2
t
(n)
3
Figure 3: Traction on an innitesimal surface
faces ACD, ABD and BCD are internal to the body. The stress components acting on
these faces are indicated in Figure 3.
The areas of the three faces ACD, ABD and BCD are given respectively by n
2
dA, n
3
dA
and n
1
dA. Now consider the equilibrium equations for this tetrahedral element. The sum
of all forces acting in the 1 direction is given by:

F
1
= ma
1
: t
(n)
1
dA
11
(n
1
dA)
21
(n
2
dA)
31
(n
3
dA) + f
1
dV = dV a
1
, (13)
where f
1
is the 1 component of a body force, is the density of the material and a
1
is the
1 component of the acceleration of the tetrahedral element. Dividing everywhere by dA,
Eq.(13) reduces to
t
(n)
1

11
n
1

21
n
2

31
n
3
+ f
1
dV
dA
=
dV
dA
a
1
. (14)
In the limit of shrinking the element to zero size, the ratio of the volume to area
dV
dA
approaches zero. Hence the relationship between the traction vector and stress tensor is
given by
t
(n)
1
=
11
n
1
+
21
n
2
+
31
n
3
(15)
t
(n)
2
=
12
n
1
+
22
n
2
+
32
n
3
(16)
t
(n)
3
=
13
n
1
+
23
n
2
+
33
n
3
, (17)
4
or, in direct notation
t
(n)
=
T
n, (18)
where the superscript T denotes the transpose of the stress tensor. In matrix-vector
notation, Eq.(18) is written as
_

_
t
(n)
1
t
(n)
2
t
(n)
3
_

_
=
_
_

11

21

31

12

22

32

13

23

33
_
_
_
_
_
n
1
n
2
n
3
_
_
_
(19)
4 Equilibrium conditions
So far in your structural mechanics studies, equilibrium is usually satised at the macro-
structural level, i.e. you write equilibrium conditions in terms of axial forces, shear forces
and bending moments. Once you obtain these quantities, they are converted to the
appropriate stresses.
Strictly speaking, these forces and moments are stress resultants, i.e. given some stress
distribution over a surface, the resultant forces and moments follow from integrating over
the surface. Figure 4 illustrates the idea of stress resultants. The total axial force in the
1 direction, which acts perpendicular to the 2-3 plane (the plane dened by the outward
unit normal e
1
), is simply
F
1
=

A
23

11
dA, (20)
where the integration is performed over the surface area A
23
. Similarly the shear forces
which acts parallel to the 2-3 plane are given by
V
2
=

A
23

12
dA (21)
V
3
=

A
23

13
dA (22)
The bending moments M
2
and M
3
follow from
M
2
=

A
23

11
x
3
dA (23)
M
3
=

A
23

11
x
2
dA, (24)
and nally the torsion T
1
follows from
T
1
=

A
23

13
x
2
dA

12
x
3
dA. (25)
In cases where the stress distribution is simple (e.g. constant or linear variation), this ap-
proach is acceptable. The stress resultants are solved from macro-structural equilibrium,
5
e
1
e
3

13
e
2

12

11
A
23
Figure 4: Illustration of stress resultants on the 2-3 plane
and the stress components then follow from simple force-stress or moment-stress relations
(such as =
F
A
, =
V
A
, =
My
I
, =
Tr
J
).
However, when the stress distribution is complex, macro-structural equilibrium conditions
are insucient to determine the stress components. We than have no choice but to solve
the equilibrium conditions expressed in terms of the stress components.
4.1 Equilibrium in terms of the stress tensor
Consider the elemental volume element, dimensions dx
1
dx
2
dx
3
, depicted in Figure 5.
In this gure, all the stress components that act on the free surfaces are indicated. Note
that in order to derive the stress equilibrium condition (which will be a partial dierential
equation), we have to assume that the stress components change with position. This
variation in a stress component with position is quantied by multiplying the appropriate
partial derivative of the stress component w.r.t. position with the spatial distance trav-
elled. As an example, the
11
stress component changes from a value of
11
on the back
face of the brick, to a value of
11
+

11
x
1
dx
1
on the front face of the brick. This change in
the stress component is purely due to the change in the x
1
coordinate.
Let us now consider the equilibrium of the volume element in the e
1
direction. As before,
6

23
+
23
x2
dx
2

32

21

22

23

11

33
+
33
x3
dx
3

32
+
32
x3
dx
3

31
+
31
x3
dx
3

13

12

31

12
+
12
x1
dx
1

13
+
13
x1
dx
1

11
+
11
x1
dx
1
f
2
dV
f
1
dV
f
3
dV
e
3

21
+
21
x2
dx
2

22
+
22
x2
dx
2

33
e
2
e
1
Figure 5: Equilibrium of an innitesimal volume
Newtons second law is applied:

F
1
= ma
1
:
_

11
+

11
x
1
dx
1
_
dx
2
dx
3
(
11
)dx
2
dx
3
+
_

31
+

31
x
3
dx
3
_
dx
1
dx
2
(
31
)dx
1
dx
2
+
_

21
+

21
x
2
dx
2
_
dx
1
dx
3

(
21
)dx
1
dx
3
+ f
1
dx
1
dx
2
dx
3
= dx
1
dx
2
dx
2
a
1
. (26)
Cancelling terms of opposite sign and dividing by dx
1
dx
2
dx
3
, Eq.(26) results in

11
x
1
+

21
x
2
+

31
x
3
+ f
1
= a
1
(27)
7
The equilibrium conditions in the 2 and 3 directions follows similarly:

12
x
1
+

22
x
2
+

32
x
3
+ f
2
= a
2
(28)

13
x
1
+

23
x
2
+

33
x
3
+ f
3
= a
3
(29)
A more convenient notation is to indicate partial derivatives with a subscripted comma,
followed by the dimension with respect to which the partial derivative is taken. This
notation results in the set of equilibrium conditions

11,1
+
21,2
+
31,3
+ f
1
= a
1
(30)

12,1
+
22,2
+
32,3
+ f
2
= a
2
(31)

13,1
+
23,2
+
33,3
+ f
3
= a
3
. (32)
Using Einsteins summation convention, the equilibrium conditions are written as

ji,j
+ f
i
= a
i
for i = 1, 2, 3. (33)
We can also make use of direct notation, in which the equilibrium conditions are stated
as

T
+f = a, (34)
where is the gradient operator, and
T
is known as the divergence of the transpose
of the stress tensor.
4.2 Symmetry of the stress tensor
Let us also consider the moments equilibrium conditions. In direct notation, the moment
equation is

M
G
= I
GG
(35)
where M
G
is the moment vector about the center of gravity, I
GG
is the inertia tensor
about the center of gravity and is the angular acceleration vector. The inertia tensor
is given by
I
GG
=
_
_
I
11
I
12
I
13
I
21
I
22
I
23
I
31
I
32
I
33
_
_
, (36)
where the diagonal entries are called the moments of inertia, and the o-diagonal terms
are known as the products of inertia. If the body is symmetric with respect to the chosen
axes, the products of inertia are zero, and the moments equations become uncoupled.
Although this is the case here, the following result is valid in general. Summing moments
about the 3 axis, we obtain
(
21
dx
1
dx
3
)
dx
2
2

_

21
+

21
x
2
dx
2
_
(dx
1
dx
3
)
dx
2
2
+ (
12
dx
2
dx
3
)
dx
1
2
+
_

12
+

12
x
1
dx
1
_
(dx
2
dx
3
)
dx
1
2
=
1
12
dx
1
dx
2
dx
3
(dx
2
2
+ dx
2
3
)
1
(37)
8
After cancelling
1
2
dx
1
dx
2
dx
3
from all the terms above, Eq.(37) reduces to

21

21
+

21
x
2
dx
2
+
12
+
12
+

12
x
1
dx
1
=
1
6
(dx
2
2
+ dx
2
3
)
1
(38)
Taking the limit as dx
1
, dx
2
and dx
3
approach zero, all terms except
12
and
21
vanish,
yielding the nal result

12
=
21
. (39)
Moment equilibrium about the 1 and 2 axes yield the results

23
=
32
and
13
=
31
(40)
respectively. This results is known as symmetry of the stress tensor. In direct notation,
we write
=
T
, (41)
and using index notation we write

ij
=
ji
for i = 1, 2, 3 and j = 1, 2, 3. (42)
This means that the 6 stress components
11
,
22
,
33
,
12
,
13
and
23
completely dene
the stress tensor . Using symmetry, the equilibrium conditions may be rewritten as

ij,j
+ f
i
= a
i
for i = 1, 2, 3 or (43)
+f = a or (44)
div +f = a (45)
which is the form I prefer and remember best.
5 Strain-displacement relations
Given a smooth displacement eld u(x
1
, x
2
, x
3
), the innitesimal strain tensor (again a
second order tensor) is dened as
=
1
2
(u +
T
u) . (46)
Instead of direct notation, we can use index notation, in which case

ij
=
1
2
(u
i,j
+ u
j,i
) (47)
It follows from the denition of the strain tensor that it is also symmetric, i.e.
=
T
(48)
As before, this means that the 6 strain components
11
,
22
,
33
,
12
,
13
and
23
completely
dene the strain tensor . Note that the o-diagonal terms, known as shear strains, has
the following relation to the shear angles :

12
=
12
+
21
(49)

13
=
13
+
31
(50)

23
=
23
+
32
(51)
9
6 Stress-strain relations
Given the strain tensor , the stress tensor is computed from the general linear consti-
tutive function
= C, (52)
where C is known as the fourth order elasticity tensor. The above relation is the most
general form that a linear constitutive relation can have. In index notation, we write

ij
= C
ijkl

kl
, (53)
where the four subscripts of the component C
ijkl
reect the fact that C is a fourth order
tensor. Eq.(53) should be interpreted according to Einsteins summation convention as

ij
=
3

k=1
3

l=1
C
ijkl

kl
for i = 1, 2, 3 and j = 1, 2, 3. (54)
Note that a general fourth order tensor in 3D problems has 81 independent constants.
However, since the stress tensor and the strain tensor are symmetric, the following re-
strictions are required for the fourth order elasticity tensor C:
C
ijkl
= C
jikl
= C
ijlk
= C
klij
. (55)
This leaves 21 independent coecients in C
ijkl
, which are required to completely describe
a linear elastic material. If however the material is isotropic (i.e. the material response is
independent of orientation), the stress-strain relation simplies to

ij
=
kk

ij
+ 2
ij
(56)
where and are known as the Lame constants. is also known as G, the shear modulus.

ij
is Kroneckers delta, dened as

ij
=
_
0 if i = j
1 if i = j
(57)
and
kk
should again be interpreted using Einsteins summation convention, which results
in
kk
=
11
+
22
+
33
. Any isotropic linear elastic material only has two independent
elastic constants, hence and can be written in terms of the more conventional elastic
modulus E and Poissons ratio :
=
E
(1 + )(1 2)
(58)
=
E
2(1 + )
(59)
All the above relations are cumbersome to work with if we want to do any hand calcula-
tions or computer implementations. Therefore more convenient relations are in use, where
the second order stress and strain tensors are represented as vectors, and the fourth order
10
elasticity tensor as a matrix. I denote vectors with curly braces { } and matrices with
square brackets [ ]. Using this notation, and making use of symmetry (recall that there
is only 6 stress and 6 strain components), the stress-strain relation is expressed as
{} = [D]{} (60)
where
{} =
_

11

22

33

12

13

23
_

_
, {} =
_

11

22

33

12

13

23
_

_
and (61)
[D] =
E(1 )
(1 + )(1 2)
_

_
1

1

1
0 0 0

1
1

1
0 0 0

1
1 0 0 0
0 0 0
12
2(1)
0 0
0 0 0 0
12
2(1)
0
0 0 0 0 0
12
2(1)
_

_
(62)
6.1 Plane problems
If certain conditions hold, three dimensional problems can be analysed by only considering
the problem in the 1-2 plane. These conditions are
The geometry is not a function of the 3 direction.
Boundary conditions are not a function of the 3 direction.
The material properties are not a function of the 3 direction.
All loading occurs in the 1-2 plane.
Due to these conditions, no shear can occur in the 1-3 or 2-3 planes, hence
13
=
23
=

13
=
23
= 0.
Two possibilities exist which can simplify the analysis even further. These are known as
plane stress and plane strain conditions. Note that these conditions are not physical types
of problem, but rather exact mathematical limits.
6.1.1 Plane stress
The plane stress problem occurs in the limit where the 3 dimension approaches zero. Since
no load is applied in the 3 direction, and since the body is very thin, the
33
stress vanishes.
11
Combined with the general plane problem simplications, the stress-strain relation is
_
_
_

11

22

12
_
_
_
=
E
1
2
_
_
1 0
1 0
0 0
1
2
_
_
_
_
_

11

22

12
_
_
_
. (63)
Note that the
33
strain is not zero, but given by

33
=

E
(
11
+
22
). (64)
6.1.2 Plane strain
The plane strain problem occurs in the limit when the 3 dimension approaches innity.
Since such a body will prevent any length change along the 3 direction, the
33
strain
vanishes. Combined with the general plane problem simplications, the stress-strain
relations reduce to
_
_
_

11

22

12
_
_
_
=
E
(1 + )(1 2)
_
_
1 0
1 0
0 0
12
2
_
_
_
_
_

11

22

12
_
_
_
. (65)
Note that the
33
stress is not zero, but given by

33
= (
11
+
22
). (66)
7 The mixed boundary value problem of solid me-
chanics
We are nally ready to pose the boundary value problem (BVP). We have a body B
with surface B and outward unit normal n. The body is loaded with a body force f
and tractions t are applied on part of the surface, denoted B
t
. On the remainder of the
surface, denoted B
u
, displacements u are prescribed. Note that either the displacements,
or the tractions are known on the boundary, i.e. B = B
u
B
t
.
12
7.1 The strong form
We must nd the stress eld (x
1
, x
2
, x
3
), the strain eld (x
1
, x
2
, x
3
) and the displace-
ment eld u(x
1
, x
2
, x
3
) that satises the equations
div +f = a in B (67)
= C in B (68)
=
1
2
(u +
T
u) in B (69)
n = t on B
t
(70)
u = u on B
u
(71)
This problem statement is known as the strong from of the BVP. A solution to the strong
form implies that we found stress, strain and displacement elds that satised Eqs.(67)
(71). Note that the total number of unknowns and the total number of equations match:
6 stress unknown, 6 strain unknowns and 3 displacements unknowns compared to 6 stress-
strain equations, 6 strain-displacement equations and 3 equilibrium equations.
How would we go about solving the mixed BVP of solid mechanics? We could propose
a method in which we attempt to solve all the unknowns simultaneously. One problem
with this approach is that the types of equations are mixed: the equilibrium equations and
the strain-displacement equations are partial dierential equations, while the stress-strain
relationship is a linear algebraic equation.
Since the unknown eld quantities are inter-related, it makes more sense to solve one eld,
and let the other eld follow from these relations. The nite element method solves the
displacement eld u, from which the strain and stress elds can then be computed using
Eqs.(69) and (68) respectively.
In general, analytical solutions to mixed BVPs in solid mechanics are very dicult to
obtain, except in cases where the geometry and loading conditions are simple. So if
an exact solution is dicult (and frequently impossible) to obtain, we have to accept
approximate solutions. Unfortunately, the strong form of the problem doesnt allow for
approximate solutions, since no mechanism exist by which we can judge the quality of the
approximations. Hence, there exists a need for a method which naturally provide a way
to distinguish between the quality of dierent approximations.
7.2 The weak form
The weak form of the solid mechanics BVP is based on the method of weighted residuals.
This is a method used to solve partial dierential equations, by converting the govern-
ing equation into an integral form. The governing dierential equation (the equilibrium
equation in the case of solid mechanics) is rst expressed as a residual R, i.e.
R = div +f a = 0. (72)
13
The dot product is then computed between the residual vector eld R and a weighting
eld w. This scalar eld is then integrated over the entire domain i.e.

B
w RdV = 0. (73)
Eq.(73) can be interpreted as a weighting of the residual (error) over the domain with
respect to a weighting function w. But does this approach retain the essence of the
original problem? It is obvious that if Eq.(72) holds, then Eq.(73) holds. But we are
interested in the opposite relation: If Eq.(73) holds, does that imply that Eq.(72) holds?
If our choice of weighting functions are naive, the answer is no. If instead the weighting
function w is allowed to be arbitrary (i.e. it can take any form, as long as it is zero along
B
u
), then a solution to the weak form will also be a solution to the strong form.
You might ask why we place the restriction that w = 0 along B
u
. Recall from your
previous structural mechanics courses that at all locations where we know displacements,
we expect to compute unknown reaction forces. Similarly, we now expect to solve the un-
known tractions

t along B
u
. It will follow later that we cannot accommodate a boundary
term with too many unknowns i.e. along B
u
we cannot allow for w to be arbitrary in
addition to an unknown

t.
One problem with the weak form as expressed in Eq.(73) is that the prescribed traction
does not appear. This presents a problem, as we will have no direct control over satis-
faction of the traction boundary condition. Fortunately, the divergence theorem of Gauss
comes to our rescue.
7.2.1 Divergence theorem of Gauss
Consider the vector eld function
F = F
1
(x
1
, x
2
, x
3
) e
1
+ F
2
(x
1
, x
2
, x
3
) e
2
+ F
3
(x
1
, x
2
, x
3
) e
3
(74)
with continuous rst and second partial derivatives, dened on a piecewise-smooth closed
surface B, which encloses a volume B. The surface has a unit outward normal n. Recall
that the divergence of such a vector eld is the scalar
divF =
F
1
x
1
+
F
2
x
2
+
F
3
x
3
. (75)
Gausss divergence theorem states that

B
divFdV =

B
F ndA. (76)
7.2.2 Divergence of a second order tensor
Weve already seen the divergence of a second order tensor when we derived the equilib-
rium equations in terms of the stress components. Recall that the divergence of a second
14
order tensor G is the vector
divG =
_
G
11
x
1
+
G
12
x
2
+
G
13
x
3
_
e
1
+
_
G
21
x
1
+
G
22
x
2
+
G
23
x
3
_
e
2
+
_
G
31
x
1
+
G
32
x
2
+
G
33
x
3
_
e
3
. (77)
Using the product rule of dierentiation, lets consider the divergence of the product
between a second order tensor G(x) and a vector g(x):
div (G
T
(x)g(x)) =div ((G
11
g
1
+ G
21
g
2
+ G
31
g
3
) e
1
+
(G
12
g
1
+ G
22
g
2
+ G
32
g
3
) e
2
+ (G
13
g
1
+ G
23
g
2
+ G
33
g
3
) e
3
)
=
G
11
x
1
g
1
+ G
11
g
1
x
1
+
G
21
x
1
g
2
+ G
21
g
2
x
1
+
G
31
x
1
g
3
+ G
31
g
3
x
1
+
G
12
x
2
g
1
+ G
12
g
1
x
2
+
G
22
x
2
g
2
+ G
22
g
2
x
2
+
G
32
x
2
g
3
+ G
32
g
3
x
2
+
G
13
x
3
g
1
+ G
13
g
1
x
3
+
G
23
x
3
g
2
+ G
23
g
2
x
3
+
G
33
x
3
g
3
+ G
33
g
3
x
3
=G
ij,j
g
i
+ G
ij
g
i,j
(78)
The inner product of two second order tensors A and B is dened as
A B =
_
_
A
11
A
12
A
13
A
21
A
22
A
23
A
31
A
32
A
33
_
_

_
_
B
11
B
12
B
13
B
21
B
22
B
23
B
31
B
32
B
33
_
_
=A
11
B
11
+ A
12
B
12
+ A
13
B
13
+ A
21
B
21
+ A
22
B
22
+ A
23
B
23
+
A
31
B
31
+ A
32
B
32
+ A
33
B
33
=A
ij
B
ij
.
(79)
Using this denition, as well as the denition of the divergence of a second order tensor,
Eq.(79) is rewritten as
div (G
T
(x)g(x)) = g(x) divG(x) +G(x)
g(x)
x
(80)
7.2.3 The useful version of the weak form of the BVP
Now that we are equipped with the divergence theorem of Gauss, we can revisit the weak
form. Expanding the terms in Eq.(73), we get

B
w (div +f a) dV = 0

B
w divdV +

B
w fdV

B
w adV = 0
(81)
15
We now apply the identity in Eq.(80) to the rst term in Eq.(81) to obtain

B
div (
T
w) dV

B

w
x
dV +

B
w fdV

B
w adV = 0. (82)
Finally, we get to apply Eq.(76) (Gausss divergence theorem) to the rst term in Eq.(82):

B
(
T
w) ndA

B

w
x
dV +

B
w fdV

B
w adV = 0. (83)
Lets consider the expression (
T
w) n. Expanding this term for the general 3D case and
then recombining terms, we obtain
(
T
w) n =
_
_
_
_

11

21

31

12

22

32

13

23

33
_
_
_
_
_
w
1
w
2
w
3
_
_
_
_
_

_
_
_
n
1
n
2
n
3
_
_
_
=
_
_
_

11
w
1
+
21
w
2
+
31
w
3

12
w
1
+
22
w
2
+
32
w
3

13
w
1
+
23
w
2
+
33
w
3
_
_
_

_
_
_
n
1
n
2
n
3
_
_
_
=
11
w
1
n
1
+
21
w
2
n
1
+
31
w
3
n
1
+
12
w
1
n
2
+
22
w
2
n
2
+

32
w
3
n
2
+
13
w
1
n
3
+
23
w
2
n
3
+
33
w
3
n
3
=
_
_
_

11
n
1
+
12
n
2
+
13
n
3

21
n
1
+
22
n
2
+
23
n
3

31
n
1
+
32
n
2
+
33
n
3
_
_
_

_
_
_
w
1
w
2
w
3
_
_
_
=
_
_
_
_

11

12

13

21

22

23

31

32

33
_
_
_
_
_
n
1
n
2
n
3
_
_
_
_
_

_
_
_
w
1
w
2
w
3
_
_
_
=(n) w
(84)
Recall from Eq.(18) and Eq.(41) that n = t. The boundary integral in Eq.(83) therefore
becomes

B
(
T
w) ndA =

B
(n) wdA
=

Bt
(n) wdA +

Bu
(n) wdA
=

Bt
(n) wdA
=

Bt
t wdA.
(85)
Recall that since w = 0 along B
u
, the boundary term over B
u
vanishes.
Note that since the substitution n = t occurs inside an integral, these quantities are
only equal in a weighted average sense.
16
Finally, the weak form of the solid mechanics boundary value problem is stated as:
Find the stress eld (x
1
, x
2
, x
3
), the strain eld (x
1
, x
2
, x
3
) and the displacement eld
u(x
1
, x
2
, x
3
) that satises the equations

B

w
x
dV =

Bt
t wdA +

B
w fdV

B
w adV (86)
= C in B (87)
=
1
2
(u +
T
u) in B (88)
u = u on B
u
(89)
for arbitrary w, as long as w = 0 along B
u
.
7.2.4 Approximate solutions to the BVP using the weak form
As mentioned earlier, the weak form can be used to obtain approximate solutions to the
BVP. This is done by limiting the weighting functions w to a nite set of chosen functions.
In the nite element method, both the displacement eld and the weighting functions
w are chosen as piecewise-continuous low order polynomial functions, dened on sub-
domains (elements) within the total domain B. Increased accuracy of the approximation
is obtained by either increasing the order of the polynomial interpolation functions, or by
reducing the size of the elements.
17