Tensors and The Equations of Fluid Motion

CHAPTER 5
Tensors and the Equations of Fluid Motion

We have seen that there are a whole range of things that we can represent on
the computer. We have solved some simple problems such as Laplaces equation
on a unit square at the origin in the rst quadrant. From the description of the
problem, you can see that it was really a very specic problem. Our objective was
to dive into the process of representing and solving partial dierential equations on
the computer and we have achieved that objective. The problem was simple enough
that we could analyse the resulting discrete equations. Similarly, we have studied
the rst order one-dimensional linear wave equation. We have also seen the heat
equation and the quasi linear version of the wave equation. We have some basic
understanding of the issues involved in representing and solving these problems on
the computer. We are now in a position to solve a larger class of problems. We will
expand on the simple problems we studied in the preceding chapters as a means
to motivate this chapter. We will then do the essentials of tensor calculus required
to derive the equations of uid motion. The equations of motion will be derived
in vector form so as to be independent of the particular coordinate system. Taken
along with the tensor calculus you should be able to specialise these equations to
any particular coordinate system.
5.1. Laplace Equation Revisited
We solved the Laplace equation on a unit square. How would we handle the
problem domain shown in Figure 5.1? Again, we are given boundary conditions on
the four sides of the quadrilateral. If we tried to solve it using a Cartesian mesh as
we did before, we would get something that looks like the domain and mesh shown
in Figure 5.2. Everything looks ne till you look at the top edge of the trapezium.
Excepting two points, there are no other grid points on the boundary. How then
can we apply the boundary condition? We could forcibly insert mesh points at the
intersection of the grid lines the boundary. This would lead to unequal distances
between the mesh points in our discretisation. Those points then have to be dealt
with as a special case. The other possibility is to consider every grid point as being
separated from neighbouring grid points by unequal distances as shown in gure
5.3. It is likely that we will have to treat some points as special points. We have
come to expect that we will treat the boundary points dierently from the rest of
the grid points anyway. After all, look at what we did in the other problems that we
have seen, especially the one-dimensional Euler equations. Since we are trying to
motivate tensor calculus, our interest lies in a third possibility. That is to generate
a non-Cartesian mesh. One such mesh is shown in Figure 5.4. If we look at the
mesh shown in the gure, we see that the mesh lines conform to the boundary of
the domain. Imagine in your mind that you have a rubber sheet the shape of this
trapezium. You could stretch the sheet so that the stretched sheet looked like a
189
CHAPTER 5. TENSORS AND THE EQUATIONS OF FLUID MOTION
x
y
L
h
H
Figure 5.1. A trapezoidal domain on which Laplace equation is
to be solved
x
y
L
h
H
Figure 5.2. Trapezoidal domain with an underlying Cartesian mesh.
square. The mesh lines would coincide with the Cartesian grid lines. This tells us
that we need to perform a transformation of our coordinates so that we are back
to solving our problem on a Cartesian mesh.
If our coordinates in the stretched sheet are (, ), the mesh lines seen in Figure
5.4 would be constant -lines and constant -lines. The Figure 5.4 is drawn in the
x y plane. Clearly, what we need is a transformation going from one coordinate
system to another. Say,
= (x, y), (5.1.1)
= (x, y), (5.1.2)
c M. Ramakrishna, Dept. of Aerospace Engineering, IITM,
Chennai, India, 600 036
5.1. LAPLACE EQUATION REVISITED
y
1
x
2
x
1
y
2
Figure 5.3. A grid point with neighbouring points placed at un-
even distances.
x
y
L
h
H
Figure 5.4. A non-Cartesian mesh in a Trapezoidal domain. The
sides of the quadrilateral are mesh lines.
and the corresponding reverse relationship
x = x(, ), (5.1.3)
y = y(, ). (5.1.4)
We are in a position where, given the solution at a point in one coordinate system,
we can provide the solution at the corresponding point in the other coordinate
system. Let us step back for a minute to see where we are.
We have Laplace equation given to us in a trapezoidal domain in the x y
coordinate system. A little stretch will give us a square in the coordinate
system, but what happens to Laplaces equation in the plane? We use the
coordinate transformation given by equations (5.1.1) to (5.1.3) along with chain
rule to transform the derivatives. For example, here are the rst derivatives in x
and y.
x
=

x
. .
A
+

x
. .
B
(5.1.5)
y
=

y
+

y
(5.1.6)
Since, the transformation is known, we can determine the partial derivative on the
right hand side of equation (5.1.5). How do we use the expression given by equation
(5.1.5)? We can take the corresponding partial derivative of . On doing this, we
get
x
=

x
+

x
(5.1.7)
y
=

y
+

y
(5.1.8)
So far it looks manageable. Since we want to solve Laplaces equation we now look
at the second derivatives. The second x derivative is
2
x
2
=

x
_
+

x
_
=

2
x
2
. .
A1
+
_
x
_
2
2
+

x
. .
A2
+

2
x
2
. .
B1
+

x
+
_
x
_
2
2
. .
B2
(5.1.9)
This is a little messy. To make sure we understand this clearly, the term A in
equation (5.1.5) results in the terms identied as A
1
and A
2
in equation (5.1.9).
The same is true of the terms marked B in the two equations. A
1
and A
2
are a
consequence of applying product rule. The two terms in A
2
emerge from applying
equation (5.1.5) to obtain the derivative of the / term with respect to x. In a
similar fashion we can write the second derivative with respect y as
2
y
2
=

2
y
2
+
_
y
_
2
2
+

y
y
2
+

y
+
_
y
_
2
2
(5.1.10)
Then the transformed Laplace equation can be written as
x
2
+

2
y
2
=
_
2
x
+
2
y
_

2
2
+ 2 (
x
x
+
y
y
)

2
+
_
2
x
+
2
y
_

2
2
+(
xx
+
yy
)

+ (
xx
+
yy
)

= 0
(5.1.11)
To keep things more compact, we decide to use the notation that the subscript
indicates dierentiation with respect to that parameter. So,
(5.1.12)
x
=

x
5.1. LAPLACE EQUATION REVISITED
Using this notation uniformly, the Laplace equation in the plane is given by
_
2
x
+
2
y
_
+ 2 (
x
x
+
y
y
)
+
_
2
x
+
2
y
_
+(
xx
+
yy
)
+ (
xx
+
yy
)
= 0
(5.1.13)
The domain for the problem has become easier, the equation does not quite t in
one line! Also, it is not in quite the right form. The coecients are still expressed
in the x, y coordinate system. We make the following observations and see if we
can clear the air a bit.
We want to solve problems that involve complicated domains. There may
be many methods to handle complicated problems, performing transfor-
mation of coordinates is denitely one way to do it.
We do not want to have to re-derive our governing equation in every new
coordinate system that we encounter. We need a general frame work in
which we can derive our equations.
The introduction of the subscript notation gave some relief in handling
the equation. So, the proper choice of notation is going to make life easier
for us. Further, we can do more complex things with the eort that we
are currently expending.
We observe that the only dierence between equation (5.1.9) and (5.1.10)
is the replacement of x with y. Again, we need the notation that will
help us to abstract these kinds of patterns out, so that we do not have to
repeat the derivation for each coordinate.
We want to solve problems in three dimensions and not just one and two
dimensions. If we are going to perform transformations in three dimen-
sions, we need to have some minimal understanding of geometry in three
dimensions.
We will address the last point here by looking at a little dierential geometry.
Coordinate lines in three dimensions are curves in three dimensions and we will try
to get a handle on them. A region of interest in three dimensions will be a volume
and it is dened using surfaces. We will take a brief look at surfaces. Tensor
calculus is a tool to address the rest of the issues raised in our list of observations.
We will do a little tensor calculus and some geometry.
As further motivation as to why one needs tensor calculus, consider the fol-
lowing conundrum. If you have learnt only calculus, this is for you to puzzle over
to show you there must be life beyond calculus. Consider a potential ow in two
dimensions. The velocity can be written in component form as (u, v) in Cartesian
coordinates. If we were to transform the velocity to some other coordinates (, )
we get
u =
dx
dt
= x
t
= x
t
+ x
t
= x
U + x
V (5.1.14)
v =
dy
dt
= y
t
= y
t
+ y
t
= y
U + y
V (5.1.15)
Where (U, V ) are the velocities in the coordinates. The matrix representation
of this transformation equation is
(5.1.16)
_
u
v
_
=
_
x
_ _
U
V
_
We also have from the denition of the potential
u =

x
=
x
=
x
+
x
=
x
U +
x
V (5.1.17)
v =

y
=
y
=
y
+
y
=
y
U +
y
V (5.1.18)
Which has a representation
(5.1.19)
_
u
v
_
=
_
x

x
y

y
_ _
U
V
_
These two equations con-
tradict each other and
are wrong
Why are these equations, (5.1.16) and (5.1.19), dierent? How can the u and v
transform in two dierent ways? One immediate conclusion that the equations are
wrong. We should be able to gure out what is wrong with these equations since
there are basically three terms involved. The left hand side of these two equations
are clearly ne since they are the quantities with which we start and are a given.
The chain rule part follows from calculus. That procedure looked right. That leaves
the U and V and of course, the = symbol. We want the equation relating velocities
in the two coordinate systems. That means there is a problem with the assumption
that the U and V in equation (5.1.16) are the same as the U and V in equation
(5.1.19). So, there may be two dierent kinds of U and V . To clear up these issues
study tensor calculus.[You93][Ari89][SS82]! We will do a very quick overview
here.
5.2. Tensor Calculus
Very often, we assume that a vector

V can be written in terms of a global basis
vectors e
1
, e
2
, e
3
as follows
(5.2.1)

V = v
1
e
1
+ v
2
e
2
+ v
3
e
3
=
3
i=0
v
i
e
i
We will see what we mean by a global basis as we go along. For now, do not confuse
the superscript on v with exponentiation. We deliberately chose superscripts and
subscripts since we anticipate that we are going to encounter two dierent kinds
of entities. We will see that superscripted entities are said to be contravariant and
subscripted entities are covariant. So, v
1
may be dierent from v
1
. We will see
what this means as we go along. If we agree that any time the index is repeated it
implies a summation, we can simply write
(5.2.2)

V = v
i
e
i
5.2. TENSOR CALCULUS
Now, THAT is compact. It is called Einsteins summation convention. It only
gets better. By itself, the equation does not even restrict us to three dimensions.
It is our assumption that we use three dimensions. In this book, we will restrict
ourselves to two / three dimensions. You should note that
(5.2.3)

V = v
i
e
i
= v
k
e
k
Since there is a summation over the index, the index itself does not survive the
summation operation. The choice of the index is left to us. It is called a dummy
index.
1
2
3
P
Q
x(P)
x(Q)
Figure 5.5. A Cartesian coordinate system used to locate the
point P and Q. x(P) gives the position vector of P in the Carte-
sian coordinate system. PQ forms a dierential element.
We now dene the notation with respect to coordinate systems. Consider Fig-
ure 5.5. It indicates a dierential line element with two points P and Q at each end
of the element. We dene x(.) as a coordinate function which returns the coordi-
nates of a point in the Cartesian coordinate system. If we had another coordinate
system overlayed on the same region, the point P will have the corresponding co-
ordinates

(P) in that coordinate system. The coordinate function is simple to
imagine if we look at it component-wise.
(5.2.4) x(P) = x
1
(P) e
1
+ x
2
(P) e
2
+ x
3
(P) e
3
Since we are dealing with Cartesian coordinates, x
i
and x
i
are the same. we have
already seen that if

P is the position vector for P then
(5.2.5) x
i
(P) =

P e
i
Consider the problem of coordinate transformations in two dimensions. Let us
restrict ourselves for the sake of this discussion to rotations. We take our standard
x y coordinate and rotate through an angle to get the coordinates. The
basis vectors in x y are e
1
and e
2
. The basis vectors in the coordinates are
1
and
2
. You can check that the basis vectors are related as follows:
(5.2.6)
_
2
_
=
_
cos sin
sin cos
_ _
e
1
e
2
_
We see that by using indices we can simply represent this as
(5.2.7)
i
= A
j
i
e
j
Now, a vector s can be represented in the xy and the coordinate systems
as
(5.2.8) s = s
i
e
i
=
i
i
Substituting for
i
from equation (5.2.7) we get
(5.2.9) s = s
i
e
i
= s
j
e
j
=
i
i
=
i
A
j
i
e
j
where i and j are dummy indices. Even though they are dummy indices, by the
proper choice of these dummy indices here we can conclude that
(5.2.10) s
j
=
i
A
j
i
= A
j
i
i
Compare equations (5.2.7) and (5.2.10). The unit vectors transform one way,
the components transform the opposite [ or contra ] way. We see that they too show
the same behaviour we saw with the velocity potential. Vectors that transform like
each other are covariant with each other. Vectors that transform the opposite way
are contravariant to each other. This is too broad a scenario for us. We will stick
with something simpler. Covariant entities will be subscripted. Contravariant
entities will be superscripted.
An example where this will be obvious to you is the case of the rotation of
the Cartesian coordinate system. Again, we restrict ourselves to two dimensions.
If you rotate the standard Cartesian coordinate system counter-clockwise, you see
that the coordinate lines and the unit vectors ( as expected ) rotate in the same
direction. They are covariant. The actual coordinate values do not change in the
same fashion. In fact, the new values corresponding to a position vector look as
though the coordinate system was xed and that the position vector was rotated in
a clockwise sense ( contra or opposite to the original coordinate rotation ). These
two rotations are in fact of equal magnitude and opposite in sense. They are, indeed,
inverses of each other. We will investigate covariant and contravariant quantities
more as we go along. Right now, we have assumed that we have a position vector.
Let us take a closer look at this.
We have made one assumption so far that the basis vector is global. We used
the term global basis in the beginning of this section. What do we mean by a
global basis? We want the basis to be the same, that is constant, at every point.
Such a set of basis vectors is also said to be homogeneous. For example, the basis
vectors in the standard Cartesian coordinate system do not depend on the (x, y)
coordinates of a point. Consider the trapezium in Figure (5.7) We see that the
tangent to the = constant coordinate lines change with . In general, the basis
vectors change from point to point. We do not have a global basis. Also, consider
e
1
1
1
2
2

1
x
2
x
2
Figure 5.6. The basis vectors rotate with the coordinates axes (
only e
1
and
1
are shown ). The coordinates of the point in the
new system are as though the point had moved clockwise and the
coordinate system was xed. That is x
2
< x
2
in this particular
case.
x
y
L
h
H
P
Figure 5.7. The basis vectors at the origin and the basis vectors
at some other point are clearly not the same. The position vector,
P is geometrically very easy to draw. It cannot be written sim-

ply as a linear combination of some basis vector in the coordinate
system shown here.
the standard polar coordinate system (see Figure 5.8). The usual symbols for the
basis vectors are e
and e
r
. Both of these vectors depend on . Again, for the
familiar and useful polar coordinate system, we do not have a global basis. That is
the basis vectors are not constant. They are not homogeneous. In fact, in the case
of polar coordinates we have as the position vector at any point

P = r e
r
. Does
the position vector not depend of at all? The fact of the matter is that the e
r
depends on , as the basis is not homogeneous. Fortunately, e
r
and e
depend only
on . So, we are still able to write

P = r e
r
.
x
y
e
r
e
P = r e
r
Figure 5.8. The position vector in polar coordinates is given by
P = r e
r
. At rst glance it seems as though there is no depen-
dence. However e
r
is a function of as is e
. Only the constant r

coordinate lines are shown.
Another example of a coordinate system with which you are familiar is used
doing log-log plots (see Figure 5.9). In this case, the basis vectors seem to be
oriented in the same fashion. However, the length of the vector seems to change.
It is clear that the notion of distance between points is an issue here.
Looking at these examples, we realise that we need to spend a little time trying
to understand generalised coordinate systems. Lets just consider one coordinate
line in some generalised coordinate system. We will see that in three dimensions,
it is a space curve. Let us rst look at space curves and some of their properties.
We are especially interested in the tangent to these curves since the tangent to the
coordinate line is a part of our basis.
Consider the coordinate line shown in Figure 5.10. The curve is determined by
a function (
1
). It is a coordinate line in a coordinate system labelled

which has
three components (
1
,
2
,
3
). The gure shows the coordinate line corresponding
x
y
P
Figure 5.9. A log-log coordinate system. We have a rectangu-
lar coordinate system, however the unit vectors are still a function
of position making it impossible to write the position vector drawn
P as a linear combination of the basis vectors.

to
2
=constant, and
3
=constant. To belabour the point, it is the
1
coordinate
line since it is parametrised on
1
. The tangent to this line is given by
(5.2.11)
1
=
d (
1
)
d
1
In fact, for any of the coordinate lines of
i
we have for the corresponding
i
(5.2.12)
i
=
d
i
d
i
, no summation on i
This basis vector is called the covariant basis vector. We note the following
In equation (5.2.12), though the subscripts are repeated, there is no sum-
mation implied over i. The subscript on the is there to conveniently
indicate three generic coordinate functions.
Some new tensor notation convention: ( see equation (5.2.12) ) the super-
script of the derivative on the on the right hand side becomes a subscript
on the left.
We consider an example to understand this process better. Let us take a look at
polar coordinates in two dimensions (
1
,
2
). A
2
= coordinate line corresponds
to a
1
= r =constant line. For the given r, the curve is parametrised as
(5.2.13) () = (
2
) = r cos() + r sin() =
1
cos(
2
) e
1
+
1
sin(
2
) e
2
As was seen earlier, e
1
and e
2
are the standard basis vectors in the Cartesian
coordinate system. You may be used to calling them and . The tangent to this
curve is
2
.
(5.2.14)
2
= r sin() e
1
+ r cos() e
2
=
1
sin(
2
) e
1
+
1
cos(
2
) e
2
We note two points here.
1
2
(
1
)
3

(
1
)
Figure 5.10. A coordinate line belonging to a three dimensional
generalised coordinate system. This line is shown embedded in our
usual Cartesian coordinate system. is shown as a function of
1
alone as the other two,
2
and
3
are held constant to obtain this
line. The local tangent vector is one of the basis vectors for the
generalised coordinates

2
is not a unit vector. If you normalise it, you get the physical basis
vector
.
e
1
and e
2
are not functions of
2
. That is the reason why we only have
two terms on the right hand side of equation 5.2.14. Otherwise we would
have had more derivative terms due to the application of product rule.
How about the other coordinate line corresponding to =
2
=constant. The
equation of such a line is given by
(5.2.15) (r) = (
1
) =
1
cos(
2
) e
1
+
1
sin(
2
) e
2
,
2
= constant
For constant
2
= , this will correspond to a radial line. The tangent vector to
this line is given by
(5.2.16)
1
=

1
= cos(
2
) e
1
+ sin(
2
) e
2
This in fact turns out to be a unit vector and is the same as
r
.
We can learn something from the study of the polar coordinate system. Why
does
2
depend on
1
?
2
is the angle measured from the x-axis. The angle
2
is
measured in radians which is the arc length at some radius that subtends
2
at the
centre nondimensionalised by that radius. Naturally, when using
2
as a coordinate,
the corresponding arc length depends on
1
.
3
X system
system
x
1
x
2
3
Q
P
d
=

(Q)
dx =

X(Q)
Figure 5.11. The origin of our Cartesian coordinate system is
moved to the point P. The dierential element PQ is now rep-
resented in terms of the translated coordinate system and similar
system of the generalised coordinates.
Lets pause and take stock of what we have and where we are. We have seen
that there are coordinate systems where the basis vectors are not homogeneous. So,
just writing a relation like equation (5.2.2),

V = v
i
e
i
, for a position vector

V may
not be possible. We will start dealing only with dierentials. A dierential element
PQ is shown in the Figure 5.11. It is represented in the X coordinate system
as dx = dx
i
e
i
. The e
i
are the basis vectors in this coordinate system. We can
transform from the X coordinates to the coordinates where the basis vectors are
i
. The dierential PQ can be written in the coordinates system as d
= d
i
i
.
How are the two representations for the given dierential element at a given point
related? Clearly, the length of the element should not depend on our choice of
the coordinate system. Or, put another way, if two people choose two dierent
coordinate systems, the length of this particular element should work out to be the
same. As we had done earlier, here are the two equations that relate the Cartesian
coordinates x
i
to the generalised coordinates
i
.
(5.2.17) x
i
= x
i
(
1
,
2
,
3
)
and
(5.2.18)
i
=
i
(x
1
, x
2
, x
3
)
In the Cartesian coordinate system the length ds is given by
(5.2.19) (ds)
2
= dx dx = dx
i
e
i
dx
j
e
j
= dx
i
dx
j
e
i
e
j
Remember that e
i
are the basis vectors of a Cartesian coordinate system and are
orthogonal to each other. Consequently, we can dene a useful entity called the
Kronecker delta as
(5.2.20)
ij
= e
i
e
j
=
_
1, i = j
0, i = j
With this new notation we can write
(5.2.21) (ds)
2
= dx dx = dx
i
dx
j
ij
= dx
i
dx
i
=
i
(dx
i
)
2
Following the convention we have used so far ( without actually mentioning it ) we
see that
(5.2.22) dx
j
ij
= dx
i
That is, j is a dummy index and disappears leaving i which is a subscript. For the
rst time we have seen a contravariant quantity converted to a covariant quantity.
If you think of matrix algebra for a minute, you will see that
ij
is like an identity
matrix. The components dx
i
are the same as the components dx
i
in a Cartesian
coordinate system. Hence, equation (5.2.21) can be written as
(5.2.23) (ds)
2
= dx
i
dx
i
=
i
(dx
i
)
2
The length of the element is invariant with transformation meaning the choice of
our coordinates should not change the length of the element. A change to the
coordinates should give us the same length for the dierential element PQ. The
length in the coordinates is given by
(5.2.24) (ds)
2
= d
= d
i
i
d
j
j
= d
i
d
j
i

j
= d
i
d
j
g
ij
= d
i
d
i
g
ij
is called the metric. Following equation (5.2.22), we have dened d
i
= g
ij
d
j
.
Why did we get g
ij
instead of
ij
? We have seen in the case of the trapezium that
the basis vectors need not be orthogonal to each other since the coordinate lines
are not orthogonal to each other. So, the dot product of the basis vectors
i
and
j
gives us a g
ij
with non-zero o-diagonal terms. It is still symmetric, though. In
this case, unlike the Cartesian situation, d
i
is dierent from d
i
.
We can dene another set of basis vectors which are orthogonal to the covariant
set as follows
(5.2.25)
i

j
=
j
i
where,
(5.2.26)
j
i
=
_
1, i = j
0, i = j
This new basis,
i
, is called the contravariant basis or a dual basis. This is
demonstrated graphically in gure 5.12. This basis can be used to dene a metric
(5.2.27) g
ij
=
i

j
3
Figure 5.12. The covariant basis vectors
1
,
2
, and
3
are shown.
In general they may not be orthogonal to each other.
3
is also
shown. It is orthogonal to
1
and
2
and
3

3
= 1
Now, is the denition given for d
i
consistent with this denition of the contravari-
ant basis? Is d
= d
i
i
? That is, if we take the dot product of a vector with a
basis vector, do we get the corresponding component? We have,
(5.2.28) d
= d
i
i
d

j
= d
i
i

j
. .
g
ij
= d
j
,
and
(5.2.29) d
= d
i
i
d

j
= d
i
i

j
= d
j
,
and
(5.2.30) d
= d
i
i
d

j
= d
i
i

j
= d
j
,
and nally,
(5.2.31) d
= d
i
i
d

j
= d
i
i

j
= d
j
,
So, to get the contravariant components of a tensor, dot it with the contravariant
basis vectors. Likewise, to get the covariant components of a tensor, dot it with
the covariant basis vectors. The eect of g
ij
on a contravariant term is to lower the
index or convert it to a covariant term. Similarly, the eect of g
ij
on a covariant
term is to raise the index or convert it to a contravariant term. So, what is g
ij
g
jk
?
(5.2.32) g
ij
g
jk
= g
k
i
=
i

j
=
k
i
The metric tensors are inverses of each other.
At this point you really can protest: Wait a minute, where is this going?
Fascinating as it is, how is it relevant to CFD? Look at the trapezium in Figure
5.4. Imagine that this trapezium represent a channel through which some uid, like
water, can ow. The top and bottom of the trapezium shown are solid walls. If
we were solving for the potential ow through a channel with the top and bottom
of the trapezium being solid walls, this tells us, we need to apply the boundary
condition /n = 0, where n is measured along a line that is perpendicular to the
surface. Look at the top of the trapezium. A zoomed view is shown in Figure 5.13.
Your coordinate line is not normal to the top surface. How do we get the derivative
along the normal. You can nd the derivatives along
1
and
2
and use Taylors
2
Figure 5.13. A zoomed view of the non-Cartesian mesh in a
Trapezoidal domain shown in Figure 5.4. The two covariant basis
vectors and one contravariant basis vector are shown.
series in two dimensions to get the normal derivative. You will nd that you are just
reinventing everything we have done so far. What you want is the contravariant
basis vector and not the covariant basis vector. Why? This is because the covariant
basis vector is along the coordinate line and the contravariant one is perpendicular
to it. The top of the trapezium is a coordinate line. The contravariant basis vector
is perpendicular to it, which is what we want. We do need this stu, so lets soldier
on. First, an assignment.
Assignment 5.1
(1) Expand the following using the summation convention assuming that we
are working in three dimensions.
(a) a
i
b
j
ij
, (b)
j
j
, (c)
j
i

i
j
, (d)
i
i

j
j
(2) Repeat the above problem assuming we are dealing with tensors in two
space dimensions.
(3) Find the covariant and contravariant bases vectors and the corresponding
metric tensors for the following coordinate systems. x
i
are the Cartesian
coordinates.
(a) Cylindrical coordinates.
1
= r,
2
= , and
3
= z in conventional
notation.
x
1
=
1
cos
2
x
2
=
1
sin
2
x
3
=
3
.
(b) Spherical coordinates.
1
= R,
2
= , and
3
= in conventional
notation.
x
1
=
1
sin
2
cos
3
x
2
=
1
sin
2
sin
3
x
3
=
1
cos
2
(c) Parabolic cylindrical coordinates.
x
1
=
1
2
_
_
1
_
2
2
_
2
_
x
2
=
1
2
x
3
=
3
.
(4) Compute the covariant and contravariant velocity components in the above
coordinate systems.
You have seen in multivariate calculus that given a smooth function , in a
region of interest, we can nd the dierential d as
(5.2.33) d =

i
d
i
Now, we also know that this is a directional derivative and can be written as
(5.2.34) d = d
i
d
i
where,
(5.2.35) =
j

j
, d
=
i
d
i
We managed to dene the gradient operator . What happens when we take the
gradient of a vector? How about the divergence? We rst write the gradients of a
scalar function and a vector function as
=
j

j
(5.2.36)
V =
j

j
(5.2.37)
If we look carefully at the two equation above, we see that equation (5.2.37) is
dierent. It involves, due to the use of product rule, the derivatives of the basis
vectors. In fact, equation (5.2.37) can written as
(5.2.38)
V =
j

j
=
j
_
v
i
i
+ v
i

i
j
_
So, what is the nature of the derivative of the basis vector? For one thing, from
the denition of the covariant basis in equation (5.2.12) we have
(5.2.39)

i
j
=

2

i
=

j
i
We have dispensed with the subscript on so as not to create more confusion.
We will use the correct corresponding to the coordinate line. We can see from
equation (5.2.39) that its component representation is going to be symmetric in
the two indices i and j. As we have already seen in equation (5.2.28), to nd the
contravariant components of this entity we can dot it with
k
to get
(5.2.40)
_
k
ij
_
=
k
j
_
k
ij
_
is called a Christoel symbol of the second kind. We took the dot product
with
k
so that equation (5.2.38) can be rewritten as
(5.2.41)
V =
j
_
v
i
i
+ v
i
_
k
ij
_
k
_
Since i and k are dummy indices (meaning we are going to sum over their values)
we swap them for a more convenient expression
(5.2.42)
V =
j
_
v
i
i
+ v
k
_
i
kj
_
i
_
This allows us to write
(5.2.43)

j
=
_
v
i
j
+ v
k
_
i
kj
__
i
In pure component form this is written as
(5.2.44) v
i
;j
=
v
i
j
+ v
k
_
i
kj
_
This is called the covariant derivative of the contravariant vector v
i
. Staying with
our compact notation, the covariant derivative is indicated by the semi-colon in the
subscript. This is so that we do not confuse it with the plain derivative v
i
/
j
.
So, if we have Christoel symbols of the second kind do we have any other
kind? Yes, there is a Christoel symbol of the rst kind. It is written as [ij, k] and
it is given by
(5.2.45) [ij, k] =
_
l
ij
_
g
lk
=

i
j

l
g
lk
=

i
j

k
The Christoel symbols of the rst kind can be directly obtained as
(5.2.46) [ij, k] =
1
2
_
g
jk
i
+
g
ki
j

g
ij
k
_
This can be veried by substituting for the denition of the metric tensor. The
peculiar notation with brackets and braces is used for the Christoel symbols (and
they are called symbols) because, it turns out that they are not tensors. That is,
though they have indices, they do not transform the way tensors do when going from
one coordinate system to another. We are not going to show this here. However,
we should not be surprised that they are not tensors as the Christoel symbols
encapsulate the relationship of the two coordinate systems and would necessarily
depend on the coordinates.
The divergence of

V is dened as the trace of the gradient of

V . That is
(5.2.47) div
V =
j
_
v
i
i
+ v
k
_
i
kj
_
i
_
Assignment 5.2
For the coordinate systems given in assignment 5.1,
(1) Find the Christoel symbols of the rst and second kind.
(2) Find the expression for the gradient of a scalar potential.
(3) Find the gradient of the velocity vector.
(4) Find the divergence of the velocity vector.
(5) Find the divergence of the gradient of the scalar potential that you just
found.
5.3. EQUATIONS OF FLUID MOTION
In the case of the velocity potential

V = we get,
(5.2.48)

V =
j

j
=
k
g
kj

j
=
k
g
kj
v
j
= v
k
k
If we now take the divergence of this vector using equation (5.2.47) we get
(5.2.49)
2
=
j
_
v
i
i
+ v
k
_
i
kj
_
i
_
=
j
j
_
g
il

l
_
i
+ v
k
_
i
kj
_
i
_
Completing the dot product we get
(5.2.50)
2
=
_

i
_
g
il

l
_
+ v
k
_
i
ki
__
Substituting for v
k
from equation (5.2.48) we get
(5.2.51)
2
=
_

i
_
g
il

l
_
+ g
kl

l
_
i
ki
__
This much tensor calculus will suce. A more in depth study can be made
using the numerous books that are available on the topic [You93], [SS82].
5.3. Equations of Fluid Motion
We have seen enough tensor calculus so that if we derive the governing equations
in some generic coordinate system, we can always transform the resulting equations
into any other coordinate system. In fact, as far as possible, we will derive the
equations in vector form so that we can pick the component form that we feel is
appropriate for us. We can conveniently use the Cartesian coordinate system for
the derivation with out loss of generality.
We will rst derive the equations of motion in integral form. We will do this in
a general setting. Let us consider some uid property Q, whose property density
is given by Q. For example, consider a situation in which we have added some ink
to owing water. At any given time, the mass of ink in a small elemental region of
interest may be dm
ink
. If the volume of the elemental region is d, then these two
measures dened on that region are related through the ink density as
(5.3.1) dm
ink
=
dm
ink
d
d =
ink
d
We would like to write out the balance laws for a general property, Q. We
arbitrarily pick a control volume. One such volume is indicated in the Figure
5.14. For the sake of simplicity, we pick a control volume that does not change in
time. This control volume occupies a region of volume . This control volume has
a surface area S. It is located as shown in the gure and is immersed in a ow eld.
Within this control volume, at an arbitrary point x, we pick a small elemental
region with volume d. From equation (5.3.1), the amount of the property of
interest at time t, dQ(x, t), in the elemental control volume is Q(x, t)d. Then the
total quantity contained in our control volume at any instant is
(5.3.2) Q
(t) =
_
Q(x, t)d
dS
d
2
3
1
x
n
V
Figure 5.14. An arbitrary control volume chosen in some uid
ow. An elemental area on the controls surface dS and and ele-
mental volume d within the control volume are also shown. Note
that in most coordinate systems we may not be able to indicate a
position vector x.
The time rate of change of this quantity is
(5.3.3)
dQ
dt
=
d
dt
_
Q(x, t)d
Then we ask ourselves the question, why is there a rate of change? There is change
because the property Q is carried / transported in and out of the control volume by
the uid. It is also possible, based on the nature of Q, that it is somehow created
or destroyed in the control volume. There may be many mechanisms by which Q
can be changed into some other property. Let us now look at the transport of Q
by the ow.
At any arbitrary point on the surface of the control volume that we have shown
in Figure 5.14, we can determine the unit surface normal vector. We can pick a
small elemental area dS at that point. The surface normal is perpendicular to this
element. By convention, we choose to pick a surface normal that points out of the
control volume. The rate at which our property Q ows out through this elemental
area is given by Q
V ndS. The total eux (outow) from the control volume is

(5.3.4)
_
S
Q
V ndS
Since this is a net eux, it would cause a decrease in the amount of Q contained
in the control volume. So, our balance law can be written as
(5.3.5)
d
dt
_
Qd =
_
S
Q
V ndS + any other mechanism to produce Q

Before going on we will make the following observation. Though the control volume
can be picked arbitrarily, we will make sure that it is smooth enough to have surface
normals almost everywhere. Almost everywhere? If you think of a cube, we cannot
dene surface normals at the edges and corners. We can break up any surface
integral in equation (5.3.5) into the sum of six integrals, one for each face of the
cube.
5.3.1. Conservation of Mass. Let us look at an example. If the property
we were considering was mass, Q
(t) would be the mass of uid in our control

volume at any given time. The corresponding Q would be mass density which we
routinely refer to as the density, . Ignoring mechanisms to create and destroy or
otherwise modify mass, we see that the production terms disappear, leaving only
the rst term on the right hand side of equation (5.3.5). This gives us the equation
for balance of mass as
(5.3.6)
d
dt
_
d =
_
S
V ndS
This equation is also called the conservation of mass equation.
5.3.2. Conservation of Linear Momentum. On the other hand, if we con-
sider the property Q to be momentum, the property density Q turns out to be
V ,
which is the momentum density. In this case, we know that the total momentum
in the control volume can also be changed by applying forces. For the sake of this
discussion, forces come in two avours. There are those that correspond to action
across a distance, these forces are often called body forces. The others that depend
on proximity are called surface forces
1
. We can write our equation of balance of
linear momentum as
(5.3.7)
d
dt
_
V d =
_
S
V

V ndS +
_
fd +
_
S
TdS
Here,

f(x) is the body force per unit volume at the point x within the control
volume.

T(x) is the traction force per unit area acting at some point x on the
control surface. If we are willing or able to ignore the body force, we are left with
the traction force to be handled. From uid mechanics, you would have seen that
we can associate at a point, a linear transformation called the stress tensor, which
relates the normal to a surface element to the traction force on that element. That
is
(5.3.8)

T = n
1
As with everything that we do in physics, what we mean by this really depends on length
scales. We have assumed that we are dealing with a continuum and that implicitly has a bifurcation
of the length scales built into it.
where,

T = T
i
i
, =
ij
j
, and n = n
k
k
. This gives us the Cauchy equation
in component form as
(5.3.9) T
i
=
ij
n
j
The momentum balance equation can be written as
(5.3.10)
d
dt
_
V d =
_
S
V

V ndS +
_
S
ndS
Combining terms we get
(5.3.11)
d
dt
_
V d =
_
S
_
V

V
_
ndS
5.3.3. Conservation of Energy. Finally, if we consider the total energy as
the property of interest so that we write out the balance law for energy. Considering
the form of the rst two equations, we will dene the total energy density as E
t
,
where E
t
is the specic total energy dened as
(5.3.12) E
t
= e +
1
2
V

V ,
Where e is the specic internal energy dened for a perfect gas as e = C
v
T. C
v
is the specic heat at constant volume and T is the temperature measured on the
Kelvin scale. We need to look at the production terms again in equation (5.3.5).
The total energy in our control volume can be changed by
(1) the forces from the earlier discussion doing work on the control volume,
(2) the transfer of energy by the process of heat through radiation and con-
duction,
(3) the apparent creation of energy through exo-thermic or endo-thermic
chemical reactions,
(4) and nally, of course, the transportation of energy across the control sur-
face by the uid.
We will ignore radiation and chemical reactions here. This results in the balance
law
(5.3.13)
d
dt
_
E
t
d =
_
S
E
t
V ndS +
_
f

V d +
_
S
T

V dS
_
S
q ndS
Here, q is the term quantifying heat. Again, if we are in a position to ignore body
forces we get
(5.3.14)
d
dt
_
E
t
d =
_
S
E
t
V ndS +
_
S
V ndS
_
S
q ndS
which we conveniently rewrite incorporating the other balance laws as
(5.3.15)
d
dt
_
Qd =
_
S
F ndS
where we have
(5.3.16) Q =
_
_
_
V
E
t
_
_
_
,

F =
_
_
_
V

V
(E
t
)
V

V q
_
_
_
where,
V is the rate at which the traction force does work on the control volume.
This, gives us a consolidated statement for the balance (conservation) of mass,
linear momentum, and energy. The great thing about this equation is that it can
be cast in any three dimensional coordinate system to get the component form. It
is written in a coordinate free fashion. Though, it is good to admire, we nally
need to solve a specic problem, so we pick a coordinate system convenient for the
solution of our problem and express these equations in that coordinate system. The
other problem is that as it is there is some element of ambiguity in the dot products
of the form (
V ) n. These ambiguities are best resolved in terms of components.

(5.3.17)

T

V = T
i
i

l
V
l
=
ij
i
(
j

k
)n
k

l
V
l
=
ij
n
j
V
i
The dierential form of equation (5.3.15) can be obtained by applying the theorem
of Gauss to the right hand side of the equation and converting the surface integral
to a volume integral.
(5.3.18)
_
_
Q
t
+ div
F
_
d = 0
The control volume is chosen arbitrarily. As a consequence, the integral needs to
be zero for any over which we integrate. This is possible only if
(5.3.19)
Q
t
+ div
F = 0
The form of equation (5.3.15) is quite general. We could add, as required, more
terms to the

F on the right hand side. We could also add as many equations as
required. If you have other properties that need to be tracked, the corresponding
equations can be incorporated. However, for our purpose, these equations are quite
general. We will start with a little specialisation and simplication.
We now decompose the stress tensor into a spherical part and a deviatoric
part. The spherical part we will assume is the same as the pressure we have in the
equation of state. The deviatoric part [ or the deviation from the sphere ] will show
up due to viscous eects. So, can be written as
(5.3.20) = p1 +
1 is the unit tensor and is the deviatoric part. Do not confuse a tensor with
the control volume . Through thermodynamics, we have an equation of state /
constitutive model for p. Typically, we use something like p = RT, where T is the
temperature in Kelvin and R is the gas constant. We need to get a similar equation
of state / constitutive model for . Assuming the uid is a Navier-Stokes uid,
that is the uid is Newtonian, isotropic and Stokes hypothesis holds we get
=
2
3
trD+ 2D, where (5.3.21)
D =
1
2
(L+L
T
), and (5.3.22)
L =
V (5.3.23)
where is the coecient of viscosity and trD is the trace of D. Which is the
sum of the diagonals of the matrix representation of the tensor. Since we are right
now looking at inviscid ow, we can ignore the viscous terms. So, for the Eulers
equation we have
(5.3.24)

T = p1 n
where, 1 is the unit tensor. The Eulers momentum conservation equation can be
written as
(5.3.25)
d
dt
_
V d =
_
S
V

V ndS
_
S
p1 ndS
Combining terms we get
(5.3.26)
d
dt
_
V d =
_
S
_
V

V + p1
_
ndS
which we conveniently rewrite as
(5.3.27)
d
dt
_
Qd =
_
S
F ndS
where we have
(5.3.28) Q =
_
_
_
V
E
t
_
_
_
,

F =
_
_
_
V

V + p1
(E
t
+ p)
V
_
_
_
giving us a consolidated statement for the conservation [ or balance ] of mass, linear
momentum, and energy. These equations are collectively referred to as the Eulers
equation. There are, as is usual, a set of auxiliary equations to complement these
equations. The constitutive model given by the equation of state is
(5.3.29) p = RT
and
(5.3.30) E
t
= e +
V

V
2
(5.3.31) e = C
v
T
With these added equations we have a closed set of equations that we should
be able to solve. The equations are in integral form. We can employ the theorem
of Gauss on the surface integral in equation (5.3.27) and convert it to a volume
integral like so
(5.3.32)
d
dt
_
Qd =
_
S
F ndS =
_
div
Fd
This gives us the following equation
(5.3.33)
_
_
Q
t
+ div
F
_
d = 0
which is valid for all possible control volumes on which we have surface normals
and can perform the necessary integration. Remember, this particular was
chosen arbitrarily. We conclude that the integral can be zero for any only if the
integrand is zero. The dierential form of the Eulers equation can be written as
(5.3.34)
Q
t
+ div
F = 0
If we use normal convention to write

F in Cartesian coordinates as
(5.3.35)

F = E + F + G
k
our governing equation in Cartesian coordinates then becomes
(5.3.36)
Q
t
+
E
x
+
F
y
+
G
z
= 0
Clearly, given any other basis vector, metrics, Christoel symbols, we can write the
governing equations in the corresponding coordinate system.
Assignment 5.3
(1) Given the concentration of ink at any point in a ow eld is given by c
i
,
derive the conservation equation in integral form for ink. The diusivity
of ink is D
i
.
(2) From the integral from in the rst problem, derive the dierential form
(3) Specialise the equation for a two-dimensional problem.
(4) Derive the equation in polar coordinates.
5.3.4. Non-dimensional Form of Equations. So far, in this book, we have
not talked of the physical units used. How do the equations depend on physical units
that we use. Does the solution depend on the fact that we use millimetres instead
of metres? We would like to solve the non-dimensional form of these equations. We
will demonstrate the process of obtain the non-dimensional form of the equation
using the two-dimensional Eulers equation written in Cartesian coordinates.
(5.3.37)
Q
t
+
E
x
+
F
y
= 0
To this end, we dene the following reference parameters and relationships.
It should be noted that the whole aim of this choice is to retain the form of the
equations.
We have a characteristic length L in the problem that we will use to scale
lengths and coordinates. For example
(5.3.38) x
=
x
L
, and y
=
y
L
We employ a reference density
r
and a reference pressure p
r
to non-dimensionlise
the density and the pressure. As a result we get the non-dimensionalisation for the
temperature through the equation of state.
(5.3.39)
r
, and p
=
p
p
r
, along with p = RT gives T
=
T
T
r
,
where,
(5.3.40) T
r
=
p
r
r
R
,
and the equation of state reduces to
(5.3.41) p
Consider the one-dimensional energy equation from gas dynamics. This relation
tells us that
(5.3.42) C
p
T
o
= C
p
T +
V
2
2
If we divide this equation through by T
r
and nondimensionalise speed with a
reference speed u
r
we get
(5.3.43) C
p
T
o
= C
P
T
+
V
2
u
2
r
2T
r
Now we see that if we dene
(5.3.44) u
r
=
_
RT
r
equation (5.3.43) reduces to
(5.3.45)

1
T
o
=

1
T
+
V
2
2
Now, consider the rst equation, conservation of mass, from equations (5.3.37).
This becomes
(5.3.46)

r
+

r
u
r
L
+

r
u
r
L
= 0
where is some characteristic time scale to be dened here. Dividing through by
r
u
r
and multiplying through by L, we get
(5.3.47)
L
u
r
= 0
Clearly, if we dene the time scale = L/u
r
we get back our original equation.
I will leave it as an exercise in calculus for the student to show that given the
following summary
x
=
x
L
, and y
=
y
L
(5.3.48)
r
, and p
=
p
p
r
, (5.3.49)
T
r
=
p
r
r
R
, and u
r
=
_
RT
r
(5.3.50)
equation (5.3.37) reduces to
(5.3.51)
Q
+
E
+
F
= 0
where
(5.3.52)
Q
=
_
t
_
_
, E
=
_
u
2
+ p
t
+ p
)u
_
, and F
=
_
v
2
+ p
t
+ p
)v
_
Very often for the sake of convenience the stars are dropped. One has to
remember that though these basic equations have not changed form. Others have
changed form. The equation of state becomes p
and the energy equation

changes form. Any other auxiliary equation that you may use has to be nondimen-
sionalised using the same reference quantities.
A careful study will show you that if L, p
r
and
r
are specied then all the other
reference quantities can be derived. In fact, we typically need to x two reference
quantities along with a length scale and the others can be determined. The other
point to note is that we typically pick reference quantities based on the problem at
hand. A review of dimensional analysis at this point would be helpful.
Assignment 5.4
(1) Non-dimensionalise the Eulers equation in the dierential form for three-
dimensional ows.
(2) Try to non-dimensionalise the Burgers equation
(5.3.53)
u
t
+ u
u
x
=
2
u
x
2
u does not have the units of speed.

Tensors and The Equations of Fluid Motion

Uploaded by

Copyright:

Available Formats

Tensors and The Equations of Fluid Motion

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

Tensors and The Equations of Fluid Motion

Uploaded by

Copyright:

Available Formats

CHAPTER 5

Tensors and the Equations of Fluid Motion

P is geometrically very easy to draw. It cannot be written sim-

. Only the constant r

P as a linear combination of the basis vectors.

V ndS. The total eux (outow) from the control volume is

V ndS + any other mechanism to produce Q

(t) would be the mass of uid in our control

V ) n. These ambiguities are best resolved in terms of components.

and the energy equation

You might also like