CH 6

Downloaded from https://onlinelibrary.wiley.com/doi/ by <Shibboleth>-member@purdue.edu, Wiley Online Library on [19/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.
com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6
Strong and Weak Forms for
Multidimensional Scalar Field
Problems
In the next three chapters, we will retrace the same path that we have just traversed for one-dimensional
problems for multidimensional problems. We will again follow the roadmap in Figure 3.1, starting with the
development of the strong form and weak form in this chapter. However, we will now consider a more
narrow class of problems; we have called these scalar problems because the unknowns are scalars like
temperature or a potential. The methods that will be developed in these chapters apply to problems such as
steady-state heat conduction, ideal fluid flow, electric fields and diffusion–advection. In order to provide a
physical setting for these developments, we will focus on heat conduction in two dimensions, but details
will be given for some of the other applications.
As can be seen from the roadmap in Figure 3.1, the first step in developing a finite element method is to
derive the governing equations and boundary conditions, which are the strong form. We will see that in two
dimensions, just as before, we will have essential and natural boundary conditions. Using a formula similar
to integration by parts, we will then develop a weak form. Finally, we will show that the weak form implies
the strong form, so that we can use finite element approximations for trial solutions to obtain approximate
solutions to the strong form by solving the weak form.
One aspect that we will stress in the extension to two dimensions is its similarity to the one-dimensional
formulation. The major equations in two dimensions are almost identical in structure to those in one
dimension, so most of the learning effort can be devoted to learning what these expressions mean in two
dimensions. The expressions for the strong and weak forms in two dimensions, by the way, are identical to
those forthreedimensions,andatthe endofthe chapterwewillgive ashortdescription ofhow they are applied
to three dimensions. In engineering practice today, most analyses are done in three dimensions, so it is
worthwhile to acquaint yourself with the theory in three dimensions. The extension from two to three
dimensions isalmosttrivial(we have usuallyavoidedtheword ‘trivial’ inthisbookbecause itisoftenmisused
in texts, for what often seems trivial to an author can be quite baffling, but the extension from 2D to 3D is
indeed trivial).
One complication in extending the methods to two dimensions lies in notation. In two dimensions,
variables such as heat flux and displacement are vectors. You have undoubtedly encountered vectors in
elementary physics. Vectors are physical quantities that have magnitude and direction, and they can be
expressed in terms of components and base vectors. We will denote vectors by superposed arrows, such as~ q,
which is the flux matrix. Let the unit vectors in the x and y directions be~i and~j; these are often called the base
A First Course in Finite Elements J. Fish and T. Belytschko

# 2007 John Wiley & Sons, Ltd. ISBN: 978-0-470-03580-1
Downloaded from https://onlinelibrary.wiley.com/doi/ by <Shibboleth>-member@purdue.edu, Wiley Online Library on [19/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
132 STRONG AND WEAK FORMS FOR MULTIDIMENSIONAL SCALAR FIELD PROBLEMS
vectors of the coordinate system. Then a vector ~

q can be expressed in terms of its components by
q ¼ qx~i þ qy~j;
~ ð6:1Þ
where qx and qy are the x and y components of the vector, respectively.
When we get to the derivation of finite element equations, it becomes convenient to use matrix notation.
A column matrix can be used to describe a vector~q by listing the components of the vector in the order as
shown below:

q
q¼ x : ð6:2Þ
qy
Though it is not crucial to deeply understand the difference between vectors and matrices at this point, a
vector differs from a matrix: a vector embodies the direction for a physical quantity, whereas a matrix is
just an array of numbers. We will give most of the formulas of the strong and weak forms in both vector
and matrix notations. In the finite element equations, we will use only matrix notation. You will see that
the derivation of weak and strong forms in matrix notation is a little awkward and differs from the forms
commonly seen in advanced calculus and physics. So if you know vector notation as taught in those
courses, you may find it preferable to use vector notation for the material in this chapter. The transition
to matrix notation is quite easy. On the contrary, some people prefer to learn both parts in matrix notation
for the sake of consistency.
An important operation in vector methods is the scalar product. The scalar product of two vectors in
cartesian coordinates is the sum of the products of the components of the vectors; the scalar product of ~ q
with a vector~r is given by
~
q ~
r ¼ qx rx þ qy ry :
The scalar product is commutative, so the order of the two vectors does not matter. If we consider two
matrices q and r that contain the components of~
q and~ r, respectively, then the scalar product is written as

r
qT r ¼ ½qx qy x ¼ qx rx þ qy ry :
ry
So writing the scalar product in terms of the matrices requires taking the transpose of the first matrix. It can
easily be shown that qT r ¼ rT q. When manipulating vector expressions in matrix form, it is important to
carefully handle the transpose operation.
Another important operation in vector methods is the gradient. The gradient provides a measure of the
slope of a field, so it is the two-dimensional counterpart of a derivative. The gradient vector operator is
defined by

~ ¼ ~i @ þ~j @ :
r
@x @y
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
~r
The gradient of the function ðx; yÞ is obtained by applying the gradient operator to the function, which
gives
~ ¼ ~i @ þ ~j @ :
r
@x @y
Notice that we have simply replaced the bold dot in ( ) by ðx; yÞ. The gradient of a function gives the
direction of steepest descent. In other words, if you think of the function as describing a ski slope, the
DIVERGENCE THEOREM AND GREEN’S FORMULA 133
gradient gives you the direction along which you would go the fastest. This is further illustrated in
Example 6.1.
The scalar product of the gradient operator with a vector field gives the divergence of the vector field. The
term divergence probably originated in fluid mechanics, where it refers to the flow leaving a point. We will
see later that the divergence of the heat flux is equal to the heat flowing from a point (the negative of the
source in a steady-state situation). The divergence of a vector~
q is obtained by taking a scalar product of the
gradient operator ~ r and ~q, which gives

~ ~ @ @ @qx @qy
r q ¼ ~i þ~j qx~i þ qy~j ¼ þ div~q:
@x @y @x @y
Notice that the divergence of a vector field is a scalar. As indicated in the last expression, the divergence
operator is often written by simply preceding the vector by the abbreviation ‘div’.
The above expressions can be written in the matrix form as follows. The gradient operator is defined as a
column matrix. So
2 3 2 3
@ @
6 @x 7 6 @x 7
=¼6 4@ 5
7 and = ¼ 6 7
4 @ 5:
@y @y
The matrix form of the divergence is written by replacing the dot in the scalar product by a transpose
operation, so
~ q ¼ =T q:
q ¼ ~
div~
It is important to notice that when we write the gradient operator invector notation, an arrow is placed on the
inverted del; in matrix notation, the arrow is omitted.
In the following, the students should use whichever notation is more natural. For those not very familiar
with either notation, they should first scan the material and see which one they can understand more readily.
For advanced students, a familiarity with both notations is recommended.
6.1 DIVERGENCE THEOREM AND GREEN’S FORMULA

The two-dimensional equations will be developed for a body of arbitrary shape. We will often refer to the
points inside the body as the domain of the problem we are treating. We will follow common practice and
draw this generic arbitrary body as shown in Figure 6.1(b); the idea of this figure is intended to convey that
we are not placing any restrictions on the shape of the body: The derivations that follow hold for arbitrary
shapes. This body is often called a potato, though heat conduction in potatoes is seldom of interest. It is
worth pointing out that the shape can actually be much more complicated: The body can have holes, it can
Γ
n
y ny
nx
n = −1 Ω = [0,l] n= 1x Ω
0 l
Γ
x
(a) (b)
Figure 6.1 (a) One-dimensional domain and (b) two-dimensional domain.

have corners and it can consist of different materials with interfaces between them. The boundary of the
domain is denoted by . Notice that our nomenclature is identical to that in the previous chapters, but now
the symbols refer to more complicated objects. The correspondence between the definitions in one and two
dimensions is readily apparent by comparing Figure 6.1(a) and Figure 6.1(b).
The unit normal vector to the domain, denoted by ~ n, is shown at a typical point in Figure 6.1(b) and is
given by
n ¼ nx~i þ ny~j;
~ ð6:3Þ
and nx and ny are the x and y components of the unit normal vector, respectively; this vector is also called the
normal vector or just the normal. As ~ n is a unit vector, it follows that n2x þ n2y ¼ 1.
The objective of this section is to develop the formula corresponding to the integration by parts formula
(3.16) for a scalar field ðx; yÞ, where ðx; yÞ is defined on the domain . Examples of the scalar fields are
temperature fields Tðx; yÞ and potential fields ðx; yÞ.
Prior to discussing the divergence theorem, it is instructive to recall the fundamental theorem of calculus
that we developed in Chapter 3: for any C0 integrable function in a one-dimensional domain, , with
boundaries , we have
Z
dðxÞ
dx ¼ ðnÞj : ð6:4Þ
dx

Recall that the boundary consists of the two end points of the domain and the unit normals point in the
negative x-direction at x ¼ 0 and positive x-direction at x ¼ l.
The generalization of this statement to multidimensions is given by Green’s theorem, which states:
If ðx; yÞ 2 C0 and integrable; then

Z I Z I
~ d ¼ ~
r n d or = d ¼ n d: ð6:5Þ

Note the similarity of (6.4) and (6.5); the operator d=dx is simply replaced by the gradient ~ r. In fact, d/dx
can be considered the one-dimensional counterpart of the gradient. So the one-dimensional form (6.4) is
just a special case of (6.5). Equation (6.5) also applies in three dimensions. The proof of Green’s theorem is
given in Appendix A4.
Using the above, we will now develop a theorem that relates the area integral of the divergence of a vector
q is C0
field to the contour integral of a vector field, which is called the divergence theorem. It states that if~
and integrable, then
Z I Z I
~ ~
r q d ¼ ~
q ~
n d or =T q d ¼ qT n d: ð6:6Þ

Note that (6.5) in two dimensions represents two scalar equations

Z Z Z Z
@ @
ðaÞ d ¼ nx d; ðbÞ d ¼ ny d: ð6:7Þ
@x @y

Letting ¼ qx in (6.7a) and ¼ qy in (6.7b), and adding them together yields
Z I Z I
@qx @qy ~ ~
þ d ¼ ðqx nx þ qy ny Þ d or r q d ¼ ~
q ~
n d; ð6:8Þ
@x @y

which is the divergence theorem given in (6.6).

Green’s formula, which is derived next, is the counterpart of integration by parts in one dimension. It
states that
Z I Z Z I Z
~ ~
wr q d ¼ q ~
w~ n d ~ ~
rw q d or w=T q d ¼ w qT n d ð=wÞT q d:

~ ðw~
To develop Green’s formula, we first evaluate r qÞ by the derivative of a product rule:
~ ðw~ @ @ @w @qx @w @qy

r qÞ ¼ ðwqx Þ þ ðwqy Þ ¼ qx þ w þ qy þ w
@x @y @x @x @y @y

@qx @qy @w @w ~ ~ ~ ~ ð6:9Þ
¼w þ þ qx þ qy ¼ wr q þ rw q:
@x @y @x @y
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
~r ~ q ~
rw ~ q
Notice that we can immediately write the last step of the above if we think of the gradient as a generalized
derivative and place dots between any two vectors.
Integrating (6.9) over the domain yields
Z Z Z
~ ðw~
r qÞ d ¼ ~ ~
wr q d þ ~ ~
rw q d: ð6:10Þ

Applying the divergence theorem to the LHS of (6.10) and then rearranging terms yields Green’s
formula:
Z I Z
~ ~
wr q d ¼ q ~
w~ n d ~ ~
rw q d: ð6:11Þ

It is interesting to observe that for a rectangular domain l 1 with one-dimensional heat flow, where
q ¼ qx~i and ~
~ n ¼ n~i, nð0Þ ¼ ~i, nðlÞ ¼ ~i, we have
Z I Z
@qx @w
w d ¼ qx wn d qx d: ð6:12Þ
@x @x

Choosing w to be only a function of x, i.e. wðxÞ, and integrating (6.12) in y the above reduces to the formula
for integration by parts in one dimension (3.16), which is repeated below:
Zl Zl
@qx @w
w dx ¼ ðqx wÞx ¼ l ðqx wÞx¼0 qx dx: ð6:13Þ
@x @x
0 0
Note the similarity of (6.11) and (6.13). For additional reading on Green’s theorem, Green’s formula and
the divergence theorem, we recommend Fung (1994) for an introductory approach and Malvern (1969) for
a more advanced treatment.
Example 6.1
Given a rectangular domain as shown in Figure 6.2. Consider a scalar function ¼ x2 þ 2y2 . Let~
q be the
gradient of defined as ~ ~ Contour lines are lines along which a function is constant.
q ¼ r.
(a) Find the normal to the contour line of passing through the point x ¼ y ¼ 0:5.
(b) Verify the divergence theorem for ~q.
The gradient vector ~
q is given as
@~ @~
~
q¼ i þ j ¼ 2x~i þ 4y~j:
@x @y
Figure 6.3 depicts the contour lines of and the gradient vector ~
q. It can be seen that ~
q is normal to the
contour lines and its magnitude represents the slope of at any point.
The gradient of at x ¼ y ¼ 0:5 is
qð0:5; 0:5Þ ¼ ~i þ 2~j:

~
At the point x ¼ y ¼ 0:5, the value of the scalar field is ð0:5; 0:5Þ ¼ 0:75. The unit normal vector to the
contour line x2 þ 2y2 0:75 ¼ 0 at the point x ¼ y ¼ 0:5 is obtained by dividing the vector ~ q by its
magnitude, which gives
1
nð0:5; 0:5Þ ¼ pffiffiffi ð~i þ 2~jÞ:
~
5
y
n(3) = j
D 1 C
n(2) = i
n(4) = − i
−1 1 x
A −1 B
n(1) = − j
Figure 6.2 Domain used for illustration of divergence theorem.

Figure 6.3 Contour lines of a function ¼ x2 þ 2y2 and its gradient.
We now verify the divergence theorem. The unit normal vectors at the four boundaries of the domain
ABCD are shown in Figure 6.2. To verify the divergence theorem (6.6), we first evaluate the integrand on
the LHS of (6.6):
~ ~ @qx @qy
r q¼ þ ¼ 2 þ 4 ¼ 6:
@x @y
Integrating the above over the problem domain gives
0 1
Z Z1 Z1
~ ~
r q d ¼ @ 6 dyA dx ¼ 24:
1 1
Evaluating the boundary integral counterclockwise gives

I Z Z Z Z
~
q ~
n d ¼ ð4yÞ |{z}
d þ d þ
2x |{z} d þ
4y |{z} ð2xÞ |{z}
d
AB dx BC dy CD dx DA dy
Z1 Z1 Z1 Z1
¼ 4 dxþ 2 dyþ 4 dxþ 2 dx ¼ 24:
1 1 1 1
Thus, we have verified the divergence theorem for this example.

C
(0,1)
n(2)
n(3)
B
A x
(2,0)
n(1)
Figure 6.4 Triangular problem domain used for illustration of divergence theorem.
Example 6.2
Given a vector field qx ¼ 3x2 y þ y3 , qy ¼ 3x þ y3 on the domain shown in Figure 6.4, verify the
divergence theorem.
The integrand on the LHS of (6.6) is given as
~ ~ @qx @qy
r q¼ þ ¼ 6xy þ 3y2 :
@x @y
Integrating the above yields

Z Z 2 Z 10:5x Z 2 h i
~ ~
r q d ¼ ð6xy þ 3y2 Þdy dx ¼ 3xð1 0:5xÞ2 þ ð1 0:5xÞ3 dx ¼ 1:5:
0 0 0

The counterclockwise computed boundary integral on AB is
Z Z Z2
~ nð1Þ d ¼
q ~ qx ð1Þ d ¼ 3x dx ¼ 6;
AB AB 0
where ~nð1Þ ¼ ~j, d ¼ dx and y ¼ 0 on AB.

For the counterclockwisepcomputed
ffiffiffi boundary integral
pffiffiffi on BC, note that equation of the line BC is given
by y ¼ 1 0:5x and~ nð2Þ ¼ 5=5ð~i þ 2~jÞ, d ¼ 5=2dx on BC. The boundary integral on BC is then
given by
Z Z pffiffiffi Z0
5 ~ ~ 1h i
~ nð2Þ d ¼
q ~ ðqx~i þ qy~j Þ ð i þ 2j Þ d ¼ ð3x2 þ y3 Þ þ 2ð3x þ y3 Þ dx ¼ 7:75:
5 2
BC AB 2
Finally, the counterclockwise boundary integral on CA is
Z Z Z0
~ nð3Þ d ¼
q ~ ðqx~i þ qy~jÞð~iÞ d ¼ y3 dy ¼ 0:25;
BC AB 1
nð3Þ ¼ ~i, d ¼ dy and x ¼ 0 on CA.

where ~
STRONG FORM 139
Adding the contributions of the three segments gives

I Z Z Z
~
q ~
n d ¼ ~
q ~
n d þ ~
q ~
n d þ ~
q ~
n d ¼ 6 þ 7:75 0:25 ¼ 1:5;
AB BC CA
which completes the demonstration that the divergence theorem holds for this case.
6.2 STRONG FORM1

To derive the strong form we will apply energy balance to a control volume. The strong form will be
completed by adding the Fourier law, which relates heat flux to the temperature gradient and the boundary
conditions. Finally, the weak form will be formulated by integrating the product of the governing equation
and the natural boundary condition with the weight function over the domains where they hold. A
symmetric form is obtained by applying Green’s formula (equivalent to the integration by parts in one
dimension). We consider only steady-state problems where the temperature is not a function of time.
Consider a plate of unit thickness shown in Figure 6.5(a); the plate contains a heat source sðx; yÞ (energy
per unit area and time). The control volume is shown in Figure 6.5(b). The balance of heat energy in the
control volume requires that the heat flux~ q flowing out through the boundaries of the control volume equals
the heat generated s. This is the same energy balance we used in Chapter 3: as the body is in steady state, the
heat energy in any control volume must stay constant, which means that the flow out has to equal the heat
energy generated by the source.
The flux vector ~ q can be expressed in terms of two components: the component tangential to the
boundary qt and the component normal to the boundary qn. The tangential component qt does not contribute
to the heat entering or exiting the control volume. Recall that
q ¼ qx~i þ qy~j;
~ n ¼ nx~i þ ny~j;
~ n2x þ n2y ¼ 1:
The normal component qn is given by the scalar product of the heat flux with the normal to the body:
qn ¼ ~
q ~
n ¼ qT n ¼ qx nx þ qy ny : ð6:14Þ
On AD, where ~ q ð~iÞ ¼ qx whereas on BC, where ~

n ¼ ~i, the heat inflow is qn ¼ ~ n ¼ ~i, the heat
inflow is qn ¼ ~ ~
q i ¼ qx .
q Δy
y qy (x , y + )
n D C 2
Ω s (x , y )
qn Δy O (x , y )
Δy
Δx Δx
Δx qx (x − , y) qx (x + ,y)
2 Δx 2
A B
x Δy
qy (x , y − )
2
(a) (b)
Figure 6.5 Problem definition: (a) domain of a plate with a control volume shaded and (b) heat fluxes in and out of the
control volume.
1
Recommended for Science and Engineering Track.
In Figure 6.5(b), only the normal components of the flux are shown, as these are the only ones that contribute
to the energy flow into the control volume. The energy balance in the control volume is given by

x x
qx x ; y y qx x þ ; y y
2 2

y y
þ qy x; y x qy x; y þ x þ sðx; yÞxy ¼ 0:
2 2
where the first four terms are the net heat inflow. Divide the above by xy and recall the definition of a
partial derivative:

x x
qx x þ ; y qx x ;y
2 2 @qx
lim ¼ ;
x!0 x @x

y y
qy x; y þ qy x; y
2 2 @qy
lim ¼ :
y!0 y @y
The above energy balance equation (after a change of sign) can then be written as
@qx @qy
þ s ¼ 0;
@x @y
or in the vector and matrix forms:

~ ~
ðaÞ r q s ¼ 0 or q s ¼ 0 or
div~ ðbÞ =T q s ¼ 0: ð6:15Þ
If we recall the definition of the divergence operator, we can see that this equation can be obtained just by
reasoning: the first term is the divergence of the flux, i.e. the heat flowing out from the point. The heat
flowing out from the point r ~ ~q must equal the heat generated s to maintain a constant amount of heat
energy, i.e. temperature, at a point, which gives equation (6.15).
Recall Fourier’s law in one dimension:
dT
q ¼ k ¼ krT:
dx
In two dimensions, we have two flux components and two temperature gradient components. For isotropic
materials in two dimensions, Fourier’s law is given by
~ ~ or q ¼ k=T;
q ¼ krT ð6:16Þ
where k > 0. As in one dimension, the minus sign in (6.16) reflects the fact that heat flows in the
direction opposite to the gradient, i.e. from high temperature to low temperature. If the conductivity k is
constant, the energy balance equation expressed in terms of temperature is obtained by substituting (6.16)
into (6.15):
kr2 T þ s ¼ 0; ð6:17Þ
where
~ ¼ =T = ¼ @ þ @ :
2 2
~ r
r2 ¼ r ð6:18Þ
@x 2 @y2
STRONG FORM 141
Equation (6.17) is called the Poisson equation and r2 is called the Laplacian operator.
The flux and the temperature gradient vectors are related by a generalized Fourier’s law:
2 3
@T

qx k kxy 6 7
6 @x 7;
¼ xx
qy kyx kyy 4 @T 5
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
D @y
or in the matrix form:
q ¼ D=T; ð6:19Þ
where D is the conductivity matrix. We write this equation only in the matrix form because the vector form
cannot be written without second-order tensors, which are not covered here.
Substituting the generalized Fourier law (6.19) into the energy balance equation (6.15) yields
=T ðD=TÞ þ s ¼ 0: ð6:20Þ
The matrix D must be positive definite as heat must flow in the direction of decreasing temperature.
For isotropic materials,

k 0
D¼ ¼ kI: ð6:21Þ
0 k
In two dimensions, the symmetry of the material is an important factor in the form of the Fourier law. A
material is said to have isotropic symmetry if the properties are the same in any coordinate system. For
example, most metals, concrete and a silicon crystal are isotropic. The form of the relation between heat
flux and temperature gradient in an isotropic material is independent of how the coordinate system is
placed. In anisotropic materials, D depends on the coordinate system. Examples of anisotropic materials
are radial tires, fiber composites and rolled aluminum alloys. For example, in a radial tire, heat flows much
more rapidly along the direction of the steel wires than in the other directions.
To solve the partial differential equation (6.20), boundary conditions must be prescribed. In multi-
dimensions, the same complementarity conditions that we learned in one dimension hold. At any point of
the boundary (see Figure 6.6), either the temperature or the normal flux must be prescribed, but they both
cannot be prescribed. Therefore, if we denote the boundary where the temperature is prescribed by T and
the boundary where the flux is prescribed by q, then we have
q [ T ¼ ; q \ T ¼ 0: ð6:22Þ
Γ = ΓT Γq
⊂
T = T on ΓT
x
Figure 6.6 Problem domain and boundary conditions.

We write the prescribed temperature boundary condition as

Tðx; yÞ ¼ Tðx; yÞ on T ; ð6:23Þ
where Tðx; yÞ is the prescribed temperature; these are the essential boundary conditions; these are also
called Dirichlet conditions. As indicated, the prescribed temperature along the boundary can be a function
of the spatial coordinates.
On a prescribed flux boundary, only the normal flux be prescribed. We can write the prescribed flux
condition as
qn ¼ ~
q ~
n¼
q on q : ð6:24Þ
These are also called Neumann conditions. For an isotropic material, the normal flux is proportional to the
gradient of the temperature in the normal direction, i.e. it follows from (6.19) and (6.21) that
qn ¼ knT rT. It can be seen that the flux depends on the derivatives of the temperature, so this is the
natural boundary condition.
The resulting strong form for the heat conduction problem in two dimensions is given in the vector form
for isotropic materials in Box 6.1 and in the matrix form for general anisotropic materials in Box 6.2. These
forms differ from what we used in one dimension in that the energy balance and Fourier’s law are not
combined. This simplifies the development of the weak form and extends the applicability of the weak form
to nonlinear heat conduction.
Box 6.1. Strong form (vector notation) for heat conduction
ðaÞ energy balance : ~ ~

r qs¼0 on ;
ðbÞ Fourier0 s law : ~
q ¼ krT~ on ;
ð6:25Þ
ðcÞ natural BC : qn ¼ ~
q ~
n¼q on q ;
ðdÞ essential BC : T¼T on T :
Box 6.2. Strong form (matrix notation) for heat conduction

ðaÞ energy balance : =T q s ¼ 0 on ;
0
ðbÞ Fourier s law : q ¼ D=T on ;
ð6:26Þ
ðcÞ natural BC : qn ¼ q n ¼
T
q on q ;
ðdÞ essential BC : T ¼T on T ;
The variables s, D, T and q are the data for the problem. These, along with the geometry of the domain ,
must be given.
6.3 WEAK FORM

To obtain the weak form we will follow the same basic procedure as for the one-dimensional problem in
Chapter 3. However, as we have already mentioned, we will develop the weak form of the balance equation
(6.15a). Then we will express the heat flux in terms of the temperature gradient by the Fourier law.
We start with the energy balance equation (6.15a) and the natural boundary condition (6.25c). We
premultiply the two equations by a weight function w and integrate over the problem domain and the
WEAK FORM 143
natural boundary q, respectively:

Z Z
ðaÞ ~ ~
wðr q sÞ d ¼ 0 8w; ðbÞ q ~
wð q ~
nÞ d ¼ 0 8w: ð6:27Þ
q
For the equivalence of the strong and weak forms, it is crucial that the weak form hold for all functions w. As
in one dimension, we will find that some restrictions must be imposed on the weight function, but we will
develop these as we need them. Applying Green’s formula to the first term in (6.27a) yields
Z I Z
~ ~
wr q d ¼ w~ q ~n d rw ~ ~ q d 8w: ð6:28Þ

Inserting (6.28) into (6.27a) yields

Z I Z Z Z Z
~
rw ~q d ¼ w~ q ~n d ws d ¼ q ~
w~ n d þ q ~
w~ n d ws d: ð6:29Þ
q T
where we have subdivided the first integral on the RHS of (6.29) into the prescribed temperature and
prescribed flux boundaries, which is permissible because of (6.22). Substituting (6.27b) into the integral on
q (6.29) yields
Z Z Z Z
~ ~
rw q d ¼ wq d þ w~ q ~
n d ws d:
q T
We now follow the same reasoning as in Chapter 3. It is easy to construct weight functions that vanish on a
portion of the boundary, so we set w ¼ 0 on the prescribed temperature boundary, i.e. the essential
boundary. Therefore the integral on T vanishes and the weak form is given by
Z Z Z
~ ~
rw q d ¼ q d ws d
w 8w 2 U0 ; ð6:30Þ
q
where U0 is the set of sufficiently smooth functions that vanish on the essential boundary, it is the space of
functions defined in (3.48). The space of admissible trial solutions U satisfies the essential boundary
conditions and is sufficiently smooth as defined in (3.47). Recall that according to the definition of these
spaces, the trial solutions and weight functions have to be C0 continuous.
Expressing (6.30) in matrix form gives
Z Z Z
ð=wÞT q d ¼ wq d ws d 8w 2 U0 :
q
The above is the weak form for any material, linear or nonlinear. To obtain the weak form for linear
materials, we substitute Fourier’s law into the first term of the above, which yields
Box 6.3. Weak form (matrix notation) for heat conduction

find T 2 U such that:
Z Z Z
ð=wÞT D=T d ¼ q d þ ws d
w 8w 2 U0 : ð6:31Þ
q
6.4 THE EQUIVALENCE BETWEEN WEAK AND STRONG FORMS2

To demonstrate the equivalence of the strong form and the weak form, it must be shown that the weak form
implies the strong form. This demonstration is similar to the one used in Chapter 3 for showing the
equivalence for one-dimensional problems: we reverse the steps that we have followed in going from the
strong form to the weak form and then invoke the arbitrariness of the weight functions to extract the strong
form from the integral equations. We will do this for the weak form for arbitrary materials.
We start with (6.30), rewritten below:
Z Z Z
~ ~
rw q d ¼ q d
w ws d:
q
Now we apply Green’s formula (6.11) to the first term, which gives
Z Z Z
~ ~
wðr q sÞ d þ q ~
wð q ~
nÞ d q ~
w~ n d ¼ 0 8w 2 U0 : ð6:32Þ
q T
We follow the same strategy as in Chapter 3. Since the weight function wðxÞ is arbitrary, it can be assumed to
be any function that vanishes on T .
We take advantage of the arbitrariness of the weight function and make it equal to the integrand that is, we
let

~ 0 on
w ¼ ðxÞðr ~ q sÞ; where ðxÞ ¼ : ð6:33Þ
> 0 on
Inserting (6.33) into (6.32) yields

Z
~ ~
ðr q sÞ2 d ¼ 0: ð6:34Þ

The boundary terms have vanished because our choice of wðxÞ, (6.33), vanishes on the boundaries. Since
ðxÞ > 0 in , the integrand in (6.34) is positive at every point in the domain. For the integral in (6.34) to
vanish, the integrand has to vanish as well. Hence, since ðxÞ > 0,
~ ~
r q s ¼ 0 in ; ð6:35Þ
which is the energy balance equation (6.15). After substituting (6.35) into (6.32) we select a weight
function that is nonzero on the natural boundary, but vanishes on the essential boundary (it does not
matter what its value is inside the domain, as by (6.35) we know that the first term in (6.32) will vanish).
So we let

0 on T
w ¼ ’ðq ~q ~
nÞ; where ’ ¼ : ð6:36Þ
> 0 on q
Substituting (6.36) into (6.32) yields

Z
q ~
’ð q ~
nÞ2 d ¼ 0: ð6:37Þ
q
2
Recommended for Advanced Track.
GENERALIZATION TO THREE-DIMENSIONAL PROBLEMS 145
z
q
n
q ⋅ n = q on Γq T = T on ΓT
k
i r y
j
x
Figure 6.7 Problem domain and boundary conditions in three dimensions.
As the integrand in (6.37) is positive on q , the quantity inside the parentheses must vanish on every point of
the natural boundary, so the natural boundary condition (6.25c) follows.
6.5 GENERALIZATION TO THREE-DIMENSIONAL PROBLEMS 3

The extension from two to three dimensions is almost trivial. The difference is not in the structure of the
strong and weak form equations, which are identical, but in the definitions of the vectors, gradient,
divergence and Laplacian operators.
In three dimensions, the base (unit) vectors are~i,~j and ~
k as shown in Figure 6.7. A vector~
q expressed in
terms of its components is
2 3
qx
q ¼ qx~i þ qy~j þ qz~
~ k; q ¼ 4 qy 5; ð6:38Þ
qz
where the matrix form is shown on the right-hand side. In three dimensions, the problem domain is a
volume (which looks like the potato in Figure 6.7) and its boundary is a surface. The progression of
dimensionality of the problem domain and its boundary from one-dimensional to three-dimensional
problems is summarized in Table 6.1.
The boundary , which is the surface encompassing the three-dimensional domain , consists of the
complementary essential and natural boundaries, as shown in Figure 6.7.
Table 6.1 Dimensionality of the problem domain and its boundary.
Entity Domain Boundary
One dimension (1D) Line segment Two end points

Two dimensions (2D) Two-dimensional area Curve
Three dimensions (3D) Volume Surface
3
The gradient operator in three dimensions in vector and matrix notations is defined as
~ ¼ ~i @ þ ~j @ þ ~
r
@
k ; ~ ¼ ~i @ þ ~j @ þ ~
r
@
k ;
@x @y @z @x @y @z
2 3 2 3
@ @
6 @x 7 6 @x 7
6 7 6 7
6@ 7 6 @ 7
=¼6 7
6 @y 7; = ¼ 6 7
6 @y 7:
6 7 6 7
4@ 5 4 @ 5
@z @z
With the above definitions of vectors and the gradient vector operator, the divergence of the vector field and
the Laplacian are
@qx @qy @qz
div ~
q¼ þ þ ;
@x @y @z
~ ¼ =T = ¼ @ þ @ þ @
2 2 2
r2 ¼ r ~ r
@x2 @y2 @z2 :
The strong form in vector and matrix notations is identical to that given in Equations (6.25) and (6.26). Note
that the Fourier law relating the three components of temperature gradient to the three flux components is
defined in terms of a 3 3 symmetric positive definite matrix D:
2 3
kxx kxy kxz
D ¼ 4 kyx kyy kyz 5:
kzx kzy kzz
The weak form is also identical to that for two-dimensional problems as given in (6.31).
6.6 STRONG AND WEAK FORMS OF SCALAR STEADY-STATE

ADVECTION–DIFFUSION IN TWO DIMENSIONS 4
The advection-diffusion equations are obtained from a conservation principle (often called a balance
principle), just like heat conduction. The conservation principle states that the species (be it a material,
an energy or a state) are conserved in each control volume of area x y and unit thickness shown in
Figure 6.8. The amount of species entering minus the amount of species leaving equals the amount
produced (a negative volume when the species decays). There are two mechanisms for inflow and outflow,
advection (or convection), which is given by~ v, and diffusion, which is given by ~
q.
Dy Dy
vyq (x , y + ) qy (x , y + )
2 2
vx q (x − D , y ) vxq (x + D , y )
x x
2 O (x , y ) 2
Dy
qx (x + D , y )
x
qx (x − D x , y ) 2
2 Dx
Dy Dy
vyq (x , y − ) qy (x , y − )
2 2
Figure 6.8 Control volume for advection–diffusion problem.

4
STRONG AND WEAK FORMS OF SCALAR STEADY-STATE ADVECTION–DIFFUSION 147
v ~
In addition, advection on each surface results in an inflow of ~ n. The conservation principle can then
be developed as in section 6.2:

x x y y
vx x ; y y þ qx x ; y y þ vy x; y x þ qy x; y x
2 2 2 2

x x y y
vx x þ ; y y qx x þ ; y y vy x; y þ x qy x; y þ x
2 2 2 2
þ xysðx; yÞ ¼ 0:
Dividing the above by xy and taking the limit x ! 0, y ! 0, we obtain
@ðvx Þ @ðvy Þ @ðqx Þ @ðqy Þ

þ þ þ s ¼ 0:
@x @y @x @y
The above can be written in the vector form as

~ ð~
r ~ ~
vÞ þ r q s ¼ 0: ð6:39Þ
This is the general form of the advection–diffusion equation. The first term accounts for the advection or
transport of the material and the second term accounts for the diffusion.
In many cases, the material carrying the species is incompressible. For steady-state problems and
incompressible materials, the rate of material volume entering control volume is equal to the rate of
material volume exiting control volume. Mathematically, this is given by

x y x y
vx x ; y y þ vy x; y x vx x þ ; y y vy x; y þ x ¼ 0:
2 2 2 2
Dividing the above by xy and taking the limit x ! 0, y ! 0 gives
@ðvx Þ @ðvy Þ
þ ¼ 0:
@x @y
The above in matrix and vector notations is
~ ~
r v ¼ 0 or =T v ¼ 0: ð6:40Þ
Equation (6.40) is known as the continuity equation for steady-state problems of incompressible
materials.
Substituting the continuity equation (6.40) into (6.39) yields the conservation equation for a species in a
moving incompressible fluid, which can be written as
~ ~ þr
v r ~ ~
q s ¼ 0 or vT = þ =T q s ¼ 0: ð6:41Þ
Assuming that the generalized Fourier’s law (6.19) holds, the conservation of species equation in the matrix
form becomes
vT = =T ðD=Þ s ¼ 0: ð6:42Þ
For isotropic materials D ¼ kI and the conservation equation reduces to
~ ~ kr2 s ¼ 0 or
v r vT = kr2 s ¼ 0; ð6:43Þ
where r2 is the Laplacian defined in (6.18). We consider the usual essential and natural boundary
conditions
¼ on ;
ð6:44Þ
~
q ~
n¼ q on q ;
where and q are complementary.

To obtain the weak form of (6.43) we multiply the conservation equation (6.41) and the natural
boundary condition by an arbitrary weight function w and integrate over the corresponding domains:
Z Z
ðaÞ wð~ ~ þr
v r ~ ~q sÞ d ¼ 0; ðbÞ wðq ~q ~
nÞ d ¼ 0 on 8w: ð6:45Þ
q
Integration by parts of the second term (the diffusion term) in (6.45a) gives
Z Z Z Z
w~ ~ d rw
v r ~ ~ q d þ wq d ws dðaÞ 8w 2 U0 ; ð6:46Þ
q
where we have exploited (6.45b) and that w ¼ 0 on .

Finally, the weak form is completed by substituting the generalized Fourier law into (6.46), which gives
find the trial solution ðx; yÞ 2 U such that

Z Z Z Z
w vT = d þ ð=wÞT D= d þ q d ws d
w 8w 2 U0 : ð6:47Þ
q
The above is the weak form for the advection–diffusion equation. Note that the first term is unsymmetric in
the weight function w and the solution . This will result in unsymmetric discrete system equations and has
important ramifications on the nature of the solutions, because as in 1D, the solutions can be unstable if the
velocity is large enough.
REFERENCES
Fung, Y.C. (1994) A First Course in Continuum Mechanics, 3rd edn, Prentice Hall, Englewood Cliffs, NJ.
Malvern, L.E. (1969) Introduction to the Mechanics of a Continuous Medium, Prentice Hall, Englewood
Cliffs, NJ.
Problems
Problem 6.1
Given a vector field qx ¼ y2 , qy ¼ 2xy on the domain shown in Figure 6.2. Verify the divergence
theorem.
Problem 6.2
Given a vector field qx ¼ 3x2 y þ y3 , qy ¼ 3x þ y3 on the domain shown in Figure 6.9. Verify the divergence
theorem. The curved boundary of the domain is a parabola.
REFERENCES 149
(0,4)
n2
x
(−2,0) (2,0)
n1
Figure 6.9 Parabolic domain of Problem 6.2 used for illustration of divergence theorem.
Problem 6.3
Using the divergence theorem prove
I
n d ¼ 0:

Problem 6.4
Starting with the strong form
dq
s ¼ 0; qð0Þ ¼
q; TðlÞ ¼ T;
dx
develop a weak form. Note that the flux q is related to the temperature through Fourier’s law, but develop the
weak form first in terms of the flux.
Problem 6.5
Consider the governing equation for the heat conduction problem in two dimensions with surface
convection:
=T ðD=TÞ þ s ¼ 2hðT T1 Þ on ;
qn ¼ qT n ¼
q on q ;
T ¼ T on T :
Derive the weak form.
Problem 6.6
Derive the strong form for a plate with a variable thickness tðx; yÞ. Hint: Consider control volume in Figure
6.5(b), and account for the variable thickness. For example the heat inflow at ðx x=2; yÞ is

x x
qx x ; y y t x ;y :
2 2
Derive the weak form for the plate with variable thickness.
y
q n = q on Γq
Γ = ΓT U Γq U Γh
T = T on ΓT
q n = h(T – T∞ ) on Γh
x
Figure 6.10 Problem domain and boundary conditions for heat conduction with boundary convection.
Problem 6.7
Consider a heat conduction problem in 2D with boundary convection (Figure 6.10).
Construct the weak form for heat conduction in 2D with boundary convection.
Problem 6.8
Consider a time-dependent heat transfer. The energy balance in a control volume (see Figure 6.5) is given
by

x x y
qx x ; y y qx x þ ; y y þ qy x; y x
2 2 2

y @T
qy x; y þ x þ sðx; yÞxy ¼ cr xy
2 @t;
where Tðx; y; tÞ, c and r denote the temperature, material specific heat and density, respectively, and t is the
time. The above equation states that the change in internal energy is not zero, but is rather governed by
density, specific heat and rate of change of temperature.
Derive the weak and strong forms for the time-dependent heat transfer problem.

CH 6

Uploaded by

Copyright:

Available Formats

CH 6

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CH 6

Uploaded by

Copyright:

Available Formats

Downloaded from https://onlinelibrary.wiley.com/doi/ by <Shibboleth>-member@purdue.edu, Wiley Online Library on [19/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.

A First Course in Finite Elements J. Fish and T. Belytschko

vectors of the coordinate system. Then a vector ~

6.1 DIVERGENCE THEOREM AND GREEN’S FORMULA

Figure 6.1 (a) One-dimensional domain and (b) two-dimensional domain.

If ðx; yÞ 2 C0 and integrable; then

Note that (6.5) in two dimensions represents two scalar equations

Letting ¼ qx in (6.7a) and ¼ qy in (6.7b), and adding them together yields

which is the divergence theorem given in (6.6).

~ ðw~ @ @ @w @qx @w @qy

qð0:5; 0:5Þ ¼ ~i þ 2~j:

Figure 6.2 Domain used for illustration of divergence theorem.

Figure 6.3 Contour lines of a function ¼ x2 þ 2y2 and its gradient.

Integrating the above over the problem domain gives

Evaluating the boundary integral counterclockwise gives

Thus, we have veriﬁed the divergence theorem for this example.

Integrating the above yields

The counterclockwise computed boundary integral on AB is

where ~nð1Þ ¼ ~j, d ¼ dx and y ¼ 0 on AB.

Finally, the counterclockwise boundary integral on CA is

nð3Þ ¼ ~i, d ¼ dy and x ¼ 0 on CA.

Adding the contributions of the three segments gives

6.2 STRONG FORM1

On AD, where ~ q ð~iÞ ¼ qx whereas on BC, where ~

or in the vector and matrix forms:

or in the matrix form:

Figure 6.6 Problem domain and boundary conditions.

We write the prescribed temperature boundary condition as

Box 6.1. Strong form (vector notation) for heat conduction

ðaÞ energy balance : ~ ~

Box 6.2. Strong form (matrix notation) for heat conduction

6.3 WEAK FORM

natural boundary q, respectively:

Inserting (6.28) into (6.27a) yields

Box 6.3. Weak form (matrix notation) for heat conduction

6.4 THE EQUIVALENCE BETWEEN WEAK AND STRONG FORMS2

Inserting (6.33) into (6.32) yields

Substituting (6.36) into (6.32) yields

Figure 6.7 Problem domain and boundary conditions in three dimensions.

6.5 GENERALIZATION TO THREE-DIMENSIONAL PROBLEMS 3

Table 6.1 Dimensionality of the problem domain and its boundary.

Entity Domain Boundary

One dimension (1D) Line segment Two end points

6.6 STRONG AND WEAK FORMS OF SCALAR STEADY-STATE

Figure 6.8 Control volume for advection–diffusion problem.

Dividing the above by xy and taking the limit x ! 0, y ! 0, we obtain

@ðvx Þ @ðvy Þ @ðqx Þ @ðqy Þ

The above can be written in the vector form as

Dividing the above by xy and taking the limit x ! 0, y ! 0 gives

The above in matrix and vector notations is

For isotropic materials D ¼ kI and the conservation equation reduces to

where and q are complementary.

where we have exploited (6.45b) and that w ¼ 0 on .

find the trial solution ðx; yÞ 2 U such that

Derive the weak form.

You might also like

where ~nð1Þ ¼ ~j, d ¼ dx and y ¼ 0 on AB.

nð3Þ ¼ ~i, d ¼ dy and x ¼ 0 on CA.

On AD, where ~ q ð~iÞ ¼ qx whereas on BC, where ~