Chapter 1

Mathematical preliminaries

In this chapter we quickly review the mathematical preliminaries required for un-
derstanding the rest of the material. This is, by no means, an exhaustive chapter
T
in vector calculus. At the end, advanced version of these preliminaries is presented
briefly to catalyze further reading. Such advanced discussions are mandatory for
AF
developing advanced theories in generalized curvilinear coordinates.
DR

1.1 Index notation

Theoretical development is concise in Cartesian coordinates x, y, z and can be
written using a short-hand notation called the index notation. Replacing x, y, z
with x1 , x2 , x3 or xi where the index i takes values 1, 2, 3. Similarly, in index
notation, unit vectors along the axes are denoted by êi , components of vector v
by vi , components of a second order tensor T by Tij and so on. The need for this
short-hand can be motivated by the following example. Consider right-handed
orthogonal axis ê1 , ê2 , ê3 and a transformed set ê′1 , ê′2 , ê′3 . A vector v can be
denoted as,
v = v1 ê1 + v2 ê2 + v3 ê3 = v1′ ê′1 + v2′ ê′2 + v3′ ê′3 . (1.1.1)

Now, to compute v1′ · · · components in terms of v1 · · · components, we will have

to write three independent equations taking dot products with ê′1 · · · each sepa-
rately. To avoid this, accepting a few rules of index notation, this relation can be
concisely written as simply,
 
vi′ = vj êj · ê′i . (1.1.2)

3
 Advanced Fluid Mechanics by Prateek Gupta

These rules are listed below with a few examples.

Summation Convention: If an algebraic index (say i) appears exactly twice in

a term, then it implies a summation over the range of that index (i) by assigning
values 1, 2, 3 to it (i), i.e.,

3
ai b i = ai bi , v = vi êi , · · · (1.1.3)
X

i=1

The repeated algebraic index is called a dummy index since it can be replaced by
any convenient algebraic index which does not appear in that term. Note that
repetition is defined here as the index appearing twice. Hence, the same index
appearing thrice in a term or a single expression is ambiguous and not allowed.
Hence, Eq. 1.1.2 can be written as all of the following,
 
3      
vi′ = vj êj  · ê′i = vj êj · ê′i = vk êk · ê′i = vl êl · ê′i · · · (1.1.4)
X

j=1
T
Same free index on both sides: On both sides of an equation, the free indices
AF
must be identical. In Eq. 1.1.2, i is the free index and j is the dummy index.
Following are some valid expressions using index notation,

aij bj = uik ck , vi = aij bj , Sij = Cijkl Tkl , · · · (1.1.5)

DR

Kronecker’s delta δij is defined by


1 if i = j
δij = ⇒ δ11 = δ22 = δ33 = 1, δ12 = δ23 = δ31 = δ21 = δ32 = δ13 = 0
0 if i =
̸ j
(1.1.6)

Note that δij = δji and êi ·êj = δij is a compact relation for the nine dot-products.
If a term with index i is multiplied with δij , then index i in the term is substituted
by j, since the only non-zero term in the sum is for i = j :
3
T...i... δij = T...i... δij = T...j... (1.1.7)
X

i=1

∂xi
For example, δik ai = ak , Akj δik = Aij , δij δij = δii = 3. Note that
= δij . The
∂xj
dot-product of vectors u = ui êi and v = vi êi can be expressed in index form
as u · v = ui vj êi · êj = ui vi .

4
 Chapter 1: CHAPTER 1. MATHEMATICAL PRELIMINARIES

1.2 Gauss’s theorem

Consider the volume as shown in figure 1.1 with one of its surface projected on
x1 − x3 plane. Consider there is field u(x) existing in the space and we require
to prove, I Z
u · n̂dS = ∇ · udV. (1.2.1)
∂S V

Let us only consider u2 component of the field, for which the above statement of
Gauss law takes the form,
Z Z
∂u2
u2 n̂2 dS = dV, (1.2.2)
∂x2
R V

where the surface integration is done on the projected surface R Consider the
RHS of the above equation, which can be written as,

ZZ f (xZ1 ,x3 )
∂u2
∂x2
dx2 dx1 dx3 =
ZZ T
(u2 (x1 , f (x1 , x3 ), x3 ) − u2 (x1 , g(x1 , x3 ), x3 )) dx1 dx3 .
AF
g(x1 ,x3 )
(1.2.3)
Here, we would like to remark that it is important to consider the total differen-
tial as an integrand. Often times, students look at the left hand side of eq. 1.2.1
DR

and think that only quantity immediate to the normal vector needs to be differ-
entiated. In such a case, eq. 1.2.3 will have additional terms due to integration
by parts. As we will see below, such terms will show that Gauss’s divergence
theorem doesn’t hold for such a partial derivative.

Consider the area integral now. The normal of a surface given by an implicit
equation F (x1 , x2 , x3 ) = 0 is given by,

∇F
n̂ = ± . (1.2.4)
|∇F |

For the two parameterized surfaces, this normal becomes,

∂f
− ê2 + ∂x
ê
∂x1 1
∂f
ê3
n̂ = ± 3
. (1.2.5)
|∇(f (x1 , x3 ) − x2 )|

With the area magnitude, the quantity n̂2 dS becomes,

n̂2 dS = ±dx1 dx3 , (1.2.6)

5
Advanced Fluid Mechanics by Prateek Gupta

Figure 1.1: A closed volume with parameterized surfaces x2 = f and x2 = g.

Projection of the surface on x2 plane is shown by R.

where the sign is decided based on top or bottom surface. Substituting in the
LHS of equation 1.2.2, we obtain,
ZZ ZZ
u2 dx1 dx3 − u2 dx1 dx3 , (1.2.7)
x2 =f (x1 ,x3 ) x2 =g(x1 ,x3 )

which is equal to the RHS of eq.1.2.3, thus proving the Gauss’s divergence theorem
T
for only one component of the field u. Similarly, other components can be shown
using the parameterized surfaces of the volume shown.
AF
Using the above, there are following corollaries (not the only ones) of the diver-
gence theorem which maybe useful,

• for u → gu
DR

Z I
∇ · (gu)dV = gu · n̂dS, (1.2.8)
∂V
V

• for u → c × u where c is a constant field

Z Z I I
∇ · (c × u)dV = − c · ∇ × udV = (c × u) · n̂dS = c · u × n̂dS,
∂V ∂V
V V
(1.2.9)
or Z I
∇ × udV = u × n̂dS, (1.2.10)
∂V
V

1.3 Stokes’s theorem

In a similar way, Stokes’s theorem can be proved which relates the integral along a
curve with the integral on the surface bounded by that curve. Consider a curved
surface A with bounding curve C in fig. 1.2. Let t̂ be unit tangent vector in
direction s along the curve C, n̂a ⊥ t̂ is in the tangent plane of A and n̂c is

6
 Chapter 1: CHAPTER 1. MATHEMATICAL PRELIMINARIES

Figure 1.2: A bounded surface A by the contour C.

defined as t̂ × n̂a . Stokes’s theorem states that,

Z Z
u · t̂ds = (∇ × u) · n̂dA. (1.3.1)
C A

Instead of outlining the full proof (refer to Kreyzig for that), we instead discuss
how surface integrals in general are represented. This becomes useful when dis-
T
cussing surface tension. To define a surface integral, take a surface S given by
parametric representation as
AF
r(u, v) = xi (u, v)êi . (1.3.2)

Assume a curve on the surface with parameter t. Then u(t) and v(t) define the
curve. Moreover,
DR

dr(t) ∂r du ∂r dv
= + = r u u′ + r v v ′ . (1.3.3)
dt ∂u dt ∂v dt

Assuming the partial derivatives r u and r v are linearly independent (which means
u and v are not parallel), the normal vector N is obtained as N = r u × r v .
Corresponding unit normal to the surface is defined as,

ru × rv
n= . (1.3.4)
|r u × r v |

The elemental area dS (or dA in figure 1.2) can then be obtained as dS = |r u ×

r v |dudv.

1.4 Tensors
Tensors are mathematical objects which are defined using transformation rules
under coordinate transforms. For right-handed orthogonal axis ê1 , ê2 , ê3 and a

7
Advanced Fluid Mechanics by Prateek Gupta

transformed set ê′1 , ê′2 , ê′3 , let aij be defined as,

aij = ê′i · êj . (1.4.1)

Then for an nth order tensor T with components Tijk · · · defined as,
| {z }
n

T = Tijk··· êi êj êk · · · , (1.4.2)

the components transform as,

′
Tpqr··· = api aqj ark · · · Tijk··· . (1.4.3)

Scalars can be identified as zeroth order tensors which have the same value under
coordinate transformation (for example temperature). Vectors can be identified
as first order tensors. In fluid mechanics, we often come across second order
tensors such as stress, strain rate, velocity gradient etc.
T
While considering constitutive modeling, we will look at an isotropic fourth order
AF
tensor. An isotropic tensor is a tensor whose components remain unaffected under
coordinate rotations. By definition, a scalar is an isotropic tensor. An isotropic
first order tensor (vector) is 0 (a non-zero vector points in some direction so
naturally changes with coordinate rotation). A second order isotropic tensor is
DR

essentially a diagonal matrix with equal values on the diagonal Aij = λδij . A
third order isotropic tensor has components proportional to ϵijk also known as
the permutation symbol. It is defined as,





1if ijk = 123, 231, 312 cyclic order
ϵijk = −1 if ijk = 321, 213, 132 anti-cyclic order (1.4.4)

 0 if at least two indices are same



Note the following :-

1. An index on ϵijk can be moved two places to right/left without changing its
value.

2. For movement of an index by one place to right/left, the sign is reversed.

3. A useful epsilon-delta identity is:

ϵijk ϵilm = δjl δkm − δjm δlk . (1.4.5)

8
 Chapter 1: CHAPTER 1. MATHEMATICAL PRELIMINARIES

4. Sometimes, cross-product is used to define the permutation symbol as,

ϵijk êk = êi × êj . (1.4.6)

5. The components of a vector c = a × b can be expressed as,

c = ci êi = aj bk êj × êk = aj bk ϵjki êi → ci = ϵjki aj bk = ϵijk aj bk . (1.4.7)

The components of a fourth order isotropic tensor Aijkl can be defined using the
linear combination of δij δkl , δil δjk , and δik δjl .

1.4.1 Tensor calculus

The most useful tool of tensor calculus for us is the Nabla operator defined using
index notation as,
∂
∇ = êi
∂xi
.
T (1.4.8)

Note that the above definition is naturally restricted to the Cartesian coordinate
AF
systems. In cylindrical coordinates, the definition is,

∂ êϕ ∂ ∂
∇ = êr + + êz . (1.4.9)
∂r r ∂ϕ ∂z
DR

∂êr ∂êϕ
Note that = êϕ and = −êr .
∂ϕ ∂r
The gradient of a field f (x) is defined as ∇f .
∂f
• For a scalar field f , ∇f = êi . The reader is encouraged to show that it
∂xi
transforms as a vector.
∂fj ∂fj
• For a vector field f , ∇f = êi ⊗ êj = êi êj . The reader is encouraged
∂xi ∂xi
to show that it transforms as a second order tensor

For tensor fields of order equal or higher than 1, divergence and curl are defined
as,
∂Tijk··· ∂Tijk···
1. Divergence : ∇ · T = êp · êi êj êk · · · = êj êk · · ·
∂xp ∂xi
∂Tijk··· ∂Tijk···
2. Curl : ∇ × T = êp × êi êj êk · · · = ϵpiq êq êj êk · · ·
∂xp ∂xp
Fields for which the divergence is 0, ∇ · T = 0 are called solenoidal. Similarly,
when ∇×T = 0, the field T is called irrotational. There are further classifications

9
Advanced Fluid Mechanics by Prateek Gupta

of fields (specifically for vectors),

1. If ∇ · a = 0 and ∇ × a = 0, then a = ∇ϕ where ϕ is a potential function.

2. Complex lamellar field : a · (∇ × a) = 0. (Note that all vector fields are

not complex lamellar or the curl is not always normal to the vector field).

3. Beltrami field : a × (∇ × a) = 0

According to Helmholtz’ theorem, a continuous vector field which vanishes at

infinity can always be written using three scalar functions as ,

a = ∇ϕ + ∇ × (ψ∇χ) , (1.4.10)

or a scalar potential function and a vector potential function,

a = ∇ϕ + ∇ × A. (1.4.11)

T
For a tensor field (of any order), Taylor’s expansion about a point x0 in the
direction n̂ yields,
AF
T (x0 + n̂dr) = T (x0 ) + n̂dr · ∇T + ∇ (∇T ) : n̂ ⊗ n̂(dr2 ) + · · · , (1.4.12)
x0 x0

In the limit dr → 0, the above expression shows that n̂ · ∇T is the directional

DR

x0
derivative of the field T at x0 in direction n̂.

Often, derivatives of the form u · ∇ appear in fluid mechanics where u is the

fluid velocity. This can be interpretted as a directional derivative along the fluid
velocity or the streamlines (curves always parallel to velocity directions).

10
 Chapter 1: CHAPTER 1. MATHEMATICAL PRELIMINARIES

Advanced tensor calculus 1.1 Optional

Tensor calculus is a vast field. One of the key assumptions made in un-
dergraduate courses is that the unit vectors êi are constant in space. In
generalized curvilinear coordinates, this is not the case. Two ubiquitous
examples of curvilinear coordinates are the cylindrical coordinates and the
spherical coordinates. Specifically, these are examples of orthogonal curvi-
linear coordinates. In this box, we will work out a sample calculation using
generalized orthogonal curvilinear coordinates. For non-orthogonal coordi-
nates, the reader is referred to Aris [2012].

1.5 Advanced summation convention

Any index repeated once in the upper position and once in the lower posi-
tion in a product is called a dummy index and summation over its values
is implied. Any index not repeated is called the free index.
T
Additionally, any free index must appear in the same position in each term
of an equation. Formula
AF
k = Bip Cq D ,
Aij (1.5.1)
pj q

is invalid.
The most important rule of this advanced index notation is that an upper
DR

affix in a denominator counts as a lower index in the numerator and vice-

∂
versa. This Di = is a valid symbol.
∂xi

1.6 Contravariant and covariant vectors

Let xj be coordinates and ξ i be transformed (general curvilinear coordi-
nates). The differentials are related as,

∂ξ i j
dξ =
i
dx . (1.6.1)
∂xj

These differentials are termed as the contravariant vectors. Any vector

components ai are called the components of a contravariant vector at a
certain point in the xj coordinate system, if under a transformation to ξ i
coordinate system, the components become,

∂ξ i j
ãi = a. (1.6.2)
∂xj

11
Advanced Fluid Mechanics by Prateek Gupta

Optional
For a moving particle, the velocity components v i = dxi /dt transform as
ṽ i = dx̃i /dt which are related as,

dx̃i ∂ x̃i dxj ∂ x̃i j

ṽ i = = = v . (1.6.3)
dt ∂xj dt ∂xj

Hence, velocity is a contravariant vector.

For a scalar function f (x1 , x2 , x3 ), derivative transform according to the
equation,
∂f ∂xj ∂f
= , (1.6.4)
∂ x̃i ∂ x̃i ∂xj
or
∂xj
ãi = aj . (1.6.5)
∂ x̃i
In this case, ai are components of a covariant vector in the coordinate
system xi , since they transform according to Eq. (1.6.5). An astute reader
would have already noticed that the affix in covariant components is a lower
T
affix, since the transformation of these vectors is similar to vectors defined
through scalar gradients. In an intuitive sense, contravariant vectors trans-
AF
form as tangents to curve would transform and covariant vectors transform
as normals (gradients) would transform. An example of a covariant vector
is the basis vectors. Let y = y 1 , y 2 , y 3 be the Cartesian coordinates of a
point and x1 , x2 , x3 its coordinates in some other curvilinear system. The
DR

∂y i
quantities for fixed i are the components of a covariant vector. For
∂xj
fixed j,
∂y
g (j) = , (notice the lower index) (1.6.6)
∂xj
is a Cartesian vector. The three Cartesian vectors g (1) , g (2) , g (3) are called
base vectors in the coordinate system xj . Any Cartesian vector can be
expressed as,
a = a1 g (1) + a2 g (2) + a3 g (3) = ai g (i) . (1.6.7)

For another system of coordinates x̃k , we would have base vectors g̃ (k)
defined. The reader is encouraged to show through transformation of ai
that they are components of a contravariant vector.

12
 Chapter 1: CHAPTER 1. MATHEMATICAL PRELIMINARIES

Optional

1.7 The metric tensor

Let the distance between two points A and B with coordinates y i and
y i + dy i respectively, be ds. Then,

ds2 = dy i dy i . (1.7.1)
X

Note that no summation is implied when both indices are upper (or both are
lower). Hence an explicit summation sign is used. Using another coordinate
system xj ,

∂y i j ∂y i k
! !
ds =
2
dx = gjk dxj dxk , (1.7.2)
X
dx
i ∂xj ∂xk

where
∂y k ∂y k
gij = (1.7.3)
X

k
T
∂xi ∂xj
is the metric tensor. A generalized curvilinear coordinate system has only
AF
diagonal terms nonzero in the metric tensor, and it is often covenient to
write the three independent diagonal components as,

g11 = h21 , g22 = h22 , g33 = h23 , (1.7.4)

DR

where hi are called the scale factors. For an infinitesimal displacement in

the direction of the xi coordinate, ds = hi dxi (no summation). Cartesian
coordinates all have the physical dimensions of length. In general, this
is not the case for curvilinear coordinates. Hence, contravariant velocity
components v i = dxi /dt would not all have the same physical dimensions.
For a general orthogonal curvilinear system,

|A|2 = (h1 A1 )2 + (h2 A2 )2 + (h3 A3 )2 , (1.7.5)

so that hi Ai have the same physical dimensions as the magnitude of A.

Similar to the basis vectors g (i) , we can define the reciprocal basis vectors
g (i) ,
g (j) × g (k)
g (i) = √ , (1.7.6)
g

13
Advanced Fluid Mechanics by Prateek Gupta

Optional
where g is the determinant of the metric tensor and g (i) ·g (j) = δji is satisfied.
Analogously, we can define the conjugate metric tensor as,

g ij = g (i) · g (j) (1.7.7)

The corresponding covariant components of a can be defined using a =

ak g (k) . The relation between the contravariant components and covariant
components of a vector can be obtained using,
 
ai g (i) · g (k) = aj g (j) · g (k) , (1.7.8)
=⇒ ai δik = aj g jk =⇒ ak = g jk aj . (1.7.9)

1.8 Derivatives and Christoffel symbols

Consider a contravariant vector a = ai g (i) . The derivative w.r.t. x(k)
(Covariant differentiation) is given by,
T
AF
∂a ∂ai ∂g (i)
= g (i) + a i
. (1.8.1)
∂xk ∂xk ∂xk

The derivative of the basis vectors w.r.t. the curvilinear coordinates, ex-
pressed in terms of the basis vectors are known as the Christoffel symbols
DR

(for basis vectors, these are Christoffel symbols of second kind) and can be
written as,
∂g (j)
Γijk g (i) = . (1.8.2)
∂xk
Now, we are in a position to define gradient, Laplacian, divergence, and
curl of tensorial quantities in generalized curvilinear coordinates.
Gradient of a scalar field is by definition a covariant vector (since the trans-
formation happens like a normal). For a scalar function ϕ, the gradients
∂ϕ/∂xi can be written in short-hand like ϕ,i which are covariant compo-
nents. The vector is given by,

∇ϕ = ϕ,i g (i) , (1.8.3)

in which summation is implied as per rules. The reader is urged to work

this out for cylindrical coordinates and confirm. The corresponding con-
travariant components of the gradient are g ij ϕ,i . To define Laplacian, we

14
 Chapter 1: CHAPTER 1. MATHEMATICAL PRELIMINARIES

Optional
differentiate this with xj to get,
!
∂ ∂ϕ ∂ϕ
g ij i + Γjjk g ik , (1.8.4)
∂xj ∂x ∂xi

which can be further simplified to familiar forms. The reader should refer
to Chapter 7 in Aris [2012] for full description.

T
AF
DR

15