MA3FM/MA4FM

A Mathematical Introduction to Fluid

Mechanics and Its Environmental Applications

Dr. A.Lukyanov

These Lecture Notes are taken from the course given by Dr. A.Lukyanov in the Autumn term
of 2020, at the Department of Mathematics, the University of Reading.
 Potential flows past a cylinder and a wing.

Introduction
Fluid Mechanics is an integral part of a more general applied mathematical discipline -
Continuum Mechanics, which deals with media motion modelled as continuum. In this
course, we will first briefly consider basic notions and principles of Continuum Mechanics,
such as tensors and operations with them, deformations, rate of deformations and stress
tensors, conservation laws, which are common for both Fluid Mechanics and Theory of
Elasticity. Then we will concentrate on the fluid mechanical description and consider in
detail the main formulation of Fluid Mechanics - the Navier–Stokes equations.

The system of Navier–Stokes equations has been the basis for theoretical consideration
of fluid motion during two centuries and arguably is the most intricate system of partial
differential equations so far. Not surprisingly, the Navier - Stokes existence and uniqueness
problem is one of the seven ”The Millennium Prize Problems”. Its first solution has
been claimed recently by Prof. Otelbayev, but it appeared that the proof had a flaw.
In the course, we will analyze both model ”academic” inviscid potential flows and more
realistic examples featured in applications such as river flows and origin of the aircraft lift.
We will consider important applications of Fluid Mechanics to environmental flows, such
as flows in porous media, concluding with the shallow water model of tsunami waves.

i
Contents

1 Tensors, Strain & Stress 2

1.1 An Introduction to Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Cartesian Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.2 Some Properties of Tensors . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Exercises - I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Elements of Continuum Mechanics 12

2.1 Eulerian and Lagrangian Coordinates . . . . . . . . . . . . . . . . . . . . . 12
2.2 Velocity and Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Rate of Change of a vector: the Material Derivative. . . . . . . . . . . . . . 18
2.4 Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.1 Analysis of Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 Deformation rate tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.6 Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.6.1 The Stress Tensor σij . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.6.2 Properties of σij . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.7 Exercises - II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 The Navier-Stokes Equations 38

3.1 The Leibniz-Reynolds Transport Theorem . . . . . . . . . . . . . . . . . . 38
3.2 Conservation of Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 Conservation of Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

ii
 3.4.1 Stress Tensor in Fluids . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4.2 The Navier-Stokes equations for incompressible fluids . . . . . . . . 45
3.4.3 The incompressible Euler equations . . . . . . . . . . . . . . . . . . 46
3.5 Boundary conditions to the Navier-Stokes equations . . . . . . . . . . . . . 48
3.5.1 Solid boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.5.2 Free surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.6 Boundary conditions to the Euler equations . . . . . . . . . . . . . . . . . 53
3.6.1 Solid boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.6.2 Free surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.7 Exercises - III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

4 Static Problems of Fluid Mechanics 59

4.1 Exercises - IV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.1.1 Reference material . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5 Inviscid Flow 70
5.1 Bernoulli Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2 Kelvin’s circulation theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3 Potential Flow: special class of inviscid flows . . . . . . . . . . . . . . . . . 79
5.4 Stagnation-point flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.5 Exercises - V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

6 Two-dimensional potential flows 86

6.1 Stream function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.2 Some basic two-dimensional potential flows . . . . . . . . . . . . . . . . . . 88
6.3 Combinations or superpositions of elementary flows: Rankine flow. . . . . . 93
6.4 Combinations or superpositions of elementary flows: potential (steady) flow
past a cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.5 The force in potential flows: d’Alembert paradox . . . . . . . . . . . . . . 96
6.6 Irrotational flow with circulation - origin of lift, the Magnus effect or playing
golf and tennis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

iii
 6.6.1 Qualitative analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.6.2 Quantitative analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.6.3 Reference material. Differential operators in polar and cylindrical
coordinate system. . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.6.4 Reference material. Some notes on the Laplace equation . . . . . . 105
6.6.5 Some particular periodic solutions of the Laplace equation in a polar
coordinate system by separation of variables. . . . . . . . . . . . . . 105
6.7 Exercises - VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

7 Viscous Fluid 110

7.1 Scaling analysis and the viscous boundary layer . . . . . . . . . . . . . . . 110
7.2 Non-dimensional Navier-Stokes equations, Reynolds number and Reynolds
number limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.2.1 How do we know that we have taken the right limit? . . . . . . . . 114
7.2.2 Viscous boundary layer . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3 Non-stationary problems of hydrodynamics: transient viscous boundary
layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.4 Some notes on the Stokes flow regime: the limit of small Reynolds numbers.121
7.5 The rectilinear flow ansatz. . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.5.1 No slip condition applied. . . . . . . . . . . . . . . . . . . . . . . . 123
7.6 Fully developed viscous flows. . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.6.1 Simple shear flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.6.2 Couette and Poisseuille flows . . . . . . . . . . . . . . . . . . . . . . 125
7.7 Exercises - VII . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

8 Flows in porous materials: Darcy’s law, filtration and making espresso.131

8.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.2 Darcy’s law. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
8.2.1 Incompressibility condition in porous media and the governing equa-
tions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
8.2.2 Geometrical model of porous media. . . . . . . . . . . . . . . . . . 134
8.2.3 One-dimensional flows in porous media. . . . . . . . . . . . . . . . . 135
8.3 Exercises - VIII . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

iv
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 1

9 The shallow water approximation - water waves and non-stationary so-

lutions 144

10 Revisions 150
10.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
10.2 Some notes from vector calculus: scalar and vector fields . . . . . . . . . . 152
10.3 Jacobian Matrix and the Jacobian . . . . . . . . . . . . . . . . . . . . . . . 153
10.4 The Inverse Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 154
10.5 The Gauss-Ostrogradsky theorem (also known as the Divergence theorem
or Gauss’s theorem). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
10.6 The Stokes’ theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

11 Self-study. 156
11.1 Summation convention: Einstein notation. . . . . . . . . . . . . . . . . . . 156
11.2 Examples of some practical use of indices, useful (later) symbols and exercises.158

(MA3FM/MA4FM) Autumn 2020

Chapter 1

Tensors, Strain & Stress

1.1 An Introduction to Tensors

The idea of introducing tensors stems from invariant forms of equations in physics and
mathematics. By the invariant forms here we mean that they are independent of the
choice of coordinate systems. Equations of mechanics and physics are often expressed in
vector notation, for example Newton’s Second Law is commonly expressed as

d2 r
M = F (t),
dt2
where M is the mass of an object and F is the force acting on that object which is
assumed to be located at point r in the three-dimensional space.
The underlying reason for writing equations representing the laws of nature in vector form
is the assumption that they (the laws) do not depend on the choice of coordinate system,
according to the covariance principle.

♦ Definition 1.1.1. Covariance Principle The fundamental covariance principle of

physics states that all physical laws can be represented in a form equally valid for all
observers.

What does the principle imply in the case of Newton’s Second Law in particular? Newton’s
Second Law is written without specifying any particular coordinate system, that is the
coordinate system is arbitrary within the class of permissible systems, which are obtained
from each other via rotation, translation with velocities less than the speed of light and
reflection. Basically, Newton’s Second Law has the same form in any admissible coordinate
system, since the vector form consisting of two vectors r and F, and two scalars M and t is
invariant of the choice of a coordinate system. Indeed, mathematically, a vector equation
A + B = C, Fig. 1.1, is insensitive to a change of the coordinate system. Apparently,
the components of each vector in the equation depend on the choice of the coordinate

2
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 3

system, but the form of the vector equation is insensitive to that choice. No matter what
coordinate system we choose, we have the same vectors A, B and C. So in fact, the
Covariance Principle in physics is essentially a mathematical statement.
In what follows, it is convenient to distinguish between active and passive transformations
to avoid confusion.
♦ Definition 1.1.2. The transformation is passive, if one merely changes the coordinate
system, while the object itself, a body for example, does not change its position. On the
other hand, the transformation is active if the body is changing its position. Apparently,
in the example about the vector equation A + B = C, we meant a passive transformation.
For simplicity of the representation, we focus on the particular type of transformation -
rotation of a coordinate system.

Formally, if we consider just one position vector X, for example, and the rotation of the
set of orthogonal axes i1 , i2 , i3 through to i1 , i02 , i03 , Fig. 1.2, that is around i1 , then the
same vector can be written in the form x = xk ik and x0 = x0m i0m . Now by the Covariance
Principle these represent the same point and thus

xk ik = x0m i0m . (1.1)

Now, before we go into any details, let’s appreciate the notation in the above equation
(1.1). Here we have used the so called summation convention. You shall read and
study this notation in detail using Self-study section 11.2. Here, let’s just consider a
few examples.
 Example 1.1.3. The form we have used already in writing (1.1) may be stated explicitly
as
3
X
xk ik = xk ik = x1 i1 + x2 i2 + x3 i3 .
k=1

Index k is called a repetitive index and summation is assumed over that index as is shown.
Note, we consider three-dimensional vector spaces only, so that indices run from 1 to 3.
If the vector space would be different, five-dimensional, for example, then

xk ik = x1 i1 + x2 i2 + x3 i3 + x4 i4 + x5 i5 .

In general, the space dimension can be anything, but throughout our module, we will only
need three-dimensional vector spaces.
Note, the only coordinate systems we are going to use will be orthogonal coordinate
systems with canonical basis, see revisions 10.1. This implies that the coordinate system
base vectors (i1 , i2 , i3 ) are orthogonal to each other (even better, they are orthonormal
to each other, since the basis vectors are unit vectors, that is for ∀ k = 1..3, |ik |2 =
ik · ik = 1), that is in general their scalar product can be expressed as

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 4

im · in = δmn
where we have introduced the Kronecker-Delta which is defined to be
(
1 if i = j,
δij =
0 if i 6= j.

Property 1.1.4.
δij fj = fi

To show that the statement is correct, consider just the case j = i, since otherwise contri-
bution is zero according to the definition of the Kronecker-Delta. Then it is straightforward
fi = fi .
Note, that in the equation
δij fj = fi
index i was a free index, which can be 1 ≤ i ≤ 3, and index j was a dummy index.
 Example 1.1.5. Another example, if we have vector functions of time t, A(t) = Ak ik
and B(t) = Bk ik , then their dot product or scalar product, see revisions 10.1,
3
X
A · B = A1 B1 + A2 B2 + A3 B3 = Ak Bk = Ak Bk
k=1

Now, consider the full derivative of the scalar product with time
dA · B d d d dA dB
= Ak Bk = Bk Ak + Ak Bk = B +A
dt dt dt dt dt dt

So, we have simply proven that

dA · B dA dB
=B +A
dt dt dt

Now, let’s return to our analysis. Taking scalar products (see revisions 10.1) in (1.1)
with i0p yields
xk ik .i0p = x0m i0m .i0p = x0m δmp = x0p ,
which by defining lpk = i0p .ik - the direction cosine between i0p and ik - yields

x0p = lpk xk . (1.2)

Note, here we used the orthogonality of the unit vectors i0m .i0p = δmp , definition and the
property of the Kronecker Delta.
It is essential that the coordinate system, we used here, was an orthogonal Cartesian
coordinate system, so that the scalar product of the basis vectors was easy to evaluate

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 5

Original system Rotated system

r
i3
r r r
i2 i3' i2'
r
i1 r
i1'

Figure 1.1: Illustration of a vector equation and two Cartesian coordinate systems, original
0 0 0
with the unit vectors (i1 , i2 , i3 ) and a rotated one with primed unit vectors (i1 , i2 , i3 ) in
a three-dimensional case.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 6

r r
i3' i3
r
i2
r
i1
X

Figure 1.2: Illustration of a vector representation in two Cartesian coordinate systems,

original with the unit vectors (i1 , i2 , i3 ) and a rotated one with primed unit vectors
0 0
(i1 , i2 , i3 ) in a three-dimensional case.

explicitly. Further, to emphasize this again, we will always assume that kind of coordinate
system. This assumption simplifies calculations, though generalization to non-orthogonal
coordinate systems is possible.
If however we take scalar products of equation (1.1), with ip then we obtain

xk ik .ip = x0m i0m .ip ,

xk δkp = x0m i0m .ip ,
xp = x0m lmp . (1.3)

From equation (1.2), we have, using matrix notation,

 0   
x1 l11 l12 l13 x1
x0 = x02  = l21 l22 l23  x2  = Lx.
x03 l31 l32 l33 x3

Similarly from equation 1.3, we have

    0
x1 l11 l21 l31 x1
x = x2 = l12 l22 l32
     x02  = LT x0 ,
x3 l13 l23 l33 x03

where, as one can be readily seen, LT is the transpose of matrix L. To remind, if the
elements of the matrix L are L = (lkp ), then the elements of the transposed matrix LT
are LT = (lpk ).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 7

So we deduce that x0 = L(LT x0 ) and similar x = LT (Lx), so that LLT = I and LT L = I

respectively, where I is the identity matrix, that is
 
1 0 0
I = 0 1 0 .
0 0 1

Therefore L indeed represents an orthogonal transformation - that is three dimen-

sional rotation of a coordinate system, which transforms an orthonormal basis to an
orthonormal basis. Note, the identity matrix components can be also represented as
I = (δij ).
Another way to represent that LLT = I and LT L = I is that

lpk lmk = δpm (1.4)

and that
lkp lkm = δpm . (1.5)

Note:
The elements of the matrix L are the cosines of the angles between the unit vectors, that
is in fact the angles of rotation. This can be seen explicitly in a two-dimensional case
shown in Fig. 1.3 where  
cos θ sin θ
L(θ) = .
− sin θ cos θ
To show that, let’s rewrite the transformation matrix in terms of the orthogonal base
vectors of the coordinate system
 0
i1 .i1 i01 .i2

L(θ) = 0 .
i2 .i1 i02 .i2

Then, as it is seen from the picture, Fig. 1.3, i01 .i1 = cos θ, i02 .i2 = cos θ, i01 .i2 =
cos(π/2 − θ) = sin θ, i02 .i1 = cos(π/2 + θ) = − sin θ and the result follows.

Now, the main result we have obtained is that under rotation of a coordinate system,
components of any vector X are transformed according to

x0p = lpk xk . (1.6)

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 8

' i2 '
i2 i1
θ i1

Figure 1.3: Illustration of rotation of coordinates in the anticlockwise direction by angle

θ in a two-dimensional case.

So, any vector satisfies the covariance principle and thus is transformed under rotation of
a coordinate system according to (1.6).
However physical quantities cannot all be classified as vectors or scalars, and the notion
of Tensors is required. For example, in fluid conductivity problems in anisotropic media
(the media where the properties are not the same in different directions), fluid current
density j, a vector, is related with another vector, the gradient of pressure ∇p through
j = S∇p, where the conductivity is a matrix S = (sij ), that is jk = skm ∂x∂pm .
We will use the transformation rule (1.6) to generalize the notion of vectors to arbitrary
tensors. The tensors enable us to extend the notion that equations are independent of
the choice of coordinate system. Cartesian Tensors are associated with Cartesian frames
of reference in Euclidean space. We shall not deal with general tensor theory and the
associated non-Cartesian frames of reference. Tensors will be dependent variables in our
discussion of the theory of fluid mechanics.

1.1.1 Cartesian Tensors

♦ Definition 1.1.6 (First Order (Rank) Cartesian Tensors). Any quantity xk which
transforms under rotation of a coordinate system as

x0p = lpk xk

is called the first order Cartesian tensor.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 9

♦ Definition 1.1.7 (Second Order (Rank) Cartesian Tensors). Any quantity aij which
transforms under rotation of a coordinate system as

a0mn = lmi lnj aij

is called the second order Cartesian tensor.

In general, one can define higher order tensor quantity by means of, for example,

♦ Definition 1.1.8 (Fourth Order (Rank) Cartesian Tensors).

a0mnkp = lmi lnj lks lpq aijsq .

0
 Example 1.1.9. Show that δij is a tensor, that is δmn = lmi lnj δij .
Using properties of the Kronecker Delta and the result (1.4), that is lpk lmk = δpm

lmi lnj δij = lmi lni = δmn

0
and hence δmn = δmn , the Kronecker delta is actually transformed to itself.

1.1.2 Some Properties of Tensors

1. Multiplying a tensor by a scalar, obtains a tensor of the same order.

2. Two tensors of the same order are added (subtracted) by adding (subtracting) the
corresponding components
 Example 1.1.10. If it is known that aij and bij are second order tensors, then
cij obtained through
C1 aij + C2 bij = cij

is also a second order tensor. Here C1,2 are constant scalars (zero order tensors).

3. Two tensors can be multiplied together, aijk blm = cijklm , this is called the outer
product.
 Example 1.1.11. If it is known that Ak and Bl are vectors (that is they are in
fact first order tensors), then their outer product will form a second order tensor

Ak Bl = Tkl .

4. If two indices are the same in the outer product, one can obtain the inner product.
In the inner product, the result is a tensor of two orders lower, that is aisk bjk =
disj . A specific example is the scalar product, which given two vectors returns a
scalar.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 10

 Example 1.1.12. If it is known that Ak and Bl are vectors, then their inner
product will form a zero order tensor, that is simply a scalar C,

Ak Bk = C.

5. If aij = aji , then aij is a symmetric tensor with 6 independent components. The
question to figure out is why do we have only 6 components independent out of total
9 components available in a second order tensor?

6. If aij = −aji , then aij is an anti-symmetric tensor with 3 independent compo-

nents. Why?

In Fluid Mechanics, we will develop a mathematical formulation solely based on tensors,

so that the obtained description will be invariant to the choice of a coordinate system.

Supplementary material 1.1.13. Sometimes it is easier to use different

N
method for testing whether a set of quantities Aij forms a second order Cartesian
tensor, rather then the definition. In these cases the Quotient Theorem provides a
test.

Theorem 1.1.14 (The Quotient Theorem). If Bij is an arbitrary second order tensor,
and one can prove/show that
P r = Aij Bij
is such that P r is a scalar, then Aij form a second order tensor.

There is a more general version of the theorem, which can be found elsewhere.

1.2 Exercises - I
1. If Aij and Bij are second order (rank) tensors, prove that Cij formed via

Cij = αAij + βBij

is a tensor, where α, β are some constant scalars.

2. If Aij and Bij are second order (rank) tensors, prove that the inner product Ckl

Akm Bml = Ckl

is a second order tensor.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 11

3. A physical law of conductivity is formulated as

jk = Skl fl .

It is known that jk and fk are both first order tensors, that is simply vectors. Demon-
strate the transformation of the above relationship during a passive transformation
of rotation given by the elements of lij .
Consider two cases: one is when Sij is known to be a second order tensor and
another one, when Sij is not a tensor.

4. Prove or demonstrate that any second order tensor Tij can be split into two parts:
a symmetric tensor Sij and an anti-symmetric tensor Aij . Hint, indeed, one can
form two quantities out of the tensor Tij components

1 1
Aij = (Tij − Tji ) ; Sij = (Tij + Tji ) .
2 2
Show that both Aij and Sij , defined this way, are second order tensors.

5. Suppose that Sij is a second order symmetric tensor and Aij is a second order
anti-symmetric tensor. Prove or show that

Sij Aij = 0.
N
6. By using the Quotient Theorem, show that the Kronecker delta δKL is a Cartesian
tensor. This question needs supplementary material, theorem (1.1.14).

(MA3FM/MA4FM) Autumn 2020

Chapter 2

Elements of Continuum Mechanics

2.1 Eulerian and Lagrangian Coordinates

Proposition 2.1.1. The Continuum Hypothesis. We assume that the highly dis-
continuous structures of real materials may be replaced by a smoothed out hypothetical
continuum, every portion of this material (no matter how small) having the same macro-
scopic physical properties as the bulk

 Example 2.1.2. Experiments indicate that the pressure P , density ρ and temperature
T of many gasses and liquids are related through a constitutive equation of the form P =
f (ρ, T ). This equation is called the equation of state. This equation remains valid over
a wide range of sample states. Continuum mechanics assumes the relationship remains
valid even for infinitesimal volumes.

♦ Definition 2.1.3. A Material Point is a mathematical entity, within a continuum and

is not to be confused with atoms or molecules.

Consider a set of material points, which form the continuum and occupy initially, at t = 0,
a volume element V (0) and after some time a volume element V (t). The fluid motion
is determined with respect to a system of coordinates of the observer, which will be
denoted x1 , x2 , x3 and is called the Eulerian coordinate system. Velocity in the Eulerian
coordinate system is simply given by a vector field

v(x1 , x2 , x3 , t),

where the independent variables are coordinates x1 , x2 , x3 and time t.

Another way to describe the motion of the continuum is to individualize the material
points. For example, one can mark the position of material points at some moment of
time, say at t = 0 and then follow their individual trajectories. These coordinates are
called Lagrangian and we will denote them with capital letters X1 , X2 , X3 . We shall

12
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 13

adopt further the convention of capital letters for the Lagrangian coordinates and lower
cases letters for Eulerian coordinates.
   
X1 x1
Lagrangian X = X2 ;   Eulerian x = x2  .

X3 x3
Then, the Lagrangian description of motion, where independent variables are (X1 , X2 , X3 )
and time t is
x1 = x1 (X1 , X2 , X3 , t),
x2 = x2 (X1 , X2 , X3 , t),
x3 = x3 (X1 , X2 , X3 , t).
Velocity in the Lagrangian coordinate system is given by partial derivative with respect
to time
∂
V (X1 , X2 , X3 , t) = x(X1 , X2 , X3 , t).
∂t
The mapping x = x(X, t) is assumed to be bijective and the inverse is given by

X = X(x, t). (2.1)

The inverse function of this transformation (see 10.3) is found uniquely in the neighbour-
hood of some point X, according to the Inverse Function Theorem, if the Jacobian of
transformation is non-vanishing, explicitly
∂x1 ∂x1 ∂x1
∂X1 ∂X2 ∂X3
∂(x1 , x2 , x3 ) ∂x2 ∂x2 ∂x2
J= = 6= 0.
∂(X1 , X2 , X3 ) ∂X1 ∂X2 ∂X3
∂x3 ∂x3 ∂x3
∂X1 ∂X2 ∂X3
provided that the partial derivatives ∂xi /∂Xj are continuous. This assumption reflects
the indestructible nature of continuum, that is there is no region of finite volume which
can be deformed into a zero or infinite volume, see Fig. 2.2. Another consequence of the
assumption is that trajectories x = x(X, t) do not cross each other. You can find more
details about the Inverse Function Theorem and the Jacobian in the Revision section.
 Example 2.1.4. It is known from Vector Calculus, that the body volume can be calcu-
lated from a triple or volume integral. That is, initial volume V (0) is, for example,
Z
V (0) = dX1 dX2 dX3 .
V (0)

The transformed volume V (t0 ) then, if we consider ’the same points of the continuum’,
Z Z
V (t0 ) = dx1 dx2 dx3 = J(t0 )dX1 dX2 dX3 ,
V (t0 ) V (0)

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 14

V (t 0 )
Transformed volume element
at time t=t0

(x 1, x 2, x 3 )
x3
Flow domain

x2
x1

( X 1, X 2, X 3 )
V (0)
Initial volume element
at time t=0

Figure 2.1: Eulerian and Lagrangian description of motion. Initial, arbitrary volume
element in the flow domain at t = 0, V (0), is transformed into V (t0 ) at time t0 . Note,
though, that the transformed volume element consists of the same liquid particles as
there were in the initial volume, that is all points in V (t0 ) are mapped from V (0) via
x = x(X, t0 ). There are two possible ways to observe the transformation (two differ-
ent observers): one is to sit at the origin of the coordinate system and measure, for
example, velocity v = v(x1 , x2 , x3 , t) at (x1 , x2 , x3 ), and another one is to follow the
transformation along the trajectories x = x(X, t), for example from some initial point
inside the volume element V (0), (X1 , X2 , X3 ), to its transformed position at t = t0 ,
(x1 (X, t0 ), x2 (X, t0 ), x3 (X, t0 )).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 15

V (0)

J (t 0 )=0

Figure 2.2: Initial finite volume element V (0) is transformed into a point of zero volume
at t = t0 . This collapse is accompanied by vanishing of the Jacobian of transformation at
that moment of time J(t0 ) = 0. This situation is assumed to be impossible in Continuum
Mechanics. (MA3FM/MA4FM) Autumn 2020
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 16

where J is the Jacobian of the transformation. It is clear now that if at some moment of
time t0 , the Jacobian vanishes J = 0, then the initial volume has been transformed into
’nothing ’. On the other hand, if J ≡ 1 always, then the volume is preserved, that is the
motion is volume preserving.

2.2 Velocity and Acceleration

The two ways of motion description in Continuum Mechanics can be further illustrated
by two examples, see Fig. 2.3 and Fig. 2.4. In the first example shown in Fig. 2.3 an
elastic deformation is depicted. The preferable way to describe elastic deformations is
the Lagrangian description. Eulerian description may be awkward, since the elastic body
after the deformation, as is shown in the figure, may move away from the observation
point A. So that there will be nothing to observe at that point. Below is the summary of
the Lagrangian description.
A. Elasticity (Lagrangian Coordinates)

• Displacement U (X, t) = x(X, t) − X = Uk (X, t)ik .

dU ∂Uk
• Velocity V = = ik = Vk ik
dt ∂t

dV ∂ ∂ 2 Uk
• Acceleration A = = Vk ik = ik
dt ∂t ∂t2

In the second example, we qualitatively consider (without dwelling into details) a steady
incompressible liquid flow through a funnel, Fig. 2.4. The total volume flux through the
funnel Φ (dimension [Φ] = L3 T −1 ) is constant because of the steady conditions of the flow
and conservation of mass. Note, the conservation of mass for incompressible flow implies
conservation of volume. The flux density φ can be defined by means of φ = Φ/S, where
S is the variable total cross-section area of the funnel. The cross-section area is changing
along the x2 -coordinate, that is along the funnel, S = S(x2 ), in particular S0 and S1
are the cross-section areas at the entrance and at the exit of the flow domain, such that
S0 > S1 . As a result of that, the volume flux density φ is also changing and is a function
of x2 , φ(x2 ), since the total flux is conserved Φ = const. Since S0 > S1 , we have from the
conservation of the total flux Φ = φ0 S0 = φ1 S1 and hence φ0 < φ1 .
The dimension of φ is [φ] = LT −1 , that is it has a dimension of velocity, and φ can be
qualitatively considered as a measure of velocity. Qualitatively, then, we have a situation
that the liquid velocity is changing along the funnel and, in particular, is a function of
x2 . In particular, the velocity is growing at the bottleneck area of the funnel.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 17

Now, if we would use the Lagrangian description of motion, we should had fixed some
points at some moment of time t = 0 and then followed their trajectories or evolution
with time t > 0. But, in this case, those points may quickly leave the observation volume.
On the other hand, in the Eulerian description we would not have such a problem. So for
fluid mechanical problems, Eulerian description is preferable. The velocity in
this description, as we have already mentioned, is simply a vector field given at points of
observation x = (xj ). So that the velocity v = v(x1 , x2 , x3 , t) is a function of independent
variables x1 , x2 , x3 and time t. In the steady flow conditions in our example, there is
no explicit dependence on time t, v = v(x1 , x2 , x3 ).
Qualitatively, this can be illustrated by the dependence of velocity component v2 as a
function of x2 and t shown in Fig. 2.5. Since φ is a measure of velocity in this case, the
velocity should grow as we move from left to right, that is through the bottleneck region of
the flow domain. At the same time, the velocity has no explicit dependence on time, as is
also shown in the figure, due to the steady flow conditions. The main question now is how
to define acceleration? If we would formally differentiate velocity v = v(x1 , x2 , x3 ) with
∂v(x1 , x2 , x3 )
respect to time, then acceleration is zero, that is a = = 0. This is shown
∂t
in Fig. 2.5. On the other hand, in the second picture, one can see that the velocity is
changing with respect to x2 , so that liquid particles do accelerate! To obtain acceleration
using the Eulerian description, one needs to introduce a special rule of differentiation with
time, the material derivative D/Dt,

d D
a= v= v.
dt Dt
The operator is defined assuming that we are taking the derivative with respect to the
moving particles, that is explicitly in Eulerian coordinates

v(x, t) = vk (x(X, t), t)ik

or
vk (x(X, t), t) = vk (x1 (X, t), x2 (X, t), x3 (X, t), t),
then by the chain rule of partial differentiation
 
D ∂vk ∂x1 ∂vk ∂x2 ∂vk ∂x3 ∂vk
v= + + + ik ,
Dt ∂t ∂t ∂x1 ∂t ∂x2 ∂t ∂x3

that is  
d D ∂vk ∂xl ∂vk
a= v= v= + ik ,
dt Dt ∂t ∂t ∂xl
∂xl
and since = vl
∂t  
D ∂vk ∂vk
v= + vl ik .
Dt ∂t ∂xl

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 18

∂vk
That is when in steady conditions the = 0, the acceleration is defined by the change
∂t
of velocity along the path of a moving liquid particle. Below is a summary of the Eulerian
description.
B. Fluid Mechanics (Eulerian coordinates)

• Velocity v = v(x, t) = vk (x, t)ik

d D
• Acceleration a = v= v
dt Dt

2.3 Rate of Change of a vector: the Material Deriva-

tive.
In general for an arbitrary vector field, we have:
In Lagrangian coordinates F (X, t) = Fk (X, t)ik , then since ∂Xk /∂t = 0;

dF ∂Fk
= ik .
dt ∂t
In Eulerian coordinates f (x, t) = fk (x, t)ik , then by the chain rule
 
d ∂fk ∂xl ∂fk
f= + ik .
dt ∂t ∂t ∂xl

∂xk
The velocity of a particle is v = vk ik = ik
∂t
   
d ∂fk ∂fk Dfk Df
f= + vl ik = ik = .
dt ∂t ∂xl Dt Dt

D ∂ ∂
The operator = + vl is known as the Material Derivative.
Dt ∂t ∂xl

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 19

Elastic deformation

Initial Deformed
x1= X 1,
x 2= X 2, A B x 1 ( X 1, X 2, X 3, t 0) ,
x3 = X 3 x 2 ( X 1, X 2, X 3, t 0 ),
t=0 x 3 ( X 1, X 2, X 3, t 0 ),
t =t 0

x3

x2
x1

Figure 2.3: An example of Lagrangian description of motion in elasticity. A body of finite

volume has been deformed as in the picture. In Lagrangian description one can follow the
trajectories of particles of the elastic body as is shown in the figure. On the other hand, if
we would try using Eulerian description, then this might be quite awkward. Consider, for
example, point A in the figure. If we would observe motion at this point, that is x1 = X1 ,
x2 = X2 and x3 = X3 , then at moment of time t = t0 point A is moved in point B.
In addition, the body is deformed such that A is out of the domain completely. That is
our observation point basically ceased to exist. For this reason, in elasticity, Lagrangian
description is preferably used.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 20

x1= X 1,
x 1 ( X 1, X 2, X 3, t 0) ,
x 2= X 2,
x 2 ( X 1, X 2, X 3, t 0 ),
x3 = X 3
x 3 ( X 1, X 2, X 3, t 0 ),
t=0
Steady flow t =t 0

 S0 S1
o o o
x =( x1 , x 2 , x 3 )

x3

x2
x1

Figure 2.4: An example of Lagrangian and Eulerian description of motion in Fluid Me-
chanics. The liquid flux Φ enters from the left into a funnel and exits from the right. One
can describe the flow using both descriptions. In Eulerian description, we can measure
and hence define the velocity vector field v(x1 , x2 , x3 , t) at any point inside the domain
of flow, for example at point (xo1 , xo2 , xo3 ). In Lagrangian description, we follow particle
trajectories from an initial position at t = 0. Apparently, after some time, the moving
particles will leave the domain of observation. For this reason, in Fluid Mechanics, Eu-
lerian description is widely used, while Lagrangian description is only used to track free
surface boundaries. Here, S0 and S1 are the cross-section areas of the funnel.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 21

,x2,x3,t)
o
o
o
1
Velocity v2(x

0
0 5 10
Time, t
,x2,x3,t)

1
1
Velocity v2(x

0
0 5 10
Distance, x2

Figure 2.5: Qualitative illustration of v2 component of velocity evolution in steady condi-

tions relevant to the example in Fig. 2.4. Note, in the first plot, the velocity component
v2 is taken at a fixed position (xo1 , xo2 , xo3 ), an observation point of the Eulerian description.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 22

2.4 Strain
In rigid body dynamics, there is no relative motion between material points, in general
however this is not true, there is compression and distortion of material. These effects
may vary throughout the continuum. A description of deformation is necessary at each
point of the continuum, the question is ‘how to describe such deformations?’.

2.4.1 Analysis of Strain

The nature of the deformation of the material in the neighbourhood of a particle labelled
X is made clear by examining the behaviour of all line vectors connecting X with X +dX.
At time t, Fig. 2.6, this line has moved between the point x(X, t) and x(X + dX, t),
such that

dx = x(X + dX, t) − x(X, t)

It is the nature of this transformation which determines the local deformation and not
the translation of the vector line segment. We denote the square of the line vector dX
as dS 2 = dXK dXK and the square of the length of the line vector dx as ds2 = dxk dxk .
We consider the difference in length
ds2 − dS 2 = dxk dxk − dXK dXK .
This quantity is determined solely by the deformation and neither by the rotation nor
other rigid body motion.
 Example 2.4.1. Consider a rigid body motion with velocity v in the X1 direction, Fig.
2.7 (a), then
x1 = X1 + vt, x2 = X2 , x 3 = X3 ,
dx1 = dX1 , dx2 = dX2 , dx3 = dX3 .
so ds2 − dS 2 = dxk dxk − dXK dXK = 0.
 Example 2.4.2. Consider a 90◦ clockwise rotation of a body about the x3 axis, Fig.
2.7 (b), then     
x1 0 1 0 X1
x2  = −1 0 0 X2  .
x3 0 0 1 X3
so
dx1 = dX2 ,
dx2 = − dX1 ,
dx3 = dX3 .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 23

dX

X
X
d
X+

,t )
x(X
dx(t)
O

x(X+dX,t)

Figure 2.6: Definition sketch of a deformation at time t.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 24

so ds2 − dS 2 = dxk dxk − dXK dXK = 0.

Note, clockwise rotation of a body about the x3 axis by an angle θ is given by
    
x1 cos θ sin θ 0 X1
x2  = − sin θ cos θ 0 X2 
x3 0 0 1 X3
and the result in the particular case of 90◦ follows.

 Example 2.4.3. Consider a body stretched in the X1 direction, Fig. 2.7 (c), (λ 6= 1)

x1 = λX1 , x2 = X2 , x3 = X3 ;
dx1 = λ dX1 , dx2 = dX2 , dx3 = dX3

ds2 − dS 2 = dxk dxk − dXK dXK ,

= dx1 dx1 − dX1 dX1 , All other terms cancel
= λ2 dX12 − dX12 ,
= (λ2 − 1) dX12 6= 0.

The quantity ds2 − dS 2 will allow us to define the strain tensor in terms of deformation.
Recall that if xk = xk (X, t), then
dxk = xk (X + dX, t) − xk (X, t),
∂xk
= dXl . (2.2)
∂Xl
Also that if Xl = Xl (x, t),then
dXl = Xl (x + dx, t) − Xl (x, t),
∂Xl
= dxk . (2.3)
∂xk
∂xk ∂xk ∂Xl ∂xk ∂Xl ∂Xm ∂xk
Note now dxk = dXl = dxm ⇒ = δkm . Similarly =
∂Xl ∂Xl ∂xm ∂Xl ∂xm ∂xk ∂Xn
δmn .
We now use these in our expression for ds2 − dS 2 .
ds2 − dS 2 = dxk dxk − dXK dXK ,
∂xk ∂xk
= dXm dXn − dXm dXm ,
∂Xm ∂Xn
∂xk ∂xk
= dXm dXn − δmn dXm dXn ,
∂Xm ∂Xn
 
∂xk ∂xk
= − δmn dXm dXn .
∂Xm ∂Xn

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 25

Translation

(a)

Rotation

(b)

Stretching
(c)

x2
x1
x3

Figure 2.7: Illustration of different transformations of a body, relevant to the examples.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 26

We have then the following definition.

♦ Definition 2.4.4 (Lagrangian Strain Tensor).

∂xk ∂xk
2Emn = − δmn
∂Xm ∂Xn

Property 2.4.5. Demonstrate that Emn is a second rank tensor.

To show the statement is correct, let’s begin with the definition of the strain tensor, that
is
ds2 − dS 2 = 2Emn dXm dXn .

The quantity in the LHS is known to be a scalar, that is

0 0
Emn dXm dXn0 = Emn dXm dXn .

Now, by the properties of tensors

Amn = dXm dXn

is a second order tensor, since dXm is a vector. Then

0
Emn lmq lnp dXq dXp − Emn dXm dXn = 0.

The dummy suffix can be renamed within one particular term, if there is no clash with
another symbol
0
Emn lmq lnp dXq dXp − Eqp dXq dXp = 0.
Since dXq dXp is arbitrary, it follows that
0
lmq lnp Emn = Eqp .

Multiplying the last relation by liq ljp , one gets

0 0
liq ljp lmq lnp Emn = δim δjn Emn = Eij0 = liq ljp Eqp .

That completes the proof.

Alternatively, one can use the supplementary material, the Quotient theorem (1.1.14).
We know that ds2 − dS 2 is a scalar and dXm dXn is a second order tensor, arbitrary.
Then, directly from the theorem Ekm is a second rank tensor, since

ds2 − dS 2 = 2Emn dXm dXn .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 27

The Lagrangian strain tensor is a symmetric tensor, which seems to be obvious, but can
be shown formally as well,
∂xk ∂xk
2Emn = − δmn ,
∂Xm ∂Xn
∂xk ∂xk
= − δnm = 2Enm .
∂Xn ∂Xm

Supplementary material 2.4.6. Alternatively, we could write

N

ds2 − dS 2 = dxk dxk − dXK dXK ,

∂Xm ∂Xm
= dxk dxk − dxk dxl ,
∂xk ∂xl
∂Xm ∂Xm
= δkl dxk dxl − dxk dxl ,
∂xk ∂xl
 
∂Xm ∂Xm
= dxk dxl δkl − .
∂xk ∂xl

We have then the following definition.

♦ Definition 2.4.7 (Eulerian Strain Tensor).

∂Xm ∂Xm
2ekl = δkl −
∂xk ∂xl

Similar to the Lagrangian Strain tensor, the Eulerian strain tensor is symmetric.

The Lagrangian Strain Tensor plays a central role in elasticity and visco-elasticity. In
Fluid Mechanics, the Deformation Rate Tensor, which will be derived later, plays an
equivalent role.

 Example 2.4.8. Consider a body stretched in the X1 direction, as in example 2.4.3

x1 = λX1 ; x2 = X 2 ; x3 = X 3 .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 28

x2 Deformed configuration
2HX 2

T12
X2

x1

Figure 2.8: Definition sketch of Simple Shear

Then the components of the Lagrangian Strain Tensor are

 2  2  2
∂xk ∂xk ∂x1 ∂x2 ∂x3
2E11 = . − δ11 , = + + − δ11 ,
∂X1 ∂X1 ∂X1 ∂X1 ∂X1
= λ2 + 0 + 0 − 1,
= λ2 − 1.

 2  2  2
∂xk ∂xk ∂x1 ∂x2 ∂x3
2E22 = . − δ22 , = + + − δ22 .
∂X2 ∂X2 ∂X2 ∂X2 ∂X2
= 0 + 1 + 0 − 1 = 0.

 2  2  2
∂xk ∂xk ∂x1 ∂x2 ∂x3
2E33 = . − δ33 , = + + − δ33 ,
∂X3 ∂X3 ∂X3 ∂X3 ∂X3
= 0 + 0 + 0 + 1 − 1 = 0.

∂xk ∂xk ∂x1 ∂x1 ∂x2 ∂x2 ∂x3 ∂x3

2E12 = . − δ12 , = + + ,
∂X1 ∂X2 ∂X1 ∂X2 ∂X1 ∂X2 ∂X1 ∂X2
= 0.

Similarly E13 = E12 = E23 = 0.

Supplementary material 2.4.9. Geometrical Interpretation of the Strain

N
Components

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 29

 Example 2.4.10 (Pure Dilatation). Consider the deformation x1 = λX1 , x2 =

λX2 , x3 = λX3 , and so the Lagrangian Strain Tensor is
∂xk ∂xk
2E11 = − 1 = λ2 − 1,
∂X1 ∂X1
2E22 = λ2 − 1,
2E33 = λ2 − 1,

E12 = E13 = E23 = 0.

The body is extended (compressed if λ < 1) equally in all directions with angles pre-
served. This change of volume with no change of shape is called pure dilatation.
E11 , E22 , E33 are called Normal Strains.

♦ Definition 2.4.11. Ekk = E11 + E22 + E33 is called the Dilation or Dilatation. It
is often denoted by ∆ and is usually associated with a change of volume.

 Example 2.4.12 (Simple Shear). Consider the deformation of a rectangle with no

deformation along the X3 axis, x1 = X1 + 2εX2 , and x2 = X2 , x3 = X3 , for ε > 0,
which is called simple shear. The effect of the deformation is to transform the initial
rectangle into a parallelogram, Fig. 2.8. Our objective is to obtain an interpreta-
tion of the off-diagonal components of the Lagrangian strain tensor in particular the
component E12 .
So we have
∂x1 ∂xk
= 2ε and otherwise = δkl
∂X2 ∂Xl
As a result, E12 = E21 = 21 (2ε + 0) = ε, E22 = 2ε2 , are the only non-zero values of
Ekl .
For small deformation |Ekl |  1, tan θ12 = 2ε = 2E12 ≈ θ12 . This implies that E12
represents half the angle of rotation. In comparison with the diagonal elements which
are usually associated with a change of volume, E12 is usually associated with a change
of shape. E12 , E13 , E23 are called Shear Strains.

2.5 Deformation rate tensor

We recall that xk = xk (X, t) is the Lagrangian description of motion. Consider the
rate-of-change of an element
∂xk
dxk = dXl .
∂Xl

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 30

Differentiating this with respect to t

 
d d ∂xk
( dxk ) = dXl ,
dt dt ∂Xl
 
d ∂xk
= dXl , dXl is Lagrangian
dt ∂Xl
 
∂ ∂xk
= dXl ,
∂Xl ∂t
∂vk
= dXl ,
∂Xl
∂vk ∂xm
= dXl , Chain Rule
∂xm ∂Xl
∂vk
= dxl .
∂xl

We now derive a formula for the deformation rate tensor by considering the rate of change
of the square of length dx:
d d
( ds2 ) = ( dxk dxk ) ,
dt dt
d
= 2 ( dxk ) dxk ,
dt
∂vk
=2 dxl dxk , Product Rule
∂xl
∂vl
=2 dxk dxl , Rate of Strain
∂xk
 
∂vk ∂vl
= + dxl dxk ,
∂xl ∂xk
= 2dkl dxk dxl .

♦ Definition 2.5.1. We define the deformation rate tensor dkl by

 
1 ∂vk ∂vl
dkl = + .
2 ∂xl ∂xk

Property 2.5.2. One can show, similar to 2.4.5, that dij is a second order tensor.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 31

Supplementary material 2.5.3. We recall that ds2 −dS 2 = 2Ekl dXk dXl .
N
We now differentiate this with respect to time:
d d d
( ds2 − dS 2 ) = ( ds2 ) = (2Ekl dXk dXl ),
dt dt dt
dEkl
=2 dXk dXl ,
dt
∂Xk ∂Xl
= 2Ėkl dxm dxn .
∂xm ∂xn

We deduce by equating to the above statement that the deformation rate tensor can be
associated with the rate of Lagrangian strain, but they are not equivalent,
∂Xk ∂Xl
dmn = Ėkl .
∂xm ∂xn

The important result here is the deformation rate tensor, which is the main object,
as we will see a little bit later, to describe the rate of change of deformations in fluid
mechanics.

2.6 Stress

2.6.1 The Stress Tensor σij

A body undergoes deformation when subjected to forces. These forces may be mechanical,
electrical etc. There are three types of forces acting on a volume Ω contained within a
smooth surface ∂Ω, Fig. 2.9,

1. Internal forces between pairs of particles belonging to Ω for example electrical.

2. External forces acting on the material within ∂Ω, known as body forces for example
gravity.

3. Surface forces, transmitted across the boundary of Ω from the region outside of Ω

Surface forces are conveniently defined by the notion of stress vector. The vector T of
surface force per unit area acting on an element dS is called the stress vector acting on
the element. Each element dS is described by the associated normal vector n, Fig. 2.9.
Note that, the normal vector is coming out of the surface into the media which is acting
with the force T on the surface.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 32

Surface forces

n
Volume or body forces
dS
∂Ω
Ω

Internal forces

Figure 2.9: Illustration of different types of forces acting on a volume element Ω.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 33

V 23 V 22
V 13 V 12 V 21

V 11

x3
x2
x1

Figure 2.10: Illustration of stress tensor components

IMPORTANT

The main result in Continuum Mechanics is that the Stress Vector is a linear
function of the normal vector, i.e.

Tj = ni σij

Thus, we defined σij to be the Cartesian stress tensor, the proof will follow.

IMPORTANT

The first suffix corresponds to the surface on which the stress acts and the second
suffix refers to the component of stress.
♦ Definition 2.6.1. σ11 , σ22 , σ33 are called the normal stress components and σ12 , σ13 , σ23
are called the shear stress components.

 Example 2.6.2.

1. Consider the surface element with the normal vector n = (1, 0, 0)T , Fig. 2.10. Let’s
calculate the surface force components acting on this element using/assuming that

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 34

components of the stress tensor are known. One has

T1 = σ11 n1 + σ21 n2 + σ31 n3 = σ11 ,
T2 = σ12 ,
T3 = σ13 .
Stress vector across the surface with unit normal n is T = (σ11 , σ12 , σ13 )T .

2. Now, consider the element, which is opposite to the one in the previous example.
The normal vector is obviously n = (−1, 0, 0)T , thus
T1 = −σ11 ,
T2 = −σ12 ,
T3 = −σ13 .
Stress vectors across the surface with unit normal n is T = −(σ11 , σ12 , σ13 )T .

3. Now, consider another element with the normal vector n = (0, 1, 0)T , Fig. 2.10.
Then
T1 = σ21 ,
T2 = σ22 ,
T3 = σ23 ,
Stress vectors across the surface with unit normal n is T = (σ21 , σ22 , σ23 )T .

2.6.2 Properties of σij

1. Show that σij is a tensor.
Let L = (lij ), and a rotation with x0 = Lx. We write the stress vectors T = Ξn
with Ξ = (σij ) representing the matrix of the stress tensor. So that T and n will
transform to T 0 = LT and n0 = Ln, so that in the transformed coordinate system
Ξ0 satisfies
T 0 = Ξ0 n0 ,
LT = Ξ0 Ln,
LΞn = Ξ0 Ln,
LΞLT = Ξ0 LLT = Ξ0 I = Ξ0 , orthogonality of L, that is LLT = I,
LΞLT = Ξ0 .
The last result can be explicitly written as
σij0 = liα mβj σαβ ,
if we designate the elements of LT = (mβj ). Then since mβj = ljβ
σij0 = liα ljβ σαβ .
Thus σij is a second order Cartesian tensor.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 35

2. Stress tensor σij is symmetric.

This is due to the fundamental fact, conservation of angular momentum, and goes
here without a proof, if interested, look for details in recommended books.

2.7 Exercises - II
1. A displacement field in a body is given by

x1 = λX1 , x2 = λX2 , x3 = λX3 ,

where (X1 , X2 , X3 ) and (x1 , x2 , x3 ) are rectangular Lagrangian and Eulerian co-
ordinates and λ > 0 is a strictly positive constant. Determine the strain tensor
components EKL . Can you make a conclusion in this case?

2. Show that an anti-clockwise rotation of coordinates in a particular case about the

x3 axis is specified by
 
cos θ sin θ 0
L(x3 , θ) =  − sin θ cos θ 0 
0 0 1

where θ is the angle of rotation.

Hence, obtain, by use of the basic definition of the strain tensor,
 
1 ∂xm ∂xm
Ekl = − δkl ,
2 ∂Xk ∂Xl
the components of Ekl under the solid body rotation in clockwise direction xk =
L(x3 , θ)km Xm . Can you explain the result?

3. If given (or postulated) that any change in volume occurs only if the Jacobian of
transformation x(X, t) is not equal to 1, that is
∂x1 ∂x1 ∂x1
∂X1 ∂X2 ∂X3
∂(x1 , x2 , x3 ) ∂x2 ∂x2 ∂x2
J= = 6= 1.
∂(X1 , X2 , X3 ) ∂X1 ∂X2 ∂X3
∂x3 ∂x3 ∂x3
∂X1 ∂X2 ∂X3
show that the deformation

x1 = λX1 + kX2 , x2 = X2 /λ, x3 = X3

preserves the volume.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 36

4. Prove that the deformation rate tensor dij is a second order symmetric tensor. Hint:
the time derivative dtd (ds2 ) is a scalar.

5. Demonstrate, that the normal component of the stress vector T on a surface with
normal vector n is given by

Tn = T.n = ni σij nj .

Can we also express the tangential component in a similar way via the stress tensor
components?

6. Assume that the stress tensor σij (r) is given in a Cartesian frame of reference by
its components σ11 (r), σ12 (r), ..., σ33 (r) in a liquid. Express the stress vector on the
surface of the unit sphere
x21 + x22 + x23 − 1 = 0
in terms of the stress tensor components assuming that the liquid fills the exterior
of the sphere.

7. Assume that the stress tensor σij (r) is given in a Cartesian frame of reference by
its components σ11 (r), σ12 (r), ..., σ33 (r) in a liquid. Consider, as an example, the
following hypothetical boundary condition - it is given that the normal component
of the stress vector on the plane, which is perpendicular to vector K = [0, 3, 4], is
equal to a constant P0 . Formulate this boundary condition in terms of stress tensor
components.

8. Assume that the stress tensor σij (r) is given in a Cartesian frame of reference by
its components σ11 (r), σ12 (r), ..., σ33 (r) in a liquid. Consider, as an example, the
following boundary condition - it is given that the normal component of the stress
vector on the plane x3 = 0 is equal to its tangential component in the direction of
x1 . Formulate this boundary condition in terms of stress tensor components.

9. Assume that the stress tensor σij is given in a Cartesian frame of reference by its
components σ11 (r), σ12 (r), ..., σ33 (r) in a closed domain up to the boundary Γ with
a given outward normal vector n as

σij = βdij − αδij ,

where dij is the deformation rate tensor and α, β = const.

Consider the following boundary condition on Γ - it is given that the tangential
component of the stress vector [Tk ]1 in the x1 -direction is proportional to the normal
velocity vn , that is
[Tk ]1 = γvn ,
where γ = const. Express the boundary condition in terms of components of Eule-
rian velocities v1 , v2 , v3 only.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 37

10. The stress tensor σij of a liquid is given by components in a Cartesian frame of
reference  
1 β x1 + β
(σij ) =  α 1 x2 + γ  .
x1 x2 1

(a) Using the basic properties of stress tensor, determine α, β and γ.

(b) Determine the force acting in the x1 - and x2 -directions per unit area of the
surface given by f (x1 , x2 , x3 ) = x21 + 2x22 + 3x23 − 1 = 0 from the liquid occupying
the exterior of the closed domain defined by f .

(MA3FM/MA4FM) Autumn 2020

Chapter 3

The Navier-Stokes Equations

3.1 The Leibniz-Reynolds Transport Theorem

Lemma 3.1.1. The Jacobian of the transformation between Eulerian (x1 , x2 , x3 ) and
Lagrangian coordinates (X1 , X2 , X3 ) defined by

∂(x1 , x2 , x3 )
J=
∂(X1 , X2 , X3 )

dJ
satisfies the equation = Jdiv(v) in a continuum with velocity field v(x1 , x2 , x3 , t).
dt
Theorem 3.1.2 (The Leibniz-Reynolds Transport Theorem - 19th century). Consider
a material volume V (t), that is a volume consisting of the same material particles at
any time, let L(x, t) be some continuously differentiable function (this may be a scalar
function, a vector function or a tensor function) defined on V (t), v(x, t) is the velocity of
particles in V (t) then
Z  Z  
d dL
L(x, t) dx1 dx2 dx3 = + Ldiv(v) dx1 dx2 dx3 .
dt V (t) V (t) dt
This is a formula for “differentiating” over a volume which is moving with the fluid.

38
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 39

Proof. If we change from Eulerian (x1 , x2 , x3 ) to Lagrangian coordinates (X1 , X2 , X3 )

Z  Z 
d d
L(x, t) dx1 dx2 dx3 = L(x(X, t), t)J dX1 dX2 dX3 ,
dt V (t) dt V (0)
Z
d
= {L(x(X, t), t)J} dX1 dX2 dX3 ,
V (0) dt
Z  
dL dJ
= J +L dX1 dX2 dX3 ,
V (0) dt dt
Z  
dL
= J + LJdiv(v) dX1 dX2 dX3 ,
V (0) dt
Z  
dL
= + Ldiv(v) J dX1 dX2 dX3 ,
V (0) dt
Z  
dL
= + Ldiv(v) dx1 dx2 dx3 .
V (t) dt
Changing back to Eulerian (x1 , x2 , x3 ) coordinates completes the proof.

3.2 Conservation of Mass

Consider as before a volume element consisting of the same liquid particles V (t) with
continuous boundary δV , Fig. 3.1. Let ρ(x, t) be the fluid density defined in the volume
element V (t), then the conservation of mass principle is given by
Z 
d
ρ dx1 dx2 dx3 = 0.
dt V (t)

That is since the volume element consist of the same particles, the total mass flux, see
Fig. 3.1 & Fig. 3.2, through the surface δV is zero, and the mass is conserved.
Applying the transport theorem yields
Z  
dρ
+ ρdiv(v) dx1 dx2 dx3 = 0.
V (t) dt
since V (t) is arbitrary the integrand is equal to zero, namely
dρ
+ ρ div(v) = 0.
dt
The full derivative with respect to t has the form,
dρ ∂ρ ∂ρ ∂x1 ∂ρ ∂x2 ∂ρ ∂x3
= + + + ,
dt ∂t ∂x1 ∂t ∂x2 ∂t ∂x3 ∂t
∂ρ
= + v.∇ρ,
∂t
Dρ
= .
Dt

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 40

δV Surface normal vector

n
V (t ) dS

Surface infinitesimal
element

Figure 3.1: Illustration of conservation of mass and momentum in the volume element
V (t) consisting of the same particles at any t ≥ 0.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 41

Surface Surface flux (density) vector f

dS

Surface normal vector n

Figure 3.2: Flux Φ through a given surface dS is equal to amount of a physical quantity
passing through the surface per unit time. Examples are mass flux (kg/s), energy flux
(J/s), momentum flux ((kg· m/s)/s). The flux density vector f is introduced via Φ =
f .n dS, where n is the normal vector to the surface element with area dS. So the flux
density is flux per unit area and per unit time, for example mass flux density (kg/(m2 s)).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 42

Thus we have that

Dρ ∂ρ ∂ρ dvk
+ ρ div(v) = 0 or + vl +ρ = 0,
Dt ∂t ∂xl dxk

which is known as the equation of continuity.

IMPORTANT
Dρ
If the density ρ is constant, as it is, for example, for water then Dt
= 0, and the conser-
vation of mass reduces to
∂vk
∇ · (v) or = 0,
∂xk
which is known as incompressibility condition in fluid mechanics.

3.3 Conservation of Momentum

By the transport theorem we consider the conservation of momentum ρv in the volume
element V (t), Fig. 3.1.
Z  Z  
d d ∂vk
ρvi dx1 dx2 dx3 = (ρvi ) + ρvi dx1 dx2 dx3 .
dt V (t) V (t) dt ∂xk

Take
d ∂vk ∂ ∂ ∂vk
(ρvi ) + ρvi , = (ρvi ) + vm (ρvi ) + ρvi ,
dt ∂xk ∂t ∂xm ∂xk
∂vi ∂ρ ∂vi ∂ρ ∂vk
=ρ + vi + ρvm + vi vm + ρvi ,
∂t ∂t ∂xm ∂xm ∂xk
 
∂vi ∂vi ∂ρ ∂ρ ∂vk
=ρ + ρvm + vi + vm +ρ ,
∂t ∂xm ∂t ∂xm ∂xk
| {z }
=0 by Conservation of mass
∂vi ∂vi
=ρ + ρvm ,
∂t ∂xm
Dvi
=ρ .
Dt

Hence Z  Z
d Dvi
ρvi dx1 dx2 dx3 = ρ dx1 dx2 dx3 . (3.1)
dt V (t) V (t) Dt

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 43

Applying Newton’s second law to the fluid in V (t), the rate of change of momentum
Z 
d
ρvi dx1 dx2 dx3 ,
dt V (t)

equals the total force acting on the volume element V (t), which consists of surface forces
transmitted across the boundary δV (t) with outward normal vector n from the region
outside V (t) and the body forces, for example, gravity. We can therefore write
Z  Z Z
d
ρvi dx1 dx2 dx3 = σij nj dS + Fi dx1 dx2 dx3 . (3.2)
dt V (t) δV (t) V (t)

Where σij nj is the ith component of stress vector acting on the area element dS and Fi is
the ith component of the body force density - that is the force acting on unit of volume.
One can see, Fig. 3.2, that σij is the force density tensor, so that σij nj dS is the total
force on the surface element dS.
Revision, see also 10.5.

Theorem 3.3.1 (The Divergence Theorem). For a volume V (t), with boundary δV (t)
with normal vector n we have, if pj is a continuously differentiable vector field, that
Z Z
∂pj
dx1 dx2 dx3 = pj nj dS.
V (t) ∂xj δV (t)

This is often expressed in vector notation as

Z Z
∇ · p dx1 dx2 dx3 = p · n dS.
V (t) δV (t)

Applying the divergence theorem to the second term in (3.2) gives

Z Z
∂σij
σij nj dS = dx1 dx2 dx3 . (3.3)
δV (t) V (t) ∂xj

Substituting (3.3) and (3.1) into (3.2) yields

Z Z  
Dvi ∂σij
ρ dx1 dx2 dx3 = + Fi dx1 dx2 dx3
V (t) Dt V (t) ∂xj

Since V (t) is an arbitrary volume, we conclude that the integrands must be equal and
thus the
Eulerian Equation of Motion, also called Cauchy momentum equation is given
by

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 44

Dvi ∂σij
ρ = + Fi .
Dt ∂xj

3.4 Fluids
For fluids the independent variables are the Eulerian coordinates. We have the equations
for conservation of mass
∂ρ ∂ρ ∂vk
+ vi +ρ = 0,
∂t ∂xi ∂xk
and for the conservation of momentum
 
∂vi ∂vi ∂σij
ρ + vl = + Fi .
∂t ∂xl ∂xj
We however require an additional equation, since we have more unknowns in the system
then the number of equations. Indeed, in general, we have ρ, three components of velocity
vk and six components of the stress tensor σij (six, since σij is a symmetric tensor). So in
total ten unknowns but only four equations. For incompressible fluids, when ρ = const,
we have nine unknowns but still four equations. So somehow, we need to reduce the
number of unknowns to four, that is basically, for an incompressible fluids to define five
components of the stress tensor.

3.4.1 Stress Tensor in Fluids

For elasticity one can relate stress to strain via the Hooke’s Law,
σij = λδij EKK + 2µEij
which is a very poor model for fluids. Instead we relate stress to the rate of deformation,
 
1 ∂vk ∂vl
dkl = + .
2 ∂xl ∂xk
♦ Definition 3.4.1. A NEWTONIAN liquid is the one which has linear, in the velocity
gradients, relationship between the stress and the rate of deformation tensors. We assume
for such fluids that
 
2
σij = −pδij + µ 2dij − dkk δij + ζδij dkk ,
3

which is called the constitutive equation; here, parameters µ > 0 and ζ > 0 are the
dynamic and volume viscosities, respectively and p is the pressure.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 45

In the case of incompressible liquids, where,

∂vk
dkk = = 0,
∂xk
the constitutive equation is simplified,

♦ Definition 3.4.2. An incompressible NEWTONIAN liquid is the one which has,

σij = −pδij + 2µdij ,

and contains only one parameter, the dynamic viscosity µ.

The dynamic viscosity is the direct measure of friction or dissipation in fluids. For exam-
ple, consider some typical fluids or fluid-like substances, where the viscosity values span
many orders of magnitude.

Material Viscosity µ- expressed in units Newton · Second · m−2

Air 15o C 1.8 × 10−5

Water 20o C ≈ 10−3
Blood 37o C ≈ 4 × 10−3
Olive oil 25o C 0.08
Honey 20o C 2 − 10
Pitch 20o C 2.3 × 108

In a frictionless fluid, with no shear forces, where µ = 0,

σij = −pδij ,

and the only non-zero components of σij are σ11 , σ22 and σ33 , the pressure in the liquid
is the same in all direction, and σ11 = σ22 = σ33 = −p.

3.4.2 The Navier-Stokes equations for incompressible fluids

We have the conservation of momentum equation

 
∂vi ∂vi ∂σij
ρ + vl = + Fi , (3.4)
∂t ∂xl ∂xj

and the stress - rate of deformation relationship

σij = −pδij + 2µdij . (3.5)

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 46

The definition of the rate of deformation tensor is

 
1 ∂vi ∂vj
dij = + . (3.6)
2 ∂xj ∂xi

Substituting (3.6) into (3.5) yields

 
∂vi ∂vj
σij = −pδij + µ + ,
∂xj ∂xi

Differentiating this with respect to xj and taking into account

∂vk
= 0,
∂xk
gives
∂σij ∂p ∂ 2 vi
=− +µ 2. (3.7)
∂xj ∂xi ∂xj

Substituting (3.7) into (3.4) gives, the Navier-Stokes Equations for incompressible fluids.

∂ 2 vi
 
∂vi ∂vi ∂p
ρ + vl =− + µ 2 + Fi .
∂t ∂xl ∂xi ∂xj

Thus, for an incompressible fluid, we have four equations to find four unknowns (vi , p);
the equations of continuity and the Navier Stokes Equations, namely

∂vk
= 0,
∂xk

∂vi ∂vi

∂p ∂ 2 vi (3.8)
ρ + vl =− + µ 2 + Fi ,
∂t ∂xl ∂xi ∂xj
or in the vector-matrix form,

 ∇.v
 = 0,
∂v (3.9)
ρ + (v∇)v = −∇p + µ∇2 v + F .
∂t

3.4.3 The incompressible Euler equations

In a frictionless fluid µ = 0, we have

σij = −pδij ,

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 47

and the Navier-Stokes equations are reduced to the system of Euler equations

∂vk
= 0,
∂xk 

∂vi ∂vi ∂p (3.10)
ρ + vl =− + Fi ,
∂t ∂xl ∂xi
or in the vector-matrix form,

 ∇.v =
 0,
∂v (3.11)
ρ + (v∇)v = −∇p + F .
∂t

Now, both (3.8) and (3.10) are systems of partial differential equations, which require
boundary conditions to fully specify fluid dynamical problems and to find unique so-
lutions. The mathematical problem in Fluid Mechanics is said to be well-defined, if the
set of boundary conditions to the system of the governing equations is sufficient to obtain
a unique solution.

IMPORTANT

We notice that the main difference between (3.8) and (3.10) is the highest order of partial
derivative with respect to space coordinates (x1 , x2 , x3 ). The order is 2 for (3.8) and
the order is 1 for (3.10). So, similar to the theory of ordinary differential equations, the
systems (3.8) and (3.10) require different number of boundary conditions. We have to
keep this fact in mind when formulating proper fluid mechanical problems.
The above statement can be illustrated by ordinary differential equations.

 Example 3.4.3. Consider a second order differential equation for f : R → R

d2 f
− f = 0.
dx2
Its general solution is f = C1 exp(x)+C2 exp(−x) with two unknown constants C1 and C2 .
To define a particular solution, one needs to have two conditions to define the constants.
On the other hand, in the case of the first order differential equation
df
−f =0
dx
the general solution is f = C exp(x), and one requires only one condition. So, each time
we solve a problem, we have to take into account the order of the differential equation.
The same sentiment is true for fluid mechanical problems.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 48

3.5 Boundary conditions to the Navier-Stokes equa-

tions
There are basically two types of boundaries and, hence, two types of boundary conditions;
solid boundaries, between a solid and a liquid, and free surfaces separating a liquid and a
gas (or vacuum or another immiscible liquid).
Note, the number and the type of admissible boundary conditions depend on the system
of partial differential equations we use. In the next discussion, we assume that we consider
three-dimensional problems for the full system of the Navier-Stokes equations, unless
otherwise specified. In solving particular problems, we will know about exceptions and
why, and how we need to treat them.

3.5.1 Solid boundaries

On a solid boundary, Fig. 3.3, moving with velocity VS , we usually require that the
velocity of the liquid is equal to the velocity of the moving substrate, that is,

v = VS ,

which can be decomposed into two equations for the normal and tangential projections
(in a similar way as any vector can be decomposed into two or three vectors
perpendicular to each other),

v · n = VS · n, (3.12)

v k = VSk , (3.13)
where n is the unit normal vector at the solid surface, which is pointed, conventionally,
into the liquid and,
v k = v − (v · n)n. (3.14)

The decomposition is due to different physics involved in the formation of the conditions
at the boundary. The first equation, (3.12), is refereed to as impermeability condition
and the second one, (3.13), is the no-slip boundary condition. Note, that the no-slip
boundary condition is a vector equation, so in fact we have two scalar conditions here.
Note:

From (3.14), one can see that the tangential component of a vector can be formally
extracted by means of  
v k j = vi (δij − ni nj ),

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 49

Liquid x3
n
Liquid‐solid boundary
x2
u Solid VS
Liquid velocity Solid velocity x1

Boundary condition
u = VS
Split into two conditions
u.n = VS .n Normal component, impermeability condition
u|| = VS || Tangential component, no‐slip boundary condition

Figure 3.3: Illustration of boundary conditions at solid substrates. In the picture, the
boundary is a plane x3 = 0 with a normal vector n = (0, 0, 1). Note, that the normal
vector is pointing into the fluid.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 50

which is equivalent to
v k = v(I − nn)
This method may be convenient in formal calculations.

Note, that there are always three boundary conditions on a solid-liquid boundary in
a general 3-D case of the full system of the Navier-Stokes equation, even if conditions
are different from what we have just learnt. In some situations boundary conditions may
differ from the standard no-slip and impermeability conditions.
It is important to remember that the number of boundary conditions depends
on the differential equation under consideration. In the above note, we have
specifically stated that we are considering 3-D cases with full system of the
Navier-Stokes equations.

Special cases: 2-D case.

If the problem dimension is reduced, for example in a two-dimensional case, when we

ignore one coordinate direction, then there are two boundary conditions left.
 Example 3.5.1. Consider the boundary value problem shown in Fig. 3.3. The bound-
ary is a plane x3 = 0 with a normal (unit) vector n = (0, 0, 1). In a 3-D case we had
two scalar conditions on the tangential velocity v1 = VS1 and v2 = VS2 - no-slip boundary
conditions, and one condition on the normal velocity v3 = VS3 - impermeability condition.
Now, presume that the flow is two-dimensional in the (x2 , x3 )-plane. Then, we will have
one condition of no-slip v2 = VS2 and one condition of impermeability, v3 = VS3 . So that
we dropped one condition in the parallel to the substrate direction, the direction we ignored
making an assumption that the flow is two-dimensional.

Sometimes, it is obvious that some condition will be satisfied automatically, due to the
problem formulation. Other special cases are when we change the order of the Navier-
Stokes equations, see static problems and inviscid and potential flows at µ = 0, for
example.

3.5.2 Free surfaces

On a moving free surface, Fig. 3.4, we require that the velocity and the stress vector are
continuous through the interface, that is,
(1) (2)
v (1) · n = v (2) · n vk = vk ,

and
(n.σ (1) ) · n = (n.σ (2) ) · n (n.σ (1) )k = (n.σ (2) )k .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 51

Gas or Liquid (1)

Continuity of
velocity v and surface stress vector T

v (1)
k T (1) (1)
k =n j σ jk
Interface

v (2k ) T (2) (2)

k =n j σ jk
n
Normal
x3 vector

Liquid (2)
x2
x1

Figure 3.4: Illustration of boundary conditions at free (flexible moving) surfaces in a 3-D
case for the full system of the Navier-Stokes equations.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 52

These conditions, the total number in a 3-D case is 6, can be represented explicitly,
(1) (2) (1) (1) (2) (2)
vk nk = vk nk , [v k ]j = vi {δij − ni nj } = vi {δij − ni nj } = [v k ]j ,
(1) (2) (1) (2)
ni σij nj = ni σij nj , ni σij (δjk − nj nk ) = ni σij (δjk − nj nk ).

The different number of boundary conditions on a free surface in comparison with the
set of boundary conditions on a solid substrate can be explained by the following line of
arguments. First of all, unlike a solid boundary, which has a well defined shape and a
given velocity, the shape of a free surface is unknown and the motion of the free surface is
not defined or given in advance in a general case - so it has to be determined. Secondary,
though this is in fact the result of flexibility of the free surface, the free surface separates
two regions, and in both of them the motion is unknown, that is also to be determined.
So, to define these extra unknowns, we need more conditions to obtain a unique solution
to a problem. Note, only normal to the interface component of velocity affects motion of
a free surface, while tangential component, simply by geometric argument does not affect
the shape or its evolution.
The role of different conditions on a free surface can be interpreted differently from the
role of boundary conditions on a solid substrate. On a solid, the continuity of velocity
simply fixes the values of liquid velocity at the boundary. You noticed, that on a free
surface, we also have continuity of velocity, but both velocities v (1) and v (2) are unknowns
in general. Consider an example from ordinary differential equations to illustrate this
point here.
 Example 3.5.2. Consider the following one-dimensional ODE defined on some inter-
vals x ∈ (a, xo ) and x ∈ (xo , b) with a common point xo ,
df
− f = 0.
dx
Its general solution is well-known f = C exp(x), where C is a free constant to be found.
So, in two intervals, we have two independent solutions f (1) = C1 exp(x) (x ∈ (a, xo )) and
f (2) = C2 exp(x) (x ∈ (xo , b)), with two constants to be found C1,2 . Now, by analogy with
the continuity conditions at a free surface, let’s set a condition at point xo ∈ (a, b), that
f (1) (xo ) = f (2) (xo ). That is, we have in fact that C1 = C2 . So, we found a connection
between two solutions, but still have one degree of freedom to fix it completely. Conclusion
- we need an extra condition to find a unique solution.

The same argument applies in general to the continuity of velocity at a free surface - we
can connect two regions separated by the flexible free boundary, but we are unable to fix
a unique solution, we need more conditions to determine it uniquely, in general, we need
exactly extra three conditions to have six conditions in total. So, one can say
that we have continuity of velocity to connect two regions and to determine the motion
of the free surface, and we need continuity of the stress surface vector to obtain a unique
solution.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 53

Special cases: 2-D case.

In a 2-D case, we can drop two continuity conditions in the direction of the coordinate
we ignore.
 Example 3.5.3. Consider the boundary value problem depicted in Fig. 3.4, but now
assuming that the flow is two-dimensional in the (x2 , x3 )-plane. Since the problem became
two-dimensional, we drop one condition of continuity of velocity and one condition of
continuity of the stress surface vector in the x3 -direction.

3.6 Boundary conditions to the Euler equations

In the case of the Euler equations, the highest order of the differential operator with
respect to spatial coordinates (x1 , x2 , x3 ) is ONE. As a consequence, we will have lesser
number of boundary conditions required in this case to obtain a unique solution. In
particular,

3.6.1 Solid boundaries

On a solid boundary, Fig. 3.3, we need only one condition, which is in the normal to the
solid interface direction, that is the impermeability condition

v · n = VS · n.

In the problems associated with the Euler system of equations, the parallel, to the solid
substrate, component of velocity is not fixed at the boundary and can take any value. An
attempt to specify it would lead to an overdetermined problem for the Euler equations.
This is often considered as a weak point of the Euler system of equations in describing
liquid flows involving interaction with a solid boundary, since apparently, the physical
no-slip condition can be easily violated. This is indeed often the case. We will return to
this issue later with more details.

3.6.2 Free surfaces

In the case of problems for the Euler equations, similar to the case of solid boundaries,
we drop both continuity conditions in the parallel to the free surface direction. The
remaining boundary conditions are, first of all, for the normal component of velocity
(1) (2)
vk nk = vk nk , which is called the ’kinematic’ boundary condition, and secondary, for the
(1) (2)
normal component of the stress vector ni σij nj = ni σij nj , which is called the ’dynamic’
boundary condition.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 54

Supplementary material 3.6.1. Kinematic boundary condition

N

The kinematic boundary condition can be represented in a different form if we presume

that the free surface is given by f (x1 , x2 , x3 , t) = 0, where f (x) is a smooth function
such that ∇f 6= 0. Then, according to the continuity of velocity, the boundary velocity
V is equal to the velocity of liquid particles associated with the free surface defined by
f = 0. That is in the limit ∆t → 0, we have
 
∂f
f (x + V ∆t, t + ∆t) − f (x, t) = ∆t + V ∇f = 0.
∂t

Then directly the kinematic condition is

∂f ∂f Df
+ V ∇f = + v∇f = =0
∂t ∂t Dt
on f = 0.

Dynamic boundary condition.

In the case of the Euler equations, the constitutive equation takes the simplest form
σij = −pδij , where p is the pressure. Then the dynamic condition is in fact the continuity
of the pressure, that is
p(1) = p(2) ,
since ni σij nj = −pδij ni nj = −pnj nj = −p and nj nj = 1 - unit vector.

Consider now a particular, but simplified example.

 Example 3.6.2. Consider an air-liquid flat stationary interface, which is geometrically

the (x1 , x2 )-plane at x3 = 0; the fluid occupies the lower half space x3 < 0.
It is assumed that the air is basically frictionless, so that the stress tensor in the air can
be approximated by

σij = −p0 δij ,

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 55

where the air pressure p0 is assumed to be a given constant.

Formulate boundary conditions at the free surface, separating the liquid and the air. The
dependent variables in the air are marked by superscript (1) and in the liquid by (2).
a) The normal vector to the free surface, pointed into the liquid, is n = (0, 0, −1). Note,
it has only one non-zero component n3 .
b) We notice that the problem formulation says, ”stationary interface”. Hence, the normal
(1) (2) (1) (2)
velocity component at the free surface is simply zero, uk nk = uk nk = 0 or u3 = u3 = 0
at x3 = 0. We have learnt, that this condition can be used to define the interface shape,
the motion of the free surface. In our case, the free surface is given, so that we may, in
principle (see the comment below), omit this condition if solving a problem in the fluid
domain.

IMPORTANT A note of caution here. While we have presumed in the formulation that
the interface shape is given, this is rather artificial. In fact, we have already circumscribed
the problem in such a way that the boundary should be flat, that is bound to be flat. This
way, we have lost a lot of degrees of freedom - not every flow will have a flat boundary.
In reality, the free surface shape is almost always unknown, that is it has to be found and
that condition uk nk = 0, if the interface is stationary, that is the flow is steady, can not
be omitted.
(1) (2) (1) (2)
c) For the tangential velocity, we have u1 = u1 and u2 = u2 . But, the air is described
by the Euler system of equations, according to the problem specification saying, ”the air
is basically frictionless”. On the other hand, the liquid is supposed to be described by the
full system of the Navier - Stokes equations. If we fix the tangential velocity in the air,
we over-determine the problem for air motion. On the other hand, for the liquid motion,
this condition is not necessary, it is only to connect to the air domain of the problem. For
this reason, we can declare that those two conditions for the tangential components are
redundant in this case.
d) Consider now the conditions for the stress vector. These conditions are necessary to
define the liquid motion in x3 < 0.
The first condition, for the normal component of the stress vector is,
(1) (2)
ni σij nj = ni σij nj .

In the air
(1)
ni σij nj = −p0 ni δij nj = −p0 .
Note n.n = ni ni = 1, since n is the unit vector.
In the liquid, which is marked by the superscript (2), we have explicitly for the normal
component
(2) (2) (2)
ni σij nj = n3 σ33 n3 = σ33 .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 56

That is  
(2) (2) ∂v3 ∂v3 ∂v3
σ33 = −p +µ + = −p(2) + 2µ = −p0 .
∂x3 ∂x3 ∂x3

The second condition is for the tangential components,

(1) (2)
ni σij (δjk − nj nk ) = ni σij (δjk − nj nk ).

In the air,
(1)
ni σij (δjk − nj nk ) = −p0 δij ni (δjk − nj nk ) =
= −p0 nj (δjk − nj nk ) = −p0 (nk − nj nj nk ) = −p0 (nk − nk ) = 0.
In the liquid,
(2) (2) (2) (2)
ni σij (δjk − nj nk ) = n3 σ3j (δjk − nj nk ) = n3 σ3k − σ33 nk .

So, we have two equations, since for k = 3, it is an identity 0 = 0,

(2)
n3 σ31 = 0,
(2)
n3 σ32 = 0.

In summary, we have three boundary conditions for the Navier-Stokes equations,

(2) (2) (2)

σ33 = −p0 , σ31 = 0, σ32 = 0.

3.7 Exercises - III

1. Demonstrate that for an incompressible fluid

J = 1,

where J is the Jacobian of transformation

∂(x1 , x2 , x3 )
J= .
∂(X1 , X2 , X3 )

2. Fluid flow in a 3-D case is given by velocity components u1 = 1 + x3 cos x1 , u2 =

1 + x2 x3 (sin x1 + 1), u3 = 1 + Ax23 . At what value of A the flow will be consistent
with that of an incompressible liquid?

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 57

3. Observations of a two-dimensional flow along the four edges of a unit square with
vertices at (0, 0), (1, 0), (0, 1) and (1, 1) has resulted in the following distributions
of velocity on the square boundaries

u = (x2 , x2 ), x1 = 0

u = (1, 1), x1 = 1

u = (x1 , x1 ), x2 = 0

u = (1, 1 + x1 (1 − x1 )), x2 = 1.

It is given that the flow domain has neither internal boundaries nor sources nor
sinks of the liquid. Could this flow be incompressible? You may, if necessary, use
the result without a proof that for a closed domain Ω with a piecewise continuous
boundary ∂Ω with a normal vector n, if p is a continuously differentiable vector
field, then ZZZ ZZ
∇ · p dΩ = p · n dS,
Ω ∂Ω
where dS is an infinitesimal element of the boundary.

4. Consider a static 3-D case, when a bulk of a stationary Newtonian (incompressible)

liquid is bounded by a solid. The solid is the stationary plane x3 = 0 and the liquid
occupies the upper half of the space, x3 > 0. Try to formulate boundary conditions
at the solid in this case, if it is known that the solid is impermeable and the liquid
does not slip on it. Assuming that the stress tensor in the liquid has been found,
determine the force (density or stress vector) acting on the solid from the liquid,
and the force (density or stress vector) acting from the solid on the liquid. Does the
viscosity coefficient play any role in this case?

5. Consider a flow of a viscous Newtonian (incompressible) liquid described by the full

system of the Navier-Stokes equations over a stationary solid substrate in a 3-D case.
It is given that the solid boundary is the plane x3 = 0, while the liquid occupies
the upper half-space, x3 > 0. Somehow, it has been found that the true boundary
conditions at the solid substrate are those of no-slip and impermeability. At the
same time, it has been found that the liquid pressure is positive p = p0 > 0 and
constant, and the velocity in the flow has only one non-zero component v2 (x3 ), which
is a function of the x3 -coordinate only. Given the explicit dependence v2 = V0 x3
(V0 > 0 is some constant), determine the force (density or stress vector) acting from
the solid on the liquid, and the force (density or stress vector) acting from the liquid
on the solid. You may presume that the viscosity µ of the liquid is known.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 58

6. Consider a flow of a viscous Newtonian (incompressible) liquid described by the full

system of the Navier-Stokes equations over a stationary solid substrate in a 3-D case.
It is given that the solid boundary is the plane x3 = 0, while the liquid occupies the
upper half-space, x3 > 0. Somehow, it has been found that the no-slip boundary
condition is violated at the solid substrate such that v(x1 , x2 , 0) = V 0 , where V 0
is some constant slip velocity. At the same time, it has been established that the
solid substrate is still impermeable. Formulate the necessary system of boundary
conditions at the solid substrate to obtain a well-defined mathematical problem.

7. Consider a flow of a viscous Newtonian (incompressible) liquid described by the full

system of the Navier-Stokes equations over a stationary solid substrate in a 3-D case.
It is given that the solid boundary is the plane x3 = 0, while the liquid occupies the
upper half-space, x3 > 0. Somehow, it has been found that the no-slip boundary
condition is violated at the solid substrate such that the liquid velocity at the
substrate v(x1 , x2 , 0) is linearly proportional (coefficient of proportionality β > 0, a
positive constant) to the tangential surface force density acting from the liquid on
the solid. At the same time, it has been established that the solid substrate is still
impermeable. Formulate the necessary system of boundary conditions at the solid
substrate to obtain a well-defined mathematical problem. You may presume that
the viscosity µ of the liquid is known. Note, the result is known now in microfluidic
community as the Navier boundary condition.

8. Consider a flow of a viscous Newtonian (incompressible) liquid described by the full

system of the Navier-Stokes equations over a stationary porous substrate in a 3-D
case. It is given that the porous boundary is the plane x3 = 0, while the liquid
occupies the upper half-space, x3 > 0. Somehow, it has been found that the no-slip
boundary condition is violated at the porous substrate such that the liquid velocity
at the substrate v(x1 , x2 , 0) is linearly proportional (coefficient of proportionality
β > 0, a positive constant) to the tangential surface force density acting from the
liquid on the solid. At the same time, it has been established that the normal
liquid volume flux density is proportional (coefficient of proportionality is γ0 > 0, a
positive constant) to the liquid pressure at the boundary such that the flux goes into
the substrate if pressure is positive. Formulate the necessary system of boundary
conditions at the solid substrate to obtain a well-defined mathematical problem.
You may presume that the viscosity µ of the liquid is known.

9. Determine boundary conditions at a stationary interface (the shape of the interface

is known) between two Newtonian incompressible and inviscid fluids.

(MA3FM/MA4FM) Autumn 2020

Chapter 4

Static Problems of Fluid Mechanics

If the liquid is at rest, then v = 0. In this case, all components of the deformation rate
tensor dij = 0, and the stress tensor is simply, according to (3.5),

σij = −pδij . (4.1)

The equation of continuity is satisfied automatically and is thus redundant, while the
Navier-Stokes equations are greatly simplified and reduced to
∂p
− + Fi = 0. (4.2)
∂xi

If the external volume force is conservative or potential, as is the case of gravitational

forces, then
F = −∇Φ,
where Φ is a scalar function, the potential of the force F .
Using this result, the Navier-Stokes equations can be represented in the following way

∇p + ∇Φ = 0.

That is
∇(Φ + p) = 0
and hence
Φ + p = C = const.
So, in a stationary case, we can find exact solution to the problem. Can we also define
the shape of a stationary interface between, say, a liquid and a gas?
In a stationary case, there is only one boundary condition, that is required. Indeed, the
continuity of velocity conditions are all redundant. Also, there is no spatial derivatives
associated with the velocity field, and off-diagonal components of the stress tensor and

59
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 60

hence all tangential components of the surface stress vector are zero, see (4.1). So the
condition of continuity of the tangential components of the stress vector are also redun-
dant. There is only one boundary condition left, which can be used to define the interface
shape - continuity of normal component of the stress vector, that is
(1) (2)
σij ni nj = σij ni nj ,

which is equivalent to continuity of pressure

p(1) = p(2) .

In the particular case when the boundary, say Γ, is between a liquid and a gas, where
the pressure is given, for example p = p0 = const, pressure in the liquid at the surface is
defined from the boundary condition

p = p0 .

At the same time, in the liquid and at the surface in particular

(p + Φ)Γ = C = p0 + ΦΓ .

Apparently, the last equation defines a surface in 3-D space or to be precise a family of
surfaces,
Φ(x1 , x2 , x3 ) = C − p0 .
An extra condition is needed to define C, but apart from this, the shape is defined.
Consider some examples.

 Example 4.0.1. Consider a liquid in a tank with solid walls and a free surface. The
external force of gravity is acting in the x3 -direction, that is the force acting on the liquid
unit volume element is F = (0, 0, −ρg0 ), where g0 is the standard gravity constant, which
is g0 = 9.8128 m/s2 and ρ is the liquid density. Define the stationary state of the liquid
in the tank.
First, let’s define the potential of the gravity force density. From

−∇Φ = F ,

we have explicitly
∂Φ ∂Φ ∂Φ
= ρg0 , = = 0,
∂x3 ∂x1 ∂x2
that is
Φ = ρg0 x3 + C0 .
Here C0 is a constant, but it is not important, since it can be merged with the constant of
integration later.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 61

Then the pressure in the liquid,

p = −ρg0 x3 + C,
where C is a constant.
Now, at the free surface,
p = p0 = −ρg0 x3 + C. (4.3)

The last equations implies that the free surface is defined by x3 = (C − p0 )/ρg0 = const,
so that the free surface is a plane, but its position in space is arbitrary at the moment and
can be fixed using some extra condition.
We can choose the origin of the coordinate system such that at the the free surface x3 = 0.
Then, the free constant is defined C = p0 , and finally the distribution of pressure in our
system is,
p = p0 − ρg0 x3 .

Note, here we got rid off the constant by using the extra condition that our coordinate
system is at the surface. But there should be here no mistake that we have found that
position in space. The trick just allowed us to simplify the final answer to get rid off the
awkward constant, but in reality the position in space is still undefined. Illustration of
this trick is given in Fig. 4.1. The absolute position of the surface, as one can see, is
determined by the particular geometry and the amount of the liquid.
To summarize, in general the free surface shape is governed by the distribution of
the external potential, similar to (4.3),
p = p0 = −Φ + C,
which is effectively Φ(x1 , x2 , x3 ) = const. So the shape of the free surface is defined by the
potential of the external field.
Note, in static problems, we do not have any specific conditions on solid
boundaries. The number of conditions on a free surface is also reduced to just
one condition.
 Example 4.0.2. Consider Φ = − (x2 +x21+x2 )1/2 , which is the global gravity force due to
1 2 3
spherical bodies, like the Earth and other planets. In this case, the surface is defined by
x21 + x22 + x23 = const
which is the equation of a spherical surface. This result demonstrates that simple, and
well-known, fact, that the ocean surface is almost spherical on large length scales. See Fig
4.2.
On the other hand, note, if the potential is disturbed, as it is often the case, by another
planet, like the Moon, then there are disturbances seen on the liquid surface. This is the
origin of tidal waves. See Fig 4.3.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 62

x3
(a)
x2
x1

(b)

x3

x2
x1

Figure 4.1: Illustration of free surface shape definition by placing the coordinate system
on the surface element. In both cases (a) and (b) the free surface is given by the same
equation x3 = 0, but as one can see the absolute position of the free surface is different
and depends on the amount of the liquid in the tank of a given volume. So the actual
extra condition will be the liquid amount (volume) in the tank of a given volume.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 63

Inspired by Ocean
static example

Figure 4.2: Illustration of the stationary state of a liquid in the potential Φ =

− (x2 +x21+x2 )1/2 . The potential is equivalent of the gravity potential of a spherical body
1 2 3
(planet) like the Earth.

Ocean and tidal waves

(static picture)
Inspired by

Actual, dynamic picture

Figure 4.3: Illustration of the tidal waves, which may be understood as a stationary state
of a liquid in the spherical body gravitational potential Φ = − (x2 +x21+x2 )1/2 disturbed by
1 2 3
the presence of another gravitating spherical body. The picture is not up to the scale
of the effect (which can be just about 1 m) and demonstrates only the principle of a
liquid in such an equilibrium. The actual pattern of the tidal wave phenomenon is more
complicated, as one can see from the illustration at the bottom (there are two bulges),
and is conditioned by the joint action of the Moon gravity and the centrifugal acceleration
of the Earth. In reality, there is also some contribution from the gravitational potential
of the Sun, which is though weaker than that of the Moon.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 64

Surface shape of a ferrofluid under the

action of magnetic forces.

Figure 4.4: Free surface of a liquid can take amazing forms. Here is a special ferro-fluid
under the action of magnetic fields.

Supplementary material 4.0.3.

N

 Example 4.0.4. Consider now a solid body of volume V0 immersed into the sta-
tionary liquid. We are going to find the total force acting on the body in the stationary
case.
The total force Π is given by the surface integral, over the total surface of the body,
of the stress vector,
I I I I I
T dS = Π, Tk dS = ni σik dS = − pδik ni dS = − pnk dS = Πk
S S S S S

The last equation can be transformed using the Corollary to the Gauss Divergence
Theorem, namely,
Corollary 4.0.5. I Z
f dS = ∇f dx1 dx2 dx3 ,
V0

to have I Z
∂p
− pnk dS = − dx1 dx2 dx3 = Πk .
S V0 ∂xk

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 65

Atmospheric pressure

p0

Buoyancy force
Π 3= ρ g 0 V 0
x3
p= p0 x 3=0
x2
x1

n
V0

p= p0− ρ g 0 x 3

Gravity force
F 3=−ρ B g 0 V 0

Figure 4.5: Illustration of the stationary state of a liquid and the buoyancy force on a
solid body of volume V0 immersed into the liquid - the Archimedes (Syracuse, 287 BC –
212 BC) Law. The liquid density is ρ and the solid body density is ρB .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 66

In the stationary case,

p = p0 − ρg0 x3 ,
and the only non-zero component of the gradient is at k = 3,
∂p
= −ρg0 .
∂x3

So, there is a net force in the x3 -direction, Π3 = ρg0 V0 . This force is directed upwards,
opposite to the net gravity force F3 = −ρB g0 V0 and is equal to the weight of fluid
displaced by the body. This is well known buoyancy force, and we have just derived
the Archimedes (Syracuse, 287 BC – 212 BC) Law.

4.1 Exercises - IV
1. Determine distribution of pressure and the shape of the free surface of a stationary
state of Newtonian incompressible liquid in the external potential field with the
potential Φ = αx3 − βx2 , α, β > 0 given in a Cartesian frame of reference. The free
surface interface is between the liquid and a gas with constant pressure p0 . One can
assume, if necessary, that the liquid is in a tank with solid walls.

2. Consider a viscous liquid contained in a cylindrical reservoir with solid walls, which
is spinning with constant angular velocity Ω around its axis of symmetry, which is
the x3 -axis of the coordinate system, as is shown in Fig. 4.6. The gravity force
F g = −ρg0 i3 (g0 is gravity constant and ρ is liquid density) is along the x3 -axis and
is directed opposite to the coordinate vector direction, that is downwards, as in the
picture. Note, one can introduce a cylindrical coordinate system for this problem
with the cylindrical or longitudinal axis in the x3 -direction. It is given that in the
coordinate system rotating with the reservoir, the liquid is at rest and there is a
force, so called centrifugal force, which is directed along the radius of the cylindrical
coordinate system, that is perpendicular to the x3 -axis. The centrifugal force acting
on unit volume of the liquid is equal to F s = ρΩ2 r, where r is the radial position
vector in the cylindrical coordinate system lying in the plane x3 = const. That is
the centrifugal force only depends on the distance from the cylindrical axis. Define
distribution of pressure in the liquid and the shape of the free surface. One can
assume that the pressure in the gas is constant and is equal to p0 .
One can use the fact that in cylindrical coordinates, the nabla operator ∇ (to
∂ ∂ 1 ∂
calculate the gradient) is given by ∇ = iz + ir + iθ , where θ is the polar
∂z ∂r r ∂θ
angle in the plane x3 = const, iz is a unit vector in the x3 -direction, ir is a unit
vector in the radial direction in the plane x3 = const and iθ is a unit vector in the

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 67

Surface shape under the action of

centrifugal and gravity forces due to a
rotation

Centrifugal
force Gravity force
F= ρ Ω2 r x3 F 3=−ρ g 0

r x2
x1

Ω
Rotation

Figure 4.6: Illustration of the free surface shape in a rotated liquid. The rotation angular
velocity Ω is constant.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 68

perpendicular to thepir -direction in the plane x3 = const. The radial position vector
r = rir , where r = x21 + x22 .

4.1.1 Reference material

♦ Definition 4.1.1 (Cylindrical coordinate system). In mathematics, the cylindri-
cal coordinate system is a three-dimensional coordinate system in which position of
each point in space is determined by a distance z from a fixed point, origin of the
coordinate system, along the longitudinal or cylindrical axis, and polar coordinates
(a distance and an angle) in the plane perpendicular to the cylindrical axis, r and θ,
Fig 4.7. If we choose x3 -axis of a Cartesian coordinate system with the same origin
in the direction of the cylindrical axis and count the polar angle anti-clockwise from
the x1 -axis, then the position of a point x1 , x2 , x3 can be represented as

x1 = r cos θ, x2 = r sin θ, x3 = z.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 69

x =( r , θ , z )
x3
z

r
θ

x1

Figure 4.7: Cylindrical coordinate system.

(MA3FM/MA4FM) Autumn 2020

Chapter 5

Inviscid Flow (µ = 0)

Taking into account the constitutive equation at µ = 0, σij = −pδij , the Eulerian equation
of motion  
∂vi ∂vi ∂σij
ρ + vl = + Fi
∂t ∂xl ∂xj
is therefore reduced to  
∂vi ∂vi ∂p
ρ + vl =− + Fi
∂t ∂xl ∂xi
or in a vector form  
∂v
ρ + (v∇)v = −∇p + F , (5.1)
∂t
which is known as the system of Euler equations with the incompressibility condition,
∂vk
= 0.
∂xk

We note here again that in the case of the Euler equations we require fewer number of
boundary conditions.

5.1 Bernoulli Theorem

This would be convenient to represent the Euler’s equations in a bit different form. But,
first of all, let’s prove a simple identity, namely,
1
(∇ × v) × v = (v∇)v − ∇(v 2 ) (5.2)
2

70
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 71

Supplementary material 5.1.1. Proof.

N

 
∂vs ∂vs
((∇ × v) × v)k = εklm εlps vm = εmkl εpsl vm =
∂xp ∂xp

∂vs ∂vk ∂vm

= (δmp δks − δms δkp )vm = vm − vm =
∂xp ∂xm ∂xk
2
∂vk 1 ∂vm 1 ∂v 2
= vm − = (v∇)vk −
∂xm 2 ∂xk 2 ∂xk

∂v
We consider a steady flow only, that is when = 0. Using the identity, the Euler
∂t
equations (5.1) become
1
ρ(∇ × v) × v + ρ∇(v 2 ) = −∇p + F .
2

Let’s multiply the above equation by the unit vector s = v/v, which is directed along the
stream lines of the velocity vector field. Taking into account that s.(∇ × v) × v = 0 and
introducing the potential of the volume forces −∇Φ = F , one has,

ρv 2
 
∂
p+Φ+ = 0,
∂s 2
2
which tells us that the quantity p + Φ + ρv2 = const is conserved along the velocity field
lines. This result forms the Bernoulli Theorem.

Theorem 5.1.2 (Bernoulli 1738). During the steady motion of incompressible ideal (invis-
cid) liquid the sum of three components along the trajectory of a liquid particle is constant,

ρv 2
p + ρg0 z + = const;
2
the gravitational field is directed along the z-axis.

 Example 5.1.3 (Tube de Prandtl - Tube de Pitot). Engineers use Pitot tube to measure
the airspeed of aircrafts. Consider the theory behind operation of this kind of devices.
Schematically, the tube is shown in Fig. 5.1 (left) and the aircraft tube is illustrated in
Fig. 5.1 (right). We consider the case of a moving incompressible liquid only. Assume
that the liquid flow is parallel to the tube-axis in the far field, that is away from the device.
The tube is pointed directly into the fluid flow and has two holes to measure the pressure
in the moving liquid. One hole is located exactly on the tip of the tube. This point is a
special one and is called the stagnation point of the flow, Fig. 5.1. Note, that the normal

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 72

component of fluid velocity is usually zero anywhere on the tube surface, but on the tip at
the stagnation point, the velocity vector is perpendicular to the surface, so that not only
normal velocity is zero, but the fluid velocity is zero us = 0 too. We will designate the
pressure at this point by ps . Another hole is to measure pressure in the undisturbed flow
and is located on the side of the Pitot tube, at some distance from the stagnation point.
Consider the flow field line ending exactly at the stagnation point. We apply the Bernoulli
equation neglecting the effects of gravity. From the equation
ρu2∞
p∞ + = ps .
2
This equation actually relates the pressure at the stagnation point with the velocity of the
fluid in the far field u∞ .
The pressure value at the side hole is known to be equal to the undisturbed pressure, p∞ .
So, the Pitot tube provides both pressures, and from the above equation, the flow velocity
u∞ is given by s
2(ps − p∞ )
u∞ =
ρ

So, if the fluid density is available (for water ρ = 103 kg/m3 ), then the pressure difference
would give the value of the flow speed. Consider typical values for water. If p∞ = 105 P a,
that is the atmospheric pressure, and u∞ = 36 km/h (that is 10 m/s), then ps − p∞ =
p∞ /2, which is very easy to measure. The above equation is roughly applicable to air flows,
though air is compressible, so that we can only demonstrate the principle. In this case,
ρ = 1.2 kg/m3 , ps − p∞ ' 5 Pa at u∞ = 10 km/h. The pressure difference is smaller, but
it is still feasible to measure accurately - the resolution of new pressure gauges is about
0.1 Pa.

 Example 5.1.4 (The flow of inviscid fluids in a curved pipe). Consider steady flow
of an inviscid and incompressible liquid along a pipe of finite length but variable cross
section. Let p1 , S1 , h1 denote the pressure, area of cross section and the height of the pipe
above the ground (in the x3 -direction) at one end of the pipe, while at the other end, the
corresponding quantities are p2 , S2 , h2 , Fig. 5.2. Also let Q be the efflux and assume that
the tube shape changes smoothly so that the velocity v 1 and v 2 at the ends of the pipe are
approximately uniform over the cross section and parallel to the pipe axis. The gravity
force is in the x3 -direction. Analyse the flow and find the force acting on the bent part of
the pipe.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 73

u=u ∞
Far field p= p∞

u=u∞ >0 Stagnation

p= p ∞ point

Pressure sensors

u. n=0
p= p s> p ∞
x2
Normal vector
n x1
x3

Pressure sensor: front

Figure 5.1: Illustrations: experimental set-up to demonstrate the principle of Pitot tube,
aircraft Pitot tube and schematic representation of physical principles behind Pitot tube
operation - stagnation point at the tip.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 74

p1 , S1 , h1 p2 , S 2 , h2
n2
n1

S
S
x3
Gravity force

x2 F = − ρg 0 i 3
x1

Figure 5.2: Steady flow in a curved pipe of variable cross section

The conservation of mass implies

Q = S1 v1 = S2 v2 .

On the other hand, from the Bernoulli’s equation

ρv12 ρv 2
p1 + + ρg0 h1 = p2 + 2 + ρg0 h2
2 2
or
ρQ2 ρQ2
p1 + + ρg h
0 1 = p 2 + + ρg0 h2
2S12 2S22
which implies
2S12 S22 (p2 − p1 + ρg0 {h2 − h1 })
Q2 = .
ρ S22 − S12

Now, if we assume S2 > S1 , the flow is possible if

p2 − p1 + ρg0 (h2 − h1 ) > 0.

Note, however, that the direction of the flow is not defined - the steady flow is possible in
either direction.
Assume that the flow is from end 1 to end 2, then even if h2 > h1 , the steady flow
is possible if p2 − p1 > ρg0 {h1 − h2 }), which is the well-known result in the theory of
siphoning.
Note, if S1 = S2 , the flux Q is not defined, but the steady flow is possible if

p2 − p1 = ρg0 {h1 − h2 }

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 75

To find the force one needs to consider conservation of momentum ρu for the liquid
contained inside the pipe, volume V , as in Fig. 5.2. Note, the volume is not the material
volume since it is not moving with the liquid, it is fixed. Then, the total change of the
momentum in steady conditions (when all variables are not explicit functions of time, for
example ∂p/∂t = 0) is zero
Z Z Z Z
d 3 3
ρvd x = 0 = ρg0 e3 d x − pndS − ρu(u · n)dS.
dt V V S∪S1 ∪S2 S∪S1 ∪S2

The first term in the RHS is due to the volume gravity force, the second term is the force
due to the liquid pressure acting on the surface of the pipe - the total surface, and the
third term is convection of momentum (that is the flux) through the surface.
On the other hand, the total force on the pipe from the moving water is
Z
Π= pn dS,
S

which is equal to
Z Z Z
3
Π= ρg0 e3 d x − pndS − ρu(u · n)dS.
V S1 ∪S2 S∪S1 ∪S2

But the normal component of velocity u · n on the tube walls, area S, is zero, so
Z Z
ρu(u · n)dS = ρu(u · n)dS.
S∪S1 ∪S2 S1 ∪S2

Z Z Z Z
3
Π= pn dS = ρg0 e3 d x − pndS − ρu(u · n)dS.
S V S1 ∪S2 S1 ∪S2

So, the total force, if one assumes that velocity and pressure are uniform across the pipe
ends, Z
Π= pn dS = ρg0 V e3 − p1 S1 n1 − p2 S2 n2 − ρS1 v12 n1 − ρS2 v22 n2 .
S

When the pipe has a uniform cross-section, away from the bent part of course, S1 = S2 .
Then, neglecting the gravity force, one has
Z
Π= pn dS = −S1 (p1 + ρv12 )(n1 + n2 ).
S

The gravity force can be neglected when v12 S1 >> g0 V = g0 S1 Lp , that is v12 >> g0 Lp ,
where Lp is the pipe length. For a standard fire hose operation Lp = 2 m, v1 = 20 m/s,
g0 = 9.8128 m/s2 and the condition is fulfilled.
Again in the case of a fire hose with S1 = 10−3 m2 , ρ = 103 kg · m−3 and v1 = 20 m/s, only
contribution from velocity (second term in the above expression) would give Π ' 400 N or
the weight equivalent 40 kg.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 76

 Example 5.1.5 (The flow of inviscid fluids in channels). Consider a steady two-
dimensional and inviscid (µ = 0) liquid flow in a canal (a river flow would be also a
good example of that kind of events) with solid walls meandering along the sloped water
bed of the landscape, see Fig 5.3. The gravity force with potential, as before, Φ = ρg0 z
(g0 is the gravity constant on the Earth) is in the z-direction, so that the z-component
∂Φ
of the force is given by Fz = − . Assume that we have all the information about the
∂z
flow at some point along the canal, say at x0 , y0 , z0 and we know the position of the free
surface h(x, y) (by some remote measurements) anywhere in the canal. We would like to
determine the velocity on the water surface anywhere.
First of all, we know that at the free surface the liquid pressure p is equal to the ambient
gas/air pressure p = p0 , which we take constant over the landscape (which is a good
approximation).
Then, we also know that the free surface is the stream line of the flow in steady conditions
when the interface is stationary. Hence, we can directly apply the Bernoulli theorem to
find out the tangential velocity of the stream anywhere on the free surface. Using the
gravity potential expression explicitly, one can form the following invariant starting from
the point where the information is available (x0 , y0 )
v02 v2
ρ + ρg0 h(x0 , y0 ) + p0 = p0 + ρ + ρg0 h(x, y).
2 2
The external gas pressure can be now eliminated

v 2 = v02 + 2g0 (h(x0 , y0 ) − h(x, y)).

If the slope of the landscape is negative, h(x0 , y0 ) − h(x, y) > 0, then v > v0 . In the
opposite case, there is a maximum height the liquid flow can go up to, defined by v = 0,
v02
h(x, y) = h(x0 , y0 ) + .
2g0
If we, in addition to, assume the flow is uniform across the canal, then we can find the
velocity distribution along the channel as well.

5.2 Kelvin’s circulation theorem

The purpose of this part is to establish a fundamental identity for inviscid fluid flows.
Lemma 5.2.1. Consider some closed contour γ moving with the liquid particles, Fig. 5.4.
Then I I
d dv
vdr = dr
dt γ γ dt

Note, vector dr is in the tangential to the contour curve direction.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 77

Inspired by
Laminar channel flow
Laminar - taking place along constant
streamlines, without turbulence

Gas pressure
h (x 0, y 0), v 0 p= p0 Gravity force

F=−ϱ g 0 i z

z
h (x , y)
y
x
Figure 5.3: Illustration of a steady (laminar) inviscid liquid flow along a canal over uneven
surface of a landscape.

Proof.
I I I I I I I
d d dv dv 1 2 dv
vdr = (vdr) = dr + vdv = dr + dv = dr
dt γ γ dt γ dt γ γ dt 2 γ γ dt
since I
dv 2 = 0
γ

Now, for an incompressible ideal (inviscid) fluid,

dv
ρ = −∇p + F = −∇(p + Φ),
dt
so, applying the Lemma,
I I I I
d dv
ρ vdr = ρ dr = − ∇(p + Φ)dr = − d(p + Φ) = 0
dt γ γ dt γ γ

This implies, I
vdr = const
γ

which is the essence of Kelvin’s circulation theorem. The theorem can be represented in
a different way using the Stokes Theorem of Vector Calculus. The theorem states that

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 78

Theorem 5.2.2 (Stokes Theorem, revision).

Z I
∇ × F · dS = F · dr,
S ∂S

Thus it relates the surface integral of the curl of a vector field over a surface S in Euclidean
three-dimensional space to the line integral of the vector field over the boundary of the
surface S. F is a smooth vector function in S. The curve of the line integral, ∂S, must
have positive orientation.

Then the Kelvin’s theorem takes the form

Theorem 5.2.3 (Kelvin’s circulation theorem, 1869). In an inviscid flow with conserva-
tive volume forces, the circulation around a closed curve (which encloses the same fluid
elements) moving with the fluid remains constant with time
I Z
v · dr = ∇ × v · dS = const
γ Sγ

♦ Definition 5.2.4.
∇×v =ω
ω is called vorticity.
♦ Definition 5.2.5. Liquid flow with zero vorticity is called irrotational.
 Example 5.2.6. Consider the stagnation point flow, which will be discussed in detail
later,
v = (x, −y, 0)

ω =∇×v =0
So, the flow is irrotational.
 Example 5.2.7. Consider the simple shear flow given by

v = (y, 0, 0)

∂vx
ω = ∇ × v = −iz = −iz
∂y
So, the flow is rotational with finite vorticity |ω| = 1.

If initially ω = 0, then, according to the Kelvin’s circulation theorem, ω=0 at any

moment of time, everywhere, since γ is an arbitrary contour.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 79

γ (t )

γ (t = 0)
dr
v
r (ξ + Δξ )
x3

r (ξ ) = ( x1 (ξ ), x2 (ξ ), x3 (ξ ))
x2
x1

Figure 5.4: Illustration of the contour traveling with the liquid. Kelvin’s circulation
theorem. The bottom part of the picture demonstrates the standard parametrisation of a
contour r = (x1 (ξ), x3 (ξ), x3 (ξ)), where ξ is the parameter. The vector dr = r(ξ + ∆ξ) −
r(ξ) is in the tangential direction to the contour in the limit ∆ξ → 0. So that v · dr is
the tangential projection of the velocity field onto the contour.

5.3 Potential Flow: special class of inviscid flows

The result of the Kelvin’s circulation theorem allows us to consider a special but a wide
class of flows, which is called irrotational flows, when ω = 0, or potential flows, since
in this case the flow velocity can be represented as

v = ∇φ.

This is due to the fact that ∇ × v = ∇ × ∇φ ≡ 0 identically.

For potential flows, we have just one scalar function, the potential of the flow field φ(x, t)
instead of three components of velocity. From the incompressibility condition it follows
that φ will satisfy the Laplace equation, indeed

∇ · v = ∇ · ∇φ = ∆φ = 0.

So, potential flows satisfy the Laplace equation,

∆φ = 0

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 80

which is the main equation to find φ and hence the velocity, v = ∇φ, with the standard
(could be different) boundary condition on the stationary solid surface

∂φ
v · n = ∇φ · n = = 0.
∂n

The pressure, on the other hand, can be found via the Bernoulli’s equation, only now,
pay attention, the quantity

ρ(∇φ)2
p+Φ+ = const .
2

is equal to the same constant everywhere. Question, why? Try to explain this.
Hint revisit the derivation of the Bernoulli’s equation and use the fact that ∇ × v = 0.
Or you can just continue reading ...
As it was in the original derivation of the Bernoulli equation, we can represent the Euler’s
equations as
1
ρ(∇ × v) × v + ρ∇(v 2 ) = −∇p + F = −∇(p + Φ).
2
but, in the case of potential flows, ∇ × v ≡ 0 identically, so the result

ρv 2
 
∇ p+Φ+ = 0,
2

is true in general, not just along some stream line.

 Example 5.3.1. Consider a uniform stream going, without loss of generality, along
the x3 -direction, v = (0, 0, v0 ), if v0 is constant, then

∂φ
= v0
∂x3
and
φ = v0 x3 ,
which obviously satisfies the Laplace equation. The pressure scalar field is given by

ρv02
p+Φ+ = const
2
which is constant if Φ = const. The free constant then can be found if we know the
reference pressure at some point.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 81

5.4 Stagnation-point flow

 Example 5.4.1. Consider a steady two-dimensional flow without volume forces with
v = (x, −y) and ρ = 1. The flow field lines are depicted in Fig 5.5 and one can see why
the flow got its name, this is because of the stagnation point at x = y = 0 where the
total velocity is zero, v = 0.
One can think of this flow as a model for a jet impinging on a solid wall at y = 0. Indeed,
the normal vector to the line y = 0 is n = (0, 1). Then the normal velocity at y = 0 is
zero, v · n = vy = 0. This implies that the standard impermeability boundary condition
for inviscid flows at a solid surface given by y = 0 is fulfilled. On the other hand, one can
think of this flow as a model flow in the corner region (x > 0, y > 0), bounded by two solid
walls, the lines y = 0 and x = 0. Indeed, at x = 0 the normal vector pointing into the
liquid is n = (1, 0). Hence we have, as in the previous case, v · n = vx = 0 - a condition,
which is equivalent to the impermeability condition at solid walls.
Let’s analyse the flow. If we substitute the velocity to the Euler equations (without volume
forces), then we find that,
∂p ∂p
= −x, = −y,
∂x ∂y
which immediately results in
1
p(x, y) = C − (x2 + y 2 ).
2

If we set pressure at x = y = 0 to p0 , then

1
p(x, y) = p0 − (x2 + y 2 ).
2

We have seen that vorticity for this flow is zero everywhere, Example 5.2.6, so the flow is
2 2
potential with φ = x −y
2
(show that).
Consider the equivalent of the Bernoulli’s equation in this case,

∇φ2 1 2 2 x2 + y 2
p+ = p0 − (x + y ) + = p0 .
2 2 2

One can see, that the sum of the two components is equal to the same constant p0 , no
matter what stream line we take. So, we have illustrated the fact that in the case
of potential flows, the Bernoulli’s equation is fulfilled everywhere.
Note, since the velocity field was given, we did not need to solve the Euler equations in this
2 2
case. We did that only for illustration. In fact, once the potential is known, φ = x −y 2
,

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 82

Stagnation point flow Liqud jet flow

Ambient gas pressure p= p0= p∞

12
v=( x ,− y , 0)
v=(0,−v ∞ , 0)
Could be
the gravity force
8

Free
Y

surface
4

-10 0 10
X
x=0

Figure 5.5: Stagnation point flow and a liquid jet flow: flow field lines of the stagnation
point flow with a centre-line at x = 0, v = (x, −y, 0) (left), flow field lines of a liquid jet
(right) with a free surface, far field velocity v∞ and pressure p = p∞ = p0 and a centre-line
at x = ξ0 .

one can always use the Bernoulli’s equation, which is a general solution to the Euler
equations, to find pressure p.
∇φ2 x2 + y 2
p = const − = const − .
2 2

 Example 5.4.2. The stagnation point flow can be used to estimate the force exerted
by an impinging jet on a solid surface once the jet velocity v∞ is given. One needs to note
though, the stagnation point flow is only a model, but it is not equivalent to a jet flow, see
illustration in Fig. 5.5.
Calculate the total force acting on the solid substrate (at y = 0) between x = −L and
x = L, L > 0, per unit length in the z-direction, in the previous example of the stagnation
point flow v = (x, −y, 0), 5.4.1. The total force is simply the line integral
 of the pressure
Z L Z L 2
x L3
distribution p(x, y) taken at y = 0, that is p(x, 0)dx = p0 − dx = 2p0 L− .
−L −L 2 3

5.5 Exercises - V
1. Show, that the velocity field

v = (Ax, −Ay, 0),

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 83

satisfies the incompressibility condition, where A is some real constant. Plot the
field lines in the region z = 0, x > 0, y > 0.
2. A liquid jet of an incompressible and inviscid Newtonian fluid of density ρ is emerg-
ing from a circular hole of radius R0 at the bottom of a solid wall tank, Fig. 5.6.
The jet flow is steady. The ambient gas is inviscid and its pressure is constant,
which is equal to p0 . The jet has axisymmetric shape and is directed downwards
along the x3 -axis, which is also the direction of the gravity force F = −ρg0 i3 . One
can assume that the flow is uniform, axisymmetric and is approximately parallel to
the x3 -axis. Note, those assumptions in a way correspond to the so called plug
flow. In the plug flow, the velocity of the fluid is assumed to be constant across
any cross-section perpendicular to the axis of the flow.
Analyse the flow using the Bernouilli’s equation. The total fluid flux Q is given.
Determine the radius of the jet as a function of x3 . Explain, how does the assumption
of the plug flow help in analysing the problem.
Supplementary material 5.5.1. How can we estimate the accuracy of
N
the plug flow assumption a posteriori? Hint: check the incompressibility condition
first. Is anything missing there? Do we have a radial (in terms of the cylindrical
coordinate system) component of velocity? Using the cylindrical coordinate system
(axis of symmetry is x3 ), try to find a correction to both the velocity field and the
missing terms in the incompressibility condition. One can use the fact that in the
cylindrical coordinate system the incompressibility condition is given by
∂v 1 ∂ 1 ∂vθ
q
+ (r vr ) + = 0, r = x21 + x22 .
∂z r ∂r r ∂θ

3. Consider a steady two-dimensional flow v = (v0 (x − x0 ), −v0 y) of an inviscid in-

compressible liquid with density ρ in the upper half-plane y > 0 in the absence of
any external volume forces. Parameter v0 > 0 is a positive constant and x0 is a
constant.
• Is this flow potential, and if it is, find the the velocity potential in this case?
• Find the distribution of pressure using the Bernoulli’s equation for potential
flows (if the flow is potential).
• Show that this flow satisfies the Euler equation in the domain and determine
the pressure field directly from the Euler equations. Compare with the pressure
field found using the Bernoulli’s equation. What is the conclusion here?
• What is the velocity at y = 0? Find the trajectory to a stagnation point, that
is the point where |v| = 0. On this trajectory, it is given that the pressure
p = p0 at y = 1, find the pressure at y=0. What is the pressure field in general?
• Can we assume that the plane y = 0 represents a solid wall boundary with the
standard impermeability condition on it?

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 84

Schematic setup

Inspired by
x3

x2
R0 x1

p= p0

Gravity force
F =−ϱ g 0

R( x 3 )
Flux Q

Figure 5.6: Illustration of the jet flow: liquid jet from a kitchen tap (left) and schematic
set-up of the problem (right).

• Can we assume that the plane x = x0 represents a solid wall boundary with
the standard impermeability condition on it? Can we assume that, if we know
that the condition of impermeability is not fulfilled for some reason?

4. Consider a steady three-dimensional flow of an inviscid, incompressible liquid with

density ρ around a porous body, Fig. 5.7. The body is at rest, while the flow in the
far field, that is far away from the body, is homogeneous and plane parallel in the
x1 -direction. It is also known that there is a field line - stream line - ending in the
perpendicular to the surface of the body direction at a critical point. Basically,
we assume here that there is such a critical point where the velocity field is strictly
perpendicular to the surface. The boundary condition on the porous body surface
is that
v · n = −αp1/2 ,
where n is the unit normal vector to the porous body surface, p is liquid pressure
and α is some positive constant parameter.

Find the velocity magnitude and the pressure at the critical point in the absence of
any external forces.

5. Consider a steady two-dimensional flow with the velocity distribution v given by

v = (3x21 − 3x22 , −6x1 x2 , 0). The liquid is assumed to be Newtonian, incompressible
and inviscid, with density ρ. The external potential Φ ≡ 0.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 85

p= p0
v=v ∞

x2

Far field x1

Figure 5.7: Flow around a porous body with a critical point. The critical point is
marked red.

• Find the stagnation point of the flow;

• Demonstrate that the flow is potential, stating clearly any results and defini-
tions you may need to use, and find the corresponding potential φ(x1 , x2 );
• Find the distribution of pressure p(x1 , x2 ), stating clearly any results you may
need to use, assuming that at x1 = 0, x2 = 0, p = p0 ;
• Find the point where the liquid pressure has a maximum;
• Sketch the flow stream lines in the region x2 > 0 and x2 < 31/2 x1 .

(MA3FM/MA4FM) Autumn 2020

Chapter 6

Two-dimensional potential flows

In the following section, we consider plane parallel two-dimensional potential flows (mostly
to simplify calculations and to get some analytic, observable results), although the theory
can be applied to a three-dimensional case. But, first of all, let’s introduce some important
function, which is widely used in the analysis of two-dimensional flows.

6.1 Stream function

The velocity field of a two-dimensional potential flow, in the (x1 , x2 )-plane, can be repre-
sented through not just one potential function, like v = ∇φ, but also using the form

♦ Definition 6.1.1 (Stream function).

 
∂ψ ∂ψ
v= ,− ,
∂x2 ∂x1

where just introduced function ψ(x1 , x2 ) is called the stream function.

This function has several important properties,

Property 6.1.2.
v · ∇ψ = 0

Proof.    
∂ψ ∂ψ ∂ψ ∂ψ ∂ψ ∂ψ ∂ψ ∂ψ
v · ∇ψ = ,− · , = − =0
∂x2 ∂x1 ∂x1 ∂x2 ∂x2 ∂x1 ∂x1 ∂x2

86
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 87

This implies ψ = const along the field lines. Indeed, it is known from Vector Calculus
that the gradient of a scalar field is perpendicular to the levels of the field, that is to the
lines ψ = const in this case. In the two-dimensional case, the velocity vector v is in the
tangential direction to the levels of ψ. Then the levels of ψ coincide with the field lines of
the velocity field.

Property 6.1.3 (Stream function and the incompressibility condition). The stream func-
tion form  
∂ψ ∂ψ
v= ,− ,
∂x2 ∂x1
satisfies the incompressibility condition identically, that is

Proof.
∂ 2ψ ∂ 2ψ
∇·v = − ≡0
∂x1 ∂x2 ∂x2 ∂x1

Property 6.1.4. The flux between two stream lines is given by the difference of the values
of the stream function taken on those lines.

The third property allows calculating the flux of the liquid between any two stream lines.
Consider two stream lines, Fig 6.1, connected by a positively oriented contour. Let’s
calculate the flux between them. Since the result is independent of the contour taken, we
can take the contour in the direction of the y-axis of a Cartesian coordinate system (more
accurate to say, we will choose the Cartesian coordinate system such that the y-axis will
be in the direction of the straight line of the contour), the normal vector to the contour
is pointed in the direction of the flow.

Z Z   Z   Z
∂ψ ∂ψ ∂ψ ∂ψ
Q= dl v · n = dy nx − ny = dy nx = dy = ψ2 − ψ1 .
∂y ∂x ∂y ∂y

Note: the flux is positive in the direction of n, if the velocity is in the same direction,
and the flux is negative if the velocity is in the opposite direction.

Note: while we introduced the stream function for potential flows, in the definition, we
had not used any assumptions that the flow must be potential. The stream function can
be defined for any incompressible flow in 2-D cases. So that, if it is known, it can be
directly used to determine liquid flux between any stream lines.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 88

\2

n
y \1

Figure 6.1: Stream function and the flux between two stream lines.

6.2 Some basic two-dimensional potential flows

Since the Laplace equation is linear, the superposition principle is applied to its solutions,
that is a linear combination of solutions is a solution. This allows to build complex flows
on the basis of a few elementary flows. Here we consider some of them. Note, we will use
polar coordinate system here. The form of our usual operators is very different in polar
coordinates. The reference material, the listing of those operators is given without proof
after the Chapter.
 Example 6.2.1 (Rectilinear constant flow. Flow I.). The first elementary flow is given
by v = (v1 , v2 , 0), where v1 , v2 are all constants.
The potential of the flow, in a Cartesian coordinate system, is given by
φ(x1 , x2 ) = v1 x1 + v2 x2 .
and the stream function
ψ = v 1 x2 − v 2 x1
The flow stream lines are illustrated in Fig 6.2.

 Example 6.2.2 (Dipole. Flow II.). The second elementary flow is called the dipole
flow.
The potential of the dipole flow of strength M placed at the origin and oriented along the
x1 -axis is given in both polar and the Cartesian coordinate systems by
M cos θ M x1
φ= =
2πr 2π(x21 + x22 )

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 89

x2

x1

Figure 6.2: Rectilinear flow - stream lines

which is one of the solutions of the Laplace equation in the polar coordinate system, see
reference material 6.6.5, and the stream function
M sin θ M x2
ψ=− =− .
2πr 2π(x21 + x22 )

Using operators in the polar coordinate system, one can show that
M cos θ M sin θ
vr = − , vθ = −
2πr2 2πr2
and
ω = 0.
Indeed, the curl in polar coordinates is given by
 
1 ∂ 1 ∂vr
ω = (∇ × v) = (rvθ ) − iz = 0.
r ∂r r ∂θ

Note that the vorticity vector is directed in the perpendicular to the flow plane direction.
So that even though we are considering two-dimensional flows, we still need the third
direction to formally define the cross - product and hence the vorticity.
If we choose, for example, a circular contour of radius R0 with its centre at the origin,
then the circulation is given by a line integral
I I Z 2π
M
vdr = vθ rdθ = − sin θdθ = 0.
2πR0 0
That is the circulation is zero for the dipole flow, as it can be anticipated. The flow stream
lines are illustrated in Fig 6.3.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 90

x2 0

-1

-2

-2 -1 0 1 2

x1

Figure 6.3: Dipole flow - stream lines. M = 2π

 Example 6.2.3 (Vortex. Flow III.). The third elementary flow despite its simple form
is very informative and has unusual properties discussed below. It is called the vortex flow.
Variations of this flow are widely observed in atmospheric and ocean currents and are at
the heart of the phenomenon of turbulence.
The potential of a vortex with its centre or vortex eye located at the origin, r = 0, is
given by, in the polar coordinate system,
Γ
φ= θ
2π
and the stream function is
Γ
ψ=− ln r.
2π
It is not difficult to find components of velocity, which are in the Cartesian coordinate

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 91

system with its centre at the origin of the polar coordinate system,
Γ x2 Γ x1
v1 = − , v2 =
2π x1 + x22
2
2π x1 + x22
2

or using operators in the polar coordinate system, one can show that
Γ
vr = 0, vθ = .
2πr

One can see that the vortex flow is ’rotating’ in anticlockwise direction, if parameter
Γ > 0 and in the clockwise direction if parameter Γ < 0, while the speed of rotation is
proportional to Γ.
If we check, it looks like that the vorticity of the vortex flow is zero everywhere and thus
circulation is expected to be zero as well,

ω = 0.

Indeed, vr = 0, vθ r = const and

 
1 ∂ 1 ∂vr
ω = (∇ × v) = (rvθ ) − iz = 0.
r ∂r r ∂θ

Consider the circulation now, choose, for example, a circular contour around the origin
(one can choose any contour as far as it contains the origin),
I I Z 2π
Γ
vdr = vθ rdθ = dθ = Γ.
2π 0
The circulation, as one can see, is not zero, opposite to what we would expect. This is due
to the fact that the velocity field at the vortex centre (the origin) is not smooth, and so
the assumptions of the theorem (10.6.1) and consequently (5.2.3) are not fulfilled in the
domain where the central point is included. Alternatively, one can think of this effect as
ω = 0 everywhere but not at the origin. The flow stream lines are illustrated in Fig 6.4.

 Example 6.2.4 (Source. Flow IV.). The fourth flow is a simple source of the liquid.
The potential of a source with its centre or eye located at the origin, r = 0, is given by
in the polar coordinate system
φ = A ln r
and the stream function is
ψ = Aθ.

It is not difficult to find components of velocity, which are in the polar coordinate system
A
vr = , vθ = 0.
r
(MA3FM/MA4FM) Autumn 2020
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 92

10

x2 0

-1 0

-1 0 0 10
x1

Figure 6.4: Vortex flow - stream lines. Γ = 2π

So that the liquid is simply moving in the radial direction from the source centre at r = 0.
The flow is potential
 
1 ∂ 1 ∂vr
ω = (∇ × v) = (rvθ ) − iz = 0
r ∂r r ∂θ

with zero circulation I I

vdr = vθ rdθ = 0.

The flow stream lines are illustrated in Fig 6.5.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 93

10

x2 0

-1 0
-1 0 0 10
x1

Figure 6.5: Stream lines of a source. A = 1.

6.3 Combinations or superpositions of elementary flows:

Rankine flow.
We will now use the linearity of the Laplace equation

∆φ = 0

to construct more complex flows using the elementary flow solutions.

The linearity implies that if φ1 and φ2 are both solutions to the Laplace equation, then
their linear combination φ = C1 φ1 +C2 φ2 is a solution as well, where C1,2 are free constants.
To find stream lines of a flow when there is a source of the liquid in a uniform stream
we can superimpose two elementary flows, Flow-I and Flow-IV. We will presume that
the uniform flow is in the direction x1 of the coordinate system. That is in the polar

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 94

coordinate system
φ = C0 φI + C4 φIV = C0 r cos θ + C4 ln r.
The stream function of this flow is of course, due to also linearity of the problem, a
combination of the two elementary stream functions

ψ = C0 r sin θ + C4 θ.

The flow field stream lines are illustrated in Fig. 6.6.

Rankine flow

ψ=r sin(θ)+2.5 θ

10

x2 0

-5

-1 0
-1 0 -5 0 5 10

10

x2 0

-5

-1 0
-1 0 -5 0 5 10

x1

Figure 6.6: Potential (steady) Rankine flow: stream lines at C0 = 1 and C4 = 2.5. The
flow can be interpreted as a liquid flow from a source immersed into a uniform stream
(top) or as a flow around a body of the shape ψ = const (bottom).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 95

6.4 Combinations or superpositions of elementary flows:

potential (steady) flow past a cylinder
Consider now two-dimensional potential steady flow past a cylinder of radius R0 with its
centre placed at the origin of the coordinate system. Assume that the undisturbed flow
is the rectilinear flow in the x1 direction. The problem is set as follows, find a harmonic
function φ(x1 , x2 ), that is a solution to

∆φ = 0

subject to the following two conditions:

at the solid surface of the cylinder

∂φ ∂φ
= =0
∂n ∂r r=R0

and in the far field, away from the cylinder r → ∞,

φ = φ∞ = v∞ x1 = v∞ r cos θ.

Potential flow: experiment

Potential flow: theory
low velocity
2

x2 0

-2
-2 0 2

x2

Flow past a sphere:

high speed experiment

Figure 6.7: Potential (steady) flow past a cylinder, theory and experiment: stream lines.
Bottom: Flow past a sphere, high speed experiment, where the flow is separated from the
surface.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 96

Since φ → ∞ as r → ∞, this Neumann boundary value problem for the Laplace equation
does not have a unique solution, see Reference material 6.6.4 ”Some notes on the Laplace
equation”.
Instead of solving the Laplace equation in this case, we will find a solution using the
linearity of the problem, by superimposing basic potential functions already found. The
purpose, though, is to satisfy the boundary condition and the condition at infinity.
Consider a combination of φ∞ (Flow-I) and the dipole flow (Flow-II), which is

M cos θ
φpast = φ∞ +
2πr
We know that φpast (as a linear combination of solutions) is a solution to the Laplace
equation, also
φpast → φ∞ , r → ∞,
as desired. So, we have to satisfy now the condition at the cylinder,

∂φpast
=0
∂r r=R0

that is
M cos θ
v∞ cos θ − =0
2πR02
and one can find M = 2πv∞ R02 .
The velocity of this combined flow is

R02 R02
   
vr = v∞ cos θ 1 − 2 , vθ = −v∞ sin θ 1 + 2
r r

and the stream function is

R02
 
ψ = v∞ sin θ r − ,
r
see Fig 6.7 for illustration. The flow structure obtained analytically is very close to that
obtained in a specially designed experiment, see Fig. 6.7.

6.5 The force in potential flows: d’Alembert paradox

We are now in the position to calculate the pressure at the cylinder surface and thus the
total force. As always for potential flows, we use the Bernoulli’s equation
2
ρv∞ ρv 2
p∞ + =p+
2 2

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 97

at the surface r = R0 assuming that in the far field p = p∞ . The velocity components,
we have found, at r = R0 are

vr = 0, vθ = −2v∞ sin θ.

So, the pressure distribution on the solid surface is given by

2
ρv∞ ρv 2
p(r = R0 , θ) = p∞ + − θ = (6.1)
2 2
2
ρv∞ 2
= p∞ + − 2ρv∞ sin2 θ.
2
The total force F acting on the cylinder surface per unit length in the perpendicular to
the flow plane direction is equivalent in our two-dimensional case to the line integral over
the circular boundary I Z 2π
Fi = −pni dl = −pni R0 dθ,
l 0

where n = (n1 , n2 ) = (cos θ, sin θ) in the Cartesian coordinate system. Note, the contour
in the line integral has been parametrized by the polar angle θ.
Now, it is not difficult to see that F = 0, since all the integrals contributing into the line
integral are identically zero. Indeed,
Z 2π Z 2π Z 2π Z 2π
3
sin θdθ = 0, cos θdθ = 0, sin θdθ = 0, sin2 θ cos θdθ = 0.
0 0 0 0

This formal result is not surprising, since pressure (6.1) is symmetric (it is insensitive to
permutations θ → π − θ and θ → −θ), and we would not expect any net force anyway.
So, the net force imposed on the cylinder by the background flow is zero, which is in con-
tradiction to common sense and to experimental observations. This result is the essence
of the d’Alembert paradox, which is true in general, not just for a cylinder or a sphere
in the three-dimensional case. The origin of this paradox is in the thin boundary layer
adjacent to the solid wall where the effects of viscosity are important. Remember, that we
have neglected viscosity everywhere and at the boundary of the solid sphere in particular.
As a result, we had to drop the no-slip boundary condition off the model, since Euler
equations are of the lesser order than the original Navier-Stokes equations. In this case,
the tangential to the solid surface component of velocity is discontinuous, since we have
found that vθ 6= 0, but from the no-slip boundary condition it should be zero vθ = 0.
The viscous processes at the boundary result in flow separation and generation of vorticity,
which is assumed to be zero in potential flows ω = 0. So, the actual flow may be (though
may be not, see Figs. 6.7, 6.8, 6.9) quite different from the potential flow, we have
just found. The boundary effects are clearly seen in the picture of some experimental
observation of a flow past a cylinder, see Figs. 6.7, 6.8, 6.9. As a result of the separation

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 98

Potential and separated flows: experiment

Figure 6.8: Flow past bodies - potential and separated. Thanks to G.M. Homsy et al.
Multimedia Fluid Mechanics, (second edition). Animations and experiments.

Potential and separated flows: experiment

Figure 6.9: Flow past bodies - potential and separated. Thanks to G.M. Homsy et al.
Multimedia Fluid Mechanics, (second edition). Animations and experiments.

of flow from the solid wall, there is a stagnation zone in the wake and a very chaotic or
turbulent, as we used to say, tail. Apparently, the pressure will be asymmetric in a real

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 99

case scenario and will produce some total force of resistance.

The d’Alembert paradox however does not imply that potential flow approximation is
completely useless. In fact, it can be used to good effect in many areas of fluid dynamics,
though with some care. For example, potential theory reproduces very well flow structure
in front of objects and in the case of thin body profiles, like some aircraft wings.

6.6 Irrotational flow with circulation - origin of lift,

the Magnus effect or playing golf and tennis
In the previous example, the flow had no circulation. Now, we will augment the flow past
a cylinder (a combination of Flow I and Flow II) with the vortex flow rotating in the
anticlockwise direction, the Flow III, with the vortex eye at the centre of the cylinder,
which has non-zero circulation. Such an addition is supposed to mimic a rotating, in the
anticlockwise direction, cylinder. So, what would happen? What shall we observe in this
case? See Figure 6.10. Note, the augmented flow will not still satisfy the no-slip boundary
condition exactly, but may correct the problem of the rotating bodies.

6.6.1 Qualitative analysis

As one can see from the stream line picture, Figure 6.10, the velocity under the cylinder
is larger than the velocity of the ambient flow vunder > v∞ , on the other hand, above
the cylinder, vup < v∞ . If we now apply the Bernoulli’s equation to this potential flow
assuming again that the pressure in the far field is p = p∞ ,
2 2
ρv∞ ρv 2 ρvup
p∞ + = punder + under = pup +
2 2 2
so pup > punder and there is a total force as is shown in the figure. The phenomenon is
known as the Magnus force.

6.6.2 Quantitative analysis

Consider now a combination of φ∞ (Flow I), the dipole flow (Flow II) and a vortex (Flow
III) with arbitrary circulation Γ. The combined potential has a form

M cos θ Γ
φpast = φ∞ + + θ.
2πr 2π
We know that the vortex velocity field has only vθ component, so that the addition of
the third flow will not affect the implementability of the boundary condition, and we can

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 100

use results obtained without the vortex field. That is, to satisfy the boundary condition
we set M = 2πv∞ R02 , where as before we assumed that the cylinder radius is R0 . The
combined potential of the velocity field is then given by

v∞ R02 cos θ Γ
φpast = v∞ r cos θ + + θ.
r 2π
One can also note that as r → ∞, the potential φpast → φ∞ = v∞ r cos θ.
The velocity of this combined flow is

R02 R2
   
Γ
vr = v∞ cos θ 1 − 2 , vθ = −v∞ sin θ 1 + 20 +
r r 2πr
and the stream function, due to the linearity of the problem, is

R02
 
Γ
ψ = v∞ sin θ r − − ln r,
r 2π
see Fig 6.10 for illustration.
The pressure distribution on the solid surface, as in the previous example without cir-
culation, is given by applying the Bernoulli’s equation with velocity components on the
solid surface at r = R0
Γ
vr = 0, vθ = −2v∞ sin θ + .
2πR0

That is
2
ρv∞ ρv 2
− θ =
p(R0 , θ) = p∞ + (6.2)
2 2
2
 2
ρv ρ Γ Γ
= p∞ + ∞ − 2ρv∞
2
sin2 θ − + ρv∞ sin θ .
2 2 2πR0 πR0

The total force F acting on the cylinder surface per unit length in the perpendicular to
the flow plane direction is as before equivalent in our two-dimensional case to the line
integral over the circular boundary
I Z 2π
Fi = −pni dl = −pni R0 dθ,
l 0

where n = (n1 , n2 ) = (cos θ, sin θ) in the Cartesian coordinate system.

We have already seen that contribution of integrals containing terms sin θ, cos θ, sin3 θ
or sin2 θ cos θ will be zero in the total force. The remaining part of the pressure pE =
Γ
ρv∞ sin θ , p(R0 , θ) = pE + pO , will produce non-zero contribution, where pO is the
πR0
part giving zero contribution.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 101

The total force then is calculated via

Z 2π
Fi = − pE ni R0 dθ.
0

The component F1 in the direction of the background flow (the x1 -direction)

Γ 2π
Z
F1 = −ρv∞ sin θ cos θdθ = 0.
π 0
So that there is still no force in the direction of the flow. Another component F2 in the
perpendicular to the background flow direction
Γ 2π 2
Z
F2 = −ρv∞ sin θdθ = −ρv∞ Γ.
π 0

Quantitatively, the lift force is proportional to the medium density ρ, circulation Γ and
the flow velocity in the background v∞ , F ∼ ρΓv∞ . In our particular case, Fig. 6.10,
F1 = 0, F2 = −ρΓv∞ . Parameters Γ > 0, v∞ > 0 and hence F2 < 0. Changing the
circulation sign Γ < 0, we can obtain the lift force, F2 > 0.

Potential flow with circulation

around a cylinder: Magnus effect.

v∞ 3

x2 0

-1

-2

-3
-3 -2 -1 0 1 2 3

Direction of the x1
background flow

Figure 6.10: Potential flow past a rotating cylinder and the Magnus effect- stream lines.

The effect has always been ascribed to H. Magnus (1852), but it has been well described by
the British engineer B. Robins (1742). The Magnus effect is commonly observed in sport
when a spinning ball exhibits a ”mysterious” curvilinear motion. It is very pronounced
in table tennis, slightly less in golf, tennis and football and almost invisible in cricket.
In general the lift force is described by Joukowski theorem - 1904-1906

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 102

Theorem 6.6.1. The theorem deals with two-dimensional flows around a body of an
arbitrary cross-section (contour C) and determines the lift generated per unit of the span
length. The direction of the lift force F is perpendicular to the direction of u∞ velocity in
the far field (upstream) and is proportional to

F = −ρu∞ Γ
H
where Γ = C vdl is the circulation (contour integral) around the two-dimensional contour
of the body (aircraft wing cross-section), where ρ and u∞ are density and velocity of the
liquid far upstream. The value of circulation Γ is determined by a postulate formulated
independently by Martin Kutta and Sergey Chaplygin. But those details are beyond the
scope of the current introductory course, let me only note that the postulate works well for
situations like that shown in Fig. 6.11.

So, mathematically,
H we can think of the origin of the lift force as due to non-zero circulation
Γ = C vdl 6= 0 around the boundary of a solid body (aircraft wing for example) in
potential flows. The smaller the region where the flow deviates from a potential one, the
better the theorem (6.6.1) works. If the flow is not potential on the other hand, we have
to appeal to the full system of the Navier-Stokes equations.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 103

Flow past an airfoil (a thin wing)

Figure 6.11: Flow past a thin wing - experiment.

6.6.3 Reference material. Differential operators in polar and

cylindrical coordinate system.
Basis vectors

Velocity in a polar or a cylindrical coordinate system, as any vector field, can be repre-
sented using basis unit vectors, ir , iθ , iz . In a polar (2-D) coordinate system
v = ir vr + iθ vθ .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 104

In a cylindrical (3-D) coordinate system

v = ir vr + iθ vθ + iz vz .

That is, we usually write v = (vr , vθ ) or v = (vr , vθ , vz ).

The stream function representation in the polar coordinate system is then
 
1 ∂ψ ∂ψ
v= ,− .
r ∂θ ∂r

The del (or nabla ∇) operator (to calculate the gradient), divergence and the
curl

In a polar (2-D) coordinate system

∂ 1 ∂
∇ = ir + iθ .
∂r r ∂θ

In a cylindrical (3-D) coordinate system

∂ 1 ∂ ∂
∇ = ir + iθ + iz .
∂r r ∂θ ∂z

The divergence in 2-D polar coordinate system

1 ∂ 1 ∂vθ
∇·v = (rvr ) + .
r ∂r r ∂θ

The divergence in 3-D cylindrical coordinate system

1 ∂ 1 ∂vθ ∂vz
∇·v = (rvr ) + + .
r ∂r r ∂θ ∂z

The curl
1 ∂ 1 ∂vr
(∇ × v)z = (rvθ ) − .
r ∂r r ∂θ

Two and three-dimensional Laplace equation in the polar and cylindrical co-
ordinate systems

1 ∂ 2φ
 
1 ∂ ∂φ
∆φ = r + = 0. (6.3)
r ∂r ∂r r2 ∂θ2

1 ∂ 2φ ∂ 2φ
 
1 ∂ ∂φ
∆φ = r + 2 2 + 2 = 0. (6.4)
r ∂r ∂r r ∂θ ∂z

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 105

6.6.4 Reference material. Some notes on the Laplace equation

In the theory of partial differential equations, solutions to the Laplace equation are called
potential or harmonic functions, and that branch of the theory of partial differential
equations is called potential theory.
It is known that there are two basic boundary value problems for the Laplace equation
(there are more, but we focus on the cases relevant to our problems),
the Dirichlet boundary value problem when one needs to find a harmonic function in some
bounded or unbounded domain Ω with boundary ∂Ω such that the function takes given
values at the boundary.
the Neumann boundary value problem when one needs to find a harmonic function in some
bounded or unbounded domain Ω with boundary ∂Ω such that the normal derivative of
the harmonic function takes given values at the boundary.
It is known, that
Theorem 6.6.2 (Uniqueness of the solution of the Dirichlet problem). If the Dirichlet
problem for a bounded domain has a solution, then that solution is unique.
If the Dirichlet problem for an unbounded domain has a solution, which is bounded at
infinity in a two-dimensional case, then that solution is unique.
Theorem 6.6.3 (Uniqueness of the solution of the Neumann problem). If the Neumann
problem for a bounded domain has a solution, then that solution is unique to within a
constant.
If the Neumann problem for an unbounded domain has a solution, which is bounded at
infinity in a two-dimensional case, then that solution is unique.

In Fluid Mechanics, we are mostly interested in the Neumann boundary value problem
for an unbounded domain with the boundary condition at a solid wall, v · n = 0,
∂φ
= 0.
∂n

6.6.5 Some particular periodic solutions of the Laplace equation

in a polar coordinate system by separation of variables.
In this part, it is demonstrated how to obtain some periodic solutions to the Laplace
equation. The elementary solutions used in our course are particular examples of such
kind of solutions obtained via separation of variables technique.
Laplace equation in a polar coordinate system can be represented by
1 ∂ 2φ
 
1 ∂ ∂φ
∆φ = r + 2 2 = 0.
r ∂r ∂r r ∂θ

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 106

We are looking for periodic solutions, such that φ(r, θ) = φ(r, θ + 2π). By separation of
variables φ(r, θ) = H(r)P (θ) and one has

1 d2 P
 
r d dH
− 2
= r = m2 .
P dθ H dr dr

From the above,

P (θ) = A cos mθ + B sin mθ,
d2 H dH
r2 + r − m2 H = 0,
dr2 dr
and from periodicity m = ±1, ±2, ....
We will seek for a solution for H(r) in the form H(r) = rλ and we have

λ(λ − 1)rλ + λrλ − m2 rλ = 0.

That is
λ = ±m.
Here, we have two different characteristic values of λ and hence two different linearly
independent solutions. As a result, from the above a general solution can be written as

φ = rm (Am cos mθ + Bm sin mθ) + r−m (Cm cos mθ + Dm sin mθ).

If m = 0, this is a special case

P (θ) = C1 θ + C2 , H(r) = C3 ln r + C4 , φ = A ln r + B + (C ln r + D)θ.

The periodicity condition is only required for the velocity components, as actual physical
quantities, that is for v = ∇φ or
∂φ 1 ∂φ
vr = , vθ = .
∂r r ∂θ
It follows then that C = 0.

6.7 Exercises - VI
1. Consider a steady two-dimensional flow v = (v0 (x − 1), −v0 y), v0 is a positive
constant, v0 > 0, in the upper (x, y) half-plane (y ≥ 0) of a Newtonian, inviscid and
incompressible liquid with density ρ in the absence of any external volume forces.

• Determine the stream function of the two-dimensional flow.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 107

• Hence or otherwise determine the flux between the stream lines starting at
x = 2, y = 1 and x = 4, y = 1.

2. Consider a steady two-dimensional flow v = (y(1 − y), 0) in the band (0 ≤ y ≤ 1)

of a Newtonian, incompressible liquid with density ρ in the absence of any external
volume forces.

• Is this flow potential? If it is, determine its potential.

• Determine the stream function of the flow.
• Find the total flux in the band.
• Can we use the Bernoulli’s equation to determine the pressure in this case?

3. Consider a steady potential flow described by a potential ϕ(x, y) in a two-dimensional

case in (x, y) plane, which is a linear combination of several, say elementary, steady
potential flows described by potential functions φ1 , φ2 and φ3 , that is ϕ = A1 φ1 +
A2 φ2 + A3 φ3 , where Ak are constants. Demonstrate that in this case the stream
function of the combined flow ψ is also a linear combination of the stream functions
of the elementary flows ψ1 , ψ2 and ψ3 .

4. Consider a combination of a rectilinear potential flow and a source in a two-dimensional

case, which is in a polar coordinate system given by the potential

φ = v0 r cos θ + A ln r,

where v0 > 0 is a real positive constant, A > 0 is a positive constant (the source
strength), r is the radius and 0 ≤ θ < 2π is the polar angle.

• Define components of velocity in the polar coordinates. You may, if necessary,

use the result, without a proof, that in the polar coordinate system with the
radius r and the polar angle θ the gradient of a scalar quantity f has the
following form
∂f 1 ∂f
∇ f = ir + iθ .
∂r r ∂θ
• Define the stream function of the combined flow. You may, if necessary, use the
result, without a proof, that the stream function representation in the polar
coordinate system  
1 ∂ψ ∂ψ
v = (vr , vθ ) = ,− .
r ∂θ ∂r
• Find the stagnation point of the combined flow, that is the point where the
velocity magnitude is exactly zero.
• Find the shape of the solid body which can be placed in the flow without
its distortion with a stagnation point on its surface, if it is known that the
stagnation point is at θs = π and rs = v0−1 .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 108

5. Consider a two-dimensional potential flow with the stream function ψ = rγ sin(γθ)

written in a polar coordinate system. Here, γ ≥ 2 is a constant, r is the radius and
θ is the polar angle.

• Define velocity components in the polar coordinate system. You may, if neces-
sary, use the result, without a proof, that the stream function representation
in the polar coordinate system
 
1 ∂ψ ∂ψ
v = (vr , vθ ) = ,− .
r ∂θ ∂r

• Find the stagnation point of the flow, that is the point where the velocity
magnitude is exactly zero.
• Determine the flow potential in the polar coordinate system. You may, if
necessary, use the result, without a proof, that in the polar coordinate system
with the radius r and the polar angle θ the gradient of a scalar quantity f has
the following form
∂f 1 ∂f
∇ f = ir + iθ .
∂r r ∂θ
6. Observations of a two-dimensional flow along the circle of radius R = 1 with its
centre at the origin of a polar coordinate system has resulted in the following dis-
tributions of velocity on the circle

ur = cos(2θ), uθ = − sin(2θ),

where θ is the polar angle in the polar coordinate system.

It is given that the flow domain has neither internal boundaries nor sources nor
sinks of the liquid. Could this flow be incompressible? Is this information sufficient
to claim that the flow is that of an incompressible fluid? You may, if necessary, use
the fact that for a domain Ω, with the boundary δΩ with a normal vector n we
have, if p is a continuously differentiable vector field, that
I I
∇ · p dΩ = p · n dS,
Ω δΩ

where dS is an infinitesimal element of the boundary.

7. Consider a two-dimensional potential flow given by its potential in a polar coordinate

system
φ = ln r + θ,
where r is the radius and θ is the polar angle.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 109

• Determine the stream function of the flow and its circulation along the unit
circle with its centre at the origin of the coordinate system. Sketch the stream
lines. You may, if necessary, use the result, without a proof, that the stream
function representation in the polar coordinate system
 
1 ∂ψ ∂ψ
v = (vr , vθ ) = ,−
r ∂θ ∂r

and that in the polar coordinate system with the radius r and the polar angle
θ the gradient of a scalar quantity f has the following form
∂f 1 ∂f
∇ f = ir + iθ .
∂r r ∂θ
• Can we identify the elementary flows involved in this case?

8. Consider an aircraft landing via a trajectory where the air density is changing with
the altitude of the aircraft position as ρ = ρ0 exp(−H/H0 ). Assume that the landing
takes place at constant vertical lift force acting on the aircraft, while the horizontal
velocity is set to change as v = v0 +(v1 −v0 ) HH0 , where v0 is an optimal landing speed
and v1 > v0 . Determine, using Joukowski theorem as a guide, how the circulation
around the aircraft wing Γa should be controlled by the pilot in this case.

(MA3FM/MA4FM) Autumn 2020

Chapter 7

Viscous Fluid

7.1 Scaling analysis and the viscous boundary layer

Consider the Navier-Stokes equations without volume forces, for simplicity,
∂vk
= 0,
∂xk

∂vi ∂vi ∂p ∂ 2 vi (7.1)
ρ + vl =− +µ 2.
∂t ∂xl ∂xi ∂xj

In the previous sections, we have studied fluid motion in the approximation of zero vis-
cosity, µ = 0. In reality, all liquid-like substances, even air, have viscosity. The question
is, how good could that approximation be and what have we actually neglected? What
does it mean that the viscosity is small? To answer these questions, we consider the so
called scaling analysis of the hydrodynamic equations.
The idea of the scaling analysis in this case is to bring the system of the governing equa-
tions into a non-dimensional form and to find the so-called non-dimensional parameters
of the system of equations defining its solutions.
It is well known, that all physical quantities are measured in units, for example length
is measured in metres or in inches or in centimetres, volume is measured in cubic metres
or in litres or in cubic centimetres. The value of the same length, let’s say a distance,
depends on the particular unit used to measure. For example, a distance L of 10 metres
would be 1000 centimetres. On the other hand, if we consider two length scales L1 and
L2 , say 10 metres and 1 metre, their ratio LL21 = 10 is non-dimensional and hence does not
depend on the particular unit used. Of course, before we calculate the ratio, we have to
bring all measurements into some standard system of units.
The dimension of a physical quantity is a combination of the base physical units (usually
mass, length, time, electric current and temperature), all other units are derived from

110
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 111

the base ones. In the standard metric system, known as SI (System International), the
main base units are metre m (length), second s (time), kilogram kg (mass) and kelvin K
(temperature). The dimension of a physical quantity is usually designated by brackets.
For example, dimension of a coordinate [x] = m, dimension of temperature [T ] = K.
There are many ways to designate dimensions, but in this course we simply designate
dimensions by SI units, as we just did.
In practice, one can also use derived units to characterize the problem. For example,
the derived unit of force in SI system of units is newton N , which is in terms of basic
units [N ] = kg m s−2 . How do we know that the dimension of the force is that we
just mentioned? This follows from the physical law F orce = M ass × Acceleration.
Acceleration has the dimension [Acceleration] = m s−2 , so the result follows. Similar,
let’s derive the SI unit of pressure. By definition, pressure is the force per unit area,
that is P ressure = F orce/Area. Then [P ressure] = [F orce]/[Area]. That is taking into
account [Area] = m2 , [P ressure] = kg m−1 s−2 .
How does the scaling analysis work?

 Example 7.1.1. Consider, for example, Newton’s law of gravity, which states that
the gravity force F acting on and between the bodies of masses M1 and M2 separated by
distance R is
M1 M2
F =G ,
R2
where G is the gravitational constant with [G] = m3 kg −1 s−2 . To bring this equation into
non-dimensional form, we will need the so-called characteristic values of mass M0 and
distance R0 to ’normalize’ M1 , M2 and R and to bring them into non-dimensional form.
Then,
GM02 M̂1 M̂2
F =
R02 R̂2
h 2i
GM
where R̂ = R/R0 , M̂1,2 = M1,2 /M0 . The dimension of R20 = m kg s−2 , which is the
0
dimension of a force measured in SI in newtons (N ), [N ] = kg m s−2 . If we now normalize
GM 2
the force F̂ = F/F0 , where F0 = R20 , we simply obtain
0

M̂1 M̂2
F̂ = ,
R̂2
which is a non-dimensional version of the Newton’s gravity law. One may notice, that the
physical law we have just brought into non-dimensional form had consistent dimensions
in the LHS and the RHS. This is no coincidence, but the rule.

 Example 7.1.2. Consider equation of motion where x is some distance,

dx
= Ax2 .
dt

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 112

First of all, one can see that A has dimension of [A] = m−1 s−1 . Indeed, both sides of the
equation should have the same dimension.
 
dx
= m s−1 = [Ax2 ] = [A]m2 .
dt
Now, to bring the equation into non-dimensional form let’s introduce characteristic length
scale of the problem L and characteristic time t0 . Then by the same procedure
dx̂
= At0 Lx̂2 .
dt̂
The combination At0 L is non-dimensional, [At0 L] = m−1 s−1 s m = 1, and it forms a
non-dimensional parameter Π0 = At0 L, so that
dx̂
= Π0 x̂2 .
dt̂
The procedure of normalization is flexible and, depending on the problem, can result in dif-
ferent forms of non-dimensional equations. For example, in this case if we would presume
that t0 = A−1 L−1 , the final equation would not have any parameters at all.

7.2 Non-dimensional Navier-Stokes equations, Reynolds

number and Reynolds number limits.
In the following, we will try to bring the system of the Navier-Stokes equations into a
non-dimensional form. We assume that it is possible to introduce only one characteristic
length sale L and only one velocity scale U , the same in all directions. We now perform
the scaling procedure with (7.1). First, let’s introduce x˜k = xk /L, v˜k = vk /U . Then the
time scale can be defined using U and L. That is t̃ = t/t0 , t0 = L/U . Note, this is the
assumption that there is no another time scale involved. The most subtle point here is to
introduce the characteristic pressure p̃ = p/p0 . There are two ways to do that.
One way is to set p0 = ρU 2 , where it is not difficult to see that [p0 ] = kg m−1 s−2 , which
is the dimension of pressure, that is the force per unit area. The introduction of this
∂vi ∂p
parameter p0 can be interpreted as ρvl ∼ .
∂xl ∂xi
∂ 2 vi ∂p
The second way is to set p0 = µU/L, according to µ 2
∼ . One can say that
∂xj ∂xi
in the first case, the pressure is set according to the characteristic pressure due to the
liquid motion with velocity U and in the second case the pressure is set according to the
characteristic pressure due to the viscous forces.

Case - I.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 113

∂ ṽk
= 0,
∂ x̃k
∂ ṽi ∂ ṽi ∂ p̃ 1 ∂ 2 ṽi (7.2)
+ ṽl =− + .
∂ t̃ ∂ x̃l ∂ x̃i Re ∂ x̃2j

Case - II.

∂ ṽk
= 0,
∂ x̃k
∂ ṽi ∂ ṽi 1 ∂ p̃ 1 ∂ 2 ṽi (7.3)
+ ṽl =− + .
∂ t̃ ∂ x̃l Re ∂ x̃i Re ∂ x̃2j

ρU L
Where in both cases we have introduced a non-dimensional parameter Re = µ
, which
is well-known as the Reynolds number in fluid mechanics.

Now, in the first case, we take the limit Re  1. Neglecting all terms having
pre-factor Re−1  1, the system (7.2) becomes
∂ ṽk
= 0,
∂ x̃k
∂ ṽi ∂ ṽi ∂ p̃ (7.4)
+ ṽl =− .
∂ t̃ ∂ x̃l ∂ x̃i
This is in fact the Euler equations for inviscid liquid flows.

It is now obvious, that we have assumed that Re → ∞ to obtain the Euler equations for
inviscid liquid flows. So, actually, it is either µ → 0 (the limit we have used to obtain the
Euler equations) or U → ∞ or L → ∞. For example if we consider water wave motion
with characteristic length scale L = 10 m and velocity U = 10 m/s, then taking water
viscosity at µ = 10−3 N · s/m2 and density at ρ = 103 kg/m3 , one obtains Re = 108  1.
This is an approximation, remember.

In the second case, we consider an opposite limit of Re  1. In this limit, neglecting

all terms having pre-factor Re  1, the system (7.3) becomes
∂ ṽk
= 0,
∂ x̃k
∂ p̃ ∂ 2 ṽi (7.5)
0=− + .
∂ x̃i ∂ x̃2j

In the limit Re  1, we have obtained the so-called system of Stokes equations. Here,
since Re  1, we have either µ → ∞ or U → 0 or L → 0. That is the obtained simplified

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 114

system of equations can be used to study flows in confined geometries, in small gaps in
solids or sand/soil media, at low velocities or in some cases at large values of viscosity.
This scenario usually occurs in the case of flows in porous materials. But remember that
the measure of that smallness is Re  1. For example if we consider a water flow inside a
porous medium with the characteristic gap scale L = 10−6 m and velocity U = 10−3 m/s,
then taking water viscosity at µ = 10−3 N · s/m2 and density at ρ = 103 kg/m3 , one gets
Re = 10−3 << 1.

7.2.1 How do we know that we have taken the right limit?

The question to ask may be how do we know that we have taken the right limit or chosen
the right length scale. Consider an example where things can go wrong.
In the first case scenario we have obtained the following non-dimensional equations:

Case - I.

∂ ṽk
= 0,
∂ x̃k
∂ ṽi ∂ ṽi ∂ p̃ 1 ∂ 2 ṽi (7.6)
+ ṽl =− + .
∂ t̃ ∂ x̃l ∂ x̃i Re ∂ x̃2j

Now, we would like to take another limit. Instead of requiring Re  1, we will take
Re  1 by multiplying the above equations through by Re. We are about to neglect then
the following terms: the whole LHS and also the term
∂ p̃
−Re .
∂ x̃i

Basically, we anticipate that

∂ p̃ ∂ 2 ṽi
Re  .
∂ x̃i ∂ x̃2j
The full, real solution is of course not aware of our manipulations with parameters. We
may find, if we measure, that real distribution of pressure at Re  1 is proportional to
µU
L
, where U and L are some values of velocity and the length scale within the range of
our real measurements of velocity and length scales of the flow structure. But, in the
scaling procedure, we have presumed that the pressure is to be normalized by p0 = ρU 2 .
Now let’s make some loose statements replacing differential operators by the values of
measurements. That is
∂ p̃ p̃ ρU L p ρU L µU
Re ∼ Re ∼ ∼ = 1,
∂ x̃i x̃i µ p0 µ ρU 2 L

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 115

∂ 2 ṽi ṽi U L2
∼ ∼ = 1.
∂ x̃2j x̃2j U L2
So, real measurements will prove us wrong, if we neglect the term
∂ p̃
Re .
∂ x̃i

The reason for that is that our assumption that the pressure scales at ’low’ velocities
as p0 = ρU 2 is incorrect. On the other hand, at ’large’ velocities, the pressure does
scale as p0 = ρU 2 . So basically at Re  1, pressure scales as p0 = ρU 2 and at Re  1,
pressure scales as p0 = µU/L. So when we take a scaling, we are bound to take the proper
limit. How do we know which one? Only from physics. Some physicists do the following
line of loose arguments or check-ups. Instead of writing the system of the Navier-Stokes
equations, they write an order of magnitude equivalent form

ρU 2 L−1 = pL−1 + µU L−2

or
ρU 2 = p + µU L−1 .
The last equation is interpreted as pressure can be either as the first term or as the third
term. If it is like the first term, then ρU 2  µU L−1 , that is Re  1. If it is like the third
term, then ρU 2  µU L−1 , that is Re  1.
Then scientists consider particular numbers coming out of experiments. For example,
take ocean waves: length 10 m, velocity 10 m/s, viscosity at µ = 10−3 N · s/m2 and density
at ρ = 103 kg/m3 . Then, ρU 2 = 105 kg m−1 s−2 , but µU L−1 = 10−3 kg m−1 s−2 . Obviously,
the pressure due to the first term is much larger. Now take the flow in sand, where length
10−5 m, velocity 10−3 m/s, then similar calculation would give ρU 2 = 10−3 kg m−1 s−2 , but
µU L−1 = 10−1 kg m−1 s−2 . So, in this case the pressure is more like the third term. Of
course, we have got plenty of room in between, where neither Re  1 nor Re  1.

7.2.2 Viscous boundary layer

Now, to understand what is going on at a solid boundary, we consider a two-dimensional

problem of a flow at a flat solid wall at x2 = 0, Fig 7.1. We assume that the length and
velocity scales in the x1 direction, L and U , are such that Re  1. On the other hand, in
the x2 direction, the length and velocity scales are assumed to be different now, though
proportionally, that is δL and δU . As a result, we have the following non-dimensional
variables x˜1 = x1 /L, v˜1 = v1 /U , x˜2 = x2 /δL and v˜2 = v2 /δU . The time scale and the
pressure scale are defined as in the first case, that is t̃ = t/t0 , t0 = L/U , p̃ = p/p0 ,
p0 = ρU 2 .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 116

x2
x1

√
Figure 7.1: Viscous boundary layer. δ = 1/ Re

Then the incompressibility condition is in the same form as before

∂ ṽk
=0 (7.7)
∂ x̃k

and
∂ ṽ1 ∂ ṽ1 ∂ p̃ 1 ∂ 2 ṽ1 1 ∂ 2 ṽ1
+ ṽl =− + + , (7.8)
∂ t̃ ∂ x̃l ∂ x̃1 Re ∂ x̃21 δ 2 Re ∂ x̃22

∂ ṽ2 ∂ ṽ2 1 ∂ p̃ 1 ∂ 2 ṽ2 1 ∂ 2 ṽ2

+ ṽl =− 2 + + . (7.9)
∂ t̃ ∂ x̃l δ ∂ x̃2 Re ∂ x̃21 δ 2 Re ∂ x̃22

Now, and it is important, we assume that δ 2 Re ∼ 1, while Re  1. Then neglecting

terms proportional to Re−1 but keeping terms ∼ δ −2 Re−1 in (7.8), keeping only terms
proportional to δ −2 in (7.9) (since Re  1 and δ 2 Re ∼ 1, one has δ −2  1)

∂ ṽk
=0
∂ x̃k

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 117

and
∂ ṽ1 ∂ ṽ1 ∂ p̃ 1 ∂ 2 ṽ1
+ ṽl =− + 2 , (7.10)
∂ t̃ ∂ x̃l ∂ x̃1 δ Re ∂ x̃22
∂ p̃
0=− .
∂ x̃2
˜ 1)
The last equations implies p̃ = p(x
The above system of equations in the viscous layer at a solid wall is called the Prandtl
model - 1904.
We have also found that the non-dimensional factor defining the thickness of the viscous
√
boundary layer, where viscous effects are important, is inversely proportional to Re

1
δ=√ .
Re

This is the main result of this section.

Some estimates and numbers for water flows. Let the length scale be L ∼ 10 cm, velocity
U ∼ 1 m/s, density ρ ∼ 103 kg/m3 , dynamic viscosity µ ∼ 10−3 N s/m2 , then Re ∼ 106 ,
and the factor δ ∼ 10−3 giving the length scale of the boundary layer about δL ∼ 0.1 mm.
The scenario, we have just examined, has a transient character. This means that during
time t0 = L/U the viscous layer of width δL has developed. To verify the statement,
consider a transient problem.

7.3 Non-stationary problems of hydrodynamics: tran-

sient viscous boundary layer.
In this section we are going to consider a time dependent problem for the first time. We
will now use the so-called rectilinear flow ansatz applied to (7.10) to obtain some closed
form analytical solutions related to the problem of the viscous boundary layer, Fig. 7.1.
We consider a 2-D flow over a lineal impermeable solid substrate, as in Fig. 7.1, with the
no-slip boundary condition at x2 = 0 (at the solid surface), that is v1 (x1 , 0) = 0, and a
rectilinear steady uniform flow in the x1 -direction in the far field at x2 → ∞, where we
set v1 = 1 and v2 = 0. The impermeability condition is satisfied automatically, since it is
assumed that v2 = 0. The condition in the far field implies that we consider a problem
of initially uniform flow (plug flow) over a solid surface, which is going to evolve in time.
Note, we drop˜for the sake of easier notations.
We will seek for a solution to (7.10) in the form v = (v1 , 0). This form is known as the
rectilinear flow ansatz.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 118

Using the ansatz, from the incompressibility condition

∂v1
= 0.
∂x1

We immediately have v1 = v1 (x2 ). Substituting the resultant form of velocity back in

(7.10), one gets
∂v1 ∂p 1 ∂ 2 v1
=− + 2 . (7.11)
∂t ∂x1 δ Re ∂x22
Now, we have that velocity v1 = v1 (x2 ) only, while p = p(x1 ) only. Then, if we are-arrange
in (7.11)
∂v1 1 ∂ 2 v1 ∂p
− 2 2
=− , (7.12)
∂t δ Re ∂x2 ∂x1
the LHS is a function of x2 , while the RHS is a function of x1 . This is only possible if
∂p
= λ = const.
∂x1

We immediately have p = λx1 + p0 . But, at x2 → ∞, the flow is a steady rectilinear flow

v1 = (1, 0), where the pressure gradient is zero. Hence, we are bound to have λ = 0 and
the pressure is constant p = p0 .
Then (7.12) can be reduced to
∂v1 1 ∂ 2 v1
= 2 . (7.13)
∂t δ Re ∂x22

Equation (7.13) is the well-known parabolic partial differential diffusion equation, which
is invariant under the similarity transformation of variables, xˆ2 = kx2 , t̂ = k 2 t. So, we
x2
will seek for a solution to (7.13) in the self-similar form v1 = v1 (η), where η = √ . Then,
2 t
(7.13) is transformed into an ordinary differential equation

d2 v1 2 dv1
= −2δ Re η . (7.14)
dη 2 dη
 
x2
This can be seen if we formally change variables from (x2 , t) to η = √ , z = t in
2 t
the partial differential equation and take into account that v1 = v1 (η) is a function of η
only. Indeed, using the chain rule for partial differentiation
∂ ∂η ∂ ∂z ∂
= + ,
∂t ∂t ∂η ∂t ∂z

∂ ∂η ∂ ∂z ∂
= + ,
∂x2 ∂x2 ∂η ∂x2 ∂z

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 119

one gets
∂ η ∂ ∂
=− +
∂t 2z ∂η ∂z

∂ 1 ∂
= 1/2 .
∂x2 2z ∂η
From the second equation,
∂2 1 ∂2
 
1 ∂ 1 ∂
2
= 1/2 = .
∂x2 2z ∂η 2z 1/2 ∂η 4z ∂η 2

∂ ∂2
Substituting the expressions for the operators , into (7.13) and neglecting contri-
∂t ∂x2
∂
butions from (since we intend to apply operators to v1 (η), which is a function of η
∂z
only) , we will get (7.14).
dv1
Equation (7.14) can be integrated twice (if first designating = f (η) as a new depen-
dη
dent variable) to obtain
√
Z δ Re η
v1 = C exp(−ξ 2 ) dξ + C1 .
0

Then, using the boundary condition v1 (0) = 0 at η = 0 (no-slip condition at x2 = 0

at t > 0, that is at η = 0), one gets C1 = 0. Using the second boundary condition
lim v1 (η) = 1 (the condition in the far field x2 → ∞ at t > 0, that is at η → ∞) we get
η→∞

2
C=√
π
taking into account the fact that
Z ∞ √
2 π
exp(−ξ ) dξ = .
0 2

So, finally, we have the result

√
Z δ Re η
2
v1 = √ exp(−ξ 2 ) dξ, (7.15)
π 0

which can be represented as  √ 

v1 = erf δ Re η ,
where erf(x) is the error function - special function in mathematical physics. The solution
is illustrated in Fig. 7.2. Note, we can also consider another, similar problem, when

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 120

initially the liquid above the solid is at rest, but at time t = 0 the solid starts to move
with velocity v s = (1, 0). In this case, we have in the far field x2 → ∞, v = (0, 0) and at
the solid substrate v1 (0) = 1, and of course v2 = 0. Then C1 = 1 and C = − √2π giving
the net result  √ 
v1 = 1 − erf δ Re η .

One can see from (7.15) that the r

width of the viscous boundary layer, if we define it at
√ 2 t
δ Re η = 1, is given by xBL = , or coming back to dimensional variables
δ Re
r
LU t
xBL = 2 . (7.16)
Re
So the width increases with time.

1.0
Error function, erf(x)

0.8

0.6

0.4

0.2

0.0
0 2 4 6 8 10
x

Figure 7.2: The erf(x) function.

From the result (7.16), we just obtained, the width of the viscous boundary layer at the
characteristic time t0 = L/U is
2L
xBL = √ ,
Re
2
which tells us that indeed δ = √ - the result that only differs by a factor of 2 from the
Re
scaling analysis estimate.
On the other hand, the time needed to obtain the viscous boundary layer of the size of L
is
ReL ρL2
tV Z = = . (7.17)
4U 4µ
This is the time, when the whole region of the size of L will become the viscous zone. A
simple estimate for water would give at L ∼ 10 m that t0 ∼ 108 s or 3 × 104 h or about

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 121

3 years. That is why for large scale water flows the approximation of inviscid liquid is
usually good, since the viscous layer has no time to penetrate into the liquid domain.
Theoretically, this would be only possible if we would observe steady conditions in a river
flow, for example, during 3 years.
♦ Definition 7.3.1. A viscous flow in the time limit t  tV Z is usually called the
developed viscous flow. So that the viscous flow is fully developed when the whole domain
of the problem is the viscous zone. We can only expect a flow to be a steady viscous flow,
if it is fully developed.

7.4 Some notes on the Stokes flow regime: the limit

of small Reynolds numbers.
In this limit Re → 0, the Navier-Stokes equations take the form of the Stokes equations.
It is important to notice that time t does not enter the system explicitly anymore, so any
time dependence can come into the model only from the boundary conditions. Inside the
liquid, everything propagates instantly. This can be seen if we consider the characteristic
time necessary to reach a fully developed viscous flow given by (7.17) in the limit Re → 0.
In the limit we have
ReL
tV Z = lim =0 (7.18)
Re→0 4U

provided that L/U ∼ 1. So in the limit the flow is bound to be developed instanta-
neously.
Moreover, if we apply ∂/∂xi to the second equation in (7.5), then
∂ 2p ∂ ∂ 2 vi
= .
∂x2i ∂xi ∂x2j
If we now assume that vi are smooth functions of the coordinates xk , then the order of
the partial differentiation does not matter and we will get
∂ 2p ∂ 2 ∂vi
= .
∂x2i ∂x2j ∂xi
∂vi
But, due to the incompressibility condition = 0 and hence
∂xi
∂ 2p
=0
∂x2i
or
∆p = 0. (7.19)
That is the pressure p in the limit of small Reynolds numbers Re  1 is a harmonic
function and satisfies the Laplace equation.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 122

7.5 The rectilinear flow ansatz.

Consider a steady, fully developed two-dimensional in the (x1 , x2 ) plane flow of a New-
tonian and incompressible fluid with density ρ and dynamic viscosity µ over an infinite
impermeable solid substrate, which is the plane x2 = 0. It is essential that we have an infi-
nite and a flat substrate, so that we expect translational symmetry along the x1 -direction.
That is the velocity may not be changing in that direction at all.
To apply the rectilinear flow ansatz, we assume that the flow is rectilinear (laminar as
opposed to turbulent) in the x1 direction, that is v = (v1 , 0, 0). Note, the condition of
impermeability at the solid substrate v2 (x1 , 0) = 0 is satisfied by default.
Then, from the incompressibility condition (taking into account that v2 = 0)
∂v1
= 0,
∂x1
one has v1 = v1 (x2 ).
Note, that we can start our analysis from a bit more general statement that v = v(x2 ),
that is the velocity field has no x1 -dependence, but may have a v2 component. Then, of
course,
∂v1
= 0,
∂x1
∂v2
= 0.
∂x1
The incompressibility condition
∂v1 ∂v2
+ = 0,
∂x1 ∂x2
then results in
∂v2
= 0,
∂x2
That is v2 = const, but at the solid surface v2 = 0, hence v = (v1 (x2 ), 0, 0).
If we now substitute the ansatz (considering that v1 = v1 (x2 )) for a rectilinear flow in the
x1 direction, v = (v1 , 0, 0), into the Navier-Stokes equations, we obtain
∂p ∂ 2 v1 ∂Φ
0=− +µ 2 − ,
∂x1 ∂x2 ∂x1
∂p ∂Φ
0=− − ,
∂x2 ∂x2
where Φ is the potential of external forces.
Integrating the second equation results in
p + Φ = f (x1 ).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 123

That is the combination p + Φ is a function of x1 only. The first equation then becomes
df ∂ 2 v1
=µ 2.
dx1 ∂x2
One may notice that the LHS of the equation is a function of x1 only, while the RHS of
the equation is a function of x2 only. This is only possible if
df ∂ 2 v1
= λ, µ =λ
dx1 ∂x22
where λ is some constant to be found.
Integrating the two equations
λ x22
p(x1 , x2 ) + Φ(x1 , x2 ) = λx1 + f0 , v1 (x2 ) = + C 1 x2 + C 2 , (7.20)
µ 2
where f0 , C1 and C2 are some constants. Note, if Φ 6= 0, we can only be certain that their
combination is independent of x2 , but not p, Φ separately.
The result (7.20) will be further used to obtain particular solutions in various situations
relevant to practice. Below, as an example, we consider application of the no-slip boundary
condition in conjunction with a finite pressure condition.

7.5.1 No slip condition applied.

Now, if we apply the no-slip boundary condition v1 (0) = 0 at x2 = 0, then immediately

from (7.20) C2 = 0 and
λ x22
v1 (x2 ) = + C1 x 2 .
µ 2

What if the domain is infinite x1 ∈ (−∞, ∞), but the pressure is finite in the absence of
external forces Φ = 0, that is there is such a constant M that for all ∀x1 ∈ (−∞, ∞),
|p| < M . In this case one has to set λ = 0 to get rid off the divergent term λx1 in the
expression for pressure and we have

p(x1 , x2 ) = f0 , v1 (x2 ) = C1 x2 .

7.6 Fully developed viscous flows.

In this section we consider the problem of fully developed steady viscous flows, when the
full set of the Navier-Stokes equations admits an analytical solution.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 124

7.6.1 Simple shear flow

Problem set up is the following. Consider a steady two-dimensional in the (x1 , x2 ) plane
flow of an incompressible fluid with density ρ and dynamic viscosity µ over an infinite
solid substrate with the no-slip boundary condition, which is the plane x2 = 0, in the
absence of volume forces. The flow is rectilinear in the x1 direction, v = (v1 , 0, 0). In the
far field, the flow is driven by constant shear stress
∂v1
σ12 = µ = S0 ,
∂x2
where S0 is a constant. We will apply the result (7.20).
In the far field, the condition is asymptotic, so that one can integrate with respect x2 to
obtain
S0
v1 = x2 + C,
µ
where C is a constant, since v1 = v1 (x2 ). Now, at the substrate, x2 = 0, v1 (0) = 0, hence
C2 = 0. In the far field x2 → ∞, v1 = Sµ0 x2 + C, so λ = 0, C = 0, C1 = S0 /µ, and we
have
S0
v1 = x2 , p = f0 = const.
µ
This is the shear flow, which has a well recognised linear profile of the horizontal velocity,
Fig. 7.3.

wv1
S0
wx2

x2
x1

Figure 7.3: Shear flow. Velocity profile and stream lines.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 125

7.6.2 Couette and Poisseuille flows

The set up of the next two problems in this section is similar to the shear flow, only
instead of one impermeable solid wall, we have two walls. So, the rectilinear flow is now
confined between two solid walls. We will again use the result (7.20), since it does not
depend on the boundary conditions.
Poisseuille flow
In this example, we consider a flow between two solid walls at rest. That is at the first
wall, which is the plane x2 = 0,
v1 = 0, v2 = 0
and similar at the second wall, which is at x2 = h,

v1 = 0, v2 = 0.

Since the flow is assumed to be rectilinear, v2 = 0 everywhere, and the two bound-
ary conditions for the normal to the solid substrates component of velocity are satisfied
identically. The remaining two boundary conditions will give us two unknown constants
C1 , C2 , namely C2 = 0, C1 = − µλ h2 , so that

λ x22 λ hx2
v1 = − .
µ 2 µ 2

Subtle point

dp
One can see, that if the pressure gradient, λ = , is zero, then there is no net flow
dx1
in this case. This is characteristic to Poisseuille-like flows to have non-zero pressure
gradient. The flow is driven by this pressure gradient and the velocity distribution has a
well recognized parabolic profile, Fig. 7.4. The stream function for this flow is simply
∂ψ λ x22 λ hx2 ∂ψ
= − , = 0,
∂x2 µ 2 µ 2 ∂x1
λ x32 λ hx22
ψ(x2 ) = const + − .
µ 6 µ 4
Using the property of the stream function, the total flow flux between the walls Q is
λ h3 dp h3
Q = ψ(h) − ψ(0) = − =− . (7.21)
µ 12 dx1 12µ

It is proportional to h3 , so if the gap between two plates would be twice larger, the flux
will be 8 times larger at the same pressure gradient.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 126

x2 h
p ( x1 ) p ( x1  Gx1 )
x2
x1 x2 0

p( x1 ) ! p( x1  Gx1 )

Figure 7.4: Poisseuille flow. Velocity profile and the stream lines.

 Example 7.6.1. In a 3-D case of a circular pipe, the total flux through the pipe Q driven
dp dp πR4
by a pressure gradient (x is along the pipe) is known to be Q = , where R is the
dx dx 8µ
pipe radius, µ liquid viscosity. So in this case the total flux is proportional to the fourth
power of the system size R4 , rather than h3 as we have obtained in the 2-D case. Consider
a simple example of crude oil (viscosity µ = 2.5×10−2 N s/m2 ) flow in a pipe R = 0.2 m. If
we would like to keep the flow at average velocity U = 10 m/s, that is Q = U πR2 , then the
dp U 8µ dp
necessary pressure gradient is given by = 2
= 5 × 106 kg m−2 s−1 or = 0.5 atm
dx R dx
5 −1 −1
per kilometre (one atmosphere of pressure atm = 10 kg m s ). So to move oil over the
distance of 100 kilometres, one needs to apply pressure difference about 50 times larger
then the normal atmospheric pressure.

Couette flow
Couette flow, unlike Poisseuille flow, is driven by the shear stress applied at the boundary.
In this example the flow is driven by the upper plate at x2 = h and the pressure is finite
in the system. The set of boundary conditions is almost the same, only now the upper
plate is moving with velocity v0 .
at x2 = 0,
v1 = 0, v2 = 0
at x2 = h,
v1 = v0 , v2 = 0.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 127

The pressure gradient is zero, λ = 0, since the pressure is finite in the system. Applying
the boundary conditions at the solid walls, we have, as before, C2 = 0 and C1 = vh0 . So
v0
v1 = x2 .
h
The velocity distribution profile is linear, similar to the shear flow, Fig. 7.5.

v0

x2
x1

Figure 7.5: Couette flow. Velocity profile and the stream lines.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 128

7.7 Exercises - VII

1. Consider a steady two-dimensional, in the (x1 , x2 ) plane, fully developed viscous
flow of a Newtonian and incompressible liquid with density ρ and dynamic viscosity
µ between two infinite impermeable solid plates, which are parallel to the x1 -axis
and located at x2 = 0 and x2 = 1 respectively. The first plate, at x2 = 0, is
stationary and the flow is driven by the second plate, at x2 = 1, which is moving
in the x1 direction such that it exerts constant shear stress (the stress in the x1
direction, parallel to the substrate) σ21 = σ0 on the liquid. It is known, that the
flow is rectilinear in the x1 direction, the no-slip boundary condition is observed on
the plate at x2 = 0, the pressure is finite in the system and the external potential
Φ is zero.

• Write down the governing equations, in the tensor form, and the boundary
conditions on both plates, such that this is sufficient to form a well-defined
fluid mechanical problem.
• Demonstrate that the distribution of velocity in the rectilinear flow in the x1
direction has the form
α 2
v1 (x2 ) = x + C 1 x2 + C 2 ,
2µ 2
where α, C1 , C2 are some constants.
• Determine all the unknown constants α, C1 , C2 and the distribution of pressure
(up to a constant).

2. Consider a steady two-dimensional, in the (x1 , x2 ) plane, fully developed viscous

flow of a Newtonian and incompressible liquid with density ρ and dynamic viscosity
µ between two infinite impermeable solid plates, which are parallel to the x1 -axis
and located at x2 = 0 and x2 = 1 respectively. The plates are stationary and
the flow is driven by applied potential external volume forces with the potential
Φ = Φ0 x1 , where Φ0 is a given constant. It is known, that the flow is rectilinear
in the x1 direction, the pressure is finite in the system and there is no slip at both
solid substrates.

• Demonstrate that the distribution of velocity in the rectilinear flow in the x1

direction has the form
α 2
v1 (x2 ) = x + C 1 x2 + C 2 ,
2µ 2
where α, C1 , C2 are some constants.
• Determine all the unknown constants α, C1 , C2 and the distribution of pressure
(up to a constant).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 129

3. ”Wet paint on a wall” problem. Consider a steady fully developed flow in

a uniform film of a viscous, Newtonian and incompressible fluid with viscosity µ
and density ρ down a vertical plane under the force of gravity, see Fig. 7.6. Find
distribution of velocity and the total volume flux using the rectilinear flow ansatz.
One can assume that the air is inviscid with constant pressure p0 and the free surface
is flat in this example. The boundary condition at the solid surface is the no-slip.

x2
x1

F=−ϱ g 0 i 1

Air
p= p0

h0

Figure 7.6: Viscous flow over a vertical plane under the action of the gravity force.

4. ”Stream driven by wind” problem. Consider a fully developed flow of a viscous

Newtonian and incompressible fluid of viscosity µ and density ρ caused by the wind
blowing over a wide, but shallow reservoir, exerting a constant shear stress S0 on
its surface in the tangential to the surface direction, as is shown in Fig. 7.7. The
reservoir has a constant depth h0 without the wind and a finite dimension in the
x1 -direction L with the no-slip condition at the solid bottom. It is known that there
is no net water flow at the boundaries. The air is almost frictionless with viscosity
µ0  µ and almost constant pressure p0 . One can presume that the motion of the
air is somehow given and the only effect of its presence is the tangential stress acting
on the boundary S0 and the normal to the interface pressure p0 .

• Analyse the flow using the rectilinear ansatz assuming that it is a steady fully
established viscous flow in the reservoir after long time. Make any necessary
assumptions to obtain a solution to this problem.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 130

F=−ϱ g 0 i 1
Air
p= p0
Wind

S0
Tangential stress x2
h0
x1

Figure 7.7: Viscous flow due to wind.

• Determine a condition, when the distortion of the initially flat free surface
profile by the wind is small defining a measure of this smallness.

(MA3FM/MA4FM) Autumn 2020

Chapter 8

Flows in porous materials: Darcy’s

law, filtration and making espresso.

8.1 Introduction.
Description of liquid flows in porous media belongs to a special category of fluid mechanical
applications. The internal structure of porous media is very complicated and disordered.
As a result, there is absolutely no chance (in most real scenarios) to describe flow behaviour
in every detail even with the use of a simplified model in the limit of low Reynolds numbers
(typical flow regime in porous media) and modern, powerful numerical tools, see for
example Fig. 8.1. Instead, a statistical approach is used in practice based on averaging
techniques. The idea of averaging is rather intuitively transparent and similar to the
continuum approach we have used to describe fluids. To apply continuum approach to
fluids, we ignored internal molecular structure of liquids and considered volume elements
containing a large number of atoms or molecules as points of continuum. In porous media,
we can perform similar kind of averaging. Every point of continuum now should contain
many elementary volume structures of the porous media. For example, if the material
is sand, then an elementary volume should contain many sand particles - elementary
structures of sand materials, see Fig. 8.2. As a result, all quantities such as pressure p
and velocity v are the averages over such macroscopic volumes.
♦ Definition 8.1.1. Porous bodies have several characteristics important in applications.
One of them is porosity. Porosity χ is defined as the ratio of the volume of voids Vo in
a sample volume V , that is χ = Vo /V . Typical values of porosity lie in the range of
20 − 30%. Another important parameter is saturation s. Saturation is defined as the ratio
of the liquid volume Vl contained in a sample volume V , to the volume of voids Vo in
that sample volume, that is s = Vl /Vo . In the course, we only consider situations when
s = 100%. This case is called fully saturated porous medium.

This means that porous media has no structure from the modelling point of view. So
that, pressure in a sample of sand, for example, is an average over some large volume,

131
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 132

containing many grains, rather than a local value of pressure in the liquid around some
individual grain particle. That is, if V is a sample volume with porosity χ, p̂(x, t) is the
local pressure distribution in that
R sample volume, then the average value of pressure p
−1 −1 3
can be calculated by p = χ V χV
p̂(x, t) d x, where the integral is taken of the volume
of voids Vo = χV .
The same statement applies to the average volumetric flux density q or average velocity
v. The flow field around a particular grain in sand can be quite complicated and does
depend on the position or the distance from a particular grain particle, but on average
the total volume flux through a surface element dS = dSn is Q = qdS. That is the
volumetric flux density can be thought as an average velocity q ∼ v. But, they are not
equal since not all the area is available for the liquid to flow, so that one needs to correct
the value of the volumetric flux density by the porosity q = vχ.

Figure 8.1: Typical pattern of flow stream lines inside porous materials.

8.2 Darcy’s law.

It turned out that such an empirical approach of averaging was very successful, so that
a phenomenological equation, a law, was postulated by Henry Darcy, a French engineer.
The Darcy’s law is fulfilled for many kinds of porous media, such as most sands, clays
and soils, and it can be formulated in a rather simple (vector) form (1856, Dijon, France)
κ
q = − ∇p. (8.1)
µ
Here p is the averaged macroscopic pressure, q is the macroscopic volumetric liquid flux
density, µ is liquid viscosity and κ(x, p, t) is the so called coefficient of permeability. One

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 133

Point of
continuum

Porous
media
continuum

Figure 8.2: Illustration of macroscopic continuum approach to porous materials.

can see that the liquid flow is driven by the pressure gradient. The permeability coefficient
has dimension [κ] = m2 of surface area, that is it is a purely geometrical characteristic of
porous materials. It is for this reason viscosity µ stands out in the Darcy’s law to make
all parametric dependencies explicit.
In isotropic media, the coefficient of permeability is a scalar function of coordinates x, time
t and possibly macroscopic pressure p. In anisotropic media, where conductivity differs
in various directions, permeability is a tensor. So in the most general form including the
presence of potential external forces, like gravity F = −∇Φ, the Darcy’s law has a form

κmk ∂(p + Φ)
qm = − . (8.2)
µ ∂xk

8.2.1 Incompressibility condition in porous media and the gov-

erning equations.

Apparently, in a general case porous media can not be considered as incompressible be-
cause it may contain some portions of a gas phase. But, in the case of fully saturated
porous samples, since both solids and liquids (no gas is involved in this case) are incom-
pressible, we can come up with pretty the same condition of incompressibility as before

∇ · v = 0.

Since q = χv, we can express the incompressibility condition as

∇ · q = 0.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 134

If we integrate the condition over some macroscopic volume element V with smooth
boundary ∂V , then in the absence of any sources or sinks of the liquid in the volume
element (including the assumption that porous particles are not absorbing the liquid
too), applying Gauss divergence theorem
Z Z
3
∇·vd x = v · n dS = 0,
V ∂V

where n is the normal vector to the surface ∂V . That is the total flux through the surface
is zero.
Consider now isotropic porous media. If we apply the nabla operator through to the
Darcy’s law in this case, we get
 
κ
∇ · q = −∇ · (∇p + Φ) .
µ
Using incompressibility condition, we arrive at

∇ · (κ∇(p + Φ) = 0.

In the case κ = κ0 = const, the above equation is simplified to

∆(p + Φ) = 0

that is to the Laplace equation for the combination of pressure and the external force
potential. We recall that a similar result has been obtained for the Stokes flows in the
limit of Re  1, see (7.19). In fact, one can observe other similarities, such as the absence
of explicit time dependence in both cases. Nevertheless, despite all those common features,
the range of applicability of the Darcy’s law is wider. It can still be applied at Re ∼ 10,
while the Stokes equations require Re  1. further, we consider one example where the
connection to the Navier-Stokes equations can be obtained explicitly and compared with
the empirical Darcy’s law.

8.2.2 Geometrical model of porous media.

As we have already mentioned in the introduction, the Darcy’s law is phenomenological.

That is only in principle it can be connected with the Navier-Stokes equations. Here we
consider an example of a rectilinear flow between a system of plates separated by distance
h, Fig. 8.3, which can be considered as a two-dimensional model porous medium. We
have already solved this problem for a steady fully developed viscous flow between two
immobile solid plates, see (7.21). According to our solution, the total flux per unit width
between any pair of the plates is
dp h3 p0 − p1 h3
Q=− =− .
dx 12µ H 12µ

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 135

Pressure
p0

Flux H

Pressure
p 1< p 0 h

Figure 8.3: Illustration of a model porous medium.

To translate the total flux per unit width into the flux density, we divide Q/h. Finally

p0 − p1 h2
q=− .
H 12µ
We can now identify the elements in the above equation. Apparently, permeability of
2
the system of plates is κ = h12 . One can imagine other examples where some analytical
results would be available, but in general all properties of porous materials are
embedded into the only parameter of the Darcy’s law, κ, which has to be determined
from experiments. The physical meaning of this parameter (to obtain an estimate) is the
characteristic area of gaps in a porous network matrix.

8.2.3 One-dimensional flows in porous media.

Consider a one-dimensional isotropic case of a flow in a fully saturated porous material,

where both velocity or the flux density and pressure depend on only one variable x. The
Darcy’s law is
κ(p) d(p + Φ)
q=− , (8.3)
µ dx
where Φ = Φ(x) is external force potential.
Using the standard procedure, that is applying the incompressibility condition ∇ · q = 0
dq
or = 0, one arrives at
dx  
d κ(p) d(p + Φ)
= 0.
dx µ dx

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 136

One can eliminate constant viscosity µ to simplify to

 
d d(p + Φ)
κ(p) = 0. (8.4)
dx dx

Consider a solution to the above equation defined on a finite interval x ∈ [0, L], which is
a subject to the following boundary conditions

p(0, t) = p0 (t), p(L, t) = p1 (t).

Integrating (8.4) once, one gets

d(p + Φ)
κ(p) = C0 (t). (8.5)
dx
Here, pay attention, C0 (t) is unknown function of time. In what follows, we consider the
case Φ = 0.
Integrating the above equation from p0 (t) to p(x, t), one gets
Z p(x,t)
κ(ξ) dξ = C0 (t)x + C1 (t).
p0 (t)

From the boundary condition at x = 0, one has C1 = 0. This is only due to our choice
of the integration limits. From the second boundary condition at x = L, one can reckon
that Z p1 (t)
κ(ξ) dξ = C0 (t)L.
p0 (t)

Thus distribution of pressure is given implicitly by

Z p(x,t)
x p1 (t)
Z
κ(ξ) dξ = κ(ξ) dξ.
p0 (t) L p0 (t)

In a particular case when κ(p) = κ0 = const,

p1 (t) − p0 (t)
(p(x, t) − p0 (t)) = x
L
and
dp p1 (t) − p0 (t)
=
dx L
giving the flux density
p1 (t) − p0 (t) κ0
q=− .
L µ
The total flux Q trough some area S can then be easily obtained Q = qS. So in a 1-D
case, the Darcy’s law can be ’integrated’ exactly (up to a quadrature).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 137

Atmospheric pressure
Flooding p= p0

H 0 (t ) Water layer
x=0

H1 Infiltration
Porous layer
x
Aquifer

Figure 8.4: Illustration of water infiltration problem after flooding.

 Example 8.2.1. Consider a one-dimensional problem of water infiltration after flood-

ing through a layer of fully saturated porous soil into the aquifer layer, Fig. 8.4. The
infiltration process takes place in the gravity field potential Φ = −ρg0 x, where coordinate
x is counted from the water-porous soil boundary, ρ is water density and g0 is the grav-
ity constant. The initial height of the water layer above the porous substrate H0 (t) is
H0 (0) = h0 , while the depth of the porous layer is H1 . The coefficient of fully-saturated-
soil permeability is assumed to be a known constant κ0 and water viscosity µ is given.
Calculate the time required for the water to soak completely into the porous layer, if it is
known that the liquid pressure at the aquifer layer is equal to the atmospheric pressure p0 .
The Darcy’s law in the soil porous layer in the one-dimensional case (8.3) according to
(8.4) assumes a form
d2 (p + Φ)
= 0,
dx2
which has a solution p(x, t) + Φ(x) = p(x, t) − ρg0 x = C0 (t)x + C1 (t). We already know
from solving static problem that pressure at the soil-porous body interface at x = 0 is equal
to p(0, t) = p0 +ρg0 H0 (t), while at aquifer the pressure is constant p(H1 , t) = p0 . Applying

the boundary conditions, we have C1 (t) = p0 + ρg0 H0 (t) and C0 (t) = −ρg0 1 + HH0 (t) 1
.

From (8.3), the volumetric water flux density

 
κ0 d(p + Φ) κ0 C0 (t) ρg0 κ0 H0 (t)
q(x, t) = − =− = 1+ .
µ dx µ µ H1
Then, from the conservation of mass in the water layer
 
dH0 ρg0 κ0 H0 (t)
= −q(x, t) = − 1+ .
dt µ H1

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 138

µH1
Normalizing H0 by H1 , Ĥ0 = H0 /H1 and time by t0 = ρg0 κ0
, t̂ = t/t0 , we arrive at

d Ĥ0
= −1 − Ĥ0 (t̂),
dt̂
which has a general solution Ĥ0 (t̂) = C exp(−t̂) − 1 with initial condition Ĥ0 (0) = h0 /H.
Note, the general solution can be found by finding a particular solution Ĥ0p (t̂) = −1, which
is not difficult to guess, and adding the general solution of homogeneous equation

d Ĥ0
= −Ĥ0 (t̂),
dt̂
which is Ĥ0a (t̂) = C exp(−t̂).
Applying the initial condition, one obtains
 
h0
Ĥ0 (t̂) = + 1 exp(−t̂) − 1.
H1
That is returning to dimensional variables

H0 (t) = (h0 + H1 ) exp(−t/t0 ) − H1 .

The moment of time when the water layer will disappear H0 (tc ) = 0 is then given by

(H1 + h0 ) exp(−tc /t0 ) = H1

or  
H1 + h0
tc = t0 ln .
H1
 Example 8.2.2. Consider the operation of a coffee Espresso machine, Fig. 8.5. To
simplify the problem, consider the flow of hot water prepared in a reservoir under pressure
P0 = 1.5 × 106 kg m−1 s−2 through ground coffee powder with known constant coefficient of
permeability κcf = 10−13 m2 , Fig. 8.6. During the preparation of one portion of coffee,
the water pressure vanishes with time as P = P0 + (Pa − P0 )(1 − exp(−t/t0 )), t0 = 10 s.
Calculate the total flux Q trough the coffee container, basically the amount of the coffee
produced, assuming that the problem is one-dimensional, the liquid is incompressible with
viscosity µ = 10−3 N · s/m2 , the coffee powder is fully saturated during the operation and
the flow obeys standard Darcy’s law. The machine operates at ambient air pressure of one
atmosphere, that is Pa = 105 kg m−1 s−2 . The height of the container is H = 5 cm and the
cross-section area is constant S = 3 × 10−3 m2 .

The total flux from the Darcy’s law is

Z ∞
κcf dP
Q = −S dt
0 µ dx

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 139

Figure 8.5: Illustration of Espresso making. Note, that the typical ground and pressed
coffee has permeability κcf = 10−12 − 10−14 m2 .

Hot water under pressure

P0

Area S H

y
x

Figure 8.6: A sketch of Espresso making process.

where x-axis is in the flow direction with the origin at the hot pressurised water entrance
to the porous coffee powder.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 140

The distribution of pressure obeys one-dimensional Laplace equation

d2 P
=0
dx2
which has a solution P = Ax + B. Using two known values of pressure, P (0, t) =
P0 + (Pa − P0 )(1 − exp(−t/t0 )) and P (H, t) = Pa , one has

x dP (Pa − P (0, t))

P = (Pa − P (0, t)) + P (0, t); = ,
H dx H
and Z ∞
S κcf S κcf
Q= (P (0, t) − Pa ) dt = (P0 − Pa ) t0 = 84 mL.
H µ 0 H µ

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 141

8.3 Exercises - VIII

1. One-dimensional flow of a Newtonian, incompressible fluid with dynamic viscosity
µ in a lineal pipe of length L with constant cross-section area S filled with a porous
material is driven by a gradient of pressure along the pipe. The porous medium is
fully saturated. The liquid in the porous medium obeys the Darcy’s law

κ(p)
q=− ∇p
µ

where κ(p) = p−1 , q is the fluid flux density and p is the pressure in the liquid.

• Show that the pressure distribution in this one-dimensional case, when both
p = p(x, t) and the flux density component along the pipe q = q(x, t) are
functions of coordinate x along the pipe and time t, obeys the partial differential
equation  
∂ 1 ∂p
= 0.
∂x p ∂x
• Hence, or otherwise, find the distribution of pressure p(x, t) and the flux density
q(x, t) along the pipe, the total flux through the pipe, starting from t = 0 and
to t = t0 assuming at x = 0, p(0, t) = p0 , and at x = L, p(L, t) = p0 exp(−t).
The pipe length is L.
• In the same set-up, but when there is applied external potential Φ(x) = −Φ0 x
and it is known that the flux everywhere in the pipe is zero, find the pressure
at x = L if at x = 0, p = p0 .

2. One-dimensional flow of a Newtonian, incompressible fluid with dynamic viscosity

µ in a lineal pipe of length L with constant cross-section area S filled with a porous
material is driven by a gradient of pressure along the pipe. The porous medium is
fully saturated. The liquid in the porous medium obeys the Darcy’s law

κ(p)
q=− ∇p
µ

where κ(p) = p4 , q is the liquid flux density and p is the pressure in the liquid.
Assuming that all variables in this one-dimensional case are functions of coordinate
x along the pipe and time t:

• Find the distribution of pressure p(x, t) and the flux density q(x, t) along the
pipe assuming that at x = 0, p(0, t) = t0 − t, and at x = L, p(L, t) = 0.
• Find the total flux through the pipe starting from t = 0 and to t = t0 .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 142

3. One-dimensional flow of a Newtonian, incompressible fluid with dynamic viscosity

µ in a lineal pipe of length L with constant cross-section area S filled with a porous
material is driven by a gradient of external force potential Φ(x, t). The porous
medium is fully saturated. The liquid in the porous medium obeys the Darcy’s law

κ(p)
q=− ∇(p + Φ)
µ

where κ(p) = κ0 = const, q is the liquid flux density and p is the pressure in the
liquid. Assuming that all variables in this one-dimensional case are functions of
coordinate x along the pipe and time t, and the potential Φ(x, t) is given:

• Find the distribution of pressure p(x, t) and the flux density q(x, t) along the
pipe assuming that at x = 0, p(0, t) = p0 , Φ(0, t) = Φ0 (t), and at x = L,
p(L, t) = p0 , Φ(L, t) = Φ1 (t).
• Find the total flux through the pipe starting from t = 0 and to t = t0 .

4. Consider a two-dimensional flow of a Newtonian, incompressible fluid with dynamic

viscosity µ taking place in the radial direction in an infinitely long cylindrical shell
filled with a porous material, Fig. 8.7. The liquid dispersion in the radial direction
is axis-symmetric, that is all dependent variables (such as liquid pressure) in the
problem are functions of only one independent variable, the distance r from the
cylinder centre. The cylinder is rotating with constant angular velocity around its
centre so that the infiltration process takes place under the action of a centrifugal
force F = Ar directed radially outward, where A is some positive constant. The
porous medium is fully saturated, so that the coefficient of permeability κ is con-
stant, which is assumed to be known. The liquid pressure at the porous domain
boundary at r = R1 is known, p(R1 ) = p1 . If it is given that the total volumetric
outward flux is Q per unit length of the cylinder, determine the value of pressure at
r = R2 .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 143

Centrifugal force F=A r

R2

R1

Porous
media

Figure 8.7: Centrifugal flow in a porous cylindrical shell. The cross-section of the cylinder
with inner and outer radii R1 and R2 respectively.

(MA3FM/MA4FM) Autumn 2020

Chapter 9

The shallow water approximation -

water waves and non-stationary
solutions

The problem is set in the following way. Consider a thin layer of an incompressible and
inviscid liquid of density ρ resting on a flat solid substrate, which is the plane x2 = 0,
with the free surface at rest at x2 = H. The external air pressure is assumed to be
constant p = pa . The liquid is subjected to the force of gravity directed in the x2 -direction
downwards, with the external potential gradient ∇Φ = −ρg0 i2 (g0 is the gravity constant),
which is normal to the solid surface. Perturbations to the free surface are introduced as
follows, h(x1 , t) = H + ξ(x1 , t). Consider the dynamics of such perturbations, their
evolution in time t.
The problem, we consider, is two-dimensional, and the liquid layer thickness H is
assumed to be much smaller than the length scale in the horizontal direction L, that is
H  L. To put that statement differently, that the parameter δ = H/L << 1. This
approximation is known as the shallow water approximation. Then one can use a variant
of the Prandtl model (7.10), neglecting now viscous terms to analyze dynamic behaviour,
but with external force potential added to the right hand side ∇Φ = −ρg0 i2 .
While it has been said that the liquid is inviscid, we actually assume that the liquid layer
thickness is larger than the width of the viscous layer.
The system of the Navier-Stokes equations in our problem can be simplified if we use the
following scaling, x˜1 = x1 /L, x˜2 = x2 /H, v˜1 = v1 /U , v˜2 = v2 /δU , δ = H/L, t̃ = t/t0 ,
t0 = L/U , p̃ = p/p0 , p0 = ρU 2 . Here U is the characteristic velocity in the x1 direction.
Now, omitting tilde (˜) to simplify notations, one has
∂vk
=0
∂xk
and
∂v1 ∂v1 ∂p
+ vl =− , (9.1)
∂t ∂xl ∂x1

144
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 145

 
∂v2 ∂v2 ∂p g0 H
+ vl δ2 = − − 2 .
∂t ∂xl ∂x2 U

Then, neglecting terms of the order of δ 2 << 1 in (9.1), one has

∂vk
=0
∂xk

and
∂v1 ∂v1 ∂p
+ vl =− , (9.2)
∂t ∂xl ∂x1
∂p g0 H
0=− − 2 .
∂x2 U
Note, all variables are non-dimensional in the above equations.

First step

At the free surface (at x2 = 1 + ξ(x1 , t)), the pressure is pa , which may be regarded as
the atmospheric pressure. Hence, from the last equation in (9.2),

g0 H
p = f (x1 ) − x2 ,
U2
f (x1 ) is some function of x1 . At x2 = 1 + ξ(x1 , t) (free surface), p(x1 , 1 + ξ) = pa , so
f (x1 ) = pa + gU0 H
2 (1 + ξ(x1 , t)) and

g0 H g0 H g0 H
p = pa + 2
(1 + ξ(x1 , t)) − 2 x2 = pa + 2 (1 + ξ(x1 , t) − x2 )
U U U

Second step

So, substituting the distribution of pressure, just found, into the second equation in (9.2),

∂v1 ∂v1 g0 H ∂ξ
+ vl =− 2 .
∂t ∂xl U ∂x1

From the above, one can see that the material derivative of v1 is a function of x1 and t
only,
Dv1 ∂v1 ∂v1 g0 H ∂ξ
= + vl =− 2 .
Dt ∂t ∂xl U ∂x1
So, if initially v1 = v1 (x1 , 0), it will remain a function of x1 only.

Third step

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 146

The vertical motion of the free surface obeys,

dξ ∂ξ ∂ξ
= + v1 = v2 (x1 , 1 + ξ).
dt ∂t ∂x1

Integrating the incompressibility condition

∂v1 ∂v2
+ =0
∂x1 ∂x2
with respect to x2 from 0 to 1 + ξ, taking into account that at the bottom the normal
velocity v2 (x1 , 0, t) = 0, we end up with
Z 1+ξ Z 1+ξ
∂v1 ∂v1 ∂ξ ∂ξ
dx2 + v2 (x1 , 1 + ξ) = dx2 + + v1 = 0.
0 ∂x1 0 ∂x1 ∂t ∂x1
If we now make use of the identity
Z 1+ξ Z 1+ξ
∂ ∂v1 ∂ξ
v1 (x1 , x2 , t) dx2 = dx2 + v1 ,
∂x1 0 0 ∂x1 ∂x1
then Z 1+ξ
∂ξ ∂
=− v1 (x1 , x2 , t) dx2 ,
∂t ∂x1 0

and, assuming that v1 = v1 (x1 , t), we have finally a system of equations, which is known
as the shallow water equations

∂ξ ∂
+ {(1 + ξ)v1 (x1 , t)} = 0,
∂t ∂x1

∂v1 ∂v1 g0 H ∂ξ
+ v1 + 2 = 0.
∂t ∂x1 U ∂x1

Approximation of small perturbations of the free surface or small amplitude

waves.

In the following, we consider small perturbations only, when |ξ| << 1. Without pertur-
(0)
bations, the system is in the stationary state, v1 = 0, and one can neglect non-linear
terms v12 , ξv1 . In this approximation, the system of the shallow water equations becomes,
∂ξ ∂v1
+ = 0,
∂t ∂x1
(9.3)
∂v1 g0 H ∂ξ
+ 2 = 0.
∂t U ∂x1

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 147

L
h ( x 1, t )

H x2

x1

Figure 9.1: Shallow water approximation H  L. Water waves.

Differentiating the first equation in (9.3) with respect to t and the second equation with
respect to x1 ,
∂ 2 ξ g0 H ∂ 2 ξ
− 2 =0
∂t2 U ∂x21
which is the well-known wave equation. It has, as you know, a general solution in the
form of travelling waves,

ξ = A g(x1 − ct) + B q(x1 + ct)

where g, q are twice-differentiable functions, in particular

ξ = A cos(x1 − ct) + B cos(x1 + ct)

p
here A, B some constants, c = g0 H/U 2 is the non-dimensional
√ wave phase velocity -
the dimensional velocity then is given by C = cU = g0 H.
 Example 9.0.1. Tsunami waves is a good example of the use of the shallow water
approximation. Typical horizontal wave length of a tsunami wave is about 100 − 200 km,
which is much bigger than the average ocean depth, about a mile. As an example, let the
depth of ocean H be H = 3×103 m, then as g0 ' 10 m/s2 , C ' 170 m/s or C ' 600 km/h,
which is well-known average speed of tsunami waves.

To conclude.

Bibliography
(MA3FM/MA4FM) Autumn 2020
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 148

Viscous - Elasticity Elasticity

Continuum

Fluid Mechanics

Inviscid flows Re>>1

Stokes flows Re<<1

Potential flows Re>>1

+ vorticity is preserved

Porous medium flows: Darcy's law

Re = 10-8 Re = 1 Re = 10 Re = 108

Reynolds number scale

Figure 9.2: Plan of the course: Conclusions.

1. G. Worster, Understanding Fluid Flow, Cambridge University Press, 2009.

2. H. Jeffreys, Cartesian Tensors, (Students edition), Cambridge University Press,

1962.

3. G. Temple, Cartesian tensors : an introduction., Dover Books on Mathematics,

2004.

4. P.C. Matthews, Vector Calculus, Springer (London), 1998.

5. S.C. Hunter, Mechanics of Continuous Media, John Wiley and Sons Inc., Ellis Hor-
wood Lim., Chichester, 1976.

6. G.M. Homsy et al. Multimedia Fluid Mechanics, (second edition).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 149

Water
waves
Soft Matter

Viscoelasticity Theory of
Mayonnaise Fluid
Elasticity Mechanics

an m
ics
u
Me tinu
Potential flows

ch
Plumbing

n
Co
Tensor Algebra and Differential Geometry

Figure 9.3: Mathematical theory of continuum.

7. B.S. Massey, Mechanics of Fluids (sixth edition), VNR International, 1989.

8. H. Lamb, Hydrodynamics (sixth edition), Cambridge University Press.

9. Landau & Lifshitz, Fluid Mechanics (second edition), Elsevier, 2004.

10. G.K. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press,

1967. This book often referred to as the ”Bible” of Fluid Dynamics, though it may
be quite difficult for the first time readers.

(MA3FM/MA4FM) Autumn 2020

Chapter 10

Revisions

10.1 Vectors

The material for revisions is partly taken from the Vector Calculus Notes of MA2VC/MA3VC
given by Dr. A. Moiola, which I advise you to revisit - the Notes or any other book rec-
ommended.
The position vector
r = x1 i 1 + x2 i 2 + x3 i 3
represents the position of a point in the three-dimensional Euclidean space relative to
the origin. Where the scalars x1 , x2 and x3 are the components, or coordinates, of
r, and i1,2,3 are three fixed orthogonal vectors that constitute the canonical basis.
When we draw vectors, we always assume that the canonical basis has a right-handed
orientation, i.e. it is ordered according to the right-hand rule: closing the fingers of the
right hand from the i1 direction to the i2 direction, the thumb points towards the i3
direction.
The magnitude of a vector or its length r = |r| is defined as
q
r = x21 + x22 + x23 .

Vectors can be added or subtracted and/or multiplied by a scalar. Also, one can form
a scalar product of vectors. These operations are defined in any vector space. In the
following we briefly recall the definitions and the main properties of the scalar product
and the vector product. The addition, the scalar multiplication and the scalar product
are defined for Euclidean spaces of any dimension, while the vector product (thus also the
triple product) is defined only in three dimensions.

150
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 151

Given two vectors u and v, the scalar product or inner product is defined (in 3-D space)
as
X3
u·v = uk vk = uv cos θ
k=1

where θ is the angle between the vectors. Two vectors are orthogonal or perpendicular
if their scalar product is zero.
Given two vectors u and w, the vector product (or cross product) gives output a
vector denoted by u × w:

i1 i2 i3
u × w : = u1 u2 u3
w1 w2 w3
u2 u3 u u3 u u2
= i1 − i2 1 + i3 1
w2 w3 w1 w3 w1 w2
= (u2 w3 − u3 w2 )i1 + (u3 w1 − u1 w3 )i2 + (u1 w2 − u2 w1 )i3 ,

where || denotes the matrix determinant. Note that the 3 × 3 determinant is a “formal
determinant”, as the matrix contains three vectors and six scalars.
The magnitude of the vector product

|u × w| = |u||w| sin θ

is equal to the area of the parallelogram defined by u and w. Its direction is orthogonal
to both u and w and the triad u, w, u × w is right-handed.
Using basic operations with vectors, one can define more complex objects with the help
of a vector operator called the Del-operator ∇ defined as
3
X ∂
∇= ik ,
k=1
∂xk

see details in 10.2. The above definition can be also written as

∂
∇ = ik ,
∂xk

using the summation convention from 11.0.1.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 152

10.2 Some notes from vector calculus: scalar and

vector fields

A scalar field is a function f : D → Re, where the domain D is an open subset of Re3 .
The value of f at the point r may be written as f (r) or f (x, y, z). (This is because a
scalar field can equivalently be interpreted as a function of one vector variable or as a
function of three scalar variables.) Scalar field may also be called “multivariate functions”
or “functions of several variables” (as in the first-year calculus modulus).
Some examples of scalar fields are
f (r) = x2 − y 2 , g(r) = xyez , h(r) = |r|4 .

Smooth scalar fields can be graphically represented using level sets, i.e. sets defined by
the equations {f (r)=constant}. The level sets of a two-dimensional field are (planar)
level curves and can be easily drawn, while the level sets of a three-dimensional field are
the level surfaces, which are harder to visualise. Level curves are also called contour
lines or isolines, and level surfaces are called isosurfaces. Note that two level surfaces
(or curves) never intersect each other, since at every point r the field f takes only one
value, but they might “self-intersect”.
Very similar to the scalar fields, a vector field F is a function of position whose output
is a vector, namely it is a function
F : D → Re3 , F (r) = F1 (r) i1 + F2 (r) i3 + F3 (r) i3 ,
where the domain D is a subset of Re3 .
Let f be a scalar field. The vector field whose three components are the three partial
derivatives of f (in the usual order) is called gradient of f and denoted by
∂f ∂f ∂f
∇f (r) := grad f (r) := i1 (r) + i2 (r) + i3 (r). (10.1)
∂x ∂y ∂z
The symbol
∂ ∂ ∂
∇ := i1 + i2 + i3 (10.2)
∂x ∂y ∂z
is called “nabla” operator, or “del”, and is usually denoted simply by ∇.
Given a smooth scalar field f and a unit vector u, the directional derivative of f in
∂f
direction u is defined as the scalar product ∂u (r) := u · ∇f (r).
Equation f (r) = const defines a surface, the level surface of f . Vector ∇f is perpendicular
to the level surface of f . Hence, one can define a normal vector n to a surface f (r) = const
through
∇f
n=± .
|∇f |

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 153

If φ(x) is a scalar field, then the gradient of the scalar field, which is a vector, can be
written using the summation convention from 11.0.1
∂φ
grad φ = ∇ φ = ik .
∂xk

If u(x) is a vector field, then its divergence, which is a scalar, (using the summation
convention from 11.0.1) is
∂uk
∇.u =
∂xk
and the curl, which is a vector,
∂uk
∇ × u = curl u = εijk ii ,
∂xj
where ii are unit vectors of the coordinate system, εijk is the Levi-Civita symbol defined
in 11.0.1 and we have used again the summation convention from 11.0.1.

10.3 Jacobian Matrix and the Jacobian

In vector calculus, the Jacobian matrix is the matrix of all first-order partial deriva-
tives of a vector-valued function. Suppose F (x) is a function from Euclidean n-space
to Euclidean n-space. Such a function is given by n real-valued component functions,
F1 (x1 , ..., xn ), ..., Fn (x1 , ..., xn ). The partial derivatives of all these functions (if they do
exist) can be organized in an n-by-n square matrix, the Jacobian matrix J of F , as follows:
 ∂F ∂F1

∂x1
1
· · · ∂x n

J =  ... ... ..  .

. 
∂Fn ∂Fn
∂x1
· · · ∂xn

This matrix is also denoted by ∂(F1 ,...,Fn )

∂(x1 ,...,xn )
. The Jacobian determinant (often simply called
the Jacobian and designated by the same letter J as the matrix itself) is the determinant
of the Jacobian matrix.
In a three-dimensional case n = 3, one can consider the transformation from one set x of
coordinates to another set X of coordinates as a vector-valued function, x(X), then the
infinitesimal volume elements in terms of both sets of coordinates are related as
∂(x1 , . . . , x3 )
dx1 dx2 dx3 = |J| dX1 dX2 dX3 = dX1 dX2 dX3 .
∂(X1 , . . . , X3 )

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 154

The last relationship is often used to change coordinates in multi-dimensional integrals

(like volume or surface integrals). For example,
Z
f (x1 , x2 , x3 )dx1 dx2 dx3 =
V
Z
f (x1 (X1 , X2 , X3 ), x2 (X1 , X2 , X3 ), x3 (X1 , X2 , X3 )) |J| dX1 dX2 dX3 .
V

 Example 10.3.1. The transformation from Cartesian coordinates (x1 , x2 , x3 ) to spher-

ical coordinates (r, θ, φ), is given by:

x1 = r sin θ cos φ
x2 = r sin θ sin φ
x3 = r cos θ.
The Jacobian matrix for this coordinate change is

∂x1 ∂x1 ∂x1

 
 ∂r ∂θ ∂φ   
 ∂x sin θ cos φ r cos θ cos φ −r sin θ sin φ
 2 ∂x2 ∂x2  

J(r, θ, φ) =   = sin θ sin φ r cos θ sin φ r sin θ cos φ  .
 ∂r ∂θ ∂φ 
 ∂x3 ∂x3 ∂x3  cos θ −r sin θ 0
∂r ∂θ ∂φ

The determinant of J is J = r2 sin θ. Then, since dV = dx1 dx2 dx3 this determinant
implies that dV = r2 sin θ dr dθ dφ, where dV is the infinitesimal volume element.
So, a volume integral will be,
Z Z
f (x1 , x2 , x3 )dx1 dx2 dx3 = fˆ(r, θ, φ)r2 sin θ dr dθ dφ.
V V

f (x1 , x2 , x3 ) → fˆ(r, θ, φ).

10.4 The Inverse Function Theorem

For a function of a single variable, the theorem states that, if f is a continuously differen-
tiable function, which has nonzero derivative at x0 , then f is invertible in the neighborhood
of x0 , the inverse function f −1 is continuously differentiable and satisfies

0 1
f −1 (f0 ) =
f 0 (x 0)

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 155

where f0 = f (x0 ).
For functions of more than one variable, the theorem states that if the Jacobian determi-
nant of F at p is nonzero, then F is an invertible function in the infinitesimal region at
p. That is, an inverse function to F exists in some neighbourhood of F (p).

10.5 The Gauss-Ostrogradsky theorem (also known

as the Divergence theorem or Gauss’s theorem).

Theorem 10.5.1. If vector field F (x) is continuously differentiable, on a bounded and

connected subset (volume) V of R3 with piecewise smooth boundary S then
ZZZ ZZ
∇.F dV = ⊂⊃ F .n dS
V S

where n is the unit normal vector to the surface/boundary S.

Corollary 10.5.2. If f (x) is is continuously differentiable scalar field, then there the
following relationship between surface and volume integrals
I Z
f dS = ∇f dV
V

10.6 The Stokes’ theorem

Theorem 10.6.1. If vector field F (x) is continuously differentiable, that is all derivatives
exist and continuous, on a piecewise smooth bounded oriented (with a unit normal vector
n) two-dimensional manifold (surface) S in R3 with piecewise smooth boundary ∂S then
Z I
∇ × F · dS = F · dr
S ∂S

The curve of the line integral, ∂S, must have positive orientation.

(MA3FM/MA4FM) Autumn 2020

Chapter 11

Self-study.

11.1 Summation convention: Ein-

J
Self-study 11.0.1.
stein notation.
To simplify notations, we shall adopt the summation convention - sometimes called
the Einstein notation. That is, when an index (a particular suffix, for example i or j)
is repeated, that is it appears twice in a single term, then summation over that
index is implied. For example

3
X
aii = aii = a11 + a22 + a33 . (11.1)
i=1
X3
a1j xj = a1j xj = a11 x1 + a12 x2 + a13 x3 . (11.2)
j=1
3 X
X 3
aij aij = aij aij , (11.3)
i=1 j=1
3
X
= (ai1 ai1 + ai2 ai2 + ai3 ai3 ),
i=1
= a211 + a212 + a213 + a221 + a222 + a223 + a231 + a232 + a233 .

In
P these examples you may observe repeated indices i and j and we have simply skipped
sign in the LHS using the Einstein convention.
Note:

156
Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 157

• We are going to work in three-dimensional, 3-D, space, so all indices run

from 1 to 3, for example 1 ≤ i ≤ 3.

• The index over which summation occurs R 1 is a dummy suffix. You can think
1
here, by analogy, of definite integrals, 0 x dx = 2 , for instance, where the result
does not depend on x, which is a ’dummy’ variable. This means, that the result
of evaluating the expression does not contain the dummy indices explicitly. For
example, in a scalar product equation

a · b + c · d = ai bi + cj dj = ak bk + cm dm

so that the dummy indices can be anything you like as far as they do not clash
with a free index, see below.

• Be careful not to accidentally use the same indices more than twice. An index
symbol may not appear more than twice within a multiplicative single expres-
sion. For example, consider a vector equation

y = (a · b)x.

If we represent the above relationship in the index form as

yi = ai bi xi

the result is ambiguous, so that the correct way to represent it is the following

y i = ak b k x i .

Here, index k is the dummy index, while i is the so called free index.

• If an index appears only once, it is called a free index: it is free to take any
value within the range (between 1 and 3 in our case), and the equation must be
valid for all values. For example, the vector equation

A=B

implies
Ai = Bi .

• Free index always appears in the resultant expression, for example index i,

Aij Bj = Ci .
Apparently, the above equation may be different at i = 1 or i = 2 or i = 3,
so that the result does explicitly depend on i. In fact, we have three different
equations.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 158

Also free indices should match in every term. For example, in the equation

ak bk xl = zl + aj aj bm bm xn xn xl

we have one free index l, which appears in both the LHS and the RHS in every
term, and the remaining indices are dummy ones. In vector notation, the above
relation is
(a · b)x = z + |a|2 |b|2 |x|2 x.

So, to summarize, we can only have two types of indices: one that appears
precisely once, the free index and one that appears precisely twice, which is a
repetitive index.
Note, in terms of notations, one can not write x = xj , since the LHS is a
vector and the RHS is scalar, a component of the vector x. But, one can write
[x]j = xj or x = (xj ).

11.2 Examples of some practical use of indices,

useful (later) symbols and exercises.

1. The dot product of two vectors A and B, as we have already seen, can be written
using the summation convention

3
X
A · B = Ak Bk = Ak Bk = A1 B1 + A2 B2 + A3 B3
k=1

2. The Kronecker Delta is defined to be

(
1 if i = j,
δij =
0 6 j.
if i =

Then one can show that

(a) δii = 3
Complete the example:

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 159

(b) aij δij = akk

Complete the example:

(c) aij δjk = aik

Complete the example:

(d) δij δij = 3.

Complete the example (hint we know already that δkk = 3):

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 160

3. In a vector form we have the following equation

3x + 5y = z.

Write it using the index notation

4. In a vector form we have the following equation

(x · y)(z · B) = 0.

Write it using the index notation

5. The Levi-Civita symbol or alternating tensor εijk is defined using cyclic

permutations by ε123 = ε231 = ε312 = 1, ε132 = ε213 = ε321 = −1 and 0 otherwise
(that is when two indices are the same). Note, that cyclic permutations (or
even permutations) are 123, 231, and 312; anti - cyclic (or odd) permutations
are 132, 213, and 321.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 161

The vector or cross product then can be presented as

u × v = εnjk uj vk in ,
or
[u × v]n = εnjk uj vk
where i1 , i2 , i3 are the unit vectors in the coordinate directions.
Show that the vector product can be presented in this way (hint, consider just
one component first):

6. For three vectors A, B and C prove using the result that

εijk εmnk = δim δjn − δin δjm

that

A × (B × C) = (A · C)B − (A · B)C :

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 162

∂ui ∂
7. If u is a vector function then div(u) = ∇ · (u1 , u2 , u3 ) = . Here ∇ = ik
∂xi ∂xk
is a vector operator, known as the Del-operator or nabla.
Write the above explicitly

∂φ
8. If φ is a scalar function then grad φ = ∇φ = ik ,
∂xk
Write the above explicitly

∂ 2φ ∂ 2φ
9. If φ is a scalar function then ∇2 φ = ∆φ = = .
∂xi ∂xi ∂x2i
Write the above explicitly

10. From the view point of the new notations (summation convention), state which
of the following expressions are well formed.

(a) aij = bi bj
(b) aii = cjj
(c) aii cjj
(d) bi ∂c
∂t
i
= bm bm
(e) aij cjj .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 163

11. Show that the matrix components (bij ) formed by

1
bij = aij − akk δij
3

are such that bii = 0.

12. The matrix product BC is defined only when the number of columns in B equals
the number of rows in C and is given by bpr crq . If the number of columns in
A and the number of rows in C equals the number of rows in BC, express the
following in the index form

(a) A(BC),

(b) C T (BC).

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 164

13. Show that

∂uk
(a) (curl u) = εijk ii ,
∂xj

(b) curl grad φ = 0,

(c) div curl u = 0.

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 165

14. Show by use of the result

εijk εmnk = δim δjn − δin δjm

that

(a) curl curl u = grad div u − ∇2 u

(b) curl u × u = ((∇ × u) × u) = (u∇)u − 12 ∇u2 .

(MA3FM/MA4FM) Autumn 2020

Alex Lukyanov A Mathematical Introduction to Fluid Mechanics and Its ... 166

(MA3FM/MA4FM) Autumn 2020