IB Fluid Dynamics (Full)

Part IB — Fluid Dynamics

Based on lectures by P. F. Linden

Notes taken by Dexter Chua

Lent 2016

These notes are not endorsed by the lecturers, and I have modified them (often

significantly) after lectures. They are nowhere near accurate representations of what

was actually lectured, and in particular, all errors are almost surely mine.

Parallel viscous flow

Plane Couette flow, dynamic viscosity. Momentum equation and boundary conditions.

Steady flows including Poiseuille flow in a channel. Unsteady flows, kinematic viscosity,

brief description of viscous boundary layers (skin depth). [3]

Kinematics

Material time derivative. Conservation of mass and the kinematic boundary condition.

Incompressibility; streamfunction for two-dimensional flow. Streamlines and path

lines. [2]

Dynamics

Statement of Navier-Stokes momentum equation. Reynolds number. Stagnation-

p oint flow; discussion of viscous boundary layer and pressure field. Conservation of

momentum; Euler momentum equation. Bernoulli’s equation.

Vorticity, vorticity equation, vortex line stretching, irrotational flow remains irrotational.

[4]

Potential flows

Velocity potential; Laplace’s equation, examples of solutions in spherical and cylindrical

geometry by separation of variables. Translating sphere. Lift on a cylinder with

circulation.

Expression for pressure in time-dependent potential flows with potential forces. Oscil-

lations in a manometer and of a bubble. [3]

Geophysical flows

Linear water waves: dispersion relation, deep and shallow water, standing waves in a

container, Rayleigh-Taylor instability.

Euler equations in a rotating frame. Steady geostrophic flow, pressure as streamfunction.

Motion in a shallow layer, hydrostatic assumption, modified continuity equation.

Conservation of potential vorticity, Rossby radius of deformation. [4]

Contents

0 Introduction

1 Parallel viscous flow

1.0 Preliminaries

1.1 Stress

1.2 Steady parallel viscous flow

1.3 Derived properties of a flow

1.4 More examples

2 Kinematics

2.1 Material time derivative

2.2 Conservation of mass

2.3 Kinematic boundary conditions

2.4 Streamfunction for incompressible flow

3 Dynamics

3.1 Navier-Stokes equations

3.2 Pressure

3.3 Reynolds number

3.4 A case study: stagnation point flow (u = 0)

3.5 Momentum equation for inviscid (ν = 0) incompressible fluid

3.6 Linear flows

3.7 Vorticity equation

4 Inviscid irrotational flow

4.1 Three-dimensional potential flows

4.2 Potential flow in two dimensions

4.3 Time dependent potential flows

5 Water waves

5.1 Dimensional analysis

5.2 Equation and boundary conditions

5.3 Two-dimensional waves (straight crested waves)

5.4 Group velocity

5.5 Rayleigh-Taylor instability

6 Fluid dynamics on a rotating frame

6.1 Equations of motion in a rotating frame

6.2 Shallow water equations

6.3 Geostrophic balance

0 Introduction

In real life, we encounter a lot of fluids. For example, there is air and water. These

are known as Newtonian fluids, whose dynamics follow relatively simple equations.

This is fundamentally because they have simple composition — they are made

up of simple molecules. An example of non-Newtonian fluid is shampoo, which,

despite looking innocent, has long chain molecules with complex properties.

In fluid dynamics, one of the most important things is to distinguish what is

a fluid and what is not. For example, air and water are fluids, while the wall is

solid. A main distinction is that we can lean on the wall, but not on air and

water. In particular, if we apply a force on the wall, it will deform a bit, but

then stop. A finite force on a solid will lead to a finite deformation. On the other

hand, if we attempt to apply a force onto air or water, it will just move along

the direction of force indefinitely. A finite force can lead to infinite deformation.

This is the main difference between solid mechanics and fluid mechanics. In

solid mechanics, we look at properties like elasticity. This measures how much

deformation we get when we apply a force. In fluid mechanics, we don’t measure

distances, since they can potentially be infinite. Instead, we would often be

interested in the velocity of fluids.

There are many applications of fluid dynamics. On a small scale, the dynamics

of fluids in cells is important for biology. On a larger scale, the fluid flow of

the mantle affects the movement of tectonic plates, while the dynamics of the

atmosphere can be used to explain climate and weather. On an even larger scale,

we can use fluid dynamics to analyse the flow of galactic systems in the universe.

1 Parallel viscous flow

1.0 Preliminaries

This section is called preliminaries, not definitions, because the “definitions” we

give will be a bit fuzzy. We will (hopefully) get better definitions later on.

We start with an absolutely clear an unambiguous definition of the subject

we are going to study.

Definition (Fluid). A fluid is a material that flows.

Example.

Air, water and oil are fluids. These are known as simple or Newtonian

fluids, because they are simple.

Paint, toothpaste and shampoo are complex or non-Newtonian fluids, because

they are complicated.

Sand, rice and foams are granular flows. These have some fluid-like properties,

but are fundamentally made of small granular solids.

In this course, we will restrict our attention to Newtonian fluids. Practically

speaking, these are fluids for which our equations work. The assumption we will

make when deriving our equations will be the following:

Definition

(Newtonian fluids and viscosity)

A Newtonian fluid is a fluid

with a linear relationship between stress and rate of strain. The constant of

proportionality is viscosity.

This, and other concepts we define here, will become more clear when we

start to write down our equations.

Definition (Stress). Stress is force per unit area.

For example, pressure is a stress.

Definition

(Strain)

Strain is the extension per unit length. The rate of strain

(strain) is concerned with gradients of velocity.

As mentioned in the introduction, for fluid dynamics, we tend to study

velocities instead of displacements. So we will mostly work with rate of strain

instead of strain itself.

These quantities are in fact tensor fields, but we will not treat them as

such in this course. We will just consider “simplified” cases. For the full-blown

treatment with tensor fields, refer to the IID Fluid Dynamics.

In this course, we are going make a lot of simplifying assumptions, since

the world is too complicated. For example, most of the time, we will make

the inviscid approximation, where we set the viscosity to 0. We will also often

consider motions in one direction only.

1.1 Stress

We first look at stresses. These are the forces that exist inside the fluid. If we

have a boundary, we can classify the stress according to the direction of the force

— whether it is normal to or parallel to the boundary. Note that the boundary

can either be an actual physical boundary, or an imaginary surface we cook up

in order to compute things.

Suppose we have a fluid with pressure

acting on a surface with unit normal

n, pointing into the fluid. This causes

solid

fluid

pressure p

Definition (Normal stress). The normal stress is

= −pn.

The normal stress is present everywhere, as long as we have a fluid (with

pressure). However, pressure by itself does not do anything, since pressure acts

in all directions, and the net effect cancels out. However, if we have a pressure

gradient, then gives an actual force and drives fluid flow. For example, suppose

we have a pipe, with the pump on the left:

high pressure low pressure p

atm

∇p

force −∇p

Then this gives a body force that drives the water from left to right.

We can also have stress in the horizontal direction. Suppose we have two

infinite plates with fluid in the middle. We keep the bottom plane at rest, and

move the top plate with velocity U.

By definition, the stress is the force per unit area. In this case, it is the horizontal

force we need to exert to keep the top plate moving.

Definition

(Tangential stress)

The tangential stress

is the force (per unit

area) required to move the top plate at speed U .

This is also the force we need to exert on the bottom plate to keep it still.

Or the horizontal force at any point in fluid in order to maintain the velocity

gradient.

By definition of a Newtonian fluid, this stress is proportional to the velocity

gradient, i.e.

Law. For a Newtonian fluid, we have

∝

Definition

(Dynamic viscosity)

The dynamic viscosity

of the fluid is the

constant of proportionality in

= µ

We can try to figure out the dimensions of these quantities:

[τ

] = ML

−1

−2





= T

−1

[µ] = ML

−1

In SI units, µ has unit kg m

−1

We have not yet said what the fluid in the middle does. It turns out this

is simple: at the bottom, the fluid is constant, and at the top, the fluid moves

with velocity U . In between, the speed varies linearly.

We will derive this formally later.

For a general flow, let

(

) be the velocity of the fluid at position

. Then

the velocity gradient is

∂µ

(x)

∂n

Hence the tangential stress is given by

= µ

∂u

(x)

∂n

and is in the direction of the tangential component of velocity. Again, the normal

vector n points into the fluid.

1.2 Steady parallel viscous flow

We are first going to consider a very simple type of flow, known as steady parallel

viscous flow, and derive the equations of motion of it. We first explain what this

name means. The word “viscous” simply means we do not assume the viscosity

vanishes, and is not something new.

Definition

(Steady flow)

A steady flow is a flow that does not change in time.

In other words, all forces balance, and there is no acceleration.

Definition

(Parallel flow)

A parallel flow is a flow where the fluid only flows

in one dimension (say the

direction), and only depends on the direction

perpendicular to a plane (say the

x −z

plane). So the velocity can be written as

u = (u(y), 0, 0).

These can be conveniently thought of as being two-dimensional, by forgetting

the z direction.

Note that our velocity does not depend on the

direction. This can be

justified by the assumption that the fluid is incompressible. If we had a velocity

gradient in the

-direction, then we will have fluid “piling up” at certain places,

violating incompressibility.

We will give a formal definition of incompressibility later. In general, though,

fluids are not compressible. For example, sound waves are exactly waves of com-

pression in air, and cannot exist if air is incompressible. So we can alternatively

state the assumption of incompressibility as “sound travels at infinite speed”.

Hence, compressibility matters mostly when we are travelling near the speed

of sound. If we are moving in low speeds, we can just pretend the fluid is indeed

incompressible. This is what we will do in this course, since the speed of sound

is in general very high.

For a general steady parallel viscous flow, We can draw a flow profile like

this:

The lengths of the arrow signify the magnitude of the velocity at that point.

To derive the equations of motion, we can consider a small box in the fluid.

x, y

x + δx

y + δy

We know that this block of fluid moves in the

direction without acceleration.

So the total forces of the surrounding environment on the box should vanish.

We first consider the

direction. There are normal stresses at the sides, and

tangential stresses at the top and bottom. The sum of forces in the

-direction

(per unit transverse width) gives

p(x)δy − p(x + δx)δy + τ

(y + δy)δx + τ

(y)δx = 0.

By the definition of τ

, we can write

(y + δy) = µ

∂u

∂y

(y + δy), τ

(y) = −µ

∂u

∂y

(y),

where the different signs come from the different normals (for a normal pointing

downwards,

∂

∂n

∂

∂(−y)

Dividing by δxδy, we get

δx

(p(x) − p(x + δx)) + µ

δy



∂u

∂y

(y + δy) −

∂u

∂y

(y)



Taking the limit as δx, δy → 0, we end up with the equation of motion

−

∂p

∂x

+ µ

∂

∂y

= 0.

Performing similar calculations in the y direction, we obtain

−

∂p

∂y

= 0.

In the second equation, we keep the negative sign for consistency, but obviously

in this case it is not necessary.

This is the simplest possible case. We can extend this a bit by allowing,

non-steady flows and external forces on the fluid. Then the velocity is of the

form

u = (u(y, t), 0, 0).

Writing the external body force (per unit volume) as (

, f

0), we obtain the

equations

∂u

∂t

= −

∂p

∂x

+ µ

∂

∂y

+ f

0 = −

∂p

∂y

+ f

The derivation of these equations is straightforward, and is left as an exercise

for the reader on the example sheet. Here

is the density, i.e. the mass per unit

volume. For air, it is approximately

1 kg m

−3

, and for water, it is approximately

1000 kg m

−3

Along with equations, we also need boundary condition. In general, Newto-

nian fluids satisfy one of the following two boundary conditions:

(i)

No-slip condition: at the boundary, the tangential component of the fluid

velocity equals the tangential velocity of boundary. In particular, if the

boundary is stable, the tangential component of the fluid velocity is zero.

This means fluids stick to surfaces.

= 0. (1.5)

(ii)

Stress condition: alternatively, a tangential stress

is imposed on the fluid.

In this case,

−µ

∂u

∂n

= τ.

The no-slip condition is common when we have a fluid-solid boundary, where

we think the fluid “sticks” to the solid boundary. The stress condition is more

common when we have a fluid-fluid boundary (usually liquid and gas), where we

require the tangential stresses to match up. In non-parallel flow, we will often

require the flow not to penetrate the solid boundary, but this is automatically

satisfied in parallel flow.

We are going to look at some examples.

Example

(Couette flow)

This is flow driven by the motion of a boundary, as

we have previously seen.

We assume that this is a steady flow, and there is no pressure gradient. So our

equations give

∂

∂y

= 0.

Moreover, the no-slip condition says

= 0 on

= 0;

. The

solution is thus

u =

Example

(Poiseuille flow)

This is a flow driven by a pressure gradient between

stationary boundaries. Again we have two boundaries

We have a high pressure

on the left, and a low pressure

< P

on the right.

We solve this problem again, but we will also include gravity. So the equations

of motion become

−

∂p

∂x

+ µ

∂

∂y

= 0

−

∂p

∂y

− gρ = 0

The boundary conditions are u = 0 at y = 0, h. The second equation implies

p = −gρy + f (x)

for some function f . Substituting into the first gives

∂

∂y

= f

(x).

The left is a function of

only, while the right depends only on

. So both must

be constant, say G. Using the boundary conditions, we get

∂

∂y

= f

(x) = G =

− P

where L is the length of the tube. Then we find

u =

2µ

y(h − y).

Here the velocity is the greatest at the middle, where y =

Since the equations of motion are linear, if we have both a moving boundary

and a pressure gradient, we can just add the two solutions up.

1.3 Derived properties of a flow

The velocity and pressure already fully describe the flow. However, there are

some other useful quantities we can compute out of these.

The first thing we consider is how much stuff is being transported by the

flow.

Definition

(Volume flux)

The volume flux is the volume of fluid traversing a

cross-section per unit time. This is given by

q =

u(y) dy

per unit transverse width.

We can calculate this immediately for the two flows.

Example. For the Couette flow, we have

q =

dy =

For the Poiseuille flow, we have

q =

2µ

y(h − y) dy =

12µ

We can also ask how much “rotation” there is in the flow.

Definition (Vorticity). The vorticity is defined by

ω = ∇ × u.

In our case, since we have

u = (u(y, t), 0, 0),

we have

ω =



0, 0, −

∂u

∂y



Example.

For the case of the Couette flow, the vorticity is



0, 0, −



This is a constant, i.e. the vorticity is uniform.

For the case of the Poiseuille flow, we have

ω =



0, 0,



y −



Recall that the tangential stress

is the tangential force per unit area

exerted by the fluid on the surface, given by

= µ

∂u

∂n

with n pointing into the fluid.

Example. For the Couette flow, we have

(

y = 0

−µ

y = h

We see that at

= 0, the stress is positive, and pulls the surface forward. At

y = h, it is negative, and the surface is pulled backwards.

For the Poiseuille flow, we have

(

y = 0

y = h

Both surfaces are pulled forward, and this is independent of the viscosity. This

makes sense since the force on the surface is given by the pressure gradient,

which is independent of the fluid.

1.4 More examples

Since we are probably bored by the Couette and Poiseuille flows, we do another

more interesting example.

Example

(Gravity-driven flow down a slope)

Suppose we have some fluid

flowing along a slope.

atmospheric

pressure

Here there is just atmosphere above the fluid, and we assume the fluid flow is

steady, i.e.

is merely a function of

. We further assume that the atmospheric

pressure does not vary over the vertical extent of the flow. This is a very good

approximation because ρ

air

 ρ

liq

Similarly, we assume

air

 µ

liq

. So the air exerts no significant tangential

stress. This is known as a free surface.

We first solve the

momentum equation. The force in the

direction is

−gρ cos α. Hence the equation is

∂p

∂y

= −gp cos α.

Using the fact that p = p

at the top boundary, we get

p = p

− gρ cos α(y −h).

In particular, p is independent of x. In the x component, we get

∂

∂y

= −gρ sin α.

The no slip condition gives

= 0 when

= 0. The other condition is that there

is no stress at y = h. So we get

∂u

∂y

= 0 when y = h.

The solution is thus

u =

gρ sin α

2µ

y(2h − y).

This is a bit like the Poiseuille flow, with

gp sin α

2µ

as the pressure gradient. But

instead of going to zero at

, we get to zero at

= 2

instead. So this is

half a Poiseuille flow.

It is an exercise for the reader to calculate the volume flux q.

For a change, we do a case where we have unsteady flow.

Example. Consider fluid initially at rest in y > 0, resting on a flat surface.

At time

= 0, the boundary

= 0 starts to move at constant speed

. There is

no force and no pressure gradient.

We use the x-momentum equation to get

∂u

∂t

= ν

∂

∂y

where

. This is clearly the diffusion equation, with the diffusivity

. We

can view this as the diffusion coefficient for motion/momentum/vorticity.

Definition (Kinematic viscosity). The kinematic viscosity is

ν =

The boundary conditions are

= 0 for

= 0 and

u →

0 as

y → ∞

for all

The other boundary condition is obviously u = U when y = 0, for t > 0.

You should have learnt how to solve this in, say, IB Methods. We will

approach this differently here.

Before we start, we try to do some dimensional analysis, and try to figure

out how far we have to go away from the boundary before we don’t feel any

significant motion.

We first list all the terms involved. We are already provided with a velocity

. We let

be our time scale. We would like to know how fast the movement

of fluid propagates up the

axis. We note that in this case, we don’t really

have an extrinsic length scale — in the case where we have two boundaries, the

distance between them is a natural length scale to compare with, but here the

fluid is infinitely thick. So we need to come up with a characteristic intrinsic

length scale somewhat arbitrarily. For example, at any time

, we can let

the amount of fluid that has reached a speed of at least

(that was completely

arbitrary).

Instead of trying to figure out the dimensions of, say,

and trying to match

them, we just replace terms in the differential equation with these quantities

of the right dimension, since we know our differential equation is dimensionally

correct. So we obtain

∼ ν

Since the U cancels out, we get

δ ∼

√

νT .

We can figure out approximately how big this is, using the table below:

µ(kg m

−1

) ρ(kg m

−3

) ν(m

−1

)

water 10

−3

−6

air 10

−5

1 10

−5

These are, in general, very tiny. So we expect to feel nothing when we move

slightly away from the boundary.

We now solve the problem properly. In an infinite domain with no extrinsic

length scale, the diffusion equation admits a similarity solution. We write

u(y, t) = Uf(η),

where

(

) is a dimensionless function of the dimensionless variable

. To turn

y into a dimensionless variable, we have to define

η =

√

νt

We substitute this form of the solution into the differential equation. Then we

get

−

ηf

(η) = f

(η),

with boundary condition f = 1 on η = 0. Then the solution is by definition

f = erfc





Hence

u = U erfc



√

ηt



δ ∼

√

ηt

We can compute the tangential stress of the fluid in the above case to be

= µ

∂u

∂y

= µ

√

νt



√



−y



y=0

µU

√

πνt

Using our previous calculations of the kinematic viscosities, we find that

√

is 1

for water and 10

−3

for air.

This is significant in, say, the motion of ocean current. When the wind blows,

it causes the water in the ocean to move along with it. This is done in a way

such that the surface tension between the two fluids match at the boundary.

Hence we see that even if the air blows really quickly, the resultant ocean current

is much smaller, say a hundredth to a thousandth of it.

2 Kinematics

So far, we have considered a rather specific kind of fluid flow — parallel flow,

and derived the equations of motion from first principles. From now on, we will

start to consider more general theory of fluid dynamics. In this section, we are

mostly making some definitions and descriptions of flows. In the next chapter,

we will take the equations of motion and try to solve them in certain scenarios.

2.1 Material time derivative

We first want to consider the problem of how we can measure the change in a

quantity, say

. This might be pressure, velocity, temperature, or anything else

you can imagine.

The obvious thing to do would be the consider the time derivative

∂f

∂t

In physical terms, this would be equivalent to fixing our measurement instrument

at a point and measure the quantity over time. This is known as the Eulerian

picture.

However, often we want to consider something else. We pretend we are a

fluid particle, and move along with the flow. We then measure the change in

along our trajectory. This is known as the Lagrangian picture.

Let’s look at these two pictures, and see how they relate to each other.

Consider a time-dependent field

(

x, t

). For example, it might be the pressure

of the system, or the temperature of the fluid. Consider a path

(

) through the

field, and we want to know how the field varies as we move along the path.

Along the path x(t), the chain rule gives

(x(t), t) =

∂f

∂x

∂f

∂y

∂f

∂z

∂f

∂t

= ∇f ·

x +

∂f

∂t

Definition

(Material derivative)

(

) is the (Lagrangian) path followed by a

fluid particle, then necessarily

x(t) = u by definition. In this case, we write

This is the material derivative.

In other words, we have

= u · ∇f +

∂f

∂t

On the left of this equation, we have the Lagrangian derivative, which is the

change in

as we follow the path. On the right of this equation, the first term

is the advective derivative, which is the change due to change in position. The

second term is

∂f

∂t

, the Eulerian time derivative, which is the change at a fixed

point.

For example, consider a river that gets wider as we go downstream. We know

(from, say, experience) that the flow is faster upstream than downstream. If the

motion is steady, then the Eulerian time derivative vanishes, but the Lagrangian

derivative does not, since as the fluid goes down the stream, the fluid slows

down, and there is a spacial variation.

Often, it is the Lagrangian derivative that is relevant, but the Eulerian time

derivative is what we can usually put into differential equations. So we will need

both of them, and relate them by the formula above.

2.2 Conservation of mass

We can start formulating equations. The first one is the conservation of mass.

We fix an arbitrary region of space

with boundary

∂D

and outward normal

n. We imagine there is some flow through this volume

∂D

What we want to say is that the change in the mass inside

is equal to the

total flow of fluid through the boundary. We can write this as

ρ dV = −

∂D

ρu · n dS.

We have the negative sign since we picked the outward normal, and hence the

integral measures the outward flow of fluid.

Since the domain is fixed, we can interchange the derivative and the integral

on the left; on the right, we can use the divergence theorem to get



∂ρ

∂t

+ ∇ · (ρu)



dV = 0.

Since D was arbitrary, we must have

∂ρ

∂t

+ ∇ · (ρu) = 0

everywhere in space.

This is the general form of a conservation law — the rate of change of

“stuff” density plus the divergence of the “stuff flux” is constantly zero. Similar

conservation laws appear everywhere in physics.

In the conservation equation, we can expand ∇ · (ρu) to get

∂ρ

∂t

+ u · ∇ρ + ρ∇ · u = 0.

We notice the first term is just the material derivative of ρ. So we get

Dρ

+ ρ∇ · u = 0.

With the conservation of mass, we can now properly say what incompressibility

is. What exactly happens when we compress a fluid? When we compress mass,

in order to conserve mass, we must increase the density. If we don’t allow

changes in density, then the material derivative

Dρ

must vanish. So we define

an incompressible fluid as follows:

Definition

(Incompressible fluid)

A fluid is incompressible if the density of a

fluid particle does not change. This implies

Dρ

= 0,

and hence

∇ · u = 0.

This is also known as the continuity equation.

For parallel flow,

= (

0). So if the flow is incompressible, we must have

∂u

∂x

= ∇ · u = 0. So we considered u of the form u = u(y, z, t).

Of course, incompressibility is just an approximation. Since we can hear

things, everything must be compressible. So what really matters is whether

the slow speed is small compared to the speed of sound. If it is relatively

small, then incompressibility is a good approximation. In air, the speed of

sound is approximately

340 m s

−1

. In water, the speed of sound is approximately

1500 m s

−1

2.3 Kinematic boundary conditions

Suppose our system has a boundary. There are two possible cases — either the

boundary is rigid, like a wall, or it moves with the fluid. In either case, fluids

are not allowed to pass through the boundary.

To deal with boundaries, we first suppose it has a velocity

(which may

vary with position if the boundary extends through space). We define a local

reference frame moving with velocity

, so that in this frame of reference, the

boundary is stationary.

In this frame of reference, the fluid has relative velocity

= u − U.

As we mentioned, fluids are not allowed to cross the boundary. Let the normal

to the boundary be n, then we must have

· n = 0.

In terms of the outside frame, we get

u · n = U · n.

This is the condition we have for the boundary.

If we have a stationary rigid boundary, e.g. the wall, then

. So our

boundary condition is

u · n = 0.

Free boundaries are more complicated. A good example of a free material

boundary is the surface of a water wave, or interface between two immiscible

fluids — say oil and water.

oil

water

ζ(x, y, t)

We can define the surface by a function

z = ζ(x, y, t).

We can alternatively define the surface as a contour of

F (x, y, z, t) = z − ζ(x, y, t).

Then the surface is defined by

= 0. By IA Vector Calculus, the normal is

parallel to

n k ∇F = (−ζ

, −ζ

, 1).

Also, we have

U = (0, 0, ζ

We now let the velocity of the fluid be

u = (u, v, w).

Then the boundary condition requires

u · n = U · n.

In other words, we have

−uζ

− vζ

+ w = ζ

So we get

w = uζ

+ vζ

+ ζ

= u · ∇ζ +

∂ζ

∂t

Dζ

So all the boundary condition says is

Dζ

= w.

Alternatively, since

is a material surface, its material derivative must vanish.

= 0,

and this just gives the same result as above. We will discuss surface waves

towards the end of the course, and then come back and use this condition.

2.4 Streamfunction for incompressible flow

We suppose our fluid is incompressible, i.e.

∇ · u = 0.

By IA Vector Calculus, this implies there is a vector potential A such that

Definition (Vector potential). A vector potential is an A such that

u = ∇ × A.

In the special case where the flow is two dimensional, say

u = (u(x, y, t), v(x, y, t), 0),

we can immediately know A is of the form

A = (0, 0, ψ(x, y, t)),

Taking the curl of this, we get

u =



∂ψ

∂y

, −

∂ψ

∂x

, 0



Definition

(Streamfunction)

The

such that

= (0

, ψ

) is the streamfunc-

tion.

This streamfunction is both physically significant, and mathematically con-

venient, as we will soon see.

We look at some properties of the streamfunction. The first thing we can do

is to look at the contours ψ = c. These have normal

n = ∇ψ = (ψ

, ψ

, 0) .

We immediately see that

u · n =

∂ψ

∂x

∂ψ

∂y

−

∂ψ

∂y

∂ψ

∂x

= 0.

So the flow is perpendicular to the normal, i.e. tangent to the contours of ψ.

Definition

(Streamlines)

The streamlines are the contours of the streamfunc-

tion ψ.

This gives an instantaneous picture of flow.

Note that if the flow is unsteady, then the streamlines are not particle paths.

Example.

Consider

= (

0). When

= 0, the velocity is purely in the

direction, and the streamlines are also vertical; at

= 1, the velocity makes an

◦

angle with the horizontal, and the streamlines are slanted:

t = 0 t = 1

However, no particles will actually follow any of these streamlines. For a particle

released at x

= (x

, y

). Then we get

˙x(t) = u = t, ˙y = v = 1.

Hence we get

x =

+ x

, y = t + y

Eliminating t, we get that the path is given by

(x − x

) =

(y − y

)

So the particle paths are parabolas.

Typically, we draw streamlines that are “evenly spaced”, i.e. we pick the

streamlines ψ = c

, ψ = c

etc. such that c

− c

= c

− c

Then we know the flow is faster where streamlines are closer together:

slow fast

This is since the fluid between any two stream lines must be between the stream

lines. So if the flow is incompressible, to conserve mass, they mast move faster

when the streamlines are closer.

We can also consider the volume flux (per unit length in the

-direction),

crossing any curve from x

to x

slow fast

Then the volume flux is

q =

u · n d`.

We see that

n d` = (−dy, dx).

So we can write this as

q =

−

∂ψ

∂y

dy −

∂ψ

∂x

dx = ψ(x

) − ψ(x

So the flux depends only the difference in the value of

. Hence, for closer

streamlines, to maintain the same volume flux, we need a higher speed.

Also, note that

is constant on a stationary rigid boundary, i.e. the boundary

is a streamline, since the flow is tangential at the boundary. This is a consequence

of u · n = 0. We often choose ψ = 0 as our boundary.

Sometimes it is convenient to consider the case when we have plane polars.

We embed these in cylindrical polars (r, θ, z). Then we have

u = ∇ × (0, 0, ψ) =



∂

0 0 ψ





∂ψ

∂θ

, −

∂ψ

∂r

, 0



As an exercise, we can verify that

∇ · u

= 0. It is convenient to note that in

plane polars,

∇ · u =

∂

∂r

(ru

) +

∂u

∂θ

3 Dynamics

3.1 Navier-Stokes equations

So far, we haven’t done anything useful. We have not yet come up with an

equation of motion for fluids, and thus we don’t know how fluids actually behave.

The equations for the parallel viscous flow was a good start, and it turns out

the general equation of motion is of a rather similar form.

Unfortunately, the equation is rather complicated and difficult to derive, and

we will not derive it in this course. Instead, it is left for the Part II course.

However, we will be able to derive some special cases of it later under certain

simplifying assumptions.

Law (Navier-Stokes equation).

= −∇p + µ∇

u + f.

This is the general equation for fluid motion. The left hand side is mass

times acceleration, and the right is the individual forces — the pressure gradient,

viscosity, and the body forces (per unit volume) respectively.

In general, these are very difficult equations to solve because of non-linearity.

For example, in the material derivative, we have the term u · ∇u.

There are a few things to note:

(i)

The acceleration of a fluid particle is the Lagrangian material derivative of

the velocity.

(ii)

The derivation of the viscous term is complicated, since for each side of the

cube, there is one normal direction and two tangential directions. Thus

this requires consideration of the stress tensor. This is only done in IID

Fluid Dynamics.

(iii)

In a gravitational field, we just have

gρ

. This is the only body force

we will consider in this course.

(iv) Note that ∇

u can be written as

∇

u = ∇(∇ · u) − ∇ × (∇ × u).

In an incompressible fluid, this reduces to

∇

u = −∇ × (∇ × u) = −∇ × ω,

where

∇ × u

is the vorticity. In Cartesian coordinates, for

, u

), we have

∇

u = (∇

, ∇

where

∇

∂

∂x

∂

∂y

∂

∂z

(v)

The Navier-Stokes equation reduces to the parallel flow equation in the

special case of parallel flow, i.e. u = (u(y, t), 0, 0). This verification is left

as an exercise for the reader.

3.2 Pressure

In the Navier-Stokes equation, we have a pressure term. In general, we classify

the pressure into two categories. If there is gravity, then we will get pressure in

the fluid due to the weight of fluid above it. This is what we call hydrostatic

pressure. Technically, this is the pressure in a fluid at rest, i.e. when u = 0.

We denote the hydrostatic pressure as

. To find this, we put in

= 0 into

the Navier-Stokes equation to get

∇p

= f = ρg.

We can integrate this to obtain

= pg · x + p

where p

is some arbitrary constant. Usually, we have g = (0, 0, −g). Then

= p

− gρz.

This exactly says that the hydrostatic pressure is the weight of the fluid above

you.

What can we infer from this? Suppose we have a body

with boundary

∂D

and outward normal n. Then the force due to the pressure is

F = −

∂D

n · dS

= −

∇p

= −

gρ dV

= −g

ρ dV

= −Mg,

where

is the mass of fluid displaced. This is known as Archimedes’ principle.

In particular, if the body is less dense than the fluid, it will float; if the body

is denser than the fluid, it will sink; if the density is the same, then it does not

move, and we say it is neutrally buoyant.

This is valid only when nothing is moving, since that was our assumption.

Things can be very different when things are moving, which is why planes can

fly.

In general, when there is motion, we might expect some other pressure

gradient. It can either be some external pressure gradient driving the motion

(e.g. in the case of Poiseuille flow), or a pressure gradient caused by the flow

itself. In either case, we can write

p = p

+ p

where

is the hydrostatic pressure, and

is what caused/results from motion.

We substitute this into the Navier-Stokes equation to obtain

= −∇p

+ µ∇

So the hydrostatic pressure term cancels with the gravitational term. What we

usually do is drop the “prime”, and just look at the deviation from hydrostatic

pressure. What this means is that gravity no longer plays a role, and we can

ignore gravity in any flow in which the density is constant. Then all fluid particles

are neutrally buoyant. This is the case in most of the course, except when we

consider motion of water waves, since there is a difference in air density and

water density.

3.3 Reynolds number

As we mentioned, the Navier-Stokes equation is very difficult to solve. So we

want to find some approximations to the equation. We would like to know if we

can ignore some terms. For example, if we can neglect the viscous term, then we

are left with a first-order equation, not a second-order one.

To do so, we look at the balance of terms, and see if some terms dominate

the others. This is done via Reynolds number.

We suppose the flow has a characteristic speed

and an extrinsic length

scale

, externally imposed by geometry. For example, if we look at the flow

between two planes, the characteristic speed can be the maximum (or average)

speed of the fluid, and a sensible length scale would be the length between the

planes.

Next, we have to define the time scale

L/U

. Finally, we suppose pressure

differences have characteristic magnitude

. We are concerned with differences

since it is pressure differences that drive the flow.

We are going to take the Navier-Stokes equation, and look at the scales of

the terms. Dividing by ρ, we get

∂u

∂t

+ u · ∇u = −

∇p + ν∇

where again ν =

. We are going to estimate the size of these terms. We get

(L/U)

U ·

Dividing by U

/L, we get

1 1

ρU

Definition (Reynolds number). The Reynolds number is

Re =

which is a dimensionless number. This is a measure of the magnitude of the

inertia to viscous terms.

So if

is very large, then the viscous term is small, and we can probably

neglect it. For example, for an aircraft, we have

U ∼

L ∼

10 and

ν ∼

−5

So the Reynolds number is large, and we can ignore the viscous term. On the

other hand, if we have a small slow-moving object in viscous fluid, then

will

be small, and the viscous term is significant.

Note that the pressure always scales to balance the dominant terms in the

equation, so as to impose incompressibility, i.e.

∇ · u

= 0. So we don’t have to

care about its scale.

In practice, it is the Reynolds number, and not the factors

U, L, ν

individually,

that determines the behaviour of a flow. For example, even though lava is very

very viscous, on a large scale, the flow of lava is just like the flow of water in a

river, since they have comparable Reynolds number.

Definition

(Dynamic similarity)

Flows with the same geometry and equal

Reynolds numbers are said to be dynamically similar.

When Re  1, the inertia terms are negligible, and we now have

P ∼

ρνU

µU

So the pressure balances the sheer stress. We can approximate the Navier-Stokes

equation by dropping the term on the left hand side, and write

0 = −∇p + µ∇

with the incompressibility condition

∇ · u = 0.

These are known as Stokes equations. This is now a linear equation, with four

equations and four unknowns (three components of

and one component of

pressure). We find

u ∝ ∇p,

and so the velocity is proportional to the pressure gradient.

When

Re 

1, the viscous terms are negligible on extrinsic length scale.

Then the pressure scales on the momentum flux,

P ∼ ρU

and on extrinsic scales, we can approximate Navier-Stokes equations by the

Euler equations

= −∇p

∇ · u = 0.

In this case, the acceleration is proportional to the pressure gradient.

Why do we keep saying “extrinsic scale”? Note that when we make this

approximation, the order of the differential equation drops by 1. So we can no

longer satisfy all boundary conditions. The boundary condition we have to give

up is the no-slip condition. So when we make this approximation, we will have

to allow the fluid at the boundary to have non-zero velocity.

So when is this approximation valid? It is obviously wrong when we are at

the boundary. If the velocity gradient near the boundary is relatively large, then

we quickly get significant non-zero velocity when we move away from boundary,

and hence obtain the “correct” answer. So we get problems only at the length

scale where the viscous and inertia effects are comparable, i.e. at the intrinsic

length scale.

Since the intrinsic length scale

is the scale at which the viscous and inertia

terms are comparable, we need

∼

νU

So we get

δ =

Alternatively, we have

Thus, for large Reynolds number, the intrinsic length scale, which is the scale

in which the viscous and boundary effects matter, is small, and we can ignore

them.

For much of the rest of this course, we will ignore viscosity, and consider

inviscid flow.

3.4 A case study: stagnation point flow (u = 0)

Example

(Stagnation point flow)

We attempt to find a solution to the Navier-

Stokes equation in the half-plane

y ≥

0, subject to the boundary condition

u → (Ex, −Ey, 0) as y → ∞, where E > 0 is a constant, and u = 0 on y = 0.

Intuitively, we would expect the flow to look like this:

The boundary conditions as

y → ∞

gives us a picture of what is going on at

large

, as shown in the diagram, but near the boundary, the velocity has to

start to vanish. So we would want to solve the equations to see what happens

near the boundary.

Again, this problem does not have any extrinsic length scale. So we can seek

a similarity solution in terms of

η =

, δ =

First of all, we make sure this is the right similarity solution. We show that

has dimensions of length:

– The dimension of E is [E] = T

−1

– The dimension of ν is [ν] = L

−1

Hence





= L

and δ has dimension L. Therefore η =

is dimensionless.

Note that we make

a function of

instead of a function of

so that we

can impose the boundary condition as y → ∞.

What would the similarity solution look like? Applying the incompressibility

conditions ∇ · u = 0, it must be of the form

u = (u, v, 0) = (Exg

(ν), −Eδg(η), 0).

To check this is really incompressible, we can find the streamfunction. It turns

out it is given by

ψ =

√

νExg(η).

We can compute

u =

∂ψ

∂t

√

νExg

(η)

= Exg

(η),

and similarly for the

component. So this is really the streamfunction. Therefore

we must have

∇ · u

= 0. Alternatively, we can simply compute

∇ · u

and find it

to be zero.

Finally, we look at the Navier-Stokes equations. The

and

components are

∂u

∂x

+ v

∂u

∂y

= −

∂p

∂x

+ ν



∂

∂x

∂

∂y



∂v

∂x

+ v

∂v

∂y

= −

∂p

∂y

+ ν



∂

∂x

∂

∂y



Substituting our expression for u and v, we get

Exg

− EδgExg

= −

∂p

∂x

− νExg

000

Some tidying up gives

x(g

− gg

) = −

∂p

∂x

− E

000

We can do the same thing for the y-component, and get

√

νEgg

= −

∂p

∂y

− E

√

νEg

So we’ve got two equations for

and

. The trick is to take the

derivative of

the first, and

derivative of the second, and we can use that to eliminate the

terms. Then we have

− gg

000

= g

(4)

So we have a single equation for g that we shall solve.

We now look at our boundary conditions. The no-slip condition gives

on y = 0. So g

(0) = g(0) = 0 when η = 0.

As as

y → ∞

, the boundary conditions for

gives

(

)

→

(

)

→ η

η → ∞.

All dimensional variables are absorbed into the scaled variables

and

. So

we only have to solve the ODE once. The far field velocity

= (

Ex, −Ey,

0) is

reached to a very good approximation when

η & 1, y & δ =

(η)

δ = O(1)

We get the horizontal velocity profile as follows:

At the scale δ, we get a Reynolds number of

Uδ

∼ O(1).

This is the boundary layer. For a larger extrinsic scale L  δ, we get

 1.

When interested in flow on scales much larger than

, we ignore the region

y < δ

(since it is small), and we imagine a rigid boundary at

at which the no-slip

condition does not apply.

When Re

 1, we solve the Euler equations, namely

= −∇p + f

∇ · u = 0.

We still have a boundary condition — we don’t allow fluid to flow through the

boundary. So we require

u · n = 0 at a stationary rigid boundary.

The no-slip condition is no longer satisfied.

It is an exercise for the reader to show that

= (

Ex, −Ey,

0) satisfies the

Euler equations in y > 0 with a rigid boundary of y = 0, with

p = p

−

ρE

+ y

We can plot the curves of constant pressure, as well as the streamlines:

As a flow enters from the top, the pressure keeps increases, and this slows down

the flow. We say the

-pressure gradient is adverse. As it moves down and flows

sideways, the pressure pushes the flow. So the x-pressure gradient is favorable.

At the origin, the velocity is zero, and this is a stagnation point. This is also

the point of highest pressure. In general, velocity is high at low pressures and

low at high pressures.

3.5 Momentum equation for inviscid (ν

= 0

) incompress-

ible fluid

We are now going to derive the Navier-Stokes equation in the special case where

viscosity is assumed to vanish. We will derive this by considering the change in

momentum.

Consider an arbitrary volume

with boundary

∂D

and outward pointing

normal n.

∂D

The total momentum of the fluid in D can change owing to four things:

(i) Momentum flux across the boundary ∂D;

(ii) Surface pressure forces;

(iii) Body forces;

(iv) Viscous surface forces.

We will ignore the last one. We can then write the rate of change of the total

momentum as

ρu dV = −

∂D

ρu(u · n) dS −

∂D

pn dS +

f dV.

It is helpful to write this in suffix notation. In this case, the equation becomes

ρu

dV = −

∂D

ρu

dS −

∂D

−pn

dS +

dV.

Just as in the case of mass conservation, we can use the divergence theorem to

write



∂u

∂t

+ ρ

∂

∂x

)



dV =



−

∂p

∂x

+ f



dV.

Since D is arbitrary, we must have

∂u

∂t

+ ρ

∂u

∂x

+ ρu

∂u

∂x

= −

∂p

∂x

+ f

The last term on the left is the divergence of

, which vanishes by incompress-

ibility, and the remaining terms is just the material derivative of

. So we

get

Proposition (Euler momentum equation).

= −∇p + f.

This is just the equation we get from the Navier-Stokes equation by ignoring

the viscous terms. However, we were able to derive this directly from momentum

conservation.

We can further derive some equations from this.

For conservative forces, we can write

−∇χ

, where

is a scalar potential.

For example, gravity can be given by

ρg

∇

(

ρg · x

) (for

constant). So

χ = −ρg · x = gρz if g = (0, 0, −g).

In the case of a steady flow,

∂u

∂t

vanishes. Then the momentum equation

becomes

0 = −

∂D

ρu(u · n) dS −

∂D

pn dS −

∇χ dV.

We can then convert this to

Proposition (Momentum integral for steady flow).

∂D

(ρu(u · n) + pn + χn) dS = 0.

Alternatively, we notice the vector identity

u × (∇ × u) = ∇



|u|



− u · ∇u.

We use this to rewrite the Euler momentum equation as

∂u

∂t

+ ρ∇



|u|



− ρu × (∇ × u) = −∇p − ∇χ.

Dotting with u, the last term on the left vanishes, and we get

Proposition (Bernoulli’s equation).

∂|u|

∂t

= −u · ∇



ρ|u|

+ p + χ



Note also that this tells us that high velocity goes with low pressure; low

pressure goes with high velocity.

In the case where we have a steady flow, we know

H =

ρ|u|

+ p + χ

is constant along streamlines.

Even if the flow is not steady, we can still define the value

, and then we

can integrate Bernoulli’s equation over a volume D to obtain

ρ|u|

dV = −

∂D

Hu · n dS.

So H is the transportable energy of the flow.

Example.

Consider a pipe with a constriction. Ignore the vertical pipes for the

moment — they are there just so that we can measure the pressure in the fluid,

as we will later see.

u, p, a

Suppose at the left, we have a uniform speed

, area

and pressure

. In the

middle constricted area, we have speed u, area a and pressure p.

By the conservation of mass, we have

q = UA = ua.

We apply Bernoulli along the central streamline, using the fact that

is constant

along streamlines. We can omit the body force

ρgy

, since this is the same

at both locations. Then we get

ρU

+ P =

ρu

+ p.

Replacing our U with q/A and u with q/a, we obtain





+ P =





+ p.

Rearranging gives



−



2(p − P )

We see there is a difference in pressure due to the difference in area. This is

balanced by the difference in heights

. Using the

-momentum equation, we get



−



2(p − P )

= −2gh.

Then we obtain

q =

2gh

√

− a

Therefore we can measure

in order to find out the flow rate. This allows us to

measure fluid velocity just by creating a constriction and then putting in some

pipes.

Example

(Force on a fire hose nozzle)

Suppose we have a fire hose nozzle like

this:

(2)

(3)

(4)

(5)(1)

u, a, p = 0

We consider the steady-flow equation and integrate along the surface indicated

above. We integrate each section separately. The end (1) contributes

ρU(−U)A − P A.

On (2), everything vanishes. On (3), the first term vanishes since the velocity is

parallel to the surface. Then we get a contribution of

0 +

nozzle

pn ·

x dS. (3)

Similarly, everything in (4) vanishes. Finally, on (5), noting that p = 0, we get

ρu

By the steady flow equation, we know these all sum to zero. Hence, the force on

the nozzle is just

F =

nozzle

pn ·

x dS = ρAU

− ρau

+ P A.

We can again apply Bernoulli along a streamline in the middle, which says

ρU

+ P =

ρu

So we can get

F = ρAU

− ρau

ρA(u

− U

) =



1 −



Let’s now put some numbers in. Suppose

= (0

πm

and

= (0

05)

πm

So we get

= 4.

A typical fire hose has a flow rate of

q = 0.01 m

−1

So we get

F =

· 1000 ·

π/40

· 10

−4





≈ 14 N.

3.6 Linear flows

Suppose we have a favorite point

. Near the point

, it turns out we can break

up the flow into three parts — uniform flow, pure strain, and pure rotation.

To do this, we take the Taylor expansion of u about x

u(x) = u(x

) + (x − x

) · ∇u(x

) + ···

= u

+ r · ∇u

with

x −x

and

(

). This is a linear approximation to the flow field.

We can do something more about the

∇u

term. This is a rank-2 tensor, i.e.

a matrix, and we can split it into its symmetric and antisymmetric parts:

∇u =

∂u

∂x

= E

+ Ω

= E + Ω,

where



∂u

∂x

∂u

∂x



Ω



∂u

∂x

−

∂u

∂x



We can write the second part in terms of the vorticity. Recall we have defined

the vorticity as

ω = ∇ × u.

Then we have

ω × r = (∇ × u) × r = r



∂u

∂x

−

∂u

∂x



= 2Ω

So we can write

u = u

+ Er +

ω × r.

The first component is uniform flow; the second is the strain field; and the last

is the rotation component.

uniform flow

pure strain pure rotation

Since we have an incompressible fluid, we have

∇ · u

= 0. So

has zero trace,

i.e. if

E =





0 0

0 E

0 0 E





then E

+ E

= 0. This justifies the picture above.

3.7 Vorticity equation

The Navier-Stokes equation tells us how the velocity changes with time. Can we

obtain a similar equation for the vorticity? Consider the Navier-Stokes equation

for a viscous fluid,



∂u

∂t

+ u · ∇u



= −∇p − ∇χ + µ∇

We use a vector identity

u · ∇u =

∇|u

| − u × ω,

and take the curl of the above equation to obtain

∂ω

∂t

− ∇ × (u × ω) = ν∇

ω,

exploiting the fact that the curl of a gradient vanishes. We now use the fact that

∇ × (u × ω) = (∇ · ω)u + ω ·∇u − (∇ · u)ω −u · ∇ω.

The divergence of a curl vanishes, and so does

∇ · u

by incompressibility. So we

get

∂ω

∂t

+ u · ∇ω −ω ·∇u = ν∇

ω.

Now we use the definition of the material derivative, and rearrange terms to

obtain

Proposition (Vorticity equation).

Dω

= ω · ∇u + ν∇

ω.

Hence, the rate of change of vorticity of a fluid particle is caused by

ω · ∇u

(amplification by stretching or twisting) and

ν∇

(dissipation of vorticity by

viscosity). The second term also allows for generation of vorticity at boundaries

by the no-slip condition. This will be explained shortly.

Consider an inviscid fluid, where ν = 0. So we are left with

Dω

= ω · ∇u.

So if we take the dot product with ω, we get



|ω|



= ω · ∇u · ω

= ω(E + Ω)ω

= ω

+ Ω

)ω

Since

is symmetric, while Ω

is antisymmetric, the second term vanishes.

In the principal axes, E is diagonalizable. So we get



|ω|



= E

+ E

wlog, we assume

0 (since the

’s sum to 0), and imagine

, E

0. So

the flow is stretched in the

direction and compressed radially. We consider

what happens to a vortex in the direction of the stretching,

= (

0). We

then get





= E

So the vorticity grows exponentially. This is vorticity amplification by stretching.

This is not really unexpected — as the fluid particles have to get closer to

the axis of rotation, they have to rotate faster, by the conservation of angular

momentum.

This is in general true — vorticity increases as the length of a material

line increases. To show this, we consider two neighbouring (Lagrangian) fluid

particles, x

(t), x

(t). We let δ`(t) = x

(t) − x

(t). Note that

(t)

= u

(t)

= u

Therefore

Dδ`(t)

= u(x

) − u(x

) = δ` · ∇u,

by taking a first-order Taylor expansion. This is exactly the same equation as

that for

in an inviscid fluid. So vorticity increases as the length of a material

line increases.

Note that the vorticity is generated by viscous forces near boundaries.

When we make the inviscid approximation, then we are losing this source of

vorticity, and we sometimes have to be careful.

We have so far assumed the density is constant. If the fluid has a non-uniform

density ρ(x), then it turns out

Dω

= ω · ∇u +

∇ρ × ∇p.

This is what happens in convection flow. The difference in density drives a

motion of the fluid. For example, if we have a horizontal density gradient and a

vertical pressure gradient (e.g. we have a heater at the end of a room with air

pressure varying according to height), then we get the following vorticity:

∇p

∇ρ

hot, light cold, dense

4 Inviscid irrotational flow

From now on, we are going to make a further simplifying assumption. We are

going to assume that our flow is is incompressible, inviscid, and irrotational.

We first check this is a sensible assumption, in that if we start of with an

irrotational flow, then the flow will continue being irrotational. Suppose we have

∇×u

= 0, and the fluid is inviscid and homogeneous (i.e.

is constant).

Then by the vorticity equation

Dω

= ω · ∇u = 0.

So the vorticity will keep on being zero.

In this case, we can write u = ∇φ for some φ.

Definition

(Velocity potential)

The velocity potential of a velocity

is a scalar

function φ such that u = ∇φ.

Note that since we are lazy, we don’t not have a negative sign in front of

∇φ

unlike, say, gravity.

If we are incompressible, then ∇ · u = 0 implies

∇

φ = 0.

So the potential satisfies Laplace’s equation.

Definition

(Potential flow)

A potential flow is a flow whose velocity potential

satisfies Laplace’s equation.

A key property of Laplace’s equation is that it is linear. So we can add two

solutions up to get a third solution.

4.1 Three-dimensional potential flows

We will not consider Laplace’s equation in full generality. Instead, we will

consider some solutions with symmetry. We will thus use spherical coordinates

(r, θ, ϕ). Then

∇

φ =

∂

∂r



∂φ

∂r



sin θ

∂

∂θ



sin θ

∂φ

∂θ



sin

∂

∂ϕ

It is also useful to know what the gradient is. This is given by

u = ∇φ =



∂φ

∂r

∂φ

∂θ

r sin θ

∂φ

∂ϕ



We start off with the simplest case. The simplest thing we can possibly imagine

is if

(

) depends only on

. So the velocity is purely radial. Then Laplace’s

equation implies

∂

∂r



∂φ

∂r



= 0.

So we know

∂φ

∂r

for some constant A. So

φ = −

+ B,

where

is yet another constant. Since we only care about the gradient

∇φ

, we

can wlog B = 0. So the only possible velocity potential is

φ = −

Then the speed, given by

∂φ

∂r

, falls off as

What is the physical significance of the factor

? Consider the volume flux

q across the surface of the sphere r = a. Then

q =

u · n dS =

dS =

∂φ

∂r

dS =

dS = 4πA.

So we can write

φ = −

4πr

When

q >

0, this corresponds to a point source of fluid. When

q <

0, this is a

point sink of fluid.

We can also derive this solution directly, using incompressibility. Since

flow is incompressible, the flux through any sphere containing 0 should be

constant. Since the surface area increases as 4

πr

, the velocity must drop as

in agreement with what we obtained above.

Notice that we have

∇

φ = qδ(x).

So φ is actually Green’s function for the Laplacian.

That was not too interesting. We can consider a less general solution, where

φ depends on r and θ but not ϕ. Then Laplace’s equation becomes

∇

φ =

∂

∂r



∂φ

∂r



sin θ

∂

∂θ



sin θ

∂φ

∂θ



= 0.

As we know from IB Methods, We can use Legendre polynomials to write the

solution as

φ =

∞

n=0

+ B

−n−1

(cos θ).

We then have

u =



∂φ

∂r

∂φ

∂θ

, 0



Example. We can look at uniform flow past a sphere of radius a.

We suppose the upstream flow is u = U

x. So

φ = Ux = Ur cos θ.

So we need to solve

∇

φ = 0 r > a

φ → Ur cos θ r → ∞

∂φ

∂r

= 0 r = a.

The last condition is there to ensure no fluid flows into the sphere, i.e.

u · n

= 0,

for n the outward normal.

Since

(

cos θ

) =

cos θ

, and the

are orthogonal, our boundary conditions

at infinity require φ to be of the form

φ =



Ar +



cos θ.

We now just apply the two boundary conditions. The condition that

φ → Ur cos θ

tells us A = u, and the other condition tells us

A −

= 0.

So we get

φ = U



r +



cos θ.

We can interpret

Ur cos θ

as the uniform flow, and

cos θ

as the dipole

response due to the sphere.

What is the velocity and pressure? We can compute the velocity to be

∂φ

∂r

= U



1 −



cos θ

∂φ

∂θ

= −U



1 +



sin θ.

We notice that u

= u

= 0 when φ = 0, π and r = a.

At the north and south poles, when θ = ±

, we get

= 0, u

= ±

So the velocity is faster at the top than at the infinity boundary. This is why it

is windier at the top of a hill than below.

Note that the streamlines are closer to each other near the top since the velocity

is faster.

To obtain the pressure on the surface of the sphere, we apply Bernoulli’s

equation on a streamline to the surface (a, θ).

Comparing with what happens at infinity, we get

∞

ρU

= p +

ρU

sin

θ.

Thus, the pressure at the surface is

p = p

∞

ρU



1 −

sin



At A and A

, we have

p = p

∞

ρU

At B and B

, we find

p = p

∞

−

ρU

Note that the pressure is a function of

sin

. So the pressure at the back of

the sphere is exactly the same as that at the front. Thus, if we integrate the

pressure around the whole surface, we get 0. So the fluid exerts no net force on

the sphere! This is d’Alemberts’ paradox. This is, of course, because we have

ignored viscosity.

In practice, for a solid sphere in a viscous fluid at high Reynolds numbers, the

flow looks very similar upstream, but at some point after passing the sphere, the

flow separates and forms a wake behind the sphere. At the wake, the pressure is

approximately

∞

. The separation is caused by the adverse pressure gradient at

the rear.

wake

Empirically, we find the force is

F = C

ρU

πα

where

is a dimensionless drag coefficient. This is, in general, a function of

the Reynolds number Re.

Example

(Rising bubble)

Suppose we have a rising bubble, rising at speed

We assume our bubble is small and spherical, and has a free surface. In this

case, we do not have to re-calculate the velocity of the fluid. We can change our

frame of reference, and suppose the bubble is stationary and the fluid is moving

past it at

. Then this is exactly what we have calculated above. We then steal

the solution and then translate the velocities back by U to get the solution.

Doing all this, we find that the kinetic energy of the fluid is

r>a

ρ|u|

dV =

ρU

where



πa



is the added mass of the bubble (and

is the mass of the fluid displaced by

the bubble).

Now suppose we raise the bubble by a distance

. The change in potential

energy of the system is

∆PE = −M

gh.

− M

gh = Energy

is constant, since we assume there is no dissipation of energy due to viscosity.

We differentiate this to know

U = M

h = M

gU.

We can cancel out the U ’s and get

U = 2g.

So in an inviscid fluid, the bubble rises at twice the acceleration of gravity.

4.2 Potential flow in two dimensions

We now consider the case where we have two-dimensional world. We use polar

coordinates. Then we have

∇

φ =

∂

∂r



∂φ

∂r



∂

∂θ

The gradient is given by

u = ∇φ =



∂φ

∂r

∂φ

∂θ



The general solution to Laplace’s equation is given by

φ = A log r + Bθ +

∞

n=1

+ B

−n

)

(

cos nθ

sin nθ

Example

(Point source)

Suppose we have a point source. We can either solve

∇

φ = qδ(r),

where

is the volume flux per unit distance, normal to the plane, or use the

conservation of mass to obtain

2πru

= q.

2πr

Hence we get

φ =

2π

log r.

Example

(Point vertex)

In this case, we have flow that goes around in circles:

Since there is no radial velocity, we must have

∂φ

∂r

= 0. So

only depends on

This corresponds to the solution φ = Bθ.

To find out the value of B, consider the circulation around a loop

K =

r=a

u · d` =

2π

· a dθ = 2πB.

So we get

φ =

2π

θ, u

2πr

It is an exercise for the reader to show that this flow is indeed irrotational, i.e.

∇ × u

= 0 for

r 6

= 0, despite it looking rotational. Moreover, for any (simple)

loop c, we get

u · d` =

(

K the origin is inside C

0 otherwise

We can interpret this as saying there is no vorticity everywhere, except at the

singular point at the origin, where we have infinite vorticity. Indeed, we have

u · d` =

ω · n dS,

where S is the surface bounded by c.

Example (Uniform flow past a cylinder). Consider the flow past a cylinder.

We need to solve

∇

φ = 0 r > a

φ → Ur cos θ r → ∞

∂φ

∂r

= 0 r = a.

We already have the general solution above. So we just write it down. We find

φ = U



r +



cos θ +

2π

θ.

The last term allows for a net circulation

around the cylinder, to account for

vorticity in the viscous boundary layer on the surface of the cylinder. We have

= U



1 −



cos θ

= −U



1 +



sin θ +

2πr

We can find the streamfunction for this as

ψ = Ur sin θ



1 −



−

2π

log r.

If there is no circulation, i.e.

= 0, then we get a flow similar to the flow around

a sphere. Again, there is no net force on the cylinder, since flow is symmetric

fore and aft, above and below. Again, we get two stagnation points at A, A

What happens when

K 6

= 0 is more interesting. We first look at the stagnation

points. We get

= 0 if and only if

cos θ

= 0. For

= 0, when

we require

K = 4πaU sin θ.

So provided

|K| ≤

πaU

, there is a solution to this problem, and we get

stagnation points on the boundary.

For

|K| >

πaU

, we do not get a stagnation point on the boundary. However,

we still have the stagnation point where

cos θ

= 0, i.e.

. Looking at the

equation for u

= 0, only θ =

works.

Let’s now look at the effect on the sphere. For steady potential flow, Bernoulli

works (i.e.

is constant) everywhere, not just along each streamline (see later).

So we can calculate the pressure on the surface. Let

be the pressure on the

surface. Then we get

∞

ρU

= p +



2πa

− 2U sin θ



So we find

p = p

∞

ρU

−

ρK

8π

ρKU sin θ

πa

− 2ρU

sin

θ.

We see the pressure is symmetrical fore and aft. So there is no force in the

direction.

However, we get a transverse force (per unit length) in the

-direction. We

have

= −

2π

p sin θ(a dθ) = −

2π

ρKU

πa

sin

θa dθ = −ρUK.

where we have dropped all the odd terms. So there is a sideways force in the

direction perpendicular to the flow, and is directly proportional to the circulation

of the system.

In general, the magnus force (lift force) resulting from interaction between

the flow U and the vortex K is

F = ρU × K.

4.3 Time dependent potential flows

Consider the time-dependent Euler equation



∂u

∂t

+ ∇



|u|



− u × ω



= −∇p − ∇χ.

We assume that we have a potential flow

∇φ

. So

= 0. Then we can write

the whole equation as

∇



∂φ

∂t

ρ|u|

+ p + χ



= 0.

Thus we can integrate this to obtain

∂φ

∂t

ρ|∇φ|

+ p + χ = f(t),

where

(

) is a function independent of space. This equation allows us to correlate

the properties of φ, p etc. at different points in space.

Example (Oscillations in a manometer). A manometer is a U -shaped tube.

y = 0

We use some magic to set it up such that the water level in the left tube is

above the equilibrium position

. Then when we release the system, the water

levels on both side would oscillate.

We can get quite far just by doing dimensional analysis. There are only two

parameters

g, H

. Hence the frequency must be proportional to

. To get the

constant of proportionality, we have to do proper calculations.

We are going to assume the reservoir at the bottom is large, so velocities are

negligible. So

is constant in the reservoir, say

= 0. We want to figure out

the velocity on the left. This only moves vertically. So we have

φ = uy =

hy.

So we have

∂φ

∂t

hy.

On the right hand side, we just have

φ = −uy = −˙gy,

∂φ

∂t

= −

hy.

We now apply the equation from one tube to the other — we get

h(H + h) +

+ p

atm

+ gρ(H + h) = f(t)

= −ρ

h(H − h) +

+ p

atm

+ gρ(H − h).

Quite a lot of these terms cancel, and we are left with

2ρH

h + 2gρh = 0.

Simplifying terms, we get

h +

h = 0.

So this is simple harmonic motion with the frequency

Example

(Oscillations of a bubble)

Suppose we have a spherical bubble of

radius

(

) in some fluid. Spherically symmetric oscillations induce a flow in the

fluid. This satisfies

∇

φ = 0 r > a

φ → 0 r → ∞

∂φ

∂r

= ˙a r = a.

In spherical polars, we write Laplace’s equation as

∂

∂r



∂φ

∂r



= 0.

So we have

φ =

A(t)

and

∂φ

∂r

= −

A(t)

This certainly vanishes as r → ∞. We also know

−

= ˙a.

So we have

φ = −

˙a

Now we have

∂φ

∂t



r=a

= −

2a ˙a

−

¨a



r=a

= −(a¨a + 2 ˙a

We now consider the pressure on the surface of the bubble.

We will ignore gravity, and apply Euler’s equation at the bubble surface and

at infinity. Then we get

−ρ(a¨a + 2 ˙a

) +

ρ ˙a

+ p(a, t) = p

∞

Hence we get



a¨a +

˙a



= p(a, t) − p

∞

This is a difficult equation to solve, because it is non-linear. So we assume we

have small oscillations about equilibrium, and write

a = a

+ η(t),

where η(t)  a

. Then we can write

a¨a +

˙a

= (a

+ η)¨η +

˙η

= a

¨η + O(η

Ignoring second order terms, we get

ρa

¨η = p(a, t) − p

∞

We also know that

∞

is the pressure when we are in equilibrium. So

(

a, t

)

−p

∞

is the small change in pressure δp caused by the change in volume.

To relate the change in pressure with the change in volume, we need to know

some thermodynamics. We suppose the oscillation is adiabatic, i.e. it does not

involve any heat exchange. This is valid if the oscillation is fast, since there isn’t

much time for heat transfer. Then it is a fact from thermodynamics that the

motion obeys

P V

= constant,

where

γ =

specific heat under constant pressure

specific heat under constant volume

We can take logs to obtain

log p + γ log V = constant.

Then taking small variations of p and V about p

∞

and V

πa

gives

δ(log p) + γδ(log V ) = 0.

In other words, we have

δp

∞

= −γ

δV

Thus we find

p(a, t) − p

∞

= δp = −p

∞

δV

= −3p

∞

Thus we get

ρa

¨η = −

3γη

∞

This is again simple harmonic motion with frequency

ω =



3γp

ρa



1/2

We know all these numbers, so we can put them in. For a

1 cm

bubble, we get

ω ≈ 2 × 10

−1

. For reference, the human audible range is 20

−20 000 s

−1

. This

is why we can hear, say, waves breaking.

5 Water waves

We now consider water waves. We are first going to use dimensional analysis to

understand the qualitative behaviour of water waves in deep and shallow water

respectively. Afterwards, we will try to solve it properly, and see how we can

recover the results from the dimensional analysis.

h(x, y, t)

5.1 Dimensional analysis

Consider waves with wave number

2π

, where

is the wavelength, on a layer

of water of depth

. We suppose the fluid is inviscid. Then the wave speed

depends on k, g and H. Dimensionally, we can write the answer as

c =

gHf(kH).

for some dimensionless function f .

Now suppose we have deep water. Then

H  λ

. Therefore

kH 

1. In this

limit, we would expect the speed not to depend on

, since

is just too big.

The only way this can be true is if

f ∝

√

Then we know

c = α

where α is some dimensionless constant.

What happens near the shore? Here the water is shallow. So we have

kH 

Since the wavelength is now so long, the speed should be independent of

. So

is a constant, say β. So

c = β

gH.

We don’t know

and

. To know that, we would need proper theory. And we

also need that to connect up the regimes of deep and shallow water, as we will

soon do.

Yet these can already explain several phenomena we see in daily life. For

example, we see that wave fronts are always parallel to the shore, regardless of

how the shore is shaped and positioned. This is since if wave is coming in from

an angle, the parts further away from the shore move faster (since it is deeper),

causing the wave front to rotate until it is parallel.

shore

wave fronts

We can also use this to explain why waves break. Near the shore, the water is

shallow, and the difference in height between the peaks and troughs of the wave

is significant. Hence the peaks travels faster than the trough, causing the waves

to break.

H + h

H − h

5.2 Equation and boundary conditions

We now try to solve for the actual solution.

z = 0

z = −H

h(x, y, t)

We assume the fluid is inviscid, and the motion starts from rest. Thus the

vorticity

∇ × u

is initially zero, and hence always zero. Together with the

incompressibility condition ∇ · u = 0, we end up with Laplace’s equation

∇

φ = 0.

We have some kinematic boundary conditions. First of all, there can be no flow

through the bottom. So we have

∂φ

∂z

= 0

when z = −H. At the free surface, we have

∂φ

∂z

∂h

∂t

+ u

∂h

∂x

+ v

∂h

∂y

when z = h.

We then have the dynamic boundary condition that the pressure at the

surface is the atmospheric pressure, i.e. at z = h, we have

p = p

= constant.

We need to relate this to the flow. So we apply the time-dependent Bernoulli

equation

∂φ

∂t

ρ|∇φ|

+ gρh + p

= f(t) on z = h.

The equation is not hard, but the boundary conditions are. Apart from them

being non-linear, there is this surface h that we know nothing about.

It is impossible to solve these equations just as they are. So we want to make

some approximations. We assume that the waves amplitudes are small, i.e. that

h  H.

Moreover, we assume that the waves are relatively flat, so that

∂h

∂x

∂h

∂y

 1,

We then ignore quadratic terms in small quantities. For example, since the waves

are small, the velocities

and

also are. So we ignore

∂h

∂x

and

∂h

∂y

. Similarly,

we ignore the whole of |∇φ|

in Bernoulli’s equations since it is small.

Next, we use Taylor series to write

∂φ

∂z



z=h

∂φ

∂z



z=0

+ h

∂

∂z



z=0

+ ··· .

Again, we ignore all quadratic terms. So we just approximate

∂φ

∂z



z=h

∂φ

∂z



z=0

We are then left with linear water waves. The equations are then

∇

φ = 0 −H < z ≤ 0

∂φ

∂z

= 0 z = −H

∂φ

∂z

∂h

∂t

z = 0

∂φ

∂t

+ gh = f(t) z = h.

Note that the last equation is just Bernoulli equations, after removing the small

terms and throwing our constants and factors in to the function f.

We now have a nice, straightforward problem. We have a linear equation

with linear boundary conditions, which we can solve.

5.3 Two-dimensional waves (straight crested waves)

We are going further simplify the situation by considering the case where the

wave does not depend on y. We consider a simple wave form

h = h

i(kx−ωt)

Using the boundary condition at

= 0, we know we must have a solution of the

form

φ =

φ(z)e

i(kx−ωt)

Putting this into Laplace’s equation, we have

−k

φ +

= 0.

We notice that the solutions are then of the form

φ = φ

cosh k(z + H),

where the constants are chosen so that

∂φ

∂z

= 0 at z = −H.

We now have three unknowns, namely

, φ

and

(we assume

is given,

and we want to find waves of this wave number). We use the boundary condition

∂φ

∂z

∂h

∂t

at z = 0.

We then get

kφ

sinh kH = −iωh

We put in Bernoulli’s equation to get

−iω

φ(z)e

i(kx−ωt)

+ gh

i(kx−ωt)

= f(t).

For this not to depend on x, we must have

−iωφ

cosh kH + gh

= 0.

The trivial solution is of course h

= φ

= 0. Otherwise, we can solve to get

= gk tanh kH.

This is the dispersion relation, and relates the frequency to the wavelengths of

the wave.

We can use the dispersion relation to find the speed of the wave. This is just

c =

tanh kH.

We can now look at the limits we have previously obtained with large and small

In deep water (or short waves), we have

kH 

1. We know that as

kH → ∞

we get tanh kH → 1. So we get

c =

In shallow water, we have

kH 

1. In the limit

kH →

0, we get

tanh kH → kH

Then we get

c =

gH.

This are exactly as predicted using dimensional analysis, with all the dimension-

less constants being 1.

We can now plot how the wave speed varies with k.

√

We see that wave speed decreases monotonically with

, and long waves travel

faster than short waves.

This means if we start with, say, a square wave, the long components of the

wave travels faster than the short components. So the square wave disintegrates

as it travels.

Note also that this puts an upper bound on the maximum value of the speed

. There can be no wave travelling faster than

√

. Thus if you travel faster

than

√

, all the waves you produce are left behind you. In general, if you

have velocity U, we can define the Froude number

F r =

√

This is like the Mach number.

For a tsunami, we have

λ ∼ 400 km, H ∼ 4 km.

We are thus in the regime of small kH, and

c =

10 × 4 × 10

= 200 m s

−1

Note that the speed of the tsunami depends on the depth of the water. So

the topography of the bottom of the ocean will affect how tsunamis move. So

knowing the topography of the sea bed allows us to predict how tsunamis will

move, and can save lives.

5.4 Group velocity

Now suppose we have two waves travelling closely together with similar wave

numbers, e.g.

sin k

x + sin k

x = 2 sin



+ k



cos



− k



since

and

are very similar, we know

≈ k

, and

−k

is small, hence

the cosine term has long period. We would then expect waves to look like this

So the amplitudes of the waves would fluctuate. We say the wave travels in

groups. For more details, see IID Fluid Dynamics or IID Asymptotic Methods.

It turns out the “packets” don’t travel at the same velocity as the waves

themselves. The group velocity is given by

∂ω

∂k

In particular, for deep water waves, where ω ∼

√

gk, we get

This is also the velocity at which energy propagates.

5.5 Rayleigh-Taylor instability

Note that it is possible to turn the problem upside down, and imagine we have

water over air:

water

air

We can imagine this as the same scenario as before, but with gravity pull upwards.

So exactly the same equations hold, but we replace

−g

with

. In deep water,

we have

= −gk.

So we have

ω = ±i

gk.

Thus we get

h ∝ Ae

√

gkt

+ Be

−

√

gkt

We thus have an exponentially growing solution. So the system is unstable, and

water will fall down. This very interesting fact is known as Rayleigh-Taylor

instability.

6 Fluid dynamics on a rotating frame

We would like to study fluid dynamics in a rotating frame, because the Earth is

rotating (hopefully). So this is particularly relevant if we want to study ocean

currents or atmospheric flow.

6.1 Equations of motion in a rotating frame

The Lagrangian (particle) acceleration in a rotating frame of reference is given

+ 2Ω × u + Ω × (Ω × x),

as you might recall from IA Dynamics and Relativity. So we have the equation

of motion



∂u

∂t

+ u · ∇u + 2Ω × u



= −∇p − ρω × (Ω × x) + gρ.

This is complicated, so we want to simplify this. We will first compare the

centrifugal force compared to gravity. We have

|Ω| =

2π

1 day

−1

≈ 2π × 10

−5

−1

The largest scales are 10 × 10

km. Compared to gravity g, we have

|Ω × (Ω × x)|

|g|

≤

(2π)

× 10

−10

. × 10

≈ 4 × 10

−3

So the centrifugal term is tiny compared to gravity, and we will ignore it.

Alternatively, we can show that

Ω ×

(

ω × x

) can be given by a scalar potential,

and we can incorporate it into the potential term, but we will not do that.

Next, we want to get rid of the non-linear terms. We consider motions for

which

|u · ∇u|  |2Ω × u|.

The scales of these two terms are U

/L and ΩU respectively. So we need

ΩL

 1.

This is known as the Rossby number. In our atmosphere, we have

U ∼ 10 m s

−1

and L ∼ 1 × 10

km. So we get

· 10

−4

≈ 0.1.

So we can ignore the non-linear terms. Thus we get

Proposition (Euler’s equation in a rotating frame).

∂u

∂t

+ 2Ω × u = −

∇p + g.

Definition

(Coriolis parameter/planetary vorticity)

We conventionally write

2Ω = f, and we call this the Coriolis parameter or the planetary vorticity.

Note that since we take the cross product of

with

, only the component

perpendicular to the velocity matters. Assuming that fluid flows along the

surface of the earth, we only need the component of

normal to the surface,

namely

f = 2Ω sin θ,

where θ is the angle from the equator.

6.2 Shallow water equations

We are now going to derive the shallow water equations, where we have some

water shallow relative to its horizontal extent. This is actually quite a good

approximation for the ocean — while the Atlantic is around

4 km

deep, it is

several thousand kilometers across. So the ratio is just like a piece of paper.

Similarly, this is also a good approximation for the atmosphere.

Suppose we have a shallow layer of depth z = h(x, y) with p = p

on z = h.

h(x, y, t)

We consider motions with horizontal scales

much greater than vertical scales

We use the fact that the fluid is incompressible, i.e.

∇ · u

= 0. Writing

u = (u, v, w), we get

∂w

∂z

= −

∂u

∂x

−

∂v

∂y

The scales of the terms are

W/H

U/L

and

V/L

respectively. Since

H  L

, we

know

W  U, V

, i.e. most of the movement is horizontal, which makes sense,

since there isn’t much vertical space to move around.

We consider only horizontal velocities, and write

u = (u, v, 0),

and

f = (0, 0, f).

Then from Euler’s equations, we get

∂u

∂t

− fv = −

∂p

∂x

∂v

∂t

+ fu = −

∂p

∂y

0 = −

∂p

∂z

− g.

From the last equation, plus the boundary conditions, we know

p = p

= gρ(h − z).

This is just the hydrostatic balance. We now put this expression into the

horizontal components to get

∂u

∂t

− fv = −g

∂h

∂x

∂v

∂t

+ fu = −g

∂h

∂y

Note that the right hand sides are independent of

. So the accelerations are

independent of z.

The initial conditions are usually that

and

are independent of

. So we

assume that the velocities always do not depend on z.

6.3 Geostrophic balance

When we have steady flow, the time derivatives vanish. So we get

u =

∂

∂y



−



∂

∂y



−

ρf



v = −

∂

∂x



−



= −

∂

∂x



−

ρf



Hopefully, this reminds us of streamfunctions. The streamlines are places where

h is constant, i.e. the surface is of constant height, i.e. the pressure is constant.

Definition (Shallow water streamfunction). The quantity

ψ = −

is the shallow water streamfunction.

In general, near a low pressure zone, there is a pressure gradient pushing the

flow towards to the low pressure area. Since flow moves in circles around the

low pressure zone, there is a Coriolis force that balances this force. This is the

geostrophic balance.

This is a cyclone. Note that this picture is only valid in the Northern hemisphere.

If we are on the other side of the Earth, cyclones go the other way round.

We now look at the continuity equation, i.e. the conservation of mass.

We consider a horizontal surface D in the water. Then we can compute

ρh dV = −

∂D

hρu

· n dS,

where u

is the horizontal velocity. Applying the divergence theorem, we get

∂

∂t

(ρh) dV = −

∇

· (ρhu

) dV,

where

∇



∂

∂x

∂

∂y

, 0



Since this was an arbitrary surface, we can take the integral away, and we have

the continuity equation

∂h

∂t

+ ∇

· (u

h) = 0.

So if there is water flowing into a point (i.e. a vertical line), then the height of

the surface falls, and vice versa.

We can write this out in full. In Cartesian coordinates:

∂h

∂t

∂

∂x

(uh) +

∂

∂y

(uh) = 0.

To simplify the situation, we suppose we have small oscillations, so we have

h = h

+ η(x, y, t), where η  h

, and write

u = (u(x, y), v(x, y)).

Then we can rewrite our equations of motion as

∂u

∂t

+ f × u = −g∇η (∗)

and, ignoring terms like uη, vη, the continuity equation gives

dη

+ h

∇ · u = 0. (†)

Taking the curl of the (∗), we get

∂ζ

∂t

+ f ∇ · u = 0,

where

ζ = ∇ × u.

Note that even though we wrote this as a vector equation, really only the

component is non-zero. So we can also view

and

as scalars, and get a scalar

equation.

We can express ∇ · u in terms of η using (†). So we get

∂

∂t



ζ −



= 0,

where

Definition (Potential vorticity). The potential vorticity is

Q = ζ −

and this is conserved.

Hence given any initial condition, we can compute

(

x, y,

0) =

. Then we

have

Q(x, y, t) = Q

for all time.

How can we make use of this? We start by taking the divergence of (

∗

) above

to get

∂

∂t

(∇ · u) − f · ∇ × u = −g∇

η,

and use (†) to substitute

∇ · u = −

∂η

∂t

We then get

−

∂

∂t

− f · ζ = −g∇

η.

We now use the conservation of potential vorticity, namely

ζ = Q

to rewrite this as

∂

∂t

− gh

∇

η + f · f η = −h

f · Q

Note that the right hand side is just a constant (in time). So we have a nice

differential equation we can solve.

Example.

Suppose we have fluid with mean depth

, and we start with the

following scenario:

Due to the differences in height, we have higher pressure on the right and lower

pressure on the left.

If there is no rotation, then the final state is a flat surface with no flow.

However, this cannot be the case if there is rotation, since this violates the

conservation of Q. So what happens if there is rotation?

At the beginning, there is no movement. So we have

(

= 0) = 0. Thus we

have

(

−

f x > 0

f x < 0

We seek the final steady state such that

∂η

∂t

= 0.

We further assume that the final solution is independent of y, so

∂η

∂y

= 0.

So η = η(x) is just a function of x. Our equation then says

∂

∂x

−

η =

= ∓

It is convenient to define a new variable

R =

√

which is a length scale. We know

√

is the fastest possible wave speed, and

thus

is how far a wave can travel in one rotation period. We rewrite our

equation as

−

η = ∓

Definition (Rossby radius of deformation). The length scale

R =

√

is the Rossby radius of deformation.

This is the fundamental length scale to use in rotating systems when gravity

is involved as well.

We now impose our boundary conditions. We require

η → ±η

x → ±∞

We also require η and

dη

to be continuous at x = 0.

The solution is

η = η

(

1 − e

−x/R

x > 0

−(1 − e

x/R

) x < 0

We can see that this looks quite different form the non-rotating case. It looks

like this:

The horizontal length scale involved is 2R.

We now look at the velocities. Using the steady flow equations, we have

u = −

∂η

∂y

= 0

v =

∂η

∂x

= η

−|x|/R

So there is still flow in this system, and is a flow in the

direction into the

paper. This flow gives Coriolis force to the right, and hence balances the pressure

gradient of the system.

The final state is not one of rest, but one with motion in which the Coriolis

force balances the pressure gradient. This is geostrophic flow.

Going back to our pressure maps, if we have high and low pressure systems,

we can have flows that look like this:

L H

Then the Coriolis force will balance the pressure gradients.

So weather maps describe balanced flows. We can compute the scales here.

In the atmosphere, we have approximately

R ≈

√

10 · 10

−4

= 10

≈ 1000 km.

So the scales of cyclones are approximately 1000 km.

On the other hand, in the ocean, we have

R ≈

√

10 · 10

−4

= 10

= 100 km.

So ocean scales are much smaller than atmospheric scales.