Lecture Notes On Fluids May 2013 PDF

LECTURE NOTES ON NEWTONIAN AND
RELATIVISTIC FLUIDS
1
Michael Tsamparlis
Athens May 2013
1 Email: mtsampa@phys.uoa.gr
2
Contents
1 The physical system continuum 1
2 A short detour in vector analysis 3

2.0.1 Tensor formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 The kinematics of a Newtonian continuum . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 The relative velocity in vector formulation . . . . . . . . . . . . . . . . . 5
2.1.2 The relative velocity in tensor notation . . . . . . . . . . . . . . . . . . . 7
2.1.3 The kinematic interpretation of relative velocity . . . . . . . . . . . . . . 8
2.1.4 The kinematic interpretation of (2.29) in terms of space intervals . . . . . 9
3 Motion of a continuum and geometry 13

3.1 The geometrization of rigid motion . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.1 The geometrization of strain motion . . . . . . . . . . . . . . . . . . . . 15
3.1.2 A geometric computation of the strain metric . . . . . . . . . . . . . . . 16
4 The stress tensor 19

4.0.3 Body and Surface forces . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.0.4 The perfect deformable continuum . . . . . . . . . . . . . . . . . . . . . 21
4.0.5 The imperfect deformable continuum . . . . . . . . . . . . . . . . . . . . 22
4.0.6 Types of strain and stresses . . . . . . . . . . . . . . . . . . . . . . . . . 24
5 The stress - strain relation for elastic continua 27

5.0.7 The stress rate - strain relation for a linear elastic isotropic continuum . 28
5.0.8 The coefficients n, k of a linear elastic isotropic continuum . . . . . . . . 30
5.0.9 A second derivation of the rate of strain stress relation from Hooke’s law 31
6 The dynamics of strain motion 35

6.0.10 The force due to a stress tensor . . . . . . . . . . . . . . . . . . . . . . . 35
6.0.11 Newton’s second law in terms of the displacement vector . . . . . . . . . 36
6.0.12 Kinematic interpretation of the equation of motion of the strain vector
in a linear elastic isotropic continuum . . . . . . . . . . . . . . . . . . . . 37
6.0.13 The dynamic equation of motion for a linear elastic isotropic continuum 40
7 Worked examples on the stress tensor 43
v
vi CONTENTS
8 Fluids 49
8.1 Mathematical concepts relevant to fluids . . . . . . . . . . . . . . . . . . . . . . 50
9 Point Transformations defined by vector fields 53

9.0.1 Flow of a vector field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
9.0.2 The point transformation induced by a vector field . . . . . . . . . . . . 54
9.0.3 The exponential map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
9.0.4 The coordinate transformation defined by the induced transformation . . 57
9.0.5 The transformation dragging along . . . . . . . . . . . . . . . . . . . . . 58
9.1 The dragging along the velocity field of a fluid . . . . . . . . . . . . . . . . . . . 59
9.1.1 Incompressible flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
9.1.2 Conservation of mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
9.2 The transport theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
9.3 Integral properties of vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . 67
9.3.1 Flux through a surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
9.3.2 Flow and circulation along a path . . . . . . . . . . . . . . . . . . . . . . 67
10 Characterization of fluids and flows 69

10.0.3 Geometric description of the flow of a fluid . . . . . . . . . . . . . . . . . 70
10.0.4 Calculating the streamlines . . . . . . . . . . . . . . . . . . . . . . . . . 70
11 The equation of continuity - conservation of mass (again) 73

11.0.5 The kinetic energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
12 The equation of motion of a perfect fluid 77

12.0.6 Euler equation for special types of incompressible flow . . . . . . . . . . 78
13 Thermodynamics and hydrodynamics 83

13.0.7 Isentropic flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
14 The Navier Stokes equation 89
15 The electromagnetic field as a viscous fluid 93

15.0.8 The stress tensor of the electromagnetic fluid . . . . . . . . . . . . . . . . 93
15.0.9 Poynting’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
15.0.10 The electromagnetic field as a viscous fluid . . . . . . . . . . . . . . . . . 97
16 Relativistic fluids 105

16.1 Fluids in Special Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
16.1.1 Relativistic fluids in Special Relativity . . . . . . . . . . . . . . . . . . . 106
16.2 The energy momentum tensor in Special Relativity . . . . . . . . . . . . . . . . 108
16.2.1 The 1+3 decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
16.2.2 The bivector metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
16.2.3 The 1+3 decomposition wrt the four-velocity ua . . . . . . . . . . . . . 113
16.2.4 The energy momentum tensor . . . . . . . . . . . . . . . . . . . . . . . . 114
16.3 The case of a single particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
CONTENTS vii
16.3.1 The case of a single particle under the action of forces . . . . . . . . . . . 119
16.4 The equations of motion of a relativistic fluid . . . . . . . . . . . . . . . . . . . 120
16.4.1 The relativistic fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
16.5 The dynamical equations of motion of a relativistic fluid: A simplified approach 123
16.6 The energy momentum tensor of a relativistic viscous fluid . . . . . . . . . . . . 125
16.7 The energy momentum tensor of the electromagnetic field . . . . . . . . . . . . . 128
16.8 The Minkowski energy momentum tensor . . . . . . . . . . . . . . . . . . . . . 132
16.8.1 The 1+3 decomposition of the EM T ab . . . . . . . . . . . . . . . . . . . . 134
17 Classical Differential Operators 135

17.0.2 The differential operators . . . . . . . . . . . . . . . . . . . . . . . . . . 136
18 Problems on Newtonian fluids 139

viii CONTENTS
Chapter 1
The physical system continuum
We consider an aggregate of particles which we assume that occupies a connected region in

space. These particles constitute the physical system which we call a continuum. In order to
study such systems we introduce physical quantities, which characterize one or more physical
aspects of the system. According to the covariance principle, these physical quantities must
A. Be described mathematically by tensor fields whose tensor character (i.e. the Jacobian
of the transformation) is defined by the invariance group of the theory of Physics with which
the continuum will be studied
b. The physical laws of the theory relating these physical quantities must involve only this
type of tensors.
The study of a continuum (and a physical system in general) is done at two levels. The
first level is the Kinematics, which concerns the description of the orbits of the particles of
the system in the space considered by the theory. The second level is the Dynamics, which
studies the evolution of a physical system in a given environment. Kinematics is predecessor
to Dynamics, because it concerns the basic assumptions of the theory e.g. the geometry of the
background space (the metric), the covariance group etc.
In the following we start our discussion with Kinematics. Mathematically the orbit of a
particle is a parametrized curve where the parameter is the ‘proper’ time of the particle. A
parametrized curve can be described by its first and second derivative, therefore the Kinematics
of a continuum involves the velocity and its first derivative1 of every and all of its particles. It is
therefore postulated that the kinematic description of the continuum involves two differentiable
vector fields in the region occupied by the continuum:
a. The velocity vector field uµ
b. The derivative vector field uµ,ν .
The Kinematics of a continuum is precisely the geometric study of these two vector fields.
The study of a vector field is done in various ways. The one closer to geometry is by means
of the study of its flow. The flow of a (differentiable) vector field is the congruence of all
integral curves of the vector field. These curves are parametrized with the arc length s (it can
be any other parameter the scenario stays the same) and every member of the congruence can
be characterized with n quantities, where n is the dimension of the space where the orbits are
traced. To understand the role of n we consider a point P in space and the integral curve
1
By derivative we mean all derivatives, i.e. if uµ is the velocity field we mean the uµ,ν which relates different
trajectories (particles) and not the acceleration which is the derivative wrt the parameter along the trajectories.
1
2 CHAPTER 1. THE PHYSICAL SYSTEM CONTINUUM
(the orbit) through P . Given the first and the second derivative of the curve at P one is able
to compute (i.e. trace) the curve near P. Therefore one can characterize the curve (orbit) by
the coordinates of P and the (affine) parameter along the curve. If in a coordinate system
the coordinates of P are y a a = 1, 2, ..., n then the set (s, y a ) characterizes the curve of the
congruence through P . However a single curve does not give information about the continuum
and one has to consider the comparative study of two curves. This study is realized by means
of relative motion as follows.
Consider the curve through the point P and a second integral curve through a point Q near2
the point P. Choose the parameter along the second curve so that at the points P, Q the value
of the parameter along each curve is the same and consider the vector P Qi joining the points
P, Q. The vector P Qi we call the connecting vector at P. Then define a vector field along
the integral curve through P by joining points along these integral curves always for the same
value of the parameter. This procedure defines a ‘transportation’ law of the connecting vector
along the integral curve through P, which we call Lie transportation along the tangent
vector of that curve. Because P is arbitrary this vector field defines a new vector field within
the continuum, which we call the connecting vector field.
In the following we study the continuum using Newtonian Physics. This means that the
orbits of the particles of the system are in the three dimensional Euclidian space E 3 where the
motion occurs. The transformation group is the Galilean (or Euclidian group) group, which is
the covariance group of Newtonian Physics. According to the covariance principle, the tensors
which will be used to describe the physical quantities are Euclidian tensors. Therefore in
Newtonian Kinematics the velocity field, the acceleration field and the connecting vector field
will be Euclidian vector fields.
2
The word near has to be defined. This we shall do as we go on
Chapter 2
A short detour in vector analysis
In the discussion to follow we shall need material from vector analysis. In the current section
we discuss this material using the language of the standard vector analysis and employ tensor
formulation, which greatly simplifies the discussion. The definitions which follow are not re-
stricted to a three dimensional Euclidian vector space and apply to an n−dimensional (real)
vector space.
Vector formulation
Consider a general vector function f (x, y, z) defined over a connected region of R3 . We define
the directional derivative (u · ▽)f of f (x, y, z) along an arbitrary vector u by the relation 1 :
[ ]
∂ ∂ ∂
(u · ▽)f ≡ ux + uy + uz f. (2.1)
∂x ∂y ∂z
For u = r and any vector u we have the identity:
(u · ▽)r = u. (2.2)
For later use we note the obvious result:
▽ × r = 0. (2.3)
Next we discuss the decomposition of an arbitrary vector u relative to the position vector
r. As it will be shown this decomposition is useful in the study of relative motion.
We start from the well known identity of vector calculus:
▽(u · v) = u × (▽ × v) + v × (▽ × u) + (u · ▽)v + (v · ▽)u (2.4)
and take v = r to find:
▽(u · r) = u × (▽ × r) + r × (▽ × u) + (u · ▽)r + (r · ▽)u. (2.5)

1
This formula is valid in Cartesian coordinates {x, y, z} only. In other coordinates cylindrical, spherical etc.
the rhs has to change accordingly
3
4 CHAPTER 2. A SHORT DETOUR IN VECTOR ANALYSIS
The first term vanishes by (2.3). The term (u · ▽)r = u by (2.2). Therefore (2.5) gives:
u= ▽ (u · r) + (▽ × u) × r − (r · ▽)u. (2.6)
This relation decomposes an arbitrary vector u into parts relative to the position vector r.
We find a new identity if we take u = v in (2.4):
▽v2 = 2v × (▽ × v) + 2(v · ▽)v (2.7)
or, solving for the directional derivative (v · ▽)v:

1
(v · ▽)v = ▽v2 − v × (▽ × v). (2.8)
2
Consider a general vector function f (xi ) i = 1, ..., n defined over a connected region of Rn .
Then the differential df of f is defined as follows:
[ n ]
∂f ∂f ∑ n
∂f ∑ ∂
df = 1 dx1 + 2 dx2 + ... = i
dxi = dxi i f ≡ (dr · ▽)f (2.9)
∂x ∂x i=1
∂x i=1
∂x
that is, the differential of f (xi ) coincides with the directional derivative of f (xi ) along the
vector dr =(dx1 , dx2 , ..., dxn ).
2.0.1 Tensor formulation

We write the above formulae using tensor (i.e. index) notation. We use Greek indices and
assume that they take the values 1, 2, 3.
The exterior product u × v in tensor formalism is written as2 :
(u × v)ρ = ερµν uµ vν (2.10)
where ερµν is the well known completely antisymmetric tensor (or Levi Civita tensor)3 . Then
(2.3) follows trivially4 form ερµν xµ,ν = 0.
The inner product u · v is written as:
u · v =uµ vµ (2.11)
The directional derivative is the vector (xµ ∂µ )f ρ .

Concerning the identity (2.4) we have:
(δ µν uµ vν ),κ = δ µν (uµ ,κ vν + uµ vν,κ )

= δ µν (δµρ δκσ − δκρ δµσ )uρ,σ vν + δ µν (δνρ δκσ − δκρ δνσ )vρ,σ uν + v ν uκ,ν + uν vκ,ν .
2
Einstein convention applies to all our tensor notation.
3
We recall the identities (Greek indices take the values 1,2,3):
µν
εµνρ εκλρ = δκλ = δkµ δλv − δkν δλµ
µνρ ρ
ε εµνλ = δλ
εµνρ εµνρ = 1
εµνρ Aµ B ν C ρ = det(A, B, C).
4
Because xµ,ν is symmetric in µ, ν.
2.1. THE KINEMATICS OF A NEWTONIAN CONTINUUM 5
The term:
δ µν (δµρ δκσ − δκρ δµσ )uρ,σ vν = δ µν ελµκ ελρσ uρ,σ vν = −εκλµ ελρσ uρ,σ v µ = εκµλ ελρσ uρ,σ v µ
therefore:
(δ µν uµ vν ),κ = εκµλ ελρσ uρ,σ v µ + εκµλ ελρσ vρ,σ uµ + v ν uκ,ν + uν vκ,ν (2.12)
To show that this expression is indeed identity (2.4) we note that in standard vector notation
the terms:
εκµλ ελρσ uρ,σ v µ = εκµλ (▽ × u)λ v µ = [v × (▽ × u)]κ
v µ uκ,µ = (v · ▽)u
from which identity (2.4) follows. Note that for v µ =uµ this reduces to identity (2.8), which
in tensor notation reads:
(uµ uµ ),κ = εκµλ ελρσ uρ,σ uµ + 2uµ uκ,µ . (2.13)
Finally for the differential of functions we have:
df ρ = f,µ
ρ
dxµ . (2.14)
2.1 The kinematics of a Newtonian continuum

2.1.1 The relative velocity in vector formulation
Let P be a point in the linear space R3 occupied by the continuum and choose the coordinate
system Σ in that region to have origin at the point P. Let Q be another point of the continuum
near P, where by near5 we understand that the position vector r =xi+yj+zk of Q in Σ is such
that squares x2 , y 2 , z 2 of the coordinates are ignorable. Let v be the velocity vector field of the
continuum and suppose that the value of the field at the points P, Q of the continuum are vP
and vQ respectively. Because the vector field is a differentiable vector field we have:
vQ = vP + dvP (2.15)
where the quantity dvP is the differential of the velocity field at the point P. The quantity dvP
we call the relative velocity of Q relative to P. According to (2.9) the differential dvP of a
velocity function vP is given by the relation:
dvP = (r · ▽)vP (2.16)
where we have replaced dr with r due to the nearness hypothesis6 we have made.
5
The assumption of nearness is necessary because otherwise there is no linear connection between the veloc-
ities of the points P, Q. Here we have the first requirement for the definition of nearness.
6
The replacement of dr with r means that we linearize the problem. That is relation (2.16) defines the dvP !
The continua which satisfy this requirement could be and should be named Linear Continua. Not all continua
are expected to be linear! In general we should write
dvP = (r · ▽)vP + O(r2 ).
We decompose dvP along r using the general identity (2.6) and write:
dvP = ▽ (dvP ·r) + (▽×dvP ) × r − (r · ▽)dvP . (2.17)
The term7 :
(r · ▽)dvP = (r · ▽)(r · ▽)vP = (r · ▽)vP = dvP (2.18)
and equation (2.17) becomes:
1 1
dvP = ▽ (dvP ·r) + (▽×dvP ) × r. (2.19)
2 2
Replacing in (2.15) we find8 :
1 1
vQ = vP + (▽×dvP ) × r+ ▽ (r·dvP ). (2.27)
2 2
7
It is easy to prove that (r · ▽)(r · ▽) = (r · ▽) by writing (r · ▽) = x ∂x
∂ ∂
+ y ∂y ∂
+ z ∂z . A shorter proof is the
following. We write x ∂µ for the term r · ▽ and have:
µ
(xµ ∂µ )(xν ∂ν )v ρ = (xµ δµν ∂ν )v ρ + xµ xν ∂µ ∂ν v ρ = (xν ∂ν )v ρ + O(xµ xν )

.
8
The study of relative motion of particles in a fluid is important from the physical point of view. Therefore
we give a second equivalent and less formal derivation.
We consider again the time moment t1 two points P, Q with position vectors r and r + δr where δr is an
infinitesimal quantity (so that its squares are neglected). We also consider the velocities of the points P and
Q the time moment t1 to be v and v + δv respectively. We note that the points P, Q are kept fixed (that is
we ”freeze” the fluid at t1 ) therefore the vectors r and r + δr as well as v and v + δv are independent of time.
This is the reason we use δr,δv to specify the position and the velocity of Q and not dr,dv.
We consider a coordinate frame Σ in which the vectors r =xi+yj+zk, v =ui+vj+wk where u(t = t1 , r),
v(t = t1 , r), w(t = t1 , r). In the same frame the vectors δr and δv are decomposed as follows:
δr = δxi + δyj + δzk, δv = δui + δvj + δwk
∂u
where δu = ∂r δr = ∂u
∂x δx +
∂u
∂y δy + ∂u
∂z δz and similarly for δv(t, r), δw(t, r). We note that we can write δu as
follows:
[ ( ) ( )] [ ( ) ( )]
∂u 1 ∂u ∂v 1 ∂v ∂u 1 ∂u ∂w 1 ∂w ∂u
δu = δx + + − − δy + + + − + δz
∂x 2 ∂y ∂x 2 ∂x ∂y 2 ∂z ∂x 2 ∂x ∂z
= exx δx + (exy − ωz )δy + (exz + ωy )δz
where
( ) ( )
∂u 1 ∂u ∂v 1 ∂u ∂w
exx = , exy = + , exz = + (2.20)
∂x 2 ∂y ∂x 2 ∂z ∂x
( ) ( )
1 ∂w ∂u 1 ∂v ∂u
ωy = − + , ωz = − (2.21)
2 ∂x ∂z 2 ∂x ∂y
It is easy to see that ωy = 12 (▽ × v) · j, ωz = 12 (▽ × v) · k. If we define the quantities:

δuS = exx δx + exy δy + exz δz, δuR = i · (ω × r) (2.22)
we have:
δu = δuS + δuR . (2.23)
The first part δuS is the contribution to δu form the local rate of change of the shape of the fluid element
whereas the second part δuR is the contribution due to the local rotation of the fluid element with angular
velocity ω without any change of shape (that is, a rigid body motion).
This is an equation where all terms are defined at the point P except the ”free” variables”
r, dvP which are used to ”measure” the relative velocity and the “nearness” of the points around
P. This equation is in vector form, that is, the same for all Newtonian coordinate systems.
We introduce the vorticity vector ω defined as follows:
1
ω = (▽×dvP ). (2.28)
2
and the velocity of the point Q is written:
1
vQ = vP + ω×r+ ▽ (r·dvP ). (2.29)
2
2.1.2 The relative velocity in tensor notation

The velocity field at Q is:
ρ
vQ = vPρ + dvPρ (2.30)
Assuming that the continuum is a linear continuum (see last footnote) we define / write:
dvPρ = vPρ,ν dxν (2.31)
Then we have:
ρ
vQ = vPρ + dvPρ = vPρ + vPρ,ν dxν
[ρ,ν] (ρ,ν)
= vPρ + vP dxν + vP dxν
(ρ,ν)
= vPρ + (δσρ δλν − δλρ δσν )vPσ,λ dxν + vP dxν
(ρ,ν)
= vPρ + εκσλ εκρν vPσ,λ dxν + vP dxν
where vP ν,ρ is the derivative of the velocity field at the point P. We define the vorticity tensor:
ω ρν = εκσλ εκρν vPσ,λ (2.32)
and get:
ρ (ρ,ν)
vQ = vPρ + ω ρν dxν + vP dxν . (2.33)
If we repeat the same calculations for the other two components δv and δw of the variation δv we find that
the change in the relative velocity δv:
δv = δvS + δvR (2.24)
where:   
exx exy exz δx
δvS =  exy eyy eyz   δy  (2.25)
exz eyz ezz δz
and
δvR = ω × r. (2.26)
We note that the variation in the shape is measured by the tensor sµν with components given by the matrix
(eµν ) above. The quantity sµν is not a vector because if we rotate the axes by 1800 it is not effected, whereas
if it was a vector it would change sign. δvS is the strain tensor.
Next we define the symmetric (0,2) tensor:

(ρ,ν)
eρν = vP (2.34)
and obtain the final answer:

ρ
vQ = vPρ + ω ρν dxν + eρν dxν . (2.35)
With the antisymmetric tensor ωµν we associate a vector ω µ , called the vorticity vector as
follows:
1
ω µ = εµνρ ωνρ ⇒ ω ρν = ερνσ ωσ . (2.36)
2
In terms of the vorticity vector the relative velocity is written as follows:
ρ
vQ = vPρ + ερνσ ωσ dxν + eρν dxν . (2.37)
2.1.3 The kinematic interpretation of relative velocity

We consider equation (2.29) and discuss its physical meaning.
Equation (2.29) implies that the velocity of the arbitrary point Q near the point P can be
described by the sum of three terms:
• The velocity vP which corresponds to a motion in which all points (including P ) are
displaced by vP dt, the same for all points near P. This type of motion we call translation.
• The velocity ω×r which corresponds to a motion in which P and the points along the line
parallel to ω and passing through P remain fixed. This type of motion we call rotation
about the axis ω with angular speed ω.
• The velocity 12 ▽ (r·dvP ) which represents a motion in which P is the only point which
remains fixed. This type of motion we call strain.
The first two motions are defined in terms of the vectors vP , ω and the last motion in terms
of the quantity dvP ·r , which we analyze below.
A motion which is either a translation, a rotation or a combination of the two we call a
rigid motion.
Aggregates of particles (continua) which can move only with rigid motion we call rigid bod-
ies or solids. All other aggregates of particles we call deformable continua. A deformable
continuum is possible under certain conditions to move like a rigid body, but in general it
does not do so. A special type of deformable bodies are the fluids, which are defined as the
deformable bodies, which is possible to sustain strain motion only.
We note that the motion strain is possible only when the material is deformable (that is
different points can have different velocities). This is expected because we never consider the
liquids to be solids!
We picture equation (2.29) by considering the three types of an aggregate of particles en-
closed in a spherical surface (see Figure 2.1).
Explanation of Figure
r′
s s′
Figure 2.1: The geometric description of a small displacement
Translation is a motion of all the points of the sphere (including its center) with no change
in shape or in volume. Note that in this type of motion the distance between any two points
of the continuum remains constant during the motion.
Rotation is a motion in which the center of the sphere and all points along the diameter
parallel to the direction of ω (the axis of rotation) remain fixed, while the rest of the points
of the sphere change position in such a manner that their distance from the axis of rotation is
constant. For this type of motion there is no change in the shape or the volume of the sphere.
We also note that, as it is the case with the translation, the distance between any two points
of the continuum remains constant during this type of motion.
The third type of motion is one for which only the center of the sphere remains fixed and
the shape of the sphere changes e.g. to a (rotational) ellipsoid. For this type of motion the
distance of all points of the continuum changes.
We draw the following important conclusions
1. Equation (2.29) is covariant (within the Euclidian 3-space!) therefore it describes an
intrinsic property of the continuum independent of the particular observer.
2. In the above approach we classify the deformable bodies not as objects but according to
the type of motion they are able to sustain:
a. If they can sustain rigid motions only, we call them rigid bodies
b. If they can sustain strain motion only, we call them fluids and
c. If they can sustain all types of motion, we call them deformable bodies.
In other words it is the type of motion sustained by an aggregate of particles (a continuum)

which qualifies the physical character of the physical system and not its material substance.
2.1.4 The kinematic interpretation of (2.29) in terms of space inter-

vals
Vector notation
We discuss the kinematic interpretation of (2.29) in terms of space intervals instead of relative
velocities. The reason for doing this is because we want to relate the kinematic properties of
the deformable body with the geometry of the space and another “internal” geometry defined
by each specific deformable body.
We consider two nearby9 points P, Q in the (linear) deformable medium and assume that
the body suffers a small deformation. Due to the deformation the points P, Q suffer a small
displacement to the points P ′ , Q′ respectively. Let s = PP′ , s′ = QQ′ be the displacement
vectors of the points P, Q. Our purpose is to calculate the new vector P′ Q′ ≡ r′ in terms of the
original vector PQ = r and the displacement vector s.
r′
s s′
Figure 2.2: The geometric description of a small displacement
We assume that the deformation of the body takes place in time dt with the velocities of
the points P, Q being vP and vQ respectively. We have then:
s′ = vQ dt, s = vP dt.
Replacing vQ from (2.29) we find10 :

1 1
s′ = s + (▽×dvP dt) × r+ ▽ (r·dvP dt). (2.38)
2 2
The linearity of the space of Figure 2.2 gives:
1 1
r′ = r + s′ −s = r + (▽×dvP )dt×r+ ▽ (r · dvP )dt. (2.39)
2 2
The first two parts correspond to a rigid body motion. Indeed the first term is a translation
by r. Concerning the second term we note that it is written as ω × rdt where ω is the vorticity
vector, hence this term encounters rotation. The strain motion is described by the third term
1
2
▽ (r · dvP )dt.
Next we consider the strain motion.
After a period of time dt the point Q has moved a distance dvP dt. Due to the differentia-
bility of the velocity and other fields associated with the continuum, this distance must be the
differential (i.e. the s′ − s) of the change s(rP ) suffered by the point P. Therefore we have the
condition11 :
dvP dt=ds(rP )
from which follows r·dvP dt= r·ds(rP ).
9
Nearness is understood as in the last section. That is, P is the origin of the coordinates and the position
vector rQ of the point Q has components rQ = xi + yj+zk where the quantities x, y, z are small in the sense
that their mutual products can be neglected.
10
Do not overlook the fact that we consider linear deformable bodies only!
11
This condition defines the linear deformable body at the level of relative position vectors, thus completing
the definition we gave previously in terms of the relative velocity.
To compute the strain motion explicitly we consider a coordinate frame, the Σ say, in which
we assume that r = (x, y, z) , vP . = (u, v, w). Then from (2.9) we have (if we replace r with dr
due to our linear approximation):
∂vP
dvP = r·
∂r
i.e. the change of the velocity in the direction of r. The inner product:
( )
∂vP
r·dvP = r· r·
∂r
[ ] [ ] [ ]
∂u ∂v ∂w
= x r· + y r· + z r·
∂r P ∂r P ∂r P

∂u ∂v ∂l
= x2 +xy +xz
∂x P ∂x P ∂x P

∂u
2 ∂v ∂l
+ yx +y +yz
∂y P ∂y P ∂y P

∂u
∂v
2 ∂l
+ zx +zy +z
∂z P ∂z P ∂z P
from which follows:
[ ( ) ( )]
1 ∂u y ∂u ∂v z ∂w ∂u
∇(r · ds(rP )) = x + + + + i
2 ∂x P 2 ∂y P ∂x P 2 ∂x P ∂z P
[ ( ) ( )]
x ∂u ∂v ∂v z ∂w ∂v
+ + +y + + j
2 ∂y P ∂x P ∂y P 2 ∂y P ∂z P
[ ( ) ( ) ]
x ∂w ∂u y ∂w ∂v ∂w
+ + + + +z k
2 ∂x P ∂z P 2 ∂y P ∂z P ∂z P
≡ eµν x j
where xi = (x, y, z) and we have introduced the strain tensor:

 ( ) ( ) 

∂u 1

∂u

∂v 1

∂w

∂u
 ∂x 2 ∂y + ∂x 2 ∂x + ∂z 
 (P ) P P ( P P ) 
 1 ∂u 
[eµν ] =  + ∂v ∂v 1 ∂w
+ ∂v . (2.40)
 2 ∂y ∂x ∂y 2 ∂y ∂z 
 ( P P ) (P ) P P 
 1 ∂w 
+ ∂u 1 ∂w
+ ∂v ∂w
2 ∂x ∂z 2 ∂y ∂z ∂z
P P P P P
An equivalent way to look upon the matrix [eij ] is the strain ellipsoid defined by the
equation:
( ) ( ) ( )
∂u 2 ∂v 2 ∂w 2 ∂u ∂v ∂w ∂u ∂w ∂v
x+ y + z + + xy+ + zx+ + yz = 1.
∂x P ∂y P ∂z P ∂y P ∂x P ∂x P ∂z P ∂y P ∂z P
(2.41)
We note that the strain tensor has dimensions [L].
For a Newtonian deformable body the strain tensor defines a positive definite quadratic
form, therefore it is always possible to choose at the point P the coordinate frame in such a
way that the matrix [eµν ] becomes diagonal. These axes correspond to the principal axes of the
strain ellipsoid and are called the principal directions of the strain ellipsoid. In this frame
along the frame axes (only!) the strain motion is a pure translation (the simplest possible!). The
translations are in general different along each principal direction, so that [eµν ] = diag(e1, e2 , e3 )
with e1, e2 , e3 different to each other. Finally we note the useful relation:
∑ ∑
T r[eµν ] = eii = si,i = ∇s. (2.42)
µ µ
Tensor notation
In the case of tensor notation we have to consider equation (2.37), i.e.:
ρ
vQ = vPρ + ερνσ ωσ dxν + eρν dxν . (2.43)
Again we recognize the translational motion (term vPρ ), the rotational mention (term ερνσ ωσ dxν )
and the strain motion (term eρν ). If we introduce the vector si = (u, v, w) = vP dt the covariant
form of the strain tensor is (compare with (2.34) ):
( ) ( )
1 ∂sµ ∂sν 1 ∂uµ ∂uν
eµν dt = + = + dt. (2.44)
2 ∂xν P ∂xµ P 2 ∂xν P ∂xµ P

∂uµ
that is, the strain tensor at P is defined by the symmetric part of the derivative ∂xν . We
P

note that the antisymmetric part of the derivative ∂u
µ
at P defines the vorticity vector ω µ .
∂xν
P
The quantity we shall use is not the strain tensor but the rate of strain tensor ẽij defined
by the relation eij = ẽij dt. Because si = ui dt where ui is the velocity of the continuum at the
position P we have: ( )
1 ∂uµ ∂uν
ẽµν = + . (2.45)
2 ∂xν P ∂xµ P
Chapter 3
The motion of a deformable body in

terms of geometry
In the previous section we considered the relative motion of a (linear) deformable body from
a kinematical point of view. We have shown that the motion consists of two different types of
motion (a) rigid body motion and (b) strain motion. In the following we relate these types of
motion with a metric and in this manner we “geometrize” the general motion of a deformable
body. We consider first the rigid body motion.
3.1 The geometrization of rigid motion

Consider the two points P, Q in the deformable body whose position vectors at some instant1
t are r and r + dr respectively. Suppose that after the action of some forces the body deforms
so that the points P, Q become P ′ , Q′ . The linearity of space gives P′ Q′ = PQ + QQ′ −PP′ or,
following the notation of the last section r′ = r + s′ −s. For rigid motion we have that2 dvP ⊥ r
(instantaneous rotation about the point P ) and form equation (2.39) we have3 :
r′ = r + s′ −s = r + ω×rdt. (3.1)
We compute the change in the magnitude of the displacement vector r. We have4 :

[ ]
r′2 = r2 + (ω×r)·(ω×r)dt2 +2r · (ω×r) = r2 + ω 2 r2 − (ω · r)2 dt2 = r2 (1 − ω 2 sin2 θdt2 ).
1
It is important that t is common for both points. In Relativity this is not possible!
2
In the following we drop the index R from sP because P is arbitrary.
3
We omit the term 21 ▽(r · wdt) because we are considering a rigid body for which the strain motion vanishes.
4
The proof of this results is as follows. The term r · (ω×r) =0 because r is normal to (ω × r). From the
identity of vector calculus
(A×B)·(C × D) =(A·C)·(B · D)−(A·D)·(B · C)
we have:
2 2
(ω×r)·(ω×r) = ω r2 − (ω · r) = r2 (1 − ω 2 sin2 θ) = ω 2 r2 cos2 θ.
Therefore:
r′2 = r2 + ω 2 r2 − (ω · r) = r2 (1 − ω 2 sin2 θ)
2
13
14 CHAPTER 3. MOTION OF A CONTINUUM AND GEOMETRY
The r2 is infinitesimal (recall that with r stands for dr because we consider infinitesimal motion
of a linear deformable body) therefore the term ω 2 r2 sin2 θdt2 is of fourth order and can be
neglected. This implies:
r′2 = r2
that is, during a rigid motion the relative (Euclidean) distance of nearby points in the de-
formable body remains invariant. We conclude then that:
Rigid motion is generated by a Euclidian isometry, that is by the Killing vectors of

the Euclidean metric.
From Geometry we know that the (continuous) Euclidean isometries constitute a six di-
mensional Lie group, the latter consisting of a closed three dimensional Abelian subgroup, the
subgroup of translations T (3) and a closed three dimensional subgroup, the subgroup of rota-
tions O(3). This implies that we can understand the motion of a rigid body as a combination
of two types of different transformations: Translations (corresponding to T (3)) and rotations
(corresponding to O(3)). But this is exactly what we have found by studying relative motion
in Newtonian Physics! In a rigid motion the space is ”frozen” into the system. Therefore the
existence of rigid motions in Newtonian Physics is equivalent to the existence of the Euclidean
metric! Looking at the same result from a different angle, we may consider the background
space to be the Euclidean space (and not simply the linear space R3 ) and understand the
rigid motion as motion of parts of space enclosing an aggregate of fixed points. This type of
geometric system we define as the physical system rigid body.
In Newtonian Physics rigid bodies are considered to be aggregates of particles moving in a
background with geometric structure defined by the Euclidian metric and not as rigid motions of
aggregates of particles developing in the linear space R3 , without any additional structure. That
is, the geometry is considered to be a property of the space not of the motion and consequently
of the structure of the aggregate of particles.
Our approach is different, in the sense that we see the geometry as a property of motion
and not as a property of space. Let us analyze this a little further.
We have defined the rigid body as an aggregate of particles, which is able to execute rigid
motions only. We have shown that this type of motion is defined by means of two vector fields
(the vP , ω) and that these vector fields generate two groups of motions, the translations T (3)
and the rotations O(3). We make the following question:
Is there a metric (that is non-degenerate second order symmetric tensor) whose

isometries (=Killing vectors) generate the Lie group with subgroups T (3) and O(3)?
If the answer is “yes”, then we have defined a metric in the linear space R3 via its group
of isometries, and in a sense we have geometrized rigid motion. The answer as to the existence
of such a metric will be given by observation. That is, we have first to observe in nature rigid
motions and thus prove that such a metric exists. In this approach geometry is defined by
physical observation and it is not an inherent property of space which modulates motion. In
fact we do not need to give space special geometric properties, besides its linearity and in a
way we are in accord with Newton, who in his celebrated book Principia defines space as:
3.1. THE GEOMETRIZATION OF RIGID MOTION 15
Absolute space, in its own nature, without regard to anything external, remains
always similar and immovable.
However, there is still a key point we have to address. This is the method of observation. As
we know, each theory of Physics postulates the procedures for observing motion. For example
in Newtonian Physics it is postulated that one observes motion using an absolute stick of unit
length and an absolute clock. In Special Relativity it is postulated that motion is measured
by means of light signals and synchronized clocks. Therefore the result of observation depends
also on the method of observation! In that sense, the geometry defined by the relative motion
in a theory of Physics and verified in the realm of this theory, is applicable to that theory only,
because the method of observation of motion is specific to one theory only.
The beauty of this approach is that everything depends upon our definition of observation5 .
There is nothing beyond and above the human power or beyond our experience. Geometry is a
human construction and it is dictated by our approach of motion. There is nothing metaphysical
in Physics!
The observational fact that there do exist rigid motions in nature, when we observe motion
with the specified Newtonian method, takes us to the conclusion that rigid motion is equivalent
to - or it can be understood in terms of - the celebrated Euclidian metric. In conclusion:
1. The Euclidian metric exists and it is the result of the observation of rigid motion in
Newtonian Physics
2. The Euclidean geometry is the result of a certain type of motion (rigid motion) and not
a property of space.
3. The Euclidean metric is proved to exist only within the observational procedures of New-
tonian observation.
4. The Euclidean metric geometrizes rigid motion in Newtonian Physics, in the sense that
instead of studying the motion per se in Physics one may study the geometric properties
of the equivalent metric and then transfer the results directly to kinematic results of rigid
motion.
3.1.1 The geometrization of strain motion

Having related the rigid motion with the Euclidian geometry we continue with the other type
of motion of a linear deformable body in Newtonian Physics, that is, the strain motion. As
5
A similar point of view seams to be taken in Cybernetics, although not justified so clearly. They say (see
http://pespmc1.vub.ac.be/):
“Among the most elementary actions known to us are small displacements ”in space”. We have put the
quotes, because people have accustomed to imagine that some entity, called ”space” exists as a primary reality,
which creates the possibility of moving from one point of this space to another. Our analysis turns this notion
topsy-turvy. Only actions constitute observable reality; space is nothing but a product of our imagination which
we construct from small displacements, or shifts, of even smaller objects called points. If x is such a shift, then
xx – the action x repeated twice – is a double shift, which we would call in our conventional wisdom a shift at
the double distance in the same direction. On the other hand, we may want to represent a shift x as the result
of another shift x’ repeated twice: x = x’x’. It so happens that we can make three different kinds of shifts, call
them x, y, z, none of which can be reduced to a combination of the other two. At the same time any shift w can
be reduced to a properly chosen combination of shifts x, y, z. So we say that our space has three dimensions. ”
we have already shown the strain motion is described by a symmetric second rank tensor (the
strain tensor). The strain tensor can be considered as a metric of signature −3 (i.e. a positive
definite metric) which is not flat, i.e. it is a Riemannian metric of non-vanishing curvature. As
was the case with the rigid motion we have for strain motion the following conclusions:
1. The strain motion defines a new metric, which is the strain tensor. This is a positive
definite metric. This metric can be degenerate for special cases of strain motion (e.g. one
dimensional strain motion)
2. The strain metric is not flat, that is, its Riemannian curvature does not vanish.
3. The symmetry group of the strain metric is not the symmetry group of the Euclidean
metric E 3 .
4. The stain motion is geometrized by the strain metric (tensor) therefore by studying the
properties of this metric with the methods of Differential geometry we can draw conclu-
sions about the strain motion in Newtonian kinematics.
The result of the above study is that the general motion of a (linear) deformable body in
Newtonian Physics can be described geometrically by means of two positive definite metrics.
The Euclidian metric which describes / corresponds to the rigid part of the motion and the
strain metric which describes / corresponds to the strain part of the motion.
3.1.2 A geometric computation of the strain metric

Having the results of the previous section as a guide it is possible to approach geometrically the
strain tensor in a more elegant fashion. Consider within the continuum the points P, Q with
position vectors x, x + dr where dr = PQ and assume that after the action of some some forces
the continuum changes its shape (is deformed) so that the points P, Q go to points P ′ , Q′ with
position vectors x + s(x), x + dr + s′ (x + dr) respectively. From the linearity of the space (see
also (2.39) ) it follows that the distance r′ = P′ Q′ :
r′ = dr + s′ (x + dr) − s(x). (3.2)
The deformation is continuous which means that:
s′ (x + dr) = s(x)+ds(r)
To write this relation quantitatively we consider a frame of reference6 , Σ say, in which the
vectors r = (x1 , x2 , x3 ), dr = (dx1 , dx2 , dx3 ) and s(r) = u(r)dt where u(r) is the velocity at the
position r. Then
∑ ∂sµ
ds(r) = ν
dxν (3.3)
ν
∂x
6
The discussion is general and applies to linear spaces of any finite dimension.
3.1. THE GEOMETRIZATION OF RIGID MOTION 17
and the squared Euclidean distance r′2 between the new points P ′ , Q′ is by virtue of (3.2) and
(3.3):
∑( ∑ ∂sµ ν 2
)
′2 µ
r = dx + ν
dx
µ ν ∂x
( ) ( )
∑ µ 2 ∑ µ ∂s
µ
ν
∑ ∂sµ ν ∂sµ ρ
= (dx ) + 2 dx dx + dx dx .
µ µ.ν ∂xν µ.ν,ρ ∂xν ∂xρ
∑
The first term (dxµ )2 = dr2 is the squared distance between original the points P, Q and
µ
corresponds to the rigid motion part of the deformation7 . The remaining two terms give the
change in squared distance between the points P, Q which is not covered by the rigid motion.
This term is quadratic in dxµ , therefore it can be manipulated to define a square matrix and
subsequently a second order symmetric tensor. We note that we can write:
( ) ( )
∑ µ ∂s
µ
ν
∑ ∂sµ ν ∂sµ ρ
2 dx dx + dx dx
µ.ν ∂xν µ.ν,ρ ∂xν ∂xρ
( ) ( )
∑ µ ∂sµ ∂sν ν
∑ ν ∂sµ ∂sµ
= dx + dx + dx dxρ
µ.ν ∂xν ∂xµ µ.ν,ρ ∂xν ∂xρ
[ ]
∑ µ ∂sµ ∂sν ∑ ∂sρ ∂sρ
= dx ν
+ µ+ ν ∂xµ
dxν .
µ.ν ∂x ∂x ρ ∂x
We define the extended strain tensor eµν with the formula:

[ ]
1 ∂sµ ∂sν ∑ ∂sρ ∂sρ
eµν = + + . (3.4)
2 ∂xν ∂xµ ρ
∂xν ∂xµ
Then the deformation is:

r′2 = dr2 + eµν dxµ dxν .
whereas the change D(dr) in the squared relative distance of two particles executing strain
motion only is (summation convention understood):
D(dr) = 2eµν dxµ dxν (3.5)

µ
We shall say that the stains are small strains if the derivatives ∂u
∂xν
are so small that their
products can be neglected. In this case the extended strain tensor reduces to the strain tensor
or strain tensor for small displacements given by:
[ ]
1 ∂sµ ∂sν
eµν = + . (3.6)
2 ∂xν ∂xµ
7
Note that
∑in full generality we may consider three types of motion of the continuum.
∂sµ ′2
ν dx = 0 ⇒ r
ν
1. ds(r) = ∂x = r2 which means that the distance (P Q) is preserved during motion. This is
ν
the rigid body motion.
∂sµ ∂sρ
2. ds(r) ̸= 0 and ∂x ν ∂xρσ = 0. In this case we have the strain motion for small strains
∂sµ
3. ds(r) ̸= 0 and ∂x ν
∂s
∂xσ ̸= 0. In this case we have the extended strain motion.
The deformable bodies, which move under small strains we call linear deformable bodies.
Because in the following we shall be concerned with motions generating small displacements,
the resulting strains will be small therefore, we shall consider linear deformable bodies only. To
simplify the wording in the following we shall drop the word linear and say simply deformable
bodies. Furthermore by strain we shall mean strain tensor for small displacements. We note
that the trace of the stain tensor (for small displacements!) is:
∂sµ
eµµ = = divw. (3.7)
∂xµ
where w is the displacement vector s.
It is possible to write the change of the squared relative deformation for small strains as
follows:
∑( )
∑ ∂sµ ν 2 ∑ µ
µ
dx + ν
dx = dx [δµν + eµν ] dxν (3.8)
µ ν ∂x µ.ν
and consider that the strain acts as a supplementary metric to the Euclidean metric.
In this approach the strain can be considered as the deformable body’s internal metric that
is, the geometry which is due only to the deformations of the body when the body is “freezed”,
in the sense that it does not suffer global translation and global rotation (global means all
the points share the same property). This is the metric which must be used in the study of
the rheological properties of a deformable body, that is the properties due only to the ”static”
deformations (=change of shape) of the body.
Using that sµ = uµ dt we obtain the rate of strain tensor we considered already (see (2.45)
), that is: [ ]
1 ∂uµ ∂uν
ẽµν = + µ . (3.9)
2 ∂xν ∂x
Note that the ‘rate of strain tensor’ does not mean change of the strain tensor in time but
instead change of the connecting vector uµ . In the following we use ẽµν to refer to the strain
motion of a linear deformable continuum.
Chapter 4
The stress tensor
As it has been shown in (2.39) the general motion of a (linear) deformable continuum consists
of three parts:
a. A pure translation r, which is a vector
b. A pure rotation 12 (▽ × vP ) × r, which is a pseudo-vector or 1-form.
c. A pure strain 12 ▽ (r · vP dt), which is a symmetric 2nd rank tensor.
According to Newton’s Second Law the first two types of motion are produced respectively
from a vector (the force F) and a pseudo-vector or 1-form (the torque M ). We expect that
the strain motion will be produced by a form of Newton’s law, which will involve a symmetric
second rank tensor. This new tensor we call the stress tensor. How the stress tensor is defined,
what measures and how it is related to the strain tensor it produces, are questions which we
address in this and the next sections.
Simple experimental observations have shown that the deformation generating a strain
motion depends on force per unit area and not on the force alone. For example, a given
load will extend a thin wire more than a thick one, because in the first case the area of the
cross section is less and therefore the force per unit area greater. The deformation resulting by
a strain motion, is due to the adjustment of the internal forces between individual particles,
and the extend of this adjustment depends on the additional forces experienced by each of the
particles affected - that is, on the additional force per particle. But each particle has a definite
effective area, hence we have to consider the area over which it acts as well as the force itself,
that is, the additional force per particle is proportional to the external force per unit area.
Therefore we define:
Stress is the force per unit area regarded as being transmitted throughout the material of the
continuum at global rest (that is, after global translation and global rotation have been
removed).
In SI units the units of stress are N m−2 i.e. pressure and in the English system pounds
per square inch (psi). The 1 N m−2 = 1P ascal; because this is a rather small stress usually
stresses are expressed as mega -Pascal (M P a) or in English units Kilo pounds per square inch
(ksi)
Since the stress is a second rank symmetric tensor in a coordinate frame it can be represented
uniquely by a square symmetric matrix whose elements are the stress components in that frame.
19
20 CHAPTER 4. THE STRESS TENSOR
We shall consider orthonormal frames and shall follow the convention:

 
σ11 σ12 σ13

[σij ] = σ21 σ22 σ23  (4.1)
σ31 σ32 σ33
If A is an orthogonal matrix (i.e. AAt = 1 where t=transpose) relating two Cartesian

frames P x1 x2 x3 and P x′1 x′2 x′3 at the point P of the continuum then the transformation of the
components of the stress tensor are:
[σij′ ] = At [σij ]A (4.2)
The stress tensor may be considered as an additional metric entering the motion of a linear
deformable continuum, which corresponds / describes / geometrizes the environment causing
the strain motion of that continuum. In this respect equation (4.2) is an isometry of the stress
metric.
4.0.3 Body and Surface forces

Two basic types of forces are easily recognized from one another and are defined as follows.
Forces which act on all volume elements of the continuum which we call body forces. One
such type of force is gravity. We designate body forces by the vector symbol bi (force per unit
mass) or by the symbol pi (force per unit volume). If ρ is the density of the continuum we have
the relation:
ρbi = pi . (4.3)
The other type of force concerns the forces which act upon and are distributed in some fashion
over a surface element of the continuum, regardless of whether that surface is part of the
bounding surface, or an arbitrary element of surface within the continuum. These forces we
call surface forces and denote with the symbol fi (force per unit area). Examples of this type
of force are the contact forces, forces which result from the transmission of forces across the
internal surface to the continuum etc.
Suppose that at a point P within a continuum the stress tensor is σij and that we place a
surface dS at P whose unit normal is nj . Then the stress vector at P is the vector1
tiS = σij nj (4.4)
which is is the surface force on the surface dS when placed at P. Different surface elements
placed at P will lead to different stress vectors at P.
At the point P of the continuum we consider the orthonormal frame P x1 x2 x3 , draw a cube
and represent the components of the stress tensor as shown in Figure ??
In drawing the components we follow the convention that a positive stress component is
represented with an arrow in the positive direction of one of the coordinate axes while acting
on a plane whose outward normal also points in a positive coordinate direction. All the stress
components displayed in Figure ?? are positive.
1
The formula tiS = σij nj is known as Cauchy stress formula.
21
x
3
Ï 33
Ï 32
Ï 31 Ï 23
Ï
Ï 21 22 x2
Ï 13
Ï 12
Ï 11
x1
Figure 4.1: Bulk Stress
The three stress components shown by arrows acting normal to the respective coordinate
planes and labeled σii i = 1, 2, 3 are called normal stresses. The six arrows lying in the
coordinate planes and pointing in the direction of the coordinate axes, namely σij i ̸= j = 1, 2, 3
are called shear stresses. Note that for these the first subscript (the i) identifies the coordinate
plane (that is, the plane normal to the i-axis) on which the shear stress acts, and the second
subscript (the j) identifies the coordinate direction in which it acts.
4.0.4 The perfect deformable continuum

Consider a linear deformable continuum at global rest (so that the continuum is possible to
execute strain motion only) which under the influence of a stress executes a strain motion. Due
to the strain motion the shape of the body changes and as a result various forces develop in
the interior of the continuum. To study these forces we immerse a small flat surface element of
area dS at various points within the continuum and study the net force exerted on the surface
at each point. There are two possibilities:
a. The net force acting on the surface dS is normal to the plane of dS.
b. The net force to the surface element dS is inclined with respect to the surface (that is,
it is not collinear to the normal vector to the plane of the surface dS).
In the first case if the surface element dS moves parallel to the plane of the surface, it does
not consume work i.e. the surface moves ”freely” into the continuum. A deformable continuum,
in which the force of the stress on an elementary surface dS placed at any of its points is normal
to the surface for all directions of the elementary surface (at the same point), we call perfect
deformable continuum. Obviously the physical properties of a perfect deformable continuum

are isotropic, that is, invariant under the action of the group of rotations O(3).
In order to compute the stress required to produce a strain motion in a perfect deformable
continuum we consider a point P within the continuum and define at P three orthogonal axes
{x, y, z}. If we place an elementary surface dS normal to the x − axis the corresponding force
will be along the x−axis i.e. we shall have Fx = (px dS, 0, 0) where px is the pressure on the
surface dS. Similarly the force on the elementary surface dS placed normal to the axes y, z are
Fy = (0, py dS, 0), Fz = (0, 0, pz dS) respectively. Because a perfect deformable continuum is
isotropic (that is invariant under the action of O(3)) all directions must be equivalent therefore
we must have2 px = py = pz = p. The common pressure p at P (the value of p in general
is different at different points) we call the isotropic pressure. We conclude that the stress
tensor producing a strain motion of a perfect deformable continuum is described by a scalar
(the isotropic pressure). Furthermore we note that the three forces Fx , Fy , Fz can be described
altogether with the diagonal matrix diag(p, p, p) so that the explicit form of the stress is the
symmetric second rank tensor pδµν .
In terms of Geometry this means that the metric of stress of a perfect linear deformable
continuum is conformally related to the (flat) Euclidian metric the conformal factor being
the isotropic pressure p. This observation shall be very important later on when we study
geometrically the dynamics of a continuum.
There are two points which are of interest in the case of a perfect deformable continuum. A
perfect deformable continuum is characterized by a mass density ρ (say) whereas its reaction
with the environment is characterized with the isotropic pressure. The fact that the stress
tensor is defined independently of the mass density makes possible the consideration of perfect
deformable bodies whose mass density vanishes. In Newtonian Physics such material continua
do not exist, however they do exist in relativistic Physics. One such continuum consists of
photons (it is known as photon gas) whose dynamics is studied e.g. one computes the pressure
it exercises on a given surface. Another extreme situation is to consider a perfect deformable
body whose mass density vanishes and the only possible pressure possible is p = 0, that is
nothing can interact with it. This continuum may be considered as the empty space. In this
approach the empty space is considered to be a physical system and not the substratum where
all types of motion take place. This cannot be said for the time whose nature is different and
must be considered as a device which we use to understand the sequence of events (i.e. motion)
of physical systems in the environment.
4.0.5 The imperfect deformable continuum

Real deformable continua are not perfect. That is, when an elementary surface is placed
inside a deformable continuum tangential components of stress (i.e. forces) always exist. The
immediate consequence of the existence of tangential forces, is that the property of the equality
of the pressure in all directions at every point no longer holds.
To study the new situation we consider again at an arbitrary point P of the deformable
body three orthogonal axes {x, y, z} and place the elementary surface dS normal to the x-
axis. In this case the force on the surface will not be normal to the surface and it will
2
This is the experimental result of Pascal with perfect fluids.
23
have in general components along all axes {x, y, z}. We write in an obvious notation Fx =
(pxx dS, pxy dS, pxz dS). Similarly the forces on elementary surfaces placed normal to the axes
y, z shall be Fy = (pyx dS, pyy dS, pyz dS) and Fz = (pzx dS, pzy dS, pzz dS) respectively. The
components pxx , pyy , pzz are normal to the surfaces on which they act, while the rest six denote
tangential components. For example pxy dS is the force in the direction y on the area dS per-
pendicular to the axis x. We shall consider the symbols pxx , pyy , pzz to be positive when they
represent tensions (that is, the force is along the outward normal to the surface), so that a
pressure is to be regarded as a negative stress. Because the surface dS does not rotate the
total moments on it must vanish, therefore we must have 3 pzy = pyx and similarly for the
other two principal axes.
We summarize the previous considerations as follows (see Figure ?? for explanation of the
indices):
1. In an arbitrary orthogonal coordinate frame {x, y, z}, the three forces acting on the ele-
mentary surface dS placed at any point P in the interior of a deformable continuum undergoing
a strain motion, define the symmetric matrix:
 
pxx pxy pxz
σµν =  pyx pyy pyz  .
pzx pzy pzz
This matrix defines (in the frame used) the components of a two index symmetric tensor (pos-
sibly degenerate), which we call the stress tensor and denote by σµν . It is this tensor which
causes the strain motion of an imperfect linear deformable continuum.
2. For a perfect linear deformable continuum we have:
pxx = pyy = pzz = p, pxy = pxz = ... = pzy = 0.
and the stress tensor σµν = pδµν .

3. As was the case with the strain tensor, the stress tensor defines the stress ellipsoid, has
stress principal axes etc. Furthermore the stress tensor can be considered as a metric on the real
space R3 , which geometrizes the external forces acting on the deformable continuum. If these
forces are due to a force field (as in the case of the gravitational field) this metric geometrizes
this force field4 .
4. For a general motion of a deformable continuum in Newtonian Physics we have three
metrics to consider :
a. The Euclidean metric geometrizing the rigid motions
b. The strain metric geometrizing the strain motions
c. The stress metric geometrizing the forces (stresses) of the environment of the deformable
continuum.
The first two metrics are independent of the linear deformable continuum and depend upon
the space where the motion occurs, whereas the last two metrics are related with the ‘equations
3
As we have said the strain is the ”motion” of a deformable continuum which remains after the global
translation and the global rotation have been removed. That is the strain is a motion in which only the shape
of the continuum changes. Because we have no rotation the stress tensor which produces the strain must be
symmetric (otherwise it would produce couples, therefore rotation about the center of mass.)
4
This observation takes us closely to the case of General Relativity.
of motion’ of the linear deformable continuum. As we shall see below these equations are the
Navier Stokes equations.
5. For a perfect linear continuum the σµν = pδµν and the stress metric is conformally
related to the space Euclidian metric, with conformal factor given by the isotropic pressure p.
4.0.6 Types of strain and stresses

For the purpose of classifying the behavior of different materials there have been recognized
three types of elementary stresses and strain.
Longitudinal Stress
The longitudinal stress (see Figure 4.2) concerns one dimensional strain and stress.
y
dw dy
w
Figure 4.2: Longitudinal stress
and results form the application of an external force F along the direction of a one dimensional
deformable continuum of cross sectional area A. It is given by the ratio F/A. If the application
of the stress results in a strain δx (elongation) of the body and the body has length x before
the application of the stress, then the measure of the longitudinal strain is given by δx/x. The
Young modulus E is the pressure defined by the ratio:
F/A
E= (4.5)
δx/x
In practice there are no one dimensional bodies therefore the longitudinal stress is an unreal
situation. In reality we have one dimensional bodies with thickness e.g. in the form of wires.
In this case it is (experimentally) observed that the continuum suffers two changes:
- longitudinal extension and
- lateral contraction.
This requires the definition of an additional parameter, which will encounter for the lateral
contraction of the continuum. This new parameter we call Poisson ratio and define as follows:
lateral contraction per unit breadth δy/y

σ= =− (4.6)
longitudinal elongation per unit length δx/x
The Poisson ration is dimensionless.
Shear Stress
Shear Stress (see Figure 4.3 ) concerns two dimensional strain and stress.
25
Figure 4.3: Shear Stress
Shear are tangential forces applied to the planes of the areas concerned, and produce a
shear stress acting through the deformable continuum. If F is the force (the continuum is
constrained so that it is possible to perform only strain motion i.e. change of shape) and the
surface of the area of application of the force is A, then the shear stress equals F/A. The shear
strain produced is a lateral displacement per unit length of the surface of the body in the
direction of the applied force and it is measured perpendicular to the direction of the force.
From Figure 4.3 we have that the shear strain equals δx y
. An equivalent measure of the shear
δx
strain is the angle of shear θ defined by tan θ = y . The ratio (pressure):
shear stress F/A

n= = (4.7)
shear strain θ
is called the modulus of rigidity. If n is a constant we say that the continuum is a perfectly
elastic material.
Bulk or volume stress

Bulk or volume stress concerns three dimensional strain and stress.
Figure 4.4: Bulk Stress
When uniform pressure is exerted equally from all sides e.g. as in a solid immersed in a
static fluid, then we have the development of a stress acting on the body, which we call bulk
or volume stress. If the pressure exerted is p N/m2 then the volume stress transmitted
throughout the material of the continuum is p N/m2 . As a result of the volume stress we have
a contraction of the volume of the continuum form V to V − δV. The quotient δV /V we call
the bulk or volume strain. The quotient (pressure):
volume stress p
k= =− . (4.8)
volume strain δV /V
we call the bulk modulus or shear modulus of the material of the continuum. The reciprocal
of the bulk modulus k we call the compressibility of the material.
In general the strain motion of a deformable continuum in real world is different from the
special types encountered above. However the strain motions we are interested in practical
situations (in general) can be understood/described in terms of these three simple types of
strain and, most important, it is possible to device experiments in which one form of the above
strains predominates and thus one can measure the values of the quantities E, σ, n, k.
All stresses have dimensions Force/Area=Pressure, that is M L−1 T −2 and all are measured
in N m−2 (Pascal) Similarly all strains are dimensionless quantities.
In general all four coefficients σ, Y, n and k of a deformable are required in order to describe
the stain motion of a linear deformable continuum under the influence of a given stress. However
for special cases there are relations amongst σ and the stress coefficients Y, n and k. For example
as we shall show in the next section (see equations (5.14), (5.10)), for a linear isotropic elastic
continuum one computes:
E E
n= , k= (4.9)
2(1 + σ) 3(1 − 2σ)
therefore only two of the four coefficients suffice for the description of the strain motion. From
the above relations we also conclude that for linear isotropic materials the Poisson ratio σ :
- Cannot exceed + 12 because k would be negative and
- Cannot be less than −1 because then n will be negative.
Therefore −1 < σ < 12 .
Chapter 5
The stress - strain relation for elastic

continua
An important type of deformable continua are the ones in which the stress is a homogeneous
function of the rate of strain. That is, when the stress and the rate of strain vanish together.
The simplest such continua are the ones for which the stress is a linear homogeneous function
of the strain, that is the following relation holds:
∑
λρ
tµν = Yµν ẽλρ (5.1)
λ,ρ
λρ
where Yµν is a tensor (the coupling tensor) whose components are functions of the coordinates.
λρ
Such deformable continua we call elastic continua. The dimensions of Yµν are [T ].
λρ
The coefficients Yµν are defined solely by the physical properties of the elastic continuum and
λρ
are called the elastic coefficients of the continuum. In general Yµν has 34 = 81 components.
However due to the symmetry of the strain and the stress tensors the independent components
of the coupling tensor reduce to 3·42
× 3·4
2
= 36.
This number of parameters is still too excessive for the level of the present notes. Therefore
we have to simplify things further and we consider the special class of linear elastic continua,
which are defined by the requirement that their physical properties are the same at all their
points. Mathematically this is stated by the condition:
λρ
Yµν ,σ = 0 (5.2)
λρ
or, equivalently, that the elastic coefficients are constants. Relation (5.1) with Yµν =constant
is known as Hooke’s law. The name is unfortunate because it is not a law of Physics, but
rather a ‘selection rule’ (i.e. a simplifying assumption) amongst the elastic continua.
In order to reduce the number of independent coefficients of a linear elastic continuum
further, we specialize further our considerations to linear elastic isotropic continua. which
are considered to be the linear elastic continua for which their physical properties are isotropic
i.e. invariant under the action of the group SO(3). Mathematically this is expressed by the
λρ
requirement that the tensor Yµν is isotropic in the Euclidian space.The generic isotropic tensor
1
of order two in the Euclidian space is the δµν . It can be shown that the most general isotropic
1
This means that this tensor is an irreducible representation of the group SO(3).
27
28 CHAPTER 5. THE STRESS - STRAIN RELATION FOR ELASTIC CONTINUA
tensor of order four is of the form2 :
Yiklm = Aδik δlm + Bδil δkm + Cδim δkl
where A, B, C are scalars (of the group SO(3)) and in our case constants. Then we have from
(5.1):
Yλρµν ẽµν = [Aδλρ δµν + Bδλµ δρν + Cδλν δρµ ] ẽµν = Aδλρ ẽµµ + Bẽλρ + C ẽρλ = (B + C)ẽλρ + Aẽµµ δλρ
where in the last step we have used that the strain tensor ẽµν is symmetric. If we set ρ = B + C
and A = λ this expression becomes3 :
∑
tµν = ρẽµν + λδµν ẽτ τ (5.3)
τ
∑
where ẽτ τ is the trace of ẽµν (an SO(3) invariant) and ρ, λ are two constant coefficients, which
τ
have to be determined. We note that the additional requirement of SO(3) symmetry reduces
λρ
the number of independent components of Yµν from 36 to 2 (the constants ρ, λ). Equation (5.3)
can be inverted to give the strain in terms of the stress as follows:
∑
ẽµν = ϕtµν + ξδµν tτ τ (5.4)
τ
where ϕ, ξ are two new constant coefficients with dimensions [T −1 ].

The value of the constant coefficients ρ, λ and ϕ, ξ is independent of the particular type of
stress and the resulting strain and are characteristics only of the material of the linear elastic
isotropic continuum. Obviously only one pair of parameters is required to be computed, the
other pair being calculated form the inverse relations. The computation of the independent
pair of parameters will be done from relation (5.3) or (5.4) provided we know both the stress
and the resulted strain.
In the following section we compute these coefficients for a linear elastic isotropic continuum.
5.0.7 The stress rate - strain relation for a linear elastic isotropic
continuum
In order to compute the parameters ρ, λ and ϕ, ξ in terms of the Young modulus E and the
Poisson ratio σ for a linear elastic isotropic continuum we consider at an arbitrary point inside
the continuum the principal axes of strain or stress. Because the material is isotropic these axes
can be rotated and still remain principal axes. We choose these axes4 so that both the strain
tensor ẽµν and the stress tensor tµν are diagonal. Suppose then that ẽµν = diag(e11 , e22 , e33 )
and tµν = diag(t11 , t22 , t33 ). In order to calculate the effect of the stress tµν we apply three
2
Has to be symmetric in the indices i, k, l, m. This is hoe we construct this tensor
3
This result holds for any symmetric second order isotropic tensors, not only for the strain and the stress
tensor.
4
Here we have two positive definite metrics which are diagonalized simultaneously, that is, their principal
axes coincide. It has been shown by mathematicians that this is always possible
29
independent longitudinal stresses along each one of the principal axis and then we sum the
result. We note that the rate of strain motion along the x−axis is:
ẽ11 = δx/x + δy/y + δz/z
where the stain motions δx/x + δy/y + δz/z are due to the stress t11 along the x−axis. A
similar result holds for the rate of stress and the stress for the other two principal axes5 .
We start with the longitudinal stress along the principal x−axis (recall that along a principal
axis the stress is only longitudinal!). This stress creates a strain motion which is elongation
along the x−axis by δx and contraction along the other two axes by δy and δz. This strain is
produced by the stress component t11 . We have for each part of the strain motion the relations:
1
1. Strain motion along the x-axes: δx/x = t
E 11
(by the definition of E)
2. Strain motion along the y axis: δy/y = −σ(δx/x) = − Eσ t11 (by the definition of σ)
3. Along the z axis: δz/z = −σ(δx/x) = − Eσ t11 (by the definition of σ).
Similar results hold for the stresses t22 , t33 applied along the directions y, z respectively and
the resulting stain motions. Therefore we have6 :
Strain motion along the x−axis:
1
ẽ11 = [t11 − σ(t22 + t33 )]
E
Strain motion along the y−axis:
1
ẽ22 = [t22 − σ(t33 + t11 )]
E
Strain motion along the z−axis:
1
ẽ33 = [t33 − σ(t11 + t22 )] .
E
As we remarked above the total strain under the action of the total stress (i.e. tµν =
diag(t11 , t22 , t33 ) ) along the principal axes is the sum of the strain motions along each principal
axis. Therefore for the total rate of stain motion we have:
1
ẽµν = diag (t11 − σ(t22 + t33 ), t22 − σ(t33 + t11 ), t33 − σ(t11 + t22 ))
E
1 σ
= diag (t11 , t22 , t33 ) − diag (t22 + t33 , t33 + t11 , t11 + t22 )
E E
1 σ σ
= diag (t11 , t22 , t33 ) + diag (t11 , t22 , t33 ) − tρ.ρ
E E E
from which follows:
1+σ σ ∑
ẽµν = tµν − δµν tρρ . (5.5)
E E ρ
5
Along the principal axes only longitudinal stresses are possible.
6
σ is the same along the z and the y direction due to the isotropy.
This is the general relation relating the applied stress and the resulting rate of strain motion
in a linear elastic isotropic continuum. Comparing with (5.4) we find that for a linear elastic
isotropic continuum:
1+σ σ
ϕ= , ξ=− . (5.6)
E E
where E, σ are the Young modulus and the Poisson ratio of the continuum.
∑ ∑
Exercise 5.0.1 Let T race(ẽ) = ẽ and T race(t) = tρρ . Show that the following relation
ρ ρρ ρ
holds:
E
T race(t) =T race(ẽ). (5.7)
1 − 2σ
Using this result show that the relation between the strain and the stress for a linear elastic
isotropic continuum is:
[ ]
E σ
tµν = ẽµν + δµν T race(ẽ) . (5.8)
1+σ 1 − 2σ
Conclude that for longitudinal stresses, the parameters ρ, λ of the linear elastic isotropic con-
tinuum of equation (5.3) are as follows:
E Eσ
ρ= , λ= . (5.9)
1+σ (1 + σ)(1 − 2σ)
Having computed the relation between the stress and the strain for longitudinal stress
applied to a linear elastic isotropic continuum in terms of the parameters E, σ we compute
the shear modulus coefficient n and the bulk modulus coefficient k for these materials, by
considering a shear stress and a bulk stress acting on such a material and using (5.5).
5.0.8 The coefficients n, k of a linear elastic isotropic continuum

F/A
We start with the shear modulus coefficient n = θ
where θ is the angle of shear. We consider
a tangential stress as in Figure 4.3 and find the stress tensor:
 
0 t12 0
tµν =  t12 0 0 
0 0 0
F
where 2t12 = S
. The rate of strain tensor7 per unit length is (for small angles θ) :
 
0 θ/2 0
ẽµν =  θ/2 0 0 .
0 0 0
We note that T race(t) = T race(ẽ) = 0 therefore the stress strain relation (5.5) for such
E
strains becomes tµν = 1+σ ẽµν . Replacing tµν , eµν from these expressions we find:
1+σ
θ/2 = t12
E
7 δx
The non-vanishing component is the y , that is, the xy and yx component only.
31
therefore the shear modulus coefficient of a linear elastic isotropic continuum is:
t12 E
n= = . (5.10)
θ 1+σ
We continue with the bulk modulus coefficient of a linear elastic isotropic continuum. Let p
be the isotropic pressure resulting from the application of a bulk stress. Then the stress tensor
is given by the relation:
tµν = −pδµν . (5.11)
Replacing in (5.5) we compute that the resulting strain tensor is:
[ ]
−p(1 + σ) σ (1 − 2σ)p
ẽµν = − (−3p δµν = − δµν . (5.12)
E E E
For strain motion due to bulk stress we have:

∆V (1 − 2σ)p
= T race(e)= − 3 . (5.13)
V E
Therefore the bulk modulus coefficient k of a linear elastic isotropic continuum is:
p E
k= = . (5.14)
−∆V /V 3(1 − 2σ)
5.0.9 A second derivation of the rate of strain stress relation from

Hooke’s law
We consider a stress tij acting on a given linear elastic continuum producing the strain eij. As
we have seen, at an arbitrary point of the continuum along the principal directions of strain
the rate of strain reduces to translations and the strain tensor is diagonal eij = diag(e1, e2 , e3 ).
Hooke’s law states that:
In a linear elastic continuum the rate of strain produced in a definite direction

is proportional to the stress (=force per unit area) in that direction.
Therefore we infer that the stress tensor tij must also be diagonal in the strain principal
directions or, to put it in an equivalent way, the principal axes of strain and stress coincide.
We write tij = diag(t1 , t2 , t3 ).
Here the geometric assumptions end and we have to take Physics into account. As we
have already mentioned it is an experimental fact that the application of longitudinal stresses
produce extensions, measured by the Young’s modulus, and transverse contractions measured
by the Poison ratio. Assuming the material to be isotropic, so that we have a linear elastic
isotropic continuum, these transverse contractions are characterized by the same Poisson ratio.
We consider an arbitrary force acting on the linear elastic isotropic continuum and analyze it
in three components along the principal axes of the strain tensor. These produce a longitudinal
strain along each direction, each strain being independent of the other two. Therefore the
relation between the stress tensor and the rate of strain tensor (per unit length of material) -
along the principal axes of strain only - is8 :
1 σ 1+σ σ
ẽ11 = t11 − (t22 + t33 ) = t1 − (t11 + t22 + t33 )
E E E E
1 σ 1+σ σ
ẽ22 = t22 − (t33 + t11 ) = t22 − (t11 + t22 + t33 ) (5.15)
E E E E
1 σ 1+σ σ
ẽ33 = t33 − (t22 + t11 ) = t33 − (t11 + t22 + t33 )
E E E E
or, in tensor notation:
1+σ σ
ẽµµ = tµµ − (t1 + t2 + t3 )δµν µ = 1, 2, 3. (5.16)
E E
We define the invariant9 :
σ
ψ = − (t1 + t2 + t3 ) (5.17)
E
and relation (5.16) is written as:
1+σ ′
ẽ′µµ = t +ψ µ = 1, 2, 3. (5.18)
E µµ
We need to emphasize that formula (5.16) applies only at a point P because the principal
directions of stress in general vary from point to point. That is, in general there does not
exist a simple coordinate system in which the stress and the strain components are related
as in (5.16). However this presents no difficulty because the stress and the strain are tensors
therefore we can compute their components in any frame if we know them in one or, to put it
in an equivalent way, equation (5.16) is covariant. Indeed let us consider a point Q far from P
µ
and let the frame we choose at Q to be {x }. If we denote the coordinate frame of the principal
directions at P with {xµ } then we have the coordinate transformation:
µ
x = aνµ xν (5.19)
where the Jacobian matrix [aνµ ] satisfies the orthogonality condition:
aνµ aµρ = δνρ . (5.20)
As it is well known this condition is a result of the Euclidean character of the space. It is
also well known that under a Euclidean (or orthogonal) transformation the trace t1 + t2 + t3 is
invariant10 . In the coordinate system at the point Q this becomes:
1+σ ′
ẽ′µν = t + ψδµν . (5.22)
E µν
8
We use the same E, σ because the continuum is isotropic, therefore the scalars on the surface of a sphere
centered at the origin (here the origin of the principal directions of strain / stress) must have the same value.
9
This is an invariant because the quantities E, σ being tensors are invariants .
10
One can prove the invariance directly using the transformation rule (5.19). Indeed we have:
∑ ∑ ∑ ∑
tµµ = aρµ aλµ tρλ = δ ρλ tρλ = tρρ (5.21)
µ µ µ µ
33
In component form (5.22) is written in terms of corresponding (symmetric) matrices as follows11 :

 ′   1+σ ′ 
ẽ11 ẽ′12 ẽ′13 t
E 11
+ ψ 1+σ ′
t
E 12
1+σ ′
t
E 13
 * ẽ′22 ẽ′23  =  * 1+σ ′
t + ψ 1+σ t′  . (5.23)
E 22 E 23
′ 1+σ ′
symmetric * ẽ33 symmetric * t +ψ
E 33
In order to invert relation (5.22) we compute the trace Ψ = e11 + e22 + e33 in terms of the
invariant ψ. From (5.22) we have in an obvious notation:
1+σ σ
Ψ= ψ + 3ψ ⇒ ψ = Ψ. (5.24)
E 2σ − 1
Then (5.22) gives:
[ ]
E ′ σ E σ
t′µν = ẽ − Ψδµν = ′
ẽ + Ψδµν . (5.25)
1 + σ µν 2σ − 1 1 + σ µν 1 − 2σ
This expression coincides with relation (5.8) we derived in the previous section using a more
mathematical method.
11
As we have already remarked the isotropy implies only that the tensors are symmetric, not diagonal!
Chapter 6
The dynamics of strain motion
6.0.10 The force due to a stress tensor

In Newtonian dynamics we have only one law: Newton’s Second Law. Therefore in order to
write down an equation of motion for a general strain motion we need to compute the force
which creates the stress. To do that we consider at a point P of the continuum an element of
mass dM whose surface area is (say) S. We consider an element of surface dσ on S and assume
that the force on the mass dM due to the rest of the material of the continuum is dF. We also
choose a coordinate frame in which both the stain and the stress tensor at P are diagonal. Then
we know1 that along each principal direction of the strain, the x1 say, the stress (=pressure) t1
is due to the action of the projection of the force dF along direction x1 acting on the projection
dσ 1 of the surface2 dσ, that is: dF1 = t1 dσ 1 . Because the stresses along the three principal
directions are independent we write in that frame3 :
dFµ = tµν dσ ν . (6.1)
where tµν = diag(t1 , t2 , t3 ) and dσ ν =(dσ 1 , dσ 2 , dσ 3 ). The force on the whole surface S is:
∫
Fµ = tµν dσ ν . (6.2)
S
Applying the divergence theorem we write:

∫∫∫
Fµ = tµν,ν dV (6.3)
V
where V is the volume enclosed by the surface S. The importance of the above expressions is
that they are covariant, hence even if they have been derived in a special coordinate system at
P (the principal axes of the stain and the stress tensors) their validity extends to all coordinate
systems, the explicit computation in a specific coordinate system is done by means of the
1
This is a key assumption!
2
The (infinitesimal) surface dσ can be associated with a vector n normal to the surface whose length equals
dσ. This is the vector dσ ν .
3
In general the density of force in the direction specified by the vector Aµ is given by tµν Aν .
35
36 CHAPTER 6. THE DYNAMICS OF STRAIN MOTION
transformation (5.19). Because the volume is infinitesimal we conclude from (6.3) that the
density of force f µ per unit volume is the divergence of the tensor tµν , that is:
f µ = tµν
..,ν . (6.4)
6.0.11 Newton’s second law in terms of the displacement vector

Before we write Newton’s second law for strain motion we have to make clear what we are
looking for. As we have repeatedly said strain motion is the residual of motion of a continuum
after the rigid motion (global translation and global rotation) have been removed. This means
that the strain motion is not a “usual” motion, that is, change of position, but a motion which
involves only the “shape” (i.e. deformation) of the continuum. Now the change in the shape
of a continuum is described by the displacement vector field, which is a vector field defined
throughout the material of the continuum. Therefore to write Newton’s second law for strain
motion means to write down the equation, which governs the evolution of this vector field in
time under the action of a given stress. This point must be clearly understood if we are going
to understand properly the Physics of strain motion. The above imply that we have to express
the strain tensor in terms of the relative displacement vector (or connecting vector) sµ that
is,equation (2.44)
1
ẽµν = (sµ,ν + sν,µ )
2
in which the rate of strain tensor is expressed in terms of the derivatives of the strain vector
sµ . We shall also use the invariant Ψ = sµ,µ = ▽s .
In the following we derive Newton’s second law for strain motion for a linear elastic isotropic
continuum. For a general strain motion of such a continuum using equation (5.25) in terms of
the strain vector is given by the following expression:
[ ]
E 2σ ρ
tµν = sµ,ν + sν,µ + s δµν . (6.5)
2(1 + σ) 1 − 2σ .ρ
Exercise 6.0.2 Show that in terms of the displacement vector s = (u, v, w) the components of
the stress tensor a linear elastic isotropic continuum are given by the entries of the following
symmetric matrix:
 [ ∂u ] ( ) ( ∂u ∂w ) 
E
+ σ
▽s E ∂u
+ ∂x∂v E
+ ∂x
 1+σ ∂x 1−2σ [
2(1+σ) ∂y
] 2(1+σ) (∂z ) 

[tµν ] =   . (6.6)
* E ∂v
+ σ
▽ s E ∂v
+ ∂w

1+σ ∂y 1−2σ
[ ∂w
2(1+σ) ∂z ∂y
]
* * E
1+σ ∂z
+ 1−2σ ▽ s
σ
Now it is an easy task to compute the density of force due to deformation stress (i.e. the
forces due to the other particles of the continuum) in terms of the derivatives of the strain
vector sµ . Indeed by taking the divergence of (6.5) and using (6.4) we find4 :
[ ] [ ]
µ E µ ν,µ 2σ ν ,µ E µ 1 ν ,µ
f = s + s.....ν + (s ) = s + (s ) . (6.7)
2(1 + σ) ...,νν 1 − 2σ .,ν 2(1 + σ) ..,νν 1 − 2σ .,ν
4
The material is assumed to be locally homogeneous and isotropic therefore E, σ are constants during the
strain motion.
37
In standard vector notation this relation is written as follows:

[ ]
E 1
f= ▽ s+
2
▽ ·(▽ · s) . (6.8)
2(1 + σ) 1 − 2σ
Exercise 6.0.3 Show that the components of the force density of a linear elastic isotropic
continuum are given by the entries of the following matrix:
 E [ 2 ] 
2(1+σ) [
▽ u + 1−2σ1 ∂
(▽s)
∂x ]
 E 
[f ]µ =  2(1+σ) ▽2 v + 1−2σ
1 ∂
(▽s)  (6.9)
[ 2 ∂y
]
E
2(1+σ)
▽ w + 1 ∂
1−2σ ∂z
(▽s)
where s = (u, v, w).
Now suppose that on a linear elastic isotropic continuum of mass density ρ(t, r), acts an
external force with volume density Rµ . Then Newton’s second law5 yields for the strain vector:
Rµ + f µ = ρsν.,tt (6.10)
or by (6.7): [ ]
µ E µ 1
R + s + (s ) = ρsν.,tt .
ν ,µ
(6.11)
2(1 + σ) ,νν 1 − 2σ .,ν
We take the partial derivative sν.,tt because the strain refers to a specific point in space and
the continuum is assumed to be under no global translational and/or global rotational motion.
In standard vector notation equation (6.11) reads:
[ ]
E 1 ∂ 2s
R+ ▽ s+
2
▽ ·(▽ · s) = ρ 2 . (6.12)
2(1 + σ) 1 − 2σ ∂t
6.0.12 Kinematic interpretation of the equation of motion of the

strain vector in a linear elastic isotropic continuum
Before we discuss the Kinematic interpretation of the equation of motion of strain we need the
following result from vector calculus.
Suppose we have a differentiable vector field f and we want to write it as the sum of two
other vector fields, one of which has divergence zero and the other curl zero, that is we want
to write:
f = ▽ × A + ▽ϕ (6.13)
where ▽A = 0. To find this decomposition6 we introduce the new vector field w with the
relations:
1 1
ϕ=− ▽ ·w, A = ▽ ×w. (6.14)
4π 4π
5
The rhs is mass density times acceleration and the lhs is density of force. We apply Newton’s Second Law
in the form:
Total force density = mass density times acceleration. Acceleration is the second order time derivative of the
strain vector.
6
We give a second proof in terms of tensors. We have:
We note that this vector field satisfies the requirement ▽A = 0. To compute the vector
field w in terms of f we replace A and ϕ from (6.14) in (6.13) and find:
1 1
f= ▽ ×( ▽ ×w)− ▽ (▽ · w).
4π 4π
Using the identity of vector calculus:
▽ × ( ▽ ×w) = ▽ (▽ · w)− ▽2 w (6.15)
we get:
▽2 w = −4πf . (6.16)
This decomposition of f is called decomposition into the sum of a solenoidal and an
irrotational vector field.
Exercise 6.0.4 Express the vector field f =yzi+xzj + (xy − xz)k as a sum of an irrotational
and a solenoidal vector field. [Hint: The solution of (6.16) is
∫ ∫ ∫
f dv
w(P ) = (6.17)
∞ r
where r is the distance of the element of integration dv from the point P.]
Let us consider now a non - dissipative7 scalar wave in the x − y plane traveling down the
x-axis with velocity V and possessing a wave profile y = f (x) at time t = 0. Then at any time
t the profile of the wave is y(x, t) = f (x − V t), that is the form of the wave is propagating
with velocity V as the waves propagates. The equation of propagation of the wave is found as
follows:
∂ 2y ∂ 2 f (x − V t) ∂ 2y 2 ∂ f (x − V t)
2
∂ 2y 1 ∂ 2y
= , = V ⇒ − = 0. (6.18)
∂x2 ∂x2 ∂t2 ∂t2 ∂x2 V 2 ∂t2
Let us consider next the equation of motion of the strain and assume that the external force
vanishes, so that the equation of motion reads:
[ ]
E 1 d2 s ∂ 2s
▽ s+
2
▽ ·(▽ · s) = ρ 2 = ρ 2 (6.19)
2(1 + σ) 1 − 2σ dt ∂t
f µ = εµνρ Aν,ρ + ϕ,µ

and we assume that the vector
Aν = ενσκ wσ,κ
so that Aν,ν = 0 and
ϕ = −w,σ
σ
hence ενσκ ϕσ,κ = 0. The vector field wµ is an auxiliary vector field. Then we have:
f µ = εµνρ ενρσ wσ,κ − (w,σ ) = − (δσµ δκρ − δσρ δκµ ) wσ,κ − (w,σ
σ ,µ
) = −w.....κ
σ ,µ µ,κ
) − (w,σ
ρ ,µ
+ (w,ρ ) = −w.....κ
σ ,µ µ,κ
.
The above remain true if we consider a constant in front of the definition of Aν , ϕ.
7
Non-dissipative means that the wave preserves its form as it propagates. By this we mean that if at time
t = 0 its form at the origin is f (r) where r is the position vector of a point in space - this is a stigmiotipo or
form of the wave - then at a time t the wave (stigmiotipo) at the point r is f (r − Vt) where V is the velocity
of the wave.
39
We use now the result above that is, that every8 vector field can be decomposed into a sum
of a solenoidal and an irrotational vector. Let s1 , s2 be these vectors, that is s = s1 + s2 where
▽ · s1 = 0 and ▽×s2 = 0. Because equation (6.19) is linear in the variable s we assume (this
is not an innocent assumption but it is ‘logical’ since the continuum is linear and isotropic and
the stress strain relation is linearized!) that it is satisfied by both s1 , s2 independently (this is
an extra restriction on s1 , s2 ). This assumption implies the equations:
E d2 s1 ∂ 2 s1
▽2 s1 = ρ 2 = ρ 2 (6.20)
2(1 + σ) dt ∂t
[ ] 2
E 1 ∂ s2
▽2 s2 + ▽ ·(▽ · s2 ) = ρ 2 . (6.21)
2(1 + σ) 1 − 2σ ∂t
However from the identity (6.15) we have:
▽ × ( ▽ ×s2 ) = ▽ (▽ · s2 )− ▽2 s2 = 0 ⇒ ▽(▽ · s2 ) = ▽2 s2 .
Replacing in the second equation we find:
E(1 − σ) ∂ 2 s2
▽2 s2 = ρ 2 . (6.22)
(1 + σ)(1 − 2σ) ∂t
Equations (6.20), (6.22) are wave equations for the vectors s1 , s2 . √

E
Equation (6.20) defines a transverse wave moving with speed Vt = 2ρ(1+σ) . Equation (6.22)
represents a longitudinal wave which is propagating with speed 9 :
√
E(1 − σ)
Vl = . (6.23)
ρ(1 + σ)(1 − 2σ)
Let us prove that the second wave is longitudinal. We assume that the wave is traveling
along the x-axis. Then, because the wave is assumed to be non-dissipative we have:
∂w ∂v
s2 = s2 (x − V t) = u(x − V t)i + v(x − V t)j + w(x − V t)k ⇒ ▽ ×s2 = − j+ k = 0
∂x ∂x
so that ∂w
∂x
∂v
= ∂x = 0 which means that v, w are independent of x − V t. We are not interested
in constant displacements therefore s2 =u(x − V t)i i.e. the displacement s2 is parallel to the
direction of propagation of the wave i.e. the wave is longitudinal.
Working similarly we prove that the wave s1 is transverse (Exercise).
In general both types of waves are present and the net strain is found from their composition.
8
To be precise a vector f is uniquely determined if its divergence and curl are known throughout all of
space, provided that f tends to zero like r12 as r → ∞. See for example Vector and tensor Analysis by H. Lass
McGraw-Hill 1950.
9
We compute the velocity from the propagation equation of a non dissipative free wave, that is:
1 ∂2s
▽2 s− = 0.
V 2 ∂t2
Exercise 6.0.5 Show that if the external forces on the linear elastic isotropic continuum are
negligible (i.e. the continuum moves ‘freely’) and if the continuum is in a state of equilibrium
then the strain satisfies the following ‘equation of motion’ :
1
▽2 s + ▽ ·(▽ · s) = 0. (6.24)
1 − 2σ
Exercise 6.0.6 Write equations (6.24) in case the strain is radial, that is s =s(r)r.
Exercise 6.0.7 A coaxial cable is made by filling the space between a solid core of radius a
and a concentric cylindrical cell of internal radius b with rubber. If the core is displaced a
small distance axially, find the displacement in the rubber. Assume that end effects, gravity
and the distortion of the metal can be neglected and the continuum is a linear elastic isotropic
continuum.
6.0.13 The dynamic equation of motion for a linear elastic isotropic

continuum
In a linear elastic isotropic continuum the stress tensor has the form tµν = pδµν where p is the
isotropic pressure. For this continuum we find that the divergence tµν,ν = p,µ . This implies
the following:
a. The displacement vector of a linear elastic isotropic continuum cannot be arbitrary but
must be a solution of the equation (see (5.13)):
(1 − 2σ)p
eµν = − δµν .
E
that is, in a linear elastic isotropic continuum one can apply only bulk stresses and can have
strain motion, which is isotropic expansion or contraction only. For example a spherical elastic
isotropic continuum cannot change shape under the action of isotropic stresses and can either
contract to a smaller sphere or expand to a larger sphere.
b. The equation of motion for a strain motion of a linear elastic isotropic continuum is:
Rµ + p,µ = ρaµ (6.25)

where aµ is the “acceleration” of the displacement vector at a point inside the medium. As we
shall show (see (8.4)) the time derivative of any vector field f is given by the formula:
df ∂f
= + (v · ▽)f
dt ∂t
where v is the first time derivative of the position vector (i.e. the velocity). Using this we write
the equation of motion (6.25) as:
( )
∂v
R + ∇p = ρ + (v · ▽)v (6.26)
∂t
or, using decomposition (2.8) we have:
1 ∂v 1
(R + ∇p) = + ▽v2 − v × (▽ × v). (6.27)
ρ ∂t 2
41
Exercise 6.0.8 Show that the term v × (▽ × v) = 2v × ω or in covariant form (v × (▽ × v))µ =

−2εµνρ vν,ρ where εµνρ is the Levi Civita antisymmetric symbol and we have used (2.28). Then
prove that the equation of motion (6.26) can be written as follows:
1 µ ∂v µ 1 ∂v2
(R + p,µ ) = + − 2εµνρ vν,ρ . (6.28)
ρ ∂t 2 ∂xµ
We shall use this expression when we study the motion of perfect fluids.
Chapter 7
Worked examples on the stress tensor
We recall some basic results from the theory of real symmetric matrices.
The eigenvectors or principal directions ni of a square matrix Tij are defined by the
requirement:
Tij nj = λni
where the scalars λ are called the principal values or the eigenvalues of Tij for the corre-
sponding eigenvector. The eigenvalues are determined from the characteristic equation:
det(Tij − λδij ) = 0
which upon expansion gives:
λ3 − IT λ2 + IIT λ − IIIT = 0
where the coefficients IT , , IIT , IIIT are called the first, the second and the third invari-
ants respectively of the tensor Tij . These quantities are defined in terms of the invariant
(trace) T rT = Tii as follows:
IT = T rT = Tii
1( i j ) 1[ ]
IIT = Ti Tj − T ij Tij = (T rT )2 − T r(T 2 )
2 2
IIIT = det T.
Results on eigenvalues
1. For a symmetric tensor with real components the roots of the characteristic equation are
all real and the eigenvectors are real. In a Euclidian space the measure of the eigenvectors is
positive, therefore without loss of generality we can take them to be unit.
2. The eigenvectors of two distinct eigenvalues are mutually perpendicular.
3. If all eigenvalues are distinct the principal directions are unique and mutually perpendic-
ular, therefore they define an orthonormal Cartesian frame. If two eigenvalues are equal, then
there is only one direction associated with the third eigenvalue, which is unique. The remaining
two can be any two directions in the plane normal to that direction, which are mutually per-
pendicular. If all eigenvalues are equal then every set of right-handed orthogonal axes qualifies
as principal axes and every direction is a principal direction.
43
44 CHAPTER 7. WORKED EXAMPLES ON THE STRESS TENSOR
For every symmetric tensor with real components there is an associated symmetric matrix
with real entries. When we write the tensor in the frame defined by principal directions this
matrix is diagonal. If the components of the tensor are given in another frame then there is
always a transformation A, say, relating that frame with the frame of principal axes. This
transformation is called a similarity transformation, is not unique and it is given by the
formula:
T ∗ = AT At
where T ∗ are the components of the tensor in the principal frame and t denotes transpose.
Concerning the principal values - that is , the entries of the diagonal matrix - we have the
following results:
- The principal values and the principal directions of T and T t are the same
- The principal values of T −1 are the reciprocals of T and the principal directions are the
same
- The product tensors T Q and QT have the same principal values
- A symmetric tensor is said to be positive (negative) definite if all its principal values
are positive (negative); and positive (negative) semi-definite if one principal value is zero
and the others positive (negative).
Example 7.0.1 Consider the second rank symmetric tensor T whose matrix representation is:
 
5 2 0
[T ] =  2 2 0 
0 0 3
Determine its principal directions.
Solution
The characteristic equation is:
(3 − λ)(6 − λ)(1 − λ) = 0
therefore there are three distinct eigenvalues the λ1 = 3, λ2 = 6, λ3 = 1.
To determine the principal direction for the eigenvalue λ3 = 1 we use the equations:
(Tij − δij )nj(3) = 0
ni(3) n(3)i = 1.
 
n1
We assume nj(3) =  n2  and find from the first the two equations:
n3
4n1 + 2n2 = 0
2n1 + n2 = 0
from which follows n1 = n2 = 0. The second equation implies then n3 = ±1 hence nj(3) =
 
0
 0 .
±1
45
Working similarly with the other two eigenvalues we find the principal directions nj(2) =
   
± √25 ∓ √15
 ± √1  , nj =  ± √2  .
5 (3) 5
0 0
From these results it follows that the transformation matrix which diagonalizes the original
matrix is:  t   
n(3) 0 0 ±1
A =  nt(2)  =  ± √5 ± √5 0 
2 1
nt(1) ∓ √5 ± √5 0
1 2
which identifies two sets of principal direction axes, related by a reflection wrt the origin. It is
easily verified the A is Euclidian orthogonal (i.e. satisfies At A = I3 ), which should be expected1 .
Exercise 7.0.9 Consider the second rank symmetric tensor represented by the matrix:
 √ 
5 1 √2

[T ] = 1√ 5√ 2 
2 2 6
a. Show that the characteristic equation is λ3 − 16λ2 + 80λ − 128 = 0.

b. Prove that the eigenvalues are λ(1) = 8, λ(2) = λ(3) = 4
c. Show
 that the principal direction corresponding to the distinct eigenvalue λ(1) = 8 is
1
j 1 
n(1) = 2 √1
2
d. Concerning nj(2) choose any unit vector perpendicular to nj(1) . An obvious choice is nj(2) =
   
−1 −1
√1  1  . Finally define n 1
−1  .
(3) = n(1) × n(2) and show that n(3) = 2
j j j j
2
0 2
e. Show that the matrix:  1 1 1

√
2 2
 2

A =  − √12 √1
2
0 
− 12 − 12 √1
2
diagonalizes the matrix [T ].
Example 7.0.2 The components of the stress tensor σij at a point P of a continuum are given
by the following symmetric matrix:
 
21 −63 42
[σij ] =  −63 0 84 
42 84 −21
Determine:
1
The principal directions of different real eigenvalues are normal (in the Euclidian sense) to each other.
a. The stress vector on a plane through the point P having the unit normal n̂ = 17 (2ê1 −
3ê2 + 6ê3 )
b. The stress vector on a plane through the point P parallel to the plane defined by he three
points A(1, 0, 0), B(0, 1, 0) and C(0, 0, 2).
Solution    
21 −63 42 69
a. The stress vector tin̂ = [n̂]t [σij ] = 17 (2, −3, 6)  −63 0 84  =  54  or
42 84 −21 −42
ti = 69ê1 + 54ê2 − 42ê3 . b. The equation of the plane through the points A, B, C is AQ ·
n̂
(BA × CA) = 0 where Q(x1 , x2 , x3 ) is an arbitrary point of the plane. This implies that the
equation of the plane is:

x1 − 1 x2 x3
 −1 1 0 = 0 ⇒ 2x1 + 2x2 + x3 = 2.

−1 0 2


2
The normal to this plane is  2  (this can be found directly2 form BA × CA) so that the
  1
2
unit normal [n̂] = 31  2  . Hence the stress vector at P on this plane is:
1
 t    
2 21 −63 42 −14
1
tin̂ = [n̂]t [σij ] =  2   −63 0 84  =  −14 
3
1 42 84 −21 77
or in vector form tin̂ = −14ê1 − 14ê2 + 77ê3 .

We note that the stress vector at the same point P is different for two different planes
passing through the same point.
Example 7.0.3 The matrix representation of the stress tensor at the axes P x1 x2 x3 at the point
P is:  
1 3 2
[σij ] =  3 1 0 .
2 0 −2
Compute the components of the shear tensor in the frame P x′1 x′2 x′3 at the point P, which
are obtained from the x1 x2 x3 by rotation at 450 counterclockwise about the x3 axis.
Solution
The transformation matrix relating the two frames is:
 
cosθ sinθ 0

A = -sinθ cosθ 0 
0 0 1
2
Or use the equation of the plane in vector form n · (r − r0 ) = 0.
47
where θ = 450 . Replacing θ we find:

 
1 1 0
1
A = √  -1 1 0 
2 0 0 1
 √ 
4 0 √ 2
The components of [σij ] in the new basis are [σij′ ] = A[σij ]At =  √
0 −2√ − 2 .
2 − 2 −2
Example 7.0.4 The components of the stress tensor at a point P of a continuum wrt the axes
P x1 x2 x3 are given by the following matrix:
 
57 0 24

[σij ] = 0 50 0 .
24 0 43
Determine the principal stresses and the principal stress directions at P.
Solution
The characteristic equation of the matrix [σij ] is found to be:
(57 − σ)(50 − σ)(43 − σ) − (24)2 (50 − σ) = 0 ⇒ (50 − σ)(σ − 25)(σ − 75) = 0
therefore the principal stress values are σ(1) = 25, σ(2) = 50, σ(3) = 75. These are the eigenvalues
of the stress tensor.
Note: Verify the invariance of the trace (=I1 ): 25 + 50 + 75 = 57 + 50 + 43.
The principal stress values being different there are three different principal stress directions
at P. For the principal stress:
σ(1) = 25 we find the principal direction n̂(1) = ± 35 ê1 ± 45 ê3
σ(2) = 50 we find the principal direction n̂(2) = ±ê2
σ(3) = 75 we find the principal direction n̂(3) = ± 45 ê1 ± 35 ê3
The transformation matrix which transforms the matrix [σij ] to its principal axes (i.e.
diagonalizes the matrix) is:
 3 
±5 0 ± 54
A= 0 ±1 0 
±5 0
4
±5 3
Note: Verify that A[σij ]At = diag(25, 50, 75) = [σij∗ ].

Chapter 8
Fluids
Fluids are special types of deformable continua which we shall define in an exact physical sense
presently. Practically every deformable continuum can be considered as a fluid with certain
properties. Therefore it is better that we continue with the term fluid in a vague manner and
see if we end up with a definition as these notes develop. For the time being we shall consider
a fluid as an aggregate of a great number of particles (a continuum) moving under mutual
interaction and external forces.
Historically there have been two methods in the description of the motion of a fluid, the
Lagrange approach and the Euler approach.
a. The Lagrange approach.
In this approach one considers an arbitrary particle in the fluid and studies its trajectory. This
is what we did in the previous sections in the study of a deformable continuum. The Lagrangian
description requires four parameters to describe the fluid motion. One parameter is the time
of the particle1 and the rest three fix the trajectory of the particle (e.g. initial position). The
Lagrange approach describes the flow of the fluid in terms of the congruence of trajectories of
the fluid particles.
b. The Euler approach.
In this approach one concentrates at a point of space - which also requires three numbers to
be specified - at all times (the fourth parameter) and describes the physical properties of the
fluid by the value of the tensor fields corresponding to the physical properties of the fluid at
that point. Assuming a continuous behavior of the fluid for each property, these values define a
tensor field throughout the fluid. Therefore in the Euler approach the fluid motion is described
by the time evolution of relevant tensor fields. For example assume that a time moment t0 at
each point in the space occupied by the fluid the matter density is ρ(t0 , r) . Then at later
times the density of the fluid at the same point r changes and this change is the value of the
scalar field ρ(t, r) at that point.
In the following we shall adopt the Euler approach description of the motion of a fluid. Let
us give a first definition of the fluid as a working tool.
Definition 8.0.1 A fluid is a deformable continuum which:

1
This parameter is common to all particles due to the absolute nature of time in Newtonian Physics. This
is not the case for relativistic fluids where universal time does not exist and one has to introduce the concept
of synchronization. For this reason we shall consider relativistic fluids separately.
49
50 CHAPTER 8. FLUIDS
a. Occupies a restricted connected region of space

b. It cannot perform rigid motion alone, but it can perform strain motion alone. It can also
perform strain and rigid motion simultaneously.
c. Its physical properties are described by a number of differentiable tensor fields the most
important and fundamental being:
a. A scalar field ρ(t, r) which we call matter density or simply density
b. A vector field v(t, r) which we call the velocity field2 The velocity field describes
the flow of the fluid.
Before we continue we emphasize that in the Eulerian approach the tensor fields describing
the physical properties of a fluid are the values of fluid characteristics at each point in the
physical space occupied by the fluid and not properties associated with the particles of the
fluid. For example the velocity field v(t, r) at a point P (t, r) in space, is the velocity of a
particle of the fluid at the point P the moment t, and the value of the field v(t′ , r) is the
velocity of a different particle of the fluid passing through the same point P the time moment
t′ .
8.1 Mathematical concepts relevant to fluids

Consider a function f defined over the four-dimensional space t, x, y, z where t and x, y, z are
independent coordinates, which are functions of t i.e. x(t), y(t), z(t). In this case formula (2.9)
gives: [ ]
∂f ∂
df = (dr · ▽)f + dt = + v · ▽ f dt. (8.1)
∂t ∂t
where vdt = dr. We introduce the operator:
D ∂
= +v·▽ (8.2)
Dt ∂t
and write:
df (r(t), t) Df
= (r(t), t). (8.3)
dt Dt
D
We note that the operator Dt gives the total change that is, the change due to both the time
D
coordinate an the space coordinate. The operator Dt is called the mobile operator for reasons
which will be given below.
As an application of the mobile operator we consider its action on the position vector r of
a moving particle. We have:
Dr dr
= =v
Dt dt
where v is the velocity.
Similarly for the acceleration we find if we take f = v in (8.3):
Dv ∂v
a= = + (v · ▽)v. (8.4)
Dt ∂t
2
Sometimes it is also called the Eulerian distribution velocity. We shall not use this terminology.
8.1. MATHEMATICAL CONCEPTS RELEVANT TO FLUIDS 51
Using (2.8) we write the acceleration in the alternative form (compare with (??)):
Dv ∂v 1
a= = + ▽v2 − v × (▽ × v). (8.5)
dt ∂t 2
We shall use this result to write Newton’s second law for the motion of an accelerated
particle of a fluid.+
Note that the equation:
Df
=0
dt
for any function f (r(t), t) implies that f is a constant for a particular fluid element. It does
not preclude that different elements having different values of f ;it just implies that each such
element will retain whatever value of f has started with.
The equation (v · ▽)f = 0 which arises frequently implies that f is constant along a
streamline. It is emphasized that this equation offers no information at all about whether f
might be a different constant on different streamlines. Suppose for instance that the flow is
everywhere in the x−direction so that this equation reduces to ∂f ∂x
= 0.This equation says that
f is independent of x, but it contains no implication about how f might depend on y, z or t.
Exercise 8.1.1 Write the acceleration in the form encountering the rotating coordinate systems
(i.e. Coriolis acceleration etc.).
D
Proposition 8.1.1 The mobile operator Dt defines a derivation in Rn , that is ,it is R−linear
D
and satisfies the Leibnitz rule. The derivative defined by Dt we call the total derivative or
(especially in fluid dynamics) the material derivative.
Proof 8.1.1 R−linearity is obvious. Concerning the Leibnitz rule we prove only the case of
the product of a vector S with a scalar and leave the general proof to the reader.
D(ρS) ∂(ρS)
= + (v·∇)(ρS)
Dt ∂t
∂(ρ) ∂S
= S+ρ + S(v·∇)ρ + ρ(v·∇)S
∂t ∂t
Dρ DS
= S+ρ .
Dt Dt
If the components of the “ velocity field ” in the coordinate frame Σ with origin at P are
v =ui+vj+wk, then in that frame:
DS ∂S ∂S ∂S ∂S
= +u +v +w . (8.6)
Dt ∂t ∂x ∂y ∂z
We justify the name mobile operator. Consider a tensorial field S(r, t) which describes a
physical or geometric property of the continuum at the point P . We consider a change δt in
t and an induced change δr = vδt in r. Then the change3 of the value of S(r, t) (at the point
P !) is given by the difference:
δS(r, t) = S(r + vδt, t + δt) − S(r, t). (8.7)
3
We use δt and not dt because this change is measured at the same point P of space. Furthermore we use
δr and not dr because the change in r is defined in terms of the change in t. If it was independent of the t then
we should use dr!
52 CHAPTER 8. FLUIDS
Expanding S(r + vδt, t + δt) in terms of δt at the point P we find:

[ ] [ ]
∂S ∂S ∂S DS
δS(r, t) = +v 2
δt + O(δt ) = + (v · ▽)S δt + O(δt2 ) = δt + O(δt2 ). (8.8)
∂t ∂r ∂t Dt
It follows that for any tensor field S:

DS δS
= lim . (8.9)
Dt δt→0 δt
D
which justifies the name mobile operator for Dt .
Let us see an application of the mobile operator.
Example 8.1.1 A fluid flows from infinity along the x−axis towards the origin with constant
speed U and passes a fixed sphere of radius R centered at the origin. Given that the velocity field
of the fluid at the point with position vector r is v = −U (1+R3 r−3 )i+3R3 r−5 xU r (U =constant)
compute the acceleration of the fluid at the point r = bi (b > R) and evaluate the maximum
value |amax | of its magnitude for the various values of b.
Solution
Because the velocity field is independent of time (in such cases we say that the motion of
the fluid is steady) the acceleration is (see (8.4)):
Dv
a= = (v · ▽)v.
Dt
At r = bi the velocity v = (2R3 b−3 − 1)U i therefore v · ▽ =v ∂x
∂
where v = (2R3 b−3 − 1)U (this
observation saves us form computing all derivatives of v). We compute:
( ) ( )
∂v 3 −4 ∂r −5 −6 ∂r −5 ∂r
= −U −3R r +3U R r r − 5r
3
xr+xr .
∂x ∂x ∂x ∂x
∂r ∂r
But ∂x
= i, ∂x
= xr , so that at r = bi we have:
∂v
= −6U R3 b−4 i.
∂x
Therefore the acceleration at the point r = bi is:
a = 6U 2 (b3 − 2R3 R3 b−7 )i.
(b3 − 2R3 R3 b−7 ) = 0 and db

2
The maximum value of the magnitude |a| occurs when db d d
2 (b −
3
2R3 R3 b−7 ) < 0. These conditions imply b = (72)1/3 R and finally |amax | = 9( 7 )7/3 U 2 /R.
2
Chapter 9
Point Transformations defined by

vector fields
A point transformation on a manifold is a map which associates points of the manifold. There
are various ways to define a point transformation on a manifold M. One particular type of
point transformations is defined by the differentiable vector fields of the manifold. Practically
this is done by considering the integral curves of a vector field and defining a mapping relating
”nearby” points on one and the same integral curve of the vector field in a (sufficiently small)
open neighborhood. These (local) point transformations we call dragging along the vector
field and are defined as explained below.
9.0.1 Flow of a vector field

Consider a (differentiable) vector field1 X ∈ F 1 (M ) which in the local chart {xa } of M is
given by X = X a ∂a . At each point P ϵM, X determines (locally) a smooth curve2 γP (t) =
(c1P (t), c2P (t), ..., cnP (t)), tϵ(−a, a), aϵR, called the integral curve of X through P, as follows:
dcaP (t)
= X a (γP (t)), caP (0) = xa (P ). (9.1)
dt
This is an autonomous system of n first order linear differential equations which, by the
theorem of existence and uniqueness of solutions of ODEs, has, at least locally in M, a unique
solution. Obviously the vector field is tangent to its integral curves at all points. Due to the
uniqueness of the solution (we repeat locally! ) two different points sufficiently close in the
same neighborhood of M are either on the same integral curve or on two different smooth
non-intersecting integral curves.
Definition 9.0.1 The set of all integral curves of a vector field X in the neighborhood U ⊂ M
we denote σX (t, U ) and call it the local flow of the vector field X in U.
The integral curve of X through P we shall denote by σX (t, P ) where t ∈ (−a, a), a ∈ R,
is the parameter along this integral curve. It is always possible (by a diffeomorphism of R)
1
It is assumed that X has no zeros, that is, it does not vanish in a finite number of points.
2 i
cP = xi ◦ γP
53
54 CHAPTER 9. POINT TRANSFORMATIONS DEFINED BY VECTOR FIELDS
to change the parameter t so that σX (0, P ) = P. In what follows we shall assume that this is
always the case.
It is important to note that a vector field has two actions on the points of the neighborhood
U. One action is to collect points along one curve (any integral curve). The second action is to
divide U in a set of 1-dimensional submanifolds (the set of integral curves).
Let us see how one computes in practice the flow of a vector field.
Example 9.0.2 Consider the space R2 with coordinate functions (x, y) and the vector field
X = −y ∂x
∂ ∂
+ x ∂y ∂
. Find the flow of X. Do the same for the vector field X = y ∂x ∂
+ x ∂y .
Solution
Let γP (t) be a curve in the flow of X passing through the point P whose coordinates are
(x0 , y0 ). The coordinates of an arbitrary point of this curve are (x(t), y(t)) and the system of
autonomous first order equations defining the flow are:
dx
= −y
dt
dy
=x
dt
with the initial conditions x(0) = x0 , y(0) = y0 . The solution of this system is:
x(t) = x0 cos t − y0 sin t

y(t) = x0 sin t + y0 cos t.
We see that the√flow of X consists of all the circles centered at the origin (0, 0) and having
radius (OP ) = x20 + y02 .
∂ ∂
For the vector field X = y ∂x + x ∂y we find in the same way that the flow consists of the
hyperbolas given by:
x(t) = x0 cosh t + y0 sinh t

y(t) = x0 sinh t + y0 cosh t. 2
9.0.2 The point transformation induced by a vector field

The flow of a vector field covers locally any open sufficiently small3 neighborhood U of the
manifold. Therefore it is reasonable to expect that it would be possible to define a point
transformation between the points of U using the flow of the vector field. This relation will
depend on the specific vector field and, furthermore, different vector fields will lead to different
point transformations.
The flow line (or integral curve) σX (t, P ) of X through P is a one dimensional submanifold
γP (t) of M, which is diffeomorphic (that is, in 1:1 and C ∞ correspondence) with the real
numbers tϵ(−a, a), aϵR where t is a parameter along the integral curve. We use this simple
observation to define a point transformation first of σX (t, P ) and then of the flow σX (t, U ),
that is, of U .
3
Meaning that each point in U lies in one and only one integral curve of X, i.e. integral curves do not
intersect.
55
Let P ϵU be a point in U ∈ M with coordinate functions {xi }, and let X = X i ∂xi be a vector
field in U . Let σX (t, P ) be the integral curve of X through the point P , such that σX (0, P ) = P.
We define the point transformation:
σt (P ) : σX (0, P ) → σX (0, P )
P → σX (t, P ) = Q
where Q is the point in the same integral curve of X through P in the neighborhood U, defined
by the parameter value t.
The transformation σt (P ) is independent of the particular point P along the orbit (prove it).
However if t is large it is possible that either the point Q is outside U or that it is not uniquely
defined. Therefore we have to restrict the values of t to a sufficiently small interval (−ε, ε) so
that σ(t, P ) is a diffeomorphism of the integral curve σX (0, P ). Such an interval always exists
because X is differentiable.
Obviously the above apply to all orbits of X in U. This leads us to the following result.
Every vector field defines a local left action4 of the open interval (−ε, ε) of
the Abelian group R on the neighborhood U as follows:
σ(X) : (−ε, ε) × U → U
(t, P ) → σ(X)(t, P ) = σX (t, P )
This action induces a group of a one parameter family of transformations of the neighbor-
hood U defined as follows:
1. Define the family of point maps F P M ≡ {σt (X) t ∈ (−ε, ε)}:
σt (X) : U → U
P → σt (X)(P ) ≡ σX (t, P )
These point maps are diffeomorphisms of U which ‘drag’ P along the integral curve of X
through P.
2. In the set of all point maps FPM define the Abelian operation ◦ as follows:
◦ : FPM × FPM → FPM

σt (X), σs (X)(t, P ) → σt+s (X)
It is easy to show that the family of point maps F P M endowed with this binary operation
becomes an Abelian group, isomorphic to the Abelian group (−ε, ε). Indeed we have:
a. The identity map of this group is σ0 (X).
b. Concerning the inverse we have:
σt (X) ◦ σ−t (X) = σ0 (X) = I =⇒ (σt (X))−1 = σ−t (X). (9.2)
This group of maps acts on the points of the open neighborhood U creating a point trans-
formation in U . This point transformation is a one parameter point transformation which we
4
Local left action means that the action is defined by a subgroup of the action group only. The group here
is the Abelian group of the real line.
call the dragging along in U generated by the vector field X. The vector field itself we call
the infinitesimal generator of the dragging along point transformation. In the following for
economy in notation we shall write 1PGT for the one parameter group of point transformations.
It is important to note that under dragging along:
• A point stays always on the same integral line of the vector field
• The orbits of the 1PGT of point transformations are identical with the integral lines of
the vector field generating the group.
In the following when the vector field X is understood we shall write σt instead of σt (X).
9.0.3 The exponential map

We introduce the symbolic notation:
σt (X) ≡ etX . (9.3)
The reason for considering this notation is due to the fact that the general point transformation
defined by the flow of the vector field σt (X) depends differentially on t therefore we can compute
i
the derivatives d σdtt (X)
i , i = 1, 2, . . .. Consequently we can consider the Taylor expansion of σt (X)
in the neighborhood of a point P = σ0 (X) and write:
( ) ( n )
dσt (X) n d σt (X)
σt (X) = 1 + t + ... + t + ...
dt dtn
[[ t=0
2
] ] t=0
d 2 d
= 1 + t + t 2 + . . . + σt (X) |t=0
dt dt
[ ]
d [ ]
= exp(t )σt (X) |t=0 = etX σX (t) |t=0 .
dt
where we have set:

d d d2
etX = exp(t
) = 1 + t + t2 2 + . . .
dt dt dt
This notation is compatible with the properties of the one parameter group of point transfor-
mations defined by X. Indeed we have:
σt (X) ◦ σs (X) = σt+s (X) =⇒ etX ◦ esX = e(t+s)X .

One other formal property of the exponential function is its behavior wrt the differentiation we
have:
dσt (X) detX
= X(σt (X)) =⇒ = X(etX ).
dt dt
We see that formally the properties of the exponential function are satisfied, therefore one
can represent the flow σt (X) produced by the vector field X as etX .
The exponential map can also be thought of as the formal integration of the first order ODE
defining the flow lines of the vector field X.
57
9.0.4 The coordinate transformation defined by the induced trans-

formation
Each point transformation on a manifold induces a coordinate transformation in M defined by
the requirement that the components of the new point in the original coordinate system equal
the components of the original point in the transformed coordinate system, i.e. x̄i (P ) = xi (Q)
where Q = σX (t, P ) = σt (X)(P ). Because the point transformation is independent of the point
P , we consider a fixed value of t and define the coordinate transformation:
x̄i = xi + σt (X)i . (9.4)
where σt (X)i is the ith component of the vector valued function σt (X). This definition gives
for the point P :
x̄i (P ) = xi (P ) + σt (X)i (P ) = xi (Q)
by the definition of the map σt (X). Because the map σt (X) relates points on the same trajectory
we have the relation:
∂
σt (X) = X. (9.5)
∂t
From (9.4) we find for the Jacobian of the transformation:
( i) ( i)
x̄ ∂ x̄
J j
= = δjι + σt (X)i,j . (9.6)
x ∂xj
Using the Jacobian matrix one transforms tensor fields5 from the point Q to the point P by
means of the requirement:
T I (Q) = T̄ I (P ) (9.7)
where I is a collective index. This definition means that the components of the dragged tensor
field at the image point Q in the original (i.e. the xi ) coordinate system equal the components
′
of the tensor field T in the new coordinate system (i.e. the xi ) at the target point P. Now
using the coordinate transformation we have:
T̄ I (P ) = T J (P )JJI (P )
where we have assumed a collective index notation for the tensor and a corresponding appro-
priate product of Jacobian matrices. Therefore:
T I (Q) = T K (P )JK
I
(P )
But the point P = σ−t (Q) hence the definition (9.7) takes the form (the vector X is understood
and we omit it):
T I (Q) = T K (σ−t (Q))JK
I
(σ−t (Q)) (9.8)
For example for a vector field we have:
Ai (Q) = Ai (σ−t (Q))Jii′ (σ−t (Q)) (9.9)
If in the coordinate system {xi } the point Q has components r, t then P has components
σ−t (r) and relation (9.9) is written:
′
Ai (r,t) = Ai (σ−t (r,t))Jii′ (σ−t (r,t)). (9.10)
The last formula we shall use extensively in the following.
5
More general geometric objects.
9.0.5 The transformation dragging along

We consider a vector field X in the augmented n + 1 dimensional space {xi , t} (therefore the
components of X might depend on all coordinates xi , t) and its integral curve c(t) in the
n−dimensional space {xi } through a point P with coordinates {xiP } defined by the equation:
dci (t, P )
= X i (P, t)
dt
t is a parameter along the integral curve. We define the transformations σt (X) in the space
{xi } associated with the vector field X by the requirement:
σt (P )i = ci (t, P )
where c = (c1 (t, xi ), ..., cn (t, xi )) with the requirement that ci (t = 0) = xi (P ). This transfor-
mation we call the dragging along the vector field. For the dragging along the coordinate
transformation (9.4) becomes:
x̄i (P ) = ci (t, xi , P )
or, because it is independent of P :
x̄i = ci (t, xi ).
The Jacobian matrix of this transformation is (see (9.6)) :
( i) ( i)
ˆ x̄ ∂ x̄
J j
= = ci,j . (9.11)
x ∂xj
The Jacobian is:

det Jˆ = J = det(ci,j ) (9.12)
In general the Jacobian is a function of all coordinates {xi , t}.
We have the following result.
Example 9.0.3 Prove that the Jacobian determinant J(r, t) satisfies the relation (we assume
a three dimensional Euclidian space and write r for xi ):
∂
J(r, t) = J(r, t) [divX(σt (r), t)] (9.13)
∂t
Solution
Let the components of c(r,t) = ξ(r,t)∂x + η(r,t)∂y + ζ(r,t)∂z ) and let the components of the
vector field be X = Xx (r,t)∂x + η(r,t)∂y + ζ(r,t)∂z so that Xx = dξ dt
, etc. According to (9.11)
the Jacobian: ∂ξ ∂η ∂ζ
∂x ∂x ∂x
∂ξ ∂η ∂ζ
J(r, t) = ∂y ∂y ∂y

∂ξ ∂η ∂ζ
∂z ∂z ∂z
We assume J(r, t) ̸= 0, which means that we may take ξ, η, ζ as new coordinates along the
integral curves of X. Then we have X = Xx (ξ, η, ζ,t)∂ξ + η(ξ, η, ζ,t)∂η + ζ(ξ, η, ζ,t)∂ζ .
9.1. THE DRAGGING ALONG THE VELOCITY FIELD OF A FLUID 59
The determinant can be differentiated by recalling that it is linear in the columns or in the
rows. Thus, holding r = r0 fixed (that is we concentrate on the same integrable curve through
the same point P (say)) we have:
∂ ∂ξ ∂η ∂ζ ∂ξ ∂ ∂η ∂ζ ∂ξ ∂η ∂ ∂ζ
∂t ∂x ∂x ∂x ∂x ∂t ∂x ∂x ∂x ∂x ∂t ∂x
∂ ∂ ∂ξ ∂η ∂ζ ∂ξ ∂ ∂η ∂ζ ∂ξ ∂η ∂ ∂ζ

J(r, t) = ∂t ∂y ∂y ∂y
+ ∂y ∂t ∂y ∂y
+ ∂y ∂y ∂t ∂y (9.14)
∂t
∂ ∂ξ ∂η ∂ζ ∂ξ ∂ ∂η ∂ζ ∂ξ ∂η ∂ ∂ζ
∂t ∂z ∂z ∂z ∂z ∂t ∂z ∂z ∂z ∂z ∂t ∂z
where the derivatives are computed at the point r0 . Now using the property that the operators
∂ ∂ξ
∂t
and ∂x commute we have :
∂ ∂ξ ∂ ∂ξ ∂ ∂Xx ∂ξ ∂Xx ∂η ∂Xx ∂ζ
= = Xx (ξ, η, ζ, t) = + +
∂t ∂x ∂x ∂t ∂x ∂ξ ∂x ∂η ∂x ∂ζ ∂x
.
∂ ∂ζ ∂ ∂ζ ∂ ∂Xz ∂ξ ∂Xz ∂η ∂Xz ∂η
= = Xz (ξ, η, ζ, t) = + +
∂t ∂z ∂z ∂t ∂z ∂ξ ∂z ∂η ∂z ∂ζ ∂z
When these are substituted in (9.14) we get:
( ) ( ) ( )
∂ ∂Xx ∂Xx ∂Xx ∂Xy ∂Xy ∂Xy ∂Xz ∂Xz ∂Xz
J(r, t) = + + J+ + + J+ + + J
∂t ∂ξ ∂η ∂ζ ∂ξ ∂η ∂ζ ∂ξ ∂η ∂ζ
∂Xx ∂Xy ∂Xz
= J+ J+ J = (divX)J(r, t).
∂x ∂y ∂z
9.1 The dragging along the velocity field of a fluid

Consider a fluid and suppose that the velocity field (not the Eulerian velocity field v(r, t)!)
of the particle trajectories (the paths) is u(r, t). According to the last section the vector field
u(r, t), being a differentiable vector field, defines a one parameter group of transformations σt (u)
of the points in the region where the fluid flows. It is accustomed to denote the transformations
σ−t (u) as ϕt (we omit u because it is understood). We call this map the fluid flow map. The
dragging along generated by this map we call moving with the fluid or comoving. That
is, the map ϕt advances each fluid particle from its position at the time t to its position at
time t = 0. For example if W is a volume in the region D where the fluid flows, the volume
ϕt (W ) ≡ Wt is the volume moving with the fluid.
EDO MPAINEI SXHMA Dragging along of a fluid region Fig 1.1.4. p8 of
Marsden
Obviously the regions W and Wt contain at all times the same particles (both in number
and identity). This is the volume we use to define the equation of motion of a fluid.
Let the path of a particle of the fluid in Cartesian coordinates {x, t, z} be r(t) = (x(t), y(t), z(t)).
Then the velocity of the particle is u(t, x(t), y(t), z(t)) = (ẋ(t), ẏ(t), ż(t), t). The acceleration of
the particle is:
d ∂u ∂x ∂u ∂y ∂u ∂z ∂u
a(t) = u(t, x(t), y(t), z(t)) = + + +
dt ∂x ∂t ∂y ∂t ∂z ∂t ∂t
∂u Du
= + u · ∇u =
∂t Dt
that is , we recover again the material derivative. This derivative takes into account the fact
that the fluid is moving (this is the term ∂u∂t
) and also that the particles within the fluid move
i.e. change positions into the fluid (this is the term u · ∇u).
As we have discussed above the dragging along a vector field defines the coordinate trans-
formation (9.4) whose Jacobian is given by (9.11).
Example 9.1.1 Consider the vector field u = (1, 0, 0) and compute the dragging along it pro-
duces in the region of its definition.
Solution
Example 9.1.2 Let the integral curve of u be c = (x(t), y(t), z(t)). Then the integral curves of
u are the solution of the system of equations:
dx dy dz
= 1, = 0, = 0.
dt dt dt
Assuming the initial conditions (x0 , y0 , z0 ) we find:
x(t) = (1 + t)x0 , y(t) = y0 , z(t) = z0 .
The flow lines are straight lines parallel to the x−axis.

The dragging along map along the fluid lines relates the points P, Q by means of relation
(9.4). That is:
ϕt (r0 ) = σt (x0 , y0 , z0 ) = c(t) = (x(t), y(t), z(t)) = ((1 + t)x0 , y0 , z0 ) .
The fluid flow map for an arbitrary point with position vector r =(x, y, z) is ϕt (r) = ((1 + t)x, y, z) .
The Jacobian matrix of ϕt (r) equals
ˆ t) = ci = ϕt (r)i = diag(1 + t, 0, 0).

J(r, ,j ,j
Therefore the Jacobian J(r, t) of ϕt (r) equals
J(r,t) = det ϕt (r)i,j = det(diag(1 + t, 1, 1)) = 1 + t.
Because J(r, t) ̸= 0 the flow is incompressible (this will be justified below). We also compute:
∂t det J(r,t) = 1 ⇒ det J(r, t)divu = 1 ̸= 0 ⇒
divu ̸= 0.
which is another criterion that the flow is incompressible.

9.1. THE DRAGGING ALONG THE VELOCITY FIELD OF A FLUID 61
9.1.1 Incompressible flow

The Jacobian is used to compute the dragging of the volume element along the velocity vector
field u(r, t). That is, if dV (r0 ,0) is the element of volume at position r0 in the region W and
dV (ϕ(r0 ,t),t) the dragged along volume after time t in the dragged along region Wt then because
the dragging along induces a coordinate change with Jacobian J we have:
dV (ϕ(r0 ,t),t) = dV (r0 ,0)J(r0 ,t). (9.15)
In the following we shall write dVt for dV (ϕ(r0 ,t),t).

For a region W moving with the fluid we have for the volume:
∫ ∫
dVt = J(r0 ,t)dV (r0 ,0). (9.16)
Wt W
We say that a flow is incompressible if the volume of the comoving region is constant as
the fluid flows. The condition for this is:
∫ ∫
d d
dVt = 0 ⇔ J(r0 ,t)dV (r0 ,0) = 0. (9.17)
dt dt
Wt W
The region W is fixed therefore the derivative dtd becomes ∂

∂t
and enters the integral:
∫
∂
J(r0 ,t)dV (r0 ,0) = 0.
∂t
W
Since this is true for all regions W we have that the condition of incompressibility is:
∂
J(r, t) = 0. (9.18)
∂t
In order to find the implications of this condition on the velocity field we consider first the
following example.
9.1.2 Conservation of mass
Let ρ(ϕ(r0 ,0) be the density of the fluid in the region W. Then the mass of the fluid in W is:
∫
mW = ρ(ϕ(r0 ,0)dV (r0 ,0).
W
Assume that we keep the region W fixed so that mass of the fluid enters and leaves the
region W. The mass in the region W the moment t is:
∫
mW = ρ(ϕ(r0 ,t)dV (r0 ,t)
W
where r0 is the position vector of an arbitrary point in the region W. The mass of the fluid the
region W changes as follows:
∫ ∫
d d ∂ρ(ϕ(r0 ,t)
mW = ρ(ϕ(r0 ,t)dV (r0 ,t) = dV (r0 ,t). (9.19)
dt dt ∂t
W W
Let ∂W the boundary of W and let n be the outward normal at any of its points. If u is
the velocity of the fluid out of the region W the mass leaving the region W is given by
∫
(ρu) · ndA
∂W
where dA is the elementary surface element. The divergence theorem gives:

∫ ∫
(ρu) · ndA = ∇(ρu)(ϕ(r0 ,t)dV (r0 ,t).
∂W
W
The principle of conservation of mass states that the rate of change in the region W equals
the rate at which mass is crossing the boundary ∂W in the inward (i.e. opposite to n) direction.
This gives the equation:
∫ ∫
∂ρ(ϕ(r0 ,t)
dV (r0 ,t)= − ∇(ρu)(ϕ(r0 ,t)dV (r0 ,t) ⇒
∂t
∫ ( )
W W
∂ρ
+ ∇(ρu) (ϕ(r0 ,t)dV (r0 ,t). (9.20)
∂t
W
Because the region W is arbitrary this implies:

∂ρ
+ ∇(ρu) =0. (9.21)
∂t
Equation (9.20) is the integral form of the law of conservation of mass and equation (9.21) is
the differential form of the law of conservation of mass.
In order to understand the concept of dragging along the velocity field u better we consider
the flow of a mass out of a region W . This rate of change of mass in W flow of mass is
measured by the vector field ρu where ρ is the density of the fluid and u the particle velocity
field. We consider this volume dragged along by the vector field u and let Wt the volume after
time t.Then at that time the flow field will be (ρu)(ϕ(r, t),t) where ϕ(r, t) is the dragged along
point. According to (9.9) we have from the conservation of mass in the volume W the relation:
(ρu)(ϕ(r,t),t)dVt = (ρu)(ϕ(r0 ,0),0)J(r0 ,t)dV (9.22)
so that the flow out of the region Wt is related to the flow out of the region W by the formula:
∫ ∫
ρu(ϕ(r,t), t)dVt = [ρu(r0 ,0)]J(r0 ,t)dV. (9.23)
Wt W
9.2. THE TRANSPORT THEOREM 63
9.2 The transport theorem

This is a theorem of fundamental importance in the study of fluids and has as follows.
Theorem 9.2.1 (Transport theorem) Let f (r, t) be an arbitrary differentiable function (

well behaved so that all operations that follow can be performed well) and let u be the velocity
field of the fluid. Then for a comoving region Wt in the fluid and every differentiable tensor
field T we have: ∫ ∫
d DT
f T dV = f dV (9.24)
dt Dt
Wt Wt
′
Proof Ai (r,t) = Ai (σ−t (r,t), t)Jii′ (σ−t (r,t), t)
We give the proof for the momentum vector field ρu.
Let Wt be the transported region W along the particle velocity vector field u. The momen-
tum of the fluid in the region Wt is
∫
P Wt = (ρu)(ϕ(r,t)dV
Wt
From (9.10) we have:

(ρu)(ϕ(r,t), t) = (ρu)(r,t)J(r,t)
hence: ∫
P Wt = (ρu)(r,t)J(r,t)dV
W
dPWt
We compute the dt
. We have:
∫
dPWt d
= (ρu)(r,t)J(r,t)dV
dt dt W
∫
D
= [ρuJ] (r,t)dV
W Dt
∂ D
∫ ∫ ∂
because along the flow the ∂t gives Dt and dtd W = W ∂t because W is fixed. The term:
D Dρu DJ
[ρuJ] = J + ρu
Dt Dt Dt
Dρ Du
= uJ+ρ J + ρu∇uJ
Dt [Dt ]
Du ∂ρ
=ρ J+ u + u∇ρu J + ρu∇uJ
Dt ∂t
[ ]
Du ∂ρ
=ρ J+ u + u∇ρu+ρu∇u J
Dt ∂t
[ ]
Du ∂ρ
=ρ J+ u + u∇(ρu) J
Dt ∂t
The term in parenthesis vanishes due to the conservation of mass. Hence we find:
∫ ∫
dPWt Du Du
= ρ JdV = ρ dV
dt W Dt W t Dt
We write this equation as:

∫ ∫
d Du
(ρu)(r, t)dV = ρ (r, t)dV
dt Wt Wt Dt
which completes the proof.6

An implication of the transport theorem is the following.
Example 9.2.1 Let u be the velocity field of a flow and let C be a closed loop in the fluid. As
the flow develops this curve is carried with the fluid7 and assume that at the moment t it is the
curve Ct (see Figure We compute the lhs). Then the following relation holds:
∫ ∫
d Du
u · ds = · ds. (9.25)
dt Dt
Ct Ct
Proof
The answer can be found immediately form the transport theorem. However we give an
independent proof.
Let r(s) sϵ[0, 1] be the parametric description of C and let Ct (r,t) = ϕt (C(r0 ,0),t) be the
dragged curve with the fluid. Then a parametrization of Ct is ϕ(r(s),s) where sϵ[0, 1]. With
6
The change in momentum is due to the surface stresses, S∂Wt say,from the other parts of the fluid and the
extermal forces acting upon all the fluid (this is Newton’s second law). If f is the density of force then the
external force on the region force Wt equals
∫
f ρdV.
Wt
Then we have the equation:

∫
dPWt
= S∂Wt + f ρdVt .
dt Wt
Therefore Newtons second law for the momentum gives:

∫ ∫
Du
ρ dVt = S∂Wt + f ρdVt .
W t Dt Wt
7
A physical picture of this is that the molecules of the fluid on the curve C as the flow is developed move in
the fluid. Then at each instant t these molecules form another curve Ct . This curve is the one which is carried
with the fluid.
Mathematically the ‘carried with the fluid‘ can be understood as follows. Consider the integral curves of the
velocity field which pass through the points of the curve. These curves are the trajectories of the fluid particles
and are parametrized by the (Newtonian) time. Move along each integral curve the same parameter value ∆t
and join the points thus defined. The curve they form is the curve Ct .This ‘transport‘ defines the Lie transport
and it is the Lie derivative of the tangent vector field of the curve along the velocity field u.
9.2. THE TRANSPORT THEOREM 65
this parametrization we find:

∫ ∫1
d d ∂ϕ(r(s), s)
u · ds= u(ϕ(r,s), s) · ds
dt dt ∂s
Ct 0
∫1 ∫1
du(ϕ(r,s), s) ∂ϕ(x(s), s) d ∂ϕ(x(s), s)
= ds+ u(ϕ(r,s), s) · ds
dt ∂s dt ∂s
0 0
du(ϕ(r,s),s) Du(ϕ(r,s),s)
The dt
= Dt
and:
d ∂ϕ(x(s), s) ∂ ∂ϕ(x(s), s) ∂u(ϕ(r,s), s)
= =
dt ∂s ∂s ∂t ∂s
∂ϕ
where in the last step we have used that ∂t
= u. Then the term:
∫1 ∫1
d ∂ϕ(x(s), s)
u(ϕ(r,s), s) · ds = u2 (ϕ(r,s), s)ds = u2 (s = 1) − u2 (s = 0) = 0
dt ∂s
0 0
because the points with parameter value s = 0 and s = 1 coincide (the curve is closed).
Replacing we find the required result.
Let us see an interesting application of the Transport theorem.

Consider f = 1, T = g(r,t) i.e. T is a scalar. Then the Transport theorem gives:
∫ ∫
d Dg
gdV = dV (9.26)
dt Dt
Wt Wt
We say that the density f (a scalar) is conserved as the volume Wt flows with the fluid if
Df
Dt
= 0 along each flow line in Wt . From the above formula we find that a scalar quantity f is
conserved iff: ∫
f (ϕ(r,t), t)dVt = cont (9.27)
Wt
which can also be written: ∫

f (ϕ(r0 ,t), t)J(r0 ,t)dV = cont.
W
One such case is when f = 1 which implies that

∫
DJ ∂J
J(r0 ,t)dV = cont ⇒ = = J∇u =0
Dt ∂t
W
However this is the condition for a flow to be incompressible. Hence (assuming J(r0 ,t) ̸=0) we
find the alternative condition for incompressible flow (flow not fluid!):
divu =0. (9.28)

Equation (9.27) gives:

∫ ∫
f (ϕ(r,t), t)dVt = f (ϕ(r0 ,0), 0)dV
Wt W
But the lhs can be written:

∫ ∫
f (ϕ(r,t), t)dVt = f (ϕ(r0 ,t), t)J(r0 ,t)dV
Wt W
hence: ∫ ∫
f (ϕ(r0 ,0), 0)dV = f (ϕ(r0 ,t), t)J(r0 ,t)dV.
W W
But the region W is arbitrary therefore:
f (ϕ(r,t), t)J(r,t) =f (r,0). (9.29)
This equation shows how a conserved scalar quantity f (r,t) is evaluated along the flow line
of the velocity field (or a vector field in general) passing from the point with position vector r.
One important application is the mass density, which is conserved (mass conservation law).
This means that if the mass density at the point with position vector r is ρ(r,0) then at the point
with position vector ϕ(r,t) (along the integral curve of the velocity field) equals ρ(ϕ(r,t), t)J(r,t)
where J(r,t) is the Jacobian of the fluid flow map.
A scalar quantity f is homogeneous (in space) if f (r,0) =constant. If the quantity f is
a conserved density then Df Dt
= 0 ⇒ ∂f ∂t
+ div(f u) = 0.The divf = 0 because f is constant in
space therefore:
∂f
+ f divu = 0.
∂t
If the flow is incompressible then divu =0 and we conclude that ∂f ∂t
= 0,hence f is also
independent of t i.e. f is constant in both space and time. Note that if f is conserved but
it is not homogeneous (in space) then it is not necessarily homogeneous in time. This means
that a fluid which is homogeneous at t = 0 but it is compressible will generally not remain
homogeneous in the course of time.
Exercise 9.2.1 Use the Transport Theorem to prove for an incompressible fluid the following
formula [known as Reynold’s formula):
∫ ∫ ∫
d ∂f
f (r, t)dVt = (r, t)dVt + f u·dS. (9.30)
dt ∂t
Wt Wt ∂Wt
Proof
From the transport theorem we have:
∫ ∫ ∫ [ ]
d Df ∂f
f (r, t)dV = (r, t)dVt = +u∇f (r, t)dVt
dt Dt ∂t
Wt Wt W
∫ ∫t ∫
∂f
= (r, t)dVt + ∇(f u)(r, t)dVt − (f ∇u)(r, t)dVt
∂t
Wt Wt Wt
Because the flow is incompressible the ∇u = 0. Replacing we find the formula.

9.3. INTEGRAL PROPERTIES OF VECTOR FIELDS 67
9.3 Integral properties of vector fields

9.3.1 Flux through a surface
Definition 9.3.1 Suppose R(t, r) is some scalar or vector physical quantity per unit volume
(i.e. density), which is carried by the particles of the fluid in their motion and let S be a
smooth geometric surface in the fluid. We define the flux of the quantity R(t, r) through S by
the surface integral: ∫∫
R(t, r)dS (9.31)
S
where dS =dSn is the outward normal element vector area of S.
For various choices of the quantity R(t, r) we have different fluxes through the surface S.
For example
∫∫ if we take R(t, r) = 1 then we have the volume flux through S given by the
integral dS. If we take R(t, r) = ρ(t, r), where ρ(t, r) is the matter density of the fluid, the
S
resulting integral: ∫∫
ρ(t, r)dS (9.32)
S
is the mass flux through S. If we take R(t, r) to be the momentum density ρ(t, r)v of the fluid
then the integral: ∫∫
ρ(t, r)v·dS (9.33)
S
defines the momentum flux of the fluid through S. In the following we shall use these three
fluxes to study the kinematics and the dynamics of Newtonian fluids.
9.3.2 Flow and circulation along a path

Definition 9.3.2 Suppose u(t, r) is a vector field in a connected region V and let C be a
non-null smooth curve joining two points P, Q in V. The flow of the vector field along the path
C from point P to the point Q is defined to be the following line integral:
∫ Q ∫ Q
dr
u(t, r) · dr = u(t, r) · ds (9.34)
CP CP
ds
where ds is arc length along the curve C .
If the vector field is irrotational there exists a velocity potential ϕ(t, r) so that u = − ▽ ϕ.
Then the flow of u is independent of the curve C joining the points P, Q and equals8 ϕ(Q)−ϕ(P ).
If the path C is closed the flow of the vector field u is called the circulation round C. If a
single valued velocity potential exists then (obviously) the circulation around any closed curve
vanishes. If the velocity potential is many valued the circulation is not necessarily zero for all
possible closed curves within the fluid.
8
Because ▽ϕ · dr = dϕ.
Chapter 10
Characterization of fluids and flows
Kinematically the fluids are classified by the properties of the Eulerian velocity field. We
consider the following types of flow of a fluid.
• The flow of the fluid is called irrotational iff there is a scalar field ϕ(t, r), called velocity
potential,
H such that v =-▽ϕ. Equivalently the velocity field is irrotational. In this case
v · dr = 0 for any closed smooth path which lies entirely into the fluid.
• The flow of the fluid is called steady or stationary iff ∂v ∂t

= 0 that is, the velocity field
(as a vector) at a point in space is constant in time. If the motion of a fluid is not steady
we call it unsteady.
Dynamically the fluids are classified in terms of the dynamic fields describing the physical
properties of the fluid. We consider the following basic types of fluids.
• A fluid is called a dust if it is characterized dynamically completely by one scalar field,

the matter density ρ(t, r). Physically a dust is an aggregate of particles which do not
interact, that is the forces in the interior of the fluid (i.e. the stresses) vanish.
• A fluid is called homogeneous (in space) if the matter density ρ(t, r) =constant in space.
• A fluid which has the properties of a perfect deformable material (that is, if we place a
small surface anywhere in the fluid and in any direction the stress force on the surface
is always normal to the surface) we call a perfect fluid. Dynamically a perfect fluid is
characterized completely by the scalar fields ρ(t, r) and p(t, r). During the motion of a
perfect fluid all stress (i.e. internal) forces are normal to the displacement of the particles
(they do not consume energy) therefore a perfect fluid cannot conduct heat and cannot
be stirred. Furthermore it is isotropic but not necessarily homogeneous.
• Any fluid which is not a perfect fluid we call an imperfect fluid. Physically an imperfect
fluid is a fluid in which the stress forces between the particles during its flow are not normal
to the (Eulerian) velocity of the fluid, therefore imperfect fluids can conduct heat and can
be anisotropic. Furthermore imperfect fluids require more tensor fields than the mater
density ρ(t, r) and the pressure p(t, r) in order to be described dynamically.
69
70 CHAPTER 10. CHARACTERIZATION OF FLUIDS AND FLOWS
10.0.3 Geometric description of the flow of a fluid

In a fluid we consider two types of lines.
a. The lines of flow or pathlines which are the trajectories in space of the individual
particles of the fluid or, equivalently, the integral lines of the particle velocity field u and
b. The streamlines, which are the integral curves of the (Eulerian) velocity vector field
v(t, r) of the fluid. In general the pathlines and the streamlines do not coincide. If they do,
the flow of fluid is called steady i.e. ∂v ∂t
= 0. If they do not the flow is called unsteady. In
the latter case the streamlines form a continuously changing pattern.
A stream surface in a fluid is a surface in space with the property that at every of its points
the (Eulerian) velocity v(t, r) is tangent at that point. Obviously a stream surface consists of
streamlines. The intersection of two stream surfaces in the fluid gives a stream line.
Given a closed curve in the fluid a stream surface is formed by drawing the streamline
through every point of the curve. A stream surface whose cross sectional area is infinitesimally
small we call a stream filament.
10.0.4 Calculating the streamlines

The streamlines are the integral curves of the (Eulerian) velocity field. From the definition
v = dr
dt
we have that the vectors v and dr are parallel, therefore the quotient of their components
in any frame of reference is the same. This implies that if in an arbitrary frame we have the
analysis v = vx (t, r)i + vy (t, r)j + vz (t, r)k and dr = dxi + dyj+dzk then the streamlines (in
that frame!) are given by the solution of the system of simultaneous equations:
dx dy dz
= = . (10.1)
vx (t, r) vy (t, r) vz (t, r)
This type of systems of simultaneous equations are called Lagrange systems.
A first integral of these equations must be of the form f (t, r) = constant. This equation
defines a stream surface in the fluid. Its intersection with a second independent solution,
g(t, r) =constant say, of the Lagrange system gives a streamline in the fluid.
Example 10.0.1 The velocity field of a fluid is v = i × r + cos atj + sinatk where a is a con-
stant (̸= 1). Find the streamlines and the pathlines. Discuss the special case a = 0.
Solution
We write r =xi + yj+zk and compute i × r = −zj+yk. Therefore the velocity field is:
v = (−z+ cos at)j + (y + sinat)k.
The Lagrange system corresponding to this velocity field is:
dx dy dz
= = .
0 −z+ cos at y + sinat
One solution of the system is x =constant, that is one family of stream surfaces are planes
normal to the x−axis. To find a second family of stream surfaces we consider the last two terms
and write:
(y + sinat)dy = (−z+ cos at)dz.
71
Integrating we find the second family of stream surfaces:
y 2 + z 2 + 2y sin at − 2z cos at = G
where G is a constant. This is a new family of stream surfaces which consists of elliptical
cylinders with axis the x − axis. The streamlines are the intersection of these two solutions/
families of stream surfaces. If a ̸= 0 the streamlines form a continuously changing pattern,
the motion being time dependent. If a = 0 then the flow is steady (because in that case
v = (−z + 1)j+yk is independent of t ) and the streamlines are fixed circles given by the
equations x =constant and y 2 + z 2 − 2z = G.
The pathlines - that is the trajectories of the particles of the fluid are given by the functions
x(t), y(t), z(t) and are found from the solution of the system of equations:
dx dy dz
= 0, = −z+ cos at, = y + sinat.
dt dt dt
The solution of the first is x =constant. To solve the other two equations we differentiate
the first and replace dz
dt
from the second. We end up with the equation:
d2 y
+ y = −(a + 1) sin at.
dt2
Assuming that a ̸= 1 the solution is:
y(t) = A cos t + B sin t + C sin at
where A, B are arbitrary constants and C = 1/(a − 1). Replacing this in the second equation
we find:
z(t) = A sin t − B cos t − C cos at.
In the special case a = 0 we find easily y 2 + (z − 1)2 =constant so that the pathlines coincide
with the streamlines as expected, because then the flow is steady.
Example 10.0.2 In a coordinate system the velocity field of a fluid is given in terms of its
2 2
components as follows u = − cr2y , v = cr2x , w = 0 where c is a constant (̸= 0) and r denotes
distance from the z−axis. Prove that the stream lines are given by the equations x2 + y 2 =
constant, z = constant, w = 0. Also prove that the velocity field admits a velocity potential
and then show that the function ϕ(x, y, z) = −c2 tan−1 ( xy ) is a velocity potential.
Solution
The velocity v of the fluid is:
c2 y c2 x
v=− i + j.
r2 r2
The stream lines are the solution of the system of equations:
dx dy dz
2 = c2 x
= .
− cr2y r2
0
The solution of the first equation is x2 + y 2 =constant, and the of the last equation
z =constant. These two equations are the equations of the stream lines of the fluid.
72 CHAPTER 10. CHARACTERIZATION OF FLUIDS AND FLOWS
To show that there exists a velocity potential, ϕ say, it is enough to show that curlv = 0.
This is left as an easy exercise to the reader. To compute the velocity potential ϕ we write
(r2 = x2 + y 2 + z 2 ):
c2 y c2 x ∂ϕ c2 y ∂ϕ c2 x
▽ϕ = −v = 2 i− 2 j ⇒ = 2, =− 2 .
r r ∂x r ∂y r
The second equation gives:
∫
x y
ϕ = −c 2
2
dy = −c2 tan−1 ( ) + C(x).
r x
Replacing in the first equation we find that C(x) = const. = C. The constant is not important
for the velocity potential and finally we have that ϕ = −c2 tan−1 ( xy ).
Example 10.0.3 Prove that ϕ = xf (r) (r2 = x2 + y 2 + z 2 , f ′ ̸= 0) is a possible form for

the velocity potential of an incompressible fluid motion. Given that the fluid speed v → 0 as
r → ∞, deduce that the surfaces of constant speed are given by (r2 − 3x2 )r−8 =constant.
Solution
The velocity v of the fluid is:
r
v = − ▽ ϕ = −f i − xf ′
r
where a prime over a letter denotes differentiation wrt r. We compute:
( ′)
xf ′ xf
▽ · v= − ▽ f i − ▽ ·r − r · ▽
r r
′
( ′ ( ′ )′ )
∂f f f f r
=− − 3x − r · i+x
∂x r r r r
( ′
)
f
= −x f ′′ + 4 .
r
The condition for an incompressible fluid is ▽ · v =0. This gives the condition (f ′ ̸= 0):
f′ f ′′ 4
f ′′ + 4 = 0 ⇒ ′ + = 0.
r f r
The solution of this equation is f (r) = − 13 Ar−3 + B where A, B are integration constants.
For this value of f (r) we find that the velocity of the incompressible fluid is:
( )
A Ax
v= 3
− B i − 5 r.
3r r
The values of the integration constants are fixed from the asymptotic conditions. We
note that when r (→ ∞ the speed ) v → −Bi so that B = 0 and v = 3rA3 (i − 3x r2
r). The
( ) 2 2 2 2 2
speed v 2 = 3rA3 1 − 6rx
r2
+ 9rr4x = 9r A
8 (r − 3x)
2
therefore v 2 =constant on the surfaces
r−8 (r − 3x)2 =constant .
Chapter 11
The equation of continuity -

conservation of mass (again)
The equation of continuity is a constraint equation, which restricts the motion of a fluid. It
is not a dynamical equation and it quantifies the assumption that the total mass of a fluid is
conserved during its motion.
In order to write down the equation of continuity we consider a simply connected region R
of space within the mass of the fluid, with volume V and bounded by a (smooth) surface S. We
also consider an elementary volume dV at a point P inside R with∫position ∫vector r at the time
moment t0 . The mass within the volume V at the moment t0 is dM ≡ ρ(t0 , r)dV. As the
R R ∫ ∫
fluid moves the mass (in the same region R!) changes. The overall change1 is dtd dM = ∂ρ ∂t
dV.
R R
This change in mass consists of the following parts:
- The mass which leaves and enters the region R . If at the point P of the surface S (enclosing
the region R ) the velocity of the fluid is v and the element of surface is dS, this change is given
by (see Figure conservation of mass):
dM dl
= −ρ · dS = −ρv · dS.
dt dt
H
The overall change is − ρv·dS where v·dS < 0 for mass entering the region R and v·dS > 0
S
for mass leaving R.
- The mass which is created by sources or destroyed in sinks within the volume V. We
measure this mass with a function ψ(t, r)dV, where we assume ψ(t, r) to∫ be positive for sources
and negative for sinks. The overall change due to sings and sources is ψ(t, r)dV.
R
Conservation of mass in the region R is expressed by the equation:
∫ ∫ I
∂ρ
dV = ψ(t, r)dV − ρv·dS. (11.1)
∂t R
R S
1 d
We enter dt into the integral because the integrating region is independent of time t.
73
74CHAPTER 11. THE EQUATION OF CONTINUITY - CONSERVATION OF MASS (AGAIN)
From Stoke’s theorem (divergence theorem) we have:

I ∫
ρv·dS = ▽(ρv)dV.
R
S
which is the integral form of the equation. Because R is arbitrary, equation (11.1) implies:
∂ρ
+ ▽(ρv) =ψ(t, r). (11.2)
∂t
This equation we call the continuity equation (for matter). 2
The continuity equation simplifies for various special types of fluids as follows:
a. If there are no sinks or sources, ψ(t, r) = 0 and the continuity equation reads:
∂ρ
+ ▽(ρv) =0. (11.3)
∂t
b. If in addition to that, the fluid is incompressible, then ρ =constant and equation (11.3)
becomes:
▽v =0. (11.4)
c. If in addition to a. and b. the velocity of fluid is irrotational, then there exists a velocity
potential ϕ such that v = − ▽ ϕ and the continuity equation becomes:
▽2 ϕ=0 (11.5)
which is Laplace’s equation.

In an incompressible fluid ∆V = 0 when ∆p ̸= 0 , that is, a certain mass within a volume3
V comoving with the fluid moving under the influence of external forces (resulting in a gradient
of pressure) is possible to change its shape but not its volume, which remains constant.
(x −y )z
Exercise 11.0.1 Show that the velocity field v = (x−2xyz
2 2
y
2 +y 2 )2 i + (x2 +y 2 )2 j + x2 +y 2
k is a possible
motion for an incompressible fluid. Is this motion irrotational?
Hint: Prove that ▽v =0. Check if ▽ × v =0.
Exercise 11.0.2 Write the continuity equation in spherical and cylindrical, coordinates.
Answer
Spherical:
∂ρ 1 ∂ 1 ∂ 1 ∂
+ 2 (ρvr r2 ) + (ρvθ sin θ) + (ρvϕ ) = ψ(t, r, θ, ϕ). (11.6)
∂t r ∂r r sin θ ∂θ r sin θ ∂ϕ
Cylindrical:
∂ρ 1 ∂ 1 ∂ ∂
+ (ρvr r) + (ρvθ ) + (ρvz ) = ψ(t, r, θ, z). (11.7)
∂t r ∂r r ∂θ ∂z
2
Every scalar field associated with the fluid which is conserved has a similar continuity equation.
3
A terminology for this volume is material volume and in descriptive terms ‘dyed blob’.
75
Exercise 11.0.3 If the velocity of a fluid is radial, that is, u = u(r, t), show that the equation
of continuity is written as follows:
∂ρ ∂ρ ρ ∂
+ u + 2 (r2 u) = ψ(t, r). (11.8)
∂t ∂r r ∂r
1
Solve this equation for an incompressible fluid if ψ(t, r) = 1/r2 and show that u = ρr .
Solution
The continuity equation in spherical coordinates is given by (11.6). In the case of radial
velocity (that is, uθ = uϕ = 0) it takes the form:
∂ρ 1 ∂
+ 2 (ρur2 ) = 1/r2
∂t r ∂r
where we have replaced ψ(t, r, θ, ϕ) with 1/r2 .
In the case the fluid is incompressible ρ =constant and the last equation reduces further to:
∂ 1
(ρur2 ) = 1 ⇒ u(r) = .
∂r ρr
11.0.5 The kinetic energy

Consider a region W in a fluid in which the mass density is ρ(r, t) and the particle velocity
field u(r, t). We define the kinetic energy EKin of the the mass in the region W of the fluid to
be the integral: ∫
1
EKin = ρ(r, t)u2 (r, t)dV (11.9)
2
W
As the region W moves with the fluid the kinetic energy changes and at time moment t it
is given by: ∫
1
EKin (t) = ρ(r, t)u2 (r, t)dV
2
Wt
where now the position vector r refers to points in Wt .The change in kinetic energy is:
∫
dEKin 1d
= ρ(r, t)u2 (r, t)dV
dt 2 dt
Wt
and applying the Transport Theorem (9.24) for f = ρ, T = u2 we find:

∫
dEKin 1 Du2 (r, t)
= ρ(r, t) dV.
dt 2 Dt
Wt
D
The operator Dt
is a differential operator therefore:
2
( ) [ ]
Du Du ∂u ∂u
= 2u· = 2u· + (u · ∇)u = 2 u· + u·(u · ∇)u
Dt Dt ∂t ∂t
from which follows: ∫ [ ]
dEKin ∂u
= ρ u· + u·(u · ∇)u dV. (11.10)
dt ∂t
Wt
76CHAPTER 11. THE EQUATION OF CONTINUITY - CONSERVATION OF MASS (AGAIN)
Chapter 12
The equation of motion of a perfect

fluid
We consider a perfect fluid which is acted upon by an external force of density f (that is force
per unit mass of the fluid). We consider an elementary volume dV in the fluid with surface area
S enclosing a mass dM = ρdV of fluid. We assume that Newton′ s second Law applies that is,
the change of momentum of the mass dM in time equals the total applied external force and
derive the equation of motion.
First we compute the total force on the mass dM. This force is due to the stress force
(i.e. the force from the remaining mass of the fluid on the boundary of dV of dM ) and to the
external force applied to the fluid as a whole (these latter are called body forces).
The stress force is given by:
I ∫
− pdS = − ▽pdV = − ▽ pdV
S=∂dV dV
where the minus sign is due the fact that we consider dS to point outwards of the surface S
and p is compression and, furthermore, we have used the fact that the fluid is perfect therefore
the force is normal to the surface dS.
Concerning the second force this equals f dM where f is the force per unit mass (=force
density).
In conclusion the total force on the mass dM is f dM − ▽pdV . The momentum of the mass
dM is dM v hence Newton’s second law gives:
D(dM v)
= (ρf − ▽p)dV.
dt
D(dM v)
The term1 dt
= dM Dv
dt
= ρ Dv
dt
dV and we obtain finally:
Dv 1
= f − ▽ p. (12.1)
dt ρ
1
The mass dM remains constant in the volume dV as the fluid moves due to the conservation of mass and
the assumption that the volume dV is comoving i.e. at all times contains the same number and type of pericles.
77
78 CHAPTER 12. THE EQUATION OF MOTION OF A PERFECT FLUID
This is Euler’s equation of motion for an elementary material volume within a perfect
fluid. To find Euler’s equation in terms of the standard derivatives we replace Dv
dt
from (8.5)
and find:
∂v 1 1
− ▽v2 + v × (▽ × v) = f − ▽ p. (12.2)
∂t 2 ρ
12.0.6 Euler equation for special types of incompressible flow

In the following we discuss Euler’s equation for a perfect fluid assuming that the flow is incom-
pressible, that is, ∇ · u = 0.
The total energy Etotol of a fluid element moving in a domain D with velocity u and contained
in a region W ⊂ D is the sum of the kinetic energy Ekin given by (11.9) and the internal energy
Eint due to the intermolecular potentials and the internal vibrations of the molecules of the
fluid which we cannot see macroscopically. If energy is supplied to the fluid or if the fluid is
allowed to do work Etotol changes. Let us consider an incompressible flow in which the internal
energy remains constant and that the energy is supplied to the fluid by the pressure and the
internal forces only. Then the first assumption gives:
∫ [ ]
dEtotal dEKin ∂u
= = ρ u· + u·(u · ∇)u dV (12.3)
dt dt ∂t
Wt
from (11.10) and the second:

dEtotal
= Wpres + Winternal (12.4)
dt
where:
I
Wpres = − pu·dA (12.5)
∂Wt
∫
Winternal = ρu · fdV. (12.6)
Wt
f is the density of internal force (i.e. force per unit mass of the fluid). From the divergence
theorem we find:
I ∫ ∫
Wpres = − pu·dA = − ∇ · (pu)dV = − ∇p · udV
∂Wt Wt Wt
Replacing in (12.3) we find for all Wt :

∫ [ ] ∫ ∫
∂u
ρ u· + u·(u · ∇)u dV = − ∇p · udV + ρu · f dV ⇒
∂t
Wt Wt Wt
[ ]
∂u
ρ + (u · ∇)u =−∇p + ρf ⇒
∂t
Du
ρ = −∇p + ρf .
Dt
We note that this equation is the same as the general Euler equation. This means that the
condition dEdt
total
= dEdtKin is equivalent to the condition of incompressibility, that is,
79
dEtotal dEKin
if the flow is such that dt
= dt
then the flow must be incompressible
(unless p = 0).
Assume that the external force is conservative with a potential function χ and that the
fluid is homogeneous and the flow is incompressible (hence ρ =constant). Then the rhs of (12.2)
is written − ▽ (χ + ρp ) and the equation of motion for an incompressible fluid in a conservative
force field is:
∂v p 1
− v × (▽ × v) = − ▽(χ + + v2 ). (12.7)
∂t ρ 2
If in addition the motion is:
a. Irrotational, then ▽ × v = 0 and the equation of motion reduces to:
∂v p 1
= − ▽(χ + + v2 ) (12.8)
∂t ρ 2
∂v
b. Steady, then ∂t
= 0 and the equation of motion reduces as follows:
p 1 2
v × (▽ × v) = ▽ (χ + + v ) (12.9)
ρ 2
From (12.9) follows that for a perfect fluid in steady motion:

[ ]
p 1 2
v· ▽(χ + + v ) = 0 (12.10)
ρ 2
This equation implies that the Eulerian velocity field v (not the velocity field of a fluid particle
!) is always tangent to the surface χ + ρp + 12 v2 =constant, therefore this surface is a stream
surface. We have proved the following result.
Proposition 12.0.1 (Bernoulli Theorem) For a homogeneous perfect fluid of density ρ

under pressure p, which flows with stationary (or steady) incompressible flow under the action
of conservative (external) forces with potential function χ the expression χ + ρp + 21 v2 is constant
along the streamlines. If the fluid moves freely (that is χ = 0) then ρp + 12 v2 is constant along
the streamlines and if in addition the fluid is incompressible the ρ =constant and p + 12 ρv2 is
constant along the streamlines (the standard Bernoulli Theorem) and an increase of speed of
the fluid at a point demands a decrease of pressure and conversely.
Example 12.0.4 Assume a perfect fluid is at rest and show that in this case ▽ × (ρf ) = 0,
hence f · ▽ ×f = 0. This is a necessary condition for equilibrium of a perfect fluid. Why ρf
must be the gradient of a scalar if equilibrium is to be possible?
Proof
Dv
From the Euler equation of motion for a perfect fluid at rest (i.e. v = 0 and dt
= 0) we
have:
1
f = ▽ p ⇒ ρf = ▽ p ⇒ ▽ × (ρf ) = 0 ⇒ ρf = − ▽ ϕ
ρ
where ϕ is a potential function. This proves the last part of the example.
Now we use the identity of vector calculus:
▽(ϕf ) = ϕ ▽ ×f + ▽ ϕ × f (12.11)
and compute:
▽(ρf )= ρ ▽ ×f + ▽ ρ × f = 0 ⇒ ρf · ▽ ×f + f · ▽ ρ × f = 0.
But f · ▽ ρ × f = ▽ ρ · f ×f = 0 therefore f · ▽ ×f = 0.
Exercise 12.0.4 A perfect fluid (which is not necessarily incompressible therefore ρ is not
constant) moves under the action of conservative force with potential function χ. If the flow of
the fluid is irrotational with velocity potential ϕ, show that the equation of motion becomes:
( )
∂ϕ p 1 2
= − χ + + v + C(t) (12.12)
∂t ρ 2
where C(t) is a smooth function. If in addition ρ = ρ(p) (this is called an equation of state
about which we shall speak later) then equation (12.12) becomes:
∫
∂ϕ 1 2 dp
+ v +χ+ = D(t) (12.13)
∂t 2 ρ
where D(t) is another smooth function.
Example 12.0.5 A sphere moves through an incompressible perfect fluid with constant velocity
v = v0 k̂. Find the velocity of the fluid relative to the sphere assuming that the motion of the
fluid is steady and irrotational.
Solution
Because the fluid is incompressible ρ = const. Because the flow of the fluid is steady and
irrotational:
∂v
= 0, v = ∇ϕ
∂t
where ϕ is a velocity potential. The incompressibility gives:
∇ · v = 0 =⇒ ∇2 ϕ = 0.
At the surface of the sphere the radial velocity of the fluid vanishes2 . This implies

∂ϕ
v|r=a = 0 =⇒ = 0.
∂r r=a
The velocity of the fluid relative to the sphere is v = ∇ϕ and in the coordinate system Σ in which the
2
sphere has velocity v0 k̂ the velocity of the fluid equals:
vΣ = ∇ϕ + v0 k̂.
.
81

We have to solve Laplace equation ∇2 ϕ = 0 with the boundary condition ∂ϕ = 0.
∂r r=a
2
We know two solutions of Laplace equations, the r and the 1/r hence we try a solution of
the form: ( )
B
ϕ(r) = Ar + 2 cos θ
r
∂ϕ
(we do not take solutions of the form ϕ(r, t) because the motion is steady, hence ∂t
= 0). Then:
( )
∂ϕ 2B a3 A
= A − cos θ = 0 =⇒ B = .
∂r r=a a3 2
We impose a second boundary condition by requiring that at infinite distance from the sphere
the velocity of the fluid vanishes, therefore the velocity relative to the sphere will be −v0 k̂.
This gives the boundary condition:

∂ϕ
= −v0 .
∂z z→∞
We compute: ( )
∂ϕ ∂ϕ
= . . . =⇒ = A = −v0 .
∂z ∂z z→∞
Therefore the solution is: ( )
a3
ϕ(r) = −v0 r + 2 cos θ.
2r
Exercise 12.0.5 A perfect fluid revolves uniformly without change in form (i.e. the relative
displacement of its particles does not change; rigid body motion) with angular velocity ω around
the z- axis. The flow is steady (because ∂v∂t
= 0) hence the Eulerian velocity field coincides with
the particle velocity field. Hence the acceleration filed is Dv
dt
.
1. Show that the velocity of a particle of the fluid with coordinates (x, y, z) is v = (−ωy, ωx, 0).
Prove that ∇v = 0, hence the flow is incompressible. Find the Eulerian velocity field and
prove that there does not exist a velocity potential. Compute the acceleration field (of the
particle),
2. Show that Euler’s equations of motion are:
1 ∂p
−ω 2 x = X −
ρ ∂x
1 ∂p
−ω 2 y = Y −
ρ ∂y
1 ∂p
0=Z−
ρ ∂z
where X, Y, Z, are the components of the external force density F acting on the fluid.
Show that from these equations we have:
1
dp = F · dr + ω 2 (xdx + ydy). (12.14)
ρ
Explain the physical meaning of this equation.
3. Show that for a homogeneous fluid and for a conservative force with potential function χ
equation (12.14) becomes:
p 1 2 2
− ω (x + y 2 ) + χ = constant.. (12.15)
ρ 2
Hint: Note that because the flow in incompressible and the fluid is homogeneous ρ is
constant.
4. Suppose that the only external force is gravity and show that equation (12.15) becomes:
p v2
− + gz = constant
ρ 2
where v = ωr, r2 = x2 + y 2 are the speed and the distance of the particle from the z- axis.
Draw a picture to describe the motion of the fluid.
Chapter 13
Thermodynamics and hydrodynamics
Newtonian continuum mechanics is concerned with the motion of continuous deformable me-
dia in three- dimensional Euclidian space and uses the notion of absolute simultaneity inher-
ent in Newtonian Physics. Thus Newtonian hydrodynamics is formulated in terms of four
-independent variables which may be taken to be the three Cartesian coordinates in the Eu-
clidian 3-space {xi } i = 1, 2, 3 and the (absolute) time t.Hydrodynamics is concerned with the
description of the state of the fluid as a function of these four independent variables. The
state is described macroscopically by five dependent variables, which may be taken to be two
thermodynamic variables, such as pressure p and the mass density ρ and three kinematic ones
such as the components of the Eulerian velocity field.
Fluids are deformable continua which consist of many fluid elements and each fluid element
consists of many particles. The state of matter in a given fluid element is determined thermody-
namically, that is in terms of a small number of variables called thermodynamic variables.
In Newtonian thermodynamics we introduce the following five quantities, which are functions
of r, t and describe macroscopically the state of the system:
p = pressure,
ρ =mass density,
T =temperature,
S =entropy,
W = enthalpy (=total energy).
In general only three thermodynamic variables are independent due to the First and the
Second Law of Thermodynamics.
Let us consider a fluid consisting of one type of particles only, and assume that the indepen-
dent variables are the S, V, N where V is the volume of the fluid and N the number of particles.
Then the enthalpy W (S, V, N ) and the combined form of the First and the Second Law is:
dW = T dS − pdV + µdN (13.1)
where µ is the new function µ is called the chemical potential of the fluid. Each term in this
equation has the following physical interpretation:
a. T dS is the ‘internal kinetic energy’ of the fluid element. At absolute temperature T = 0
this energy vanishes and the molecules of the fluid element are still in space.
83
84 CHAPTER 13. THERMODYNAMICS AND HYDRODYNAMICS
b. pdV is the mechanical work which can be produced by the fluid element if it is allowed to
expand in volume dV under constant external pressure p (i.e. forces applies on the boundary
of the fluid).
c. µdN is the ‘internal’ potential energy of the fluid element , that is, the energy stored in
the molecules of the fluid element and remaining at T = 0.
The enthalpy W is the total energy of the fluid element. From (13.1) follows that the
temperature, the pressure and the chemical potential in terms of the independent variables
S, V, N are determined as follows:
( ) ( ) ( )
∂W ∂W ∂W
T = ,p = − ,µ = . (13.2)
∂S V,N ∂V S,N ∂N S,V
The thermodynamic variables are classified in extensive and intensive. A thermodynamic

variable X is called extensive if it changes its value to kX when the independently variables
Y change their value to kY and intensive if it is not extensive. For example consider as
independent variables the S, V, N.Then as N → kN (the other two remaining constant) the
W → kW so that W is an extensive variable. The temperature T is obviously an intensive
variable. This invariance1 allows us to find a relation among the thermodynamic variables
known as Euler relation (also as fundamental relation) as follows.
Let W̃ = kW, S̃ = kS, Ṽ = kV, Ñ = kN be the extensive variables after a change of the
system size. The thermodynamic laws give for the new values of the variables:
dW̃ = T dS̃ − pdṼ + µdÑ
ReplacingW̃ , S̃, Ṽ we find:
dW̃ = T dS̃ − pdṼ + µdÑ

= k(T dS − pdV + µdN ) + (T S − pV + µN )dk
= kdW + (T S − pV + µN )dk
But from W̃ = kW we also have:
dW̃ = kdW + W dk.
Comparing these relations we infer the Euler relation:
W = T S − pV + µN. (13.3)
The extensible variables are the W, S, N therefore we define the specific variables, that
is the variables per unit mass. We have then the specific enthalpy s = S/M and the specific
entropy W/M where M is the mass of the fluid. In terms of the specific variables equation
(13.1) reads:
1
w = T s − p + µn (13.4)
ρ
1
Invariance means symmetry and each symmetry leads to a constraint which can be used to eliminate one
of the degrees of freedom of the system.
85
where n = N/M is the particle number density of the fluid element. This equation contains
only intensive thermodynamic variables (that is, independent of the size of the system) and it
is used to reduce the number of variables required for a complete specification of the state of
the fluid element form three to two.
We may choose therefore as free variables the s, p and write for the rest of the variables:
w(s, p), ρ(s, p), µ(s, p).

The first law of thermodynamics becomes in terms of these variables:
1
dw = T ds + dp (13.5)
ρ
the temperature and the mass density now being given by the equations:
( ) ( )
∂w ∂w
T = , ρ = 1/ . (13.6)
∂s p ∂p s
The type of fluid is described by specifying its specific internal energy ε which is defined
as follows:
p
ε=w− . (13.7)
ρ
Each specification of ε is called a caloric equation of state and prescribes the specific internal
energy ε as a a function of pressure and density. The quantity of fluid is described by the number
of particles in each fluid element.
Using as free variables the s, ρ (so that the pressure p(s, ρ) and the specific enthalpy w(s, ρ))
we find from (13.5) :
p
dε = T ds + 2 dρ. (13.8)
ρ
It follows that in this case: ( )
2 ∂ε
p=ρ .
∂s ρ
13.0.7 Isentropic flow

The flow of a fluid is called isentropic if ds = 0. Physically this condition means that during
the flow the ‘internal kinetic energy‘ of the fluid (i.e. the entropy) does not change. For example
the temperature of the fluid remains constant. For such a flow the first law of thermodynamics
(13.5) implies:
1 1
dw = dp ⇔ gradw = gradp (13.9)
ρ ρ
that is, a flow is isentropic if the specific enthalpy satisfies (13.9). If for an isentropic flow the
pressure is a function of ρ only (and not of s!) then the specific enthalpy:
∫ρ
dp/dλ
w= dλ. (13.10)
λ
For isentropic flow the internal energy becomes:

( )
p ∂ε
dε = 2 dρ ⇒ p = ρ2 (13.11)
ρ ∂ρ s
and if p is a function of ρ only:

∫ρ
p(λ)
ε= dλ. (13.12)
λ2
Using (13.9) we find that for isentropic flow Euler equation of motion reads:
Dv
= f − ∇w. (13.13)
dt
Exercise 13.0.6 Assume that the flow of a fluid is isentropic and the internal forces vanish.
Show that in this case if W is a fixed volume in space (not a volume comoving with the fluid!)
the equation of conservation of energy gives:d
∫ ( ) ∫ ( )
d 1 2 1 2
ρv + ρε dV = − ρ v + w u · dA. (13.14)
dt 2 2
W ∂W
( )
Due to this result the quantity ρ 12 v2 + w u is called the energy flux vector. Note that
in Bernoulli Theorem this vector is preserved.
The Circulation Theorem

The Circulation Theorem concerns the isentropic flow of a perfect fluid. Let u be the velocity
field of a flow and let C be a simple oriented closed contour in the fluid.
As the flow develops this curve is carried with the fluid2 and assume that at the moment t
it is the curve Ct .The circulation ΓCt around the curve Ct is defined by the integral:
∫
ΓCt = u · ds (13.15)
Ct
We have the following result.
Theorem 13.0.1 (Kelvin’s circulation theorem) For isentropic flow without external forces
the circulation ΓCt is constant in time.
2
A physical picture of this is that the molecules of the fluid on the curve C as the flow is developed move in
the fluid. Then at each instant t these molecules form another curve Ct . This curve is the one which is carried
with the fluid.
Mathematically the ‘carried with the fluid‘ can be understood as follows. Consider the integral curves of the
velocity field which pass through the points of the curve. These curves are the trajectories of the fluid particles
and are parametrized by the (Netonian) time. Move along each integral curve the same parameter value ∆t
and join the points thus defined. The curve they form is the curve Ct .This ‘transport‘ defines the Lie transport
and it is the Lie derivative of the tangent vector field of the curve along the velocity feold u.
87
Proof 13.0.1 Taking into account (9.25) we find:

∫ ∫
dΓCt d Du
= u · ds = · ds
dt dt Dt
Ct Ct
But for isentropic flow without external forces Euler equation gives Du
Dt
= −∇w, hence:
∫
dΓCt
= − ∇w · ds =w(1) − w(0) = 0
dt
Ct
because Ct is closed.
Suppose now that the contour C is the boundary of a surface S. Then Stoke’s theorem gives:
∫
ΓCt = u · ds =S (∇ × u)·dS =S ω·dS
Ct
where ω is the vorticity vector. Thus we have the following corollary of the circulation
theorem.
For isentropic flow without external forces the flux of the vorticity across a surface moving
with the fluid is constant in time.
The integral curves of the vorticity vector are called vortex lines. Vortex lines generate
surfaces which we call vortex sheets. A vortex line is the intersection of two vortex sheets.
We have the following result (see Figure vertex lines and vertex sheet):
If the flow is isentropic then the vortex lines and the vortex sheets are preserved.
Euler’s equation for isentropic flow can be written in terms of the vorticity vector. Indeed
we have:
∂u
+ (u · ∇)u = f − ∇w
∂t
But we have the identity:
1
∇(u · u) = u × (∇ × u)+(u · ∇)u = u×ω+(u · ∇)u
2
hence:
∂u 1
+ ∇(u · u) − u×ω =f − ∇w.
∂t 2
Taking the curl and using that the curl of the div vanishes we find:
∂ω
− ∇ × (u×ω) =∇ × f
∂t
From the identity:
∇ × (A × B) =A(∇ · B) − B(∇ · A)+(B·∇)A−(A·∇)B

we have:
∇ × (u×ω) =u(∇ · ω) − ω(∇ · u)+(ω·∇)u−(u·∇)ω.
Now applying the property divcurlA = 0 for any differentiable vector field A to the vector
field ω=curlu we find ∇·ω=0, hence Euler’s equation in terms of the vorticity vector becomes:
∂ω
+ u(∇ · ω) + ω(∇ · u)−(ω·∇)u=∇ × f ⇒
∂t
Dω
+ ω(∇ · u)−(ω·∇)u=∇ × f .
Dt
The continuity equation gives:
∂ρ Dρ
+ ∇(ρu) = 0 ⇒ + ρ∇ · u =0.
∂t Dt
We find now:
[ ]
D(ω/ρ) Dω 1 1 Dρ Dω 1 1 1 Dω
= − ω= + ω(∇ · u) = + ω(∇ · u)
Dt Dt ρ ρ2 Dt Dt ρ ρ ρ Dt
1
= [∇ × f +(ω·∇)u]
ρ
and finally the equation of motion for the vorticity field is:
D(ω/ρ) ω 1
− ( ·∇)u = ∇ × f . (13.16)
Dt ρ ρ
A result which follows form the vorticity equation is the following.
Consider the quantity G(r, t) = ∇ϕt (r, t)·ω(r, 0) where ϕt is the flow map. We compute:
∂G(r, t) ∂ ∂ϕt (r, t)

= ∇ϕt (r, t) · ω(r, 0) = ∇ · ω(r, 0) = (∇·u(ϕt (r, t))) · ω(r, 0)
∂t ∂t ∂t
= (∇·u(r, t))∇ϕt (r, t) · ω(r, 0) = (∇·u(r, t))G(r, t) = G(r, t)·u
We note that the quantity G(r, t) and ω/ρ satisfy the same equation in the case ∇ × f i.e.
either the external forces are absent or they are conservative. This implies that:
For isentropic flow in the absence of external forces (or for conservative external
forces) the quantity:
(ω/ρ)(ϕt (r, t), t) = ∇ϕt (r, t) · (ω/ρ)(r, 0) (13.17)
that is, the value of the quantity (ω/ρ)(r, 0) propagates with the divergence of the
flow map.
Chapter 14
The Navier Stokes equation
As it is the case with deformable materials real fluids are not perfect, which means that during
their motion develop stress forces (friction like) between adjacent layers and if a test surface is
immersed in a fluid the net force on the surface is not normal to the surface. This implies that
we have to describe the physical properties of a real fluid by means of more tensor fields than
the mass density field ρ(t, r) and the velocity field v(t, r). A fluid which is not perfect we call
it an imperfect fluid.
The Navier Stokes equation is the equation of motion of a viscous fluid (to be defined below)
under the action of stress and external (body) forces. In order to understand the Physics behind
this equation it is important that we reconsider Euler’s equation of motion of a perfect fluid
i.e. :
Dv
ρ = ρf − ▽p. (14.1)
dt
The term ρ Dvdt
is the kinematic part (i.e. mass times acceleration) of Newton’s second law. It
contains the density instead of mass because we use the density of force and not the force itself.
The term ρf is the total external force density acting on the fluid (recall that f is force per unit
mass) and last the term − ▽ p is the pressure thrust on the surface of an elementary volume
within the fluid (p is the isotropic pressure inside the fluid). If the external force vanishes the
fluid moves under the action of the pressure term − ▽ p only.
How can we understand a perfect fluid in the context of deformable continua? We claim
that the perfect fluid is a deformable continuum on which the only possible stress tensor is the
bulk stress given by the relation:
tµν = −pδµν (14.2)

where p is the isotropic pressure. This means that the only strain motion which can take
place under stress in a perfect fluid is contraction and expansion. In order our claim to be
valid it must conform with Euler’s equation of motion. As we have seen the force resulting
from the application of the stress (giving rise to the corresponding strain) is f µ = tµν
,ν . Placing
tµν = −pδµν in this expression we find f = −p = −(▽p) which proves that our choice of
µ ,µ ,µ
tµν is correct.
The strain tensor of a perfect fluid can be computed from (5.3) if we set tµν = −pδµν . We
find:
ẽµν = −p(ϕ + 3ξ)δµν = −pβδµν (14.3)
89
90 CHAPTER 14. THE NAVIER STOKES EQUATION
where β is a coefficient characteristic of the material of the perfect fluid. We note that for
a perfect fluid the principal axes of strain and stress coincide. From a geometric point of
view we infer that the strain and the stress metric eµν and tµν respectively are conformally (or
homothetically if p =constant) related to the Euclidean metric δµν .The homothety factor for
the stress metric tµν is the isotropic pressure − 21 p and for the strain metric the factor − 12 pβ.
We consider now the case of a real fluid. A real fluid differs form a perfect fluid in the
property that rapidly moving adjacent layers of the fluid tend to drag along the slower layers of
fluid, and, conversely, the slower layers tend to retard the faster layers. In other words there is
a ”friction” amongst the various layers in the fluid which is due to forces which are tangent to
the relative velocity of the surfaces. The phenomenon that the forces between adjecent moving
fluid surfaces is not normal to the relative velocity of the surfaces at each point of their interface
we call viscosity. Viscosity is the transfer of the usual concept of friction experienced by the
solid bodies in relative motion. Concerning the value of the viscosity it has been found by
experiment that (for standard fluids) the force due to viscosity is directly proportional to the
common area A of the layers and to the gradient of the relative velocity normal to the flow.
We determine the relation between the stress and the strain in a real fluid. We restrict our
considerations to real fluids described by linear elastic isotropic continua, which under bulk
stress alone behave as perfect fluids. These type of continua we shall call viscous fluids.
According to the previous sections the stress and the strain tensor of a viscous fluid must
be related by equation (5.3), that is:
∑
tµν = 2ηẽµν + λδµν ẽρρ (14.4)
ρ
where η, λ are coefficients (not necessarily constants) characteristic of the fluid. The stress
tensor tµν we call the viscous stress tensor. Because for a perfect fluid the stain tensor is
proportional to δµν we rewrite this equation as follows:
[ ] ( ) ∑
1 2
tµν = 2η ẽµν − T race(ẽ)δµν + λ + η δµν ẽρρ
3 3 ρ
that is, we brake the strain tensor in a trace and a traceless part. The first part takes care the
“perfect fluid” character of the fluid and the second the(viscous) character
∑ of the fluid.
Because the bulk stress is given by −pδµν we set λ + 32 η ẽτ τ = −p, where p is the
τ
isotropic pressure within the fluid and we write for the stress tensor of a viscous (not a general!)
fluid:
1
tµν = 2η(ẽµν − ẽδµν ) − pδµν (14.5)
3
∑
where ẽ = ẽρρ and ẽµν is the strain tensor of the viscous fluid. The coefficient η we call the
ρ
coefficient of viscosity. A perfect fluid is defined by the requirement η = 0, therefore the
viscous fluids are a distinct class containing perfect fluids.
From (14.5) we compute the force density on a viscous fluid by making use of equation (6.4).
We have:
1 ,µ
..,ν = 2η(ẽ ,ν − ẽ δµν ) − p .
F µ = tµν µν ,µ
(14.6)
3
91
where we have assumed η,µ = 0 i.e. η is constant. We recall (see (6.5)) that the strain tensor
in terms of the strain vector for a general strain motion of a linear elastic isotropic continuum
is given by the expression:
1
ẽµν = (uµ,ν + uν,µ ). (14.7)
2
Replacing in (14.6) we find:
2 ( ρ ),ν
F µ = tµν ν − δµν ) − p,µ .
µ,ν ν,µ
..,ν = η(u ν +u u
3 ,ρ
In standard vector notation this is written as:
2
F = η(▽2 u − ▽ ·(▽ · u)) − ▽p. (14.8)
3
If the density of the applied external force on the fluid is f µ , Newton’s second law gives for
the equation of motion of a viscous fluid:
Duµ
ρ = ρf µ + F µ (14.9)
dt
where F µ is the force density due to the stress on the fluid. Using (14.8) this equation can be
written:
Duµ η 2 ( ρ ),ν 1
= f µ + (uµ,ν ν + uν,µ ν − u,ρ δµν ) − p,µ (14.10)
dt ρ 3 ρ
or in standard vector notation:
[ ]
Du η 2 1
= f + ▽ u − ▽ ·(▽ · u) − ▽ p.
2
(14.11)
dt ρ 3 ρ
Du
or, replacing dt
:
[ ]
∂u η 2 1
+ (u · ∇)u = f + ▽ u − ▽ ·(▽ · u) − ▽ p.
2
(14.12)
∂t ρ 3 ρ
Equation (14.10) or (16.4) is the Navier Stokes equation for the motion of a viscous fluid.
The quotient ηρ is called the Kinematic coefficient of viscosity. We note that for a perfect
fluid η = 0 so that this equation reduces to Euler’ s equation of motion (12.1).
For an incompressible viscous fluid ▽ · v =0 equations (14.10) and (16.4) become respec-
tively:
Dv µ η 1
= f µ + (uµ,ν ν + uν,µ ν ) − p,µ (14.13)
dt ρ ρ
Du η 1
= f + ▽2 u − ▽ p. (14.14)
dt ρ ρ
This equation of motion and the continuity equation are the equations which govern the motion
of an incompressible Newtonian viscous fluid.
92 CHAPTER 14. THE NAVIER STOKES EQUATION
Example 14.0.6 Prove Archimedes’ Law : The force exerted by a static fluid on a body im-
mersed in the fluid (without changing its shape) equals the weight of the displaced fluid and
points in the direction opposite to the force of gravity.
Solution
Because the body (a) is at rest and (b) does not change its shape the velocity v µ and the
stress tensor tµν vanish (the fluid need not be perfect i.e. η does not necessarily vanish). Then
the Navier Stokes equation reads:
ρf µ = p,µ
where f µ is the external force density (i.e. force per unit mass). The only external force acting
on the fluid is the force of gravity, whose density is the acceleration of gravity g µ . Then we find
that the gradient of pressure is:
p,µ = ρg µ .
Suppose the body has a volume V with boundary the surface S. Then the force on the body
is: ∫
F = − pnµ dS
µ
where p is the isotropic (inward) pressure in the mass of the fluid and the (inward) pressure
on the surface of the body. dS is the (outward) element of surface of the body. Applying the
divergence theorem we find:
∫
F = − pnµ dS = −V p,µ dV = −gVµ ρdV = −g µ m
µ
where m = V ρdV is the mass of the fluid in the volume occupied by the body.
Chapter 15
The electromagnetic field as a viscous

fluid
15.0.8 The stress tensor of the electromagnetic fluid

The viscous fluid is a state of mater and not a specific substance like, for example water or oil.
In this section we shall study the electromagnetic field from the point of view of a viscous fluid,
that is, we shall consider a fluid of “photons“ and assuming that the fluid satisfies Maxwell
equations we shall compute the stress tensor and from the equations of motion we shall find
the momentum of the electromagnetic field etc.
In our calculations we shall need identity (2.7) which we rewrite here for convenience:
▽v2 = 2v × (▽ × v)+2(v · ▽)v (15.1)
and in tensor notation (see (2.12)) :
(δ ij ui uj ),k = 2ηikjl uj,l ui + 2ui uk,i .
Maxwell equations in SI units for the electromagnetic field in a material are written as follows1 :
▽·B=0 (15.2)
▽·D=ρ (15.3)
∂D
▽ × H= j+ (15.4)
∂t
∂B
▽ × E= − (15.5)
∂t
In 3-d tensor notation these equations are written as follows:
µ
B,µ =0 (15.6)
µ
D,µ =ρ (15.7)
ν,ρ
ηµνρ H = jµ + Dµ,t (15.8)
ηµνρ E ν,ρ = −Bµ,t . (15.9)
1
Equation ▽ · E = ερ0 in general is ▽ · D = ρ.
Recall that for empty space B = µ0 H and D = ε0 E.
93
94 CHAPTER 15. THE ELECTROMAGNETIC FIELD AS A VISCOUS FLUID
In empty space the catastatic relations are D =ε0 E, B = µ0 H where ε0 is the electric per-
meability and µ0 is the magnetic permeability of the empty space. We have the relation
ε0 µ0 = c12 .
As it is well known the quantity c12 E2 + B2 is the energy density of the electromagnetic
field. We will compute the div of this quantity using identity (15.1) and replacing ▽ × E and
▽ × B from Maxwell equations. We find for the electric field:
( )
∂B
▽E = 2E × (▽ × E)+2(E · ▽)E =2E× −
2
+ 2(E · ▽)E
∂t
∂(E × H) ∂E
= −2µ0 +2 ×B + 2(E · ▽)E
∂t ∂t
∂S ∂E
= −2µ0 +2 ×B + 2(E · ▽)E (15.10)
∂t ∂t
where:
S = E × H = ϵµνρ Eν Hρ (15.11)
is the Poynting vector. Similarly for the B2 we have:
( )
∂E
▽B = 2B × (▽ × B)+2(B · ▽)B =2B×µ0 j + ε0
2
+ 2(B · ▽)B
∂t
2 ∂E
= −2µ0 (j × B)− 2 ×B + 2(B · ▽)B. (15.12)
c ∂t
Adding (15.10) and (15.12) we find:
( )
1 1 2 µ0 ∂S 1
▽ 2E + B + 22
= −µ0 (j × B)+(B · ▽)B+ 2 (E · ▽)E. (15.13)
2 c c ∂t c
It is easier if we continue with tensor formalism. The term:

ρ ν
(E · ▽)E =E µ ∂µ E ν = ∂µ (E µ E ν ) − (∂µ E µ )E ν = ∂µ (E µ E ν ) − E (15.14)
ε0
and similarly:
(B · ▽)B =B µ ∂µ B ν = ∂µ (B µ B ν ) − (∂µ B µ )B ν = ∂µ (B µ B ν ). (15.15)

( µ ν ( ))
E E 1 µν E2 ∂S ν
∂µ +B B − δ
µ ν
+ B2
+ µ 0 = −µ0 (ρE + j × B)ν .
c c 2 c2 ∂t
The term on the rhs is the Lorentz force density2 (in SI units) F = ρE + j × B, therefore
we obtain the final form:
( µ ν ( ))
1 E E 1 µν E2 ∂S ν
ν
f + ∂µ +B B − δ
µ ν
+ B 2
= − . (15.16)
µ0 c c 2 c2 ∂t
2
The current j = ρv for empty space. For materials the rhs contains more terms.
95
Equation (15.16) is the equation of motion for the ”electromagnetic fluid”.

We define the symmetric second rank tensor:
( )
Eµ Eν 1 µν E2
µν
T = +B B − δ
µ ν
+B 2
(15.17)
c c 2 c2
and the equation of motion becomes:

1 µν ∂S ν
fν + T ,µ = − . (15.18)
µ0 ∂t
This equation coincides with the equation of motion of a viscous fluid (see equation (14.9)
if we identify the Lorentz force F µ with the external force and the tensor T µν with the stress
tensor of the electromagnetic fluid. In that case the vector S ν is the momentum density of the
fluid provided (S · ∇)S = 0.
It is important to note that if there are no charged particles (ρ = 0) then there is still a
change in the fluid momentum, which is due to the stress T µν . In other words if we consider
the charges as being the ”particles” of the electromagnetic fluid then there is still a part of
the fluid which survives. This part of the fluid is the electromagnetic field whose ”particles“
have momentum but zero mass and zero charge. The tensor T µν is called the Maxwell stress
tensor and does exactly what has been said above.
( 2 )
Example 15.0.7 a. Prove that Trace(T)=T µµ = gµν T µν = − 21 δ µν Ec2 + B2 .
b. Prove that in a Cartesian coordinate system {x, y, z} the components of the electromag-
netic stress tensor are as follows:
 2 
Ex + Bx2 − W Ex Ey + Bx By Ex Ez + Bx Bz
Tµν =  Ex Ey + Bx By Ey2 + By2 − W Ey Ez + By Bz  (15.19)
Ex Ez + Bx Bz Ey Ez + By Bz Ez + Bz − W
2 2
( 2 )
where W = 12 Ec2 + B2 is the energy density of the electromagnetic field in vacuum.
15.0.9 Poynting’s Theorem

To find a physical interpretation of the Poynting vector S µ = εµνρ Eν Hρ we consider its diver-
gence. We compute:
µ
S,µ = η µνρ (Eν Hρ ),µ = η µνρ (Eν,µ Hρ + Eν Hρ,µ )
= η µνρ Eν,µ Hρ + η µνρ Eν Hρ,µ = −η µνρ Eµ,ν Hρ + η µνρ Eρ Hµ,ρ . (15.20)
From Maxwell equations (15.8) and (15.9) we have:
η µνρ Eν,µ = −B µ ,t η µνρ Hρ,µ = −(j µ + Dµ ,t ).

µ
S,µ = −(−B µ ,t )Hµ − (j µ + Dµ ,t )Eµ ⇒
µ
S,µ + j µ Eµ = −(B µ ,t Hµ + Dµ ,t Eµ ) (15.21)
Equation (15.21) is known as Poynting’s Theorem.
In order to give the physical interpretation of Poynting’s Theorem we note the following.
The term Dµ Eµ is the energy density of the electric field and the term B µ Hµ is the energy
density of the magnetic field stored within the ‘fluid’. Therefore the rhs is the total energy
loss of the field. Part of this energy is used to create the current j µ into the conductor (Joule
heat loss) and part of it is used to the divergence of the vector S µ . As we shall show using the
integral form of Poynting’s Theorem the vector S µ is the spatial density of the energy flow.
To find the integral form of Poynting’s Theorem we consider a smooth surface S with unit
outward nµ which encloses a space of volume V and charge density is ρ. Then (15.21) takes the
following integral form:
H ∫∫∫ d ∫∫∫ µ
Sµ nµ dσ + j µ Eµ dV = − (B Hµ + Dµ Eµ )dV. (15.22)
S V dt V
This equation states that in order to determine the time rate of energy loss in a given volume
V,we have to find the flux through the boundary surface S of V of the vector S µ and add
to this the rate of generation of heat within the volume V. Therefore it is natural to interpret
the Poynting’s vector as the density of energy flow. For this reason Poynting’s Theorem is
considered as the energy conservation law in electrodynamics.
We assume now that the material is homogeneous with electric permeability ε and magnetic
permeability µ. Then B µ = µH µ , Dµ = εE µ and equation (15.21) becomes:
µ
S,µ + j µ Eµ = −(µH 2 + εE 2 ),t . (15.23)
This result is Poynting’s Theorem for homogeneous materials. If we consider a smooth

surface S with unit outward nµ which encloses a space of volume V, then the integral form of
Poynting’s Theorem for homogeneous materials takes the form:
∫
d
Sµ nµ dσ +V j µ Eµ dV = − (µH 2 + εE 2 )dV. (15.24)
dt V
S
The physical explanation of this equation is as follows:
[Flux of S µ through S] + [Mechanical work (heat) done on the current in V ] (15.25)

= [Loss of energy of the field in V ]
Example 15.0.8 Prove Poynting’s Theorem using standard vector calculus.
Solution
We use the vector identity:
▽ · (u × v) = (▽ × u) · v−(▽ × v) · u (15.26)
97
to compute the divergence of the vector S = E × H. We have:
▽ · S = ▽ · (E × H) =(▽ × E) · H−(▽ × H) · E
( )
∂B ∂D
= (− )·H− j+ ·E
∂t ∂t
where in the last step we used Maxwell equations (11.4) and (11.5). This equation gives then
the general differential form of the Poynting’s Theorem:
( )
∂B ∂D
▽·S+j·E=− ·H+ ·E . (15.27)
∂t ∂t
In the special case of a homogeneous medium we write as before B =µH, D =εE and equation
(15.27) becomes:
d
▽ · S + j · E = − (µH 2 + εE 2 ) (15.28)
dt
which is the differential form of the Poynting’s Theorem for a homogeneous medium.
Exercise 15.0.7 Find the value of E and H on the surface of an infinite cylindrical wire
carrying a current. Show that the Poynting vector represents the flow of energy into the wire,
and show that this flow is just enough to supply the energy which appears as heat.
Exercise 15.0.8 Find the Poynting vector around a uniformly charged sphere placed in a uni-
form magnetic field.
15.0.10 The electromagnetic field as a viscous fluid

As we have seen the stress tensor of the electromagnetic field is (see ((15.17)):
[ µ ν ( )] ( )
E E 1 µν E2 1 µν E2
µν
T = +B B − δ
µ ν
+B 2
− δ +B 2
c c 3 c2 6 c2
But we know that the stress tensor of a viscous fluid is given by the relation (see (14.5)):
1
tµν = 2η(eµν − eδµν ) − pδµν
3
Comparing these two relations we infer that for the electromagnetic viscous fluid:
a. The (isotropic) pressure is: ( )
1 E2 2
p= +B
6 c2
b. The strain tensor is: ( )
1 Eµ Eν
eµν = + BµBν
2 c c
( )
1 E2
c. The trace e = 2 c2
+ B2 = u where u is the energy of the electromagnetic field.
d. For the electromagnetic viscous fluid the equation of state3 is:

1
p= u
3
e.The viscosity of the electromagnetic viscous fluid equals η = 1.
Exercise 15.0.9 1+2 decomposition of T µν

Suppose we have a direction (in 3-d space) specified by the unit vector nµ . Let hµν = δµν −
nµ nν be the associated normal projector to the vector nµ . Prove that the decomposition of 1 + 2
of the tensor T µν wrt a vector nµ gives the following irreducible parts:
1. The scalar: T µν nµ nν
2. The vector: T µν nµ hρν
3. The tensor: T µν hρµ hλν
Show that the physical meaning of each part is the following:

The scalar T µν nµ nν is the isotropic pressure on a surface whose normal is nµ . Obviously
the pressure depends on the orientation of the surface.
The vector T µν nµ hρν is the force density (=force per unit area) normal to nµ THIS MUST
BE CHECKED!!!!
The tensor T µν hρµ hλν is I DO NOT KNOW!!!!. IT IS THE SCREEN PROJECTION OF
µν
T .
Consider the electric and the magnetic fields E and H of the electromagnetic field in a
coordinate frame Σ. Then in Σ Maxwell equations read:
▽·B=0 (15.29)
ρ
▽·E= (15.30)
ε0
∂D
▽ × H= j+ (15.31)
∂t
∂B
▽ × E= − (15.32)
∂t
where ρ is the charge density in Σ and j is the current in Σ. Being on the same frame Σ
consider the following transformation of the fields:
E= E′ cos α + H′ sin α, D = D′ cos α + B′ sin α (15.33)

H= −E′ sin α + H′ cos α, B = −D′ sin α + B′ cos α (15.34)
and show that the quantities E × H (the Poynting vector) , E · D + B · H (the energy of the
field) and also the components of the Maxwell stress tensor which are the physical quantities
of the electromagnetic field are covariant (i.e. retain their form). This transformation is
3
The equation of state is a relation between the isotropic pressure and the energy density. This equation we
meet frequently in General Relativity.
99
called the dual transformation of the electromagnetic field. (Obviously it is not a coordinate
transformation).
Show that under this transformation Maxwell equations change and find their new form.
Now assume that we extent Maxwell transformations by considering a magnetic charge den-
sity, which will be related to the magnetic field in exactly the same way as the electric charge
density is related to the electric field plus that the magnetic charge satisfies the continuity equa-
tion. In consequence we write Maxwell equations in symmetric form as follows:
ρm
▽·B= (15.35)
µ0
ρe
▽·E= (15.36)
ε0
∂D
▽ × H= jm + (15.37)
∂t
∂B
▽ × E= −je − . (15.38)
∂t
Show that these equations remain the same (=are covariant) if in addition to the transfor-
mation of the fields considered above we also consider the dual transformation of the sources:
ρe =ρ′e cos α + ρ′m sin α, je = j′e cos α + j′m sin α (15.39)

ρm =−ρ′e sin α + ρ′m cos α, jm = −j′e sin α + j′m cos α. (15.40)
Also show that under the transformations (15.33) - (15.34) and (15.39) - (15.40) the
standard Maxwell equations (15.35) - (15.38) remain the same for the dashed quantities (they
are covariant). Conclude that if one postulates the magnetic charge ρm and the magnetic charge
density jm in addition to the known electric charge and electric current density, then Maxwell
equations are generalized to the symmetric form (15.35) - (15.38).
The invariance of the new equations wrt the transformations (15.33) - (15.34) and (15.39)
- (15.40) shows that to a great extent it is a question of convention what one calls the magnetic
charge of a particle. Because if all particles have the same ratio between the electric and the
magnetic charge, the angle α can be chosen universally such that:
( )
′ ρ′m
ρm = ρe − sin α + ′ cos α = 0.
ρe
In this case:
( ) ( )
j′ ρ′m
jm = je − sin α + m′ cos α = je − sin α + ′ cos α = 0.
je ρe
For this special angle α, equations (15.35) - (15.38) reduce to the standard Maxwell equa-
tions and no change occurs. Then one is possible to fix the charge of the electron by the
convention:
qe = −e qm = 0.
and measure the charges of all other particles using these units.
Solution
The generalized Maxwell equations are:
∂D
∇ · D = ρ ef ∇×H =
+ Je
∂t (15.41)
∂B
∇ · B = ρmf −∇ × E = + Jm
∂t
where ρef is the free electric charge density and ρmf is the free magnetic charge density.
We consider the transformation
E = Ẽ cos a + H̃ sin a D = D̃ cos a + B̃ sin a

(15.42)
H = −Ẽ sin a + H̃ cos a B = −D̃ sin a + B̃ cos a
Maxwell equations remain unchanged if we transform the sources according to:
ρef = ρ̃ef cos a + ρ̃mf sin a Je = J̃e cos a + J̃m sin a

(15.43)
ρmf = −ρ̃ef sin a + ρ̃mf cos a Jm = −J̃e sin a + J̃m cos a
Replacing (15.42) in (15.41) yields (the ∇ operator does not apply to the angle a):
∇ · D̃ cos a + ∇ · B̃ sin a = ρef (15.44)

−∇ · D̃ sin a + ∇ · B̃ cos a = ρmf (15.45)
∂ D̃ ∂ B̃
−∇ × Ẽ sin a + ∇ × H̃ cos a = cos a + sin a + Je (15.46)
∂t ∂t
∂ D̃ ∂ B̃
−∇ × Ẽ cos a − ∇ × H̃ sin a = − sin a + cos a + Jm (15.47)
∂t ∂t
Replacing the sources from the transformation equations (15.43) into (15.41) through (15.44)
we get:
∇ · D̃ cos a + ∇ · B̃ sin a = ρ̃ef cos a + ρ̃mf sin a =⇒

(∇ · D̃ − ρ̃ef ) cos a + (∇ · B̃ − ρ̃mf ) sin a = 0
−∇ · D̃ sin a + ∇ · B̃ cos a = −ρ̃ef sin a + ρ̃mf cos a

(∇ · B̃ − ρ̃mf ) cos a + [−(∇ · D̃ − ρ̃ef )] sin a = 0
For the system of equations
(∇ · D̃ − ρ̃ef ) cos a + (∇ · B̃ − ρ̃mf ) sin a = 0 (15.48)

(∇ · B̃ − ρ̃mf ) cos a + [−(∇ · D̃ − ρ̃ef )] sin a = 0 (15.49)
101
to posses a non zero solution for the unknowns cos a, sin a, we demand that the determinant
vanishes:
( )
(∇ · D̃ − ρ̃ef ) (∇ · B̃ − ρ̃mf )
det = 0 =⇒
(∇ · B̃ − ρ̃mf ) [−(∇ · D̃ − ρ̃ef )]
−(∇ · D̃ − ρ̃ef )2 − (∇ · B̃ − ρ̃mf )2 = 0 (15.50)

Since the final equation is a sum of squared terms, each term must equal zero:
∇ · D̃ − ρ̃ef = 0 and
∇ · B̃ − ρ̃mf = 0
The first 2 Maxwell equations are:
∇ · D̃ = ρ̃ef and (15.51)
∇ · B̃ = ρ̃mf (15.52)
∂ D̃ ∂ B̃
−∇ × Ẽ sin a + ∇ × H̃ cos a = cos a + sin a + J̃e cos a + J̃m sin a
( ) ( ) ∂t ∂t
∂ B̃ ∂ D̃
− ∇ × Ẽ + + J̃m sin a + ∇ × H̃ − − J̃e cos a = 0
∂t ∂t
Working in a similar manner we show:

∂ D̃ ∂ B̃
−∇ × Ẽ cos a − ∇ × H̃ sin a = − sin a + cos a − J̃e sin a + J̃m cos a
( ) ( ) ∂t ∂t
∂ D̃ ∂ B̃
∇ × H̃ + + J̃e sin a − ∇ × Ẽ + + J̃m cos a = 0
∂t ∂t
In order the system of equations:

( ) ( )
∂ B̃ ∂ D̃
− ∇ × Ẽ + + J̃m sin a + ∇ × H̃ − − J̃e cos a = 0 (15.53)
∂t ∂t
( ) ( )
∂ D̃ ∂ B̃
∇ × H̃ + + J̃e sin a − ∇ × Ẽ + + J̃m cos a = 0 (15.54)
∂t ∂t
to have non zero solutions (wrt sin a, cos a unknowns) the determinant of the coefficients must
vanish:
 ( ) ( ) 
− ∇ × Ẽ + ∂t + J̃m
∂ B̃
∇ × H̃ − ∂t − J̃e
∂ D̃
det  ( ) ( ) = 0 =⇒
− ∇ × H̃ − ∂∂tD̃ − J̃e − ∇ × Ẽ + ∂∂tB̃ + J̃m
( )2 ( )2
∂ B̃ ∂ D̃
∇ × Ẽ + + J̃m + ∇ × H̃ − − J̃e = 0 =⇒
∂t ∂t
∂ B̃ ∂ B̃
∇ × Ẽ + + J̃m = 0 =⇒ −∇ × Ẽ = + J̃m (15.55)
∂t ∂t
∂ D̃ ∂ D̃
∇ × H̃ − − J̃e = 0 =⇒ ∇ × H̃ = + J̃e (15.56)
∂t ∂t
We conclude that all four Maxwell equations remain unchanged under the action of this
transformation.
Terminology
We give below the definition of the various terms discussed in the previous sections.
Stress Terms
Stress is defined as force per unit area. It has the same units as pressure, and in fact pressure
is one special variety of stress. However, stress is a much more complex quantity than pressure
because it varies both with direction and with the surface it acts on.
Compression
Stress that acts to shorten an object.
Tension
Stress that acts to lengthen an object.
Normal Stress
Stress that acts perpendicular to a surface. Can be either compressional or tensional.
Shear
Stress that acts parallel to a surface. It can cause one object to slide over another. It also
tends to deform originally rectangular objects into parallelograms. The most general definition
is that shear acts to change the angles in an object.
Hydrostatic
Stress (usually compressional) that is uniform in all directions. A scuba diver experiences
hydrostatic stress. Stress in the earth is nearly hydrostatic. The term for uniform stress in the
earth is lithostatic.
Directed Stress
Stress that varies with direction. Stress under a stone slab is directed; there is a force in
one direction but no counteracting forces perpendicular to it. This is why a person under a
thick slab gets squashed but a scuba diver under the same pressure doesn’t. The scuba diver
feels the same force in all directions4 .
Strain Terms
Strain is defined as the amount of deformation an object experiences compared to its original
size and shape. For example, if a block 10 cm on a side is deformed so that it becomes 9 cm
long, the strain is (10-9)/10 or 0.1 (sometimes expressed in percent, in this case 10 percent.)
Note that strain is dimensionless.
Longitudinal or Linear Strain
Strain that changes the length of a line without changing its direction. Can be either
compressional or tensional.
4
In practice we see stress as it deforms materials. Even if we were to use a strain gauge to measure in-situ
stress e.g. in the rocks, we would not measure the stress itself. We would measure the deformation of the strain
gauge (that’s why it’s called a ”strain gauge”) and use that to infer the stress.
103
Compression
Longitudinal strain that shortens an object.
Tension
Longitudinal strain that lengthens an object.
Shear
Strain that changes the angles of an object. Shear causes lines to rotate.
Infinitesimal Strain
Strain that is tiny, a few percent or less. Allows a number of useful mathematical simplifi-
cations and approximations.
Finite Strain
Strain larger than a few percent. Requires a more complicated mathematical treatment
than infinitesimal strain.
Homogeneous Strain
Uniform strain. Straight lines in the original object remain straight. Parallel lines remain
parallel. Circles deform to ellipses. Note that this definition rules out folding, since an originally
straight layer has to remain straight.
Inhomogeneous Strain
How real geology behaves. Deformation varies from place to place. Lines may bend and do
not necessarily remain parallel.
Terms for Behavior of Materials
Elastic
Material deforms under stress but returns to its original size and shape when the stress is
released. There is no permanent deformation. Some elastic strain, like in a rubber band, can
be large, but in rocks it is usually small enough to be considered infinitesimal.
Brittle
Material deforms by fracturing. Glass is brittle. Rocks are typically brittle at low temper-
atures and pressures.
Ductile
Material deforms without breaking. Metals are ductile. Many materials show both types of
behavior. They may deform in a ductile manner if deformed slowly, but fracture if deformed
too quickly or too much. Rocks are typically ductile at high temperatures or pressures.
Viscous
Materials that deform steadily under stress. Purely viscous materials like liquids deform
under even the smallest stress. Rocks may behave like viscous materials under high temperature
and pressure.
Plastic
Material does not flow until a threshold stress has been exceeded.
Viscoelastic
Combines elastic and viscous behavior. Models of glacio-isostasy frequently assume a vis-
coelastic earth: the crust flexes elastically and the underlying mantle flows viscously.
Chapter 16
Relativistic fluids
16.1 Fluids in Special Relativity

The concept of fluid is of fundamental importance in Newtonian Physics therefore must be
extended to the Theory of Special (and General) Relativity. Up to now we have been dealing
with single particles or systems with a finite number of particles which are characterized by
four vectors only. However a Newtonian fluid requires for its description the strain tensor which
is a second order tensor. This tensor cannot be generalized to a 4-vector hence in order to
generalize the concept of fluid in Special Relativity we have to introduce a second order tensor.
We shall restrict our considerations to the generalization of viscous fluids only. One should
expect that there will be relativistic fluids which do not have a Newtonian analogue, but these
will not concern us here.
As we have shown, in Newtonian Physics a viscous fluid is a linear elastic isotropic continuum
which is described by the scalar field ρ(t, r) of mass density and the (Eulerian) velocity vector
field v(t, r). These fields satisfy the following dynamical equations:
1. The continuity equation for the mass density corresponding to the conservation of mass:
Dρ ∂ρ
= + ▽(ρv) = 0 (16.1)
dt ∂t
2. The Navier Stokes equation:

Dv µ
ρ = ρf µ + F µ . (16.2)
dt
In terms of the velocity vector v µ equation (16.2) is written:
Dv µ η 2 1
= f µ + (v µ,ν ν + v ν,µ ν − v,ρρ δµν ) − p,µ (16.3)
dt ρ 3 ρ
or, in standard vector notation:

[ ]
Dv η 2 1
= f + ▽ v − ▽ ·(▽ · v) − ▽ p.
2
(16.4)
dt ρ 3 ρ
105
106 CHAPTER 16. RELATIVISTIC FLUIDS
Making use of the continuity equation (16.1), equation (16.3) can be written in terms of
the momentum density pµ = ρv µ as follows:
Dpµ
= ρf µ + ηπ µν ,ν − p,µ (16.5)
dt
where in equation (16.5) πµν = (v µ,ν + v ν,µ − 23 v,ρρ δµν ) is the traceless part of the stress tensor,
p is the isotropic pressure (trace of the stress tensor) and f µ is the external force density. A
viscous fluid for which π µν = 0 is called perfect.
Because in Special (and in General) Relativity the mass density and the 3−momentum
density are both components of the 4-momentum density we expect that equations (16.1) and
(16.5) are the ones we should use in the dynamics of the relativistic viscous fluid. Furthermore,
due to the fact that the Navier Stokes equation contains the stress tensor, the relativistic
generalization must contain a second order symmetric tensor, which will generalize in some
way the stress tensor of the Newtonian viscous fluid.
16.1.1 Relativistic fluids in Special Relativity

The first assumption we make towards the definition of a relativistic viscous fluid is that a
relativistic viscous fluid:
a. Consists of (relativistic) particles as a Newtonian fluid which move in a connected region
of the three dimensional space.
b. The world lines of the particles of a fluid define a timelike or null congruence of curves
(world lines) in spacetime. This congruence consists respectively of the integral curves of a
timelike or null differentiable vector field. This spacetime description of the fluid contains all
information concerning the motion (flow) of the fluid.
Concerning the description of the fluid itself we introduce the following quantities:
1. At each point P in the space where the fluid flows one chooses an infinitesimal volume
dV (in a coordinate frame Σ because the 3-dimensional volume is not a Lorentz invariant) and
count the number n of the particles of the fluid which flow through dV per unit of time (in Σ).
If the number n it does not change in time we say that (during that period of time) there are
not sinks and sources in the volume dV.
2. We also consider at the point P and at a moment (in an arbitrary frame Σ) an elementary
volume dV0 (in Σ) containing a number of particles and then propagate dV0 with the fluid so that
at all times contains the same particles (both in number and identity). This new elementary
volume we call comoving volume.
3. At each point P there exists an observer whose four-velocity at P is tangent to the
worldline of the particle of the fluid passing the point P at that moment. This observer we call
the comoving observer at the point P .
Using the above assumptions we are able to introduce various physical quantities which can
be used for the description of the “motion” of the fluid.
Let ua be the four-velocity of the comoving observers in the comoving infinitesimal volume
dV0 in the fluid. We define the physical quantity particle number density ρ characterizing
the “motion” of the relativistic fluid as follows:
dn
ρ0 = . (16.6)
dV0
16.1. FLUIDS IN SPECIAL RELATIVITY 107
ρ0 is a scalar (not invariant!) whose value is determined in the comoving frame. To compute
its value ρΣ in another Lorentz frame, Σ say, whose γ−factor wrt the comoving frame is γ , we
have to know how both dn and dV0 transform. Concerning the volume dV0 we know that
under a Lorentz transformation it transforms as:
dV0
dV = . (16.7)
γ
Concerning dn we assume that it is an invariant. Therefore equation (16.6) gives:
ρΣ = ρ0 γ. (16.8)
a
Using the physical quantities ρ0 and ua we define the particle number current j as
follows:
a
j = ρ0 u a . (16.9)
a
To compute the components of the vector j in the Lorentz
( frame
) Σ referred before we
γc
recall that in Σ the components of the four-velocity are ua = , therefore:
γv Σ
( ) ( )
a ρΣ γc ρΣ c
j = = . (16.10)
γ γv Σ ρΣ v Σ
a
We compute the divergence of j in order to find out if it satisfies a continuity equation,
a
that is an equation of the form j ,a = 0. We find:
a 0 µ ∂(ρΣ c) ∂ρ c Dρ
j ,a = j ,0 + j ,µ = + ▽(ρΣ v) = Σ + ▽(ρΣ v) = Σ . (16.11)
∂t ∂t dt
a
It follows that the vanishing of the divergence of the current j is equivalent to the re-
quirement that the number of particles in the volume dV does not change in time1 , that is, the
volume dV is comoving. This result is known as the law of preservation of particle number
and it is expressed by the continuity equation2 :
a
j ,a = 0. (16.12)
a
The vector j characterizes the number of particles in dV but not the type of the particles.
Therefore we have to look for more current vectors defined for the fluid.
In Special Relativity the type of a particle is characterized (to the lowest degree) by its
proper mass. We assume that the fluid consists of one type of particles of proper mass m0 c2
and define the new scalar matter density ρ0 of the fluid in the comoving frame as follows:
ρ 0 = ρ 0 m0 c 2 . (16.13)
1
It is easy to see that ρΣ dV = ρ0 dV0 .
2
A continuity equation corresponds to the conservation of a scalar quantity which defines a vector. For
example in the present case the conservation of the scalar quantity which multiplies the four velocity in the
corresponding current is the particle number density and the continuity equation is the conservation of this
quantity. Another example from Newtonian Physics is the conservation of mass which concerns the divergence
of the current j = mv and it is given by Dm dt = 0.
If ρΣ is the energy density of the fluid in the Lorentz frame Σ considered above, then using
the transformation of energy:
EΣ = γE0 = γm0 c2 (16.14)
we find3 :
ρΣ = ρ0 γ 2 . (16.15)
Using the energy density we define the matter density current j a to be the 4-vector:
j a = ρ0 u a . (16.16)
In the frame Σ the components of the vector j a are:

( ) ( )
a ρΣ γc 1 ρΣ c
j = 2 = . (16.17)
γ γv Σ γ ρΣ v Σ
We compute again the divergence of the matter density current. We find:

[ ]
1 ∂(ρΣ c) γ,α a 1 DρΣ γ,α a
a
j,a = + ▽(ρΣ v) − j = − j . (16.18)
γ ∂t γ γ Dt γ
We note that the equation of continuity for the energy density 4-current is not equivalent to
the mass continuity equation Dρ
dt
Σ
. This is to be expected because the energy is not an invariant
in Special (and in General) Relativity.
16.2 The energy momentum tensor in Special Relativity

The energy density ρ and the energy density 4-current j a exhaust all the physical quantities one
can define in terms of the number, the type of the fluid particles and the comoving observers.
We have shown that these quantities can take into account the mass continuity equation of a
Newtonian fluid. Concerning the Navier Stokes equation of motion we note that this equation
requires the stress tensor which is a second rank symmetric tensor, hence in the definition of a
relativistic fluid we must introduce a symmetric second rank tensor, Tab say. This tensor will
be build from the available elements we have, that is, the number density ρ0 , the 4-velocity ua
and the Lorentz metric ηab . The new tensor we call the energy momentum tensor.
Before we proceed with the Physics we look at the tensor Tab form a mathematical point of
view. This means that we shall answer the question:
How much information a symmetric second order tensor can provide to the co-
moving observers ua of the fluid?
3
A more ”classical“ proof is the following:
dm dE γdm0 c2
ρ= = 2 = 21 = γ 2 ρ0 .
dV c dV c γ dV0
One can prove the same result using (16.8) and (16.14).
16.2. THE ENERGY MOMENTUM TENSOR IN SPECIAL RELATIVITY 109
Let us do some counting. A general tensor of second rank in a four dimensional space has
2
4 = 16 components. But we only need four, as many as the equations we wish to derive (that is
equations (16.1) and (16.5)). This implies that we have to restrict T ab by imposing conditions,
which will restrict the number of free components to the number four. The first requirement
we impose is that it will be symmetric, that is T ab = T ba . This is logical because it must
correspond somehow to the stress tensor of the Newtonian fluid and the later is symmetric.
This requirement reduces the number of independent components to 4×5 2
= 10, still too many.
ab
We further demand that the divergence of this tensor T..,b shall be equal to a definite four-vector
and we end up with the correct number of equations.
The energy momentum tensor describes the environment while the fluid is described by
its four-velocity vector ua say. Therefore the interaction of the fluid with the environment
will be done by the interaction of these two fields. This interaction is described by means of
the irreducible parts of the tensor resulting from its 1+3 decomposition wrt the vector field
ua . This decomposition is covariant, in the sense that under a coordinate transformation each
irreducible part transforms as an independent tensor. Each irreducible part corresponds to a
physical quantity describing the flow of the fluid under the given environment (i.e. the Tab ).
16.2.1 The 1+3 decomposition

The concept of energy momentum tensor is present in all branches of Physics, even in Newtonian
Theory, therefore it is useful to derive it using general concepts and covariant methods, which
(with the appropriate adjustments) apples to all physical theories. This means that the concept
of the energy momentum tensor has a geometric nature which is revealed in each theory if we
decompose it wrt a non-zero vector field. The results obtained are valid equally well in all
Newtonian Physics, Special Relativity and General Relativity i.e. one obtains the formulae
in each theory by replacing in the general results the special metric and/or the decomposing
vector field. This remark will be made clear from the discussion that follows.
Consider a spacetime endowed with a metric gab and let q a be a non-null vector (i.e.
ε(q)q 2 ≡ gab q a q b ̸= 0 where ε(q) = ±1 is the sign of q a ) in this spacetime. Then the ten-
sor:
ε(q)
hab (q) = gab − qa qb (16.19)
q2
projects normal to ua i.e. hab (q)q b = 0. Furthermore it is easy to prove that the tensor hab (q)
satisfies the properties:
hab (q) = hba (q), haa (q) = 3, , hab (q)hbc (q) = δac . (16.20)
Using the projection tensor hab (q) we can 1+3 decompose any tensor along and normal to
the vector q a .
Exercise 16.2.1 Let Aa be a vector in the same space as q a (q 2 ̸= 0, Aa can be null).

(a) Show the identity:
Aa =∥ Aa +⊥ Aa
0
) ( a
) (
A ∥A
where ∥ Aa = q12 (qb Ab )q a , ⊥ Aa = hab (q)q b . Symbolically we write Aa = = a
A Σ ⊥A
a
where Σ is an arbitrary coordinate frame. This (covariant) decomposition of A we call the
1+3 decomposition of Aa wrt q a .
(b) Show that ⊥ (⊥ Aa ) =⊥ Aa and ∥ (∥ Aa ) =∥ Aa
Exercise 16.2.2 Consider a second rank tensor Aab and by contracting with q a q b , q a hbc (q), hac (q)q b , hac (q
show that the following identity (we emphasize: mathematical identity!)holds:
ε(q) ε(q)
Acd = (A(ab) q a q b )qc qd − 2
A(ab) q a hbc (q)qd − 2 A[ab] hac (q)q b qd +A(ab) hac (q)hbc (q)+A[ab] . (16.21)
q q
This identity involves the irreducible parts:
A(ab) q a q b , A(ab) q a hbc (q), A(ab) hac (q)q b , A(ab) hac (q)hbc (q), A[ab]
and defines the 1+3 decomposition of the tensor Aab along the vector q a . In matrix form we
write symbolically:
( )
A(ab) q a q b − 21 ε(q) A (ab) q a b
h (q) ε(q)
c
− 2 A[ab] hac (q)q b qd + A[ab] (16.22)
q 2
Aab = 1 ε(q) a b a b
- 2 q2 A(ab) hc (q)q A(ab) hc (q)hd (q) q
Σ
where Σ is an arbitrary coordinate frame. This tensor can be decomposed further if we write
the symmetric tensor ha(c (q)hbd) (q) in terms of a traceless part and a trace as follows:
1 1
ha(c (q)hbd) (q) = (hr(a hsb) − hab hrs ) + hab hrs .
3 3
In this case the 1+3 decomposition of Aab reads:
( )
ε(q) ε(q) 1 ε(q)
Acd = (A(ab) q q )qc qd − 2 A(ab) q hc (q)qd − 2 A[ab] hc (q)q qd + hc hd +
a b a b a b a b
hcd hab
A(ab)
q q 3 q2
(16.23)
1 ε(q) ab
− (h A(ab) )hcd + A[cd]
3 q2
and leads to the symbolic formula in matrix form:
( )
A(ab) q a q b − 12 ε(q) A(ab) q a hbc (q) ε(q)
Aab = ( q2 ) − A[ab] hac (q)q b qd +A
1 ε(q) a
- 2 q2 A(ab) hc (q)q b
hc hd + 3 q2 hcd h A(ab) −
a b 1 ε(q) ab 1 ε(q)
3 q2
(hab A(ab) )hcd q2
Σ
(16.24)
We conclude that by the 1+3 decomposition of a general 2-index tensor Aab is specified by
(and specifies) six different tensors:
a. Two scalars: A(ab) q a q b , hab A(ab)
b. Two vectors: 12 ε(q) A(ab) q a hbc (q), ε(q) a
2 A[ab] hc (q)q
q(
b
q2 )
c. One traceless symmetric tensor: hac hbd − 13 hcd hab A(ab)
d. One antisymmetric second rank tensor: A[ab] .

These tensors are the maximum possible elements that one can employ to describe the
tensor in reference to a non-null vector field q a .
For a symmetric tensor we have that the available quantities are:
a. Two scalars: A(ab) q a q b , hab A(ab)
b. One vector: 21 ε(q) A(ab) q a hbc (q)
q2 ( )
c. One traceless symmetric tensor: hac hbd − 13 hcd hab A(ab) .
These last quantities we shall use to describe the dynamics of a relativistic fluid.
16.2.2 The bivector metric

Having given the basics on the 1+3 decomposition of a vector field and a second rank tensor
we turn back to the projection tensor hab (q) and look at it from a different angle. We consider
the three dimensional space4 of all vectors ⊥ Aa . In that space we define the length ⊥ A2 of the
three vector ⊥ Aa as follows:
2
⊥A = hab (q)⊥ Aa ⊥ Ab = gab a b
⊥A ⊥A .
This implies that we can consider the tensor hab (q) as a metric in the 3- space. However we are
interested in a metric defined in the whole 4-dimensional space. For that we note the identity5 :
ε(q)
hab (q) = 2
(gab gcd − gac gbd )q c q d . (16.25)
q
Identity (16.25) expresses the projection tensor hab (q) in terms of two tensors. The tensor:
gabcd = gab gcd − gac gbd (16.26)
which depends only on the metric of the 4-space and it is independent of the decomposing
c qd
vector field q a and the tensor ε(q)q
q2
which depends solely on the decomposing vector field q a .
The new tensor gabcd we call the bivector metric of the space. This metric is the same for
the projective tensors (therefore all 1+3 decompositions) of all non-null vectors q a ! It is an easy
exercise to show that the bivector metric satisfies the following symmetry properties6 :
gabcd = g[ab]cd = gab[cd] = ga[bc]d = gcdab (16.27)

4
The discussion in this section is independent of the dimension of the space, therefore it holds for an
n−dimensional metric space of any signature.
5
Proof
ε(q) ε(q) ε(q)

hab (q) = gab − 2
qa qb = gab − 2 gac gbd q c q d = 2 (ε(q)q 2 gab − gac gbd q c q d )
q q q
ε(q) ε(q)
= 2 (gab gcd − gac gbd )q c q d = 2 gabcd q c q d .
q q
6
The symmetries of gabcd are the same as the (index) symmetries of the (Riemann) curvature tensor. This
tensor also is involved in the decomposition of the Riemann tensor.
where according to our standard convention round brackets ( ) denote symmetrization wrt the
indices they enclose and square brackets [ ] antisymmetrization7 .
Exercise 16.2.3 Show that the bivector metric can be written in terms of the antisymmetric
Levi Civita tensor as follows8 :
1
gabcd = − ηabrs ηcdrs . (16.29)
2
In terms of the bivector metric the projection tensor hab (q) is written as follows (see (16.25)
):
ε(q)
hab (q) = gabcd q c q d . (16.30)
q2
Consequently the length ⊥ A2 =⊥ Aa ⊥ Aa of the normal component ⊥ Aa of a 4-vector Aa is
given by the formula9 :
2 ε(q)
⊥A = 2
gabcd q c q d Aa Ab . (16.31)
q
This formula shows that the length ⊥ A2 depends on two parts:
a. The part gabcd Aa Ab which depends to the decomposed vector Aa and it is independent
of the decomposing vector q a
b. The part ε(q)
q2
q c q d which depends on the vector q a only.
This leads us to define a new second rank tensor Eab depending only on the vector Aa and
the space bivector metric gabcd by the formula:
Eab = Xgabcd Ac Ad . (16.32)
where X is an invariant (constant) to be determined. The tensor Eab is a symmetric second

rank tensor therefore it can be 1+3 decomposed wrt the vector field q a . We expect that the
irreducible parts of this decomposition will provide the quantities which we need to describe
the“motion” of the vector Aa in space..
7
The following result is not true and has to be checked and if necessary corrected.
Show that the bivector metric can be written in terms of the antisymmetric Levi Civita tensor as follows:
1
gabcd = − ηabrs ηcdrs . (16.28)
2
8
Proof:
1 1
− ηabkl ηcdkl = − (−1)(δak δbl −δal δbk )gkc gld = −(−gac gbd +gbc gda ) = −(gad gbc −gac gbd ) = gac gbd −gad gbc = gabcd .
2 2
9
Proof
2 ε(q)
⊥ A = hab (q)⊥ Aa ⊥A
b
= hab (q)Aa Ab = c d a b
q 2 gabcd q q A A .
16.2.3 The 1+3 decomposition wrt the four-velocity ua

We specialize the results of the last section to the case q a is the 4-velocity ua (ua ua = −c2 ) of
some observers. From (16.25) and (16.30) we have for the projection tensor hab (q) :
1 1
2
ua ub = − 2 gabcd uc ud .
hab (u) = gab + (16.33)
c c
a
For an arbitrary vector A this implies the1+3 irreducible parts:
1 1
∥A
a
=− 2
(Ab ub )ua , ⊥A
a
= hab (u)Ab , ⊥A
2
=− 2
gabcd uc ud Aa Ab . (16.34)
c c
For a general (that is, not necessarily symmetric) tensor Mab of type (0,2) formula (16.23)
gives:
( )
1 1 11
Acd = (A(ab) u u )uc ud + 2 A(ab) u hc (u)ud + 2 A[ab] hc (u)u ud + hc hd − 2 hcd h
a b a b a b a b ab
A(ab)
c c 3c
(16.35)
11
+ 2 (hab A(ab) )hcd + A[ab]
3c
which in matrix form is:
( )
A(ab) ua ub (A(ab) ua hbc (u) ) 1
Aab = A ha (u)ub ud +A[ab] .
+
A(ab) hac (u)ub hac hbd − 13 c12 hcd hab A(ab) + 1 1
3 c2
2 [ab] c
(hab A(ab) )hcd
Σ c
(16.36)
From this relation ( mathematical identity!) we conclude that in Special (and General)
Relativity a 2-index tensor wrt the 4-velocity (or equivalently wrt any constant timelike vector
filed) is specified by (and specifies) the following six different tensors fields:
a. Two scalars: A(ab) ua ub , hab A(ab)
b. Two vectors: c12 A(ab) ua hbc (u)ud , c12(A[ab] hac (u)ub ud )
c. One traceless symmetric tensor: hac hbd − 13 hcd hab A(ab)
d. One antisymmetric second rank tensor: A[ab] .
This implies that if we consider a second order tensor in the description of the motion of a
physical system then any class of observers ua will have at its disposal only the above tensor
fields to study the development of the system in spacetime.
Let us apply the above to the kinematics of the relativistic observers ua and the dynamics
of matter as observed by these observers.
The kinematic quantities of the 4-velocity ua

Given a 4-velocity field ua one can define the second rank tensor ua,b . This tensor is not
symmetric and contains information on the 4-velocity only. Therefore it can be used to describe
the “kinematics” of the observers ua . The 1+3 decomposition of this tensor is given by the
following identity:
1 .
ua,b = σab + ωab + θhab − ua ub (16.37)
3
where we have set10 :

a. θ = hab ua,b
.
b. ua = ua,b ub
c. σab = (hra hsb − 13 hab hrs )ur,s
d. ωab = u[a,b] .
The geometric meaning of each tensor is obtained from the study of the geometry of the
integral curves of ua . These curves make up a timelike congruence whose geometry is studied
by means of the connecting vector and its Lie transport along the curves of the congruence or,
in physical terms, the relative velocity of the congruence of observers. We shall not comment
further on that. Note that the four-velocity ua is a vector field in a geometric space, therefore
one has to define what “motion” means. Finally Lie transportation defines the concept of
“motion’ both in Newtonian Physics and in Relativity.
.
The study of kinematics shows that the quantities σab , ωab , ua , θ have the following physical
meaning:
θ : Measures the change of the relative distance between two elements in the spacelike plane
normal to the observer worldline
.
ua : Is the 4-acceleration of the congruence of observers
σab : Measures the anisotropy of the relative motion in the spacelike plane normal to the
observer congruence
ωab : Measures the relative rotation in the spacelike plane normal to the observer congruence
Due to their kinematic role we name θ the expansion, σab the shear (strain), ωab the rota-
.
tion and ua the 4-acceleration of the timelike congruence of the observers ua . The quantities
.
(i.e. σab , ωab , ua , θ) are the fundamental physical quantities in both relativistic and Newto-
nian kinematics and Physics. Details on the significance of these quantities will be discussed
elsewhere.
16.2.4 The energy momentum tensor

We apply the general formula (16.35) to the tensor Aab = Tab where Tab = Tba is a symmetric
(0,2) tensor. In this case one shows easily the following decomposition (mathematical identity! ):
Tab = µua ub + phab + 2q(a ub) + πab (16.38)
where we have introduced the tensors/ irreducible parts:
µ = Tab ua ub (16.39)
1
p = hab Tab (16.40)
3
q a = hab Tbc uc (16.41)
1
πab = (hra hsb − hab hrs )Trs (16.42)
3
As noted already the 1+3 decomposition of Tab produces two scalar fields (µ, p) one spacelike
vector (q a , qa ua = 0) and one traceless symmetric 2-tensor (πab , gab π ab = 0).
10
Note that ua,b ua = 0 because ua ua = −c2 .
16.3. THE CASE OF A SINGLE PARTICLE 115
The tensor Tab is independent of the observers ua and we assume that it geometrizes the
environment which modulates the motion of the relativistic fluid (i.e. the timelike congruence
of the worldlines.). The tensors µ, p, qa , πab are the quantities which describe the interaction of
the fluid with the environment and we consider them as the dynamical variables describing the
evolution of the fluid11 . The same Tab for different fluids gives different dynamical quantities
µ, p, qa , πab !
To show the appropriateness of the energy momentum tensor in the description of the
dynamics of a relativistic fluid we turn to the Newtonian Physics (which is the only theory
with direct observational evidence) and examine the dynamic fields involved. The description
of the dynamic behavior of a Newtonian fluid requires the following quantities:
a. Mass density
b. Isotropic pressure
c. Heat flux (conductivity)
d. Strain
That is one needs two scalars, one vector and one symmetric traceless tensor, exactly as
many as the irreducible parts of the 1+3 decomposition of a symmetric tensor Tab ! Therefore we
are covered mathematically and have only to consider which tensor corresponds to which phys-
ical quantity. To decide on this correspondence we employ the bivector metric we developed
above.
16.3 The case of a single particle

We start with the simple case of an isolated particle of mass m0 c2 . We assume that in a
coordinate system Σ the particle has a velocity v, acceleration a while moving under the
action of forces, some of them being conservative (in Σ!) - given by the potential U - and some
of them being non-conservative given by the cumulative force F. The equation of motion for
the particle is:
dpa
= fa (16.43)
dτ
where τ is the proper time of particle and pa is the 4-momentum of the particle. Let us assume
that the four-velocity of the observer comoving with the frame Σ is ua . Then the four-velocity
v a of the particle is 1+3 decomposed wrt ua as follows:
v a = v 0 ua + hab (u)v b =∥ v a + ⊥v
a
(16.44)
where v 0 = − c12 (ua va ). In terms of components we have12 :

( ) ( )
a γc a c
v = ,u = . (16.45)
γv Σ
0 Σ
11
We note that in this approach the observers are not considered to be part of the physical system. This dis-
tinction between observers and observed systems reflects the sharp distinction between observer and observation
which lies in the roots of non-quantum Physics.
12
Note that:
hab v b = (gab + c12 ua ub )v b = va + c12 (uv)ua = va + c12 (−γc2 )ua = va − γua . In a Lorentz frame the coordinates
are: hab v b = (−γc, γv) − γ(−c, 0) = (0, γv) =⊥ va . We shall need this result below.
As we have seen the length of the space part γv of the 4-velocity v a is given by the formula
(see (16.34)):
1
(γv)2 = − 2 gabcd uc ud v a v b . (16.46)
c
The kinetic energy of the particle in Σ (!) equals:
p2 1 m0
T = = m0 (γv)2 = − 2 gabcd uc ud v a v b . (16.47)
2m0 2 2c
This can be written:
2T = Eab ua ub (16.48)
where Eab is the symmetric symmetric second rank tensor we introduced in (16.32) with X =
− mc20 :
m0
Ecd = − 2 gabcd v a v b . (16.49)
c
We note that the tensor Ecd depends only on the metric of the space and the four-velocity of
the particle while it is independent on the particular frame of reference Σ where the motion of
the particle is studied. Therefore it describes the covariant (that is, independent of the reference
frame) part of the motion of the particle in spacetime. The specialization of the motion to the
frame Σ is made by the specification of the four-vector ua .
We 1+3 decompose Eab wrt ua by computing its contractions with ua ub , ua hbc , hac hbd . We
find:
Ecd uc ud = 2T
by its definition. Therefore the invariant Ecd uc ud is related to the kinetic energy of the
particle (see also (16.48) ) .
For the vector term we compute:
m0
Ecd uc hdl = − 2 gabcd v a v b uc hdl
c
m0
= − 2 (gab gcd − gac gbd )v a v b uc hdl
c
m0
= − 2 (−c2 ud hdl − (uv)hal v a )
c
m0 m0
= 2 (uv)hal v a = 2 (−γc2 )hal v a
c c
= −m0 γhal v a = −hal pa (16.50)
where pa = m0 γv a is the four-momentum of the particle.
We can write this relation differently. We have:
1 1
hac v c = (gac + 2 ua uc )v c = va + 2 (uv)ua (16.51)
c c
therefore:
( )
1
Ecd u hl = −m0 γ vl + 2 (uv)ul
c d
c
( )
1
= −m0 γ vl + 2 (−γc )ul 2
c
= −m0 γ (vl − γul ) .
In the proper frame Σ of the four-velocity ua in which the four-velocity of the particle is
va = (−γv, γv)Σ we have13 :
∗
Ecd uc hdl = −m0 γ 2 v (16.52)
that is, this term is the linear momentum of particle in Σ.

Concerning the space part hac hbd Eab we compute14 :
m0
Ecd hcr hds = − gabcd v a v b hcr hds
c2
m0
= − 2 (gab gcd − gac gbd )v a v b hcr hds
c
m0
= − 2 (−c2 hrs − har v a hbs v b )
c( )
1 a b
= m0 hrs + 2 hra v hbs v (16.54)
c
( ( )( ))
1 1 1
= m0 hrs + 2 vr + 2 (uv)ur vs + 2 (uv)us
c c c
( )
1 [ ]
= m0 hrs + 2 vr vs − γ(vr us + vs ur ) + γ ur us .2
c
In the proper frame of ua we have:

( ) ( )
1 ∗ 1 2
Ecd hcr hds a b
= m0 hrs + 2 hra v hbs v = m0 δrs + 2 γ vr vs . (16.55)
c c
13
With an asterisk over the = sign we mean that the equality holds in the specific frame it refers, here the
proper frame of the observers ua .
14
There is another way to write this result. We have:
m0
Ecd hcr hds = − (−c2 hrs − har v a hbs v b )
c2 [( ) ]
m0 1
=− 2 hab − 2 ua ub v v hrs − har v hbs v
a b a b
c c
[ ]
m0 1
= − 2 − 2 (uv) hrs + hab hrs v v − har v hbs v
2 a b a b
c c
[ ]
m0 1
= − 2 − 2 (−γc ) hrs + (hab hrs − har hbs ) v v
2 2 a b
c c
m0 [ 2 2 ]
= − 2 −γ c hrs + (hab hrs − har hbs ) v a v b
c
m0
= m0 γ 2 hrs − 2 (hab hrs − har hbs ) v a v b
c
E2 m0
= hrs − 2 habrs v a v b
m0 c4 c
where E = m0 γc2 is the energy of the particle and:
habrs = hab hrs − har hbs (16.53)
is a bivector metric in the rest space of the four-velocity ua .

We brake this result in its trace and the trace-free part. For the trace we compute:
1 rs
hrs Ecd hcr hds = m0 (3 + h vr vs )
c2
1 1
= m0 (3 + 2 (−c2 + 2 (uv))
( c c)
1
= m0 2 + 4 (−γc2 )2
c
2
= m0 (2 + γ ). (16.56)
For the traceless part we find:

( ) ( )
1 cd 1 [ ] 1
hr hs − h hrs Ecd = m0 hrs + 2 vr vs − γ(vr us + vs ur ) + γ ur us − m0 (2 − γ 2 )hrs
c d 2
3 c 3
( )
1 1 [ ]
= m0 (1 − γ )hrs + 2 vr vs − γ(vr us + vs ur ) + γ ur us
2 2
(16.57)
3 c
We collect the above results in the following formal matrix (note that the (0, 0) element is
1
c2
E ab ua ub ):
( )
2T − 1c h(al pa )
Eab = . (16.58)
− 1c hal pa m0 hrs + 1
c2
[vr vs − γ(vr us + vs ur ) + γ 2 ur us ]
There is a more compact and instructive way to write Eab . Indeed from (16.54) we have:
( )
c d 1 c d
Ecd hr hs = m0 hrs + 2 hr hs vc vd
c
( )
c d 1 c d
= m0 hr hs gcd + 2 hr hs vc vd
c
( )
c d 1
= m0 hr hs gcd + 2 vc vd
c
c d
= m0 hr hs h(v)cd (16.59)
where h(v)cd = gcd + c12 vc vd is the projection tensor associated with the four-velocity v a . With
this observation we write:
( )
2T − 1c hal pa
Eab = . (16.60)
− 1c hal pa m0 hcr hds h(v)cd
We note that the (2,2) term in the Eab matrix is of a geometric nature, the particle entering
only through the (invariant) mass m0 .
In the Newtonian limit c → 0, γ → 1 and the proper frame of ua is the universal Newtonian
frame. In this frame the above general formula reads:
( )
2T −p
Eab = . (16.61)
−p m0 δrs
that is, Eab is defined solely in terms of the kinematic quantities of the particle and the metric
of the Euclidian space. We can extract the invariant m0 outside of the rhs and write15 :
( 2 )
v -v
Eab = m0 (16.62)
-v δrs
16.3.1 The case of a single particle under the action of forces

We come now to the case of a particle which is moving under the action of forces. In this case
the equation of motion of the particle is given by the (generalized) Newton’s law:
dpi
= fi (16.63)
ds
i
where f i is the collective four-force. The dp
ds
= ṗj = 1c pi;j uj .
We are calculating the derivative Ėab and show that this quantity contains all the information
concerning the equation of motion. We have16 :
m0
Ėcd = − 2
gabcd (v a v b ).
c
1 1
=− 2
gabcd (pa pb ). = − 2
(gab gcd − gac gbd )(pa pb ).
m0 c m0 c
1
=− (−m20 c2 gcd − pc pd ).
m0 c2
2 2
= ṗ (c pd) = f(c pd)
m0 c2 m0 c 2
2
= 2 f(c vd) . (16.64)
c
We conclude that:
2
Ėab ua ub = (fa ua )(vb ub )
c2
1
Ėab ua hbc = 2 (fa vb + fb va ) ua hbc
c
1 [ ]
= 2 (fa ua )hbc vb + (uv)hbc fb
c
2
Ėab hac hbd = 2 hac hbd f(a vb) .
c
15
In this case we can consider the symmetric tensor:
( 2 )
v -v
g(v)ab =
-v δrs
as a metric in a four-dimensional velocity space whose vectors are of the form TO BE FOUND. SEE IF THAT
OBSERVATION GIVES ANYTHING NEW.
16
Note that the derivative of the metric vanishes because the metric is flat, therefore in a Cartesian frame
the components are constant.
From these relations we calculate:

c2
fa ua = Ėab ua ub (16.65)
2(uv)
1 [ 2 ]
hbc fb = c Ėab ua hbc − (fa ua )hbc vb (16.66)
(uv)
that is the components of the four-force in the 1+3 decomposition wrt the four-vector ua (i.e.
in Σ).
It is more instructive to determine the force in the proper frame of v a . Indeed in this case
we have:
Ėab v a v b = −2(fa v a )
Ėab v a hbc (v) = −hbc fb
from which follows:

1
f r = (Ėab v a v b )v r − Ėab v a hbc (v). (16.67)
2
We conclude that the tensor Eab determines completely the motion of the particle.
16.4 The equations of motion of a relativistic fluid

In section ?? we have shown that with each particle one can associate the second rank tensor Eab
whose components are the kinematic quantities (mass and momentum) of the particle and that
the tensor Eab determines completely the motion of the particle. In this section we generalize
this concept to the case of a fluid and show that the equations of motion are obtained by the
divergence of the tensor Eab. To do that we consider an elementary comoving volume dV0 in a
relativistic fluid consisting of one type of particles and assume that the matter density of the
fluid in the proper frame of the comoving volume dV0 is ρ0. Next we let ua to be the four-velocity
of the comoving observers in dV0 and assume that in their proper frame, Σ say, the velocity of
the fluid is v. This implies that the matter density of the fluid in the frame Σ is ρ = ρ0 γ 2 .
ab
We compute the divergence E..,b in the frame Σ and compare the result with the equations
of motion of the fluid, that is, equations (16.1), (16.2) . We have17 :
∗ 1 1
E 0a ,a = E 00 ,0 + E 0µ ,µ = − (m 0 c 2 2 2
γ β ),0 − (m0 γ 2 vµ ),µ
c2 c
∗ 1 1
= − 2 (m0 c2 (γ 2 − 1)),0 − (m0 γ 2 vµ ),µ
c c
∗ 1
= (−m0 γ 2 + m0 ),0 − (m0 γ 2 vµ ),µ
c
∗ 1
= −ρ,0 + ρ0,0 − (ρvµ ),µ )
c
∗ 1
= − [cρ,0 + (ρvµ ),µ )] + ρ0,0 . (16.68)
c
17
Recall that when we rise the 0 index we change sign!
16.4. THE EQUATIONS OF MOTION OF A RELATIVISTIC FLUID 121
Concerning the space part we have:

( )
νa ν0 νµ ∗1 2 ν µν 1 2 µ ν
E ,a =E ,0 +E ,µ = (m0 γ v ),0 + m0 δ + 2 γ v v
c c ,µ
∗ 1 1
= (ρvν ),0 + 2 (ρvµ vν ),µ
c c
∗ 1 1 ν 1 1
= ρ,0 v + ρv,0 + 2 ρ,µ vν vµ + 2 ρ(vµ vν ),µ
ν
c c c c
∗ 1 1 µ 1 µ ν 1 ν 1
= ( ρ,0 + 2 ρ,µ v + 2 ρv,µ )v + ρ(v,0 + vν ,µ vµ )
c c c c c
∗ 1 µ ν 1 ν ν µ
= 2 (cρ,0 + (ρv ),µ )v + 2 ρ(cv,0 + v ,µ v ) (16.69)
c c
Now if the four-force density on the fluid element is f a we demand the equation of motion
of the fluid is:
1
E ab;b = f a . (16.70)
c
By the extension of Newton’s Second Law to Special Relativity in Σ we have that the four-force:
dpa 1 1 1
fa = = pa,b ub = (m0 ua ),b ub = m0,0 ua + m0 ua ,b ub = ρ0,0 ua + ρ0 u̇a (16.71)
dτ c c c
where u̇a = 1c ua ,b ub is the four-acceleration of the fluid. To see if (16.70) is the required
expression we have to show that in the case of a Newtonian fluid it gives the correct result,
that is the continuity equation and the Navier Stokes equation in Σ.
For that we consider this equation in the proper frame Σ of the comoving observers ua . In
this case E ab;b is given by (16.68), (16.69). the Replacing in (16.70) and comparing terms we
obtain:
∗
cρ,0 + (ρvµ ),µ = 0 Mass continuity equation
ν ∗
cv,0 + vν ,µ vµ = f ν Equation Navier Stokes.
where:
c ν ∗ c ∗ 2 ν ∗ γ̇ ν
fν =
u̇ = (γ γ̇v ν
+ γ a ) = c( v + aν ).
γ2 γ2 γ
is the spacelike part of the four-force in Σ.
We conclude that the tensor Eab determines completely the dynamics of a relativistic fluid
and furthermore the equation of motion for the fluid is equation (16.70).
16.4.1 The relativistic fluid

Following the results of section 16.4 and the 1+3 decomposition of the symmetric tensor Tab
we define a relativistic fluid by defining its physical variables by means of the following corre-
spondence:
The scalar µ corresponds to mass density,
The scalar p corresponds to isotropic pressure,
The vector q a corresponds to heat flux vector or momentum transfer vector
The traceless symmetric tensor corresponds to the traceless stress tensor.

It is assumed that all these quantities refer to that dynamic quantities of the fluid as
measured by the observers ua . Another class of observers will give different dynamical variables
for the same fluid which will not be related to those of ua unless the two observers are related!
This observation is crucial in understanding the dynamical description of a relativistic fluids.
Following the above correspondence we classify the relativistic fluids by their “complexity”
in terms of the physical properties as shown in Table ..... Each type of fluid describes certain
simplified types of a general fluid.
TABLE: Types of Energy - Momentum Tensors

a ab
µ p q π Tab Type of fluid
0 0 0 0 0 Empty space
̸ 0
= 0 0 0 Tab = µua ub Dust
̸= 0 ̸ 0
= 0 0 Tab = µua ub + phab Perfect fluid
̸= 0 ̸= 0 ̸ 0
= 0 Tab = µua ub + phab + 2q(a ub) Isotropic heat conducting non-perfect fluid
̸= 0 ̸= 0 0 ̸= 0 Tab = µua ub + phab + πab . Anisotropic fluid without heat flux
̸= 0 ̸= 0 ̸= 0 ̸= 0 Tab = µua ub + phab + 2q(a ub) + πab . General anisotropic fluid
In the following sections we discuss the physics resulting from the above classification and
show how the Navier Stokes equation of motion is expressed in terms of the geometric fluid
variables.
16.5. THE DYNAMICAL EQUATIONS OF MOTION OF A RELATIVISTIC FLUID: A SIMPLIFIED
16.5 The dynamical equations of motion of a relativistic

fluid: A simplified approach
As we have remarked above in order to define the “strain tensor” of a relativistic fluid we have
to look for a second rank tensor From the results of the previous section it follows that it is
not possible to describe relativistically a classical fluid by means of the number density ρ0 and
the 4-velocity ua , or equivalently the particle number current and the matter density current.
Therefore we have to introduce new tensor fields and our next choice is a second order tensor
T ab , which will be build from the available elements we have, that is, the number density
ρ0 , the 4-velocity ua and the Lorentz metric ηab . For reasons we have mentioned in the last
section this tensor must be symmetric and that the equations we are looking for are of the form
T ab ,b = f a where f a is the external ”force” density - which is not part of the fluid and refers
to the environment. This requirement defines the equation of motion of the relativistic fluid in
Special Relativity.
In order to define the tensor T ab we note that the possible choices we have using the available
tensors are the following:
Tab = ρ0 ηab , Tab = ρ0 ua ub , Tab = ρ0 ua ub + Aηab .
The first choice is rejected because it does not contain the 4-velocity ua of the fluid. The
second choice is the simplest and it is contained in the last. We choose the second because the
last contains the unknown quantity A. If necessary we shall return to it.
We consider Tab = ρ0 ua ub and compute its components in a general Lorentz frame Σ. We
have: ( ) ( )
γc γc
Tab = ρ0 ⊗
γv Σ
γv Σ
where ⊗ denotes tensor product. It follows:
( 2 )
c cvµ
Tab = ρ (16.72)
cv µ v µ vν Σ
where18 ρ = ρ0 γ 2 . We note immediately that the unwanted quantities γ have been absorbed
into the matter density ρ, which is one of the quantities we want to use in our equations.
We compute next the divergence T ab ,b and from that the equations resulting if we set it
equal to f a . We write for the coordinates in Σ x0 = ct and xµ = {x, y, z} and find:
[ ]
∂ρ
0a 00 0µ 2 µ
T ,a = T ,0 + T ,µ = (ρc ),0 + (ρcv ),µ = c + ▽(ρv) .
∂t
If we demand T 0a ,a = 0 (we assume that there are no sinks or sources) then it follows that
in Σ this condition gives:
∂ρ
+ ▽(ρv) = 0 (16.73)
∂t
which is the continuity equation for the mass density. Obviously equation (16.73) is not
covariant because T 0a ,a is a component of the four vector T ab,b . However the form of the equation
18
Recall that according to our conventions lower indices denote rows and upper indices denote columns
has been derived for an arbitrary Lorentz frame Σ and the difference from another frame Σ′
will only be in the particular values of the non-relativistic quantities ρ and v not in its form.
We continue with the spatial components of the vector T ab ,b . We compute in Σ:
T νa ,a = T ν0 ,0 + T νµ,µ = (cρv ν ),0 + (ρv µ v ν ),µ

= cρ,0 v ν + cρv,0
ν
+ ρ,µ v ν v µ + ρ(v µ v ν ),µ
= (cρ,0 + ρ,µ v µ + ρv,µ
µ
)v ν + ρ(cv,0
ν
+ v ν ,µ v µ )
= (cρ,0 + (ρv µ ),µ )v ν + ρ(cv,0
ν
+ v ν ,µ v µ ).
We note that the first part in the rhs is the continuity equation for the mass density in Σ
and the second part is the lhs of Euler’s equation of motion in Σ. Therefore if we assume that
the continuity equation is satisfied we find that Euler’s equation of motion is satisfied provided
the space part of the vector T,bab satisfies the condition:
T νa ,a = −p,ν + f µ (16.74)
where p is the isotropic pressure of the fluid, and f µ the external force density. We conclude
that:
a. The choice of T ab = ρ0 ua ub defines at most perfect fluids
b. The time part T 0a ,a = 0 of the vector T,bab corresponds to the continuity equation (for
matter)
c. The space part T νa ,a of the vector T ,b must equal T ,a = −p + f in order that Euler’s
ab νa ,ν µ
equation of motion is satisfied.

We note that if the pressure p is a constant then T ab = ρ0 ua ub satisfies both equations. A
perfect fluid for which the pressure p is zero, is described by the tensor T ab = ρ0 ua ub and we
call it dust. Dust is the simplest perfect fluid and it is described only in terms of the matter
density and the four-velocity.
The isotropic pressure is an internal element of the Newtonian fluid therefore it must be
included in the relativistic description of the fluid. This means that we are looking for a
generalization of T ab which will be such that the Euler’s equation of motion will follow form
the condition T νa ,a = f µ . We consider:
T ab = ρ0 ua ub + sab (16.75)
where sab is a symmetric tensor such that:

a. sab ub = 0 (in order to leave the T 0a ,a component the same - because we have shown
that it works)
b. sνa ,a = p,ν (in order to have that T νa ,a = f µ ).
c. The tensor sab will be constructed from the available elements: p, ηab , ua .
The most general tensor which can be constructed from the elements ηab , ua is:
sab = Aua ub + Bηab .
Condition sab ub = 0 implies:
−Ac2 ua + Bua = 0 ⇒ B = Ac2

16.6. THE ENERGY MOMENTUM TENSOR OF A RELATIVISTIC VISCOUS FLUID125
so that sab becomes:

1
sab = A(ηab + ua ub ) = Ahab (16.76)
c2
where:
1
hab = ηab +
ua ub (16.77)
c2
is the tensor which projects normal to the vector ua , that is hab ub = 0.
We examine next condition sνa ,a = p,ν . We consider the comoving frame in which the
( )
a c
four-velocity u = and using that in a Lorentz frame ηab = diag(−1, 1, 1, 1) we find:
0
( ) ( )
1 c c
hab = diag(−1, 1, 1, 1) + 2 ⊗ = diag(0, 1, 1, 1). (16.78)
c 0 0
It follows that in that frame:
sab = Adiag(0, 1, 1, 1) ⇒ sab ,b = A,a
and condition sνa ,a = p,ν implies A = p (up to a constant which we take it to be zero, the
case of dust). Therefore we have:
sab = phab
ab
and we conclude that the tensor T we are looking for is:
T ab = ρ0 ua ub + phab . (16.79)
Tab has the property that its divergence T ab,b produces both the energy continuity equation
and Euler’s equation of motion for a perfect fluid provided we set T,bab = f a . An equivalent
form of the tensor T ab is:
T ab = (ρ0 + p)ua ub + pηab . (16.80)
The tensor T ab we call the energy-momentum tensor of a (relativistic) perfect fluid.
It contains all dynamics of the fluid because:
(a) It is constructed from ρ0 and p which are the relativistic invariants describing the perfect
fluid
(b) its divergence produces the two equations which characterize the development of the
fluid.
16.6 The energy momentum tensor of a relativistic vis-

cous fluid
The energy momentum tensor of a viscous fluid is a tensor T ab such that the vanishing of its
divergence T ab,b will produce both:
a. The continuity equation for the energy density Dρ
dt
=0
b. The Navier Stokes equation of motion:
Dv µ
ρ = ρf µ + ηtµν ,ν − p,µ (16.81)
dt
where ϕµν = η(v µ,ν ν + v ν,µ ν − 23 v,ρρ δµν ) is the traceless stress tensor of the viscous fluid, p is the
isotropic pressure of the perfect fluid part of the fluid and f µ is the external force density. v µ
is the four-velocity vector. This equation can be written as:
Dv µ
ρ + p,µ = ρf µ + tµν ,ν .
dt
The lhs is the Euler equation of motion for a perfect fluid, therefore it will be produced
from the Tab = ρ0 ua ub + phab . There remains only the term ϕµν,ν to be considered. We define:
Tab = ρ0 ua ub + phab + ϕab (16.82)
where ϕab is the stress tensor corresponding to the strain motions in the relativistic fluid. In
order to compute the form of the tensor ϕab we note that ϕab is symmetric, traceless and it is
restricted by the condition:
ϕab ub = 0.
Using the 1+3 decomposition wrt the 4-velocity vector ua we find that the most general form
of the tensor ϕab is:
ϕab = 2q(a ub) + πab (16.83)
where the vector qa = hba qb (⇔ qa ua = 0) and the traceless tensor πab is such that πab =
hca hcb πcd or, equivalently, πab ua = πab ub = 0. We conclude that the required form of the energy
momentum tensor for a viscous fluid is:
Tab = ρ0 ua ub + phab + 2q(a ub) + πab . (16.84)
The physical significance of each term in (16.84) is as follows:

- ρ0 is the energy density of the fluid (in the local rest frame)
- p is the isotropic pressure in the fluid
- qa is a vector which has to do with the flux of energy in the fluid due to internal friction
forces (this will be proved later).
- πab is the traceless stress tensor in the fluid responsible for the development of strains in
the fluid.
It is important to note that expression (16.84) is the 1+3 decomposition of Tab wrt the
four-velocity vector ua which we could write immediately. However with the analysis we have
made we have identified a physical role for each irreducible part of the decomposition. The
particular form of each irreducible part will be determined form the condition T ab ,b = f a where
f a is the external force.
Example 16.6.1 It is possible to associate a second rank symmetric tensor with a relativistic
particle in a different way. Consider a particle of four-momentum pa moving in a coordinate
frame of four-momentum ua in a spacetime with metric gab and show that the kinetic energy of
the particle can be written in the form:
1
T = Bab g ab (16.85)
2
where:
1 1
Bab = pa pb − p(a ub) .
c2 pc uc
16.6. THE ENERGY MOMENTUM TENSOR OF A RELATIVISTIC VISCOUS FLUID127
Compute the matrix representation of Bab and the divergence of Bab and show that it gives the
continuity equation and the Navier Stokes equation.
Solution
We have:
p2 c2 1 [ 2 ] 1 [ 2 ] c2
2T = = E − m2 c 2 = E + c2 gab pa pb = E + gab pa pb
E E E E
( ) ( )
E/c c
The inner product pa ua = = −E therefore:
p Σ
0 Σ
[ 2 ]
c2 c a b
2T = −pc u −
c
gab p p = −gab
a b
p p + p u = −Bab g ab
(a b)
pc uc pc u c
2
where Bab = pccuc pa pb + p(a ub) .
Obviously 2T equals the minus the trace of Bab .Concerning the 1+3 decomposition of Bab
we have the following.
Bab ua ub = c2 pc uc − c2 pc uc = 0
c2 c2
Bab ua hbc (u) = c2 hbc (u)pb − hbc (u)pb = hbc (u)pb
2 2
2
c
Bab hac (u)hbd (u) = d
hac (u)pa hbd (u)pb .
pd u
c2 cd
Bab hab (u) = h (u)pc pd
pe u e
( ) ( )
1 ab c2 1 ab
Bab hc (u)hd (u) − h (u)hcd (u) =
a b
hc (u)hd (u) − h (u)hcd (u) pa pb .
a b
3 pe u e 3
Therefore:
( c2 b
)
0 h (u)pb [ 2 ]
[Bab ]Σ = c2 b
2 c
c2
( a )
h (u)pb
2 c pe ue
hc (u)hbd (u) − 13 hab (u)hcd (u) pa pb + 31 pecue hef (u)pe pf hcd (u)
Concerning the divergence of Bab we have:
0a 00 0µ 0µ c2 ( b )
B;a = B;0 + B;µ = B;µ = hc (u)pb ;µ = ...
2
µa µ0
B;a = B;0 + B;νµν = ...
TO BE CONTINUED. THE FIRST MUST GIVE THE ZERO COMPONENT OF THE

FOUR-FORCE AND THE SECOND THE SPATIAL COMPONENT OF THE FOUR-FORCE.
POSSIBLY UP TO A FACTOR OF 2.
16.7 The energy momentum tensor of the electromag-

netic field
We consider an electromagnetic field in empty space described by the field tensor Fab which
satisfies Maxwell equations (in SI units):
F ab ,b = −µ0 j a (16.86)
3F{ab,c} = Fab,c + Fbc,a + Fca,b = 0. (16.87)
where j a is the four-current. The requirement of charge conservation implies the continuity
equation for the charge density (provided there are no sinks or sources):
a
j,a = 0. (16.88)
The Lorentz force acting on the four-current is given by the formula:
1
f a = − F ab j b . (16.89)
c
We note that the Lorentz force is normal to the four-current j b .
We wish to determine an energy momentum tensor T ab for the electromagnetic field. We
require that this tensor:
a. Must be of the form (16.82)
b. The stress tensor ϕαb of the electromagnetic field must satisfy the requirement ϕµb ,b =
− 1c F ab j b .
To find this tensor we note from (16.89) that due to Maxwell equations the Lorentz force
can be written as follows:
1 1 ab d
f a = − F ab j b = F Fb,d . (16.90)
c cµ0
From this we conclude that the required tensor must be quadratic in the tensor F ab and it
can also contain the Lorentz metric ηab . The most general19 form of a symmetric second rank
tensor satisfying these properties is the following:
ϕab = AF ac Fc b
+ Bη ab (F cd Fcd ) (16.91)
where the scalars A, B must be determined form the requirement ϕµb ,b = − ρc0 F µb ub . We
compute the divergence ϕab ,b :
ϕab ,b = AF ac ,b Fc b
+ AF ac Fc b ,b + 2Bη ab F cd Fcd,b
[ ]
= AF ac Fc b ,b + F cb AF a c,b + 2Bη ad Fcb,d
[ ]
= A.cµ0 f a + F cb AF a c,b + 2Bη ad Fcb,d
19
Modulo a term with vanishing divergence.
16.7. THE ENERGY MOMENTUM TENSOR OF THE ELECTROMAGNETIC FIELD 129
The second term is given as follows:

[ ] [ ]
F cb AF a c,b + 2Bη ad Fcb,d = Fcb 2BF cb,a + AF ac,b
[ ( ) ]
= Fcb 2B −F ba,c − F ac,b + AF ac,b
[ ( ) ]
= Fcb 2B −F ac,b − F ac,b + AF ac,b
= (−4B + A)Fcb F ac,b .
Therefore:
ϕab ,b = Acµ0 f a + (−4B + A)Fcb F ac,b . (16.92)
and the requirement ϕµb ,b = f µ gives:
1 1
A= , B= .
cµ0 4cµ0
We conclude that the required energy momentum tensor of the electromagnetic field is:
[ ]
ab 1 b ac 1 ab cd
ϕ = F F + η (F Fcd ) . (16.93)
cµ0 c 4
Exercise 16.7.1 Prove that the trace of the energy momentum tensor for the electromagnetic
field vanishes, that is, ηab ϕab = 0.
In order to compute the stress tensor ϕab of the electromagnetic field in terms of the electric
filed E and the magnetic field H in a frame Σ we use the familiar expression of the field tensor
F ab in terms of the components of the electric and the magnetic fields:
 
0 Ex /c Ey /c Ey /c
 -Ex /c 0 −Bz By 
Fab =  -Ey /c Bz
. (16.94)
0 −Bx 
-Ey /c -By Bx 0
The term F cd Fcd is the invariant −2X of the electromagnetic field. Therefore we have::
( )
1 2
F Fcd = −2X = −2 2 E − B .
cd 2
(16.95)
c
Concerning the term Fc b F ac we let the reader to show20 that equals the energy momentum
tensor of the electromagnetic field we computed in (15.17) i.e.:
[ ]
F ac Fcb = −E2 δ0a δb0 − Sµ δ0a δbµ + (E µ )2 + (Bµ )2 − B2 δµa δbµ + [E µ E ν + B µ B ν ] δµa δbν (16.96)
20
We give the computation in brief.
The 00− term:
F 0c Fc0 = −F 0c F0c = −E µ Eµ = −E2
The 0µ - term:
E
F 0c Fcµ = F 0ν Fνµ = −E ν Fνµ = ... = −( × B)µ = −Sµ
c
The µν- term:
Fc µ F νc = F0 µ F ν0 + Fρ µ F νρ = E µ E ν + Fρ µ F νρ
where S µ = ( Ec × B)µ is the Poynting vector. Then the energy momentum tensor (or the stress
tensor) of the electromagnetic field is:
1
cµ0 ϕab = F ac Fcb + η ab (F cd Fcd )
4 [ ]
1 E2 1 E2
= − ( 2 + B )δ0 δb − Sµ δ0 δb + (E ) + (Bµ ) − ( 2 + B ) δµa δbµ + [E µ E ν + B µ B ν ] δµa δbν
2 a 0 a µ µ 2 2 2
2 c 2 c
(16.97)
and in matrix form: ( 1 E2

)
(
2 c2
+ B2 ) −Sµ
cµ0 ϕab = (16.98)
−S µ
tµν
where the tensor tµν = ϕµν (the stress tensor of the electromagnetic filed) is given by (compare
also with (15.19)):
 2 
Ex + Bx2 − X Ex Ey + Bx By Ex Ez + Bx Bz
tµν =  Ex Ey + Bx By Ey2 + By2 − X Ey Ez + By Bz  (16.99)
Ex Ez + Bx Bz Ey Ez + By Bz Ez + Bz − X
2 2
We conclude that in Special Relativity the energy momentum tensor of a charged perfect
fluid interacting with the electromagnetic field (it turns out that it is the same in General
Relativity with the partial derivative being replaced with the covariant derivative) in terms of
the 3-vectors E and B (not of the four vectors E a and B a !) is:
[ ]
ab 1 ac b 1 ab cd
T = ρ0 ua ub + phab + F F.c + η (F Fcd )
cµ0 4
( 1 E2 )
( 2 + B2
) + ρ 0 −S µ
= 2 c
−S µ pδµν + tµν
Concerning the last term we find for µ = ν = 1 :
Fρ 1 F 1ρ = −(F 12 )2 − (F 13 )2 = (F 23 )2 − B2 = Bx2 − B2
We conclude that for µ = ν we have:
Fρ µ F µρ = (E µ )2 + (B µ )2 − B2 .
For µ ̸= ν we have:
Fρ µ F νρ = F1 µ F ν1 + F2 µ F ν2 + F3 µ F ν3
= F1 3 F 31 (δ2µ δ3ν + δ2ν δ3µ ) + F2 1 F 32 (δ1µ δ3ν + δ1ν δ3µ ) + F3 1 F 23 (δ1µ δ2ν + δ2ν δ1µ )
= Bz By (δ2µ δ3ν + δ2ν δ3µ ) + Bx Bz (δ1µ δ3ν + δ1ν δ3µ ) + Bx By (δ1µ δ2ν + δ2ν δ1µ )
= BµBν
Collecting all these we have (no sum over the indices µ, ν):
[ ]
F ac Fcb = −E2 δ0a δb0 − Sµ δ0a δbµ + (E µ )2 + (Bµ )2 − B2 δµa δbµ + [E µ E ν + B µ B ν ] δµa δbν .
16.7. THE ENERGY MOMENTUM TENSOR OF THE ELECTROMAGNETIC FIELD 131
Exercise 16.7.2 Show that the Lorentz force equation f a = 1c F ab jb can be written as f a =
∂EM T ab ab 1
[ ac b 1 ab cd
]
∂x b where EM T = cµ0
F F c + 4
η (F F cd ) is the energy momentum tensor of the
electromagnetic field.
16.8 The Minkowski energy momentum tensor

This section will be revised with previous notes made on the electromagnetic field and its energy
momentum.
The Minkowski energy momentum tensor for the electromagnetic field is defined as follows:
1
EM Tab = Fac Kb c − (Fcd K cd )gab (16.100)
4
where the field tensors Fabc , Kab have been defined in (??) and (??). We have the following
result.
Proposition 16.8.1 In terms of the electromagnetic fields E a , B a , H a , Da , H a the energy mo-

mentum tensor EM Tab is given by the expression:
1
EM T
ab
= (ua ub + g ab )(E c Dc + B c Hc ) − (E a Db + H a B b ) + (ua S b + ub P a ) (16.101)
2
where:
S a = η abcd Eb Hc ud (16.102)
P a = η abcd Db Bc ud (16.103)
In he rest frame of the observer ua the 4-vector S a we call the Pointing vector and the
4-vector P a we call the electromagnetic momentum density vector.
Proof 16.8.1 We have:

( )
F ac Kbc = −E a uc + E c ua + η acde Bd ue (−Eb uc + Ec ub + ηbcrs B r us )
= (E c Dc )ua ub + S b ua − E a Db + P a ub + g bh η acde ηhcrs ud Be ur H s
The last term gives upon expanding the contraction η acde ηhcrs :g bh η acde ηhcrs ud Be ur H s = g ab (B c Hc )+
ua ub (B c Hc ) − B b H a
Replacing we find:
F ac Kbc = ua ub (E c Dc + B c Hc ) + (S b ua + P a ub ) − (E a Db + B b H a ) + g ab (B c Hc ) (16.104)
Contract a, b to find:
F ab Kab = −2(E c Dc + B c Hc ) + 4(B c Hc ). (16.105)
Replacing in (16.100) we find the required result.
From the above result we note the following:

ab
1. The Minkowski energy momentum tensor EM T is not symmetric.
2. By taking Kab = Fab in (16.105) we find the useful result (this is also one of the invariants
of the electromagnetic field):
F ab Fab = −2(E 2 − H 2 ). (16.106)

16.8. THE MINKOWSKI ENERGY MOMENTUM TENSOR 133
In the rest frame of ua , EM T

ab
has the following components:
∗ 1
EM T00 = (E c Dc + B c Hc ) (16.107)
2
∗
EM T0µ = Sµ (Pointing vector) (16.108)
∗
EM Tµ0 = Pµ (Electromagnetic 4-momentum vector) (16.109)
∗ 1
EM Tµν = gµν (E c Dc + B c Hc ) − (E µ Dν + H µ B ν ) (Maxwell spatial stress tensor). (16.110)
2
The vector P a is obviously spacelike. In order to compute its coordinates in the Lorentz
orthogonal rest frame of ua we note first that in that frame ua = δ0a , ua = −δa0 and gab =
diag(−1, 1, 1, 1).Then21 :
P µ = η µ0νρ u0 Dv Bρ = εµνρ Dν Bρ = (D × B)µ . (16.111)
Similarly we show that the components of the Pointing vector in this frame are:
S µ = η µ0νρ u0 Ev Hρ = εµνρ Eν Hρ = (E × H)µ . (16.112)

In the isotropic case with the constitutive relations Da = εEa , Ba = λHa where ε is the
electric permittivity and λ is the magnetic permeability the energy momentum:
1
EM T
ab
= (ua ub + g ab )(εE 2 + λB 2 ) − (εE a E b + λH a H b ) + (ua S b + ub P a ) (16.113)
2
and in the rest frame:
∗ 1
EM T00 = (εE 2 + λH 2 ) (16.114)
2
∗
EM T0µ = (E × H)µ (Pointing vector) (16.115)
∗
EM Tµ0 = ελ(E × H)µ (Electromagnetic 4-momentum vector) (16.116)
∗ 1
EM Tµν = gµν (εE 2 + λH 2 ) − (εE µ E ν + λH µ H ν ) (Maxwell spatial stress tensor). (16.117)
2
We note that the EM T ab is not symmetric even in the case of isotropic material with con-
stitutive relations Da = εEa , Ba = λHa 22 .
In the following we assume that the material is isotropic therefore the constitutive equations
Da = εEa , Ba = λHa are applicable. This implies that the electromagnetic 4-momentum vector
P a is=related to the Pointing vector as follows:
P a = ελS a . (16.118)
Finally starting from the definition of K ab and replacing Da = εEa , Ba = λHa we show
easily that:
1 ελ − 1 a b
K ab = F ab + (u E − ub E a ). (16.119)
λ λ
21
Recall that η abcd = |g|11/2 εabcd .
22
As we shall see it is symmetric in the MHD approximation of infinite electic conductivity.
ab
16.8.1 The 1+3 decomposition of the EM T
The energy momentum tensor EM T ab can be 1+3 decomposed wrt the vector ua as it is done
with the general (not necessarily symmetric) energy momentum tensor. To do that we note
that EM T ab in (??) can be written as follows:
ab 1 1
EM T = (εE 2 + λB 2 )ua ub + (εE 2 + λB 2 )hab + (ua S b + ub ελS a )
2 6
1
+ (εE 2 + λB 2 )hab − (εE a E b + λH a H b )
3
= µem ua ub + pem hab + λµqem
a
ub + ua qem
b ab
+ πem
where:
1
µem = (εE 2 + λB 2 ) (16.120)
2
1
pem = (εE 2 + λB 2 ) (16.121)
6
a
qem = Sa (16.122)
1
ab
πem = (εE 2 + λB 2 )hab − (εE a E b + λH a H b ). (16.123)
3
ab
We note that πem is traceless as it is expected.
On application of the above result is the determination the equation of state for isotropic
radiation (for example the background radiation in the Universe). In this case:
a ab
qem = 0, πem =0 (16.124)
and the energy momentum tensor becomes:
ab 1 1
EM T = (εE 2 + λB 2 )ua ub + (εE 2 + λB 2 )hab (16.125)
2 6
from which follows that:
1
pem = µem . (16.126)
3
Chapter 17
Classical Differential Operators
Note: Under construction

Suppose is a system of curvilinear coordinates in E 3 which is related to the standard Carte-
sian coordinate system {x = x1 , y = x2 , z = x3 } by the relations xµ = xµ (q 1 , q 2 , q 3 ). The q µ −
coordinate line is the set of points with position vector:
r = x(q ν̸=µ = const)i + x(q ν̸=µ = const)j + x(q ν̸=µ = const)k (17.1)
∂r
The tangent to this line is ∂q µ
. Let:

∂r
hµ = µ (17.2)
∂q
∂r µ ∂r
be the length of ∂q µ and eµ the unit tangent to the q coordinate line. Then ∂q µ = hµ e µ .
The coordinate q µ is the parameter along the q µ − coordinate line. An affine parameter sµ
along the q µ − coordinate line is defined as that parameter for which the length of the tangent
vector is unity, that is the tangent vector is eµ . This implies:
∂r ∂r ∂sν ∂sν ∂sν
µ
= ν µ
⇒ hµ eµ = eν µ ⇒ hµ = µ (eµ · eν ) (no summation over µ)
∂q ∂s ∂q ∂q ∂q
from which follows:
(eµ · eν )dsν = hµ dq µ (no summation over µ) (17.3)
We say that the coordinate system {q µ , µ = 1, 2, 3} is orthogonal if eµ · eν = δµν and

orthonormal if in addition hµ = 1 (for all µ, ν = 1, 2, 3).
For an orthogonal coordinate system we have:
dsµ = hµ dq µ (no summation over µ). (17.4)
It is useful to find an expression of the arc length in terms

√ of the metric tensor components.
The component gµν of the metric tensor is defined as gµν ≡ ∂q µ · ∂q ν where ∂q µ is the tangent
∂r ∂r ∂r
vector to µ−coordinate line. The arc length ds2 is defined by the relation:
ds2 = gµν dq µ dq ν (17.5)
135
136 CHAPTER 17. CLASSICAL DIFFERENTIAL OPERATORS
where {dq µ } is the dual basis of {eµ }. The element of arc length along the µ−coordinate line
is:

√ ∂r µ
dsµ = gµµ dq = µ dq = hµ dq µ ⇒
µ
∂q
√
hµ = gµµ and dsµ = hµ dq µ (no summation over µ). (17.6)
17.0.2 The differential operators

The differential operators in classical vector theory is the grad, div, curl. In the following we
compute define these operators and we compute them in orthogonal curvilinear coordinates.
The operator grad is defined as follows (summation over µ!):
∂
grad := eµ (17.7)
∂sµ
where sµ is arc length along the coordinate q µ and eµ is the unit tangent to the corresponding
coordinate line. From (17.4) we have:
1 ∂
grad := eµ . (17.8)
hµ ∂q µ
The div operator is defined as follows. Consider a region R of volume V enclosed by a

(smooth) surface S. Suppose that in R there is defined a vector field A. We define the divA as
the limit of the flux of A through S as V → 0, that is we define:
∫∫
1
divA = lim A · ndS (17.9)
V →0 V
S
The divergence of a vector field is a scalar function of r and it is a scalar field (i.e its value is
independent of the coordinate system used). The scalar field divA exists only if the components
µ
Aµ (in a system of rectangular Cartesian coordinates x1 , x2 x3 ) and their first derivatives ∂A ∂xµ
(summation over µ) exist and are continuous. From Gauss Theorem we have:
∫∫ ∫∫∫
1 1 ∂Aµ
A · ndS = dV.
V V ∂xµ
S V
µ
If V shrinks to some interior point P of R then the right hand side becomes ∂A
∂xµ
evaluated
at that point, hence:
∂Aµ ∂A1 ∂A2 ∂A3
divA = µ = + + . (17.10)
∂x ∂x1 ∂x2 ∂x3
In orthogonal coordinates {q µ } divA has a different expression. We note that Gauss
Theorem can be written: ∫∫ ∫∫∫
A · ndS = divAdV. (17.11)
V
S
137
This form of Gauss’Theorem is called the Divergence Theorem. Geometrically this theorem
says that the integral of the divergence of a vector field over a (connected) volume V equals the
flux of the vector field through the surface S bounding V (provided the vecotr field is suitably
smooth inside V and on S). We can write the divergence as the inner product of ▽ operator
and the vector field A as follows:
∂Aµ ∂
divA = µ
= iµ µ · A = ▽ ·A. (17.12)
∂x ∂x
Besides the divergence of a vector field A there is another important operator the curl A.
This is defined as follows. Consider a region R of volume V enclosed by a (smooth) surface S.
Suppose that in R there is defined a vector field A. We define the divcurlA as the limit of the
flux of A through S as V → 0, that is we define:
∫∫
1
curlA = lim n × AdS (17.13)
V →0 V
S
2 2 2 2
Finally we have the operator ∆ = divgrad = ∂xµ∂∂xµ = (∂x∂ 1 )2 + (∂x∂ 2 )2 + (∂x∂ 3 )2 in (Euclidean)
Cartesian coordinates only.
In orthogonal coordinates {q µ } we have the formulae:
[ ]
1 ∂ ∂ ∂
divA = (A1 h2 h3 ) + 2 (A2 h3 h1 ) + 3 (A3 h1 h2 ) (17.14)
h1 h2 h3 ∂q 1 ∂q ∂q
[ ]
1 ∂ ∂
curlA= (A3 h3 ) − 3 (A2 h2 ) e1
h2 h3 ∂q 2 ∂q
[ ]
1 ∂ ∂
+ (A1 h1 ) − 1 (A3 h3 ) e2
h3 h1 ∂q 3 ∂q
[ ]
1 ∂ ∂
+ (A2 h2 ) − 2 (A1 h1 ) e3 (17.15)
h1 h2 ∂q 1 ∂q
[ ( ) ( ) ( )]
1 ∂ h2 h3 ∂f ∂ h3 h1 ∂f ∂ h1 h2 ∂f
∆f = div gradf = + 2 + 3 (17.16)
h1 h2 h3 ∂q 1 h1 ∂q 1 ∂q h2 ∂q 2 ∂q h3 ∂q 3
Examples
a. Cylindrical coordinates
Transformation equations:
x = r cos ϕ, y = r sin ϕ (17.17)

q 1 = r, q 2 = ϕ, q 3 = z.
We compute hµ form the formula:
ds2 = (dx1 )2 + (dx2 )2 + (dx3 )2 = h21 (dq 1 )2 + h22 (d22 )2 + h23 (dq 3 )2
138 CHAPTER 17. CLASSICAL DIFFERENTIAL OPERATORS
It follows:
h1 = 1, h2 = r, h3 = 1 (17.18)
Then we find:
1 ∂f ∂f
gradf = er + eϕ + ez (17.19)
r ∂ϕ ∂z
1 ∂ 1 ∂Aϕ ∂Az
divA = (rAr ) + + (17.20)
r ∂r r ∂ϕ ∂z
( ) ( ) ( )
1 ∂Az ∂Aϕ ∂Ar ∂Az 1 ∂(rAϕ ) ∂Ar
curlA= − er + − eϕ + − ez (17.21)
r ∂ϕ ∂z ∂z ∂r r ∂r ∂ϕ
( )
1 ∂ ∂f 1 ∂ 2f ∂ 2f
div curlf = r + 2 2+ 2 (17.22)
r ∂r ∂r r ∂ϕ ∂z
a. Spherical coordinates
Transformation equations:
x = r cos ϕ sin θ, y = r sin ϕ sin θ, z = r cos θ (17.23)

q 1 = r, q 2 = θ, q 3 = ϕ.
We compute hµ form the formula:
ds2 = (dx1 )2 + (dx2 )2 + (dx3 )2 = h1 (dq 1 )2 + h2 (d22 )2 + h3 (dq 3 )2
It follows:
hr = 1, h2 = r, h3 = r sin θ (17.24)
Then we find:
∂f 1 ∂f 1 ∂f
gradf = er + eθ + eϕ (17.25)
∂r r ∂θ r sin θ ∂ϕ
1 ∂ 2 1 ∂ 1 ∂Aϕ
divA = 2
(r Ar ) + (Aθ sin θ) + (17.26)
r ∂r r sin θ ∂θ r sin θ ∂ϕ
( ) ( ) ( )
1 1 ∂Az ∂Aϕ ∂Ar ∂Az 1 ∂(rAϕ ) ∂Ar
curlA= − er + − eϕ + − ez (17.27)
r sin θ r ∂ϕ ∂z ∂z ∂r r ∂r ∂ϕ
( )
1 ∂ ∂f 1 ∂ 2f ∂ 2f
div curlf = r + 2 2+ 2 (17.28)
r ∂r ∂r r ∂ϕ ∂z
Chapter 18
Problems on Newtonian fluids
We say that a viscous fluid has laminar flow or that it moves in a plane parallel shear
flow if there exists a rectangular coordinate system in which the velocity of the fluid is of the
form:
u = (u(y, t), 0, 0)
where the axes x, y are on the plane of the laminar flow. For a laminar flow we have that:
1. The fluid must be incompressible (because ∇u = 0 by definition). Therefore its density
ρ =constant.
2. The velocity of all the particles of the fluid at all times are (say) parallel to the fixed
x−direction therefore the only possible change is in the speeds.
3. The stream surfaces are planes parallel to each other
The particles which flow on the plane (stream surface) z=constant always stay on that plane
(because the velocity does not have a component out of that plane)
4. The flow lines for a laminar flow coincide with the path lines. To prove this we compute
the flow lines form the solution of the Lagrange system:
dx dy dz
= = ⇒ x = u(y, t)dt, y = C1 , z = C2 ???????????
u(y, t) 0 0
where λ is an arbitrary parameter. The pathlines are the solution of the system of equations:
dx dy dz
= u(y, t), = = 0 ⇒ x(t) = u(C1 , t)dt, y = C1 , z = C2 .
dt dt dt
The Navier Stokes equation for a fluid moving in laminar flow gives:
∂u 1 ∂p ∂ 2u
=− +ν 2 (18.1)
∂t ρ ∂x ∂y
∂p ∂p
= =0 (18.2)
∂x ∂z
where ν = ηρ is the kinematic viscosity.
∂p
From (18.2) follows that the pressure p = p(x, t).From (18.2) we have that the term ∂x
equals the difference of two terms which are not functions of x.This implies that:
p = −p0 x + f (t) (18.3)
∂u ∂ 2u p0
−ν 2 =− (18.4)
∂t ∂y ρ
139
140 CHAPTER 18. PROBLEMS ON NEWTONIAN FLUIDS
where −p0 = ∂x∂p

(p0 > 0) is a constant. We have taken −p0 (with (p0 > 0!!) because the gradient
of pressure is negative in the direction of increasing velocity (since during the motion under
pressure the velocity increases and the pressure is starts decreases from higher pressure to lower
pressure).
For steady flow the speed ∂u ∂t
= 0 hence u=u(y) only and the Navier Stokes equation gives:
∂ 2u
p0 = −η (18.5)
∂y 2
An alternative way to write this equation is:
∂u
τxy = −η (18.6)
∂y
where τxy is the only surviving component of the stress tensor. A viscous fluid which moves so
that relation (18.6) is satisfied we call a Newtonian fluid. Note that a given viscous fluid is
possible to move in a way so that (18.6) is not satisfied. Therefore it is misleading to speak
about Newtonian fluids and one should talk about Newtonian flows. In fact, to take it
further, the term ‘Newtonian fluid’ should be discarded as misleading and the term laminar or
preferable plane parallel shear flow should be used instead.
To show the equivalence of the equations (18.6) and (18.6) for steady laminar flows we
consider the derivation of (18.5) from first principles.
Suppose an incompressible fluid moves in laminar flow in the y − z plane with speed u(y)
(i.e. the velocity is parallel to the z−axis). Consider a thin rectangular slab (shell) of thickness
∆y perpendicular to the x- direction extending a distance W in the x- direction and a distance
L in the z- direction as shown in Figure.... A ’rate of z- momentum’ balance over this thin
shell has the general form:
Rate of z- momentum: In - Out + Generation = Accumulation
At steady-state, the accumulation term is zero. Momentum can go ’in’ and ’out’ of the shell
by both the convective and molecular mechanisms1 . Since vz (x) is the same at both ends of the
system, the convective terms cancel out because2 (ρvz vz W ∆y)|z=0 =(ρvz vz W ∆y)|z=L . Only the
molecular term (W Lτyz ) remains to be considered, whose ’in’ and ’out’ directions are taken in
the positive direction of the z- axis. Generation of z- momentum occurs by the pressure force
acting on the surface [pW ∆y] and gravity force acting on the volume (ρL∆yg).
The different contributions may be listed as follows:
• rate of z- momentum in by viscous transfer across planar surface at y is (W Lτyz )|r
• rate of z- momentum out by viscous transfer across planar surface at y + ∆y is

(W Lτyz )|y+∆y
• rate of z- momentum in by overall bulk fluid motion across planar surface at z = 0

is (ρvz vz W ∆y)|z=0
1
The convective mechanism refers to the flow froma macroscopic point of view and the molecular mechanism
to the interation of the shell with the rest of the fluid. The molecular mechanism is the ‘friction’ mechanism
which is responsible for the development of stresses in the fluid.
2 dpz d d
dt = dt (ρV vz ) = dt (ρW ∆yzvz ) = ρvz vz W ∆y
141
• rate of z- momentum out by overall bulk fluid motion across planar surface at z = L
is (ρvz vz W ∆y)|z=L
• pressure force acting on planar surface at z = 0 is p(0)W ∆y
• pressure force acting on planar surface at z = L is −p(L)W ∆y
• gravity force acting in z−direction on volume of shell is (ρW ∆yLg).
On substituting these contributions into the z- momentum balance, we get:
(LW τyz )|y − (LW τyz )y+∆y + (p(0) − p(L)) W 394y + ρgW L394y = 0 (18.7)
Dividing this equation by LW 394y yields:

( )
(τyz )|y+∆y − (τyz )|y p0 − pL + ρgL
= (18.8)
∆y L
On taking the limit as ∆y → 0, the left-hand side of the above equation is the definition
of the first derivative. The right-hand side may be written in a compact and convenient way
by introducing the modified pressure P , which is the sum of the pressure and gravitational
terms. The general definition of the modified pressure isP = p + 3c1gh, where h is the height
(in the direction opposed to gravity i.e. −z) above some arbitrary preselected datum plane.
The advantages of using the modified pressure P are that:the fluid may flow as a result of a
pressure difference, gravity or both.
Here h is negative since the z- axis points downward, giving h = −z and therefore P =
p − ρgz. Thus, P0 = p(0) at z = 0 and PL = p(L) − ρgL at z = L giving p0 − pL + 3c1gL =
P0 − PL ≡ ∆P . Thus, equation (18.8) yields:
d ∆P
(τyz ) = . (18.9)
dy L
The above differential equation can be integrated to give the following expression for the
shear stress distribution:
∆P
τyz = y + C1 . (18.10)
L
The constant C1 is determined from the boundary conditions.
Since equations (18.9) and (18.10) have been derived without making any assumption about
the type of fluid, they are applicable to both Newtonian and non-Newtonian fluids.
Now integrating (18.5) once we find:
∂u
p0 y + C1 = −η
∂y
But p0 = − ∂x
∂p
= ∆P
L
the (constant) rate of change of pressure, therefore:
∂u
τyz = −η
∂y
which is the condition for a Newtonian fluid or Newton’s viscosity law.

Thus we have proved that the Navier Stokes equation for the case of laminar steady state
flow coincides with Newton’s viscosity law of flow of a fluid.
Let us discuss some examples.
Example 18.0.1 An incompressible fluid of density ρ and viscosity µ flows in a plane parallel
shear flow (or laminar flow) in a plane narrow slit of length L and width W formed by two
parallel walls which are at a distance 2B apart. It is assumed that B ≪ W ≪ L so that end
effects may be neglected. The fluid flows under the influence of a pressure difference ∆p and
the effect of gravity.
a. Determine the velocity profile of the flow for steady-state shear stress
b. Determine the maximum velocity, the average velocity and the mass flow rate for slit
flow.
a. 1rst Solution
Using first principles.
Consider the Figure ... and take the momentum balance along the z− axis. Then we have
(as has been shown above) the shear stress:
∆P
τyz = y + C1 .
L
where Newton’s viscosity law gives:
∂u
τyz = −η
∂y
from which we compute:
∆P 2 C1
vz = − y − y + C2
2ηL η
where the constants C1 , C2 are computed from the boundary conditions. These conditions are
the no-slip boundary conditions, that is the fluid at the two fixed walls has speed zero:
BC1 : At x = B the vz (B) = 0

BC2 : At x = −B the vz (−B) = 0.
∆P B 2
Using these conditions we calculate C1 = 0, C2 = 2ηL
which results in:
( )
∆P B 2 y2
vz = 1− 2
2ηL B
It follows that the velocity distribution for laminar, incompressible flow of a Newtonian fluid
in a plane narrow slit is parabolic.
2nd Solution
Using the Navier Stokes equation.
We select the coordinate axes so that the planes of the laminar flow are parallel to the y − z
plane and the direction of the velocity of the fluid is parallel to the z− axis (see Figure ...). Then
the velocity of the fluid is u = (u(y, z, t), 0, 0) and satisfies the condition of incompressibility
143
∇u = 0 trivially. Because the flow is under steady - state stress the velocity is independent of
time and furthermore it doers not depend on the z−coordinate. These imply u = (u(y), 0, 0).
The density of the external force on the fluid is g and the pressure in the interior of the fluid
is p(x, y, z) .
The equation of motion is the Navier Stokes equation:
[ ]
∂u η 2 1
+ (u · ∇)u = f + ▽ u − ▽ ·(▽ · u) − ▽ p
2
∂t ρ 3 ρ
which becomes:
1
0 = f +ν ▽2 u − ▽p
ρ
where ν is the kinematic viscosity.3 The components of the various vectors are:
u = (u(y), 0, 0), f = (g, 0, 0), ▽p = (p,x , p,y , p,z ).
The Navier Stokes equation gives:

1
0 = (g, 0, 0) + ν (u,yy , 0, 0) − (p,x , p,y , p,z ) ⇒
ρ
1
u,yy = p,x +g (18.11)
η
p,y = p,z = 0.
From (18.11) follows that the term p,x is a constant C = − ∆p

L
, (∆p > 0) hence:
∆p
p(y) = − y
L
∆p
u,yy = − − g = −K.
ηL
The solution of this equation is:
1
u(y) = − Ky 2 + C1 y + C2
2
where C1 , C2 are constants. To determine the constants we assume the no slip conditions at
the fixed plates bounding the motion. These conditions are:
u(y = B) = u(y = −B) = 0.
We find:
1
KB 2 + C1 B + C2 = 0
2
1
KB 2 − C1 B + C2 = 0
2
For laminar flow the term (u · ∇)u = 0 as it is easy to prove. This is easy to understand becasue for lamilar
3
flow the acceleration vanishes which means that Dudt = 0 which is teh lhs of the Navier Stokes equation.
The solution of this system of simultaneous equations is:

1
C1 = 0, C2 = KB 2
2
which implies: ( )
∆P B 2 y2
u(y) = 1− 2 −B ≤y ≤B
2ηL B
dvz 2
b. The maximum velocity occurs at x = 0 (where dx
= 0, ddxv2z < 0). Therefore,
∆P B 2
vz max = vz |x=0 = (18.12)
2µL
The average velocity is obtained by dividing the volumetric flow rate by the cross-sectional
area, ∫B ∫ B
v W dx
−B z 1 ∆P B 2 2
vz avg = ∫ B = vz dx = = vz max (18.13)
W dx 2B −B 3µL 3
−B
Thus, the ratio of the average velocity to the maximum velocity for Newtonian fluid flow in a
narrow slit is 23 .
The mass rate of flow is obtained by integrating the velocity profile over the cross section
of the slit as follows. ∫ B
w= ρvz W dx = 2ρW Bvz avg . (18.14)
−B
Substituting vz avg from (18.13) we have the final expression for the mass rate of flow.
2∆P B 3 ρW
w= (18.15)
3µL
Example 18.0.2 Differential Shell Momentum Balance in Cylindrical Coordinates

A shell momentum balance is used below to derive a general differential equation that can be
then employed to solve several fluid flow problems in cylindrical coordinates. For this purpose,
consider an incompressible fluid in laminar flow under the effects of both pressure and gravity in
a system of length L, which is at an angle β to the vertical. End effects are neglected assuming
the dimension of the system in the radial direction is relatively very small compared to that in
the axial direction (L)
Figure to be placed here. Use Corel to get the figure from file.
Since the fluid flow is in the axial direction, vr = 0, vθ = 0, and only vz exists. For small
flow rates, the viscous forces prevent continual acceleration of the fluid. So, vz is independent
of z and it is meaningful to postulate that velocity vz = vz (r) and pressure p = p(z). The only
non-vanishing components of the stress tensor are τrz = τzr , which depend only on r.
Consider now a thin cylindrical shell perpendicular to the radial direction extending a
distance L in the z-direction. A ’rate of z- momentum’ balance over this thin shell of thickness
∆r in the fluid is of the form:
145
Rate of z- momentum: In - Out + Generation = Accumulation
At steady-state, the accumulation term is zero. Momentum can go ’in’ and ’out’ of the
shell by both the convective and molecular mechanisms. Since vz (r) is the same at both ends
of the system, the convective terms cancel out because (ρvz vz 2πr∆r)|z=0 =(ρvz vz 2πr∆r)|z=L .
Only the molecular term (2πrLτrz ) remains to be considered, whose ’in’ and ’out’ directions are
taken in the positive direction of the r-axis. Generation of z- momentum occurs by the pressure
force acting on the surface [p2πr∆r] and gravity force acting on the volume [(ρg cos β)2πr∆rL].
The different contributions may be listed as follows:
• rate of z- momentum in by viscous transfer across cylindrical surface at r is (2πrLτrz )|r

• rate of z- momentum out by viscous transfer across cylindrical surface at r + ∆r is
(2πrLτrz )|r+∆r
• rate of z- momentum in by overall bulk fluid motion across annular surface at z = 0
is (ρvz vz 2πr∆r)|z=0
• rate of z- momentum out by overall bulk fluid motion across annular surface at z = L
is (ρvz vz 2πr∆r)|z=L
• pressure force acting on annular surface at z = 0 is p0 2πr∆r
• pressure force acting on annular surface at z = L is −pL2πr∆r
• gravity force acting in z−direction on volume of cylindrical shell is (ρg cos 3b2)2πr∆rL
On substituting these contributions into the z-momentum balance, we get:

(2πrLτrz )|r − (2πrLτrz )|r+∆r + (p0 − pL )23c0r394r + (ρg cos 3b2)23c0r394rL = 0 (18.16)
Dividing the equation by 2πL394r yields:
( )
(rτrz )|r+∆r − (rτrz )|r p0 − pL + ρgL cos 3b2
= (18.17)
∆r L
On taking the limit as ∆r → 0, the left-hand side of the above equation is the definition of
the first derivative. The right-hand side may be written in a compact and convenient way by
introducing the modified pressure P , which is the sum of the pressure and gravitational terms.
The general definition of the modified pressure is P = p + 3c1gh, where h is the height (in the
direction opposed to gravity) above some arbitrary preselected datum plane. The advantages
of using the modified pressure P are that: (i) The components of the gravity vector g need not
be calculated in cylindrical coordinates;
(ii) The solution holds for any flow orientation; and
(iii) The fluid may flow as a result of a pressure difference, gravity or both.
Here,h is negative since the z-axis points downward, giving h = −z cos β and therefore
P = p − ρgz cos β. Thus, P0 = p0 atz = 0 and PL = pL − 3c1gL cos β at z = L giving
p0 − pL + 3c1gL cos β = P0 − PL ≡ ∆P . Thus, equation (18.17) yields:
d ∆P
(rτrz) = r. (18.18)
dr L
The above differential equation can be integrated to give the general expression for the
momentum flux (or shear stress) distribution in cylindrical coordinate systems as
∆P C1
τrz = r+ . (18.19)
L r
For Newtonian fluids, Newton’s law of viscosity may be substituted for τrz in equation
(18.19) to obtain:
dvz ∆P C1
−µ = r+ (18.20)
dr L r
Integrating the above differential equation one obtains the following general expression for
the velocity profile for Newtonian fluids:
∆P 2 C1
vz = − r − ln r + C2 (18.21)
4µL µ
where C1 , C2 are integration constants, which are determined using appropriate boundary con-
ditions based on the flow problem.
Since equations (18.18) and (18.19) have been derived without making any assumption
about the type of fluid, they are applicable to both Newtonian and non-Newtonian fluids. On
the other hand, equation (18.21) is applicable to only Newtonian fluids. Some of the axial flow
problems in cylindrical coordinates where these equations may be used as starting points are
given below.
Example 18.0.3 Steady, laminar flow occurs in the space between two fixed parallel, circular
disks separated by a small gap 2b. The fluid flows radially outward owing to a pressure difference
(P 1 − P 2) between the inner and outer radii r1 and r2 respectively. Neglect end effects and
consider the region r1 ≤ r ≤ r2 only. Such a flow occurs when a lubricant flows in certain
lubrication systems.
Figure 2.TO BE PLACED HERE

a) Simplify the equation of continuity to show that rvr = f , where f is a function of z only
b) Simplify the equation of motion for incompressible flow of a Newtonian fluid of viscosity
µ and density ρ.
c) Obtain the velocity profile assuming creeping flow.
d) Determine an expression for the mass flow rate by integrating the velocity profile.
Solution 2
a) Since the steady laminar flow is directed radially outward, the velocity vector is v = (vr , 0, 0) .
Because the flow is incompressible it must hold:
∇v = 0 (18.22)
This equation in cylindrical coordinates gives:
∂ 1 ∂vθ ∂vz
(rvr ) + + =0⇒
∂r r ∂θ ∂z
∂
(rvr ) = 0 ⇒
∂r
rvr = f (θ, z)
147
Since the solution is expected to be symmetric about the z-axis there is no depended on the angle
θ (equivalently the f (θ, z),θ = 0), hence:
rvr = f (z) . (18.23)
b) For a Newtonian flow the Navier-Stokes equation is:

Dv
ρ = −∇P + µ∇2 v (18.24)
Dt
The cylindrical components of this equation are:
( )
∂vr ∂P ∂ 2 vr
r : ρ vr =− +µ 2 (18.25)
∂r ∂r ∂z
∂P
θ:0= (18.26)
∂θ
∂P
z:0= (18.27)
∂z
From (18.26) and (18.27) it is easy to prove that P = P (r). Also substituting vr from (18.23)
into (18.25) we find:
f 2 (z) dP (r) µ d2 f (z)
−ρ 3 = − + (18.28)
r dr r dz 2
These is the equation of motion for an incompressible, steady laminar flow of a viscous fluid of
viscosity µ and density ρ.
c) The equation of motion (18.28) is a non linear differential equation. To find a solution we
2
neglect the term ρ f r(z)
3 have to neglected and write:
dP (r) d2 f (z)
r =µ (18.29)
dr dz 2
The lhs of (18.29) is a function of r, and the rhs is a function of z. Therefore each term must
be a constant (a0 ∈ R) :
dP (r) d2 f (z)
r =µ = a0 .
dr dz 2
This results in two two differential equations:
dP (r)
= a0 (18.30)
dr
d2 f (z)
µ = a0 . (18.31)
dz 2
Integrating the first and substituting a0 in the second we find:
P (r2 ) − P (r1 ) 2
f (z) = ( ) z + a1 z + a2 (18.32)
2 ln rr21
where a1 , a2 are integration constants. We take as boundary conditions the non-slip conditions,
that is:
vr (r, z = +b) = 0
vr (r, z = −b) = 0
which when used in (18.32) give:

( )
P (r2 ) − P (r1 ) 2 z2
f (z) = ( ) b 1− 2 (18.33)
2 ln rr21 b
We conclude that assuming creeping flow4 , the velocity profile is:

( )
P (r2 ) − P (r1 ) 2 z2
vr (r, z) = ( ) b 1− 2 (18.34)
2r ln rr21 b
Furthermore the expression for the pressure P (r) is given from (18.30) and is:
( )
ln rr2
P (r) = P (r2 ) + (P (r1 ) − P (r2 )) ( )
ln rr12
where a0 is computed when one substitutes f (z) in (18.31).
d) To determine mass flow rate w we consider the surface area dS with normal vector n =
(1, 0, 0) (see Figure....) and have:
∫ b
w= ρ (n, v) dS
−b
∫ b
= ρvdS
−b
∫ b
= ρvr (dS) 2πrdz
−b
Replacing vr we find after integration:
P (r2 ) − P (r1 ) 3
w = 4π ( ) bρ (18.35)
3 ln rr21
4
What is this flow??
Index
1PGT, 56 elastic continuum, 27

electric permeability, 93
Navier Stokes equation, 91 empty space, 22
angle of shear, 24 energy density current, 108
Archimedes’ Law, 92 energy flux vector, 86
energy momentum tensor, 108
bivector metric, 111 Euler approach, 49
body forces, 77 Euler relation
bulk modulus, 25 fundamental relation, 84
bulk stress, 25 Euler’s equation of motion, 78
extended strain tensor, 17
caloric equation of state, 85
extensive property, 84
characteristic equation, 43
chemical potential, 83 flow
circulation, 67 irrotational
circulation Theorem, 86 steady, 69
coefficient of viscosity, 90 flow line, 54
comoving flow of a vector field, 1
observers, 106 flow of the fluid, 50
volume, 106 fluid, 8, 49
compressibility , 25 flux, 67
connecting vector, 2
connecting vector field, 2 homogeneous fluid, 69
conservation of mass, 62 homogeneous quantity, 66
continuum, 1 Hooke’s law, 27
Covariant Principle, 1
current imperfect fluid, 69, 89
energy density, 108 incompressible flow, 61
particle number, 107 infinitesimal generator, 56
integral curve, 53
deformable continua, 8 intensive property, 84
dragging along, 56, 58 invariants, 43
dragging along a vector field, 53 irrotational flow, 69
dual transformation of electromagnetic field, isentropic flow, 85
99 isotropic pressure, 21
dust, 69
Kinematic coefficient of viscosity, 91
eigenvalue, 43
elastic coefficients, 27 Lagrange approach, 49
149
150 INDEX
Lagrange system, 70 small strains, 17

law of conservation of mass, 62 solid, 8
linear continuum, 5, 7 specific internal energy, 85
linear elastic continuum, 27 specific variables, 84
linear elastic isotropic continuum, 27 steady flow, 70
lines of flow steady flow
pathlines, 70 stationary flow, 69
local flow, 53 strain, 8
local left action, 55 strain ellipsoid, 11
longitudinal stress, 23 strain tensor, 11, 17
stream filament, 70
magnetic permeability, 93 stream surface, 70
mass flux, 67 streamlines, 70
material derivative stress tensor, 23
total derivative, 51 stress tensor of the electromagnetic fluid, 93
matter density, 50 stress vector, 20
Maxwell stress tensor, 95 surface forces, 20
modulus of rigidity, 25
momentum flux, 67 thermodynamic variables, 83
total derivative
Newtonian viscous fluid, 91 material derivative, 51
normal stress, 21 translation, 8
transport theorem, 63
one parameter group of transformations, 55
unsteady flow, 69, 70
particle number current, 107
particle number density, 85, 106 velocity potential, 69
pathlines viscosity, 90
lines of flow, 70 viscous fluids, 90
perfect deformable continuum, 21 viscous stress tensor, 90
perfect fluid, 69 volume flux, 67
perfectly elastic material, 25 volume stress, 25
positive definite tensor, 44 vortex lines, 87
Poynting vector, 94 vortex sheets, 87
Poynting’s Theorem, 96
preservation Young modulus, 24
particle number , 107
principal value, 43
rigid bodies, 8
rigid motion, 8
rotation, 8
shear modulus, 25
shear strain, 24
shear stress, 21, 24
similarity transformation, 44

Lecture Notes On Fluids May 2013 PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Lecture Notes On Fluids May 2013 PDF

Caricato da

Copyright:

Formati disponibili

LECTURE NOTES ON NEWTONIAN AND

Athens May 2013

1 The physical system continuum 1

2 A short detour in vector analysis 3

3 Motion of a continuum and geometry 13

4 The stress tensor 19

5 The stress - strain relation for elastic continua 27

6 The dynamics of strain motion 35

7 Worked examples on the stress tensor 43

9 Point Transformations deﬁned by vector ﬁelds 53

10 Characterization of ﬂuids and ﬂows 69

11 The equation of continuity - conservation of mass (again) 73

12 The equation of motion of a perfect ﬂuid 77

13 Thermodynamics and hydrodynamics 83

14 The Navier Stokes equation 89

15 The electromagnetic ﬁeld as a viscous ﬂuid 93

16 Relativistic ﬂuids 105

17 Classical Diﬀerential Operators 135

18 Problems on Newtonian ﬂuids 139

The physical system continuum

We consider an aggregate of particles which we assume that occupies a connected region in

A short detour in vector analysis

For u = r and any vector u we have the identity:

For later use we note the obvious result:

▽(u · v) = u × (▽ × v) + v × (▽ × u) + (u · ▽)v + (v · ▽)u (2.4)

and take v = r to ﬁnd:

▽(u · r) = u × (▽ × r) + r × (▽ × u) + (u · ▽)r + (r · ▽)u. (2.5)

▽v2 = 2v × (▽ × v) + 2(v · ▽)v (2.7)

or, solving for the directional derivative (v · ▽)v:

2.0.1 Tensor formulation

(u × v)ρ = ερµν uµ vν (2.10)

The directional derivative is the vector (xµ ∂µ )f ρ .

(δ µν uµ vν ),κ = δ µν (uµ ,κ vν + uµ vν,κ )

2.1 The kinematics of a Newtonian continuum

(xµ ∂µ )(xν ∂ν )v ρ = (xµ δµν ∂ν )v ρ + xµ xν ∂µ ∂ν v ρ = (xν ∂ν )v ρ + O(xµ xν )

It is easy to see that ωy = 12 (▽ × v) · j, ωz = 12 (▽ × v) · k. If we deﬁne the quantities:

2.1.2 The relative velocity in tensor notation

dvPρ = vPρ,ν dxν (2.31)

ω ρν = εκσλ εκρν vPσ,λ (2.32)

Next we deﬁne the symmetric (0,2) tensor:

and obtain the ﬁnal answer:

2.1.3 The kinematic interpretation of relative velocity

Figure 2.1: The geometric description of a small displacement

In other words it is the type of motion sustained by an aggregate of particles (a continuum)

2.1.4 The kinematic interpretation of (2.29) in terms of space inter-

Figure 2.2: The geometric description of a small displacement

Replacing vQ from (2.29) we ﬁnd10 :

where xi = (x, y, z) and we have introduced the strain tensor:

The motion of a deformable body in

3.1 The geometrization of rigid motion

We compute the change in the magnitude of the displacement vector r. We have4 :

Rigid motion is generated by a Euclidian isometry, that is by the Killing vectors of

Is there a metric (that is non-degenerate second order symmetric tensor) whose

3.1.1 The geometrization of strain motion

3.1.2 A geometric computation of the strain metric

r′ = dr + s′ (x + dr) − s(x). (3.2)

The deformation is continuous which means that:

We deﬁne the extended strain tensor eµν with the formula:

Then the deformation is:

D(dr) = 2eµν dxµ dxν (3.5)

The stress tensor

We shall consider orthonormal frames and shall follow the convention:

If A is an orthogonal matrix (i.e. AAt = 1 where t=transpose) relating two Cartesian