Derivation of the basic equations of fluid flows.

No
particle in the fluid at this stage (next week).
week)

•Conservation of mass of the fluid.

•Conservation of mass of a solute (applies to non-sinking

particles at low concentration).

•Conservation of momentum.

•Application of these basic equations to a turbulent fluid.

fluid
A few concepts before we get to the meat…

Tensor (Stress), Vectors (e.g. position, velocity) and

scalars (e.g. T, S, CO2).

We need to define a coordinate system, and an

((infinitesimal)) element of volume.

We assume a continuous fluid, and

that all the fields of interest are
differentiable.
differentiable
The Lagrangian framework is the framework in which the
laws of classical mechanics are often statedstated. The
v
coordinates of a point x (t ) describe the trajectory from
v v
x0 = x (t = 0) . Density, ρ, can evolve along the trajectory.
By the chain rule,rule along a parcel trajectory:
v
dρ dx ⎛ ∂ρ ⎞ ⎛ ∂ρ ⎞ v ∂ρ
= ⎜ v⎟ +⎜ ⎟ = u ⋅ ∇ρ +
dt dt ⎝ ∂x ⎠t =const ⎝ ∂t ⎠ xv =const ∂t
v
dρ ( x (t ), t ) ⎛ ∂ v ⎞ v
→ = ⎜ + u ⋅ ∇ ⎟ ρ (x , t )
dt ⎝ ∂t ⎠
D ∂ v
≡ + u ⋅ ∇ ÅConversion from Lagrangian to Eulerian
Dt ∂t
∂t
Example:
Let’s assume that we are in a river that feeds on glacial
melt. The water warms at a constant rate that is a
function of distance from the source. If we drift down
river (A la ‘Huckleberry Fin’), the temperature increases
with time (DT/Dt>0). At one point along the river,
however we may see no change in temperature with time
however,
(∂T/∂t=0), as the water arriving there is always at the
same temperature. The heat flux is advective, (u∂T/∂x>0).

In short, the convective derivative is:

D ∂ ∂ ∂ ∂ ∂ v
= + u + v + w = + u ⋅∇
Dt ∂t ∂x ∂y ∂z ∂t
Mass conservation (Eulerian, differential approach):
Accountingg for the change
g in mass inside a fixed,, constant-
size volume: (x-Δx/2,y+Δy/2,z+Δz/2)
(x+Δx/2,y+Δy/2,z+Δz/2)

Az =ΔxΔy
(x-Δx/2,y-Δy/2,z+Δz/2)
Ay=ΔxΔz
z Δy Δy
= Δz = Δ z
Ax Ax
A Δ Δ
Ay=ΔxΔz
y (x+Δx/2,y+Δy/2,z-Δz/2)

x
Az =ΔxΔy
Mass=ρV
Mass (x+Δx/2,y-Δy/2,z-Δz/2)
(x-Δx/2,y-Δy/2,z-Δz/2)
∂ ( ρV )
∂t
( ) ( ) (
= − Ax ρu + Δx / 2 − ρu − Δx / 2 − Ay ρv + Δy / 2 − ρv − Δy / 2 − Az ρw + Δz / 2 − ρw − Δz / 2 )
∂ρ
∂t
=−
1
Δx
(
ρu + Δx / 2 − ρu − Δx / 2 −
1
Δy
) (
ρv + Δy / 2 − ρv − Δy / 2 −
1
Δz
)
ρw + Δz / 2 − ρw − Δz / 2 ( )
∂ρ ∂ ∂ ∂ ∂ρ v
+ (ρu ) + (ρv ) + (ρw) = + ∇ ⋅ (ρu ) = 0
∂t ∂x ∂y ∂z ∂t
 Mass conservation (Eulerian, integral approach):
Accounting for the change in mass inside a fixed,
fixed
constant-volume volume (V0):

d v v
∫
dt V0
ρdV = − ∫ ρu ⋅ n dS
∂V0

∂ v
→ ∫ ρdV = − ∫ ∇ ⋅ ρudV
V0
∂t V0

∂ v
→ ρ + ∇ ⋅ ρu = 0
∂t
Where we used the divergence
g theorem:

It states that the volume total of all sinks and sources, the volume integral of
the divergence, is equal to the net flow across the volume's boundary (WIKI).
Reiteration (no sinks/sources):

Mass conservation (Lagrangian, integral):

D
∫
Dt V
ρdV = 0

Mass conservation (Eulerian,

(E l integral):
l)

d v v
∫
dt V0
ρdV = − ∫ ρu ⋅ ndS
∂V0
Mass conservation:
Note that:

∂ρ r
+ ∇ ⋅ ( ρu ) = 0
∂t
∂t
Can be written as:

1Dρ r
+ ∇ ⋅u = 0
ρDt
D

The 2nd term is the fluid divergence (rate of outflow of volume per unit
volume). This can be nonzero only for compressible fluids. It is the rate of
loss of density due to compression/expansion.
r
For both
F b th water
t and d air
i we can assume ththatt ∇ ⋅ u = 0 iin terms
t of
f their
th i
dynamics (we need compressibility to pass sound…).
Mass balance for conserved scalar:
Adding
ng mo
molecular
cu ar diffusion:
ffus on

∂C r r
∫V ∂t dV = −∫S (Cu − K∇C )⋅ ndS
Where V is the volume of the control volume and S its surface, and
using Fick’s law. By the help of the divergence theorem:

⎡ ∂∂C
C r ⎤
∫V ⎢⎣ ∂t + ∇ ⋅ (Cu − K∇C )⎥⎦dV = 0
Since the volume is arbitrary, this can be true if and only if:

∂C r
+ ∇ ⋅ (Cu ) = ∇ ⋅ (K∇C )
∂t
Momentum balance (Navier-Stokes):
Newton’s 2nd law of motion states that the time rate of changeg of
momentum of a particle is equal to the force acting on it. This law is
Lagrangian, the “time rate of change” is with respect to a reference system
following the particle.

d v v v
∫ ρudV = ∫ ρ gdV + ∫ T dS
dt V (t ) V (t ) ∂V ( t )

Where g is the body force per unit mass (e.g. gravity) and T is the surface
force per unit surface area bounding V.
If
f the
h volume
l is small
ll enough,
h the
h integrands
d can be
b taken
k out of f the
h integral:
l

d v d ⎛⎜ v ⎞ d v
⎟ = (ρuδV )
dt V∫(t ) dt ⎜⎝ V∫(t ) ⎠⎟ dt
ρ u dV = ρu dV

v v
d v du v d (ρδδV ) du
(ρuδV ) = ρδV + u = ρδV
dt dt dt dt
Momentum balance (Navier-Stokes):
Th body
The b d force
f is
i similarly
i il l treated:
t t d

v v
∫ ρ gdV = ρgδV
V (t )

Defining a stress tensor (expanded on the next slide):

v v
T = Τ⋅n
A d applying
And l i ththe divergence
di th
theorem:
v
∫ T dS =
∂V ( t )
∫ ∇ ⋅ ΤdV = ∇ ⋅ ΤδV
V (t )

v
Du v
→ρ = ρg + ∇ ⋅ Τ
Dt
Surface forcing:
For an inviscid fluid,, the surface force
v exerted by y the surrounding
g
v
fluid is normal to the surface, i.e. T = − p ⋅ n , and p is called the
pressure force.

In general,v viscous stress force, S,v is also present. For viscous

v v v
fluids: T = − p ⋅ n + S . By
v definition T = Τ ⋅ n , and we now have
v
T = − pI + ∑ , where S = ∑ ⋅n and I is the identity tensor.

For Newtonian fluids,

v
∇ ⋅ Σ = μ∇ u 2

And the resultant Navier

Navier-Stokes
Stokes equations for incompressible fluids are:
v
Du v 2v
ρ = ρg − ∇p + μ∇ u
Dt
Rotational symmetry:

⎛τ xx τ xy τ xz ⎞ ⎛τ xx τ xy τ xz ⎞
⎜ ⎟ ⎜ ⎟
T = ⎜τ yx τ yy τ yz ⎟ = ⎜τ xy τ yy τ yz ⎟
⎜τ τ τ ⎟ ⎜τ τ τ ⎟
⎝ zx zy zz ⎠ ⎝ xz yz zz ⎠

− p ≡ (τ xx + τ yy + τ zz )
1
3
Total stress tensor, Newtonian fluid:
⎛ ⎛ ∂u ∂v ⎞ ⎛ ∂u ∂w ⎞ ⎞⎟
⎜ − p + 2 μ ∂u μ ⎜⎜ + ⎟⎟ μ ⎜ + ⎟ ∑ ij
⎜ ∂x ⎝ ∂y ∂x ⎠ ⎝ ∂z ∂x ⎠ ⎟
⎜ ⎛ ∂u ∂v ⎞ ∂v ⎛ ∂v ∂w ⎞ ⎟ ⎛ ∂u j ∂ui ⎞
T = ⎜ μ ⎜⎜ + ⎟⎟ − p + 2μ μ ⎜⎜ + ⎟⎟ ⎟ = − pδ ij + μ ⎜⎜ + ⎟
⎟
⎜ ⎝ ∂y ∂x ⎠ ∂y ⎝ ∂z ∂y ⎠ ⎟ ⎝ ∂xi ∂x j ⎠
⎜ ⎛ ∂u ∂w ⎞ ⎛ ∂v ∂w ⎞ ∂w ⎟
⎜ μ⎜ + μ ⎜⎜ + ⎟⎟ − p + 2 μ ⎟
⎜ ⎝ ∂z ∂x ⎟⎠ ⎝ ∂z ∂y ⎠ ∂z ⎠⎟ i,, j={1,
j { , 2,, 3}={x,
} { , y, z}}
⎝
Stokes, 1845:
1 Σij linear function of velocity gradients.
1. gradients
2. Σij should vanish if there is no deformation of fluid elements.
3. Relationship between stress and shear should be isotropic.

∂T ∂T ∂T ∂p ∂ 2
u ∂ 2
u ∂ 2
v
ˆi ⋅ (∇ ⋅ T ) = xx + yx + zx = − + 2 μ +μ 2 +μ
∂x ∂y ∂z ∂x ∂x 2
∂y ∂y∂x
v
∂u2
∂ w
2
∂p ∂ (∇ ⋅ u ) ∂p
+μ 2 +μ = − + μ∇ u +
2
= − + μ∇ 2u
∂z ∂z∂x ∂x ∂x ∂x
 Navier-Stokes equations:
v
Du ∇p μ 2 v v v
=− + ∇ u + g, ∇ ⋅ u = 0
Dt ρ ρ
∂u ∂u ∂u ∂u 1 ∂p μ ⎛ ∂ 2u ∂ 2u ∂ 2u ⎞
+u +v +w =− + ⎜⎜ 2 + 2 + 2 ⎟⎟
∂t ∂x ∂y ∂z ρ ∂x ρ ⎝ ∂x ∂y ∂z ⎠
∂v ∂v ∂v ∂v 1 ∂p μ ⎛ ∂ 2 v ∂ 2 v ∂ 2 v ⎞
+u +v +w = − + ⎜⎜ 2 + 2 + 2 ⎟⎟
∂t ∂x ∂y ∂z ρ ∂y ρ ⎝ ∂x ∂y ∂z ⎠
∂w ∂w ∂w ∂w 1 ∂p μ ⎛ ∂ 2 w ∂ 2 w ∂ 2 w ⎞
+u +v +w =− + ⎜⎜ 2 + 2 + 2 ⎟⎟ + g
∂t ∂x ∂y ∂z ρ ∂z ρ ⎝ ∂x ∂y ∂z ⎠
∂u ∂v ∂w
+ + =0
∂x ∂y ∂z
Coriolis is added when moving the framework to an accelerating
framework. Have to add boundary & initial conditions.
Navier-Stokes equations (Boussinessq approximation):
Separate balance of fluid at rest from moving fluid.

p = p0 ( z ) + p ' ( x , y , z , t )
ρ = ρ 0 + ρ ' ( x, y , z , t )
First order balance (hydrostatic):

1 ∂p0
0=− +g
ρ 0 ∂z
2 d order
2nd d b balance:
l :
v
Du 2v v
ρ0 ∇p '+ μ∇ u + gρ 0 , ∇ ⋅ u = 0
= −∇
Dt
 Example: steady flow under gravity down an inclined plane.

∂u ∂w From: Acheson
Acheson, 1990
+ =0
∂x ∂z
1 ∂p ∂u2
0=− +ν 2 + g sin α
ρ 0 ∂x ∂z
BCs: z = 0 : w = u = 0
1 ∂p
0=− + g cos α ∂u
ρ 0 ∂z z = h: μ = 0, p = pa
∂z
Example: steady flow under gravity down an inclined plane.

Solution:

p − pa = ρ 0 g (h − z ) cos α
g
u= z (2h − z )sin α
2ν
Q: what ν should we use?
 Reynolds decomposition of the N-S equations

Assume a turbulent flow. At any given point in space we

separate the mean flow (mean can be in time, space, or
ensemble) and deviation from the mean such that:

p = p + p ' , u = u + u ' , u = v + v' , u = w + w'

p = p, p ' = 0, etc'
Substituting into the continuity equation (linear):

∂u ∂v ∂w ∂u ' ∂v' ∂w'

+ + = 0; + + =0
∂x ∂y ∂z ∂x ∂y ∂z
Substituting into x-momentum Navier-Stokes equation:

∂u ∂u ∂u ∂u 1 ∂p μ ⎛ ∂ 2u ∂ 2u ∂ 2u ⎞
+u +v +w =− + ⎜⎜ 2 + 2 + 2 ⎟⎟
∂t ∂x ∂y ∂z ρ ∂x ρ ⎝ ∂x ∂y ∂z ⎠
⎛ ∂ (u ' u ') ∂ (v' u ') ∂ (w' u ') ⎞
− ⎜⎜ + + ⎟
⎟
⎝ ∂x ∂y ∂z ⎠
The evolution of the mean is forced by correlations of fluctuating properties.
The correlation terms time density are the “Reynolds stresses”.
Substituting into a scalar conservation equation
equation:

∂C ∂C ∂C ∂C K ⎛ ∂ 2C ∂ 2C ∂ 2C ⎞
+u +v +w = ⎜⎜ 2 + 2 + 2 ⎟⎟
∂t ∂x ∂y ∂z ρ ⎝ ∂x ∂y ∂z ⎠
⎛ ∂ (u ' C ') ∂ (v' C ') ∂ (w' C ') ⎞
− ⎜⎜ + + ⎟
⎟
⎝ ∂ x ∂ y ∂z ⎠
Note that the Reynolds stress tensor is symmetric (as is
the viscous stress tensor):

⎛ u ' u ' u ' v' u ' w' ⎞

⎜ ⎟
τ = − ρ 0 ⎜ u ' v' v' v' v' w' ⎟
⎜⎜ ⎟⎟
⎝ u ' w' v ' w' w ' w' ⎠
The closure problem: to develop equations for the evolution of the Reynolds
stresses theselves, higher order correlation are needed (e.g. w’u’u’) and so
on. For this reason theories have been devised to describe τij in terms of
th mean flow.
the fl

For more, see: http://www.cfd-

online.com/Wiki/Introduction
onl ne.com/W k /Introduct on_to_turbulence/Reynolds_averaged_equat
to turbulence/Reynolds averaged equations
ons
One solution to the closure problem is to link the Reynolds’ stress to mean-flow
Quantities. For example:

∂U
− ρ 0 w' u ' = K eddy
dd
∂z
∂T
− ρ 0 w' T ' = K eddy
∂z
This type of formulation is appealing because it:
a. Provide for down-gradient flux.
b Is reminiscent of
b. f molecular
l l diffusion
d ff and
d viscosity.
c. Provide closure to the equations of the the mean fields.

This type of formulation is problematic because:

a. Keddy is a property of the flow and not the fluid.
b. Keddy is likely to vary with orientation, unlike molecular processes.

How is Keddy related to the turbulence?

Assume a gradient in a mean property (momentum, heat, solute, etc’.
Remember: no mean gradientÆno flux). Assume a fluctuating velocity field:

l’ is the distance a parcel travels before it loses its identity. The rate of upward
vertical turbulent transfer of <Ψ> is down the mean gradient:

⎛ ∂ψ ⎞ ∂ψ ∂ψ
w' ⎜ψ + l ' ⎟ = w' l ' = − K eddy
⎝ ∂z ⎠ ∂z ∂z
How is Keddy related to the turbulence?

∂U
− ρ 0 w' u ' = ρ 0 K eddy
∂z
∂T
− ρ 0 w' T ' = ρ 0γ T
∂z

Tennekes
T k and d Lumley
L l (1972) approach h thi
this problem
bl f
from di
dimensional
i l analysis
l i
based on assuming a single length scale-l and a single velocity scale ω=<w’2>1/2.

− ρ 0 w' u ' = cρ 0ω ; c ~ O(1)

2

The eddies involved in momentum transfer have vorticities,

vorticities ω/l; this
vorticity is maintained by the mean shear (l is the length scales of the
eddies, e.g. the decorrelation scale).

∂U
ω /l = c ; c ~ O(1)
∂z
It follows that:

∂U
K e ddy ~ lω ~ l 2
dd
∂z

In analogy with momentum flux, for heat we have:

∂T
− ρ 0 c p w' T ' = ρ 0 c pγ T
∂z
It is most commonly assumed, and verified that γT=Keddy.
 Eddy-diffusion: perspective from a dye patch (figures from lecture notes of
Bill Young, UCSD)

Dye patch << dominant scale of eddies. Dashed circle denotes initial position of patch.
Dye patch ≈ dominant scale of eddies

Dye patch >> dominant scale of eddies

Cheat sheet:
1 G
1. Gradient
di of
f a scalar
l ((a vector):
)

∂φ ˆ ∂φ ˆ ∂φ
∇φ = i
ˆ +j +k
∂x ∂y ∂z
2. Divergence of a vector (a scalar):
v ∂ϕ x ∂ϕ y ∂ϕ z
∇ ⋅ϕ = + +
∂x ∂y ∂z
3 Divergence of a Tensor,
3. Tensor T12=T(face)(direction):

⎛ ∂Txx ∂Tyx ∂Tzx ⎞ ∂T ∂T ∂T

ˆj ⎛⎜ xy + yy + zy ⎞⎟ +
∇ ⋅ T = i ⎜⎜
ˆ + + ⎟⎟ + ⎜ ∂x ⎟
⎝ ∂x ∂y ∂z ⎠ ⎝ ∂y ∂z ⎠
⎛ ∂Txz ∂Tyz ∂Tzz ⎞
k ⎜⎜
ˆ + + ⎟⎟
⎝ ∂x ∂y ∂z ⎠
Cheat sheet (cotinued):
4 Laplacian
4. L l i of f a vector (a
( vector):
)

2v v ⎛ ∂ 2
ϕ ∂ 2
ϕ ∂ 2
ϕx ⎞
∇ ϕ = ∇ ⋅ (∇ ⋅ ϕ ) = i ⎜⎜ 2 + 2 + 2 ⎟⎟ +
ˆ x x

⎝ ∂x ∂y ∂z ⎠
⎛ ∂ 2ϕ y ∂ 2ϕ y ∂ 2ϕ y ⎞ ⎛ ∂ 2ϕ z ∂ 2ϕ z ∂ 2ϕ z ⎞
ˆj ⎜ + + ⎟ + ˆ⎜
k + 2 + 2 ⎟⎟
⎜ ∂x ∂ ∂ ⎟ ⎜
y z ⎠ ⎝ ∂x ∂y ∂z ⎠
2 2 2 2
⎝