2

The equations of fluid motion

In this chapter we briefly review the Navier–Stokes equations which gov-

ern the flow of constant-property Newtonian fluids. More comprehensive
accounts can be found in the texts of Batchelor (1967), Panton (1984), and
Tritton (1988). Two topics that are important in the study of turbulent flows,
that are not extensively discussed in these texts, are the Poisson equation
for pressure (Section 2.5), and the transformation properties of the Navier–
Stokes equations (Section 2.9). The equations of fluid motion are expressed
either in vector notation or in Cartesian tensor notation, which is reviewed
in Appendix A.

2.1 Continuum fluid properties

The idea of treating fluids as continuous media is both natural and familiar.
It is, however, worthwhile to review the continuum hypothesis – that reconciles
the discrete molecular nature of fluids with the continuum view – so as to
avoid confusion when quantities such as ‘fluid particles’ and ‘infinitesimal
material elements’ are introduced.
The length and time scales of molecular motion are extremely small
compared with human scales. Taking air under atmospheric conditions as
an example, the average spacing between molecules is 3 × 10−9 m, the mean
free path, λ, is 6 × 10−8 m, and the mean time between successive collision
of a molecule is 10−10 s. In comparison, the smallest geometric length scale
in a flow,
, is seldom less than 0.1 mm = 10−4 m, which, for flow velocities
up to 100 m s−1 , yields a flow timescale larger than 10−6 s. Thus, even for
this example of a flow with small length and time scales, these flow scales
exceed the molecular scales by three or more orders of magnitude.

10
 2.1 Continuum fluid properties 11

The separation of the length scales is quantified by the Knudsen number

Kn ≡ λ/
. (2.1)
In the above example, Kn is less than 10−3 , while in general the continuum
approach is appropriate for Kn  1.
For very small Kn, because of the separation of scales, there exist interme-
diate length scales
∗ , such that
∗ is large compared with molecular scales,
yet small compared with flow scales (i.e., λ 
∗ 
). Roughly speaking,
the continuum fluid properties can be thought of as the molecular properties
averaged over a volume of size V =
∗3 . Let Vx denote a spherical region of
volume V centered on the point x. Then, at time t, the fluid’s density ρ(x, t)
is the mass of molecules in Vx , divided by V .
Similarly the fluid’s velocity U (x, t) is the average velocity of the molecules
within Vx . Because of the separation of scales, the dependence of the contin-
uum properties on the choice of
∗ is negligible.
(While the approach presented in the previous paragraph is standard (see,
e.g., Batchelor (1967) and Panton (1984)), as Exercise 2.1 illustrates, more
care is needed to provide a proper definition of the continuum properties in
terms of averaging over a scale
∗ . In fact, continuum fields are best defined
as expectations of molecular properties, see Chapter 12.)
It is important to appreciate that, once we invoke the continuum hypothe-
sis to obtain continuous fields, such as ρ(x, t) and U (x, t), we can leave behind
all notions of the discrete molecular nature of the fluid, and molecular scales
cease to be relevant. We can talk meaningfully of ‘the density at x, t,’ even
though (in the microscopic view) in all likelihood there is no matter at (x, t).
Similarly, we can consider differences in properties over distances smaller
than molecular scales: indeed we do so when we define gradients,
 
∂ρ 1
≡ lim [ρ(x1 + h, x2 , x3 , t) − ρ(x1 , x2 , x3 , t)] . (2.2)
∂x1 h→0 h

EXERCISE
2.1 In the flow of an ideal gas, let m(i) , x(i) (t) and u(i) (t) be the mass,
position, and velocity of the ith molecule. As a generalization of the
standard continuum hypothesis, consider the definition
 (i) (i)
m u K(|r (i) |)
U (x, t) ≡ i (j) , (2.3)
j m K(|r |)
(j)

where r (i) ≡ x(i) − x, and K(r) is a smooth kernel such as

K(r) = exp(− 12 r2 /
∗2 ), (2.4)
12 2 The equations of fluid motion

with
∗ being a specified length scale. Show that the velocity gradients
are
 (i) 
∂Uk m (Uk − u(i)
k )K (|r |)r
/|r |
(i) (i) (i)
= i  (j) , (2.5)
j m K(|r |)
∂x
(j)

where K  (r) = dK(r)/dr.

(Evidently, the continuum field defined by Eq. (2.3) inherits the
mathematical continuity properties of the kernel. If, as in the standard
treatment, K(r) is piecewise constant, i.e.,

1, if r ≤
∗ ,
K(r) = (2.6)
0, if r >
∗ ,

then U (x, t) is piecewise constant, and hence not continuously differ-

entiable.)

2.2 Eulerian and Lagrangian fields

The continuum density and velocity fields, ρ(x, t) and U (x, t), are Eulerian
fields in that they are indexed by the position x in an inertial frame. The
starting point for the alternative Lagrangian description is the definition of
a fluid particle – which is a continuum concept. By definition, a fluid particle
is a point that moves with the local fluid velocity: X + (t, Y ) denotes the
position at time t of the fluid particle that is located at Y at the specified
fixed reference time t0 , see Fig. 2.1. Mathematically, the fluid particle position

X+(t1, Y )

t0 t1 t

Fig. 2.1. A sketch of the trajectory X + (t, Y ) of a fluid particle in x–t space, showing
its position Y at the reference time t0 , and at a later time t1 .
 2.2 Eulerian and Lagrangian fields 13

X + (t, Y ) is defined by two equations. First, the position at the reference time
t0 is defined to be
X + (t0 , Y ) = Y . (2.7)

Second, the equation

∂ +
X (t, Y ) = U (X + (t, Y ), t), (2.8)
∂t
expresses the fact that the fluid particle moves with the local fluid velocity.
Given the Eulerian velocity field U (x, t), then, for any Y , Eq. (2.8) can be
integrated backward and forward in time to obtain X + (t, Y ) for all t.
Lagrangian fields of density and velocity, for example, are defined in terms
of their Eulerian counterparts by

ρ+ (t, Y ) ≡ ρ(X + (t, Y ), t), (2.9)

U + (t, Y ) ≡ U (X + (t, Y ), t). (2.10)

Note that the Lagrangian fields ρ+ and U + are indexed not by the current
position of the fluid particle, but by its position Y at the reference time t0 .
Hence, Y is called the Lagrangian coordinate or the material coordinate.
For fixed Y , X + (t, Y ) defines a trajectory (in x–t space) that is the fluid-
particle path, and similarly ρ+ (t, Y ) is the fluid-particle density. The partial
derivative ∂ρ+ (t, Y )/∂t is the rate of change of density at fixed Y , i.e.,
following a fluid particle. From Eq. (2.9) we obtain
∂ + ∂
ρ (t, Y ) = ρ(X + (t, Y ), t)
∂t ∂t
   
∂ ∂ + ∂
= ρ(x, t) + Xi (t, Y ) ρ(x, t)
∂t x=X + (t,Y ) ∂t ∂xi x=X + (t,Y )
 
∂ ∂
= ρ(x, t) + Ui (x, t) ρ(x, t)
∂t ∂xi x=X + (t,Y )
 
D
= ρ(x, t) , (2.11)
Dt x=X + (t,Y )

where the material derivative, or substantial derivative, is defined by

D ∂ ∂ ∂
≡ + Ui = + U · ∇. (2.12)
Dt ∂t ∂xi ∂t
Thus the rate of change of density following a fluid particle is given by the
partial derivative of the Lagrangian field (i.e., ∂ρ+ /∂t) and by the substantial
derivative of the Eulerian field (i.e., Dρ/Dt).
14 2 The equations of fluid motion

Similarly, for fixed Y , U + (t, Y ) is the fluid particle velocity, and

 
∂ + D
U (t, Y ) = U (x, t) (2.13)
∂t Dt x=X + (t,Y )

is the rate of change of fluid particle velocity, i.e., the fluid particle accelera-
tion.
A fluid particle is also called a material point and we have seen that it
is defined by its position Y at time t0 and by its movement with the local
fluid velocity (Eq. (2.8)). Material lines, surfaces, and volumes are defined
similarly. For example, consider at time t0 a simple closed surface S0 that
encloses the volume V0 . The corresponding material surface S (t) is defined
to be coincident with S0 at time t0 , and by the property that every point of
S (t) moves with the local fluid velocity. Thus S (t) is composed of the fluid
particles X + (t, Y ), which at t0 compose the surface S0 :
S (t) ≡ {X + (t, Y ): Y ∈ S0 }. (2.14)
Because a material surface moves with the fluid, the relative velocity between
the surface and the fluid is zero. Consequently a fluid particle cannot cross
a material surface; neither is there a mass flux across a material surface.

EXERCISE
2.2 Consider two fluid particles that, at the reference time t0 are located
at Y and Y + dY , where dY is an infinitesimal displacement. At time
t, the line between the two particles forms the infinitesimal line element
s(t) ≡ X + (t, Y + dY ) − X + (t, Y ). (2.15)
Show that s(t) evolves by
ds
= s · (∇U )x=X + (t,Y ) . (2.16)
dt
(Hint: expand U + (t, Y +dY ) = U (X + (t, Y )+s(t), t) in a Taylor series.
Since s is infinitesimal, only the leading-order terms need be retained.)

2.3 The continuity equation

The mass-conservation or continuity equation is
∂ρ
+ ∇ · (ρU ) = 0. (2.17)
∂t
The derivation and interpretation of this equation in terms of control volumes
and material volumes should be familiar to the reader and are not repeated
 2.3 The continuity equation 15

here. A further useful interpretation is in terms of the specific volume of the

fluid ϑ(x, t) = 1/ρ(x, t). Manipulation of Eq. (2.17) yields
D ln ϑ
= ∇ · U. (2.18)
Dt
The left-hand side is the logarithmic rate of increase of the specific volume,
while (as Exercises 2.3 and 2.4 show) the dilatation ∇·U gives the logarithmic
rate of increase of the volume of an infinitesimal material volume. Hence
the continuity equation can be viewed as a consistency condition between
the change of the specific volume following a fluid particle, and the change
in the volume of an infinitesimal material volume element.
In this book we consider constant-density flows (i.e., flows in which ρ is
independent both of x and of t). In this case the evolution equation Eq. (2.17)
degenerates to the kinematic condition that the velocity field be solenoidal
or divergence-free:
∇ · U = 0. (2.19)

EXERCISES
2.3 Let V (t) be a material volume bounded by the material surface S (t).
Show from geometry that the volume of fluid V (t) within V (t) evolves
by
dV (t)
= U · n dA, (2.20)
dt S(t)

where dA is an area element on S (t), and n is the outward pointing

normal. Use the divergence theorem to obtain
dV (t)
= ∇ · U dx. (2.21)
dt V(t)

Show that, for the infinitesimal volume dV (t) of an infinitesimal

material volume dV ,
d
ln dV (t) = ∇ · U . (2.22)
dt
2.4 The determinant of the Jacobian
 + 
∂Xi (t, Y )
J(t, Y ) ≡ det (2.23)
∂Yj
gives the volume ratio between an infinitesimal material volume
dV (t) at time t, and its volume dV (t0 ) at time t0 . To first order in
the infinitesimal dt, show that
Xi+ (t0 + dt, Y ) = Yi + Ui (Y , t0 ) dt, (2.24)
16 2 The equations of fluid motion

 
∂Xi+ (t0 + dt, Y ) ∂Ui
= δij + dt, (2.25)
∂Yj ∂xj Y ,t0

J(t0 + dt, Y ) = 1 + (∇ · U )Y ,t0 dt. (2.26)

Hence show that
 
∂
ln J(t, Y ) = (∇ · U )Y ,t0 . (2.27)
∂t t=t0

2.5 The volume V (t) defined in Exercise 2.3 can be written

V (t) = dx = J(t, Y ) dY . (2.28)

V(t) V(t0 )

Differentiate the first and last expressions in this equation with

respect to time, and compare the result with Eq. (2.21) to obtain
∂
ln J(t, Y ) = (∇ · U )X + (t,Y ) . (2.29)
∂t
Hence argue that, in constant-density flows, J(t, Y ) is unity.

2.4 The momentum equation

The momentum equation, based on Newton’s second law, relates the fluid
particle acceleration DU /Dt to the surface forces and body forces experi-
enced by the fluid. In general, the surface forces, which are of molecular
origin, are described by the stress tensor τij (x, t) – which is symmetric, i.e.,
τij = τji . The body force of interest is gravity. With Ψ being the gravitational
potential (i.e., the potential energy per unit mass associated with gravity), the
body force per unit mass is
g = −∇Ψ. (2.30)
(For a constant gravitational field the potential is Ψ = gz, where g is the
gravitational acceleration, and z is the vertical coordinate.) These forces
cause the fluid to accelerate according to the momentum equation
DUj ∂τij ∂Ψ
ρ = −ρ . (2.31)
Dt ∂xi ∂xj
We now specialize the momentum equation to flows of constant-property
Newtonian fluids – the fundamental class of flows considered in this book.
In this case, the stress tensor is
 
∂Ui ∂Uj
τij = −P δij + µ + , (2.32)
∂xj ∂xi
 2.4 The momentum equation 17

where P is the pressure, and µ is the (constant) coefficient of viscosity.

Recalling that (for the constant-density flows considered) the velocity field is
solenoidal (i.e., ∂Ui /∂xi = 0), we observe that Eq. (2.32) expresses the stress
as the sum of isotropic (−P δij ) and deviatoric contributions.
By substituting this expression for the stress tensor (Eq. (2.32)) into the
general momentum equation Eq. (2.31) (and exploiting the facts that ρ and
µ are uniform and that ∇ · U = 0), we obtain the Navier–Stokes equations
DUj ∂ 2 Uj ∂P ∂Ψ
ρ =µ − −ρ . (2.33)
Dt ∂xi ∂xi ∂xj ∂xj
Further, defining the modified pressure, p, by

p = P + ρΨ, (2.34)

this equation simplifies to

DU 1
= − ∇p + ν ∇2 U , (2.35)
Dt ρ
where ν ≡ µ/ρ is the kinematic viscosity. In summary: the flow of constant-
property Newtonian fluids is governed by the Navier–Stokes equations
Eq. (2.35) together with the solenoidal condition ∇ · U = 0 stemming from
mass conservation.
At a stationary solid wall with unit normal n, the boundary conditions
satisfied by the velocity are the impermeability condition

n · U = 0, (2.36)

and the no-slip condition

U − n (n · U ) = 0, (2.37)

(which together yield U = 0).

It is sometimes useful to consider the hypothetical case of an ideal (inviscid)
fluid, which is defined to have the isotropic stress tensor

τij = −P δij . (2.38)

The conservation of momentum is given by the Euler equations

DU 1
= − ∇p, (2.39)
Dt ρ
which follow from Eqs. (2.31), (2.34), and (2.38). Because the Euler equations
do not contain second spatial derivatives of velocity, they require different
18 2 The equations of fluid motion

boundary conditions than those of the Navier–Stokes equations. At a sta-

tionary solid wall, for example, only the impermeability condition can be
applied, and in general the tangential components of velocity are non-zero.
While it is preferable to obtain the Euler equations (and other equa-
tions derived from them) directly from the definition of τij being isotropic
(Eq. (2.38)), it may nevertheless be observed that the Euler equations can
be obtained from the Navier–Stokes equations by setting ν to zero. It is
important to appreciate, however, that ν = 0 is a singular limit: solutions to
the Navier–Stokes equations in the limit of vanishing viscosity (ν → 0) are
different than solutions to the Euler equations. For one thing, even in this
limit, the equations require different boundary conditions.

2.5 The role of pressure

The role of pressure in the (constant-density) Navier–Stokes equations re-
quires further comment. First we observe that isotropic stresses and conser-
vative body forces have the same effect, which is expressed by the modified
pressure gradient. Hence the body force has no effect on the velocity field
and on the modified pressure field. (This is, of course, in contrast to variable-
density flows, in which buoyancy forces can be important.) Henceforth, we
refer to p simply as ‘pressure’.
We may be accustomed to thinking of pressure as a thermodynamic
variable, related to density and temperature by an equation of state. However,
for constant-density flows, there is no connection between pressure and
density, and a different understanding of pressure is required.
To this end, we take the divergence of the Navier–Stokes equations
Eq. (2.35), without assuming the velocity field to be solenoidal, but instead
writing ∆ for the dilatation rate (i.e., ∆ = ∇ · U ). The result is
 
D
− ν ∇ ∆ = R,
2
(2.40)
Dt
where
1 ∂Ui ∂Uj
R ≡ − ∇2 p − . (2.41)
ρ ∂xj ∂xi
Consider the solution to Eq. (2.40) with initial and boundary conditions
∆ = 0. The solution is ∆ = 0 if, and only if, R is zero everywhere, which in
turn implies (from Eq. (2.41)) that p satisfies the Poisson equation
∂Ui ∂Uj
∇2 p = S ≡ −ρ . (2.42)
∂xj ∂xi
 2.5 The role of pressure 19

Thus, we conclude that the satisfaction of this Poisson equation is a necessary

and sufficient condition for a solenoidal velocity field to remain solenoidal.
At a stationary, plane solid surface, the Navier–Stokes equations Eq. (2.35)
reduce to
∂p ∂ 2 Un
=µ , (2.43)
∂n ∂n2
where n is a coordinate in the wall-normal direction, and Un is the velocity
component normal to the wall. This equation provides a Neumann boundary
condition for the Poisson equation, Eq. (2.42). Given Neumann conditions of
this form, the Poisson equation Eq. (2.42) determines the pressure field p(x, t)
(to within a constant) in terms of the velocity field at the same instant of time.
Thus, ∇p is uniquely determined by the current velocity field, independent
of the flow’s history.
The solution to the Poisson equation Eq. (2.42) can be written explicitly
in terms of Green’s functions. Consider the Poisson equation
∇2 f(x) = S(x). (2.44)
in a domain V . The source S(x) can be written

S(x) = S(y)δ(x − y) dy, (2.45)

V

where y is a point in V , and δ(x − y) is the three-dimensional Dirac delta

function1 at y. A solution to the Poisson equation
∇2 g(x|y) = δ(x − y) (2.46)
is
−1
g(x|y) = . (2.47)
4π |x − y|
(As implied by the notation, the solution depends both on x and on the
location of the delta function, y.) When it is multiplied by S(y) and integrated
over V , Eq. (2.46) becomes ∇2 f = S (i.e., Eq. (2.44)), and hence Eq. (2.47)
becomes a solution:
−1 S(y)
f(x) = g(x|y)S (y) dy = dy. (2.48)
V 4π V |x − y|
The solution to the Poisson equation for pressure Eq. (2.42) is, therefore,
 
(h) ρ ∂Ui ∂Uj dy
p(x, t) = p (x, t) + , (2.49)
4π V ∂xj ∂xi y,t |x − y|
1 The properties of Dirac delta functions are reviewed in Appendix C.
20 2 The equations of fluid motion

where p(h) is a harmonic function (∇2 p(h) = 0) dependent on the boundary

conditions. (It is possible to express p(h) in terms of surface integrals over the
boundary of V , see e.g., Kellogg (1967).)

EXERCISES
2.6 Show that (away from the origin)
∂2
∇2 |x|−1 = (xj xj )−1/2 = 0. (2.50)
∂xi ∂xi

2.7 A simple numerical method for solving the Navier–Stokes equations

for constant-property flow advances the solution in small time steps
∆t, starting from the initial condition U (x, 0). On the nth step the
numerical solution is denoted by U (n) (x), which is an approximation
to U (x, n ∆t). Each time step consists of two sub-steps, the first of
which yields an intermediate result U
(n+1) (x) defined by

∂2 Uj(n) ∂Uj(n)

j(n+1) ≡ Uj(n) + ∆t ν
U − Uk(n) . (2.51)
∂xi ∂xi ∂xk

The second sub-step is

(n)

j(n+1) − ∆t ∂φ ,
Uj(n+1) = U (2.52)
∂xj

where φ(n) (x) is a scalar field.

(a) Comment on the connection between Eq. (2.51) and the
Navier–Stokes equations.
(b) Assuming that U (n) is divergence-free, obtain from Eq. (2.51)

(n+1) .
an expression (in terms of U (n) ) for the divergence of U
(c) Obtain from Eq. (2.52) an expression for the divergence of
U (n+1) .
(d) Hence show that the requirement ∇ · U (n+1) = 0 is satisfied if,
and only if, φ(n) (x) satisfies the Poisson equation
(n)
∂Uk(n) ∂Uj
∇2 φ(n) = − . (2.53)
∂xj ∂xk

(e) What is the connection between φ(n) (x) and the pressure?
 2.6 Conserved passive scalars 21

2.6 Conserved passive scalars

In addition to the velocity U (x, t), we consider a conserved passive scalar
denoted by φ(x, t). In a constant-property flow, the conservation equation
for φ is
Dφ
= Γ ∇2 φ, (2.54)
Dt
where Γ is the (constant and uniform) diffusivity. The scalar φ is conserved,
because there is no source or sink term in Eq. (2.54). It is passive because
(by assumption) its value has no effect on material properties (i.e., ρ, ν, and
Γ), and hence it has no effect on the flow.
The scalar φ can represent various physical properties. It can be a small
excess in temperature – sufficiently small that its effect on material properties
is negligible. In this case Γ is the thermal diffusivity, and the ratio ν/Γ is
the Prandtl number, Pr. Alternatively, φ can be the concentration of a trace
species, in which case Γ is the molecular diffusivity, and ν/Γ is the Schmidt
number, Sc.
An important property of the scalar is its boundedness. If the initial and
boundary values of φ lie within a given range
φmin ≤ φ ≤ φmax , (2.55)
then φ(x, t) for all (x, t) also lies in this range: values of φ greater than φmax
or less that φmin cannot occur.
To show this result we examine local maxima in the scalar field. Suppose
that there is a local maximum at x̄ at time t̄, and we choose a coordinate
system such that ∂2 φ/(∂xi ∂xj ) is in principal axes there. The mathematical
properties of a maximum imply that
(∇φ)x̄,t̄ = 0, (2.56)
and that the second derivatives ∂2 φ/∂x21 , ∂2 φ/∂x22 , and ∂2 φ/∂x23 are negative
or zero. Consequently, for their sum, the Laplacian, we have
(∇2 φ)x̄,t̄ ≤ 0. (2.57)
Then, from the conservation equation for φ (Eq. (2.54)), we obtain
   
Dφ ∂φ
= + V · ∇φ = Γ(∇2 φ)x̄,t̄ ≤ 0, (2.58)
Dt x̄,t̄ ∂t x̄,t̄

for every vector V ; showing that, following any trajectory from the local
maximum, the value of φ does not increase. Consequently, there is no way
in which φ can increase beyond the upper bound φmax imposed by the initial
22 2 The equations of fluid motion

and boundary conditions. Obviously, a similar argument applies to the lower

bound, φmin .

2.7 The vorticity equation

An essential feature of turbulent flows is that they are rotational: that is,
they have non-zero vorticity. The vorticity ω(x, t) is the curl of the velocity

ω = ∇ × U, (2.59)

and it equals twice the rate of rotation of the fluid at (x, t).
The equation for the evolution of the vorticity is obtained by taking the
curl of the Navier–Stokes equations Eq. (2.35):
Dω
= ν ∇2 ω + ω · ∇U . (2.60)
Dt
The pressure term (−∇ × ∇p/ρ) vanishes for constant-density flows.
The equation for the evolution of an infinitesimal line element of material
s(t) (see Eq. (2.16)) is
ds
= s · ∇U , (2.61)
dt
which, apart from the viscous term, is identical to the vorticity equation.
Hence, in inviscid flow, the vorticity vector behaves in the same way as
an infinitesimal material line element (Helmholtz theorem). If the strain
rate produced by the velocity gradients acts to stretch the material line
element aligned with ω, then the magnitude of ω increases correspondingly.
This is the phenomenon of vortex stretching, which is an important pro-
cess in turbulent flows, and ω · ∇U is referred to as the vortex-stretching
term.
For two-dimensional flows, the vortex-stretching term vanishes, and the
one non-zero component of vorticity evolves as a conserved scalar. Because of
the absence of vortex stretching, two-dimensional turbulence (which can oc-
cur in special circumstances) is qualitatively different than three-dimensional
turbulence.

EXERCISES
2.8 Use suffix notation to verify the relations:

∇ · ω = 0, (2.62)

∇ × ∇φ = 0, (2.63)
 2.8 Rates of strain and rotation 23

∇ × (∇ × U ) = ∇(∇ · U ) − ∇2 U , (2.64)

U × ω = 12 ∇(U · U ) − U · ∇U . (2.65)
Are the expressions in Eqs. (2.64) and (2.65) tensors?
2.9 Show that the Navier–Stokes equations (Eq. (2.35)) can be written
in the Stokes form
 
∂U p
− U × ω + ∇ 2U · U +
1
= ν ∇2 U . (2.66)
∂t ρ
Hence obtain Bernoulli’s theorem: for a steady, inviscid, constant-
density flow, the Bernoulli integral,
p
H ≡ 12 U · U + , (2.67)
ρ
is constant
(a) along streamlines,
(b) along vortex lines (i.e., lines parallel to ω), and
(c) everywhere in irrotational flow (ω = 0).
2.10 Show that the vorticity squared – or enstrophy – ω 2 = ω · ω evolves
by
Dω 2 ∂Ui ∂ωi ∂ωi
= ν ∇2 ω 2 + 2ωi ωj − 2ν . (2.68)
Dt ∂xj ∂xj ∂xj

2.8 Rates of strain and rotation

The velocity gradients ∂Ui /∂xj are the components of a second-order tensor,
the general properties of which are described in Appendix B. The decom-
position of ∂Ui /∂xj into isotropic, symmetric-deviatoric, and antisymmetric
parts is
∂Ui
= 13 ∆δij + Sij + Ωij , (2.69)
∂xj
where the dilatation ∆ = ∇ · U is zero for constant-density flow, Sij is the
symmetric, deviatoric rate-of-strain tensor
 
1 ∂Ui ∂Uj
Sij ≡ + , (2.70)
2 ∂xj ∂xi
and Ωij is the antisymmetric rate-of-rotation tensor
 
1 ∂Ui ∂Uj
Ωij ≡ − . (2.71)
2 ∂xj ∂xi
(For variable-density flow, Sij is defined as Sij ≡ 12 (∂Ui /∂xj +∂Uj /∂xi )− 13 ∆δij .)
24 2 The equations of fluid motion

It may be observed that the Newtonian stress law (Eq. (2.32)) can be
re-expressed as
τij = −P δij + 2µSij , (2.72)

showing that the viscous stress depends linearly on the rate of strain, inde-
pendent of the rate of rotation.
The vorticity and the rate of rotation are related by

ωi = −εijk Ωjk , (2.73)

Ωij = − 12 εijk ωk , (2.74)

where εijk is the alternating symbol. Thus Ωij and ωi contain the same
information, but (as discussed in Appendix A) Ωij is a tensor whereas ωi is
not.

EXERCISES
2.11 From Eq. (2.16), derive an equation for the evolution of the length of
an infinitesimal material line element. Show that the rate of growth
of the line depends linearly on the rate of strain, and is independent
of the rate of rotation.
2.12 Show that the vorticity equation (Eq. (2.60)) can alternatively be
written
Dωi ∂2 ωi
=ν + Sij ωj . (2.75)
Dt ∂xj ∂xj
Re-express the source in the Poisson equation for pressure (Eq. (2.42))
in terms of Sij and Ωij .
2.13 In a simple shear flow, all the velocity gradients are zero except for
∂U1 /∂x2 . For this case write down the components of Sij and Ωij (as
matrices) and of ω.

2.9 Transformation properties

By studying the behavior of the Navier–Stokes equations when they are
subjected to various transformations, we are able to deduce important prop-
erties of the fluid flows that they describe. The most important of these
properties are Reynolds number similarity, invariance under fixed rotations
and reflections of the coordinate axes, Galilean invariance, and the lack of
invariance under frame rotations.
 2.9 Transformation properties 25

x2 , x2
x2

x1 x1 , x1
(a) Reference (b) Change of scale

x2 x2 x2

x2

x1
x1
x
x1 x1
(c) Shift in space (d) Change in orientation

x2 x2 x2 V

x1 , x1

x1

x2 x1
(e) Reflection (f) Uniform motion

x2 x2 x2

x2

x1
x1
V(t) Ω

(g) Rectilinear acceleration x1 (h) Frame rotation x1

Fig. 2.2. Sketches of experiments used to study the transformation properties of

the Navier–Stokes equations: (a) reference experiment (referred to the E coordinate
system); (b)–(h) other experiments (referred to the Ē coordinate system).

Consider a particular fluid-mechanics experiment performed in a labora-

tory, and consider a second experiment, that is similar to the first, but differs
in some respect. For example: the second experiment could be performed
at a different time; the apparatus could be placed in a different location; it
could be orientated differently; it could be placed on a moving platform;
26 2 The equations of fluid motion

a different fluid could be used; or a second apparatus that is geometrically

similar to the first, but of a different scale, could be constructed. For each
of these differences we can ask whether the velocity fields in the two exper-
iments are similar. That is, are the velocity fields the same when they are
appropriately scaled and referred to appropriate coordinate systems? These
questions can be answered by studying the transformation properties (also
called invariance properties or symmetries) of the Navier–Stokes equations.
These are important considerations in the modelling of turbulent flows. A
model will be qualitatively incorrect unless its transformation properties are
consistent with those of the Navier–Stokes equations.
Figure 2.2(a) is a sketch of the apparatus considered in the first (reference)
experiment. The size of the apparatus is characterized by the length scale L,
and the initial and boundary conditions on the velocity are characterized by
the velocity scale U . The coordinate system (denoted by E, with orthonormal
basis vectors ei ) has its origin and axes fixed relative to the apparatus, which
is at rest in an inertial frame.
The length scale L and the velocity scale U are used to define the non-
dimensional independent variables
x̂ = x/L, t̂ = t U /L, (2.76)
and dependent variables
Û (x̂, t̂) = U (x, t)/U , p̂(x̂, t̂) = p(x, t)/(ρU 2 ). (2.77)
On applying these simple scaling transformations to the continuity equation
Eq. (2.19), the Navier–Stokes equations Eq. (2.35), and the Poisson equation
Eq. (2.42), we obtain
∂Ûi
= 0, (2.78)
∂x̂i

∂Ûj ∂Ûj 1 ∂2 Ûj ∂p̂

+ Ûi = − , (2.79)
∂t̂ ∂ x̂i Re ∂ x̂i ∂ x̂i ∂ x̂j

∂2 p̂ ∂Ûi ∂Ûj
=− , (2.80)
∂x̂i ∂x̂i ∂x̂j ∂x̂i
where the Reynolds number is
Re ≡ UL/ν. (2.81)
Evidently, the Reynolds number is the only parameter appearing in these
equations.
 2.9 Transformation properties 27

Reynolds-number similarity
The experiment shown in Fig. 2.2(b) has a different length scale Lb , velocity
scale Ub , and fluid properties, νb and ρb . If the scaled variables are defined in
an analogous way (x̂ = x/Lb , Û = U /Ub , etc.) then the boundary conditions
(expressed in terms of Û (x̂, t̂)) in two experiments are the same, and the
transformed Navier–Stokes equations are the same as Eqs. (2.78)–(2.80),
except that Re is replaced by

Reb ≡ Ub Lb /νb . (2.82)

Thus, if the Reynolds numbers are the same (Re = Reb ), then the scaled
velocity fields Û (x̂, t̂) are also the same, because they are governed by
identical equations with identical initial and boundary conditions. This is the
property of Reynolds-number similarity.
The scaled Euler equations are the same as Eq. (2.79), but with the
omission of the term in Re. The scaled velocity fields Û (x̂, t̂) given by the
Euler equations are therefore the same, irrespective of Lb , Ub , and ρb : they
exhibit scale similarity and the Euler equations are said to be invariant with
respect to scale transformations.

Time and space invariance

The simplest invariance properties of the Navier–Stokes equations are their
invariances with respect to shifts in time and space. As depicted in Fig. 2.2(c)
we consider the second experiment performed a time T later than the
reference experiment, with the apparatus translated by an amount X . The
velocity field in the second experiment is referred to the Ē coordinate system
shown in the Fig. 2.2(c), which has orthonormal basis vectors ēi . With the
scaled independent variables defined by

x̂ = x̄/L = (x − X )/L, (2.83)

t̂ = (t − T ) U /L, (2.84)

it is trivial to show that the transformed Navier–Stokes equations are iden-

tical to Eqs. (2.78)–(2.80).

Rotational and reflectional invariance

Figure 2.2(d) shows the apparatus with a different orientation than that in
the reference experiment; the appropriate Ē coordinate system is obtained
by a rotation of the reference (E) coordinate axes. Figure 2.2(e) shows a
different apparatus, constructed to be the mirror image of the reference
28 2 The equations of fluid motion

apparatus. In this case, the appropriate Ē coordinate system is obtained by

a reflection of a coordinate axis.
These coordinate transformations – rotations and reflections of the axes –
are precisely those considered in Cartesian tensors (see Appendix A). With
aij ≡ ei · ēj being the direction cosines, the scaled variables are
x̂i = x̄i /L = aji xj /L, (2.85)
Ûi = aji Uj . (2.86)
It follows immediately from the fact that the Navier–Stokes equations can
be written in Cartesian tensor notation that the transformed equations
are identical to those in the reference system (Eqs. (2.78)–(2.80)). Thus
the Navier–Stokes equations are invariant with respect to rotations and
reflections of the coordinate axes.
In these considerations it is important to distinguish between two kinds of
‘rotations.’ Here we are considering the Ē coordinate system obtained by a
fixed rotation of the E coordinate axes. By ‘fixed’ we mean that the direction
cosines aij do not depend on time. In contrast, we consider below rotating
frames, so that the direction cosines are time dependent.
The invariance with respect to reflections has a physical significance and
a mathematical consequence which are discussed at greater length in Ap-
pendix A. The physical significance is that the Navier–Stokes equations
contain no bias toward right-handed or left-handed motions. Of course such
bias can occur in a flow – most dramatically in a tornado – but it arises
from the initial or boundary conditions, or from frame rotation, not from
the equations of motion (expressed in an inertial frame).
Any equation written in Cartesian tensor notation ensures invariance
under rotations and reflections of coordinate axes. In contrast, an equation
written in vector notation and involving pseudovectors (e.g., vorticity), or
written in suffix notation using the alternating symbol εijk , does not ensure
these invariance properties.

Time reversal
Analogous to the reflection of a coordinate axis (e.g., x̄2 = −x2 ), we can
consider the reversal of time by defining
t̂ = −t U /L, (2.87)
Û (x̂, t̂) = −U (x, t)/U . (2.88)
It is readily shown that the corresponding transformed Navier–Stokes equa-
tions are the same as Eqs. (2.78)–(2.80), except that the sign of the viscous
 2.9 Transformation properties 29

term (proportional to Re−1 ) is altered. Thus, the Navier–Stokes equations

are not invariant under a time reversal; but the Euler equations are.

Galilean invariance
The remaining topics in this section are concerned with moving frames. We
consider first, as depicted in Fig. 2.2(f), the apparatus moving at a fixed
velocity V , so that both coordinate systems (E and Ē) are in inertial frames.
The transformations between the coordinate systems are
x̄ = x − V t, t̄ = t, (2.89)

Ū (x̄, t̄) = U (x, t) − V . (2.90)

A quantity that is the same in different inertial frames is said to be Galilean
invariant. From Eqs. (2.89) and (2.90) we obtain
∂Ūi ∂Ui
= , (2.91)
∂x̄j ∂xj

∂Ūi ∂Ui ∂Ui

= + Vj , (2.92)
∂t̄ ∂t ∂xj

DŪi ∂Ūi ∂Ūi DUi

≡ + Ūj = , (2.93)
Dt̄ ∂t̄ ∂x̄j Dt
showing that the velocity gradients and the fluid acceleration are Galilean
invariant, whereas the velocity and its partial time derivative are not. Other
quantities that are Galilean invariant include scalars such as φ(x, t) and
pressure p(x, t), and quantities related to velocity gradients, e.g., Sij , Ωij , and
the vorticity ω.
It is simply shown that the transformed Navier–Stokes equations (written
for Û = Ū /U in terms of x̂ ≡ x̄/L, etc.) are identical to Eqs. (2.78)–(2.80),
and hence are Galilean invariant. Just like all phenomena described by
classical mechanics, the behavior of fluid flows is the same in all inertial
frames.

EXERCISE
2.14 Which of the following are Galilean invariant:
(a) a streamline (which by definition is a curve that is everywhere
parallel to the velocity vector),
(b) a vortex line (which by definition is a curve that is everywhere
parallel to the vorticity vector),
30 2 The equations of fluid motion

(c) the helicity, which is defined as U · ω

(d) the enstrophy, which is defined as ω · ω
(e) material lines, surfaces, and volumes and,
(f) for a scalar field; ∂φ/∂t, ∂φ/∂xi , and Dφ/Dt?

Extended Galilean invariance

A peculiar property of the Navier–Stokes equations is that they are invariant
under rectilinear accelerations of the frame. We consider, as depicted in
Fig. 2.2(g), the second experiment being performed on a platform moving
at a variable velocity V (t), but with no rotation of the frame, so that the
coordinate directions (e.g., e1 and ē1 ) remain parallel. With the transformed
variables x̄, t̄, and Ū defined by Eqs. (2.89) and (2.90), the transformed
Navier–Stokes equations are
∂Ūj ∂Ūj ∂2 Ūj 1 ∂p
+ Ūi =ν − − Aj , (2.94)
∂t̄ ∂x̄i ∂x̄i ∂x̄i ρ ∂x̄j
where the additional term on the right-hand side is the acceleration of the
frame, A = dV /dt. The last two terms can be written
1 ∂p 1 ∂
+ Aj = (p + ρx̄i Ai ), (2.95)
ρ ∂x̄j ρ ∂x̄j
showing that the frame acceleration can be absorbed in a modified pressure.
Consequently the Navier–Stokes equations for the transformed variables
Û ≡ Ū /U , p̂ ≡ (p + ρx̄ · A)/(ρ U 2 ), (2.96)
are identical to Eqs. (2.76)–(2.80). Thus the scaled velocity Û and modified
pressure p̂ fields in the experiment in the frame with arbitrary rectilinear
acceleration are identical to those in the inertial flame. This is extended
Galilean invariance (which applies only to constant-density flows).

Frame rotation
Finally, we consider the second experiment being performed in a non-inertial
rotating frame, Fig. 2.2(h). In the Ē coordinate system, the time-dependent
basis vectors ēi (t) evolve by
d
ēi = Ω̃ij ēj , (2.97)
dt
where Ω̃ij (t) = −Ω̃ji (t) is the rate of rotation of the frame. Note that, in this
case, the direction cosines aij (t) ≡ ei · ēj (t) are time-dependent.
The Navier–Stokes equations transformed to the non-inertial frame are
 2.9 Transformation properties 31

the same as Eq. (2.94), but with the frame acceleration −Aj replaced by the
fictitious force
dΩ̃ij
Fj = −x̄i Ω̃ik Ω̃kj − 2Ūi Ω̃ij − x̄i (2.98)
dt̄
(see Exercise 2.15). The three contributions to F represent the centrifugal
force, the Coriolis force, and the angular acceleration force. The centrifugal
force can be absorbed into a modified pressure, but the remaining two forces
cannot. As is well known in meteorology and turbomachinery, Coriolis forces
can have significant effects on flows in rotating frames.
A quantity that is the same in rotating and non-rotating frames is said to
possess material-frame indifference. Evidently, the Navier–Stokes equations
do not have this property.
The effect of frame rotation is also evident in the vorticity equation. In
the non-inertial Ē coordinate system, the equation for the evolution of the
vorticity
∂Ūk
ω̄i ≡ εijk , (2.99)
∂x̄j

obtained from the Navier–Stokes equations (i.e., Eq. (2.94) with Fj in place
of −Aj ), is

∂ω̄i ∂ω̄i ∂2 ω̄i ∂Ūi ∂Ū
dΩ̃jk

+ Ūj =ν + ω̄j − 2εijk Ω̃
k − εijk . (2.100)
∂t̄ ∂x̄j ∂x̄j ∂x̄j ∂x̄j ∂x̄j dt

Evidently, because of the last two terms – which correspond to Coriolis and
angular acceleration forces – the vorticity equation in a rotating frame is
different than that in an inertial frame (Eq. (2.60)).

EXERCISE
2.15 Let X (t) be the position of a moving point relative to the origin of
the E coordinate system in an inertial frame. Let Y (t) = ēi (t)Yi (t) be
the position of the same point relative to the non-inertial frame Ē.
The origin of the Ē frame moves with velocity V (t), and its basis
vectors ēi evolve according to Eq. (2.97). If the origins are coincident
at time t = 0, then
t
X (t) = Y (t) + V (t ) dt . (2.101)
0

Show that the velocity and acceleration (relative to the inertial

32 2 The equations of fluid motion

frames) are
Ẋ = V + ēj (Ẏj + Yi Ω̃ij ), (2.102)
˙ ),
Ẍ = V̇ + ēj (Ÿj + Yi Ω̃ik Ω̃kj + 2Ẏi Ω̃ij + Yi Ω̃ (2.103)
ij

where an overdot indicates differentiation with respect to time.

Two-dimensional flows
Another peculiar property of the Navier–Stokes equations is that, for two-
dimensional flow (in the x1 –x2 plane, say), they are invariant with respect to
steady rotations of the frame in the plane of the flow (i.e., rotations about
the x3 axis), see Speziale (1981). For two-dimensional flows, it is sometimes
useful to re-express the Navier–Stokes equations in terms of streamfunction
and vorticity. The streamfunction ψ(x1 , x2 , t) is such that the velocities are
given by
∂ψ ∂ψ
U1 = , U2 = − , (2.104)
∂x2 ∂x1
and the only non-zero component of the vorticity is
∂U2 ∂U1
ω3 = − . (2.105)
∂x1 ∂x2
For steady rotations of the frame, the final term in Eq. (2.100) is zero,
and, for the two-dimensional flows considered, explicit evaluation of the
penultimate term reveals that it too is zero (see Exercise 2.17). Thus, for
this special case, the vorticity is unaffected by frame rotation, and it follows
that the Navier–Stokes equations exhibit material-frame indifference (in this
restricted sense).

EXERCISES
2.16 For two-dimensional flow, with U3 = 0 and U1 and U2 given by
Eqs. (2.104), show that the divergence of velocity is zero for all
streamfunctions. Show that the streamfunction and vorticity are re-
lated by the Poisson equation
 2 
∂ ∂2
+ ψ = −ω3 . (2.106)
∂x21 ∂x22
2.17 With reference to the penultimate term in Eq. (2.100), consider the
quantity
∂Ū

Ω∗i ≡ εijk Ω̃
k , (2.107)
∂x̄j
 2.9 Transformation properties 33

for a two-dimensional flow (in the x1 –x2 plane), and for frame rota-
tions in the same plane. Which components of ∂Ū
/∂x̄j and Ω̃
k are
zero? Show that Ω∗1 and Ω∗2 are zero. Obtain the result
 
∗ ∂Ū1 ∂Ū2
Ω3 = Ω̃12 + , (2.108)
∂x̄1 ∂x̄2
and hence argue that Ω∗ is zero for the class of flows considered.