8
2. VISCOSITY AND LAMINAR FLOW: BA-
SICS
Let’s start by looking at some simple fluid flow so-
lutions. We’ll consider steady-state, incompressible
flows. Some very simple solutions can be found when
viscosity is important to the flow. Because viscosous
stresses can, however, get very complicated, we’ll be-
gin with simple examples, then see the general stuff.
A. One-dimensional Laminar Flows
Think about a long, 1D channel such as in Figure 2.1.
The flow is driven by a pressure gradient, p2 < p1 ; and
we assume the velocity must go to zero at the walls (this
is a no-slip boundary). Because the flow is incompress-
ible – has the same density everywhere – the velocity
profile must be independent of x, in order to enable
mass conservation (the same flow rate across any point
within the channel).
y
dx
(p1)
2b dy
x
Figure 2.1. Defining the geometry for this problem. The
flow is driven by a pressure gradient; p2 < p1 . This driver,
combined with the no-slip condition assumed at the walls
and the action of viscous stresses within the flow, leads to
the quadratic velocity profile shown.
Now, think about a little differential volume dxdy,
as shown in the figure. It feels a net force to the right,
due to dp/dx; what balances this force to keep the flow
from accelerating? This must be viscosity: friction on
the little volume due to the shear in the flow. We as-
sume the shear stresses are linearly proportional to the
velocity gradient, transverse to the surface (this is called
a Newtonian flow). The little volume will feel friction
forces on its top and bottom surfaces; the net force on
the volume will be the difference between the forces
on the top and bottom. If any one surface feels a force
ρν(∂v/dy), the net force (per length in the z direction)
on our element is
" What does ∇2 a mean, if a is a vector? For Cartesian it’s simple:
 #
∂2v
  
∂v ∂v
ρν − ρν dx = ρν
∂y y+dy ∂y y dy 2
(2.1)
In this last I’ve assumed dy is small, of course, and also
that ν is independent of y. I’ve also written the constant
of proportionality as ρν, and the constant ν is what I
like to call the coefficient of viscosity; I’ll return to this
shosrtly.
Beware: you must be careful when working with
viscosity. The definitions, symbols and even units
of viscosity vary from author to author. I pre-
fer the choice, that ν∇v has the dimensions of a
force per area per gram: and that ρν∇v is a force
per area. Thus, the dimensions of ν are [L]2 /[T].
Many authors (including Faber) include the den-
sity; he uses η (dimensions [M]/[L][T]) which is
the same as our ρν. Still other authors use µ where
Faber uses η.
So: putting this back into vector form, we’ve discov-
ered that our governing equation for this simple system
should be1
∇p = ρν∇2 v
But now ... looking back to equation (1.10), our re-
sult here suggests that we should add a new term to de-
scribe the effects of viscous stresses. Taking this as true
for the moment (we’ll derive it more generally below),
we have the basic equation governing viscous flow:
v
(p2 ) ∂v
ρ + ρ(v · ∇)v + ∇p = ρg + ρν∇2 v (2.2)
∂t
Compare this back to our first version of the momentum
equation, namely (1.10), which we called Euler’s equa-
tion. This version of the force equation, with viscosity
included, is called the Navier-Stokes equation.
In the rest of this chapter, we look at smooth (time-
steady) flows in which viscosity plays the dominant
role. This situation is often called laminar flow. Faber
describes laminar flow in a more restricted sense, im-
plying that “the fluid can be treated as an assembly of
laminae of uniform thickness, whose boudaries remain
fixed as the flow moves between them”. The more
general, and more usual, definition, uses laminar flow
to mean smooth, non-turbulent flow. In most applica-
tions, we note that a real fluid does not slip freely past
a boundary (as was the implicit assumption in potential
flow theory), but rather sticks: we will generally use
no-slip boundary conditions.
1
∇2 a = ∇2 a x , ∇2 a y , ∇2 a z
` ´
dx
where
∂2 ∂2 ∂2
„ «
∇2 = , ,
∂x2 ∂y 2 ∂z 2
is the usual Laplacian operator. For cylindrical and spherical co-
ordinates, ∇2 a is a more complicated form – check the vector
references I’ll put up on the web, or your favorite vectors-in-
funny-coordinates reference book.
 9

B. Steady Flow: Cartesian Applications

In particular, in this section and the next, we consider
steady flows, where gravity is unimportant. Thus, we
are solving

ρ(v · ∇)v + ∇p = ρν∇2 v (2.3)

and the second derivative gives this equation the prop-

ertires of a diffusion equation.
1. FLOW BETWEEN PARALLEL PLATES

In this section, we consider flow along x̂, with gradients

in the ŷ direction v = (v(y), 0, 0). Thus, the system
(2.3) simplifies still further to Figure 2.2. Plane-flow solutions for various boundary
conditions. From Kundu figure 9.4.
∂p ∂2v
= ρν 2
∂x ∂y open surface allows us to assume the pressure is ev-
(2.4)
∂p erywhere equal to atmospheric pressure, so that ∇p is
=0
∂y zero. Keep the Cartesian coordinates aligned with the
flow bed. The basic equation is then
Note, we have eliminated the interial term utterly, by
picking our geometry carefully: so that the gradients d2 v
are perpendicular to the velocities. The second of these ρν = −ρg sin α (2.7)
dy 2
equations says that the pressure can vary only with x. In
the first, then, the first term can be only a function of x, The boundary conditions are now v = 0 at the lower
while the second can be only a function of y; thus, both surface, and dv/dy = 0 at the upper (free, open) sur-
must be constant. Thus, dp/dx = constant, (2.4) be- face, a distance b away. The latter is due to the fact
comes an ODE, and our general solution for v(y) comes that shear stresses are continuous at a boundary, and
from the atmosphere (having very low viscosity) is assumed
to carry no shear. This solves to
y 2 dp
ρνv(y) = Ay + B + (2.5)
2 dx 1
g sin α 2by − y 2


v(y) = (2.8)
Thus, the transverse velocity profile is a quadratic in y. 2ν
Specific examples come from various imposed bound- which gives a mass flux,
ary conditions, as illustrated in Figure 2.2.
Z b
One particular case is plane Poiseuille flow, in 1
Q= ρv(y)dy = ρgb3 sin α (2.9)
which an external pressure gradient drives a flow 0 3ν
through two stationary walls. This is what we intro-
duced in Figure 2.1. The velocity profile here is a cen- 3. HELE SHAW FLOW
tered parabola, the flow being fastest in the middle (as
you’ll see in the homework). The mass flux per unit This describes the means by which the striking two-
width is dimensional images used to illustrate potential flow are
Z b made. Inject the fluid through closely spaced parallel
1 dp plates, with some obstacle in the middle. We want the
Q= ρv(y)dy = − b3 (2.6) vertical direction (between the plates; call it ẑ) to be
0 3ν dx
governed by viscosity, while the two horizontal direc-
Other variants, shown in Figure 2.2, will appear in the tions (say the flow goes along ẑ) can be treated as po-
homework. tential flow. Let the plate separation be b, the scale of
the central obstacle be L, and the flow be driven by a
2. FLOW IN AN OPEN CHANNEL
fixed, externally imposed dp/dx.
This comes from Faber. Let the channel have an open First, use our parallel-plate solution to find the dis-
surface, and be inclined to horizontal at an angle α. The charge rate Q. From this define a characteristic velocity
 10

(strictly, the mean over the z-direction): get (2.5), we argue here that the second term in (2.14)
must be constant, giving a flow profile
Q b2 dp
V = =− (2.10) r 2 dp
b 12ρν dz v(r) = + A ln r + B (2.14)
4ρν dz
(the minus sign just notes that V points opposite to the
pressure gradient). But I can write the right hand side where A, B are again integration constants, set by the
as the gradient of a scalar: boundary conditions.
 2 
b dp
V =∇ (2.11)
12ρν dz

That is, the horizontal velocity can be treated in terms

of a potential, φ = −(b2 /12ρν)(dρ/dz). This shows
that all the potential flow apparatus of chapter 2 can
be used here; and, coversely, that flows where this
Figure 2.3. The circular Poiseuille flow solution,
works are good illustrations for two-dimensional poten- showing the velocity field and also the stress field (τ ; our Σ)
tial flow solutions. From Kundu figure 9.5.
But then: what does it take to justify this? We must
be able to ignore the viscous forces in the x and y direc- But now, A = 0 to keep v finite as r → 0; and a
tions, and to ignore the inertial force in the z direction. no-slip boundary at r = a gives
That is, we need
r 2 − a2 dp
∂vx ∂ 2 vx ∂ 2 vx ∂ 2 vx v(r) = (2.15)
vx ≪ν 2 ; ≪ 4ρν dz
∂x ∂z ∂y 2 ∂z 2
that is, another centered parabolic profile. The mass
But now: the second condition simply requires that flux here is
L ≫ b; while the first requires Z a
πa4 dp
V2 νV b2 1 Q= 2πvrdr = − (2.16)
≪ 2 ; ≪ (2.12) 0 8ν dz
L b L2 Re
This is called Poiseuille’s Law, or the Hagen-Poiseuille
Thus, the Reynolds number determines the required di- law.
mensions of the apparatus.
2. CIRCULATING FLOW
C. Steady Flow: Cylindrical Applications
Another case is flow between two concentric, rotating
We can repeat the exercise for cylindrical geometry; cylinders, called circular Couette flow.
differences from the planar cases highlight the effects
of geometry.
1. PIPE FLOW

Now, consider steady flow in a pipe of radius a. This

is circular Poiseuille flow. The radial component of the
momentum equation again required dp/dr = 0, ie no
transverse pressure gradients. Assuming a very long
pipe, so that things only vary with r, the z-component
of the momentum equation is
 
dp ρν d dv Figure 2.4. The geometry and the velocity field for
= r (2.13) circular Couette flow. From Kundu figure 9.6.
dz r dr dr

As there is only a z-component to the velocity, we have Referring to (2.41) and (2.42) in Appendix 2, and not-
dropped the subscript. By the same argument as used to ing that only vφ 6= 0, the two components of the equa-
 11

tion of motion are the shear force points along the interface between the
vφ2 two fluid parcels). The coefficient ν is the viscosity or
1 dp
= shear viscosity.
r ρν dr
  (2.17)
d 1 d
0=ν (rvφ )
dr r dr
The general solution for the azimuthal velocity is
B
vφ (r) = Ar + (2.18)
r
and the pressure gradient, required to offset centrifugal
force, is
B 2
 
dp ρ
= Ar + (2.19)
dr r r
Using boundary conditions vφ (R1 ) = Ω1 R1 and
vφ (R2 ) = Ω2 R2 gives the specific values for A and
B.
This has some interesing limits. One has Ω = Ω2 Figure 2.5. The two-dimensional geometry for the stress
and R1 = 0, that is a rotating cylindrical tank. This has tensor. pi are the normal forces on the fluid square, and sij
are the shear forces. From Faber, Figure 1.2
vφ = Ωr (2.20)
showing that the fluid inside goes into solid body ro- Life is a bit more complicated, however. Consider
tation. Alternatively, if the outer cylinder is at infin- a two-dimensional Cartesian system, where s12 is the
ity, with Ω2 = 0, and the inner one has R = R1 and force in direction 2, due to adjacent fluid in direction 1;
Ω = Ω1 , then the solution is and s21 is the force in direction 1, due to adjacent fluid
in direction 2. These two forces must be equal: s12 =
ΩR2 s21 ; if not we would have a net torque on the parcel of
vφ = (2.21)
r fluid. Thus, we must symmetrize our expression for the
We’ll see this again in chapter 4 when we work with the shear force:
 
irrotational vortex. ∂v1 ∂v2
s12 = s21 = ρν + = s3 (2.22)
∂x2 ∂x1
D. Viscous Stresses, Generally
where the last equality simply defines s3 , to shorten the
OK, we’ve avoided this long enough .. we need to write notation.
down the more general form of the viscous stresses. There are also normal forces on the parcel: due to
This will be a tensor. We start by determining the sur- the fluid pressure, and also due to the fluid deformation.
face forces on a piece of fluid, due to deformations Faber approaches this by considering a frame rotation,
(arising from velocity gradients) of the adjacent bits of into the 45◦ degree primed frame. From force balance,
fluid. we find that
1. DO IT PHYSICALLY FIRST 1
s′3 = (p1 − p2 )
2 (2.23)
Following Faber here. Consider a planar flow in the x̂ 1
′
direction, with transverse velocity gradients (along ŷ). p1 = (p1 + p2 − 2s3 )
2
We expect the shear stress to be linear: ν x̂dvx /dy is the
force between adjacent streams in the fluid.2 (Note that and also, from a Taylor expansion, we find that
 ′ 
∂v1 ∂vs′
s3 = ρν −
2
This is a BIG assumption – its traditional, but not obvious to ∂x′1 ∂x′2
me that it should always hold. Look ahead to the effects of a from which we get
magnetic field and j × B forces, for a possible counter-example.
Luckily, however, we rarely need to worry about viscous stresses ∂v1 ∂v2 ∂v3
in MHD applications ....
p1 + 2ρν = p2 + 2ρν = p3 + 2ρν (2.24)
∂x1 ∂x2 ∂x3
 12

But now, taking the mean (isotropic) pressure to be Next, write a general velocity gradient in terms of sym-
metric (Dij ) and antisymmetric (Ωij ) parts:
1
p= (p1 + p2 + p3 ) (2.25) ∂vi
3 = Dij + Ωij
∂xj
   
we end up with an expression for the normal force: 1 ∂vi ∂vj 1 ∂vi ∂vj
= + + −
2 ∂xj ∂xj 2 ∂xj ∂xj
 
2 ∂v1 ∂v2 ∂v3 (2.29)
p1 = p − ρν 2 − − (2.26)
3 ∂x1 ∂x2 ∂x3
and we note that the Ωij term is 1/2 of the usual vor-
So, generalizing to three dimensions, we have all 9 ticity, Ω = ∇ × v. Also, note that the trace of Dij
components (6 independent) of the surface forces act- is
ing on this parcel, which we collect as the tensor ~~σ : ∂v1 ∂v2 ∂v3
Dmm = + + =∇·v (2.30)
  ∂x1 ∂x2 ∂x3
p s s Now, we argue that Σij should depend only on the sym-
 1 12 13 
σij = − s21 p2 s23 
 
(2.27) metric part of the velocity gradient, Dij (otherwise,
  simple rotation, such as solid body, would lead to a spu-
s31 s32 p3 rious stress term). Thus, requiring linearity, we have in
general
(the minus sign has been added for consistency with the
Σij = αijmn Dmn (2.31)
usual treatment).
which involves no fewer than 81 separate α’s. However,
we can make arguments about symmetry and isotropy
to reduce this to the most general form useful for us:
 
1
Σij = 2ρν Dij − δij Dmm + ρνb δij Dmm (2.32)
3

That is, we have only two α’s – ν, the shear viscosity,

usually just called the viscosity, and νb , the bulk viscos-
ity. Thus, the general stress tensor is
 
1
σij = −pδij + 2ρν Dij − δij Dmm + ρνb δij Dmm
3
(2.33)
From this, we have the full expression for the vector
surface force per area:
Figure 2.6. The surface forces on a three-dimensional
fluid cube. The notation here uses τ for our σ. The normal T = eˆk σik (2.34)
forces are τii , and the shear forces are τij . From Kundu
figure 2.2. if eˆk is the unit vector in the kth direction.
In many applications the bulk viscosity is ignored.
This is called the Stokes hypothesis; Thompson dis-
2. THEN DO IT FORMALLY cusses it nicely. This hypothesis is physically true
(at the kinetic level) only for simple systems – dilute,
We can also approach this more formally (following monatomic gases. For other fluids, νb ∼ > ν is possi-
Thompson). First, we want to write the surface forces ble. However, in many applications the flows are in-
as a general stress tensor, compressible, and (from 2.30 or 2.33) the effects of the
bulk viscosity term can be ignored.
σij = −pδij + Σij (2.28)
 13

E. The Navier-Stokes Equation (in Cartesian)

Now, we want to add viscous stresses to the force equation. Remember that we’re still working in Cartesian. We
need to generalize the ∇p term in the Euler equation (e.g., 1.10) to the full stress tensor. With the stress term added,
the force equation is called the Navier-Stokes equation. Recall, g = −∇Φ is the body force (derivable from a
potential). In Cartesian, in index notation, we have the form
 
∂vi ∂vi ∂σki
ρ + vj = ρgi + (2.35)
∂t ∂xj ∂xk
and in vector notation, we have the form
 
∂v 2 1
ρ + ρ(v · ∇)v + ∇p = ρg + ρν∇ v + ρνb + ρν ∇(∇ · v) (2.36)
∂t 3
If Stokes’ hypothesis holds, this simplifies a bit to
∂v 1
ρ + ρ(v · ∇)v + ∇p = ρg + ρν∇2 v + ρν∇(∇ · v) (2.37)
∂t 3
which is still general (allows variation of ρ). If we now assume the fluid is incompressible, ∇ · v = 0, (2.37)
becomes
∂v
ρ + ρ(v · ∇)v + ∇p = −ρ∇Φ + ρν∇2 v (2.38)
∂t
which is the most commonly used form of this important equation, and recovers our “guess” in (2.2).

References
I’ve mostly followed Kundu and Faber here, as well as Thompson (whose discussion of viscosity and stress
tensors I like). For the non-Cartesian forms of tne NS equation, Pozrikidis is one good reference.

F. Appendix: Navier-Stokes in other coordinates

It is also worth storing the stress tensor and the N-S equation in curvilinear coordinates (taken from Pozrikidis).
Here, we are using only the shear viscosity, and dropping the bulk viscosity term.
1. C YLINDRICAL COORDINATES

The 6 independent components of the stress tensor are

 
∂vr ∂  vφ  2 ∂vr
σrr = −p + 2ρν σrφ = ρν r +
∂r ∂r r r ∂φ
   
∂vr ∂vz 2 ∂vφ 2
σrz = ρν + σφφ = −p + ρν + vr
∂z ∂r r ∂φ r
 
∂vφ 1 ∂vz ∂vz
σφz = ρν + σzz = −p + 2ρν (2.39)
∂z r ∂φ ∂z
 14

and the 3 componts of the NS equation are:

2
vφ ∂vφ vφ
 
∂vr ∂vr ∂vr 1 ∂σzr 1 ∂(rσrr ) 1 ∂σrφ σφφ
+ vz + vr + − = gr + + + − (2.40)
∂t ∂z dr r dφ r ρ ∂z r ∂r r ∂φ r

1 ∂(r 2 σrφ ) 1 ∂σφφ

 
∂vφ ∂vφ ∂vφ vφ ∂vφ vr vφ 1 ∂σzφ
+ vz + vr + + = gφ + + 2 + (2.41)
∂t ∂z ∂r r ∂φ r ρ ∂z r ∂r r ∂φ
 
∂vz ∂vz ∂vz vφ ∂vz 1 ∂σzz 1 ∂(rσrz ) 1 ∂σφz
+ vz + vr + = gz + + + (2.42)
∂t ∂z ∂r r ∂φ ρ ∂z r ∂r r ∂φ
Alternatively, we can write the NS equations out explicitly for the case of constant viscosity:
2
∂vr ∂vr ∂vr vφ ∂vφ vφ
+ vz + vr + −
∂t ∂z dr r dφ r
 2 (2.43)
1 ∂ 2 vr
  
1 ∂p ∂ vr ∂ 1 ∂(rvr ) 2 ∂vφ
=− + gr + ν + + 2 − 2
ρ ∂r ∂z 2 ∂r r ∂r r ∂φ2 r ∂φ

∂vφ ∂vφ ∂vφ vφ ∂vφ vr vφ

+ vz + vr + +
∂t ∂z ∂r r ∂φ r
 2 (2.44)
1 ∂ 2 vφ 2 ∂vr
  
1 ∂p ∂ vφ ∂ 1 ∂(rvφ )
=− + gφ + ν + + 2 +
ρr ∂φ ∂z 2 ∂r r ∂r r ∂φ2 r ∂φ

∂vz ∂vz ∂vz vφ ∂vz

+ vz + vr +
∂t ∂z ∂r r ∂φ
 2 (2.45)
1 ∂ 2 vz
  
1 ∂p ∂ vz 1 ∂ ∂vz
=− + gz + ν + r + 2
ρ ∂z ∂z 2 r ∂r ∂r r ∂φ2

2. S PHERICAL P OLAR C OORDINATES

The 6 independent components of the stress tensor are
 
∂vr ∂  vθ  1 ∂vr
σrr = −p + 2ρν σrθ = ρν r +
∂r ∂r r r ∂θ
   
1 ∂vr ∂  vφ  ρν ∂vφ
σrφ = ρν +r σφφ = −p + 2 + vr sin θ + vθ cos θ
r sin θ ∂φ ∂r r r sin θ ∂φ
   
sin θ ∂  vφ  1 ∂vφ 2 ∂vθ 2vθ
σθφ = ρν + σθθ = −p + ρν + (2.46)
r ∂θ sin θ r sin θ ∂φ r ∂θ r
and the 3 componts of the NS equation are:

∂vr ∂vr vθ ∂vr vφ ∂vr vθ2 + vφ2

+vr + + −
∂t ∂r r ∂θ r sin θ ∂φ r
 2  (2.47)
1 1 ∂(r σrr ) 1 ∂(σθr sin θ) 1 ∂σφr σθθ + σφφ
+gr + + + −
ρ r2 ∂r r sin θ ∂θ r sin θ ∂φ r

∂vθ ∂vθ vθ ∂vθ vφ ∂vθ vr vθ cot θ

+vr + + + − vφ2
∂t ∂r r ∂θ r sin θ ∂φ r r
 2  (2.48)
1 1 ∂(r σrθ ) 1 ∂(σθθ sin θ) 1 ∂σφθ σrθ σφφ
+gθ + + + + − cot θ
ρ r2 ∂r r sin θ ∂θ r sin θ ∂φ r r
 15

∂vφ ∂vφ vθ ∂vφ vφ ∂vφ vφ

+vr + + + (vr + vθ cot θ)
∂t ∂r r ∂θ r sin θ ∂φ r
(2.49)
1 1 ∂(r 2 σrφ ) σrφ 1 ∂σθφ 2 cos θ
 
1 ∂σφφ
= gφ + + + + σθφ +
ρ r2 ∂r r r ∂θ r r sin θ ∂φ

Once again, we can write these out explicitly:

vθ + vφ 2 2
∂vr ∂vr vθ ∂vr vφ ∂vr
+vr + + −
∂t ∂r r ∂θ r sin θ ∂φ r
    (2.50)
1 ∂p 2 2vr 2 ∂vθ 2 ∂vφ
=− + gr + ν ∇ vr − 2 − 2 + vθ cot θ − 2
ρ ∂r r r ∂θ r sin θ ∂φ

∂vθ ∂vθ vθ ∂vθ vφ ∂vθ vr vθ cot θ

+vr + + + − vφ2
∂t ∂r r ∂θ r sin θ ∂φ r r
  (2.51)
1 ∂p 2 2 ∂v r vθ 2 cos θ ∂vφ
=− + gθ + ν ∇ vθ + 2 − 2 2 − 2 2
ρr ∂θ r ∂θ r sin θ r sin θ ∂φ

∂vφ ∂vφ vθ ∂vφ vφ ∂vφ vφ

+vr + + + (vr + vθ cot θ)
∂t ∂r r ∂θ r sin θ ∂φ r
  (2.52)
1 ∂p vφ 2 ∂vr 2 cos θ ∂vθ
=− + gφ + ν ∇2 vφ − 2 2 + 2 + 2 2
ρr sin θ ∂φ r sin θ r sin θ ∂φ r sin θ ∂φ
In this set, for f some scalar function,

∂2f
   
1 ∂ ∂f 1 ∂ ∂f 1
∇2 f = r2 + sin θ + 2
r 2 ∂r ∂r r 2 sin θ ∂θ ∂θ r 2 sin θ ∂φ2

and this is the end of this chapter.