Fluid Dynamics 3 - 2011/2012

Jens Eggers

Preliminaries
Course information
Lecturer: Prof. Jens Eggers, Room SM2.3 Timetable: Weeks 1-12, Tuesday 2.00 (SM2), Thursday 4.10 (SM1) and Friday 11.10 (SM1). Course made up of 32 lectures. Oce hours: my oce hours for this course are Tuesday 3.00 in my room, SM2.3 Prerequisites: Mechanics 1, APDE2 and Calc2. Need ideas from vector calculus, complex function theory, separation solutions and PDEs. Homework: Questions from 10 worksheets set and marked during the course. Will be handed out each Friday, starting in the rst week. Solutions to be returned the following Friday in the box marked Fluids3. Assessment: 2.5 hours exam in April, best 4 out of 5 questions. No calculators allowed. Credit points: will be awarded for serious attempts at 40% of homework. Only comes into play for exam marks in the range 30-40. Web: Standard unit description includes detailed course information. Also a uids 3 web page (http://www.maths.bris.ac.uk/~majge/fluids3.html), which contains: Homework and solution sheets as well as lecture notes (will be put up on the web as soon as possible after each lecture). Lectures: There is no need to take notes during the lecture, as all material relevant for the exam will be put on the web. The main purpose of the lecture is the live development of the material, and a chance for you to ask questions!

Recommended texts
1. A.R. Paterson, A First Course in Fluid Dynamics, Cambridge University Press. (The recommended text to complement this course - costs 50 from Amazon; there are 6 copies in Queens building Library and 3 copies in the Physics Library) 2. D.J. Acheson, Elementary Fluid Dynamics. Oxford University Press 3. L.D. Landau and E.M. Lifshitz, Fluid Mechanics. Butterworth Heinemann

Films
There is a very good series of educational lms on Fluid Mechanics available on YouTube, produced by the National Committee for Fluid Mechanics Films in the US in the 1960s. Each lm is also accompanied by a set of notes. I recommend them highly, and will point out the appropriate ones throughout this course.

The following 3 sections are useful for the course. For the purposes of an examination, I would expect you to know the denition of div, grad, curl and the Laplacian in Cartesians and grad and the Laplacian in plane polars. Other denitions would be provided.

Revision of vector operations

Let u = (u1 , u2 , u3), v = (v1 , v2 , v3 ) be Cartesian vectors. Let (x) be a scalar function and f (x) = (f1 (x), f2 (x), f3 (x)) a vector eld of position x = (x, y, z ) (x1 , x2 , x3 ). Then The dot product is u.v = u1 v1 + u2 v2 + u3 v3 The cross (or vector) product is uv = (u2 v3 u3 v2 ) x+(u3v1 u1 v3 ) y+(u1 v2 u2 v1 ) z The gradient is = , , x1 x2 x3 f2 f3 f1 + + x1 x2 x3 + x f3 f1 x3 x1 + y f1 f2 x1 x2 z

The divergence is f = The curl is f =

f3 f2 x2 x3

The Laplacian is 2 = =

2 2 2 + 2+ 2 x2 x2 x3 1

Formulae in cylindrical polar coordinates

Coordinate system is x = (r, , z ) where the relationship to Cartesians is x = r cos , = x cos + y sin , sin + y cos and y = r sin . The unit vectors are r=x z. In the z. r + f + fz following, f = (fr , f , fz ) fr The gradient is = r 1 + +z r r z 1 (rfr ) 1 f fz + + r r r z

The divergence is f =

r r z 1 The curl is f = /r / /z . r fr rf fz The Laplacian is 2 = 1 2 2 2 1 + + 2 + r 2 r r r 2 2 z

Formulae in spherical polar coordinates

Coordinate system is x = (r, , ) where the relationship to Cartesians is x = r sin cos , y = r sin sin , z = r cos . = x sin cos + y sin sin + z cos , cos cos + The unit vectors are r = x cos sin z sin and = x sin + y cos . y + f . In the following, f = (fr , f , f ) fr r + f The gradient is = r 1 1 + + r r r sin 1 (sin f ) 1 f 1 (r 2 fr ) + + r 2 r r sin r sin r r r sin . /r / / fr rf r sin f r2 r + 1 r 2 sin sin + 2 1 r 2 sin2 2

The divergence is f =

1 The curl is f = 2 r sin The Laplacian is 2 = 1 r 2 r

1
1.1

Introduction & Basic ideas

What is a uid ?

In this course we will treat the laws governing the motion of uids and gases. A uid or gas is characterised by the fact that there is no preferred rest state for the parts it is composed of. If a hole is made in a water bottle, the water will ow out. If a drop is placed on a solid surface, it will spread. By contrast, a solid retains a memory of its original state. If one deforms a piece of metal, it will relax back to its original state once the force is no longer applied. Another important property of liquids and gases is that they are featureless. Viewed from a particular point in space, all directions are equivalent. Solids, on the other hand, often have an internal (lattice) structure. As a result, it makes a dierence in which direction they are deformed relative to their internal structure. Fluid dynamics is an example of continuum mechanics: Defn. A continuum is any medium whose state at a given instant can be described in terms of a set of continuous functions of position x = (x, y, z ). E.g. Density, (x, t), velocity, u(x, t), temperature T (x, t). (These functions usually also depend upon time, t). We know that this description fails if we observe matter on small enough length scales: e.g. typical molecule size ( 109 m) or typical mean free path in a gas ( 107 m). The miracle is that on a scale only slightly larger than that, all microscopic features can be ignored, and we end up with a universal description of all things uid. We will also ignore all eects of incompressibility, so for the purposes of this course, uids and gases are the same thing. We will almost always speak of a uid, but which can mean either a uid or a gas for the purposes of this course. The motion of uids and gases is governed by the same underlying principles

1.2

Viscosity
plate moving at speed U
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

tangential force F

u(y ) = y x

plate area, A

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

stationary plate A very important approximation we will be making throughout this course is the neglect of internal friction (or viscosity) of the uid. For example, consider two plates of area A being moved at speed V relative to each other (see gure). It is common sense that

a certain force is necessary to maintain this motion, just like it takes a certain force to rub one hand against the other, and there is friction between the two. In the case of a uid, the origin of this force is that molecules rub against each other. This way a force is transmitted from one plate to the other. One expects that the necessary shear force is proportional to 1. the speed U 2. the area A and inversely proportional to 3. the distance between the plates. Thus the force is given by UA , (1) where the constant of proportionality is called the viscosity of the uid. In this course we will neglect all viscous eects, we will pretend that the viscosity is small. Note that what constitutes a small viscosity depends on the dimensions of the system: if is suciently small, the force can be made to be signicant, no matter how small the viscosity! There is a wonderful way to avoid further discussion of the matter: if we can neglect all viscous eects, we speak of Ideal Fluids. F =

1.3

What we would like to do

In this lecture course, we will rst develop an equation of motion for the velocity eld u(x, t), which gives the uid velocity at any instant in time t, everywhere in space. This equation (or set of equations) will necessarily have the form of a partial dierential equation. It will be based on Newtons equations of motion, but for a continuum of particles, distributed over space. With the equations in hand, it is down to our ability to deal with the mathematical complexities of solving a partial dierential equation (PDE) to solve physical problems. Some examples of problems dealt with rather successfully using the concept of ideal uids are the following:

There are many wonderful phenomena described quite well within the ideal framework of this course. One of them

are impact problems. The Jesus Christ lizard (see Figure) can walk on water by hitting its surface with its feet. At each impact, it receives an upward impulse, which compensates its weight.

Another spectacular success is the theory of ight. The ideal ow of air around a wing is able to describe the lift necessary for ight, and much more. What is much more dicult is the theory of drag. Inviscid theory suggests that there should be no energy cost to ight at all!

Another huge (and hugely important) area is the theory of water waves. This includes waves from the scale of millimetres up to huge tsunami waves. The absence of any solid boundary results in very little friction, so the ideal theory works very well.

1.4

Outside the realm of ideal uids

Of course in reality there is always at least a little bit of viscosity, so a viscous description will always be more accurate than an inviscid one. If the viscosity is large (think for example of Lyles golden syrup), a theory based on neglecting viscosity is of course spectacularly wrong. We will not be able to describe the pouring of syrup shown in the picture (note the interesting shape of bubbles trapped inside the syrup)! The same is true if the geometry under consideration is small and conned. Then there is a lot of shearing going on, and frictional forces are large.

Unfortunately, there are also many cases where inviscid theory fails even if the viscosity is small, as illustrated in the gure showing the ow around an obstacle. The uid ow does not remain attached to the body, but rather separates from it at the rear. We will see that ideal theory usually predicts the ow lines to hug the body, even in the back. The reason for the failure of ideal uid theory is very subtle. There is a very small region near the body (called the boundary layer), where viscosity becomes important. It is in this small region, not described properly by inviscid theory, where separation occurs. More on this in Fluids 4...

1.5

Lagrangian and Eulerian descriptions of the ow

We now begin to develop a dynamical description of uid ow, which will lead us to formulate a PDE for uid motion, known as the Euler equation. Before we can do that, we must understand the motion of uids a little better. The description of motion is called kinematics. In this chapter, we will deal with kinematics. I encourage you to look at the lm Fluid Mechanics (Eulerian and Lagrangian description) parts 1-3 on YouTube. There are two very dierent ways of describing uid motion, known as the Eulerian and Lagrangian description. Ultimately, they are equivalent, as they describe the same thing. However, they serve dierent purposes, so we need them both. Eulerian description of the ow. This is what the stationary observer sees. Choose a xed point, x to measure, for e.g. the velocity u(x, t). This provides a spatial distribution of the ow at each instant in time. This is the way continuum equations are usually formulated, and our equation of motion will indeed be an equation for the Eulerian eld u(x, t). If the ow is steady, then u does not depend on time, t: u = u(x). Why do we need anything else? The reason is that to make contact with Newtons equations, we need to describe the ow as a moving particle would see it. This is the Lagrangian description of the ow. The observer moves with the uid. Choose a uid particle (for example, we can place a small drop of ink in the uid), and follow it through the uid. Measuring its velocity at a given time, t gives its Lagrangian velocity. Now we describe the whole velocity eld this way, by labelling all material points. A convenient way of doing so is to choose an initial time t0 , and to label all uid particle by their position x = a at that time. Then at time t > t0 , the particle is at x = x(a, t), where x(a, t0 ) = a. Of course, x(a, t) is very interesting in its own right. For example, it describes the course of a balloon, launched at time t0 and at position a into the atmosphere. The Lagrangian velocity is dened as v(a, t) = x t .
af ixed

By denition, it is the velocity of a particle going with the ow. This is precisely how velocity is dened in Newtonian Mechanics. As we said earlier, the Eulerian and the Lagrangian velocity elds contain the same information; the relation between the two is: v(a, t) = u(x(a, t), t). (2)

An example: Consider logs owing along a narrowing section of river. A xed observer measures the velocity by observing the velocity of logs at a given point in space. By observing many logs at dierent positions, he will be able to obtain the entire Eulerian velocity eld u(x, t). Unless the ow conditions are changing, this eld will be timeindependent, u(x, t) = u(x). Now imagine each log being ridden by a moving observer, each of whom reports his velocity as time goes by. If all observers are labelled by their position a at some reference time t0 , this will produce the Lagrangian velocity eld v(a, t).

The logs travel with the uid and will see the ow accelerating, as the river bed becomes narrower. Thus v(a, t) is manifestly time-dependent although the Eulerian eld is not!
Observer moving with the flow sees the logs accelerate

Stationary observer sees logs passing at constant speed.

Defn: A ow is two-dimensional if it is independent of one of its components (in some xed frame of reference). E.g. u = (u, v, 0). Example 1: Let us consider a very simple model ow for the river shown above. Consider the two-dimensional, stationary Eulerian velocity eld u = (u, v, 0), dened by u = kx, v = ky.

As can be seen in the Figure, the ow speeds up as the river contracts, which we can think of as being conned by the river banks, shown as the red lines.

1.6

Particle paths and streamlines

Defn. The particle paths (pathlines) are the paths followed by individual particles. They are determined by the solution of the dierential equation: dx = u(x, t), dt with initial condition x(t0 ) = a = (a1 , a2 , a3 ) (3)

where u is assumed given. This system of equations species a unique curve. For some (simple) u, can be integrated using elementary methods, but in general not. Let u = (u(x, t), v (x, t), w (x, t)), then in components, (3) is dx = u(x, y, z, t) dt dy = v (x, y, z, t) dt dz = w (x, y, z, t) dt

with x(t0 ) = a1 , y (t0 ) = a2 , z (t0 ) = a3 . Example 1: In the example above, the equations become x = kx, y = ky.

The solutions, with initial positions a = (a1 , a2 ) at time t = 0, are x(t) = a1 ekt , y (t) = a2 ekt ,

which describes the path of particles in the ow. Using the relation (2) between Eulerian and Lagrangian elds, we nd v(a, t) = ka1 ekt , ka2 ekt . In particular, the Lagrangian velocity eld is indeed time-dependent, while the Eulerian eld was steady. Now let us consider a case where the Eulerian eld is also time dependent. Example 2: Two-dimensional ow, u = (y, x) cos t, with x(t0 ) = (1, 0). As seen in the Figure, this is a vortex whose strength (and sense of rotation) oscillates.

Then which we can combine as.

dx = y cos t, () dt

dy = x cos t, dt

0=x so that the solution is

dx dy d 1 2 1 2 +y = (2 x + 2 y ), dt dt dt x2 + y 2 = 1.

Now we know that the particle paths lie on a circle, which is not a surprise when looking at the velocity eld. We still need to nd the time dependence x(t) and y (t). We can describe the curve by x(t) = cos (t), y (t) = sin (t),

 = y and from () must have = cos t or = sin t + const. and so dx/dt = sin The constant is sin t0 , since = 0 at t = t0 . Thus the particle paths oscillate around a circle, with angle (t) = sin t0 sin t. As shown in the Figure below, the particle position oscillates back and forth in a section
y e.g. t0=0 =1, t= 3/2 t=0,,2 x

of the circle.

=1, t= /2

Defn: A streamline of a ow u(x, t) at a given instant in time t0 , is a curve which is everywhere parallel to u(x, t0 ). Thus, along a streamline, dx = u(x, t0 ). (4) dt Eliminating t, we nd dx dy dz = = (= dt), t = t0 . (5) u v w Notice that time doesnt go into the denition of a streamline, and we could have taken any quantity to parameterise the curve. In general, a streamline varies with time, but are time-independent for a steady ow. Streamlines can be visualised by taking a short-time exposure of illuminated particles in a ow. Taking an arbitrary starting point, one always continues in the local ow direction. For steady ows, streamlines and particle paths coincide. Evidently, the dening equations (4) and (3) become the same, if u does not depend on time. Example 1: u = (kx, ky ). Since this is a steady ow, we could eliminate t from the pathline to obtain the streamline. Alternatively, we have dy dx = kx kv according to (5). Integrating, we nd ln |x| + ln |y | = C , or |x||y | = eC , which is the equation for a hyperbola. Any such hyperbola is a possible equation for the river bank, which must be a streamline of the ow.

Example 2: u = (y, x, 0) cos t. dx dy dx y = so that = and then integrating, x2 + y 2 = const. At t = t0 , y cos t0 x cos t0 dy x Thus streamlines are circles, and are dierent from pathlines, which just cover a section of a the circle. It is also important to draw arrows into the streamlines, to indicate the local direction of the ow. This has to be done only once for each streamline, since the direction cannot reverse; it is enough to compute the velocity at one point along the streamline. As seen in the Figure below, the direction of ow reverses periodically in our
t0< /2 y /2 < t0< 3/2 y

example. Before we go on, we will revise some material from vector calculus. Appendix A goes in here

1.7

The Lagrangian derivative

(a.k.a. the convective derivative, or the material derivative). We know how to measure the time derivative of a physical quantity associated with the uid, for example that of the temperature T (x, t), at a xed point in space (the Eulerian derivative). Its just T . t This quantity will describe the change of temperature at a xed location, for example air temperature in Bristol. However, this quantity will not be a measure of how a mass of air heats up or becomes colder. The reason is that air is swept away by the prevailing ow eld u(x, t). In other words, to describe the change of temperature of a piece of air, we need to consider the rate of change of T (x, t), following a uid particle. This is called the Lagrangian derivative.

x=ut t x

t+ t x+ x O

Suppose a particle at time t is positioned at x, then at time t + t it has moved to x + ut. The change in T in the instant t is T (x + ut, t + t) T (x, t) Taylor expanding about x, t we get T (x, t) + utTx (x, t) + vtTy (x, t) + wtTz (x, t) + tTt (x, t) + O ((t)2 ) T (x, t). The Lagrangian derivative is (6) divided by t as t 0, so that we get T DT + ( u ) T . t Dt (7) (6)

A particularly important example is the change in velocity (the acceleration) of a uid particle, which we need to apply Newtons equations to uid motion. Of course, the velocity is a vector quantity, which means we have to apply (7) to each component: Du = Dt Du1 Du2 Du3 , , Dt Dt Dt .

The the acceleration of a uid particle becomes u Du = + (u )u. Dt t (8)

Note: The Lagrangian derivative (i.e. following uid particles) is given in terms of Eulerian (i.e. xed point) measurements. It is vital to understand exactly how to compute expressions like (u )u for a given velocity.

Example: Consider an accelerating uid ow, such as the log owing through a narrowing channel. Suppose u = (U + kx, U ky, 0). Then The ow is steady since The advective term is ( u ) u = (U + kx) + (U ky ) x y (U + kx, U ky, 0). u = 0. t

Hence the acceleration of the log is Du = (k (U + kx), k (U ky ), 0) . Dt

1.8

Mass conservation

One of the fundamental laws of continuum mechanics is the law of mass conservation. That is, uid is neither created or destroyed. Consider an arbitrary nite volume, V , which is xed in a xed frame of reference. V is bounded by the surface S and n represents a unit normal on S outward from V . A uid occupies the space of which V is a subset. The uid has velocity u(x, t) and density (x, t). Fluid can ow in and out of V , and its density can change (within V ).

ds V S

The mass contained in V is

V

(x, t)dV . u ndS . The quantity j = u is

called the mass ux density.

The ux of mass into V (n points outward) is

u n

In time dt the uid on the surface dS is transported distance u n in a direction perpendicular to dS. Thus, the mass transported is (u n)dtdS and the rate of mass transport is this quantity divided by dt.

dS

So the rate of change of mass in V must equal the rate of change of mass in/out of V through S . So d (x, t)dV = u ndS dt V S Since V does not change with t and using the divergence theorem for the RHS we get dV = t (u)dV

or
V

+ (u) dV = 0 t

This is true for any xed V , so must have + (u) = 0 t (9)

at every point in the uid. This is called the mass conservation equation or the continuity equation. Remark: The structure of this equation is very general, and applies to any conservation law, which can be written + j = 0. (10) t Here j is the ux of the quantity in question.

1.9

Incompressibility

Defn A uid is said to be incompressible if the density of each uid particle is constant, D (i.e. = 0). Dt From (9) we have 0= D + (u) = + u + u = + u t t Dt

So ( > 0) an incompressible uid satises u = 0. (11)

No uid is completely incompressible, but even gases are often suciently incompressible for (11) to apply to their motion. Incompressibility is a valid approximation if The ow speed is much less than the velocity of sound; Timescales are much larger than (sound frequency)1 . Of course this leaves out everything to do with sound waves, which are due entirely to compressible eects. Before we go on, we will revise more material from vector calculus. Appendix B goes in here

1.10

Streamfunctions (for incompressible ows)

If u = 0, then it follows that there exists a vector eld A(x, t) s.t. u=A The so-called vector vector potential A is complicated unless the ow is two-dimensional in which case it is very simple.

1.10.1

Two-dimensional ows

(i) Cartesians: Here A = (x, y, t) z. In other words, the ow is determined by a single function . This corresponds exactly to the expected degrees of freedom of the ow: there are two components of the velocity, and one constraint u = 0. Then u=A= (and clearly u = 0). Reminder: streamlines are given by v dx udy = 0, dx dy = , where u = (u, v, 0). So u v dx + dy = d = 0 x y , ,0 y x

(by Chain rule). Therefore (x, y, t) = const on a streamline of the ow. Defn: We call the function (x, y, t) the streamfunction of the ow.

= constant

Note: For steady ows, the streamlines do not cross each other and uid does not cross the streamlines. (ii) Cylindrical polar coordinates: Now let A = (r, , t) z in cylindrical polars (r, , z ). Then (conrm ...) u=A= 1 , ,0 r r

where u = (ur , u , 0) (check (i) u = 0 satised again and (ii) streamlines given by = const)

Example 1: A simple source: the ow is purely radial. So u = 0 and ur is independent of . 1 u r = f (r ) = r u = 0 = r Must have = () but / independent of . So = A where A constant. Then f (r ) = A/r . Defn: The source strength is the ux of uid from the source point. The ow is incompressible, so the ux at the origin equals the ux through any closed boundary surrounding the origin. Found, most conveniently, by measuring the ux through a circle radius r centred at the origin. The source strength is
2 2

m=

circle

u ds =

u.nrd =
0 0

A rd = 2A. r

where n = r = (1, 0, 0) in polars. This means the mass ow through a circle surrounding the origin is independent of the radius of the circle, and equals the source of mass ux at the origin. In fact, one can show that the mass ux through any closed surface is m. (how?) Note: In Cartesians, = A tan1 (y/x).

Example 2: The same can be done for a 3D point source The ow is once more radially symmetric, that is we expect u = f (r ) r with the source at the origin. Now the total ux through a sphere of radius r (which should give the source strength) is m= sphere u ndS = 4r 2 f (r ).

Therefore the radial dependence must be f (r ) = m/4r 2, and thus u= m r, 4r 2 (12)

the ow of a three-dimensional point source of strength m.

2
2.1

Flow dynamics for an incompressible inviscid ow

Forces on a uid

Fluids move in response to the forces that on each uid particle. These forces are of two types: Force on parcel of uid, volume V may be proportional to V . These are known as body forces (e.g. gravitational force, V g). Force is transmitted across the surface element S of a uid parcel. I.e. force exerted by the uid on the exterior of V or vice versa. These are surface forces. When the uid is a rest, the surface force must be in the direction of the normal, n. (The denition of a uid is one that remains at rest unless a tangential force is applied). When a uid is in motion, tangential components of the force on S can occur. These are associated with viscosity which describes the eect that one layer of uid in motion has on an adjacent layer. Defn An uid is said to be inviscid when viscous eects are small enough to be idealised as zero (e.g. slidy uids but depends on lengthscales). For an inviscid uid, the surface stress (i.e. Force per unit area) is in the direction n normal to the surface S even when the uid is in motion. That is, the surface force Fs = pnS = force exerted by exterior uid on uid inside V . The magnitude of this force is pS where p(x, t) is the pressure and it is directed inwards because uids are usually in a state of compression.

2.2

Equation of motion

We apply Newtons law to a xed (arbitrary) volume V with surface S . The momentum within V may change as a result of (i) body forces (ii) surface forces, and (iii) uid momentum ow out of V .

n S u dS z y 0 x V dS ( u.n) t

Total momentum in V =
V

udV

In time t, the uid momentum leaving a small section of surface S is u(u n)tS . So rate of momentum transport (i.e. the momentum ux) is u(u n)S . Thus d dt udV = u(u n)dS pndS +
S V

FdV

In each i = 1, 2, 3 components, (ui )dV = t ui (uj nj )dS pni dS +

S V

Fi dV.

since V is xed. Now apply the divergence theorem to get (ui )dV = t (ui uj )dV xj p dV + xi Fi dV.
V

Now since V is an arbitrary volume within the uid, we must have p (ui ) + (ui uj ) = + Fi . t xj xi (13)

This is (one form of) the equation of motion, but it can be simplied by expanding as ui ui ui p + + uj + ui (uj ) = + Fi t t xj xj xi

and using the continuity equation (9) in component form: + (uj ) = 0 t xj so that we have ui ui p + uj = + Fi . t xj xi u + u u = p + F t

Finally, we can identify this in vector notation as (14)

or, using the convective derivative, as Du = p + F. Dt (15)

This is the momentum equation or Eulers equation (due to Euler 1756).

2.3

Hydrostatics

If the body forces are conservative we can write F = where is a potential. Then, if = const, (15) becomes Du = P, Dt where P = p + is a modied pressure (16)

For a uid at rest, u = 0 and so it follows P = 0 , I.e. P = const. In the case of a uid at rest under gravity, the gravitational force on a uid parcel of volume V is V g and so the force per unit volume is g k, where k is the unit vector vertically upwards. Then = gz so that F = (gz ). So, for = const, P = p + gz = const, And for = (z ) (as in the oceans, atmospheres),
z

or, in general, p = F

in the uid

p = (z )g k, 2.3.1
patm

p(z ) = p(z0 ) g

(z )dz
z0

Example: The milk bottle. r

h c

z R

Suppose a milk bottle consists of a smaller cylinder on top of a larger cylinder. Cream of density c occupies the smaller cylinder and oats on milk of density m > c in the larger cylinder below. p c , = g where = m , z mospheric pressure) on z = H + h and p must be rst p(z ) = patm + c g (H + h z ), We need to solve then H <z <H +h and p = patm (at0<z<H continuous at z = H . This gives H <z <H +h 0<z<H

The pressure at the base is p1 (0) = patm + c gh + m gH .

p(z ) = patm + c gh + m g (H z ),

Compare with homogenised (shaked-up) mixture: here h = 0 < z < H + h. So h = m and p(z ) = patm + h g (H + h z ), The pressure on the base is now p2 (0) = patm + h g (H + h) = patm + m g (H + h) r2h (m c ) R2 H + r 2 h

r 2 hc + R2 Hm , for R2 H + r 2 h

0 < z < H + h.

r 2 h(H + h) g (m c ) R2 H + r 2 h

The dierence between the base pressures in problems 1 and 2 is p1 (0) p2 (0) = r 2 h(H + h) hH (r 2 R2 ) h g ( ) = g (m c ) < 0 m c R2 H + r 2 h R2 H + r 2 h

since R > r . So pressure is greater for mixture. Why ?

2.3.2

Archimedes principle ( 250BC)

z S V n Stationary fluid, density

Force on submerged body = = =

V

pndS
S

pdV,

Divergence Theorem k unit normal in z -direction

g kdV,

= gV k Therefore force on body = weight of uid displaced is gV . What about a body intersecting two uids of dierent densities (air/water) ?

2.3.3

Example: Lock gates

patm p atm H1 p(z)

1

p (z) 2

H2

In x < 0, solve

p1 = g with p = patm on z = H1 : z p1 (z ) = patm + g (H1 z ), 0 < z < H1

and similarly, in x > 0, p2 ( z ) = patm , z > H2 , patm + g (H2 z ),

H1

0 < z < H2 ,

Force exerted by uid on gate from x < 0 is F1 =

0 2 1 gH1 p1 dz = patm H1 + 2

Force exerted by uid on gate from x > 0 is

H1

F2 =
0

2 1 gH2 p2 dz = patm H1 + 2

2 2 Net force is 1 g (H1 H2 ). What about turning moment ? 2

2.4

The momentum integral

Mij (ui ) + =0 t xj (17)

From 2.2, equation (13),

where Mij = ui uj + P ij is called the momentum ux (recall P = p + ) and ui is called the momentum density. Note that (17) has the same fundamental structure than (9), as is typical for conservation laws. For example, if I dene the ux of x-momentum px by (p ) fj x = M1,j , then it follows from (17) that px + f (px ) = 0, t which describes the conservation of x-momentum. The equations for the other components are analogous.

For steady ows, LHS of (17) is zero. So 0=

V

Mij dV = xj

Mij nj dS
S

by the divergence theorem. So, back in vector form,

S

u(u n) + P ndS = 0

for any closed surface S in/bounding the uid. This is the momentum integral theorem

2.5

Example: Axisymmetric jet impinging on a wall

jet

. What is the force exterted by the ow on the wall ? Ignore the eects of gravity. n= z u = U z S
film

p = patm

p = patm r u = Ur

n= r

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

n = z

wall
r

Take a control surface S of cylindrical shape that is composed of the free surface of the jet on one side, and the wall on the other. The surface is closed by a circular surface through the jet far from impact (marked jet), and a cylindrical surface where the uid ows parallel to the wall (marked lm). Assume uniform ow parallel to wall in the . Also assume lm region far from impact and across the cross section of the jet u = U z pressure on all boundaries apart from the wall is patm . Then P = p (no body forces) and
S

u(u n) + pndS = 0. patm ndS = (p patm ) zdS

Now pndS =
S S

(p patm )ndS +

wall

since
S

patm ndS =
V

patm dV = 0 and p = patm everywhere else. Also,

S

u(u n)dS =

dS + U 2 z
jet f ilm

2 Ur rdS

 since u n = 0 on the wall. Bringing it all together and equating terms in direction z gives
wall

(p patm )dS = U 2 A,

where A is the cross section of the jet. So there is an extra pressure, and the force exterted by the uid on the wall is U 2 A. [What about Ur and U2 ? Later!]

2.6

Dynamic boundary condition

Consider a surface of discontinuity, C (e.g. interface between oil/water). oil n V C

water

Integrate momentum equation throughout a small disc V of thickness and let 0:

V

(ui )dV 0, t

as 0. Mij nj dS = [Mij nj ]+ (18)

Therefore 0 = lim So
0 V

Mij dV = lim 0 xj

[u(u n) + pn]+ = 0

Provided = const in each layer, the kinematic boundary condition (see 1.9) implies [u n]+ = 0. So component of (18) in direction n is [p ] + = 0 Which just says pressure is continuous across an interface.

2.7

Bernoullis equation for steady ows

1 u u) u u u = ( 2

We start with the vector identity (see Appendix A) where = u is the vorticity. Write u2 = u u and use above in (16) u u2 ) u = (P/ + 1 2 t The ow is steady, so = 0 and taking the dot-product with u on both sides gives t
1 2 u (u ) = 0 = u (P/ + 2 u)

(19)

The denition of a streamline is

dx = u (s measured along a streamline), so ds d dxi 1 2 (P/ + 1 (P/ + 1 u2 ) = u2 ) = u (P/ + 2 u )=0 2 2 ds ds xi along any streamline in the ow (20)

Consequently,
1 2 1 2 P/ + 2 u = p/ + + 2 u = const,

This equation can be seen as a statement of energy conservation, where u2 /2 is the kinetic energy, and the potential energy. 2.7.1 Example: Flow out of a tank

A tank of uniform cross section A0 with a small hole, area Ae at a height h above the base. The height of the uid above the hole is H patm = patm area

Ao

U o

area Ae U e

streamlines

Q: What is the ow speed out of the drain ? First, conservation of mass gives A0 U0 = Ae Ue (21)

(mass uxes across two boundaries equal). Assume steady ow (at any instant, ow is approximately steady). Here, = gz . Streamlines connect the surface to the exit. Along any one of these streamlines apply Bernoullis equation.

1 2 At surface: patm + g (h + H ) + 2 U0 =C 2 Ue =C At exit: patm + gh + 1 2

where C = const. Eliminating C and patm and use (21) to give

2 Ue [1 (Ae /A0)2 ] = 2gH Reasonable to assume that Ae A0 so Ue 2gH . 1) Draining time found by integrating up

Ae dH = U0 dt A0 A0 Ae

2gH.

If the draining time is td and the initial height above the drain is H0 then we get td = 2H0 /g

2) Distance travelled by the jet. The ow speed is a maximum at t = 0 when H = H0 and Ue = 2H0 g . Assume projectile motion of the jet of water. Then time taken to hit 1 2 gt and distance travelled is x = Ue t simplies to x = 2 H0 h. the oor found from h = 2 To nd the value of h for which x is a maximum, for a given ll height Hf = H0 + h we 1 Hf . Then xmax = Hf . need to maximise x = 2 (Hf h)h w.r.t. h and this gives h = 2 2.7.2 Example: Flow through a slowly diverging channel

area A1 U1 streamline

area A2 U2

Mass conservation: A1 U1 = A2 U2 . Bernoullis equation (ignoring gravity):

1 2 U1 2 2 + p1 = const = 1 U2 + p2 . 2

[Note: fast ow implies low pressure and vice versa]

1 2 2 2 2 Hence, p = p1 p2 = 1 (U2 U1 )= 2 U2 (1 A2 2 /A1 ). 2

2.8

The vorticity equation

We have seen that the velocity eld can be decomposed into a straining part and a rotational part, which is given by the vorticity . We now try to nd out what governs the vorticity? We start with Eulers equation written as in (19), namely u u = (P/ + 1 u2 ) 2 t

Taking the curl of this equation gives (u ) = 0 t We can use a vector identity: (u ) = u( ) ( u) + ( )u (u ) = ( )u (u ) since the ow is incompressible and = ( u) = 0. Hence + u = u t or This is called the vorticity equation. In 2D ows, u = (u(x, y ), v (x, y ), 0) and so, by denition = (0, 0, (x, y )) = (x, y ) z. It is then clear that the term u = (x, y ) and so u =0 z D = u Dt

D =0 Dt That is, vorticity is conserved as it moves with the ow. Note: If = 0 at time t = 0, then = 0 for all time. Vorticity cannot be generated in a 2D ow.

2.9

Kelvins circulation theorem

=
C (t)

Defn: Circulation of a velocity eld is dened to be u dl

where C (t) is a closed loop which moves with the uid and dl is an innitesimal line segment along C (t). Note: By Stokes theorem =
C (t)

u dl =

S (t)

( u) ndS =

S (t)

ndS

where S (t) is a surface whose edges connect with C (t). D = 0. Kelvins theorem states that Dt Proof: Consider two points on C (t), x1 and x2 s.t. dl = x2 x1 . In a short time t, x1 x1 + u(x1 )t Now, by Taylor expanding, x2 x2 + u(x2 )t dl dl + (u(x2 ) u(x1 ))t

which is the derivative in the dl-direction, so that dl dl + (dl )ut. That is D d l = ( d l ) u Dt Now Du D Du D (u dl) = dl +u dl = dl + u (dl )u. Dt Dt Dt Dt u u) since, using (22) it is The last term can be written as dl ( 1 2 ui So now, Du D u2 ) (u dl) = dl + ( 1 2 Dt Dt Finally, using Eulers equation (16) (for const) we have D 1 2 u dl = (P/ 2 u ) dl = 0 Dt C (t) C (t) because v d l =
S (t)

u(x2 ) = u(x1 + dl) = u(x1 ) + (dl )u,

(22)

1 ui ui D (dli ) ui = ui (dlj ) = dlj 2 . Dt xj xj

( v ) ndS = 0. Hence result.

Note: Irrotational ow can have circulation if it has point sources of vorticity. E.g in polars, suppose u = (0, A/r, 0). Then = u = 0 apart from at r = 0. But A rd = 2A. r C 0 (steady ow, C is a circle radius r , so dl = (0, rd, 0)). = u dl =
2

Irrotational ows: potential theory

Kelvins circulation theorem states that the circulation around any loop convected by the ow cannot change, if the uid is inviscid. In particular, if the circulation is zero, it will remain zero. Now if there is no vorticity present in the ow at some intitial time, the circulation around any loop in the ow is zero. Hence Kelvins circulation theorem guarantees that there will never be any circulation in the ow, and thus = u = 0 throughout the domain. No vorticity can be created in a ow domain if the viscosity is small. However, we will see in a little while that problems arise with this argument when applied near solid boundaries. The reason is that the circulation theorem applies to loops. However, it is impossible to surround a point on a solid boundary by a loop that stays in the ow. Thus no conclusions can be drawn on parts of the ow that originate from a solid wall. For the time being we simply concentrate on those parts of the ow which are reasonably free of vorticity. In that case, very powerful tools from potential theory can be brought to bear on uid problems, as we will see now.

3.1

The velocity potential

If = u = 0 throughout the domain (apart from at isolated singularities) - i.e. irrotational - it follows that there exists a (x, t) s.t. u = . Then is called the velocity potential or simply the potential. If the velocity eld u is given, the potential is calculated from the path integral over u,
x

=
x0

u dl,

provided of course the ow is indeed potential, i.e. there is path-independence. If the uid is also incompressible then, u = 0 and ( ) = 2 = 0 This is Laplaces equation. When the uid domain communicates with a (moving) solid boundary, we impose the Kinematic boundary condition, 1.9, u.n = f (x) or n. = f (x) for some given f . The eld equation (Laplaces) and the boundary conditions are an example of a Neumann boundary-value problem. An important feature of the equations are that they are linear and so we can use superposition of solutions. This may seem odd, since the Euler equation is a nonlinear equation, and does not obey the superposition principle. However, as we will see very soon, the pressure depends on in a non-linear way (later), which reects the nonlinear character of the Euler equation. 3.1.1 Uniqueness of the velocity potential

Suppose an incompressible irrotational uid occupies a simply connected domain D , so u = and that on the boundaries S of D , u n = f (x).

Suppose that is non-unique. I.e. let 1 and 2 satisfy 2 i = 0 , s.t. 1 = 2 . Let = 1 2

V

and

n i = f (x),

for i = 1, 2

| |2 dV

dV = =

( ) 2 dV, ndS = 0,

and 2 = 0

using divergence theorem

since n = 1 n 1 n = 0 on S . So must have | | = 0 throughout the domain, so = const. So 1 and 2 only dier by a constant, which is irrelevant, as only terms in appear.

3.2
3.2.1

Some simple ows and their potentials

Uniform ow

If the velocity eld is constant, u = U = const, a simple integration yields = U x. , then = Ux, and the streamfunction If U is chosen to point in the x-direction, U = U x is = Uy . 3.2.2 A three-dimensional source

The ow eld has been given in section 1.10.1 cf. (12). Integrating radially from innity, we nd m . = 4r It is shown readily that Laplaces equation is indeed satised: i2 1 xi 3 x2 i = i 3 = 3 + 3 5 = 0. r r r r

The Stokes stream function (which has to be used because the ow is axisymmetric) for this ow becomes simplest in spherical polars, and is dened by A= so (r, ) , r sin

1 1 , u = . sin r sin r Since u = 0 it follows that = (), and then ur = r2 m 1 = 2 . 2 4r r sin

If we dene = 0 for = 0 it follows that m () = (1 cos ). 4 Using simple geometry, can also be expressed in cylindrical polars as = 3.2.3 Stagnation point ow
y

m 4

z2

z + r2

stagnation point

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x

We begin with the two-dimensional case of a ow onto a at plate, see Figure. We consider the simplest case in which the ow is two-dimensional, u = (u, v, 0). There must be a stagnation point u = 0 somewhere along the plate, which we place at the origin, x = y = 0. The x-component will increase linearly as one moves away from the line of symmetry: u = Ax, and correspondingly v = Ay if one moves away from the plate against the ow. Integrating, one nds the potential: A = (x2 y 2 ) 2 It is easy to conrm that this is a solution of Laplaces equation, and thus satises the uid equations even away from the stagnation point. More importantly, it will automatically give the structure of the uid ow close to a (two-dimensional) stagnation point. The stream function is = Axy.

The axisymmetric version of stagnation point ow is relevant if any axisymmetric body is placed in an oncoming stream. The potential is in that case (exercise) = A 2 (r 2 z 2 ) 4

3.3

Bernoullis theorem for unsteady, irrotational ows

Within potential ow theory, we have the slightly confusing situation that we are seemingly solving ow problems only by solving Laplaces equation, which in turn comes from the incompressibility condition u = 0. We are not using Eulers equation of motion (14) to solve ow problems (at least this is almost true; we have been using (14) indirectly inasmuch we used it to derive Kelvins theorem, which motivated the potential ow assumption). However, the Euler equation does come in, albeit in a sneaky way, as we will now show. From (19), the modied version of Eulers equation is u u = (P/ + 1 u2 ) 2 t For irrotational ows, u = , and = 0 so It follows that P 1 2 u + +2 t =0

P u2 = C (t) + +1 2 t

(23)

where C (t) is an arbitrary function of time. Remember P = p + where F = is the body forces in the uid (e.g. P = p + gz for gravity forces). Note: (23) applies throughout the uid, not just on a streamline. Note: By dening a new velocity potential =
t

C (t )dt ,

leads to the function C (t) can always be eliminated from the problem. The potential the same ow eld! (why?) The Bernoulli equation (23) is best viewed as an equation for the pressure. Solving a potential ow problem then becomes a two-step process: First, solve Laplances equation for . Second, given , use (23) to compute the pressure. The pressure depends on u in a non-linear way, a reection of the nonlinearity of the Euler equation.

3.4

The collapse of a spherical bubble

Suppose that at t = 0 we have a spherical bubble, initially at rest, radius R0 in an innite uid. Approximations: (i) neglect gravity; (ii) neglect pressure in the bubble (p = 0); (iii) pressure at innity is constant: p0 > 0.

The pressure dierence causes the bubble to collapse, so R(t) < R0 , where R(t) is radius at time t. Problem: nd R(t) and time of collapse. Introduce velocity potential, u = . The ow is spherically symmetric, so (r, t) = satises 2 = 0, and A = A(t). Now ur = Boundary condition is dR , ur = R dt and hence on r = R, A = R2 R A = 2 r r A r

R2 R r Still dont know R(t) of course. Use unsteady Bernoullis eqn so that (r, t) = 1 +p+ 2 u2 = C (t) = p0 t

since 0 and u 0 as r . Now, on surface of bubble p = 0 (by assumption) and |u| = R, Substituting in gives and + 2R R 2 R2 R = . t R

, so that which is an ODE for R(t). Trick: multiply by 2R2 R d 2 = 3R 2 R 3 + 2R RR 3 = p0 2 R 2 R R3 R dt So 2 = 2 p0 R 3 + C R3 R 3

2 = p0 / 2R 2 + 1R RR 2

= 0 when R = R0 . Hence where C is a constant, determined by the initial condition R 2 = 2 p0 R 3 or dR = dt 2 p0 3

3 R0 R3 R3 3 R0 1 R3 1/2

(24)

If the bubble collapses at time t = tc then R(tc ) = 0 implies (R = R0 u)

1 0

< 0 by assumption. Integrating up where the negative sign must be chosen because R gives 1/2 1/2 t R(t) 2 p0 2 p0 R3/2 dR = d t = t . 3 R3 )1/2 (R0 3 3 0 R0 u3/2 du tc = 3 1 / 2 (1 u ) R0 2 p0 3
1/2

and the integral can be evaluated numerically to a value of 0.747. Hence tc = 0.915 p0
1/2

R0

We can also analyze the collapse during its nal moments, when the radius goes to zero. Such singular events are usually described by power laws, so we try R = A(tc t) . 2 (tc t)22 , and the right hand side of (24) scales as R3 (tc t)3 . Thus Now R 2 2 = 3, and the power law exponent is found to be 2 = . 5 Plugging this back into (24), we nd R=
3 2p0 R0 3 1/5

(tc t)2/5 .

What is remarkable about the collapse is that the inertia of the surrounding liquid is focused into a point by the converging motion. If one calculates the speed of the collapse, one nds (tc t)3/5 , R which is diverging as the radius goes to zero. If the cavity is lled with gas, the gas can be compressed to very high pressures during the last moments, reaching around 10,000 kelvins in the interior of the bubble. This may cause ionization of the gas, which may start to glow. This phenomenon is known as sonoluminescence, since the collapse of the bubble is started o by an external sound eld.

3.5

Flow past a sphere

n

R z

One of the most fundamnetal problems of uid mechanics is to understand the ow around obstacles, which stand in the way of a ow. Equivalently (think of Galilean invariance) one can think of a body moving inside a uid which is at rest (for example an aeroplane). Let us begin studying this problem by considering the perhaps simplest possible body, a sphere of radius R. The sphere is placed inside a uniform stream, which . we choose in the direction of the z or symmetry axis: u = U z Far from the body, the ow will be uniform. The sphere introduces a perturbation to the ow which decays as r , i.e. as one moves away from the body. According to the superposition principle (the Laplace equation is a linear equation), this perturbation is to be added to the potential = Uz describing the uniform ow. The simplest such ow we know is a source, with potential 1/r . However, such a ow cannot describe the physical situation at hand. Far from the sphere, the total mass ow through a closed control surface should be zero, not nite as for a source. Another way of putting it is that the ow eld of a source decays too slowly (like 1/r 2 ) away from the body. This can be helped by taking the derivative of the source solution. Taking the derivative in the -direction, one arrives at the potential x 1 = = 3 , r r which is called the dipole ow (why?). The dipole is oriented in the -direction and has strength ||. This ow is also a solution of Laplaces equation, since 1 r = 1 r = 0.

In summary, our ansatz for the potential is = Uz + The velocity eld is ui = i = Ui3 + i or + u = Uz i xi j xj = Ui3 + 3 3j xj 5 3 r r r x . r3

x 1 3 x 2 . r3 r

Now we have to satisfy the boundary condition on the surface of the sphere, which is u n = 0. The normal vector to the sphere is n = x/r . Now un=U z x 2 4 , r r

UR3 will guarantee that this is zero for r = R. In summary, the z and the choice = 2 solution we seek, which satises all the boundary conditions is = Uz + and the velocity eld is + u = Uz U 2 R r UR3 z , 2 r3
3

(25)

z z3 n . r

(26)

Now we want to calculate the pressure eld, and in particular the pressure on the surface of the sphere. This will permit us to calculate the total force on the sphere. On the surface we have z 3U n , z u= 2 r so 9U 2 z2 u2 = 1 2 . 4 r According to Bernoullis equation p = patm + using that z = cos . r U 2 (U 2 u2 ) = patm + 9 cos2 5 , 2 8

This result has the remarkable property that the pressure distribution is symmetric around the equator of the sphere = /2. Thus while the sphere is pushed by the ow in the downstream direction, there is an equally high overpressure in the back. As a result, without further calculation, we can conclude that the total drag force on the sphere is zero. By denition, the drag on a body in a uniform stream is the force in the direction of the stream. The force perpendicular to the stream is called the lift, which is of course zero if the ow is axisymmetric, as is the case here.

3.6

dAlamberts paradox

At rst sight, one might think that the fact that there is no force acting on a sphere in a stream at steady state is related to the high symmetry of the sphere. In fact, this is not the case: we now show that the force F acting on a body of arbitrary shape vanishes at steady state. At least as far as the component of F in the direction of U is concerned, this is a statement about energy conservation, since F U is the power. Since there is no friction, this energy input cannot be dissipated; the only other possible option would be for the energy to be transported o to innity. The fact that this cannot occur is a statement about how fast the perturbation introduced by the body into the stream decays at innity. S V U S n F

Let us write the total velocity eld in the form u = U + v, where v = is the perturbation introduced by the body. As argued before for the case of the sphere, we will assume that = O r 2 for r . Now the force on the body is calculated from integrating the pressure force over the body: F= pndS ;
S

the minus sign comes from the normal pointing outward, opposite the direction of the = 0, since the ow is pressure force. To nd p, we use the Bernoulli equation (23) with steady: p + U 2 /2 + v2 /2 + U v = patm + U 2 /2. In other words, F= patm ndS +
S

v2 ndS +
S S

(U v)ndS.

The rst integral on the right is zero, as we have seen before. As a mathematical corrolary, we will show below that v2 ndS = 2 v(v n)dS. (27)
S S

But on the surface of the body the normal component of the total velocity vanishes: n (U + v) = 0. Armed with this insight, we have thus shown that Fi = Uj
S

(vi nj vj ni ) dS.

(28)

is zero. However, the components of F normal to U are also zero. Let V be the volume exterior to the body but bounded by a closed surface S very far from the body (see Figure). Then it follows from the divergence theorem that vj vi xj xi dV = (vi nj vj ni ) dS + (vi nj vj ni ) dS ;

From the symmetry of this expression it is clear immedeately that the drag of the body, which is dened as D = FU

note that n points into the volume V . If v decays faster than 1/r 2 at innity, the integral over S goes to zero as the radius of S goes to innity. The volume integral has a component of v as its integrand, which is zero. It follows that the surface integral appearing on the left hand side of (28) is zero, and thus the force is altogether zero. The corollary We now demonstrate (27). From the divergence theorem we conclude that ()2 dV = ()2 ndS + ()2 ndS.

The integrand of second integral on the right behaves like 1/r 6, so its contribution cearly vanishes. We have shown in Appendix A that v2 = 2(v )v for an irrotational ow, and ((v )v)i = vj j vi = j (vi vj ) for an incompressible uid. Thus
S

()2 ni dS =

i ()2 dV = 2

j (vi vj ) dV = 2
V S

vi vj nj dS

by the divergence theorem and using the boundary condition, which proves the corollary. This is a truly remarkable result of potential ow theory, but it also presents a series problem, as absence of drag is clearly not in accord with observation. If our argument were completely airtight, airline companies would surely be doing some serious overcharging! (maybe they do anyway).

3.7

Drag

The experimental picture of a sphere in a stream reveals what the problem is: the ow around the sphere is far from rear-aft symmetric. While the ow in front of the sphere is nicely attached to the body and well described by (26), the back is very dierent. The so-called Reynolds number of the ow is Re = UR 1.5 104 (29)

which is a dimensionless measure of the size of the viscosity . Since the Reynolds number is very large, one might believe that viscous eects are small and the potential ow assumption represents a faithful represenation. However, this is not the case. In particular, the ow does not remain attached to the body, and rather there are uid particles which originate from the surface of the sphere and the enter the liquid. This is also the place where vorticity enters the ow. Clearly, vortices have formed in so-called wake of the ow. Apart from the vortices, the wake appears to be a reatively stagnant region of the ow, which is thus at constant pressure p = pcav , provided it remains closed (which it seems to be in the eperimental Figure). What is the value of pcav ? Let us consider the streamline that separates from the body and bounds the cavity. Since the pressure is constant in the cavity, the pressure on this streamline is also constant, by continuity of the pressure. Such a streamline is called a free streamline. According to the steady Bernoulli equation (20) the velocity along this streamline is also constant. Therefore, there is a jump of the velocity at the free streamline to the almost vanishing velocity in the wake. (Nothing prevents the velocity to be discontinuous if there is no viscosity). Such a surface over which the velocity changes abruptly is called a shear layer or a vortex sheet. Now pressure in the cavity will be the pressure on the separating streamline, as it leaves the body. To do any calculation of the eect of the wake, one needs an estimate of the point where the ow leaves the body, to produce the wake. A very simple criterion (which can be justied in more detail using full viscous theory), is the concept of the adverse pressure gradient. At the front of the body the pressure maximum, and then decreases to a minimum value at the equator, where the speed is maximum; from there the pressure decreases again. It is intuitive and can be motivated in greater detail from the viscous ow theory, that the ow along the body is stable as long as the pressure decreases along the path of a uid particle. If however a uid article has to push against an increasing pressure (adverse pressure gradient), it prefers to leave the surface of the body. Thus on the basis of the pressure distribution which we have calculated, one can give an estimate of the place where eparation is going to occur, which is at = /2 in the case of a sphere. Now we can attempt to recalculate the drag force on the sphere, assuming separation at s = /2. The pressure in the cavity will be 5 pcav = patm U 2 : 8 in our calculation, there is an underpressure in the cavity which pulls the sphere along. We are now in a position to calculate the drag on the sphere, doing the front and the back separately. The result cannot depend on patm which exters equal forces on front and back; therefore, we can put it to zero. Since n z = cos, the force from the front in the ow direction is Fz(f ront) = p() cos d = U 2 R2 4 f ront
/2

sin cos (9 cos2 5)d =

 2 2 U R 4 and from the back

1 0

x(9x2 5)dx =

U 2 R2 , 16

/2

Fz(back) = 2R2

sin cos pcav d =

0

5 2 2 U R 4

xdx =
0

5 2 2 U R . 8

Thus the total force or drag is D = Fz = 9 9 2 2 U R = AU 2 , 16 16

where A = R2 is the projected area of the sphere. For a general body, the drag coecient Cd is dened by D = U 2 ACd , 2 and so Cd = 9/8 for our calculation of the sphere. We hasten to add that our calculation is not particularly good, see the Figure. In the relevant range of Reynolds numbers, CD 0.5. Our intention is to give a general idea of where the drag is coming from, and

how dAlamberts paradox arizes.

3.8

Impact theory

If a body impacts on a uid surface, an impulse upward force results. The surroundings of the projectile show a characteristic splash prole, over which the the parts close to

the projectile show the greatest upward velocity. We now calculate both the impulse as well as the velocity prole right after impact. Our assumption is that the speed of impact is so high that the acceleration of the uid

u is large, but there is no signicant transport yet, so (u )u is small in comparison. t Thus the Euler equation simplies to 1 u = p. t p
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx ta tb

Now the acceleration and thus the pressure will be very great at the moment of impact, and small before and after: the pressure as function of time has a very sharp peak. In = F we have to calculate the time integral over the peak view of Newtons equation p (the area of the shaded region) to calculate the total momentum transferred to the body. This object is known as the impulse. We thus have 1 u(a) u(b) = (a) (b) =
ta tb

pdt ,

where is the impulse per unit area, the superscripts refer to a time tb just before impact, and a time ta just after impact. Now (b) will be constant (describing a uid at rest before impact), and integrating we nd (a) = + const.

free surface

U
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx body xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

=0

n = n U Clearly, the impulse is zero along the unperturbed free surface of the liquid (see Figure). Thus the potential is constant along the free surface; we can choose this constant to be zero. On the lower side of the body, the uid has to move with the impact velocity U of the body: U n = u n = n . This denes the boundary conditions for on the boundaries of the uid domaine V , bounded by the free surface on one hand, and the underside of the body on the other.

In general this problem is quite dicult to solve. However in the special case that the free surface lies in a symmetry plane of the body, the calculation becomes equivalent to the problem of the body moving with speed U in an innite expanse of uid. As an example, let us consider a sphere impacting on a bath of uid, such that its lower half is wet by the uid. Equation (25) is the potential for a stationary sphere in a stream of , this is equivalent to the sphere moving in speed U . If one subtracts the velocity u = U z the negative z -direction in a stationary uid. Thus the solution we seek (sphere moving in the z -direction) is UR3 z . = 2 r3 It is checked directly that = 0 in the plane of symmetry z = 0, and by construction on the sphere (there only is a z -component), we have to integrate nz = cos over the front Sf of the sphere: Fz dt = On the surface of the sphere, (a) = Fz dt =
Sf

U n = n . Thus this is the solution we seek. To calculate the total impulse

Fz dt

nz dS.
Sf

UR cos , and so 2
1

(a) nz dS =

UR 2

/2 0

2R2 sin cos cos d = UR3

x2 dx = UR3 3

(pointing opposite the impact direction).

The velocity in the z = 0 plane is U = uz = z 2 R r

3

and ur = 0. After the impact, uid particles will continue to travel along straight lines in the direction of this velocity eld, so the free surface will have a shape given by the above velocity prole! This is conrmed by comparison to the experimental picture.

There is a reptile, called the Jesus Christ lizard or Basiliscus basiliscus, who supports its weight by a rapid succesion of impacts with their feet. (There are some nice movies of it on youtube). To do that, they have to satisfy the equation Fz dt = MgTstep For more information, see J. W. Glasheen and T. A. McMahon, Nature 380, 340 (1996). As a model for the feet of the lizard, they choose a disc of radius R. Using elliptical coordinates, it can be shown that the potential of a disc moving at speed U along its axis z is 2UR (1 arctan ) = in front of the disc, where z = R, r=R (1 2 )(1 + 2)

are cylindrical coordinates. In the plane z = 0 we have = 0, and thus = Thus the impulse on the plate is
1

2UR

r R

Fz dt =
Sf

(a) nz dS = 4UR3

1 2 d =

4UR3 3

(not too far from the result for a sphere). Glasheen and McMahons measurements in Phys. Fluids 8, 2078 (1996) agree very well with this answer.

Two-dimensional ows

If the ow is in the plane, and there are only two independent variables x, y , the ow problem is much simplied. Moreover, we will see that some of the physics is fundamentally dierent from three dimensions. One fact we know of already is that no vorticity is created in two dimensions, is simply convected with the ow. Of course, real ows are never truly two-dimensional; however if the geometry extends very far in one direction (think of the wing of an aeroplane), a two-dimensional description is a good approximation.

4.1

Flow past a cylinder

. This is very Find the ow around a stationary cylinder in a steady stream U = U x closely analogous to the ow around a sphere; the eect of the sphere is modelled by a (two-dimensional) dipole, which is the derivative of a source, whose velocity eld behaves like ur 1/r (section 1.9.1). Thus the potential is = ln r , and a dipole in the -direction has potential x = ( ) ln r = 2 , r The ansatz for the velocity potential is thus = Ux + and so x , r2

2( x)x . 2 r r4 Now the boundary condition for r = R is u n = u x/r = 0, and from the velocity eld we nd x x un=U 3 . r r Thus the boundary condition is satised if we choose + u = Ux = R2 U . Thus the nal answer for the velocity eld is u= R2 [U 2n(U n)] + U . r2 (30)

Now we plug this into Bernoullis equation to compute the pressure: u2 p U 2 patm + = + . 2 2 On the surface u2
r =R

= (2U 2n(U n))2 = 4U 2 4(U n)2 = 4U 2 (1 cos2 ) = 4U 2 sin2 , p = patm +

U 2 (1 4 sin2 ). 2 The pressure is once more symmetric about = /2, i.e. about the midsection of the cylinder. It follows that the total force on the cylinder is again zero.

so in other words

4.2

Non-uniqueness of the potential

One particularly important aspect of two-dimensional ow is that any solid body placed in the ow domain will create a domain that is no longer simply connected. Defn: A closed curve C is reducible in a domain D if it can be shrunk to a point without ever leaving D . If every closed curve is reducible then D is simply connected. E.g. (i) if D is the interior of a circle, then it is simply connected. (ii) if D is the exterior of a circle then it is not simply connected.

ii

D D
Example: Let D be the domain a < r < b, 0 < < 2 in cylindrical polars. D is not simply connected. Dene a ow in this domain, u = (0, /2r, 0), which is called a point vortex of strength . At r = 0 the ow is singular, but outside it is irrotational [Check] and has circulation . Hence there is a velocity potential s.t. 1 = 0, u = = /(2r ). The solution is ur = r r (r, ) = ( + A), 2

for any constant A. Theres a problem here: for example, (r, 0) = (r, 2 ) so the potential is discontinuous along = 0 or, alternatively, is multivalued. Evidently, whenever there is circulation = dl ,
C

a nite amount of potential is picked up as one goes around the curve. Such a ow is thus necessarily multivalued. In practice, one places a branch-cut in D to make the domain simply connected. In simply-connected domains, is unique:

b a Needs a branch cut to make domain simply connected and the potential singlevalued

4.3

Steady ow past a circular cylinder with circulation

In the previous sections we have shown that the potential for the steady ow past a cylinder of radius R is R2 =U r+ cos r (cylinder xed at origin, ow speed U from left to right). Suppose now that there is circulation round the cylinder, by adding a vortex rotating in the clockwise direction: =U r+ R2 r cos 2

This is not a single-valued function. The streamfunction is =U r R2 r sin + log(r/a) 2

satisfying = 0 on r = R. Other streamlines given by = const. Find the stagnation points (u = (0, 0)) with, on r = R ur = u = So u = 0 implies sin = 1 = U r =0 r 1+ R2 R2 sin 2R

4UR

which has two roots in 0 < < 2 (and hence two stagnation points) if /UR < 4 .

3 . If /UR > 4 then there If /UR = 4 then two stagnation points coincide at = 2 are no stagnation points: To calculate the force on the cylinder, use unsteady Bernoulli, 1 p+ 2 u2 = C

noting that the ow is steady (C (t) = C , /t = 0). With p patm , u2 U 2 as r , 1 U 2 1 u2 . p = patm + 2 2 On cylinder surface, u = (ur , u ) = (0, /r )|r=R = (0, 2U sin + /2R) and with n = (cos , sin ),
2

F=

pnR d
0

= Fx = 0 (no drag force, as before). But the transverse force is So F x

2

= Fy = F y =
0

0 2

1 U 2 1 (2U sin + /2R)2 sin R d p0 + 2 2 1 4U 2

sin (/2R) sin R d

= U so there is an upward lift force on the cylinder (Magnus Eect - 1853). This can explain the forces acting on spinning objects (for example ping-pong balls). The eect is also closely related to the lift forces on airplane wings, to which we will come back in more detail. The big question in that case of course is what sets the value of !

4.4

The complex potential

We now introduce the formulation of potential ow using a complex formulation, where each point in the xy plane is represented by a complex number z . Only this formulation reveals the full power of two-dimensional methods. The rst remarkable result is that the potential and the streamfunction are simply the real and imaginary part of the same complex potential.

Now let z = x + iy and dene the complex potential w (z ) = (x, y ) + i (x, y ). If w (z ) = w (x, y ) is analytic then dw = +i = i + dz x x y y

Let u = (u, v, 0). The ow is irrotational and incompressible as in chapter 3, so u = and exists such that (see 3.2.1) = u= x y (31) = v= y x

(w (z ) is independent of the direction of the derivative). Eqns (31) are the Cauchy-Riemann equations and are satised by any analytic functions. Properties: (i) 2 = 2 = 0 follow directly from (31) (ii) = + = 0 meaning equipotential curves (where = const) x x y y are perpendicular to streamlines, (where = const)

(iii)

dw = u iv = q ei where q = (u2 + v 2 )1/2 is the speed of the ow, is the angle dz dw the ow makes to the x-axis; is known as the complex velocity (note the minus dz sign)! Example: straining ow

4.4.1

Hence w (z ) = + i = k (x2 y 2) + i2kxy = kz 2 is the complex potential. 4.4.2 Example: Line source

Straining ow, = 2kxy , k a constant. Then u= = 2kx = y x , = 2ky = v= x y

= k (x2 y 2)

In three dimensions, a two-dimensional source is a line of sources pushing uid radially outward. The ow eld (see 1.9.1 (ii)) is u= m r, 2r m . 2

while the potential and stream functions are = m log r, 2 =

It is recognised immediately that these are the real and imaginary parts of the complex potential m log z, w= 2 since in polars z = r ei where r = |z | and = arg(z ) and so log(z ) = log(r ) + log(ei ) = log(r ) + i. 4.4.3 Example: Line vortex u= , 2r

A vortex has the ow eld

, which is the same as the streamfunction of a source! Thus and the potential is = 2 we know directly that the complex potential of a vortex must be the same as that of a source, up to a complex rotation: i log z. 2 In particular, we can conclude without further calculation that the stream function is w= ln r. 2 This is a simple example for the application of a powerful theorem from complex analysis: once one knows either the real or the imaginary part of an analytic function, the whole complex function is determined completely. In three dimensions, this will be a line vortex which is completely straight and perpendicular to the xy plane. In general, line vortices are curved, but we disregard this eect here. = Im{w } =

4.5

Interaction of vortices

A line vortex intersects a plane in one point, in two dimensions one can thus describe the vortex by a point (the position z0 = x0 + iy0 of the vortex, and its strength . According to Kelvins theorem, the circulation is conserved, and thus the strength of the vortex is constant in time. The only thing we need to keep track of is the position of the vortex. In complex notation, the velocity eld generated by a vortex located at z0 is u iv = i 1 i z z0 i d log(z z0 ) = = 2 dz 2 z z0 2 |z z0 |

Now each vortex moves in the velocity eld generated by all the other vortices. In the simplest case of two vortices of strength 1 and 2 , located at (x1 , y1) and (x2 , y2), respectively, the equations of motion are 2 y1 y2 dx1 , = dt 2 (x1 x2 )2 + (y1 y2 )2 dy1 2 x1 x2 , = dt 2 (x1 x2 )2 + (y1 y2 )2

and correspondingly for the other vortex. The generalisation for N vortices is obvious and involves the sum over all the vortices except for the one that is being convected. This system of equations is beautiful and simple, but things are even better: the system has a Hamiltonian structure! The Hamiltonian is H= 1 2 ln (x1 x2 )2 + (y1 y2 )2 . 4 |i |(xi , sign(i )yi ), then the equations become dy i H = , dt x i

If now one denes variables ( xi , y i ) =

H dx i = , dt y i

which are the usual Hamiltons equations. One particular of this is that since H is conserved, the distance between the two vortices must stay the same. How does the motion of two vortices look like ? Let us look at the two possible cases: 4.5.1 Two corotating vortices
1 2h

h 1
2 2h

G R1 R2

The vortex strengths are 1 and 2 , both being positive (the two vortices rotate counterclockwise). Let the distance between the two vortices be h. The speeds of the two vortices are then 2 1 U1 = , U2 = , 2h 2h respectively. From the construction it is clear that at each instant, the vortices move in a direction perpendicular to the line connecting them. This means they have to go around in circles, whose centre G is along the line connecting the vortices. The point G is stationary in space. The some of the radii R1 , R2 must satisfy R1 + R2 = h. Since both vortices make a full turn in the same time, we must have 2 R1 = . R2 1

For two equal vortices of strength 1 = 2 = , it follows that R1 = R2 = R. This means the vortices are always at the opposing sides of a circle of radius R. The whole conguration rotates at an angular velocity of = 4.5.2 Two counterrotating vortices . 4R2

Now consider the case that the two vortices have opposite sign. This situation is produced easily by the stroke of a paddle. In fact since the total vorticity of the system should remain zero according to Kelvins theorem, typically 1 = 2 . After the paddle is withdrawn, the vortices continue to move according to the other vortex velocity eld. Let the distance between the two vortices again be h. R2
2 2hh 1 2h

1 R1

The vortices now move in the same direction, and rotate about the same centre. The radii of this circular motion obeys the same relationship has before. However, we now have 2 h R2 R1 = h, and thus R1 = . Thus when the vortices are equal and opposite, the 1 2 radii of both circles tend to innity. In this case the vortices move in a straight line, and translate uniformly at speed U = . This is whats seen in the experimental pictures 2h of two vortices being created behind a paddle.

4.6
4.6.1

The method of images

A vortex next to a wall
2

1 b

=0

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

2 Consider the motion of a vortex located at some position z0 = a + ib in the upper half of the complex plane. The x axis is the location of a solid wall. Thus we have to satisfy the boundary condition of no ow through the wall: v = 0 for y = 0. Simply taking the ow eld of the vortex, u iv = i x a i(y b) , 2 (x a)2 + (y b)2 (32)

does not satisfy this condition: the ow eld must be modied by the presence of the wall.

The method of images is a technique that permits to nd solutions to Laplaces equation which satisfy the right boundary condition on the solid wall. Near the vortex at z0 , the solution should look like (32). The method of solution is indicated in the gure: imagine replacing the wall by another image vortex that is placed at an equal distance on the other side of the wall. Its position is where an image in a mirror would appear. And equally like a mirror image, the sense of rotation of the image vortex is reversed. By symmetry it is clear that the v component of the image vortex has opposite sign, so on the line of symmetry (the locus of the wall), the two contributions cancel and v = 0. The boundary condition on the wall is satised and we have solved our problem! In other words, the ow problem we want to solve (one vortex in the presence of a wall) is exactly the same as the ow problem of two vortices (the original vortex and its image) without

the wall. Lets check if our reasoning was correct. It is easiest to do the calculation using complex notation. If the vortex is at z0 , its image is at z0 . Since the image also has the opposite sense of rotation, there is a minus sign in front of it. The total potential (vortex + image) becomes i i w (z ) = log(z z0 ) + log(z z0 ). 2 2 We have to verify that Im{w } = = const for z = x real! Now since ln(z ) = ln(z ) (why?) we have i i w (z ) = log(z z0 ) log(z z0 ). 2 2 But if z is real, z = z , and thus w (x) w (x) = 0, and so = 0 along the wall. In fact, we have shown that the problem of a vortex near a wall is mathematically equivalent to the problem of two counterrotating vortices of equal strength, which in the previous section we showed to move at constant speed in the direction perpendicular to the line connecting them, in other words, parallel to the wall. Let us check this fact as well, using complex notation. The vortex in question moves in the ow eld of its image, i.e. i x a i(y + b) , u iv = 2 (x a)2 + (y + b)2 u iv = i 2 2ib 4b2 = , 4b

evaluated at z = a + ib. Thus

and the vortex moves with a horizontal speed of , which corresponds to the earlier 4b result with h = 2b. 4.6.2 A recipe

This leads us to the following result, which permits to nd the ow generated by any combination of singularities next to a wall. Let the wall be the real axis of the complex plain, and let f (z ) describe all singularities (vortices, sources, etc.) present in the ow i ln(z z0 ) for a vortex and f (z ) = domain y > 0. For example, f (z ) = 2 2 (z z0 ) for a dipole. Thus w (z ) = f (z ) would be the complex potential of the ow without the wall. We claim that w (z ) = f (z ) + f (z ) is the potential in the presence of the wall. In the second member, we take the complex conjugate of everything except of z itself. We have to prove that = Im{w (z )} = 0 if z = x is real. But if this is the case then z = z , and so w (x) w (x) = f (x) + f (x) (f (x) + f (x)) = 0, which proves the theorem.

To apply this, consider a source of strength m placed at z = ib above a wall. Then m m f (z ) = ln(z ib) and f (z ) = ln(z + ib). Thus the solution we are seeking is 2 2 w (z ) = m m (ln(z ib) + ln(z + ib)) = ln(z 2 + b2 ). 2 2

4.7

von K arm an vortex street

The Figure shows a cylinder in a uniform stream at a Reynolds number Re = 105. The ow is not steady, but the cylinder sheds a periodic sequence of counterrotating vortices, which form a characteristic staggered pattern. Von K arm an became interested in the problem problem when he was appointed as a graduate assistant in Goettingen in Prandtls laboratory in 1911. Prandtl had a doctoral candidate, K. Hiemenz, to whom he had given the task of constructing a water channel in order to observe the separation of the ow behind a cylinder. Much to his surprise, Hiemenz found that the ow in his channel oscillated violently, and he failed to achieve symmetrical ow about a circular cylinder. Von K arm an writes: When Hiemenz reported this to Prandtl, the latter told him: Obviously your cylinder is not circular. However, even after very careful machining of the cylinder, the ow continued to oscillate. Then Hiemenz was told that possibly the channel was not symmetric, and he started to adjust it.

I was not concerned with this problem, but every morning when I came into the laboratory I asked him, Herr Hiemenz, is the ow steady now? He answered very sadly, It always oscillates. xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx b xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx

a
Von K arm an modelled the ow by an innite array of point vortices as pictured. Suciently far behind the obstacle that produces the street, the eect of the obstacle can be neglected. The characteristic feature of the street is that the pattern is stationary in the frame of reference of the obstacle. To achieve this, x a particular vortex and consider the velocity eld it experiences. The velocity generated by all the other vortices has to compensate exactly the velocity u = U of the stream. We take the street as innitely extended in both directions. Thus there is one set 1 of vortices of strength at z = na, and another row of vortices at z = (n + )a + ib, 2 where n = 0, 1, 2, . . .. Just a little bit of thought shows that all the contributions to the velocity of one vortex coming from the same row cancel exactly, so we only have to consider the other row. Now x a vortex, at z = a/2 + ib, say. The contribution of each of the members of the i other row is ln(z na), which is the same (up to a constant), as 2 Thus the total contribution is i w= 2

i z ln 1 . 2 na z i ln z = na 2 = z i a ln sin , 2 a

, n=0

ln 1

z2 i ln z 1 2 2 2 na n=1

using a well-known relation from complex functions. Now the complex velocity is u iv = and at the position of the vortex dw dz =
z =a/2+ib

i z dw = cot , dz 2a a tanh 2a b . a

z i tan 2a a

We thus have the condition that U= tanh 2a b a

for the vortex street to stay in place.

xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx b xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxxxxxxx

a
Why have we been assuming a staggered conguration rather then a symmetric one, as shown above? The reason is that this conguration is unstable. Imagine shifting one of the vortices to the right. The main contribution to its motion comes from the vortex just below. As a result, the vortex will move farther away from it, and the horizontal velocity it receives becomes smaller. As a result, it will be convected even further downstream. Clearly, this is an unstable conguration. In the staggered conguration, this argument does not apply: while the vortex moves farther away from its upstream neighbour, it comes closer to its downstream neighbour. However, there is only one particular ratio of a/b for which the staggered conguration is completely stable. A more detailed calculation shows that cosh or b a = 2,

b = 0.283 if the conguration is to be stable. a

4.8

Mappings and transformations

The most powerful tool of complex analysis consists in mapping the ow domain into another. As soon as I have found a mapping that is conformal, I have transformed the problem from one geometry to another.
xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx D xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx z0 xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

= f (z )

H 0

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Imagine we want to solve a ow problem in a given domain D . This means we are looking for a function w (z ) (the complex potential) with the following properties: w (z ) is analytic in the domain D , except perhaps for one or more singularities at z0 (and other places).

 Im{w (z )} = const on the boundaries of the domain. Now imagine that = f (z ) is a mapping that is analytic everywhere in D , and which maps onto a new domain D1 . Then the singularity at z0 is mapped to a new place 0 = f (z0 ) and the boundary of D is mapped to the boundary of D1 . For D1 one will choose a very simple domain such as the upper half plane H, for which I can solve the ow problem; let this solution be w1 ( ) (which has a singularity at 0 ). Then w (z ) = w1 (f (z )) is a solution to the original ow problem. To prove this statement, it is enough to observe that since w1 ( ) is analytic in D1 , w (z ) is analytic in D . This is the rst requirement. In addition, Im{w ( )} is constant on the boundary of D1 . But this means that Im{w (z )} = Im{w1 (f (z ))} is also constant, which meets the second requirement.

4.8.1

Example: A source in a corner

The function = f (z ) z 2 , z = 1/2 maps the rst quadrant onto the half-plane. Namely, if z = rei , then = r 2 e2i = r 2 ei . Thus the domain 0 < /2 is mapped 2 onto 0 < . The source at z0 maps to a source at 0 = z0 . Now the solution in the -domain is m m w 1 ( ) = log( 0 ) + log( 0 ) 2 2 as we have seen before. So w (z ) = w 1 (z 2 ) = m 2 log(z 2 z0 ) + log(z 2 z 2 0) 2 m = (log(z z0 ) + log(z + z0 ) + log(z z 0 ) + log(z + z 0 )) 2

The plot is for the case z0 = 2 + i. 4.8.2 Example: rotation

Consider any ow problem, for example the ow around a cylinder: w 1 ( ) = U + R2 .

Let = z ei f (z ), which is a rotation of the axes by in the anticlockwise direction. Namely, for z = r ei , = r ei() .

Then in the z -plane, w (z ) = w1 (zei ) = U Note: Velocities are most easily found from dw dw1 d = u iv = dz d dz E.g. in above u iv = (u1 iv1 )ei . 4.8.3 Example: a channel z ei + R2 i e . z

Another useful mapping is the exponential = f (z ) eKz , (z = ln /K ), which maps the upper half plane to a channel of width /K . Namely, if = | |ei then z = ln | |/K + i/K . Since 0 < , the imaginary part of z varies between 0 and /K . Here the positive x-axis is mapped onto the lower boundary of the channel, the negative x-axis on the upper boundary. Now imagine a vortex situated in the channel of width at z0 = i. Then 0 = ei = cos(1) + i sin(1). The potential in the half space is w 1 ( ) = and thus w (z ) = is the solution. i ln( 0 ) ln( 0 ) , 2 i ez ei ln z 2 e ei

4.9

The Joukowski mapping: circles to ellipses

A particularly useful application of the mapping idea concerns the ow around bodies. We have solved the problem of the ow around a cylinder. Thus if we can nd a conformal mapping between the unit circle and any given shape, we have solve the ow problem around this shape. b2

z=+

R z 2c
Consider the (inverse) mapping z=+ b2 , (33)

2d

called the Joukowski mapping. Consider a circle of radius R, whose surface in the -plane is described by the polar representation = Rei . Under the Joukowski mapping, z = Rei + b2 i e = R R+ b2 R cos + i R b2 R sin = c cos + id sin

is an ellipse with axes length 2c and 2d. The semiaxes come out to be c=R+ b2 , R d=R b2 . R

Now consider the uniform ow past an ellipse. To model an arbitrary angle between the direction of the ow and the semi-major axis, we consider the ow around a cylinder that approaches the x-axis under an angle : w 1 ( ) = U ei + R2 i . e

In principle, one can nd w (z ) using the inverse of the Joukowski mapping 1 = f (z ) = (z + z 2 4b2 ), 2 so that w (z ) = w1 (f (z )). However, the resulting expressions are often not so useful. For example, to nd the streamlines, it is much easier to nd the streamlines of the w1 ( ) in the -plane, and then to transform them using (33). This is how the pictures were produced. If one wants to calculate the velocity, one uses u iv = dw dw1 1 U (ei R2 ei / 2 ) = = . dz d dz/ d (1 b2 / 2 ) 2iU sin( ) U (ei ei e2i ) = i . 2 2 2 i (1 (b /R )e ) (e (b2 /R2 )ei )

On the cylinder, = Rei , so u iv =

This means there are stagnation points at = and = + . This point is where a streamline leaves the surface. In other words, this streamline (plotted in red) has the same value of the streamfunction then the surface of the ellipse.

4.10

Lift

Now we want to y! In principle, we know how to construct the ow around a wing of arbitrary shape, we only have to nd the transformation, starting from a circle. We have seen already that the key ingredient is to have circulation around the wing. We have calculated the lift in the case of a cylindrical cross-section, but what is it for an arbitrary shape?

4.10.1

Blasius theorem

y t = (cos , sin ) n = (sin , cos ) x

We now calculate the force on a xed body in a stream, using complex notation. This is a close relative of the calculation done in section (3.6), leading to (28). Suppose a xed rigid body, boundary C is in a steady ow, generating a potential w (z ). We know dw = u iv = q ei dz so |u| = q (reminder: is the angle the ow direction makes to the horizontal). So, according to Bernoulli (no gravity):
1 q 2 , p = p0 2

where p0 = patm + U 2 /2 is a constant. Let s be the arclength along C , which we use to integrate over the surface of the body. Now the line element along C can be written in complex notation as dz dx + idy = dx dy +i ds ds ds.

Since the surface of the body is a streamline, the velocity vector (u, v ) is parallel to the tangent on the surface. Thus using the angle we have dz = dx + idy = ei ds = (cos + i sin ) ds. Multiplying by i we achieve a rotation by ninety degrees in the clockwise direction, which gives the direction of the outward normal: n = (sin , cos ). Now we are in a position to write the total force F = (Fx , Fy ) F= pn ds
C

in complex form, dening a complex force F = Fx iFy . Then F = Fx iFy =

C

p(sin + i cos ) ds = i = i = i 2

pei ds
C

p0 (dx idy ) + (q ei )2 dz =

i 2

q 2 e2i ei ds
C

i 2

dw dz

dz

(the term proportional to p0 vanishes because the integral of a total dierential over a closed loop is zero). This is Blasius theorem.

Example A cylinder in a stream with circulation: w (z ) = Uz + U so R2 i log z z 2

R2 i dw =U U 2 dz z 2z So Blasius around the circle C says Fx iFy = i 2 i = 2 R2 i dz 2 z 2z iU + terms {z 2 , z 3 , z 4 } U z U U dz = 2i, z

2

dz

Now use Cauchys residue theorem

C

dz = 0, n = 1 and zn

Fx iFy = iU , which is the same as in 4.3. 4.10.2 The Kutta-Koukowski lift theorem

From the previous example one appreciates that Blasius theorem yields a much more general result, since only the residue of the integral comes into play. Consider a body of arbitrary cross-section C , in a uniform stream, U , which generates circulation of strength . Far away from the body, i log z w (z ) Uz 2 (origin inside C ) so i dw U , z dz 2z and it must be analytic outside C and so dw dz
2

dz = lim

| z | =R

dw dz

dz

since there are no singularities in between C and |z | = R, a circle of large radius (basic theorem from C.F.T). So dw dz
2

dz =
| z | =R

i 2z

dz = 2U ,

using the same reasoning as for the cylinder. Hence we have shown that Fx iFy = iU. So the drag on an arbitrary body is zero and the lift force is Flif t = U. This is the Kutta-Joukowski lift theorem. Note that the lift force is by denition the force in the direction normal to the ow.

4.11
.

Oblique ow past plates

We have developed a wonderful theory for lift. The only problem is that we dont know how to determine the value of which determines the lift. Indeed, the wings of an airplane do not have circular cross section: if cylinder is placed in a uniform stream (left side of the Figure), there is no reason why the ow should turn either way, and thus there is no circulation. The secret is to fashion the shape so as to induce lift. Namely, the tail of wing is shaped to have a sharp corner, so as to force the ow to separate at that point. As the simplest possible model for such a situation consider the ow around a at plate, which is placed in a uniform stream at an angle . A at plate is obtained from an ellipse by letting d 0, that is putting b = R in the Joukowski transformation (33): z=+ R2 .

The the width of the plate is 2c = 4R. As is seen from the Figure, the ow separates from the plate beyond the trailing edge, so that uid particles very close to the plate have to go around the sharp corner before leaving the plate. This situation is reected by the velocity on the surface of the plate, which is u iv = U sin( ) , sin

and so u as 0, (at the sharp edges of the plate). This is a physically untenable situation; uid particles are not likely to make such a sharp turn without leaving the plate altogether, especially not at innite speed! The same conclusion can be drawn from the concept of adverse pressure gradient, introduced earlier. The speed is very great at the trailing edge and decreases as one moves up the plate. According to Bernoulli, this means that the pressure increases, in other words uid particles experience an adverse pressure gradient. Instead of following a path of adverse pressure, uid particles prefer to leave the body, producing a point of separation at the trailing edge. This requirement is known as the Kutta condition.

As we have seen in our calculation of the ow around a cylinder with circulation, the point of separation can be made to move by adding circulation: w 1 ( ) = U Then the velocity is U sin( ) sin 4R sin Let us focus on the trailing edge, corresponding to 0. (assume for example that the front of the plate is slightly rounded, so that separation is not as important there). Thus we want u iv to be nite for 0, a requirement which is another incarnation of the Kutta condition. This can be achieved by choosing u iv = = 4RU sin , which is the case shown in the above Figure. Now the singularity cancels and the velocity becomes U (sin( ) + sin ) U cos , as 0. u iv = sin This means that the ow leaves the plate smoothly in the direction of the orientation of the plate, as seen in the above Figure. ei + R2 i e i log . 2

The wonderful thing is that we have now determined the circulation uniquely, so we can calculate the lift using Blasius theorem: Flif t = U = 4RU 2 sin , where 4R is the width of the plate. Once more, note that this is the force acting normal to the ow, so it is at an angle relative to the orientation of the plate. It is customary to dene a lift coecient cL by Flif t = cL Aw U 2 , 2

where Aw = 4R is the area of the wing (per unit length), also called the chord. Thus we have found that for the plate CL
(plate)

= 2 sin 2

In the Figure, this result is compared to experiment, and it works really well, as long as the angle of attack is small. If however becomes too large the theory fails abruptly. The reason is clear from the two photographs below. As long as is small, the ow remains laminar and attached to the wing. As is too great, the ow separates and a completely dierent type of description must be sought.

Finally, we comment on the presence of circulation around the wing, which is the crucial ingredient needed for ying. In the rst instance, there is nothing wrong with that from the point of view of a potential ow description. However, where is this circulation coming from when one imagines starting up the ow, with zero circulation. Since the net circulation in a large circle around the wing must vanish initially, and the Kutta condition requires a circulation < 0 around the wing, this means that in the process of the point of separation moving to the trailing edge, vortices of positive circulation (rotating counterclockwise) are shed from the wing. Eventually the vortices are convected downstream, and no longer matter for the problem.

4.12

Joukowski wings

wing

R R 2 R

Now we try to model a wing in a slightly more realistic fashion, as proposed by Joukowski. In particular, we account for the fact that a real wing will be rounded at the leading edge (for the oncoming ow to go around it smoothly) and to be very sharp at the trailing edge (for the ow to separate).

Mapping an ellipse, we have seen that if b < R, the Joukowski transformation maps a circle to an ellipse; If b = R, the ellipse degenerates to a at plate. Now let us take the Joukowski transformation R2 , z=+ but shift the circle by to the left, and with radius R + . The equation of the circle in the -plane is = + (R + )ei , = 0..2. Thus it touches the circle of radius R at the right, but lies inside it everywhere else. In other words, the aerofoil is described by the equation z = + (R + )ei + R2 . + (R + )ei

It is clear that this shape is rounded everywhere, but sharp at the trailing edge to the right, where it has a cusp (it looks locally line the sharp end of a plate), as seen in the Figure. The length of the wing in the horizontal is called the chord. From the above

mapping, back and front ends are at = R and = R 2, respectively. Thus the chord is R2 4(R + )2 A = 2R + R + 2 + = . R + 2 R + 2 Now the complex potential around the shifted circle is evidently w 1 ( ) = U ( + )ei + (R + )2 i e ( + ) i log( + ), 2

and the complex velocity around the aerofoil is dw = dz U e

i

R+ ( + )

i 2 ( + )

R2 1 2

Once more, there are potential singularities at = R. However, the singularity = R lies inside the wing, and is therefore harmless. On the other hand, the point = R lies on the surface, and will lead to a singularity, unless is chosen appropriately. At = R, the term in braces becomes U ei ei which vanishes for In other words, using Blasius theorem, the lift coecient for the Joukowski wing is Flif t and the lift coecient is cL
(wing ) (wing )

i i = 2iU sin , 2 (R + ) 2 ( R + )

= 4U (R + ) sin .

= 4U 2 (R + ) sin . 2 (R + 2) sin . R+

wing

R R

It is seen from the above formula that there is no lift if the angle of attack vanishes. In reality, wings are cambered: the top is rounded, and the bottom is hollowed out. This can be achieved by shifting the centre of the circle to the position + I in the -plane. To produce a sharp trailing edge, the circle still has to touch the point = R, and thus its equation is = + i + (R + )2 + 2 ei .

For future convenience, we dene the angle = arctan R+

(see Figure). The chord is the same as for the prole without camber.

The resulting cambered prole is shown as the green outline above. To compute the ow, we note the complex potential in the -plane: w 1 ( ) = U ( + i)ei + (R + )2 + 2 i e ( + i) i log( + i), 2

and the velocity is now u iv = U e

i

(R + )2 + 2 i e ( + i)2

i 2 ( + i)

R2 1 2

As before, the term in braces has to vanish for = R for the velocity to remain nite, so this condition gives U ei (R + )2 + 2 i e (R + i)2 i = 0. 2 (R + i)

Noting that R + + i =

(R + )2 + 2 ei , we nd = 4 (R + )2 + 2 sin( + )

as the Kutta condition. The lift coecient becomes cL

(wing )

(R + )2 + 2 (R + 2) sin( + ) , ( R + )2

and so indeed there is lift for = 0. As a result, the angle of attack can be kept small,

and stall is less likely. The theoretical result for such a cambered prole is compared to experiment in the Figure. The pressure distribution over the wing as well as the total lift predicted by Joukowski theory is in good agreement with experiment.

Waves and free surface ows

Free surface ows are in some ways very dierent from what we have done before. In all problems considered so far, the domain D in which to solve the problem is given (for example some box or the exterior of an aeroplane wing). A free surface, on the other hand, moves, so the domain varies in time. The key feature, however, is that the domain does not vary according to some predetermined program. Instead, it moves in response to the ow itself, that is in response to the ow solution (which of course itself depends on the domain). This leads to solutions which are quite dierent in character to the xed-boundary solutions we were considering so far (and generally speaking much more interesting solutions)! Another nice feature of free surface ows is that they are easy to observe experimentally, one just needs to track the motion of the free surface.

5.1

Kinematic boundary condition

If a uid particle is adjacent to a boundary then we must impose a condition which links the velocity of the boundary to that of the particle. This is known as the kinematic boundary condition. Let S (x, t) = 0 describe the equation of a surface in (or on the boundary of) the uid. As the ow evolves, particles remain on the surface S if S DS = + u.S = 0 Dt t Proof: Taylors theorem: 0 = S (x, t) S (x + ut, t + t) = t u.S + S t

Oil S=0 Water Example: Consider two uids bounded by an interface S (x, t) = 0 (e.g. water/oil). Then DS S = + u(oil) .S = 0, on S = 0 from above Dt t DS S = + u(water) .S = 0, on S = 0 from below Dt t Now S is normal to the surface S = const, so n = S/|S | is the unit normal to the surface S = 0. It follows that
u(oil) .n = u(water) .n, on S = 0 (34)

I.e. the normal component of the velocity on either side of the interface must be equal (intuitive). At a uid/solid boundary, (34) holds, with u(oil) replaced by u(solid) , the velocity vector of the boundary. If xed, u(water) n = 0 on xed surface S (x) = 0 with normal n.

5.2

Linear gravity waves

The simplest approximation (but still v. good in many cases): waves of small amplitude. Flow is irrotational and uid incompressible. Assume 2D ow (for now).

Pressure = patm

z x

(x,t) = z z=0

Density = z=h
So u = , 2 = 0, where u = (u, 0, w ) and = (x, z, t) now. Choose z = 0 to coincide with the undisturbed free surface Bottom on the uid is at z = h. Let the surface of the water in motion be given by z = (x, t). Boundary conditions: (i) On z = h, 0 = n. = = w (Kinematic B.C.) z

(ii) On z = (x, t), Kinematic B.C. on a moving surface is

DS = 0 (see section (1.10)), Dt where S = z (x, t) = 0 denes the surface of the uid. This implies that particles on the surface, remain there time. I.e. +u +w t x z (z (x, t)) = + =0 t x x z on z =

(iii) Dynamic boundary condition: p = patm (const) on z = . So unsteady Bernoulli on surface gives
1 patm / + 2 ||2 +

+ g = C (t), t

on z =

1. If we assume that x the ow is driven by the motion of the free surface (no additional driving from below the surface), it must be that the ow is also week: || 1. Thus we will throw away all terms quadratic in the two variables and , i.e. containing 2, 2 , or products . Note that (iii) contains the velocity at quadratic (nonlinear) order, which is a remnant of the nonlinear character of the Euler equation, as it remains in Bernoullis equation. The , which is quadratic as well. Both terms kinematic boundary condition contains x x will be neglected in our linearised treatment. However, the most signicant problem is that in the formulation of the boundary conditions, the position of the free surface is part of the solution, so dont know where to By the small amplitude assumption, | | h and so

apply B.Cs ! We can deal with this diculty by linearising about z = 0 using Taylors expansion about z = 0 for all z -dependent terms. E.g. t =
z =

+
z =0

+ ...
z =0

and throw away any products involving and/or , because quadratic terms are negligible. So now (i) On z = h, (ii) On z = 0, (iii) On z = 0, =0 z

= t z + g = C (t) patm /. t

are time dependent only, so they dont aect the ow velocities, Laplaces equation or the B.Cs. So now (dropping primes) (iii) is replaced with (iii) On z = 0, + g = 0. t

Last condition is ugly, so redene = patm t/ +

C (t )dt . The extra terms

Linearised pressure in the uid is p(x, z, t) + + gz = C (t) t and because of this transforms to p(x, z, t) patm = gz t (35)

Now look for particular solutions to 2 = 0 with (i), (ii), (iii) motivated by observations. Assume surface of the form (x, t) = H sin(kx t + ) max/min of is H . is just a phase shift (in space or time). periodic with period = 2/ ( is angular frequency) Repeats in space every = 2/k ( is wavelength) Moves to the right with speed c = /k (c is phase speed). Why ? Let = x ct, then = H sin(k + ) is unchanged for constant so 0 = d/dt = dx/dt c and dx/dt = c = speed. (36)

Now need . Since

= g reasonable to write t (x, z, t) = cos(kx t + )Z (z )

for some Z (z ) this is a separation solution = X (x)Z (z ). Then from 2 = 0, k 2 cos(kx t + )Z (z ) + cos(kx t + )Z (z ) = 0 and so Z (z ) k 2 Z (z ) = 0. Then Z (z ) = A cosh k (z + h) + B sinh k (z + h) (= A ekz + B ekz ) for constants A, B, A , B . Because of (i), need B = 0, so Z (z ) = A cosh k (z + h) Still have (ii) and (iii) to apply (but only A to nd !) From (iii) rst, g = /t|z =0 gH sin(kx t + ) = A sin(kx t + ) cosh kh and then A = gH/( cosh kh). Then from (ii), /z |z =0 = /t H cos(kx t + ) = kA cos(kx t + ) sinh kh implies H = k or gH cosh kh sinh kh

2 = k tanh kh g

(37)

Notes: Changing k k , (37) still holds. So then (x, t) = H sin(kx + t ) is a wave travelling to the left, speed c = /k . k = 2/ is called the wavenumber (the number of wavelengths that can be t into 2 ). The principal result of this calculation is the so called dispersion relation (37) of the wave problem. It establishes how the frequency of a monochromatic wave is related to its wavelength. Very roughly, large small k small large = 2/ . So long wavelengths imply long periods, and vice versa. If the dispersion relation between and k is non-linear, one speaks of a dispersive problem: parts of a wave containing dierent frequencies travel at dierent speeds, and thus disperse. Two important special cases of (37) exist:

5.2.1

Shallow depth

This is the case kh 1 (or /h 1), which means that the wave length is long compared to the depth. Then tanh kh kh or = ghk, so the dispersion relation is a linear function. The constant of proportionality is the (constant) wave speed c = /k = gh. Evidently, the shallow water problem has no dispersion. 5.2.2 Innite depth

If on the other hand kh 1 (or /h 1), the water is deep relative to the length of the wave. Then tanh kh 1 and the dispersion relation is = gk. Thus the wave speed c = /k = g/k = g/2 does depend on the wavelength, and there is dispersion. The velocity potential corresponding to the wave form (36) is = H kz e cos(kx t + ). k w = H ekz cos(kx t).

Neglecting the irrelevant phase factor , the velocity components are u = H ekz sin(kx t),

Assuming that a particle with position (x , z ) in the ow only departs by a small amount from its mean position (x, z ), one can write dx = H ekz sin(kx t), dt x = H ekz sin(kx t), dz = H ekz cos(kx t). dt z = H ekz cos(kx t).

This is integrated easily to yield the particle trajectory

Thus a particle moves along a circular path, whose radius decreases exponentially with depth. Most of the motion is concentrated near the surface, and the wave energy decreases exponentially as one descends into the water.

5.3

Group velocity

Now we investigate the propagation of so-called wave-packet, that is a burst of nite spatial extend, as it might be produced by throwing a stone into the water. We write this situation as a superposition of innitely extended plane waves:

a(k )ei(kxt) dk.

The amplitude of each part of the wave is given by a(k ). It is easy to show (a version of the uncertainty relation), that the more peaked the distribution a(k ) is around some mean value k0 , the more oscillations the wave packet contains, and thus the broader it becomes. Let us assume that a(k ) is suciently narrow relative to k0 . In other words, it is safe to write d (k k 0 ). (k ) = (k 0 ) + dk k=k0 But this means the wave packet can be written

(x, t) = ei(k0 x(k0 )t)

a(k )ei(kk0 )(xcg t) dk,

where cg

d dk

(38)
k =k 0

is called the group velocity. The rst factor in the representation of the wave packet represents a harmonic modulation with a high-frequency wave. The second factor is envelope, giving the shape of the wave packet. The key observation is that x only appears in the combination x cg t, and so the entire wave packet moves with the group velocity cg . This velocity is usually more signicant than the phase velocity c = /k . For example, the energy of a wave is concentrated where the wave packet is localised. Therefore, energy transported by a wave moves at speed cg . Taking waves of innite depth as an example, the phase velocity is c = g/k, while cg = g/k/2, or half the value!

5.3.1

Example: Tsunamis

The ocean is deep 4km, but in this case the waves are very long ( 200km). As a result, /h 1 and the shallow water approximation applies. There is no dispersion, but the wave speed c = /k = gh (which is the same as the group velocity) depends on the depth. Thus as the ocean depth changes, the Tsunami wave moves in a medium of variable refractive index, and does not move along a straight path. Instead, it is refracted in unexpected directions, and can turn corners. Here we only want to discuss the amplication of the wave amplitude near the shore. The Tsunami moves very fast, the wave speed is c = /k = gh = 10.4000 = 200ms1 = 400mph. Owing to energy conservation, the wave energy must be conserved as h changes, and thus the energy ux of the wave must be constant. The wave contains potential and kinetic energy to equal proportion. From elementary mechanics, the potential energy of a liquid column (per unit area) of height is 2 . 2 0 Now for a sinusoidal wave, the average of the evaluation is gzdz = g 2 = H2 2
2 2

sin(x)dx =
0

H2 , 2

so the potential energy becomes gH /4. Since the same energy is contained in the kinetic part, the total energy (per unit area) is e = gH 2. 2 Since the energy moves with the group velocity, the energy ux becomes 1 1 ghgH 2 . fe = cg gH 2 = 2 2 If this is to be constant for all x, this implies that H 4 h = const, and therefore H 1/h1/4 . If the depth goes from 4000m to 4m then factor of 1000 and the wave height increases by a factor of 10001/4 6. So a 0.5m wave in the ocean is a 3m wave at the coast.

5.4

Oscillations in a container

Liquids readily slosh back and forth in a closed container (e.g. tea in a tea-cup). There are many associated important practical problems: sloshing in road tankers, water on decks of ships, resonance in harbours, etc...
z z= (x,t) x z=0

z=h x=0 x=L

Example: consider a two-dimensional rectangular box with rigid walls at x = 0, L and a bottom on z = h, lled with uid to z = 0. Use small-amplitude theory from before, so that linearised equations are 2 = 0 for (x, y, z ) in box, (i) = 0 on z = h. z = on z = 0. t z

(ii) Kinematic: (iii) Dynamic:

= g on z = 0. t

Also need no-ow conditions (kinematic) on walls: (iv) = 0 on x = 0, L; x

Dierent to before - no travelling waves - but still periodic in time, so write (x, y, t) = (x, y ) sin t and assume we can separate variables in : (x, y, z, t) = X (x)Z (z ) cos t. Then Laplaces equation gives X (x)Z (z ) cos t + Z (z )X (x) cos t = 0 which implies X (x)Z (z ) + Z (z )X (x) = 0 This separates: X (x) Z (z ) = = k 2 X (x) Z (z )

where k 2 is the separation constant. Solving for Z (z ), with (i) gives Z (z ) = A cosh k (z + h) as before. Combining (ii) and (iii) to eliminate (thats g (ii) +/t(iii) gives 2 = g , 2 t z which implies and it follows that 2 /g = k tanh kh as before. (We expect this, as the vertical structure of the uid is independent of the vertical lateral walls) Now solving for X (x) gives X (x) = B cos(kx) + C sin(kx) subject to (from (iv)), X (0) = X (L) = 0. Easy to show that must have C = 0 and k = n/L and so X (x) = B cos(nx/L) So pulling everything together we have (x, z, t) = A cos(nx/L) cosh k (z + h) cos t for some constant A = AB while from (iii) the free surface is given by (x, t) = A cosh kh cos(nx/L) sin t g 2 X (x)Z (0) cos t = gX (x)Z (0) cos t on z = 0

and here, k = n/L, so that (37) reads

2 2 /g = (n/L) tanh(nh/L) n /g,

n = 0, 1, 2, . . .

and therefore denes a set of discrete wave frequencies at which these oscillations may occur. Here, n is a mode number and tells you how many oscillations are occurring across the box. We cannot have n = 0 as then = 0 and = 1, and = 0. So there is no motion in the uid. (In fact, you can discount a sloshing mode with no x-dependence a at surface oscillating up and down as it would violate mass conservation in the tank). The Fundamental frequency is the lowest frequency, given here by n = 1. E.g. Let L = 1m, h = 20cm. Then the fundamental frequency (n = 1) is 1 = 9.81 tanh( 0.2) = 4.14s1

given a period of = 2/4.14 = 1.51 seconds. The shape of the surface is given by the variation of x in (x, t), which is cos(x/L) and it is modulated by the signal sin t.

ce

n=2 large n n=1 The n = 2 mode gives a period of 0.86 seconds and a shape function of cos(2x/L) etc... As n , tanh(nh/L) 1 and so n gn/L.

5.5

Liquid sheets
A A

CC

We now move on some of examples of truly nonlinear solutions of free surface ow problems. The rst example is that of a sheet coming out of a slit at the bottom of a container. One might think that the sheet has the same thickness as the width of the slit. However, the ow converges as it enters the slit, so the sheet is actually thinner; it contracts as it ows out of the container. We want to nd out what the ow is like exactly near the entrance, and in particular what the shape of the free surface is like. This is a two-dimensional problem, so we will use complex mapping techniques. They permit to deal with the diculties associated with the non-linearity of free surface ow. We begin with the trivial problem of an innite sheet owing in the negative y -direction. With view to what is to come, we want to solve the problem by mapping to solve the problem by mapping to the upper half plane, see the Figure above. We have already seen in section 4.8.3 that the map which accomplishes this is the logarithm, which produces a channel oriented in the x-direction. Thus we introduce a ninety degree rotation, and consider the map z = i ln , which maps points A,B,C,C,B,A as indicated. For simplicity, we consider a sheet of width ; it is easy to change the length scales later on. In particular, the lower limit of the sheet is near the origin of the -plane. Therefore, one might guess that the representation of the sheet is a sink located at the origin. According to section 4.4.2 the potential is w ( ) = ln . (39)

We will henceforth adopt the somewhat more lax, but useful notation w1 w if w is represented as function of . Now in the z -plane this becomes w = z/i = iz , and thus u iv = i. Thus in the sheet there is a ow with unit speed in the negative y -direction, exactly as anticipated.

container B z C,C A B B A A

A C sheet C

Of course the real problem at hand is much more dicult, as shown on the right of the above Figure. Between A and B as well as between B and A, the ow is bordered by hard walls. At least in these places we know where the boundary is. However, at the free surface of the sheet, which is between B and C as well as B and C, the shape of the domain becomes part of the problem. This is also the reason why the problem becomes non-linear. Complex mapping techniques are extremely powerful at approaching this kind of problem, since the whole ow domain (bordered along A,B,C,C,B,A) can still be mapped onto the upper half plane. The name of the game is to nd a complex mapping z ( ) which maps the upper half plane conformally onto the ow domain. Once represented in the -plane, the ow will still be described by the potential (39). The reason is that the only singularity of the problem is the sink located at = 0, which describes the sheet owing o to innity. How do we nd the mapping z ( )? The main idea is due to Helmholtz and Kirchho (1869), with additional simplications suggested by Planck and Love. The trick is to map the ow domain onto a domain of known shape, for which also some of the boundary conditions are known. According to complex mapping theory, this then determines the map uniquely. What are the boundary conditions? At a solid border, the ow must be parallel to the wall. Thus representing the velocity in complex notation, dw = u iv = q ei , dz the ow direction is = 0 along B,A, and = along A,B. At the free surface, the pressure is constant (we are not considering gravity). This is why the surface is called a free streamline. According to Bernoulli, the ow speed q is also constant along the free streamline. We will normalise such that q = 1 all along the sheet. Now the ingenious idea is to introduce the function dz = ln q + i. dw This maps the ow onto a rectangular domain, see above. Namely, B , A is mapped onto the straight line {} = 0, and A, B onto the straight line {} = . The free surface, on the other hand, is mapped onto a straight line parallel to the imaginary axis: since q = 1, we have {} = 0. Thus all the boundaries are as drawn in the Figure. As a check of consistency, let us calculate at the points C,C innitely far down the sheet. The ow will be that of an unperturbed sheet: w = iz , and thus = ln(i) = ln ei/2 = i/2. Indeed, this is midway between the upper and lower boundary of the domain. Now we construct the mapping between the and planes. This mapping is guaranteed by the so-called Schwarz-Christoel mapping theorem, but we will take a more = ln

constructive approach: we consider the derivative / . On the boundary, this expression has singularities at = 1 and = 1, corresponding to points B and B , where the boundary of makes 90 turns. We have considered this situation before in problem 1 of worksheet 9. There we have shown that a turn by /2 corresponds to a singularity /1 = 1/2 . Two such turns are to be executed at = 1, so we have d = d 1 2 1 .

This function does exactly what we need: For > 1 the argument of the square root is positive, and thus the expression on the right is real and positive. Thus as one integrates from A to B along the -axis, one moves to the left parallel to the real z -axis. At = 1, the right hand side becomes becomes purely imaginary, with a nagative coecient. Thus the path of integration moves up parallel to the imaginary axis as one goes from B to B. Finally, at = 1, the right hand side is once more real, and one moves back parallel to the real z -axis. Integrating, we have = Arcosh + 0 . Now for > 1 we must have = i , and so since Arcosh is real, it follows that {0 } = . In fact 0 = i ; for 1 < < 1, must be purely imaginary: = i2 . Thus cosh i2 = cos 2 = cosh(Arcosh i ) = , and 2 runs from at = 1 to 0 at = 1, exactly as required. In summary, = Arcosh i. (40)

This is the mapping we are looking for, which will essentially be the solution to the problem. Streamlines, represented in the -plane, are straight lines approaching the origin = 0, thus with the mapping (40) streamlines can be represented in the -plane, as shown. To nd z ( ), we write dz dw dw e + 2 1 dz = = e = = , d dw d d where we have used Arcosh = ln + 2 1 and ei = 1. Another integration will yield the desired mapping. Since we are mainly interested in the shape of the sheet for 0 < < 1, we write 2 1 = i 1 2, and the integral gives z = +i 1 2 ln 1+ 1 2 + z0 .

The constant of integration z0 is merely a shift, but we want to place the sheet symmetrically about the y -axis. In the limit of 0, we have z i ln as before for the free sheet, which lies between and 0. Thus we have z0 = /2 and nally z = +i 1 2 ln 1+ 1 2 + . 2 (41)

In particular, for 0 < < 1, x = {z } = + /2, and so y = {z } = 1 (x /2)2 ln 1+ 1 (x /2)2 x /2

is the shape of the sheet. Of particular interest is the contraction of the sheet. The width x of the sheet decreases from 2(1 + /2) at the opening to far downstream. Thus the sheet suers a contraction by the ratio 0.611 < 1. Cc = 2+ The experimental value, found by J.S. McNown and E.-Y. Hsu, is Cc good!
(exp)

= 0.632: pretty

Appendix A: Vector calculus

We shall revise some vectors operations that you should have already met before this course. These may be presented slightly dierently to the way you have previously seen them.

A.1 Sux notation and summation convention

Suppose that we have two vectors u = (u1 , u2 , u3) and v = (v1 , v2 , v3 ). dot product is dened to be
3

Then the

uv =

ui vi
i=1

or, more simply, write

u v = ui vi

(drop the summation symbol on the understanding that repeated suces imply summation. Defn A.1.2: The Kronecker delta is dened by ij = 1, 0, i=j i=j

So in summation convention, ij aj = ai since

3

ij aj Examples: 1. ii = 3 2. ij ui vj = uj vj u v.

ij aj = ai
j =1

Defn A.1.3: The antisymmetric symbol ijk is dened by 123 = 1 ijk is zero if there are any repeated suces. E.g. 113 = 0. Interchanging any two suces reverses the sign: e.g. ijk = jik = kji Above implies invariant under cyclic rotation of suces: ijk = jki = kij With this denition all 27 permutations are dened. There are only 6 non-zero components, 123 = 231 = 312 = 1, 213 = 132 = 321 = 1, Defn A.1.4: The cross product is dened by w = u v = (u2 v3 u3 v2 ) x + (u3 v1 u1 v3 ) y + (u1 v2 u2 v1 ) z

But can now be written in component form as wi = ijk uj vk where summation over j and k occurs. [Check]. Example: Consider the triple product, w (u v) = wi ijk uj vk where summation is over i, j , k so result is scalar. It follows that w (u v ) = = = = jki wi uj vk = u (v w) kij wi uj vk = v (w u) jik wi uj vk = u (w v) ikj wi uj vk = w (v u)

Defn A.1.5: The double product of ijk is ijk ilm = jl km jm kl [Check] Defn A.1.6: The vector triple product is dened by the result w (u v) = (w v)u (w u)v Proof: [w (u v)]i = = = = = True for i = 1, 2, 3, hence result. ijk wj [u v]k ijk wj klm ul vm kij klm wj ul vm (il jm jl im )wj ul vm wj ui vj wj uj vi = (w v)ui (w u)vi

A.2 Dierential operations

Here we consider operations on a function (x) and a vector eld f (x) where x = , , . Without us(x1 , x2 , x3 ). One can regard as the vector operator x1 x2 x3 ing any information (but always remembering the true meaning of the symbol!), one can also use the much sleaker notation (1 , 2 , 3 ). Then The gradient is . So []i = i xi

 The divergence is f =

fi i fi (in summation convention) xi fk ijk j fk . xj

The curl is f where [ f ]i = ijk Examples: 1. [(xj )]i = 2. x = xj = ij xi

xi = ii = 3 xi xk = ijk kj = ijj = 0 xj

3. [ x]i = ijk

4. r = x , where r 2 x x. r

i x2 xj i xj xj ij xi j = = . = Proof: [r ]i = i xj xj = r r r 2 x2 j

A.2.1 Useful vector identities fi (fi ) 1. (f ) = = + fi = f + f xi xi xi 2. (f ) = ( f ) + ( f ) Proof: [ (f )]i = ijk (fk ) fk = ijk + fk ijk = [ f ]i + [ f ]i . xj xj xj

f f ) ( f ) f 3. f ( f ) = ( 1 2 Proof: [f ( f )]i = ijk fj [ f ]k = ijk fj klm fm fm = kij klm fj xl xl 1 fj fj fm fj fi = (il jm im jl )fj = fj fj = 2 (f )fi xl xi xj xi

Hence result.

A.3 Integral results

A.3.1 The divergence theorem

dS V S

Very important. Consider a volume V bounded by a closed surface S with outward unit normal n. Then for a vector eld f (x) f dV = f ndS

Corollary: Let f (x) = a(x) where a is an arbitrary constant vector, and (x) a scalar function. Since (a(x)) = a (x), so the divergence theorem reduces to a True for any a, so
V V

dV = a

ndS
S

dV =

ndS
S

A.3.2 Stokes theorem

n ds S C dl
Let C be a closed curve bounding a surface S with unit normal n. Then for a vector eld f (x), f dl = ( f ) ndS

where dl is a line element on C . Corollary: Let f (x) = a(x) where a is an arbitrary constant vector. Then a dl = ( a) ndS

and from A.2.1(2), ( a) = ( a) + ( a) = ( a). Using the vector triple product result, ( a) n = a ( n) and then a Therefore
C C

dl = a dl =

ndS

ndS

Appendix B: Curvilinear coordinate systems

q1 increasing r(q1 , q2 )

e2 e1

q2 increasing

Many problems can be approached more simply by choosing a coordinate system that ts a given geometry. Instead of writing the position vector r as a function of Cartesian coordinates (x, y, z ), r is now written as function of three new coordinates: r(q1 , q2 , q3 ). The coordinate lines are swept out by varying one of the coordinates, keeping the other two constant. We will deal only with the by far most important case of orthogonal coordinate systems, in which the coordinate lines always intersect one another at right angles. r is a vector which points in the direction of the i-th coordinate line, Evidently, qi see the gure. If each of these vectors are normalized to unity, we obtain the local basis system: r 1 r i = , hi . (42) q hi qi qi The quantities hi (q1 , q2 , q3 ) are called scale factors or metric coecients. The fact that i , computed at a point (q1 , q2 , q3 ), the coordinate system is orthogonal means that all q i of course changes as one goes along the are orthogonal. However, the direction of q coordinate lines. Example: The cylindrical polar coordinate system. The position vector is r = (r cos , r sin , z ) , and the coordinates are (r, , z ). Thus the scale factors become h1 = 1, h2 = r , and h3 = 1, and the local basis is r= r = (cos , sin , 0) , r = 1 r = ( sin , cos , 0) , r r = (0, 0, 1). z= z

, z are indeed mutially orthogonal. It is clear that r, Remark: As the coordinates are varied by q1 , q2 , and q3 , respectively, r describes a volume whose sides are orthogonal. The length of each side is hi qi , and thus the volume

of the cuboid is d3 x = h1 h2 h3 dq1 dq2 dq3 . Thus integration in a curvilinear coordinate system is achieved by the formula f (x)d3 x =
V V

f (q1 , q2 , q3 )h1 h2 h3 dq1 dq2 dq3 .

(43)

In cylindrical polars, the volume element is rdrddz . To do vector calculus in the curvilinear coordinate system, we have to work out what the -operator is. For any scalar function f (r), rj f f = = qi qi rj r f = hi ( qi ) f qi

by the chain rule. This menas that in the local basis, has the representation = 1 2 3 q q q + + . h1 q1 h2 q2 h3 q3 1 r+ z, + r r z (44)

Thus in cylindrical polars, the gradient of a scalar function is =

in agreement with chapter 0. Now we are almost there. The only but crucial thing that remains to be considered is the fact that both scale factors and unit vectors depend on the coordinates qi , so they i q need to be dierentiated. In particular, we want to represent in terms of the basis qj vectors. It is not dicult to write down general expressions for the derivatives of the basis vectors if only the hi are known. However, for a given coordinate system, it is generally much easier to simply calculate the derivatives directly. For example, in cylindrical polars r = , = r, and all the other derivatives are zero. Now it is easy to derive the other formulae in chapter 0. In cylindrical polars, the Laplacian becomes = = ( r r) 2 1 + r 2 r r + + z r r z 1 + 2 r r r + + z r r z =

2 2 1 + ( + z z ) = r 2 2 z 2

1 2 2 2 1 + + + 2. r 2 r r r 2 2 z + uz z . Then the r + u In local coordinates, a vector eld is written as u = ur divergence is u= r + +z r r z ur ur 1 u uz + uz z = r + u ur + + + , r r r z

again in agreement with chpater 0.

Appendix C: Streamfunctions
If u = 0, then it follows that there exists a vector eld A(x, t) s.t. u = A. Conversely, it is clear from this representation that the ow is incompressible. The representation is particularly useful if there are two independent coordinates, in which case A can be written in terms of a single scalar function, the streamfunction. By denition, the streamfunction is constant along streamlines. There are two important cases: two-dimensional ows and three-dimensional, axisymmetric ows. In the latter case the streamfunction is known as the Stokes streamfunction. 5.5.1 Two-dimensional ows

(i) Cartesians: Consider Cartesian coordinates (x, y, z ), and take the ow in the (x, y )plane. Then A = (x, y, t) z, and u= , y v= . x

Now conrm that is indeed constant along streamlines, which are dened by dx = u, ds and thus dx = uds and dy = vds. It follows that d = dx + dy = v dx udy = 0, x y

if indeed is varied along streamlines. Thus is constant along a streamlines, as advertised. Defn: We call the function (x, y, t) the streamfunction of the ow.

= constant

(ii) Cylindrical polar coordinates: Now consider ow in the same two dimensional plane z = 0, which is represented in cylindrical polars (r, , z ). The relation to Cartesian coordinates is x = r cos , y = r sin . Since the stream function is dened completely by the geometry of the streamlines, it must be the same in any coordinate system. In other words if c (x, y, t) is the Cartesian version, and p (x, y, t) the streamfunction in polar coordinates, then p (r, , t) = c (r cos , r sin , t). As before, the vector potential is A = (r, , t) z, and thus from the curl in polar coordinates 1 , u = , ur = r r . r + u where u = ur

Now verify that (r, ) is indeed constant along streamlines. In polar coordinates, x = r r, and thus d d d dr dr r dx r+r r+r = = , ds ds ds d ds ds d r where we have used that = . In other words, we nd that dr = ur ds and rd = d u ds. Now as before, dr + d = u dr + rur d = 0, r if is indeed varied along streamlines. d = 5.5.2 Three-dimensional axisymmetric ows

This is a dierent physical situation, as the ow is now three-dimensional. However, since it is axisymmetric, it can be written in terms of just two coordinates. Once more, we choose two dierent coordinate systems. (i) Cylindrical polars: Coordinates (r, , z ) and and the velocity eld is independent of and so u = ur (r, z ) r + uz (r, z ) z. For this to be true, A must be in the third coordinate direction: . A= r The denition is such that (r, z ) is indeed a streamfunction, as we will now conrm; it is called the Stokes streamfunction. First, calculating A gives 1 1 , uz = . r z r r Streamlines given by dr = ur ds and dz = uz ds, so ur = d =

dr + dz = ruz dr rur dz = 0. r z
z

Therefore = const on streamlines (actually stream surfaces as 3D ow).

z

x Spherical Polars

xaxis Cylindrical polars

(ii) Spherical polars: The coordinates are (r, , ) and the velocity eld is indepen , and A must be in the -direction: dent of . Thus u = ur (r, ) r + u (r, ) A= (r, ) . r sin

We conrm that (r, ) is the streamfunction, and thus it is the same as dened previously in cylindrical polar coordinates. First, computing A we nd ur = r2 1 , sin u = 1 . r sin r

As in polar coordinates, streamlines are given by dr = ur ds and rd = u ds. Thus d = dr + d = r sin u dr + r 2 sin ur d = 0 r

from the denition of the stream function. Thus (r) is indeed constant along streamlines. If (, , z ) are cylindrical polars, we have z = r cos and = r sin , and thus sp (r, ) = cp (r sin , r cos ), dened in spherical polars and cylindrical polars, respectively.