Professional Documents
Culture Documents
Fluid Dynamics
1
1 Forces
1.1 Introduction
See Section 1.1 in Batchelor.
There are a few instances (of the order of ten) when Science had the greatest impact
on people’s life. Fluid Dynamics provided at least one of them. The list of greatest impacts
could include, for example,
What is a fluid? Fluids (unlike solids) can be easily deformed: Small forces (no matter
how small) can produce large deformations. A sphere-like blob of fluid can easily become
a spagetti-like blob
• Turbulence.
Richard Feynman: The problem of turbulence is “common for many sciences”; it is
“the central problem that people need to solve”. He gives two examples of turbulence:
(1) water flow in a pipe, (2) the motion of galaxies (with stars instead of molecules).
2
The fluidity at the molecular level. Consider the force between two molecules.
When the distance d between them is of the order 1 Angstrem= 10−8 cm, they experi-
ence a strong force of quantum origin. There are two possibilities: (1) chemical bond ⇒
solid, (2) no chemical bond ⇒ fluid. The force can be schematically shown as follows
F>0: repulsion
F<0: attraction
Let us define d0 as the distance where the force F changes sign. In liquids the typical
distance between molecules is of the order d0 . In gases the typical intermolecular distance
≫ d0 (e.g. 10d0 − 100d0 under normal conditions).
Melting ⇒ density falls by several percent (water is exception). Amazing: Such a
small change in the intermolecular spacing leads to such dramatic change in the molecular
mobility.
Our goal in this Chapter is to write Newton’s second law for fluid motion. For this we
will need to describe forces in fluids.
Fluid mechanics is normally concerned with the behavior of fluids on scales (called
macroscopic) large compared with the intermolecular distance.
In most cases Fluid dynamics studies variations in dynamic properties over distances
−3
10 cm. At the same time the intermolecular distance even in gases (at normal conditions)
is less than 100d0 = 10−6 cm. So, macroscopic scales contain at least billions of molecules
[(10−3 /10−6 )3 = 109 ].
The continuum hypothesis: Physical quantities (e.g. mass, momentum) are spread
uniformly over macroscopically small volumes. In strict reality they are concentrated in
tiny fractions of these volumes (namely in the nuclei of atoms).
Variables to describe the motion of a fluid:
)
ρ(x, t) density
at point x at instant t
u(x, t) velocity
3
u is not the velocity of a molecule
N
1 X
ρ(x, t)u(x, t) = mvn
N n=1
2. Exchange of momentum when molecules cross the imaginary boundary between two
fluid elements.
Example. Exchange of coal between two rail cars moving with different speeds on
parallel tracks.
How to describe such a force? It is determined not only by the point x and instant
t, but also by the orientation of the surface, passing through that point (i.e. by the unit
normal vector n).
4
n unit normal vector to the imaginary boundary
between the interacting fluid elements
xt
in
po
e
th
The penetration depth is supposed to be small compared with the linear dimensions of
the surface element δA. This is possible if the fluid dynamic quantities (like ρ(x, t) and
u(x, t)) change on much larger scale than the penetration depth.
The force per unit area, Σ(n, x, t) is called stress.
The stress is an odd function of the normal n. Indeed, by Newton’s third law,
Σ(−n, x, t) = −Σ(n, x, t)
5
c
nnn
δA1
δA2
b
δA3
The large face has area δA and the unit normal n (pointing outside of the tetrahedron).
What are the forces acting on this tetrahedron element of fluid?
Let a, b, c be the unit vectors along the axis X1 , X2 , X3 respectively. The total surface
force is
Σ(n)δA + Σ(−a)δA1 + Σ(−b)δA2 + Σ(−c)δA3
(the dependence of Σ on x and t is not displayed here, because these variables have the
same value — approximately in the case of x).
Newton’s law for the tetrahedron:
mass × acceleration = total body force + total surface force.
Let the linear dimensions of the tetrahedron be l, and let it approach zero, l → 0:
the volume of the tetrahedron δV ∼ l3 , and so are the mass and the body forces (provided
the density ρ is finite). At the same time the area scales like l2 . Thus, the total surface
force equals zero.
By geometric considerations
.
Recall also that stress is odd.
Thus,
Σ(n) = Σ(a)a · n + Σ(b)b · n + Σ(c)c · n
or in writing by components
σij is the i-component of the stress (short-range force per unit area) exerted on a
surface element which has a normal n pointing in the j-direction.
6
These relations follow from the balance of the angular momentum (the angular momentum
argument).
Consider a small rectangular domain in the fluid
X2
δx2
δx3
δx1
X1
X3
For simplicity, forget about the third dimension X3 . Next picture shows the outside unit
normal vectors
b
X2
-a
a
X1
-b
7
X2
σ22(δx1δx3)
σ12(δx1δx3)
σ21(δx2δx3)
σ11(δx2δx3) .O σ11(δx2δx3)
σ21(δx2δx3)
σ12(δx1δx3)
σ22(δx1δx3)
X1
Q: Why is — according to this figure — the total surface force equal zero?
A: There are some volume forces and acceleration terms in the Newton law. However,
they are all of the order of δV ∼ l3 (l is a typical linear size). At the same time, the
surface force is of the order δA ∼ l2 .
The balance of the surface forces is, actually, not exact. However, the difference is again
of the order l3 , e.g. the sum of the tangential forces on the lower and upper faces is
−[σ12 (δx1 δx3 )]δx2 + [σ21 (δx2 δx3 )]δx1 = O(l3 ) · O(l)
So, as l → 0, we find σ12 = σ21 . Similar, we can establish the other two symmetries.
This equation is useful if the Force can be expressed in terms of positions and velocities
of all the particles.
So, our goal now is to express the stress tensor in terms of the functions that determine
the state of the fluid (in particular u(x, t) and ρ(x, t)).
8
1.7 The stress tensor in a fluid at rest (u ≡ 0)
See Section 1.3 (the second part).
We have defined a fluid as being unable to withstand any shear forces (which deform
the fluid without changing its volume).
It means that for a fluid at rest, the stress tensor should be a diagonal matrix:
σ11 0 0
σ = 0 σ22 0 .
0 0 σ33
X2
σ22(δx1δx3)
σ12(δx1δx3)
σ12(δx1δx3)
σ22(δx1δx3)
X1
and u 6= 0.
Q: What matrix is diagonal in any orthonormal basis?
A: All eigenvalues of such matrix should be equal. So, such matrix is a multiple of the
unit matrix:
σij = −pδij .
This formula includes “−” sign in the r.h.s., because then p = p(x, t) is the usual pres-
sure: For positive pressure p, the force exerted on the fluid element through its boundary
is opposite to the outside normal to this boundary.
9
We will assume that the stress tensor is a linear function of the velocity gradients (at
the same point and at the same instant):
∂uk
σij = Bij + Aijkl
∂xl
Remarks:
1. σij cannot depend on the velocity itself, since a uniform motion (with constant
speed) should not be related to stress (a Galilean transformation should not change
the stress).
∂uk
2. The relation is local in space and time: σij at (x, t) depends on ∂xl at the same
(x, t).
3. This relation holds with excellent accuracy for many fluids — in particular, for water
and air.
Fluids satisfying this relation are called Newtonian
What kind of stress tensor can take place for the non-Newtonian fluids? For instance,
∂uk
• σij may be a nonlinear function of ∂xl ,
∂uk
• σij may be a nonlocal function: It could depend on ∂xl at the preceding instants
in time or at different points in space.
Bij = −pδij .
2. The stress tensor σij is symmetric w.r.t. the transposition of its indices i ↔ j (see
the angular momentum argument). Therefore, Aijkl should be symmetric w.r.t. the first
pair of indices, e.g. A12kl = A21kl .
3. Aijkl should be symmetric w.r.t. the last pair of indices. This can be shown by the
following the rotation argument.
Consider a rotation of the fluid as a whole around axis X3 . Then the velocity of the
fluid
a b c −ωx2
u = ω × x = 0 0 ω = +ωx1 .
x1 x2 x3 0
10
Calculations give
0 −ω 0
∂uk
= ω 0 0 ,
∂xl
0 0 0
and so,
σij = Bij + Aij12 × (−ω) + Aij21 × (ω) = Bij + ω[Aij21 − Aij12 ].
The forces should have the same value if ω is replaced by −ω. This is only possible if
the bracket equals zero. Thus, Aij12 = Aij21 . Similar, we can establish the other two
symmetries.
Due to this symmetry, the stress tensor is a linear function of the so-called rate-of-
strain tensor
1 ∂uk ∂ul
ekl = + , σij = Bij + Aijkl ekl .
2 ∂xl ∂xk
Both the stress and the rate-of-strain tensors are symmetric matrices.
A fluid flows and mixes. If there is no preferred direction, Aijkl should be isotropic
tensor: When we change basis, the coefficients Aijkl should remain the same. It is shown
that any symmetric tensor of the fourth order is a sum of products of Kronecker delta
tensors:
Aijkl = λδij δkl + µδik δjl + γδil δjk .
Now, we have only 4 independent scalars
11
Our pressure p is thermodynamic equilibrium pressure, which we will denote here as
pe . It is related to other thermodynamic quantities by the equilibrium thermodynamics
relations, observed in the fluid at rest.
So,
2
pm = pe − (λ + µ) ∇ · u.
3
2
The coefficient κ = λ + 3 µ is called the expansion viscosity. In most of the physical
situations
κ = 0 ⇒ pm = pe .
Batchelor considers the difference between the two pressures in Section 3.4, p.154.
We will see soon that if the fluid is incompressible, ∇ · u = 0. Then certainly the two
pressures are the same pm = pe = p and the second viscosity λ drops out
∂uk ∂ul
σij = −pδij + µ + .
∂xl ∂xk
X2
ξ(x, t)
x x+ξ(x,t)
X1
A: We can use the deformation vector ξ(x, t) which shows the shift at the instant t of a
point that was at x, when forces were absent.
Solids also subject to the long-range and short-range forces. The latter are also ex-
pressed by the stress tensor σij .
A: We can use the deformation vector ξ(x, t) which shows the shift at the instant t of
a point that was at x, when forces were absent.
Q: How is the stress tensor can be expressed in terms of ξ(x, t)?
12
A: Hook’s law:
∂ξk
σij = Cijkl
∂xl
.
Q: How many numbers are in Cijkl ?
A: 34 = 81. But not all of them are independent.
Similar to the fluid case, Cijkl has the following symmetries:
• Cijkl is symmetric w.r.t. the first pair of indices (the angular momentum argument).
• Cijkl is symmetric w.r.t. the last pair of indices (the rotation argument).
There is one more symmetry, which was absent in the case of fluids.
• Cijkl is symmetric w.r.t. the transposition of the first and the last pairs of indices
Cijkl = Cklij . This takes place if the system is conservative.
1 2 3 4 5 6
1
15
2
(i,j)
3
6
4
15
5
6
13
Due to the symmetry of Cijkl w.r.t. the last pair of indices, the stress tensor depends
on the symmetric combination of the deformation derivatives
!
1 ∂ξi ∂ξj
εij = + ,
2 ∂xj ∂xi
which is called the strain tensor.
In fluid mechanics, the stress tensor depends on the symmetric combination of the
velocity derivatives
1 ∂uk ∂ul
ekl = +
2 ∂xl ∂xk
which is called the rate-of-strain tensor.
Force. The force is the total of the volume forces and surface forces.
14
X2
.x δx2
δx3
δx1
X1
X3
15
(which runs the whole upper face when each of the parameters α and γ changes from −1 to +1) the force
is equal to
α 1 γ ∂σi1 α ∂σi2 1 ∂σi3 γ
σ(x1 + δx1 , x2 + δx2 , x3 + δx3 ) = (δx1 )2 (δx3 ) + (δx1 )(δx2 )(δx3 ) + (δx1 )(δx3 )2 .
2 2 2 ∂x1 2 ∂x2 2 ∂x3 2
It might seem strange that we neglect the first and the third terms, but leave the second one, although all
of them are of the same order. Similar expression we find for the bottom face. However in their sum the
terms with α and γ cancel each other, and we find the expression, as if the force on the each of the upper
and the bottom faces were constant.
Now we derive the Navier-Stokes equation more rigorously: We write Newton’s second
law for an arbitrary material volume τ , enclosed by the material surface S.
Recall Newton’s mechanics for a system of N particles:
N N
mn an = Fn
X X
(n = 1, .., N )
n=1 n=1
The forces Fn acting on the particles in the system can be divided into the forces acting
between the particles (internal forces) and the forces acting on the particles from the
outside sources (outside forces). The sum of the internal forces equals zero according to
Newton’s third law, and so the sum of all forces in the r.h.s. can be replaced by the sum
of the outside forces.
For the system of fluid particles within the material volume τ we have
N
Dui
Z
n n
X
m a = ρdτ
n=1
Dt
Thus, we have an equation with three terms, each being an integral over the volume τ .
Since τ is arbitrary, we arrive at the equation (1).
Remarks.
• The equation is essentially nonlinear, even for Newtonian fluids when the forces
depend linearly on the velocity field. The reason is the fluidity. The form of the
nonlinearity (see the l.h.s.) is called hydrodynamic nonlinearity.
16
• The viscosities µ, λ are functions of position x and time t, because, for instance,
they depend on the temperature T (x, t). See p. 147.
• The Navier-Stokes equation should be supplemented by the boundary conditions.
No-slip condition: The velocity u(x, t) of the fluid on the solid boundary should
be equal to the velocity of the boundary.
See Section 3.3 (fourth part).
u(y)
d
Assume that the bottom plate is at rest, while the upper plate is moving with some velocity
u0 in the X direction, parallel to the plane. d is the distance between the planes
x u(y)
x = y , u = v = 0 , F = 0.
z w=0
In most of situations, the Navier-Stokes equation cannot be solved analytically because
this equation is nonlinear. But in this case the nonlinearity vanishes: One can calculate
that ∂u ∂u ∂u ∂u
∂t + u ∂x + v ∂y + w ∂z
Du ∂v ∂v ∂v
= ∂t + u ∂x + v ∂y + w ∂v
∂z = 0.
Dt ∂w ∂w ∂w ∂w
∂t + u ∂x + v ∂y + w ∂z
The stress tensor you calculated in the HW:
µ ∂u
−p ∂y 0
∂u
σ = µ ∂y −p 0 .
0 0 −p
So, the Navier-Stokes equation gives
∂p 2
0 = − ∂x + µ ∂∂yu2
∂p
0 = − ∂y
∂p
0 = − ∂z
17
From the second and the third equations, the pressure p depends only on x; then from the
first equation,
∂p ∂2u
= µ 2 = G = const,
∂x ∂y
and therefore,
G 2
p = Gx + p0 , u= y + C1 y + C2 .
2µ
2-D Coette flow. There is no pressure gradient, G = 0, and u = ud0 y (it grows
linearly from u = 0 to u = u0 ). The force that we need to apply to the upper plane (per
unit area) in order to move it with the speed u0
force u0
=µ .
area d
This formula can be used to determine the viscosity of particular fluids. In practice, people
use two coaxial cylinders (instead of the two parallel planes), one being at rest, the other
rotating with some non zero angular velocity.
Ω
2-D Poiseuille flow. The upper plane is also at rest, u0 = 0, and u = G2 y(d − y)
(parabolic profile). One can calculate the flux Q of the fluid volume through a cross-
section of the pipe. In the HW you need to find this flux for the 3-D pipe with circular
cross-section.
18
This solution (with parabolic profile) exists for all possible pressure gradients G. How-
ever, when G is sufficiently large, this laminar flow becomes unstable. Absolutely different,
turbulent, flow is realized in practice for large G. Nobody was able to calculate theoret-
ically the dependence Q vs. G. This example is referred by Richard Feynman as an
example of the Turbulence Problem.
19
2 Conservation laws
See the last subsection of Section 3.1, pp. 135-136.
In the absence of sources and sinks of the fluid, this mass changes due to the net flux of
the fluid through the boundary. How much fluid mass enters through the boundary?
I
− ρu · ndA.
Thus,
d
Z I
ρdV = − ρu · ndA.
dt
On the left, differentiate in t under the integral sign (remembering that the volume V is
fixed). On the right transform the surface integral to the volume integral using the Gauss
theorem.
∂ρ
Z
+ ∇(ρu) dV.
∂t
Since the volume V is arbitrary,
∂ρ
+ ∇ · (ρu) = 0 (2)
∂t
This equation can be written in a different form
Dρ
+ ρ∇ · u = 0.
Dt
So, what is an incompressible fluid?
The density of an incompressible fluid does not haven to be a constant throughout the
fluid. The density of a fluid element doesn’t change if Dρ
Dt = 0, i.e. ∇u = 0. This is the
condition of incompressibility.
For an incompressible fluid, the second viscosity λ drops out, and the Navier-Stokes
equation (1) becomes
( !)
Dui ∂p ∂ ∂ui ∂uj
ρ = ρFi (x, t) − + µ + .
Dt ∂xi ∂xj ∂xj ∂xi
20
2.2 Conservation of momentum
Conservation of momentum is equivalent to Newton’s law, and so the differential form of
the momentum conservation for fluids should be equivalent to the Navier-Stokes equation.
Consider again the volume V whose position is fixed relative to the coordinate axis.
The total momentum in the volume V is
Z Z
ρudV or ρui dV.
ρFi dV
All the terms of the momentum balance are expressed as integrals over the volume V . Since
the volume V is arbitrary, we have the differential form of the momentum conservation
∂ρui ∂ρui uj ∂σij
+ = ρFi + . (3)
∂t ∂xj ∂xj
This equation should be equivalent to the Navier-Stokes equation. It is, which we can
easily note differentiating the products
" # " #
∂ρ ∂ρuj ∂ui ∂ui ∂σij
ui + +ρ + uj = ρFi + .
∂t ∂xj ∂t ∂xj ∂xj
The first bracket vanishes due to the mass-conservation equation. The rest is the Navier-
Stokes equation (1).
The system of the Navier-Stokes equation and the mass-conservation equation has 4
equations and 5 unknowns u(x, t) = (u, v, w), ρ(x, t), and p(x, t).
So, we need one more equation. This will be the energy conservation equation; it
expresses the thermodynamic balance of the fluid, so that the fluid dynamics includes
thermodynamics.
There is, however, an important case when the system of 4 equations contains only 4
unknowns. This is the case of a uniform fluid, when the density ρ and the viscosity µ
are constants throughout the fluid. Then the NS equation becomes
Dui 1 ∂p µ
= Fi (x, t) − + ν∇2 ui where ν= .
Dt ρ ∂xi ρ
21
The coefficient ν is called the kinematic viscosity, while µ is often referred as the
dynamic viscosity.
Under normal conditions (15 degrees C; 1 atm.)
kg 2
for air: µ = 1.8 · 10−5 m·s , ν = 1.5 · 10−5 ms ,
kg 2
for water: µ = 1.1 · 10−3 m·s , ν = 1.1 · 10−6 ms .
It is interesting to note that when the coefficient ν is more relevant than µ, the water
is less viscous than the air.
For a uniform fluid, the system of the Navier-Stokes equation and the mass-conservation
equation
Du p
= −∇ + ν∇2 u + F(x, t),
Dt ρ
∇ · u = 0.
Next we will consider two examples in which compressibility is important (the last
system cannot be used). However, the pressure and the density are directly related,
and so the system of the Navier-Stokes equation and the mass-conservation equation is
sufficient: It contains 4 equations and 4 unknowns. To determine this relation between
the pressure and the density we will need to use thermodynamics. So these two examples
will serve us as motivation to consider thermodynamic balance. After these two examples,
we will consider general situation with five differential equations (the mass-conservation,
the momentum-conservation, and the energy-conservation).
22
Take the system of the Navier-Stokes equation and the mass-conservation equation
(compressibility is important).
Neglect viscousity.
Neglect body forces (in particular, the gravity).
Du
ρ = −∇p,
Dt
∂ρ
+ ∇ · (ρu) = 0.
∂t
Exact solution: u ≡ 0, ρ(x, t) ≡ ρ0 = const, p(x, t) ≡ p0 = const. This is an equilib-
rium (still air).
Now, let us linearize our equations around this exact solution: Take
∂ 2 ρ̃ ∂ũ
2
= −∇ · ρ0 = ∇2 p̃.
∂t ∂t
Already Newton did this calculation. To calculate the speed of sound, he needed the
relation between the pressure and the density (then he would have 1 equation with 1
unknown). Some years before Newton, Boyle did such experiments
He found that
p p0
= = const (Boyle’s law).
ρ ρ0
Then our equation becomes the wave equation
∂ 2 ρ̃
= c2 ∇2 ρ̃
∂t2
p
with the speed of sound c = p0 /ρ0 .
23
For normal air (at 15 degrees C),
kg kg
p0 = 1atm = 105 , ρ0 = 1.2 ⇒ c = 290m/s.
m · s2 m3
You can imagine how greatly Newton was disappointed when he compared his theoretical
value with the experimental value c = 340m/s.
What is wrong?
The internal energy (E) of a uniform mass of fluid is changed by a gain of heat (Q)
and by the performance of work (W ):
∆E = Q + W
(Batchelor considers all these quantities for a fluid element of unit mass. But here, the
mass is arbitrary).
mechanical
work W
Internal energy
heat Q
The internal energy is a function of the parameters of the system, and so, ∆E depends
only on the initial and final states. On the contrary, Q and W depend on the way of
transition between the initial and final states.
In the infinitesimal form, the first law of thermodynamics
dE = δQ + δW.
Now, back to the sound speed. Our goal is to find dependence p vs. ρ in the adiabatic
process.
24
For adiabatic process (without heat exchange), δQ = 0.
Suppose, the fluid is compressed slowly enough for the pressure p to equilibrate to a
uniform pressure throughout the gas element; then δW = −pdV , where dV is differential
of the volume of the fluid element.
The energy differential is dE = M Cv dT , where M is the mass of the gas element, dT
is the temperature differential, and Cv is the specific heat (the sub index v means that the
volume is held fixed, so that Cv shows how much heat we need to transfer to 1 kilogram of
the gas to increase its temperature by 1 degree K, if the volume of the gas is held fixed).
Thus, the first law of thermodynamics gives
M Cv dT = −pdV.
This is one equation relating three quantities temperature T , pressure p and volume V
(or density ρ = M/V ) of a fluid element with a constant mass M . We need one more
equation to find the relation between p and ρ. This will be the equation of state
M
pV = RT
m
where R is the universal gas constant, m is the mass of one mole of the gas.
Q: What is a mole?
A: A mole of something is the Avogadro number NA (≈ 6 · 1023 ) of these things.
Examples.
25
By Taylor’s formula expansion,
dp p0
p − p0 = (ρ − ρ0 ), ⇒ p̃ = γ ρ̃.
dρ at ρ0 ρ0
We again have the wave equation, but with a different speed
p0
r
c= γ .
ρ0
For the air, γ = 7/5, and the sound speed is c = 340m/s, which agrees with experiments.
√
The new value is bigger than Newton’s one by the factor γ.
Q: We said that in the sound wave the compression is so fast that the fluid elements
have no time to exchange heat.
We also said that the compression is slow enough that the pressure is in time to equilibrate
to a uniform thermodynamic pressure.
How can it be that the compression is fast in one respect and slow in another?
A: The pressure equilibrates much faster (with the sound speed) than temperature
(that equilibrates diffusively).
Consider
compression
λ
2
Heat exchange: Fourier’s law Tt = k∇2 T ; k = 2 · 10−5 ms for air at the normal conditions
(1 atm, 15 degrees C).
2
The characteristic time of heat exchange: τh = λk .
The characteristic time of pressure equilibration: τp = λc .
kν
τp ≪ τh ⇔ ≪1
c2
26
where ν is the frequency of the sound; people can hear in the range 20Hz < ν < 20KHz.
So,
2
τp kν 2 · 10−5 ms · 20000s−1
= 2 < ≈ 10−5.5 ;
τh c (340m/s)2
clearly, the heat exchange is irrelevant.
If the fluid were incompressible... ... The atmosphere would stable if ρ(z) were a
decreasing function.
Why?
If a fluid element is dispaced upward, it would find itself among lighter fluid, and it would
be forced back.
In the compressible case, the decrease of the density ρ(z) is not enough, the density
should not be just decreasing function of z, but it should decrease sufficiently fast.
Indeed, when a fluid element at a level z, with density ρ(z), is dispaced upward to
level z + dz, it would feel less pressure, and therefore it would expand, and its density
would decrease to ρ∗ = ρ(z) − δρ. The atmosphere is stable if the displaced fluid element
is forced back, i.e. the new density ρ⋆ exceeds the surrounding density ρ(z + dz).
27
p(z+dz)
δΑ
ρg
δΑ
p(z)
upward
displacement
The temperature T also changes with height. ⇒ The displaced fluid element would
find itself surrounded by gas at a different temperature. ⇒ Heat exchange would oc-
cur. However, it is a slow process. We can assume that the fluid element is dispaced
adiabatically γ
p − ρgdz ρ∗ ρ2 g
= ⇒ ρ∗ = ρ − dz.
p ρ γp
We should compare this density ρ∗ with the surrounding density ρ + ρ′ dz. Thus, the
condition for the stability of the atmosphere is
dρ ρ2 g
<− .
dz γp
28
The density of the atmospere should not just decrease with height, but decrease sufficiently
fast.
How does the temperature need to vary with height in order for the atmosphere to be
stable? We will answer this question assuming that the atmosphere consists of a perfect
gas
R p′ ρ′ T ′
p = ρT ⇒ = + .
m p ρ T
Using the hydrostatic balance, p′ = −ρg, and the stability condition in terms the density
ρ,
T′ ρg ρ′ ρg ρg
=− − >− + .
T p ρ p γp
Applying again the equation of state for a perfect gas, we find
dT m γ−1
>− g .
dz R γ
The earth atmosphere:
J K
γ = 1.4, m = 0.029kg/mol, R = 8.3 , g = 9.8m/s2 ⇒ T ′ > 0.01 .
mol · K m
So, the temperature can decrease with height at most 10 degrees (K or C) per kilometer.
The lower atmosphere turns out to be close to being nutrally stable (because of turbulence),
and so this rate is usually observed.
∆E = Q + W.
Let E be the internal energy per unit mass. Consider some volume V (fixed in the
laboratory frame) surrounded by a boundary A. The total energy in the volume V
u2i
Z
ρ(E + ) dV
2
is changing in time due to:
1. The flux of the energy through the boundary
∂
I Z
− ρ(E + u2i /2)uj nj dA =< Gauss theorem >= − [ρ(E + u2i /2)uj ]dV.
∂xj
29
3. The work (per unit time) by the surface forces
∂
I Z
uj σij nj dA =< Gauss theorem >= [ui σij ]dV.
∂xj
Here we have assumed that the heat flux is defined by a vector q proportional to
the temperature gradient (q = −k∇T , the Fourier law). The coefficient k is the
thermal conductivity of the fluid.
All these quantities are expressed as integrals over the volume V . And since V is
arbitrary, we find the differential equation of the energy conservation
" #
∂ ∂ ∂ ∂ ∂T
[ρ(E + u2i /2)] = − [ρuj (E + u2i /2)] + ρFi ui + [ui σij ] + k . (4)
∂t ∂xj ∂xj ∂xj ∂xj
The system of the conservation equations (2), (3), (4) (respectively of mass, momen-
tum, and energy) make the the Everest of Fluid Dynamics.
Now we have 5 equations with 7 unknown fields. We should supplement this system by
the equation of state, which enables us to express all thermodynamic quantities in terms
of any two of them. Note: The coefficients µ, λ, k can be functions of (x, t) (because, in
particular, they depend on the temperature T , which can vary in (x, t). However, we can
express them in terms of the two thermodynamic quantities. Thus, we supplement this
system by the relations of the equilibrium thermodynamics, expressing all thermodynamic
quantities in terms of two, say the density and the temperature
Remarks.
1. Batchelor (Section 3.4) derives the energy equation for a material fluid element of
unit mass. Our energy equation can be readily reduced to his equation if we differentiate
the products in our energy equation and use the mass-conservation equation (2) and the
Navier-Stokes equation (1).
Indeed,
" # " #
∂ρ ∂(ρuj ) ∂(E + u2i /2) ∂(E + u2i /2)
+ (E + u2i /2) + ρ + uj =
∂t ∂xj ∂t ∂xj
" #
∂ ∂ ∂T
= ρFi ui + [ui σij ] + k .
∂xj ∂xj ∂xj
The first bracket is zero, due to the mass-conservation equation. Differentiating the prod-
ucts again,
" #
DE Dui ∂σij ∂ui ∂ ∂T
ρ + ρui = ρFi ui + ui + σij + k .
Dt Dt ∂xj ∂xj ∂xj ∂xj
The second term on the left and the first and second terms on the right represent the NS
equation, multiplied by ui , and so, they cancel. This is the reason for Batchelor (p. 152,
30
∂σ
after formula (3.4.1)) to write that ui ∂xijj contriburtes to the gain of the kinetic energy
∂ui
u2i /2, while σij ∂x j
contributes to the internal energy E
" #
DE ∂ui ∂ ∂T
ρ = σij + k .
Dt ∂xj ∂xj ∂xj
(recal, summation over the repeated indices i, j is implied). Thus, “any viscous motion is
inevitably accompanied by unidirectional transfer of energy from the mechanical agencies
causing the motion to the internal energy” (this is the dissipation); the viscosities µ and
λ need to be positive. See pp. 153-154.
3. When all the coefficients of molecular transport µ, λ, and k turn into zero, we have
DE p ∂uk
=− .
Dt ρ ∂xk
By virtue of the mass conservation equation, the r.h.s. of this equation is
p Dρ D 1
2
= −p ,
ρ Dt Dt ρ
so the fluid motion is adiabatic: dE = −pdV ≡ pd ρ1 and δQ = 0. Since the heat Q
is related to the enstrophy S by the formula δQ = T dS (for reversible processes), the
adiabatic fluid motion is also called isentropic flow (the entropy of any fluid element
remains constant). If the entropy (per unit mass) is constant throughout the fluid, the
flow is called homentropic.
The path of a fluid element (or of a tracer particle, following passively the fluid flow) does
not in general cincide with the a streamline, although it does so when the fluid flow is
steady.
A stream-tube is the surface formed instanteneously by all streamlines that pass
through a given closed curve.
Assume:
31
1. Fluid is frictionless, λ = µ = 0, and does not conduct heat, k = 0.
3. The body force F is potential: There is a function Ψ(x, t) such that F = −∇Ψ.
u2 p
H= +E+ +Ψ (5)
2 ρ
2’
2
δΑ2
x2
1’
1
δΑ1
x1
′
The fluid between cross-sections 1 and 2 during small time from t to t has moved along
′ ′
the streamtube to the volume between the cross-sections 1 and 2 . The change of its
energy (it can be calculated considering the end portions only, since the flow is steady)
u2 ′ ′
−( + E + Ψ)at x1 ρ1 δA1 |u1 |(t − t) + (...)at x2 ρ2 δA2 |u2 |(t − t)
2
′
is due to the work of the pressure force during time (t − t)
′ ′
p1 δA1 |u1 | (t − t) − p2 δA2 |u2 | (t − t).
Q: Why isn’t there work accross the latteral boundary of the stream tube?
A: There the velocity is orthogonal to the force.
′ ′
By virtue of the mass conservation, ρ1 δA1 |u1 |(t − t) = ρ2 δA2 |u2 |(t − t), and we
establish the constancy of the function (5) along the streamlines.
Remark. Batchelor’s derivation is a little more general, and then he considers how the
constant H varies from one streamline to another in steady isentropic flow (See Section
3.5, zero subsection).
In the next couple sections, we consider two applications of the Bernoulli theorem.
32
2.8 Efflux
See Section 6.3, first subsection.
Streamlines
h
u0
m = ρS|u0 |,
33
However, this is not what observed in experiments.
The streamline of the jet are converging for a few diameter after the orifice. The cross-
sectional area of the jet — when the streamlines become parallel — is smaller by some
factor α than the area S of the orifice, and so the efflux
m = ραS|u0 |.
In the case of a circular hole in a thin wall, the experiments give 0.61 < α < 0.64.
There are two situations when the efflux can be calculated analytically.
α=1
Streamlines
h
u0
Borda’s mouthpiece.
34
α=1/2
u0
S
The region,
where the fluid velocity
is not negligible, and so,
near the wals, the pressure
is hydrostatic.
Each second, the fluid acquires momentum m|u0 |. Why? Due to the force from the vessel
= ρghS. Thus, ραS|u0 |2 = ρghS. Together with the Bernoulli’s expression for the speed
|u0 |, this gives α = 1/2.
streamline
|u|, T
35
Example. Ping-pong ball in an air jet.
We compare pressure on different streamlines. Why can we do this?
Far up-stream, the velocity and pressure are constant, and the Bernoulli constant H is
the same across streamlines.
remains constant.
Indeed, the curve Γ(t) can be parametrized by some parameter s, x = X(s, t), say
0 ≤ s ≤ 1. Since the curve Γ is moving with the fluid,
∂X
= u.
∂t
∂X
Since dl = ∂s ds, Z 1 ∂X
C(t) = u(X(s, t), t)
ds
0 ∂s
(Such an expression is often the definition of a curvilinear integral.) So,
C(t)
Z 1 Du ∂X
Z 1 ∂u
= · ds + u ds.
dt 0 Dt ∂s 0 ∂s
∂ 2
In the second integral in the r.h.s. the integrand is 21 ∂s u , and since the points corre-
sponding s = 0 and s = 1 coincide, the integral vanishes. In the first integral on the right,
the integrand — due to the NS equation — equals
1 ∂X
− ∇p + ∇Ψ · ,
ρ ∂s
and since ρ is a function of p, the integrand is also a full derivative of some function φ(s)
with φ(0) = φ(1); integration of this function also vanishes. Thus, we prove the constancy
of the circulation (6).
Remarks.
36
1. The theorem holds even if the fluid domain is not simply connected, and the closed
curve Γ cannot be spanned by a surface wholly lying in the fluid.
2. The proof requires that the viscosity is zero on the curve Γ (not everywhere in the
fluid).
37
3 Airplane lift
Having reached the Everest of Fluid dynamics, we can decend into places of interest. Our
first goal is to obtain the lift force, that allows airplanes to fly. We will consider the
simplest possible situation that allows us to find realistic lift. In this Section, we will
simplify the fluid dynamics description assuming that the fluid flow is
1. Inviscid
3. Irrotational
4. 2D (two-dimensional)
we assume that the fluid density is uniform throught the fluid (not only constant for
motion of a fluid element).
The order of spatial derivatives has dropped.
We are unable now to satisfy the no slip boundary condition.
Instead we require that only normal component of the velocity field is equal to the normal
component of the rigid surface (normal to the surface). If the rigid body is at rest,
We assume that near rigid boundaries there are some thin boundary layers where the fluid
velocities adjust to velocities of the solid bodies, so that the no slip boundary condition
is actually satisfied.
Our goal is to consider a solid body moving with constant velocity through the fluid
at rest. Using the Galilean Invariance, we consider the body at rest in a fluid stream,
uniform at infinity.
The conditions for Kelvin’s theorem are satisfied.
Upstream the vorticity ω = ∇ × u is zero, and the circulation equals zero along any closed
curve. However, the topology in the 2D and 3D are different: In the 3D, the fluid domain
is singly connected, but in 2D, it is doubly connected.
38
Then
u = ∇φ
where the potential φ should satisfy the Laplace equation
∇2 φ = 0.
A linear equation! The nonlinear momentum equation becomes only equation for the
calculation of pressure after the velocity field is found from the linear equation. This
is a cruciakl simplification that allows to employ powerful mathematical techniques. In
particular, the velocity field (but not the pressure field) has the superposition principle.
Use the iidentity
1
∇(u2 ) = (u · ∇)u + u · [∇ × u].
2
When u = ∇φ, we can integrate the momentum equation (p.382)
∂φ u2 p
+ + + Ψ(x) = const throughout the fluid
∂t 2 ρ
(Ψ is the potential of the force F). This fact is similar to the Bernoulli theorem. The
difference: (1) In the Bernoulli theorem the flow is required to be steady, (2) here the
const is the same throughout the entire fluid.
2D. By Kelvin’s theorem, the circulation should be zero along any curve that can
be continuously deformed from a curve upstream. The circulation can be non-zero for
curves enclosing the body. The circulation should remain constant on any such curve.
The circulation along any curve should be a multiple of some constant (vortex strength).
The potential function still exists (u = ∇φ), but the function φ(x, t) can be multivalued.
Examples.
1. The uniform flow
u = (U, 0, 0), φ = U x.
2. The flow near a stagnation point
39
κ q
y
φ=− θ (r = x2 + y 2 , θ = arctan ).
2πr x
This is an example of multivalued potential.
The flow is irrotational everywhere, besides the origin.
The circulation along any curve enclosing the origin equals κ×the number of times the
curve goes around the origin.
Stream function.
In 2D the incompressibility condition implies the existence of a function ψ(x, y, t) such
that
∂ψ ∂ψ
u= , v=− .
∂y ∂x
The function ψ(x, y, t) is called the stream function; it should satisfy equation
∂ 2 ∂ψ ∂∇2 ψ ∂ψ ∂∇2 ψ
∇ ψ− + = 0.
∂t ∂x ∂y ∂y ∂x
The streamlines are the level lines of the stream function.
If the flow is steady, then the values of the stream function is constant along a streamline.
The stream function is constant along a rigid boundary at rest.
w = φ + iψ
z = x + iy −→ ζ = ξ + iη.
40
3.2 Magnus effect
Imagine a circular cylinder moving orthogonal to its axis with the velocity (U, V, 0) in
some fluid. Will it have some lift force?
We will see that it is possible to find lift considering irrotational, incompressible, 2D
flow around a circle.
In the frame fixed at circle’s center, the fluid has velocity (−U, −V ) at infinity. The
circle might rotate around its center, but this rotation has no effect if the fluid is inviscid.
Our strategy:
1. Find the velocity field. For this we will construct the complex potential.
2. Using the integrated momentum equation find the pressure along the boundary of
the circle. Then calculate the total force on the circle, integrating the pressure over
the boundary.
∇2 φ = 0,
φ(x, y) = −U x − V y when (x, y) → ∞,
∇φ · n = 0on the boundary of the circle x2 + y 2 = a2 .
Recall method of images. Suppose we wish to solve the Poisson equation in a half
plane x ≥ 0 with the boundary condition of vanishing of the normal component of the
gradient ∇φ.
Since now the source is an even function of x, the solution also is an even function of x,
and the boundary condition ∂φ
∂x = 0 is satisfied automatically.
We can also consider images in the circular boundary x2 + y 2 = a2 :
a2
z = x + iy −→
z∗
and construct the solutions of the Poisson equation in a region with a circular boundary.
For instance, we can do it with the aid of a conformal transformation. The final result
that we need is stated in the following theorem.
41
The circle theorem (Milne-Thomson, 1940). (p. 422-423). Suppose that in the
absence of the circle |z| = a the complex potential is w = f (z), the function f (z) being
free from cingularities in the region |z| < a. Then the complex potential
" !#∗
a2
w = f (z) + f
z∗
has the following two properties (1) it has the same singularities outside the circle (|z| > a)
as the original complex potential f (z), and (2) the boundary of the circle (|z| = a) is a
streamline (and therefore, the normal component of the velocity vanishes).
2
Indeed, the points az ∗ lies inside the circle, if z lies outside. Since f (z) has no singular-
ities inside the circle, the new complex potential has the same singularities as the original
potential outside the circle.
On the boundary of the circle (z = aeiθ ) the value of the complex potential
h i∗
w = f (aeiθ ) + f (aeiθ )
Consider a circle held in a stream of the uniform velocity (−U, −V ) at infinity. In the
absence of the circle, the flow has complex potential f (z) = (U − iV )z (one singularity is
at infinity). So for the flow with the circle, the complex potential is
a2
w = (U − iV )z + (U + iV ) .
z
It is not the only irrotational flow (satisfying the boundary conditions at ∞ and on
the circle boundary); we can superimpose (the velocity is defined by a linear equation) a
vortex
a2 iκ z
w(z) = (U − iV )z + (U + iV ) − log .
z 2π a
The corresponding stream pattern is shown figure 6.6.1 (see pp.424-425). The fluid velocity
is obtained by the differentiating the complex potential. For now (considering a circular
body), we can assume that the relative motion of the circle and the fluid is parallel to the
x-axis, i.e. V = 0,
dw a2 iκ
u − iv = = −U + U 2 − .
dz z 2πz
The fluid velocity at the circle bondary is obtained putting z = aeiθ :
κ
u + iv = ieiθ (2U sin θ + )
2πa
(we also made the conjugation u + iv = [u − iv]∗ ). So, the velocity is obtained from eiθ ×(a
real number) by π2 rotation (multiplication by i = eπ/2 ). There are two stagnation points
(when the velocity vector vanishes)
κ
sin θ = −
4πaU
42
on the lower part of the circle. These stagnation points on the boundary really exist if the
absolute value of the r.h.s. does not exceed 1.
the x-component of the force (parallel to the relative velocity) equals zero.
Signs of U and κ are discussed on p. 426. Pressure should be higher on the lower
part than on the top part of the circle. So, the velocity should be larger near the top and
smaller near the bottom. Therefore, the force is positive if the circle moves in the positive
direction (U > 0) with a counterclockwise circulation (κ > 0).
4 Boundary layers
5 Waves
6 Instabilities
7 Turbulence
43