Introduction To The Calculus of Variations

MS327
Deterministic and stochastic dynamics
Unit 5
Introduction to the calculus of variations

This publication forms part of an Open University module. Details of this and other Open University
modules can be obtained from Student Recruitment, The Open University, PO Box 197, Milton
Keynes MK7 6BJ, United Kingdom (tel. +44 (0)300 303 5303; email [email protected]).
Alternatively, you may visit the Open University website at www.open.ac.uk where you can learn
more about the wide range of modules and packs offered at all levels by The Open University.
Note to reader
Mathematical/statistical content at the Open University is usually provided to students in
printed books, with PDFs of the same online. This format ensures that mathematical notation
is presented accurately and clearly. The PDF of this extract thus shows the content exactly as
it would be seen by an Open University student. Please note that the PDF may contain
references to other parts of the module and/or to software or audio-visual components of the
module. Regrettably mathematical and statistical content in PDF files is unlikely to be
accessible using a screenreader, and some OpenLearn units may have PDF files that are not
searchable. You may need additional help to read these documents.
The Open University, Walton Hall, Milton Keynes, MK7 6AA.

First published 2016.
c 2016 The Open University
Copyright
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, transmitted
or utilised in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without
written permission from the publisher or a licence from the Copyright Licensing Agency Ltd. Details of such
licences (for reprographic reproduction) may be obtained from the Copyright Licensing Agency Ltd, Saffron
House, 6–10 Kirby Street, London EC1N 8TS (website www.cla.co.uk).
Open University materials may also be made available in electronic formats for use by students of the
University. All rights, including copyright and related rights and database rights, in electronic materials and
their contents are owned by or licensed to The Open University, or otherwise used by The Open University as
permitted by applicable law.
In using electronic materials and their contents you agree that your use will be solely for the purposes of
following an Open University course of study or otherwise as licensed by The Open University or its assigns.
Except as permitted above you undertake not to copy, store in any medium (including electronic storage or
use in a website), distribute, transmit or retransmit, broadcast, modify or show in public such electronic
materials in whole or in part without the prior written consent of The Open University or in accordance with
the Copyright, Designs and Patents Act 1988.
Edited, designed and typeset by The Open University, using the Open University TEX System.
Printed in the United Kingdom by Latimer Trend and Company Ltd, Plymouth.
ISBN 978 1 8487 3962 8

1.1
Contents
Contents
Introduction 5
1 Shortest distance between two points 6
1.1 Distance between two points on a given curve 7
1.2 Stationary paths 9
2 Some examples of functionals 15
2.1 Brachistochrone 16
2.2 Minimal surfaces of revolution 17
2.3 A problem in navigation 18
2.4 Catenary 18
3 Reminder about partial and total derivatives 19
4 Euler–Lagrange equation 21
4.1 Derivation of the Euler–Lagrange equation 21
4.2 Alternative forms of the Euler–Lagrange equation 25
5 Applications of the Euler–Lagrange equation 28
5.1 Fermat’s principle 28
5.2 Revisiting the brachistochrone 34
5.3 Revisiting minimal surfaces of revolution 40
Solutions to exercises 51
Acknowledgements 63
Index 64
Introduction
Introduction
In Book 1, we introduced various types of dynamical systems. These
systems could be classified as:
• conservative, where the total energy of the system is constant
(i.e. conserved) throughout the motion
• non-conservative, where energy is not conserved, as for example in the
presence of forcing or damping
• deterministic, where the equations describing the motion have no
randomness in them
• non-deterministic (or random or stochastic), where the equations
describing the motion have randomness built in, as for example in a
random walk.
In Book 2, we explore some aspects of conservative and non-conservative
deterministic systems in more detail, before moving on to discuss random
systems in Book 3.
In earlier modules, you may have seen how to derive the equations of
motion of simple mechanical systems using Newton’s laws of motion. In
Unit 6, we will introduce you to another method called Lagrangian
mechanics, which applies to conservative systems, and is arguably much
easier to use and more powerful than Newtonian mechanics – although
perhaps less intuitive since it is based on energies rather than forces.
However, before we get to Unit 6, we first need to set up some important
mathematical apparatus called the calculus of variations: this is the main
purpose of this unit.
In ordinary calculus, we often work with real functions, which are rules for
mapping real numbers to real numbers; for example, the function sin x
maps the whole of the real line to the interval [−1, 1]. Functions can have
various properties: for instance, they can be continuous and differentiable,
and they can have stationary points and local maxima and minima.
In the calculus of variations, we work with functionals: these are objects
that map functions to the real line. For example,
Z 1
S[f ] = (f ′ (x))2 dx
0
is a functional, which for any sufficiently well-behaved real function f gives
us a real number S[f ]. (Here and throughout this unit and the next, f ′ (x)
denotes the derivative df /dx evaluated at x.) The square bracket notation
S[f ] is used to emphasise the fact that the functional S depends on the
choice of function.
5
Unit 5 Introduction to the calculus of variations
Functionals share many of the same properties as functions. In particular,

the notion of a stationary point of a function has an important analogue in
the theory of functionals, which gives rise to the calculus of variations, as
you will see. The calculus of variations is a hugely important topic in the
natural sciences. It leads naturally to the Lagrangian formulation of
mechanics, mentioned above, from which most dynamical equations of
mathematical physics can be derived. It is also a powerful mathematical
tool, finding applications in subjects as diverse as statics, optics,
differential geometry, approximate solutions of differential equations and
control theory. In addition to this, the calculus of variations is an active
topic of study in its own right.
The structure of this unit is as follows. In Section 1 we introduce many of
the key ingredients of the calculus of variations, by solving a seemingly
simple problem – finding the shortest distance between two points in a
plane. In particular, this section introduces the notion of a functional and
that of a stationary path. In Section 2 we briefly describe a few basic
problems that can be formulated in terms of functionals, in order to give
you a feel for the range of problems that can be solved using the calculus
of variations; this section will not be assessed. Section 3 is a short interlude
about partial and total derivatives, which are used extensively throughout
the rest of the unit. Section 4 is the most important section of this unit. It
contains a derivation of the Euler–Lagrange equation, which will be used
throughout the rest of this unit and the next. Finally, in Section 5 we
apply the Euler–Lagrange equation to solve some of the problems
discussed in Section 2, as well as a problem arising from a new topic, called
Fermat’s principle, which serves as a bridge to the next unit.
1 Shortest distance between two

points
In this section we take a swift tour through variational principles by
considering perhaps the simplest physical example possible – the shortest
distance between two points in a plane.
You may think that this is a somewhat trivial example, since the shortest
path between two points in a plane is the straight line joining them.
However, it is almost always easiest to understand a new idea by applying
it to a simple familiar problem, so here we introduce the ideas of the
calculus of variations by proving this trivial fact. The algebra involved may
seem over-complicated for such a basic problem, but far more complicated
problems (like the problem of finding the shortest distance between two
points on a curved surface) can be solved using the same principles.
6
1 Shortest distance between two points
We begin in Subsection 1.1 by showing that the distance between two

points can be expressed as a functional. Then in Subsection 1.2 we show
that the shortest distance between two points in a plane is a straight line.
1.1 Distance between two points on a given curve

First, we need an expression for the length of a curve between two given
points in a plane.
Suppose that we are given two points Pa and Pb with Cartesian
coordinates (a, A) and (b, B), respectively. Furthermore, suppose that
y = y(x), where a ≤ x ≤ b, is a smooth curve that joins Pa at x = a to Pb
at x = b, so that y(a) = A and y(b) = B, as shown in Figure 1.
y
Pb
B
y(x)
Pa
A
a b x
Figure 1 Graph of the curve y = y(x) passing through points Pa and Pb
You may have derived a formula for the length of this curve in earlier
modules. It is obtained by dividing the interval a ≤ x ≤ b into N intervals
of length δx = (b − a)/N . Let us denote the endpoints of these intervals by
x0 , x1 , . . . , xN , in that order, so that x0 = a, xN = b and xk+1 − xk = δx
for k = 0, 1, . . . , N − 1, as shown in Figure 2.
y
Pb
B
y(x)
Pa
A
x0 x1 x2 x3 x4 x5 x
a b
Figure 2 Subdivision of the interval [a, b] into N equal intervals; in this
case N = 5
7
The curve is now approximated by a sequence of N straight-line segments.

Let δsk be the length of the segment above the interval [xk , xk+1 ]. This
segment is illustrated in Figure 3.
y
y(xk+1 )
δsk
δy
y(xk )
δx
xk xk+1 x
Figure 3 Segment of the curve y = y(x) between xk and xk+1
The length of each segment can be determined using Pythagoras’ theorem:

s 2
p δy
2 2
δsk = δx + δy = δx 1 + .
δx
However, δy/δx is just the gradient of the segment, which in the limit as
N → ∞ is given by δy/δx = y ′ (xk ) (i.e. the derivative dy/dx of y evaluated
at x = xk ). Therefore
p
δsk = δx 1 + y ′ (xk )2 . (1)
The approximate distance from (a, A) to (b, B) along the curve y = y(x) is
given by the sum of all the segment lengths:
N
X −1 N
X −1 p
S[y] ≃ δsk = δx 1 + y ′ (xk )2 .
k=0 k=0
In the limit as N → ∞, this sum becomes the integral
Z bp
S[y] = 1 + y ′ (x)2 dx.
a
The above expression gives the length of the curve y = y(x) between the
points Pa and Pb . The length changes for different functions y(x). For
instance, if Pa = (−1, 0) and Pb = (1, 0), then the distance from Pa to Pb
along the straight line joining these points is 2, whereas the distance is π
along the semicircle of which this straight line is a diameter.
We denote the numerical value of the length of the curve by S[y], which
emphasises the fact that the length depends on the function y(x). We do
not write S(y), because we wish to distinguish S[y] from the real-valued
function y(x). The quantity S[y] is our first example of a functional; it
maps functions y(x) that satisfy y(a) = A and y(b) = B to the length of
the curve y = y(x) between x = a and x = b. More generally, a functional
is any map from functions to real numbers. Sometimes the set of functions
on which a functional acts is restricted, as in the example that we have
just considered.
8
A functional S[f ] is a map from functions f to the real numbers.
Distance along a curve

The distance between two points (a, A) and (b, B) along a curve
y = y(x) that passes between them is given by the functional
Z bp
S[y] = 1 + y ′ (x)2 dx, y(a) = A, y(b) = B. (2)
a
Exercise 1
Let Pa = (−1, 0) and Pb = (1, 0). Verify the following statements. y
(a) If y = y(x) is the straight line joining Pa to Pb (i.e. the red line in
Figure 4), then the functional (2) has the value S[y] = 2.
(b) If y = y(x) is the upper semicircle with diameter the straight line from
Pa to Pb (i.e. the blue curve in Figure 4), so that x2 + y 2 = 1, then Pa Pb
S[y] = π. −1 0 1 x
(Hint: You may
Z assume the following standard integral, given in the Figure 4 The two paths in
1 Exercise 1
Handbook: √ dx = arcsin x.)
1 − x2
Exercise 2
Determine the value of the functional
Z 1
S[y] = y ′ (x)2 dx
0
for the following functions.
(a) y(x) = x
(b) y(x) = x3
1.2 Stationary paths

In this subsection we consider how the length of the path varies with y(x),
and we define the notion of a stationary path; we also show that the
straight line through the endpoints is a stationary path.
Recall that in ordinary calculus, a stationary point x of a function y(x) is
a point at which y ′ (x) = 0, that is, a point x for which the tangent to the
graph of y(x) at x is parallel to the x-axis. Once a stationary point has
been found, further work is required to determine whether the function has
a maximum, a minimum or a point of inflection there.
9
We want to define something analogous to this for functionals. To do this,

let us think of a concrete problem – finding the path with shortest distance
between the points (a, A) and (b, B) in the plane. Our problem then is to
find y = y(x) that will make
Z bp
S[y] = 1 + y ′ (x)2 dx, y(a) = A, y(b) = B,
a
as small as possible.
Suppose that y = y(x) is the desired ‘minimal path’. Then let g(x) be any
other (smooth) function that satisfies
g(a) = g(b) = 0, but otherwise g(x) is an arbitrary function of x.
Then we can define a new path by
ye(x, ε) = y(x) + ε g(x), where ε is a real number.
For each value of ε, this defines a new path ye that also passes through the
points (a, A) and (b, B), because ye(a, ε) = y(a) + ε g(a) = A and
ye(b, ε) = y(b) + ε g(b) = B. As g was defined arbitrarily (apart from its
y endpoints), every sufficiently well-behaved curve from a to b can be written
(b, B)
in this way. The situation is illustrated in Figure 5.
ye(x) Now, the length of the new path ye is given by
(a, A) Z bp
y(x) S[y + εg] = 1 + ((y + εg)′ )2 dx
g(x) a
Z bp
x = 1 + (y ′ + εg′ )2 dx, (3)
a
Figure 5 A minimal length and for fixed y and g, the functional S[y + εg] gives a different path length
path y(x) and a new path
for each value of ε. That is, for fixed y and g, S[y + εg] takes as input a
ye(x) = y(x) + ε g(x), both
passing through the points real number ε and outputs the length of y + εg, another real number, so
(a, A) and (b, B) S[y + εg] is a real-valued function. Furthermore, since y is our desired
minimal path, S[y + εg] must take its minimum value at ε = 0.
Mathematically, this translates to

d
S[y + εg] = 0,
dε ε=0
and this must be true for all functions g(x) that satisfy g(a) = g(b) = 0.
(The notation f (x)|x=a means evaluate f (x) at x = a; that is,
evaluate f (a).) It should be clear that this condition guarantees only that
S[y + εg] is stationary at ε = 0, and further work is required to determine
whether this is a maximum, a minimum or a point of inflection. A function
y(x) that satisfies this equation is said to be a stationary path of S.
Although we have derived this result for functionals that are path lengths,
the definition of stationary paths generalises to all functionals that we
consider in this module.
10
Stationary path of a functional

Let S[y] be a functional that maps functions y that satisfy y(a) = A
and y(b) = B to the real numbers. Any such function y(x) for which

d
S[y + εg] = 0, (4)
dε ε=0
for all functions g(x) that satisfy g(a) = g(b) = 0, is said to be a
stationary path of S, or alternatively a stationary curve or a
stationary function of S.
Also, if S and y(x) satisfy (4), then we say that S is stationary at
y(x), and sometimes we abbreviate this by merely saying that S[y] is
stationary.
Let us now use equation (4) to find the path between two points in a plane
with shortest length. From equation (3) we have
Z b
d d p
S[y + εg] = 1 + (y + εg ) dx
′ ′ 2 .
dε ε=0 dε a ε=0
The integration limits a and b are independent of ε, and the order of
integration and differentiation can be interchanged using Leibniz’s integral
rule, which we state (slightly informally) here.
Leibniz’s integral rule

For any sufficiently well-behaved function f (x, ε), we have
Z b Z b
d d
f (x, ε) dx = f (x, ε) dx,
dε a a dε
where a and b are fixed constants (that do not depend on ε).
Applying this rule gives

Z b p
d d
S[y + εg] = 1 + (y ′ + εg′ )2 dx
dε ε=0 a dε ε=0
Z b !
(y ′ + εg′ )g′

= p dx
a 1 + (y + εg ) ε=0
′ ′ 2
Z b
y′ g′
= p dx.
a 1 + (y ′ )2
If y(x) is a stationary path of S, then it follows, by definition, that
Z b
y ′ (x)
p g′ (x) dx = 0 (5)
′
1 + y (x)2
a
for all functions g(x) for which g(a) = g(b) = 0.
11
To solve this equation, we integrate by parts:

Z b Z b
′
b
u(x) g (x) dx = u(x) g(x) a − u′ (x) g(x) dx,
a a
p
with u = y ′ (x)/
1+ y ′ (x)2 .
The first term on the right-hand side vanishes
because g(a) = g(b) = 0, so we get
Z b Z b
ug′ dx = − u′ g dx,
a a
and hence equation (5) becomes
Z b !′
y ′ (x)
p g(x) dx = 0.
a 1 + y ′ (x)2
Because g is arbitrary (apart from the restriction g(a) = g(b) = 0), this
integral can vanish only if
!′
y ′ (x)
p = 0.
1 + y ′ (x)2
(In Section 4 we will give a more careful explanation of why this is so.)
Integrating both sides with respect to x gives
y ′ (x)
p = α, (6)
1 + y ′ (x)2
for some constant α. We could do some algebra here to solve for y ′ in
terms of α, but this is not necessary as the left-hand side of equation (6)
can be constant only if
y ′ (x) = m,
where m is some function of α – another constant.
Integration now gives the general solution
y(x) = mx + c,
for yet another constant c; this is the equation of a straight line, as
expected. The constants m and c are determined by the condition that the
straight line passes through Pa and Pb ; that is, y(a) = A and y(b) = B. It
is straightforward to show that this gives m = (B − A)/(b − a) and
c = (Ab − Ba)/(b − a), so
B−A Ab − Ba
y(x) = x+ . (7)
b−a b−a
This analysis shows that the functional S defined in equation (2) is
stationary along the straight line joining Pa to Pb .
Let us summarise what we have just learned.
12
Stationary path of the curve-length functional

The functional
Z bp
S[y] = 1 + y ′ (x)2 dx, y(a) = A, y(b) = B. (8)
a
gives the length of the curve y = y(x) that passes between the points
Pa = (a, A) and Pb = (b, B). The stationary path y of this functional
satisfies
y ′ (x) = constant.
Solving this differential equation and applying the boundary
conditions y(a) = A and y(b) = B gives the equation of a straight line
through Pa and Pb , which is the stationary path.
The main features of the procedure that you have just seen for finding the
stationary path between two points are common to all variational problems
in this module. That is, we start with some functional and some boundary
conditions on the functions that it maps; we use equation (4) to find the
functions y(x) that make it stationary; this leads to a differential equation
for the stationary function; then we solve the differential equation subject
to the boundary conditions to find the stationary function.
Classification of stationary paths

We have not shown that a solution y(x) of equation (4) is a path of
minimum distance between Pa and Pb . To do this, we must proceed
in the usual manner of ordinary calculus; that is, we must show that

d2
S[y + εg] > 0.
dε2
ε=0
We omit the algebra involved as it is tedious, and in any case, it is
physically obvious that a straight line gives the shortest distance.
More generally, our ultimate goal is in applying variational principles
to dynamics where one is interested only in finding functions that
make the functional stationary, not in finding maxima or minima. So
we concentrate on the former task.
In the next example and exercise we find the stationary paths of two more
functionals, in the same way that we found the stationary path of the path
length functional. In practice we rarely use this method to determine
stationary paths. Instead, we use the powerful Euler–Lagrange equation,
which is discussed in Section 4. To appreciate the derivation of the
Euler–Lagrange equation, however, you first need to understand the
method used in this section. This method will not be assessed in TMAs or
the examination.
13
Example 1
Let B ≥ −1. Consider the functional
Z 1p
S[y] = 1 + y ′ (x) dx, y(0) = 0, y(1) = B.
0
(a) Show that on a stationary path y(x) of S, we have y ′ (x) = constant.
(b) Hence show that the stationary path is the straight line y(x) = Bx.
(c) Also show
√ that the value of the functional on this line is
S[y] = 1 + B.
Solution
(a) In this case, the endpoints of the paths are (a, A) and (b, B), where
a = 0, A = 0 and b = 1. In order to find the stationary function, we
need to compute equation (4):

d
S[y + εg] = 0, where g(0) = g(1) = 0.
dε ε=0
Here
Z 1p
S[y + εg] = 1 + (y + εg)′ dx
0
Z 1p
= 1 + y ′ + εg′ dx,
0
so
Z 1
d d p

S[y + εg] = 1 + y + εg dx
′ ′ .
dε ε=0 dε 0 ε=0
The integration limits 0 and 1 are independent of ε, so the order of
integration and differentiation can be interchanged using Leibniz’s
integral rule. This gives
Z 1 p
d d
S[y + εg] = 1 + y ′ + εg′ dx
dε ε=0 0 dε ε=0
Z
1 1 g′

= √ dx
2 0 1 + y + εg ε=0
′ ′
Z
1 1 g′
= √ dx.
2 0 1 + y′
If S[y] is stationary, then it follows, by definition, that
Z 1
g′ (x)
p dx = 0
0 1 + y ′ (x)
for all functions g(x) for which g(0) = g(1) = 0. To solve this equation,
we integrate by parts:
Z 1 Z 1
′
1
u(x) g (x) dx = u(x) g(x) 0 − u′ (x) g(x) dx,
0 0
√
with u = 1/ 1 + y ′ .
14
2 Some examples of functionals
However, the first term on the right-hand side vanishes because

g(0) = g(1) = 0, so we get
Z 1 !′
1
p g(x) dx = 0.
0 1 + y ′ (x)
Because
p of the freedom in choosing g, this integral can vanish only if
(1/ 1 + y ′ (x))′ = 0. Integrating both sides with respect to x gives
1
p = α,
1 + y ′ (x)
where α is a constant. Rearranging this equation, we obtain y ′ = m,
for some constant m.
(b) Integrating the equation y ′ = m gives y = mx + c, for some constant c.
The boundary conditions y(0) = 0 and y(1) = B then give y = Bx,
which is the equation of the stationary path.
(c) With this value for y(x), the functional is
Z 1p Z 1
√ √
S[y] = 1 + y ′ (x) dx = 1 + B dx = 1 + B.
0 0
This is real if B ≥ −1.
Exercise 3
Consider the functional
Z 2
S[y] = x(y ′ )2 dx, y(1) = 0, y(2) = 1.
1
(a) Show that the stationary path satisfies x y ′ (x) = constant.
(b) Hence show that the stationary function is y(x) = ln x/ ln 2.

In this section we briefly describe a few problems that can be formulated
in terms of functionals (some of which are derived later), and which have
solutions that are stationary paths of the functionals. The list of problems
illustrates some of the types of question for which variational principles are
useful, and exposes you to the sorts of functionals that are commonly used.
The examples have been chosen because they require no specialist scientific
knowledge to understand them. As such, the list is by no means
exhaustive, and you should be aware that variational principles are applied
to a much wider class of problems than those considered here.
15
There are no exercises for this section as it will not be assessed, but it is
important for you to read it in order to understand the sections that follow.
2.1 Brachistochrone
0 b x The problem here is to find the shape of a wire joining two given points, so
that a bead will slide down the wire under gravity from one point to the
Pa other (without friction) in the shortest time (see Figure 6).
B Pb The name given to this curve is the brachistochrone, which comes from
the Greek brachystos, meaning shortest, and chronos, meaning time. If the
y
y-axis is vertical, and the two given points at the ends of the wire are
Figure 6 A bead sliding Pa = (0, 0) and Pb = (b, B), and the particle starts from rest, then it can
from point Pa to point Pb be shown that the time taken along the curve y(x) is
Z bs
1 1 + (y ′ )2
T [y] = √ dx, y(0) = 0, y(b) = B, (9)
2g 0 y
where g is the magnitude of the acceleration due to gravity. The map T is
a functional of the wire shape y(x): it maps functions y(x) to the times
taken for beads to descend them. The problem then is to find the function
y(x) that minimises T .
In Subsection 5.2 we will derive the functional T and obtain the equations
for the brachistochrone, the curve that minimises T . The curve turns out
to be a segment of a cycloid, which is the curve traced out by a point on
the rim of a circular wheel rolling in a straight line (see Figure 8).
Figure 7 A cycloid: the path of a point P on the rim of a wheel
History of the brachistochrone

The problem discussed in this subsection was first considered by
Galileo Galilei (1564–1642) in 1638, but, lacking the necessary
mathematical techniques, he concluded erroneously that the solution
is the arc of a circle passing vertically through Pa .
It was Johann Bernoulli (1667–1748) who made the problem famous
when in June 1696 he challenged readers of the scientific journal Acta
Eruditorum to solve it, reassuring them that the curve was well
known to geometers. He also stated that he would demonstrate the
solution at the end of the year, provided that no one else had.
16
In December 1696, Bernoulli extended the time limit to Easter 1697,

though by this time he was in possession of the solution by Gottfried
Wilhelm Leibniz (1646–1716), sent in a letter dated 16 June 1696 –
Leibniz having received notification of the problem on 9 June. Isaac
Newton (1642–1727) also solved the problem quickly, apparently on
the day of receipt, and published his solution anonymously.
2.2 Minimal surfaces of revolution

Here the problem is to find the surface of minimal area that is generated
by revolving a curve y(x) about the x-axis, where y(x) passes through two
given points Pa = (a, A) and Pb = (b, B), as shown in Figure 8.
y
(b, B)
(a, A)
a b x
Figure 8 Cylindrical-type shape produced when a curve y(x) is rotated

about the x-axis
The area of this surface is shown in Subsection 5.3 to be

Z b p
S[y] = 2π y 1 + (y ′ )2 dx, y(a) = A, y(b) = B. (10)
a
The map S[y] is a functional: it maps curves y(x) between Pa and Pb to
the areas of the surfaces formed by rotating such curves about the x-axis.
The problem is to find the curve that minimises S.
We will see that this problem has solutions that can be expressed in terms
of differentiable functions only for certain combinations of A, B and b − a.
Soap films
Many of us will have created soap film bubbles by dipping a loop of wire
into soap solution and blowing on it. It transpires that there is a close
connection between soap film surfaces and problems in the calculus of
variations.
To a good approximation, a soap film will try to minimise its surface area,
consistent with any wire boundary that contains it; it does this because it
wants to minimise its energy, which is proportional to its surface area.
17
For example, a soap film supported by a circular loop of wire will be a flat
circular disc.
The minimal surface of revolution can be created by dipping a pair of wire
hoops into a soap solution, and gently pulling them apart. The soap film
will be suspended between the two hoops, and the surface that it forms
will be the ‘minimal surface of revolution’, as shown in Figure 9.
2.3 A problem in navigation

Consider a river with straight, parallel banks a distance a apart, and a
boat that can travel with constant speed c in still water. The problem is to
Figure 9 Minimal surface of cross the river in the shortest time, starting and landing at given points,
revolution formed by a soap when there is a current.
film
We choose the y-axis to be the left bank, the line x = a to be the right
bank, and the starting point to be the origin. The water is assumed to be
y moving parallel to the banks with speed v(x), a known function of the
distance from the left bank. Then the time of passage along the path y(x)
from the point (0, 0) on the left bank to the point (a, A) on the right bank,
assuming that c > max(v(x)), can be shown to be
v(x) Z ap 2
c (1 + (y ′ )2 ) − v(x)2 − v(x) y ′
(a, A) T [y] = dx, y(0) = 0, y(a) = A.
0 c2 − v(x)2
(11)
The map T is a functional: it maps paths y(x) to the times to cross the
river, and the problem is to find paths that minimise T [y]. Notice that T
also depends explicitly on the speed profile v(x). This problem is
illustrated in Figure 10.
(0, 0) x 2.4 Catenary

A catenary is the shape assumed by an inextensible chain of uniform
Figure 10 Path of a boat density hanging between supports at both ends. In Figure 11 we show an
across a flowing river example of such a curve when the points of support, (−a, A) and (a, A),
have the same height.
If the shape of the chain is described by the function y(x), then it can be
shown that the potential energy E of the chain is proportional to the
y functional
Z a p
A
S[y] = y 1 + (y ′ )2 dx, y(−a) = A, y(a) = A, (12)
−a
which you encountered earlier, in equation (10).
x The function y that minimises this functional, subject to the length of the
−a 0 a
curve
Z ap
Figure 11 The catenary
formed by a uniform chain L= 1 + (y ′ )2 dx
hanging between two points −a
at the same height remaining constant, is the shape assumed by the hanging chain.
18
3 Reminder about partial and total derivatives
It turns out that the shape of the catenary is given by y = c cosh(x/c) + d,

where c and d are constants determined by the length of the chain and the
values of a and A. Proving this, however, requires minimisation of
functionals subject to constraints, which is beyond the scope of this
module.
3 Reminder about partial and total

derivatives
In the next section we will extend the kind of analysis that we performed
in Subsection 1.2 to more general types of functionals. To do this, we must
consider functions of an independent variable x, and of dependent
variables y(x) and y ′ (x) = dy/dx, for example F (x, y, y ′ ) = xy 2 + 3y ′ , and
we will have to differentiate such functions with respect to all three
variables. Although differentiating such functions is straightforward, there
are aspects of the process that can cause confusion at first. In this short
section we explain what is required.
Consider a function F (x, u, v) of three variables; for instance,
F (x, u, v) = xu2 + 3v.
Remember that each of the partial derivatives of a function of several
variables is obtained by differentiating with respect to one variable while
holding the others constant. The partial derivatives of F (x, u, v) are
∂F ∂F ∂F
= u2 , = 2xu and = 3.
∂x ∂u ∂v
If we make the simple substitutions u → y(x) and v → y ′ (x), and treat y
and y ′ as independent variables, then we obtain the function
F (x, y, y ′ ) = xy 2 + 3y ′ ,
and the partial derivatives
∂F ∂F ∂F
= y2, = 2xy and = 3.
∂x ∂y ∂y ′
Next, remember that the total derivative dF/dx is the rate of change of F
with respect to x, without holding the other variables constant. Because y
and y ′ depend on x, we must use the chain rule to find the total derivative
of F (x, y, y ′ ):
d ∂F ∂F dy ∂F dy ′
F (x, y, y ′ ) = + + ′
dx ∂x ∂y dx ∂y dx
∂F ∂F ′ ∂F ′′
= + y + ′y .
∂x ∂y ∂y
So in this case we obtain
dF
= y 2 + 2xyy ′ + 3y ′′ . (13)
dx
19
We can check that this expression gives the right result for particular
choices of y(x). For example, suppose that y = x3 ; then it follows that
y ′ = 3x2 and y ′′ = 6x. Substituting these values into the formula
F (x, y, y ′ ) = xy 2 + 3y ′ gives
dF
F (x, x3 , 3x2 ) = x7 + 9x2 , = 7x6 + 18x.
so
dx
On the other hand, equation (13) gives
dF
= (x3 )2 + 2x(x3 )(3x2 ) + 18x = 7x6 + 18x,
dx
so the two methods for finding the total derivative agree.
Here are a few exercises for you to try.
Exercise 4
p
Let F = x y 2 + (y ′ )2 , where y and y ′ are functions of x.
∂F ∂F ∂F dF
(a) Find , and ′
, and hence find .
∂x ∂y ∂y dx
dF
(b) If y = sin x, then find by
dx
(i) substituting y = sin x into the formula for F and differentiating
(ii) using the result from part (a).
Exercise 5
p
For the function F = x2 + y(y ′ )2 , where y and y ′ are functions of x, find
∂F ∂F ∂F dF
, , and .
∂x ∂y ∂y ′ dx
Exercise 6
For the function F = x2 y 3 , where y is a function of x, find
∂F ∂F dF
, and .
∂x ∂y dx
Also, show that

d ∂F ∂ dF
= .
dx ∂y ∂y dx
The formula

d ∂F ∂ dF
=
dx ∂y ∂y dx
established in Exercise 6 is valid for any sufficiently well-behaved function
F (x, y), where y is a function of x.
20
4 Euler–Lagrange equation
In this section we extend the analysis of Subsection 1.2 to find the
stationary paths of more general functionals. In particular, we derive the
Euler–Lagrange equation, which is the main result of this unit. The
importance of the Euler–Lagrange equation is that it can be used to find
stationary paths of a wide class of functionals without going through the
type of analysis that we performed in Subsection 1.2.
4.1 Derivation of the Euler–Lagrange equation

We have seen that the functionals given in equations (9)–(12) describe a
variety of physical situations. Furthermore, all of these functionals can be
written in a unified way:
Z b
S[y] = F (x, y, y ′ ) dx, y(a) = A, y(b) = B, (14)
a
where F (x, y, y ′ ) is an expression containing at least one of the functions y
and y ′ , and possibly also the variable x. (In contrast, notice that only y
appears as an argument of the functional S; that is, we write S[y].)
For example, equation (10) for the area of a surface of revolution has
p
F = 2πy 1 + (y ′ )2 ,
so F is a function of y and y ′ , but not of x. Also, equation (11) for the
time of navigation has
p
c2 (1 + (y ′ )2 ) − v(x)2 − v(x) y ′
F = ,
c2 − v(x)2
so F is a function of y and y ′ , and also of x because of the additional
function v(x), which measures water speed (for example, we may know
that v(x) = x(k − x), for some constant k).
With all these problems, we need to find a path y(x) that makes the
functional (14) stationary. To do this, we proceed in exactly the same
manner as in Subsection 1.2. That is, we start by assuming that y(x) is
the required stationary path. We then define a neighbouring path
ye(x, ε) = y(x) + ε g(x), where g(a) = g(b) = 0.
Here ε is a small real number, and g(x) is any sufficiently well-behaved
function that satisfies g(a) = g(b) = 0, so that the neighbouring path ye
passes through the same endpoints as y.
Now, the value of the functional along the neighbouring path is given by
Z b
de
y
S[ey] = F (x, ye, ye ′ ) dx, where ye ′ = .
a x
21
For fixed y and g, different values of ε give different values of

S[ey ] = S[y + εg], hence we can think of S[e y ] as just a function of ε. But y
is defined to be a stationary path, so S[ey ] must be stationary at ε = 0; that
is,

d
y ]
S[e = 0.
dε ε=0
This is exactly the same argument as in Subsection 1.2, and it leads to the
same result, equation (4).
The rate of change of the functional is given by
Z b
d d
S[e
y ] = F (x, ye, ye ) dx
′
dε ε=0 dε a ε=0
Z b
d
= F (x, ye, ye ′ ) dx, (15)
a dε ε=0
where we have used the fact that the integration limits a and b are
independent of ε, so that the order of integration and differentiation can be
interchanged using Leibniz’s integral rule.
Remember that ye and ye ′ both depend on ε, but x does not, so we can
apply the chain rule to give

d
′ ∂F dey ∂F de y ′
F (x, ye, ye ) = + .
dε ε=0 ∂ey dε ∂e y ′ dε ε=0
However, ye = y + εg, hence ye ′ = y ′ + εg′ , so

d
′ ∂F ∂F ′
F (x, ye, ye ) = g+ g .
dε ε=0 ∂e
y ∂ey′ ε=0
Now, F = F (x, ye, ye ′ ) depends on ε, whereas g = g(x) is independent of ε.
As ε → 0, we have ye → y and ye ′ → y ′ . Therefore

d
′ ∂F ∂F
F (x, ye, ye ) = g + ′ g′ ,
dε ε=0 ∂y ∂y
where now on the right-hand side F = F (x, y, y ′ ). The right-hand-side
expression is independent of ε.
Substituting this expression into equation (15) gives
Z b
d ∂F ∂F
S[e
y ] = g(x) ′
+ g (x) ′ dx.
dε ε=0 a ∂y ∂y
The second term in this integral can be rewritten using integration by
parts, to give
Z b Z b
′ ∂F ∂F b d ∂F
g (x) ′ dx = g(x) ′ − g(x) dx.
a ∂y ∂y a a dx ∂y ′
22
But g(a) = g(b) = 0, so the first term on the right-hand side vanishes, and
the rate of change of the functional becomes
Z b
d ∂F d ∂F
S[e
y ] = g(x) − g(x) dx
dε ε=0 a ∂y dx ∂y ′
Z b
d ∂F ∂F
=− g(x) − dx. (16)
a dx ∂y ′ ∂y
If S[y] is stationary, then by definition this integral vanishes for all
functions g(x). Because g is arbitrary, this can occur only if

d ∂F ∂F
′
− = 0, y(a) = A, y(b) = B. (17)
dx ∂y ∂y
Deriving the differential equation

You can see how (17) follows from (16) using a ‘proof by
contradiction’, as follows.
Suppose that the bracketed expression on the right of equation (16) is
not zero everywhere. Then because g(x) is chosen arbitrarily, we can
choose it to have the same sign as the bracketed expression for every
value x. Hence for this choice of g, the integrand is positive whenever
the bracketed expression is not zero. As a consequence, the integral is
positive too, which contradicts the fact that y is a stationary
function. Hence the bracketed expression must be zero everywhere,
which establishes equation (17).
Equation (17) is known as the Euler–Lagrange equation, and is hugely

important in mathematical physics. Solutions of the Euler–Lagrange
equation are stationary paths of the functional S[y] in equation (14).
When applied to a variational problem, the Euler–Lagrange equation
usually produces a second-order differential equation, which must be solved
subject to the boundary conditions in equation (14) in order to obtain the
stationary path.
Note that it is usually much easier to use the Euler–Lagrange equation to
find stationary paths than it is to use the type of analysis that we
performed in Subsection 1.2.
The Euler–Lagrange equation is the main result of this unit, and it is
essential that you remember the equation and how to use it (although you
do not need to remember how it is derived).
23
Euler–Lagrange equation
The functional
Z b
S[y] = F (x, y, y ′ ) dx, y(a) = A, y(b) = B, (18)
a
has stationary paths that are given by solving the Euler–Lagrange
equation

d ∂F ∂F
′
− = 0, y(a) = A, y(b) = B. (19)
dx ∂y ∂y
Let us apply the Euler–Lagrange equation in an example and some

exercises.
Example 2
Find the stationary path of the distance functional
Z bp
S[y] = 1 + y ′ (x)2 dx, y(a) = A, y(b) = B, (20)
a
considered in Section 1 (equation (2)).
Solution
Comparing equations (18) and (20), we see that the integrand is given by
p
F (x, y, y ′ ) = 1 + y ′ (x)2 .
To use the Euler–Lagrange equation, we need to calculate ∂F/∂y and
∂F/∂y ′ . The first thing to notice is that the integrand depends explicitly
only on y ′ , not on x or y. Therefore
∂F y′ ∂F
′
= p and = 0.
∂y ′
1 + (y )2 ∂y
Hence the Euler–Lagrange equation (19) becomes
!
d y′
p = 0. (21)
dx 1 + (y ′ )2
We could differentiate this to get a second-order differential equation for y,
which we could then solve by integration. However, equation (21) is
already in a convenient form to be integrated directly.
Integrating both sides of equation (21) with respect to x gives
y′
p = α, (22)
1 + (y ′ )2
for some constant α. This is identical to equation (6), which is the
equation that we derived in Subsection 1.2 for the stationary path. Its
solution (equation (7)), was shown to be
B−A Ab − Ba
y= x+ .
b−a b−a
24
Exercise 7
Z b

S[y] = (y ′ )2 + y dx, y(0) = 0, y(1) = 2.
a
(a) Show that the stationary paths of this functional must satisfy
2y ′′ − 1 = 0, y(0) = 0, y(1) = 2.
(b) Solve this differential equation to show that the stationary path is
y = 14 x2 + 74 x.
Exercise 8
(a) Show that the Euler–Lagrange equation for the functional
Z X

S[y] = (y ′ )2 − y 2 dx, y(0) = 0, y(X) = 1,
0
where 0 < X < π, is
y ′′ + y = 0.
(b) Hence show that the stationary function is y = sin x/ sin X.
Exercise 9
(a) Show that the Euler–Lagrange equation for the functional
Z 1

S[y] = (y ′ )2 + y 2 + 2xy dx, y(0) = 0, y(1) = α,
0
where α is a constant, is
y ′′ − y = x.
(b) Hence show that the stationary function is

α+1
y= (ex − e−x ) − x.
e − e−1
4.2 Alternative forms of the Euler–Lagrange

equation
In Example 2, you saw an example of an integrand
p
F (x, y, y ′ ) = 1 + y ′ (x)2
used to define a functional, which did not depend explicitly on y. In this
case, ∂F/∂y = 0, and the Euler–Lagrange equation (19) reduces to

d ∂F
= 0.
dx ∂y ′
25
After integrating with respect to x, we obtain

∂F
= constant.
∂y ′
(Compare these two equations with equations (21) and (22).) Thus we
arrive at the following result.
Special case of the Euler–Lagrange equation

If a functional has the form
Z b
S[y] = F (x, y ′ ) dx, y(a) = A, y(b) = B, (23)
a
in which the integrand does not depend explicitly on y, then the
Euler–Lagrange equation can be integrated to give
∂F
= constant.
∂y ′
If, on the other hand, the integrand does not depend explicitly on x, so
that the functional has the form
Z b
S[y] = F (y, y ′ ) dx, y(a) = A, y(b) = B,
a
then it can be shown that the Euler–Lagrange equation (19) reduces to
∂F
y′ − F = c, y(a) = A, y(b) = B,
∂y ′
where c is a constant determined by the boundary conditions. We will
derive this equation in Exercise 12. The expression on the left-hand side of
the equation is called the first integral of the Euler–Lagrange equation.
First integral of the Euler–Lagrange equation

If a functional has the form
Z b
S[y] = F (y, y ′ ) dx, y(a) = A, y(b) = B, (24)
a
where the integrand does not depend explicitly on x, then the
first integral of the Euler–Lagrange equation is
∂F
y′ − F = c, y(a) = A, y(b) = B, (25)
∂y ′
where c is a constant determined by the boundary conditions.
The stationary path of the functional S is determined by solving this
first-order differential equation.
26
Equations (23) and (25) are important because they both give rise to
first-order differential equations, whereas the most general form of the
Euler–Lagrange equation gives rise to a second-order differential equation
– and first-order equations are usually easier to solve than second-order
equations.
Exercise 10
Z 2
S[y] = (ln y ′ + y) dx, y(1) = ln 2, y(2) = ln 3.
1
(a) Show that the first integral of the Euler–Lagrange equation is
dy
= αe−y , y(1) = ln 2, y(2) = ln 3,
dx
for some constant α.
(b) Hence show that the stationary function is y = ln(x + 1).
Exercise 11
Z 1p
S[y] = y(1 + (y ′ )2 ) dx, y(−1) = y(1) = A,
−1
where A ≥ 1.
(a) Show that the first integral of the Euler–Lagrange equation for this
functional is
1p
y′ = y − k2 , y(−1) = y(1) = A.
k
(b) Hence show that the functional is stationary on the paths
x2 p
y(x) = h + , where h = 12 A ± A2 − 1 .
4h
Exercise 12
(a) Suppose that F (y, y ′ ) does not depend explicitly on x, so that
∂F/∂x = 0. Show that

d ′ ∂F ′ d ∂F ∂F
y −F =y − .
dx ∂y ′ dx ∂y ′ ∂y
(Hint: Evaluate the derivative on the left-hand side.)
(b) Hence show that equation (25) is true if either (i) y(x) satisfies the
Euler–Lagrange equation, or (ii) y(x) is constant.
27
5 Applications of the Euler–Lagrange

equation
In this section we consider some applications of the Euler–Lagrange
equation to physical problems. We begin with a discussion of Fermat’s
principle, which describes the trajectories of light rays. This topic is the
most important one of this section because it forms a useful stepping stone
to the next unit, on Lagrangian mechanics.
After Fermat’s principle, we revisit two of the examples of Section 2 – the
brachistochrone and the soap film. These are interesting applications that
expose you to the sorts of method used to construct functionals, and they
exhibit the type of behaviour that can be modelled using the calculus of
variations. However, solving the Euler–Lagrange equation for these two
problems is beyond what we would expect from you in assessment of this
module. So do not worry if some of the details of Subsections 5.2 and 5.3
seem more advanced than the rest of this unit. You will not be assessed on
the fine details of these topics. However, the examples and exercises
distributed throughout this section are typical of the sorts of question on
which you will be assessed. It is therefore very important that you follow
the examples and complete the exercises.
5.1 Fermat’s principle

In most circumstances, light can be considered to travel along lines from a
source (such as a light bulb) to an observer (such as a human eye); these
lines are called light rays, or just rays, and are often straight lines. This
is why most shadows have distinct edges, and why eclipses of the Sun are
so spectacular.
A simple experiment with light rays involves reflecting light in a plane
mirror, as shown in Figure 12. In the diagram there is a fixed point source
of light S, which emits light rays in all directions. One light ray is reflected
from the mirror at R and travels to an observer fixed at O, as shown in the
diagram. Labelled in the figure are the angle θ1 between the line SR and
the vertical line through R, and the angle θ2 between the line RO and the
vertical line through R. (The plane of the mirror is perpendicular to the
page, and it is assumed that the plane SRO is in the page.)
We wish to determine the location of R. The key to this, obtained from
experiment, is to observe that for light rays reflected in this way, the
incoming angle θ1 , called the angle of incidence, is equal to the outgoing
angle θ2 , called the angle of reflection. Using the fact that θ1 = θ2 , we
can determine the x-coordinate of R using elementary geometry. (In brief,
because θ1 = θ2 , the two right-angled triangles with hypotenuses SR and
OR are similar, from which it follows that x/h1 = (d − x)/h2 . Rearranging
this equation gives x = dh1 /(h1 + h2 ). We will not use this exact formula
for x again.)
28
5 Applications of the Euler–Lagrange equation
h2
S
h1 θ1 θ2
R
x d−x
d
Figure 12 A light ray emitted from S, reflected at R, and observed at O
This observation about the two angles has an interesting consequence.

Consider any path from S to R to O of the type shown in Figure 12 (not
necessarily the path of a light ray; θ1 and θ2 may differ). By applying
Pythagoras’ theorem to the triangles, we can see that the total length of
the path is
q q
f (x) = x2 + h21 + (d − x)2 + h22 .
It turns out (see Exercise 13) that this function of x has a minimum when
θ1 = θ2 ; so the light ray that travels from S to O via reflection in the
mirror minimises the distance that it travels.
This result was generalised by the French mathematician Pierre de Fermat
(1601–1665) into what is now known as the principle of least time.
The principle of least time states that the path taken between two
points by a ray of light is the path that can be traversed in the least
time.
(Note that this principle is not quite correct; an amended statement
will be given shortly.)
This principle means that light will always travel in straight lines when it
passes through a single uniform medium like air, because it has a constant
speed, so the shortest distance is always the fastest time of passage.
However, light has different speeds in different media: for example, in a
vacuum the speed of light is c ≃ 2.9 × 108 m s−1 , whereas in water it is
cwater ≃ 2.25 × 108 m s−1 . (Note that the letter c is conventionally used to
denote the speed of light in vacuum. Take care not to get this mixed up
with any constants of integration that you introduce!) Generally, if the
speed of light in a uniform medium is cm , then we define the refractive
index of the medium to be the ratio n = c/cm ; so the refractive index of a
medium is just another way of characterising the speed of light that passes
through it.
29
Consider now a general (possibly non-uniform) medium in the Cartesian

plane, with refractive index n(x, y) at the point (x, y). Then the speed of
light will vary in space according to the formula cm (x, y) = c/n(x, y), and
the time taken for light to travel along an infinitesimal line segment of
length δs is
δs 1
δt = = n(x, y) δs.
cm (x, y) c
However,
p in Subsection 1.1 we showed that along a path y = y(x), we have
δs = 1 + y ′ (x)2 δx (see equation (1)). Therefore
1 p
δt = n(x, y) 1 + y ′ (x)2 δx.
c
So for light moving in the Cartesian plane, in a medium with refractive
index n(x, y), with the source at the point (a, A) and observer at the point
(b, B), the time of passage T along a path y(x) joining these points is
Z p
1 b
T [y] = n(x, y) 1 + (y ′ )2 dx, y(a) = A, y(b) = B. (26)
c a
In the context of the calculus of variations, the principle of least time states
that the path taken will minimise this functional. In fact, this is not quite
correct: light rays actually take paths that make this functional stationary;
that is, light rays follow paths that are stationary paths of equation (26).
Normally, such paths do minimise the time of passage, but there are
situations in which they also maximise it. We call our revised principle
Fermat’s principle; it supersedes the principle of least time stated earlier.
Fermat’s principle says that the path taken between two points by
a ray of light is a stationary path of the time functional (26).
In other texts, the phrase ‘principle of least time’ is used for the statement
that we call Fermat’s principle. Here we reserve the phrase ‘Fermat’s
principle’ for the correct statement about the path travelled by light rays.
For light travelling through a uniform material (i.e. n(x, y) = n, for some
constant n and all points (x, y)), the functional (26) is proportional to the
distance functional of Subsection 1.1. Therefore the stationary paths are
straight lines. For a general non-uniform medium, in which the refractive
index depends on the position, light rays follow curved paths. Mirages are
one consequence of a position-dependent refractive index, where light
passes through layers of air of differing temperature and therefore density,
leading to distorted images – see Figure 13.
Figure 13 Sunset mirage
30
Snell’s law of refraction

Fermat’s principle can be used to show that for light reflected in a
mirror, the angle of incidence equals the angle of reflection. For light
crossing the boundary between two media, it gives Snell’s law,
named after the Dutch astronomer and mathematician Willebrord
Snellius (1580–1626), which says that
sin α1 c1
= ,
sin α2 c2
where α1 and α2 are the angles between the ray and the normal to
the boundary, and c1 and c2 are the speeds of light in the media, as
shown in Figure 14.
Air α1
N
α2
S′
Water
Figure 14 A ray of light is refracted as it passes from water to air
In water the speed of light is approximately c2 = c1 /1.3, where c1 is

the speed of light in air. It follows that 1.3 sin α2 = sin α1 , so α1 > α2 .
In Figure 14, the light from an object S in water travels to the
observer O in the air along the two straight lines SN and N O. The
observer perceives the object to be at S ′ , on the straight line
through ON . This explains why, for instance, a stick partly
submerged in water appears bent.
Intuitively, we can see why light bends in this way: its speed is
greater in air than in water, so to minimise the time of travel, it needs
to extend its path in air and reduce its path in water.
The situation is similar to that of a lifeguard on a beach, who wants
to reach a swimmer in distress in the fastest possible time. Since the
lifeguard can run faster than she can swim, she must extend her path
on the sand and reduce her path in the water, as shown in Figure 15.
31
Beach
Sea Help!
Figure 15 Path taken by a lifeguard to reach a distressed swimmer
Connection to Lagrangian mechanics

Equation (26) is sometimes written in the form
Z b
T [y] = L(x, y, y ′ ) dx, y(a) = A, y(b) = B, (27)
a
where
1 p
L(x, y, y ′ ) =n(x, y) 1 + (y ′ )2
c
is called the optical Lagrangian and T is called the optical action.
Fermat’s principle tells us that light rays passing between (a, A) and (b, B)
travel along paths that make the optical action stationary.
This is analogous to what we will find in the next unit. There we will
demonstrate that the equations of motion of mechanical objects can be
obtained by finding the stationary paths of an action like (27), but with a
different form for the Lagrangian L, and where the independent variable is
time t instead of distance x.
Example 3
Consider a layer of transparent medium with refractive index
√
n(x, y) = 4 y that is located in the part of the plane
given by 18 ≤ y ≤ 2.
Findthe path of a light ray from a source at −1, 54 to an observer at
1, 54 .
Solution
By Fermat’s principle, the light ray will be a stationary path of the
functional
Z
4 1p
T [y] = y(1 + (y ′ )2 ) dx, y(−1) = 54 , y(1) = 54 .
c −1
32
Apart from the factor 4/c, this is exactly the functional of Exercise 11,
with the same boundary conditions if A = 54 .
From the solution to Exercise 11(b), the functional is stationary on the two
paths
x2 p
y(x) = h + , where h = 12 A ± A2 − 1 .
4h
As A = 54 , we see that h = 1 or h = 14 , so there are two solutions:
y+ (x) = 1 + 14 x2 and y− (x) = 1
4 + x2 .
described by y+ and y− are shown in Figure 16. The observer

The paths
5
at 1, 4 sees two light rays coming from different directions.
y
2

−1, 54 1, 54
y+
1
1 y−
4
−1 0 1 x
Figure 16 Graph of solutions
Note that the paths of the light rays depend on the refractive index n, but
they do not depend on c, the explicit value of the speed of light in a
vacuum.
Here are some exercises for you to try.
Exercise 13
Show that the function
q q
f (x) = x2 + h21 + (d − x)2 + h22 ,
where h1 , h2 , x and d represent the distance measurements of Figure 12, is
stationary when θ1 = θ2 , where
x d−x
sin θ1 = p and sin θ2 = p .
x2 + h21 (d − x)2 + h22
Show that at this stationary value, f (x) has a minimum.
33
Exercise 14
Consider a layer of transparent medium with refractive index given by
n(x, y) = 2/y that is located in the part of the plane given by 12 ≤ y ≤ 2.
(a) Use Fermat’s principle to show that the path of a light ray located
within this medium is described by the differential equation
dy 1 p
= 1 − k2 y 2 for 12 ≤ y ≤ 2,
dx ky
where k is a constant.
(b) Find the path of a light ray from source at (−1, 1) to observer at (1, 1).
5.2 Revisiting the brachistochrone

In Subsection 2.1, you learned about the problem of finding the shape of a
wire joining two given points, such that a bead starting from rest from one
point will slide down the wire under gravity to the other point, without
friction, in the shortest time. Here you will first see how to derive the
functional for this system, and then learn how to solve it.
Deriving the functional of the brachistochrone problem

Let us formulate the functional for the brachistochrone problem by
obtaining an expression for the time of passage between given points
Pa = (a, A) and Pb = (b, B) along a curve y(x), using the principle of
conservation of energy. Such arguments involving energy are possible only
because we have assumed that the bead slides without friction.
We use a coordinate system with the y-axis oriented vertically downwards.
Let us suppose that the particle starts from rest at the origin; that is,
a = 0 and A = 0. To simplify the problem further, we assume that the end
of the wire is a unit distance below the starting point; that is, Pb = (b, 1),
where b is positive and is allowed to vary. A typical curve y(x) is shown in
Figure 17. Labelled in that figure is the distance s(x) from the origin to a
point P = (x, y(x)) on the curve.
O b x
(0, 0)
s(x)
1
(b, 1)
y
Figure 17 A point P on the curve y(x)
34
The speed of the bead at the point P = (x, y(x)) is v = ds/dt. Let m
denote the mass of the bead. Then the kinetic energy of the bead at P is
1 2
2 mv , and the potential energy of the bead at that point is −mgy. (The
sign of the potential energy is negative because the y-axis points
downwards; we therefore expect potential energy to decrease as y
increases.)
The total energy E of the bead is the sum of its kinetic and potential
energies:
E = 12 mv 2 − mgy.
Initially, the value of E is 0, because the initial values of v and y are 0.
Moreover, because there is no friction, energy is conserved, so E is
always 0; that is,
1 2
2 mv − mgy = 0. (28)
As we saw earlier (Figure 3), small changes in s are given by
δs2 = δx2 + δy 2 , so
2 2 2
ds dx dy
= +
dt dt dt
2 2 , 2 !
dx dy dx
= 1+
dt dt dt
2
dx
= 1 + (y ′ )2 .
dt
Here we have used the fact that
dy dy dx
= ,
dt dx dt
by the chain rule. As v = ds/dt, we can write equation (28) as
2
dx
1
2m 1 + (y ′ )2 − mgy = 0.
dt
Rearranging this gives
2
dx 2gy
= .
dt 1 + (y ′ )2
Now, the time of passage from x = 0 (at time t = 0) to x = b (at time
t = T ) is given by the integral
Z T Z b Z bs
1 1 + (y ′ )2
T = dt = dx = dx.
0 0 dx/dt 0 2gy
Therefore the functional for the time taken for a bead to slide down the
wire with shape y(x), starting from rest at the origin, is
Z bs
1 1 + (y ′ )2
T [y] = √ dx, y(0) = 0, y(b) = 1. (29)
2g 0 y
35
Exercise 15
Use equation (29) to show that the time taken for a particle of mass m to
slide down the curve y = x/b from rest at the origin to the point (b, 1) is
s
b2 + 1
T =2 .
2g
Minimising the functional of the brachistochrone problem

To determine the shape of the brachistochrone, we need to minimise the
functional (29). The integrand
s
1 1 + (y ′ )2
F (y, y ′ ) = √
2g y
of this functional is independent of x, so we can use the first integral of the
Euler–Lagrange equation (25):
∂F
y′ − F = constant.
∂y ′
We obtain
s
1 (y ′ )2 1 1 + (y ′ )2
√ p −√ = constant.
2g y(1 + (y ′ )2 ) 2g y
Simplifying the left-hand side of this equation, and multiplying by −1,
gives
1 1
p = ,
′ 2
y(1 + (y ) ) c
√
where c is a constant. (We have absorbed the factor 2g into the
constant.) Squaring both sides and rearranging gives y 1 + (y ′ )2 = c2 ,
from which we get
s s
dy c2 c2 − y
=± −1=± .
dx y y
This first-order differential equation is separable and can be solved. First,
however, note that because the y-axis is vertically downwards, we expect
the solution y(x) to increase as x increases, so we take the positive sign in
this equation. Separating the variables now gives
Z Z r
y
dx = dy.
c2 − y
Of course, the integral on the left-hand side is just x (plus a constant,
which we absorb into the constant from the right-hand side).
36
To solve the integral on the right-hand side, we substitute y = c2 sin2 φ, so

that dy = 2c2 sin φ cos φ dφ, to obtain
Z r Z s
y 2 c2 sin2 φ
dy = 2c sin φ cos φ dφ.
c2 − y c2 − c2 sin2 φ
Using the trigonometric identities 1 − sin2 φ = cos2 φ and
sin2 φ = 12 (1 − cos 2φ), we obtain
s
Z r Z
y 2 sin2 φ
dy = 2c sin φ cos φ dφ
c2 − y cos2 φ
Z
2
= 2c sin2 φ dφ
Z
2
=c (1 − cos 2φ) dφ
= 12 c2 (2φ − sin 2φ) + d,

where d is a constant.
In summary, we have
x = 12 c2 (2φ − sin 2φ) + d and y = c2 sin2 φ = 12 c2 (1 − cos 2φ),
which are solutions of the functional in parametric form (where the
parameter is φ). When φ = 0, these equations give x = d when y = 0.
However, we chose our axes such that x = 0 when y = 0; hence d = 0, and
the equations become
x = 12 c2 (2φ − sin 2φ) and y = 12 c2 (1 − cos 2φ). (30)
The curve described by these equations passes through the origin, and the
constant c is determined by the fact that the curve also passes through
(b, 1). Suppose that φ = φb when (x, y) = (b, 1). Then
b = 12 c2 (2φb − sin 2φb ) and 1 = 12 c2 (1 − cos 2φb ),
so 12 c2 = 1/(1 − cos 2φb ). Therefore our parametric equations are
2φ − sin 2φ 1 − cos 2φ
x= and y= for 0 ≤ φ ≤ φb , (31)
1 − cos 2φb 1 − cos 2φb
where φb is given by
2φb − sin 2φb b
b= . (32)
1 − cos 2φb
12
Clearly, if φ = 0, then (x, y) = (0, 0), and if φ = φb , then (x, y) = (b, 1). 10
8
Equation (32) cannot be solved analytically for φb (there is no closed 6
formula for φb ), and instead we must solve it numerically. Figure 18 shows 4
a graph of b against φb . The graph starts at the origin, and b increases as 2
φb increases. You can see from the graph that b → ∞ as φb → π, a fact
easily verified using equation (32). The graph shows us that for every 0 1 2 φb
value of b, a unique value of φb can be found. You are asked to reproduce Figure 18 Graph of b as a
this graph in Exercise 18. function of φb
37
To solve the brachistochrone problem for any value b, we must solve

equation (32) for φb , then substitute the result into the parametric
equations (31). The result can be plotted using suitable computer
software. Figure 19 shows graphs of the stationary paths for various values
of b; all the curves start at (0, 0) and end at (b, 1).
0 1 2 3 4 x
0.5
1.5
y
Figure 19 Graphs of the stationary paths joining the points (0, 0) and
(b, 1) for b = 0.2 (red), b = π/2 (black), b = 3 (blue) and b = 4 (green)
From the figure, we see that for small values of b, the stationary path is
close to that of a straight line, as you might expect. In fact, if b < π/2,
then the stationary path crosses the y = 1 line just once, at the terminal
point. For b > π/2, the stationary path crosses the y = 1 line twice. The
critical value of b, where the stationary path is tangential to the line y = 1
at its terminal point, occurs at b = π/2 (see Exercise 16(c) for a derivation
of this result).
Cycloid
A cycloid is the curve traced by a point on the rim of a circular
wheel as the wheel rolls along a straight line without slippage.
Figure 20 illustrates a cycloid.
y
θ
C B
r
O A D x
Figure 20 Cycloid (in red) traced out by a point on the rim of a

rolling wheel
In this figure, a circle of radius r rolls along the x-axis, starting with
its centre on the y-axis. Fix attention on the point P attached to the
circle, initially at the origin O. As the circle rolls, P traces out the
cycloid through O, P and D.
38
The cycloid has been studied by many mathematicians from the time
of Galileo (1564–1642), and was the cause of so many controversies
and quarrels in the 17th century that it became known as ‘the Helen
of geometers’ (a reference to ‘Helen of Troy’ from Greek mythology).
Galileo named the cycloid but had limited understanding of the
mathematics behind it. He did, however, suggest that it would make
an attractive arch for a bridge. This suggestion was implemented in
1764 with the building by James Essex (1722–1784) of Trinity College
Bridge, in the grounds of Trinity College, over the river Cam in
Cambridge, which has three cycloidal arches. The cycloid is still
occasionally used in architecture today. Other notable
mathematicians to have studied the cycloid include René Descartes
(1596–1650), Pierre de Fermat (1601–1665), Blaise Pascal
(1623–1662), Christiaan Huygens (1629–1695) and Christopher Wren
(1632–1723), the architect of St Paul’s Cathedral.
The equation of the cycloid is obtained by finding the coordinates
of P . In Figure 20, after the circle has rolled through an angle θ, the
length of the longer circular arc between P and A is rθ. Because
there is no slipping, OA = rθ also, so C has coordinates (rθ, r). Next,
by applying trigonometry to the triangle with vertices P , B and C,
we see that P B = −r cos θ and BC = −r sin θ, so P has coordinates
x = r(θ − sin θ) and y = r(1 − cos θ),
which together give the parametric equation of the cycloid. These are
identical to equation (30) if we make the transformation θ = 2φ and
r = c2 /2. So the shape of the brachistochrone has a simple geometric
interpretation: it is a segment of a cycloid.
Exercise 16
(a) Show that the gradient of the brachistochrone described by
equations (31) and (32) is
dy sin 2φ 1
= = .
dx 1 − cos 2φ tan φ
(Hint: Use the identities sin 2φ = 2 sin φ cos φ and
cos 2φ = 1 − 2 sin2 φ.)
(b) Deduce that at the origin (0, 0), the brachistochrone hangs vertically
downwards.
(c) Show that the critical brachistochrone, which is the brachistochrone
with parameter b that is tangent to the line y = 1 at its terminal point
(the black curve in Figure 19), satisfies b = π/2.
39
Exercise 17
Z 1p
S[y] = 1 + x + (y ′ )2 dx, y(0) = 0, y(1) = 1.
0
(a) Show that the function y(x) defined by the relation
p
y ′ = c 1 + x + (y ′ )2 ,
where c is a constant, makes S[y] stationary.
(b) By expressing y ′ in terms of x, solve this equation to obtain
(1 + x)3/2 − 1
y= .
23/2 − 1
Exercise 18 Computing – optional

(a) Use computer software and equation (32) to create a plot of b
against φb similar to Figure 18.
(b) Modify the code given in the section ‘Example of plotting a
brachistochrone’ of the Computer Exploration Worksheet to create a
plot of four brachistochrones with parameters b = 0.2, π/2, 3, 4, similar
to Figure 19.
5.3 Revisiting minimal surfaces of revolution

Minimal surfaces of revolution were briefly discussed in Subsection 2.2.
The problem is to find the function y(x) with given endpoints y(a) = A
and y(b) = B, where A, B ≥ 0, such that the area of the surface formed by
rotating the curve y(x) about the x-axis is minimised. You saw that to a
good approximation, this surface is the same as that of a soap film
suspended between a pair of wire hoops that lie in planes perpendicular to
the x-axis, with radii A and B, and centres (a, 0) and (b, 0), respectively.
This subsection has several parts. First we derive the functional S[y] for
surface area. Then we apply the Euler–Lagrange equation to obtain the
differential equation that a function y(x) must satisfy to make the
functional stationary. Next we solve this differential equation in a simple
case, discovering that even this special case has subtleties. Finally, we
discuss the physical interpretation of our results.
40
Deriving the surface of revolution functional

Figure 21(a) shows the surface of revolution of a curve y(x) that satisfies
y(a) = A and y(b) = B. To obtain an expression for the area of this
surface, we first find the area of a thin disc of width δx, such as that shown
in Figure 21(b). The small segment of length δs can be approximated by a
straight line provided that δx is sufficiently small, so using the analysis of
Subsection 1.1 (see equation (1)), we obtain
p
δs = 1 + y ′ (x)2 δx.
The area δS traced out by this segment as it rotates about the x-axis is
p
δS = 2π y(x) δs = 2π y(x) 1 + y ′ (x)2 δx.
Hence the area of the whole surface from x = a to x = b is
Z b p
S[y] = 2π y(x) 1 + y ′ (x)2 dx, y(a) = A, y(b) = B, (33)
a
where A, B ≥ 0.
y y
(b, B)
(a, A) δs
x x
δx
(a) (b)
Figure 21 (a) Surface of revolution; (b) a narrow cross-section of that

surface
Exercise 19 y (b, B)
(a) Show that the equation of the straight line joining the points (0, 0) and
(b, B) is
B (0, 0) x
y= x.
b
(b) Use equation (33) to show that the area of the cone (without a base)
shown in Figure 22 is
p Figure 22 Cone of height b
S = πB b2 + B 2 . and base radius B, base not
included
41
Applying the Euler–Lagrange equation to the surface of

revolution functional
The integrand of the functional (33) is 2π F (y, y ′ ), where
p
F (y, y ′ ) = y 1 + (y ′ )2 .
This function does not depend explicitly on x, so we can use the first
integral of the Euler–Lagrange equation. As
∂F yy ′
= p ,
∂y ′ 1 + (y ′ )2
we see that
∂F y(y ′ )2 p y
y′ ′ − F = p − y 1 + (y ′ )2 = − p .
∂y 1 + (y ) ′ 2 1 + (y ′ )2
Applying the first integral of the Euler–Lagrange equation, equation (25),
gives
y
p = y0 , y(a) = A, y(b) = B,
1 + (y ′ )2
for some constant y0 (this constant absorbs the factor 2π that we
suppressed earlier). By squaring both sides of this equation and
rearranging, we get y 2 /y02 = 1 + (y ′ )2 , from which we obtain the first-order
differential equation
p
dy y 2 − y02
= , y(a) = A, y(b) = B. (34)
dx y0
(Note that the constant y0 , which could be positive or negative, but not
zero, absorbs the factor ±1 that we obtain by taking square roots.) The
solutions of equation (34) are stationary paths of the functional (33).
Solving the differential equation for the minimal surface of

revolution
The differential equation (34) can be solved by separating variables. The
solution involves the hyperbolic cosine function (also known as cosh), but
it is not necessary to know many properties of hyperbolic sines or cosines
to follow the subsequent analysis. However, if you have not come across
these functions previously, then you might like to read the following box,
which contains a brief summary of their properties, before continuing.
Brief summary of hyperbolic functions

Recall that Euler’s formula eix = cos x + i sin x enables us to express
sines and cosines in terms of complex exponentials:
1 1
cos x = (eix + e−ix ) and sin x = (eix − e−ix ).
2 2i
42
The hyperbolic sine function, also called sinh (pronounced ‘shine’,

or sometimes ‘sinch’), and the hyperbolic cosine function, also
called cosh (pronounced ‘cosh’), are defined in analogy to these
formulas, but with real exponentials:
cosh x = 12 (ex + e−x ) and sinh x = 12 (ex − e−x ).
Using these definitions, we can show that hyperbolic sines and cosines
have many similar (although not identical) properties to ordinary
sines and cosines. For example,
d d
cosh x = sinh x and sinh x = cosh x.
dx dx
Their graphs, however, look quite different, as we can see from
Figure 23(a).
The inverses of the hyperbolic functions can be shown to be
p
arccosh x = ln(x + x2 − 1), x ≥ 1,
p
arcsinh x = ln(x + x2 + 1).
The graphs of the inverse hyperbolic functions are shown in
Figure 23(b).
y y
y = cosh x y = arcsinh x
3
2 2
1
y = sinh x y = arccosh x
−2 −1 0 1 2 x −10 −5 0 5 10 x
−1
−2 −2
−3
(a) (b)
Figure 23 Graphs of: (a) cosh and sinh; (b) arccosh and arcsinh
Separating the variables in equation (34), we obtain

Z Z
dy 1
p = dx.
y 2 − y02 y0
Then, referring to the table of standard integrals in the Handbook, we find
that

y x+k
arccosh = ,
y0 y0
where k is a constant.
43
Rearranging, we obtain

x+k
y = y0 cosh . (35)
y0
The constants y0 and k can be determined using the boundary conditions
y(a) = A and y(b) = B. We will see that, depending on the values of a, A,
b and B, suitable solutions of the minimal surface of revolution problem do
not always exist, and that when they do, it may be necessary to carry out
further work in order to determine the nature of the stationary point.
Solution of the minimal surface of revolution problem in a

special case
In order to understand the nature of the solution (35), let us consider a
simple physical example – a soap film suspended between two identical
circular hoops of radius 1, centred at (±L/2, 0), and each lying in a plane
perpendicular to the x-axis (see Figure 24). In this case, a = −L/2,
b = L/2 and A = B = 1.
(−L/2, 1) (L/2, 1)
Figure 24 Symmetric soap film
Let us use physical intuition to determine one of the constants. Because

the rings are the same size and placed symmetrically about the y-axis,
y(x) must be an even function, that is, y(x) = y(−x). However, because
cosh is also an even function, there is only one way that solution (35) can
be symmetric – namely, if k = 0 (the same result is obtained using algebra
in Exercise 20). Hence the solution for this symmetric case is

x
y = y0 cosh , y(−L/2) = 1, y(L/2) = 1.
y0
Setting x = 0 gives y(0) = y0 , so the constant y0 is simply the value of y at
the origin. Applying the boundary conditions gives y0 cosh(L/2y0 ) = 1, so
our solution becomes

x L
y = y0 cosh , where y0 cosh = 1. (36)
y0 2y0
The equation y0 cosh(L/2y0 ) = 1 cannot be rearranged to give a closed
formula for y0 ; instead, we must solve the equation numerically on a
computer. Figure 25 shows a plot of y0 against L. You are asked to
reproduce this graph in Exercise 22.
44
y0
1.0
0.848
0.5
0.235
0 0.5 1.0 1.5 L

1.3255
Figure 25 Graph of y0 as a function of L
From the graph, you can see that for L > 1.3255 (approximately) there is
no value of y0 that satisfies the equation y0 cosh(L/2y0 ) = 1, whereas for
L < 1.3255 there are two solutions of this equation, corresponding to the
red and blue parts of the curve. Let us examine each of these cases in turn.
Case 1: L < 1.3255

In this case there are two values of y0 that satisfy the equation
y0 cosh(L/2y0 ) = 1. For example, at L = 1 (when the radius of the hoops
equals their distance apart), Figure 25 shows that the solutions
are y0 ≃ 0.848 and y0 ≃ 0.235. In Figure 26, we plot the corresponding
solutions of equation (36), namely
x x
yc (x) ≃ 0.848 cosh and yd (x) ≃ 0.235 cosh
0.848 0.235
(note that yc (0) ≃ 0.848 and yd (0) ≃ 0.235).
y
1.0 0.848 yc
0.5
yd
0.235
−0.5 −0.25 0 0.25 0.5 x
Figure 26 Graph of yc and yd

45
We see that yd has a steep profile, so the shape of the soap film is like a
dumbbell, while yc has a shallow profile, so the soap film approximates a
cylinder. These results were derived with L = 1, but they are typical for
other values L < 1.3255; solutions with smaller values of y0 (shown in red
in Figure 25) correspond to dumbbell-like solutions, and those with larger
values of y0 (shown in blue in Figure 25) correspond to cylinder-like
solutions. But which solutions corresponds to real soap films; that is,
which solutions have minimal surface area?
The areas of the cylinder-like and dumbbell-like solutions, as functions of
the distance L between the hoops, are plotted in Figure 28. You are asked
to reproduce this graph in Exercise 22. From the graph, we see that the
cylinder-like solutions always have smaller area than the dumbbell-like
solutions, so they give the shapes of soap films. It turns out that the
dumbbell-like solutions are not even local minima (they are points of
inflection), but the proof of this is beyond the scope of this module.
Case 2: L > 1.3255

When the distance L between the hoops exceeds 1.3255, there is no
solution of the equation y0 cosh(L/2y0 ) = 1, so there is no smooth function
y(x) that gives a minimal surface of revolution. If we think of the hoops as
y very far apart, then there is an obvious non-smooth function y(x) that
gives rise to a surface of revolution of minimal area – the function whose
1 graph is shown in Figure 27. This function is called the Goldschmidt
discontinuous solution, or just the Goldschmidt solution, and it
−L/2 0 L/2 x corresponds to a flat soap film over each hoop. The area of this solution is
just the area of both hoops, namely 2π ≃ 6.2832. This area is independent
Figure 27 Graph of the of the distance between the hoops, and it is represented by the green
Goldschmidt solution horizontal line in Figure 28. In Exercise 22 we show numerically that this
line crosses the cylinder-like line at L ≃ 1.0554.
s
Dumbbell-like
6 Goldschmidt
solution
Cylinder-like
2
0 0.5 1 L
1.3255
1.0554
Figure 28 Graph of the surface areas of the dumbbell-like, cylinder-like
and Goldschmidt solutions as functions of the distance L between the
hoops
46
Notice that because the Goldschmidt solution is not a smooth function (it
is not differentiable at x = ±L/2), it cannot be obtained from the
Euler–Lagrange equation.
Physical interpretation of the minimal surfaces of revolution

Let us summarise what we have learned about soap films forming minimal
surfaces of revolution, with reference to Figure 28. Suppose that we dip
two circular wire hoops of unit radius into a soap solution, and gradually
pull the hoops apart, keeping them parallel. What does the preceding
analysis predict will happen?
When the distance between the hoops is L < 1.0554, the soap film will
assume the cylinder-like solution of equation (36), as shown in
Figure 29(a).
L L
(a) (b)
Figure 29 Soap film corresponding to: (a) the cylinder-like solution;

(b) the Goldschmidt solution
As we move the rings farther apart, and once L > 1.3255, there will no
longer be a minimal surface area that joins the two rings, and the soap film
will burst – either disappearing altogether, or assuming the Goldschmidt
solution shown in Figure 29(b), which is the minimal surface area solution.
For 1.0554 < L < 1.3255, the Goldschmidt solution exists and has the
smallest area. However, with these values of L, the cylinder-like solution
can exist, but it is only a local minimum, so small perturbations in the
shape of the cylinder give surfaces of larger area, but larger perturbations
in the shape of the cylinder can give rise to surfaces of smaller area.
We carried out the preceding analysis for circular hoops of radius 1. But
since we never specified any units in our problem, it should be clear that
this analysis carries through for hoops of any radius R, provided that we
scale all distances by the number R.
This relatively simple problem of finding the minimal area of a soap film
provides some idea of the possible complications that can arise in
variational problems.
47
Exercise 20
Apply the boundary conditions y(−L/2) = 1, y(L/2) = 1, to the equation

x+k
y = y0 cosh ,
y0
which is equation (35), to deduce that k = 0.
(Hint: Use the identities
cosh(x + y) = cosh x cosh y + sinh x sinh y
and
cosh(x − y) = cosh x cosh y − sinh x sinh y.)
Exercise 21
As you have seen, the area of the soap film with hoops of unit radius
separated by a distance L is given by
Z L/2 p
′ 2
x
S[y] = 2π y 1 + (y ) dx, where y(x) = y0 cosh .
−L/2 y0
Show that

L L
S[y] = πy02 sinh + . (37)
y0 y0
(Hint: Use the identities
cosh2 x − sinh2 x = 1 and cosh2 x = 12 (cosh 2x + 1).)
Exercise 22 Computing – optional

(a) Use the computer software for this module to plot L against y0 , where
y0 cosh(L/(2y0 )) = 1 (see Figure 25).
(b) Show numerically that this curve has a maximum Lmax ≃ 1.3255 at
y0 ≃ 0.5524.
(c) Read the section ‘Plotting stationary curves’ in the Computer
Exploration Worksheet. Use the functions yc0(L) and yd0(L) defined
there, and equation (37) of Exercise 21, to create a plot similar to
Figure 28.
(d) Show that the area of the Goldschmidt solution and the area of the
cylinder-like solution are equal when L ≃ 1.0554.
48
More about soap films

You have seen that the behaviour of soap films can be complicated,
even for simple surfaces such as those formed between two hoops. The
complexity of their behaviour is further compounded when one
realises that there can be minimal energy solutions of a quite
unexpected form, like that shown in Figure 30. We do not expect the
theory described in this subsection to find such a solution because the
mathematical formulation of the physical problem makes no allowance
for this type of behaviour.
Figure 30 A physically possible soap film, produced when a circular

film, perpendicular to the axis, is formed in the centre of a soap film
held by two circular hoops
Soap films can form a variety of complex shapes with other boundary
conditions. For example, consider a cubical frame of wire dipped into
a soap solution and then taken out. Films of local minimum energy –
that is, minimum area – will form on this frame. It transpires that
the possible shapes formed are many and varied, and they are often
counter-intuitive. Some of the possible shapes are shown in Figure 31.
Figure 31 Soap films on cubical frames of wire
These examples illustrate the fact that there may be more than one
minimal surface with the same boundary conditions. They also
demonstrate that some physical problems that are simple to state can
have bizarre solutions, which are difficult to describe mathematically.
49
We finish with a discussion of other types of functionals that are used in

mathematics and physics.
Further developments in the calculus of variations

In this unit we have considered functionals of the type
Z b
S[y] = F (x, y, y ′ ) dx, y(a) = A, y(b) = B, (38)
a
which involve one independent variable x, a single dependent variable
y(x), and its first derivative y ′ (x). There are many useful and
important extensions to this type of functional, some of which we
mention here.
• A simple generalisation of equation (38) is to integrands that
depend on several dependent variables yk (x), k = 1, 2, . . . , n, and
their first derivatives, so that these have the form
F (x, y1 , y1′ , y2 , y2′ , . . . , yn , yn′ ). Functionals of this type will be
studied in the next unit, on Lagrangian mechanics.
• Another generalisation involves integrands that are functions of
second or higher derivatives of y, for example F (x, y ′ , y ′′ ). Such
functionals are occasionally encountered in practical problems;
for example, in engineering they are used to describe the
mechanics of beams under stress. The differential equations that
describe the stationary paths of these functionals are higher than
second order. However, such functionals are not the norm, so we
do not consider them any further.
• An important generalisation involves integrands that depend on
two or more independent variables. For example, if there are two
independent variables x1 and x2 , then the dependent variable y is
a function of both of these, and the generalisation of
equation (38) has the form
Z Z
∂y ∂y
J[y] = F x1 , x2 , y, , dx1 dx2 .
D ∂x1 ∂x2
In this case the integral defining the functional is over a region D
in the (x1 , x2 )-plane, rather than over a line as in (38). Now the
Euler–Lagrange equation becomes a partial differential equation.
Such functionals and their generalisations belong to a branch of
mathematical physics called classical field theory, which
describes many important areas of physics such as the wave
equation, electromagnetism and general relativity.
• One can also consider broken extremals, which are continuous
solutions of the Euler–Lagrange equation with first derivatives
that are discontinuous at a finite number of points. That such
solutions are important is clear by observing collections of soap
bubbles that form a composite shape with sharp corners. A
simple example of such a solution is the Goldschmidt solution.
50
Solutions to exercises
Solution to Exercise 1
(a) On this straight line y = 0, so the value of the functional is
Z 1
√ 1
S[y] = 1 + 0 dx = x −1 = 1 − (−1) = 2.
−1
√
(b) As y = 1 − x2 , we have
dy x
= −√ ,
dx 1 − x2
so the value of the functional is
Z 1r Z 1
x2 1
S[y] = 1+ 2
dx = √ dx.
−1 1 − x −1 1 − x2
Using the standard integral given in the hint, we obtain
1 π π
S[y] = arcsin x −1 = − − = π.
2 2
(a) If y = x, then y ′ = 1 and the functional is
Z 1
S[y] = dx = 1.
0
(b) If y = x , then y ′ = 3x2 and the functional is

3
Z 1
S[y] = 9 x4 dx = 95 .
0
(a) In this case the endpoints (a, A) and (b, B) of the paths satisfy
a = 1, A = 0 and b = 2, B = 1, so
Z 2
S[y + εg] = x((y + εg)′ )2 dx
1
Z 2
= x(y ′ + εg ′ )2 dx.
1
To find the stationary function, we use equation (4), giving
Z 2
d d
S[y + εg] = x(y + εg ) dx
′ ′ 2
.
dε ε=0 dε 1 ε=0
The integration limits 1 and 2 are independent of ε, so the order of
integration and differentiation can be interchanged using Leibniz’s
integral rule. This gives
51
Z 2
d d

S[y + εg] = ′ ′ 2
x(y + εg ) dx
dε ε=0 1 dε ε=0
Z 2
=2 (xg ′ (y ′ + εg ′ ))|ε=0 dx
1
Z 2
=2 xy ′ g ′ dx.
1
If S[y] is stationary, then it follows, by definition, that
Z 2
x y ′ (x) g ′ (x) dx = 0
1
for all functions g(x) for which g(1) = g(2) = 0.
To solve this equation, we integrate by parts:
Z 2 Z 2
′
2
u(x) g (x) dx = u(x) g(x) 1 − u′ (x) g(x) dx,
1 1
′
with u = xy . However, the first term on the right-hand side vanishes
because g(1) = g(2) = 0, so we get
Z 2
′
(xy ′ ) g(x) dx = 0.
1
′
This integral can vanish only if (xy ′ ) = 0, which gives
x y ′ (x) = α,
where α is a constant.
(b) The equation xy ′ = α is separable and is equivalent to
dy α
= .
dx x
Integrating both sides with respect to x gives y = α ln x + β, where β
is a constant. But y(1) = 0, so β = 0, and y(2) = 1, so α = 1/ ln 2.
Therefore the required solution is y = ln x/ ln 2.
(a) The partial derivatives are
∂F p ∂F xy ∂F xy ′
= y 2 + (y ′ )2 , =p , = p .
∂x ∂y y + (y ′ )2
2 ∂y ′ y 2 + (y ′ )2
The total derivative is therefore
dF ∂F ∂F ′ ∂F ′′
= + y + ′y
dx ∂x ∂y ∂y
p xyy ′ xy ′ y ′′
= y 2 + (y ′ )2 + p +p
y 2 + (y ′ )2 y 2 + (y ′ )2
y 2 + (y ′ )2 + xyy ′ + xy ′ y ′′
= p .
y 2 + (y ′ )2
52
(b) (i) If y = sin x, then y ′ = cos x and y ′′ = − sin x.

Substituting
p these values into the formula for F gives
2
F = x sin x + cos2 x = x (where we have used the identity
sin2 x + cos2 x = 1). Hence dF/dx = 1.
(ii) Using the result from part (a), we obtain
dF sin2 x + cos2 x + x sin x cos x − x cos x sin x
= p
dx sin2 x + cos2 x
1
= √ = 1,
1
in agreement with part (b)(i).
The partial derivatives are
∂F x ∂F (y ′ )2 ∂F yy ′
=p , = p , = p .
∂x x + y(y ′ )2
2 ∂y 2 x2 + y(y ′ )2 ∂y ′ x2 + y(y ′ )2
The total derivative is
dF ∂F ∂F ′ ∂F ′′
= + y + ′y
dx ∂x ∂y ∂y
x (y ′ )3 yy ′ y ′′
=p + p +p .
x2 + y(y ′ )2 2 x2 + y(y ′ )2 x2 + y(y ′ )2
The partial derivatives are
∂F ∂F
= 2xy 3 and = 3x2 y 2 ,
∂x ∂y
and the total derivative is
dF ∂F ∂F ′
= + y = 2xy 3 + 3x2 y 2 y ′ .
dx ∂x ∂y

d ∂F ∂F
Now, is the total derivative of , which itself is a function of
dx ∂y ∂y
x and y. Hence

d ∂F ∂ ∂
= (3x2 y 2 ) + y ′ (3x2 y 2 ) = 6xy 2 + 6x2 yy ′ .
dx ∂y ∂x ∂y
Also, using the expression above for dF/dx, we obtain

∂ dF
= 6xy 2 + 6x2 yy ′ .
∂y dx
53
(a) With the notation of equation (18), the integrand is given by
F (x, y, y ′ ) = (y ′ )2 + y,
with a = 0, b = 1 and A = 0, B = 2. As
∂F ∂F
= 2y ′ and = 1,
∂y ′ ∂y
the Euler–Lagrange equation (19) is
2y ′′ − 1 = 0, y(0) = 0, y(1) = 2.
(b) The differential equation can be written as
y ′′ = 12 ,
which can be solved by direct integration. Integrating both sides with
respect to x gives y ′ = 21 x + c, and integrating again gives the general
solution
y = 14 x2 + cx + d,
where c and d are constants. As y(0) = 0, we see that d = 0. As
y(1) = 2, we see that 41 + c = 2, so c = 74 . Therefore the stationary
path is
y = 14 x2 + 47 x.
(a) In this case F (x, y, y ′ ) = (y ′ )2 − y 2 , so
∂F ∂F
= 2y ′ and = −2y.
∂y ′ ∂y
This gives the Euler–Lagrange equation 2y ′′ − (−2y) = 0, or
equivalently,
y ′′ + y = 0.
(b) The equation y ′′ + y = 0 models a simple harmonic oscillator (see
Unit 2), and has general solution
y = A cos x + B sin x,
where A and B are constants determined by the boundary conditions.
As y(0) = 0, we see that A = 0. As y(X) = 1, we see that B sin X = 1,
so B = 1/ sin X. Therefore the stationary function is
sin x
y= .
sin X
54
(a) In this case F = (y ′ )2 + y 2 + 2xy, so
∂F ∂F
= 2y ′ and = 2y + 2x.
∂y ′ ∂y
This gives the Euler–Lagrange equation 2y ′′ − (2y + 2x) = 0, or
equivalently,
y ′′ − y = x.
(b) The equation y ′′ − y = x is a second-order inhomogeneous equation, of
the type that you studied in Unit 1. The solution is given by
y = yc + yp , where yc is the complementary function of the differential
equation, and yp is a particular integral.
The auxiliary equation is λ2 − 1 = 0, with solutions λ = ±1, so the
general solution of the homogeneous equation y ′′ − y = 0 is
yc = Cex + De−x , where C and D are constants.
The inhomogeneous part is the expression x, so to find a particular
integral we try yp = Ax + B. Substituting this into the differential
equation y ′′ − y = x gives −Ax − B = x. Comparing coefficients, we
see that A = −1 and B = 0, so yp = −x.
Therefore the general solution of the equation y ′′ − y = x is
y = Cex + De−x − x.
To find C and D, we use the boundary conditions. As y(0) = 0, we see
that C + D = 0, so y = C(ex − e−x ) − x. As y(1) = α, we see that
α = C(e − e−1 ) − 1. Thus

α+1
y= (ex − e−x ) − x.
e − e−1
If you are familiar with hyperbolic functions (a brief summary of their
properties is given in Subsection 5.3), then you can simplify this
expression for y. Since sinh x = 12 (ex − e−x ), we can write
sinh x
y = (α + 1) − x.
sinh 1
(a) The integrand F (y, y ′ ) = ln y ′ + y does not depend on x, so it is valid
to find the first integral of the Euler–Lagrange equation. As
∂F/∂y ′ = 1/y ′ , we can use equation (25) to obtain
y′
− (ln y ′ + y) = c,
y′
for some constant c. Rearranging this equation, we get
ln y ′ = 1 − c − y.
55
By taking the exponential of both sides, we obtain

dy
= e1−c−y = e1−c e−y = αe−y ,
dx
where α = e1−c is another constant.
(b) The differential equation dy/dx = αe−y is separable. Separating the
variables gives
Z Z
y
e dy = α dx,
hence
ey = αx + β.
Taking the logarithm of both sides gives
y = ln(αx + β).
The boundary conditions tell us that
ln(α + β) = ln 2 and ln(2α + β) = ln 3;
that is, α + β = 2 and 2α + β = 3. Therefore α = β = 1. It follows
that the stationary function is
y = ln(x + 1).
p
(a) The integrand F (y, y ′ ) = y(1 + (y ′ )2 ) does not depend explicitly
on x, so it is valid to find the first integral of the Euler–Lagrange
equation. We obtain
√ ′ 2 p
∂F y(y )
y′ ′ − F = p − y(1 + (y ′ )2 ) = k,
∂y 1 + (y ′ )2
for some constant k. Simplifying this gives
r
y
= −k.
1 + (y ′ )2
Now square both sides to obtain
y
(y ′ )2 = 2 − 1,
k
so
r
′ y 1p
y =± − 1 = y − k2,
k2 k
where in the final expression we have absorbed the factor of ±1 into
the constant k, which could be positive or negative (but not zero).
56
(b) The equation for y ′ is separable. Separating variables, we obtain

Z Z
dy
k p = dx.
y − k2
p
Integrating both sides gives 2k y − k 2 = x + α, for some constant α,
so
(x + α)2
y = k2 + .
4k 2
The boundary conditions at x = ±1 give
(α + 1)2 (α − 1)2
A = k2 + and A = k 2
+ .
4k 2 4k 2
By inspection (or by solving for α), these equations can be true only
if α = 0, so
x2 1
y = k 2 + 2 and A = k 2 + 2 .
4k 4k
By rearranging the latter equation, we obtain h2 − Ah + 41 = 0, where
h = k 2 ,√which isa quadratic equation in h that has solutions
1
2 A± A2 − 1 . Therefore the stationary paths are given by
x2 p
y(x) = h + , where h = 12 A ± A2 − 1 .
4h
If A > 1, then there are two stationary paths, but if A = 1, then there
is only one stationary path.
(a) Evaluating the derivative on the left-hand side, we obtain

d ′ ∂F ′′ ∂F ′ d ∂F dF
y −F =y +y − .
dx ∂y ′ ∂y ′ dx ∂y ′ dx
Now,
dF ∂F ∂F ′ ∂F ′′ ∂F ′ ∂F ′′
= + y + ′y = y + ′y ,
dx ∂x ∂y ∂y ∂y ∂y
because ∂F/∂x = 0. Combining these two equations, we find that

d ∂F d ∂F ∂F
y′ ′ − F = y′ − ,
dx ∂y dx ∂y ′ ∂y
as required.
(b) The right-hand side of the preceding equation is zero if either (i) y(x)
satisfies the Euler–Lagrange equation

d ∂F ∂F
− = 0,
dx ∂y ′ ∂y
or (ii) y(x) is constant (so that y ′ (x) = 0). In both cases, we can
integrate to obtain
∂F
y′ − F = c,
∂y ′
for some constant c. This is the first-order differential equation of
equation (25).
57
The derivative of f (x) is
x d−x
f ′ (x) = p −p .
x2 + h21 (d − x)2 + h22
However, from Figure 12 we have
x d−x
sin θ1 = p and sin θ2 = p .
x2 + h21 (d − x)2 + h22
It follows that
f ′ (x) = sin θ1 − sin θ2 .
Therefore the function f (x) is stationary when sin θ1 = sin θ2 , that is,
θ1 = θ2 . Furthermore, as
h21 h22
f ′′ (x) = + > 0,
(x2 + h1 )3/2 ((d − x)2 + h22 )3/2
2
the stationary point is a minimum.
(a) From Fermat’s principle, the light ray will be a stationary path of the
functional
Z
2 b 1p
T [y] = 1 + (y ′ )2 dx, y(a) = A, y(b) = B.
c a y
As the integrand
2p
F = 1 + (y ′ )2
cy
does not depend on the independent variable x, we can use the first
integral of the Euler–Lagrange equation (see equation (25)) to obtain
(y ′ )2 1p
p − 1 + (y ′ )2 = k,
y 1 + (y ′ )2 y
where k is a constant into which wep have absorbed the common factor
2/c.
p Multiplying throughout by y 1 + (y ′ )2 and simplifying gives
2
ky 1 + (y ) = −1. Then square both sides and solve for y ′ to obtain
′
r
dy 1
=± −1
dx k y2
2
1p
=± 1 − k2y2 .
ky
Because k is an arbitrary constant, which could be positive or negative
(but not zero), we can remove the factor ±1 to obtain
dy 1p
= 1 − k 2 y 2 , where 12 ≤ y ≤ 2.
dx ky
58
(b) The preceding differential equation is separable. Separating variables

gives
Z Z
ky
p dy = dx.
1 − k2y2
As
dp k2y
1 − k2y2 = − p ,
dy 1 − k2y2
we can integrate both sides of the equation above to give
1p
− 1 − k 2 y 2 = x + α,
k
where α is a constant. Squaring both sides gives 1/k 2 − y 2 = (x + α)2 ,
thus
r
1
y= − (x + α)2 ,
k2
where we take the positive sign because y ≥ 21 .
To find k and α, we apply the boundary conditions y(−1) = y(1) = 1.
We obtain
r r
1 2
1
2
− (α − 1) = 1 and − (α + 1)2 = 1.
k k2
By inspection, these equations can both be true only if α = 0 (you can
also show this
p by squaring both sides of these equations and solving
for α). So 2
√ 1/k − 1 = 1. Solving this equation for k gives
k = ±1/ 2, and hence the path of the light ray (as shown in the
figure) is
p
y = 2 − x2 , where −1 ≤ x ≤ 1.
y
2
x
−1 1
As y = x/b, we have y ′ = 1/b, so
s s r
1 + (y ′ )2 1 + 1/b2 b2 + 1 1
= = √ .
y x/b b x
59
Therefore equation (29) tells us that the time T satisfies

s Z s s
b2 + 1 b 1 b2 + 1 √ b b2 + 1
T = √ dx = 2 x 0=2 .
2gb 0 x 2gb 2g
(a) Equation (31) says that
2φ − sin 2φ 1 − cos 2φ
x= and y = .
Therefore the gradient of the curve y = y(x) is

dy dy dx 2 sin 2φ 2 sin φ cos φ 1
= = = 2 = .
dx dφ dφ 2(1 − cos 2φ) 2 sin φ tan φ
(b) At the origin, we have x = y = 0, so φ = 0. Therefore tan φ = 0, which
implies that dy/dx is infinite, so the brachistochrone hangs vertically
downwards at (0, 0).
(c) The brachistochrone is parallel to the x-axis at points where the
gradient is 0, that is, when tan φ is infinite. This occurs only when
φ = π/2, in our range. At this value of φ, we have
π 2
x= and y = .
The brachistochrone terminates at (x, y) = (b, 1), so the critical
brachistochrone must satisfy
π 2
= b and = 1.
Dividing one equation by the other gives b = π/2, as required.
(a) In this example
p the integrand of the functional has the form
F (x, y ) = 1 + x + (y ′ )2 , and
′
∂F ∂F y′
=0 and = p .
∂y ∂y ′ 1 + x + (y ′ )2
Hence the Euler–Lagrange equation (19) becomes
!
d y′
p = 0.
dx 1 + x + (y ′ )2
Integrating both sides with respect to x gives
p
y ′ = c 1 + x + (y ′ )2 ,
for some constant c, as required.
60
(b) Squaring and rearranging the preceding equation gives

2
dy c2
= k 2 (1 + x), where k 2 = .
dx 1 − c2
Taking square roots, we obtain y ′ = ±k(1 + x), but we may as well
write this as
y ′ = k(1 + x),
because k is a constant that could be positive or negative. By
separating variables in this equation, we obtain
Z Z
√
dy = k 1 + x dx,
so
y = α(1 + x)3/2 + β,
where α = 32 k and β are constants. Applying the boundary conditions
gives
α+β=0 and 23/2 α + β = 1,
which have solution α = −β = 1/(23/2 − 1). Hence
(1 + x)3/2 − 1
y= .
23/2 − 1
(a) The general equation of a non-vertical straight line is y = mx + c,
where m and c are constants. The line passes through (0, 0), giving
c = 0, and through (b, B), giving m = B/b. Therefore
B
y= x.
b
(b) Substituting the preceding equation into equation (33) gives
Z b r
B B2
S[y] = 2π x 1 + 2 dx
0 b b
B p b
= π 2 b2 + B 2 x2 0
bp
= πB b2 + B 2 .
61
The boundary conditions give

2k + L 2k − L
y0 cosh = 1 and y0 cosh = 1.
2y0 2y0
Because cosh is an even function, it should be fairly clear that these
equations can be satisfied only if k = 0. However, let us establish this fact
algebraically.
Using the identity given in the hint with x = k/y0 and y = L/(2y0 ), we
obtain
y0 cosh(k/y0) cosh(L/(2y0)) + y0 sinh(k/y0) sinh(L/(2y0)) = 1,
y0 cosh(k/y0) cosh(L/(2y0)) − y0 sinh(k/y0) sinh(L/(2y0)) = 1.
Subtracting the second equation from the first gives
2y0 sinh(k/y0) sinh(L/2y0) = 0, which, because L > 0, can be true only if
y0 sinh(k/y0) = 0. As y0 is not zero, we see that sinh(k/y0) = 0, so k = 0.
Observe that y ′ (x) = sinh(x/y0 ), so
Z L/2 s
x 2 x
S[y] = 2πy0 cosh 1 + sinh dx
−L/2 y0 y0
Z L/2
x
= 2πy0 cosh2 dx,
−L/2 y 0
where we have used the first identity from the hint. Now use the second
identity from the hint to obtain
Z L/2
2x
S[y] = πy0 cosh + 1 dx
−L/2 y0
L/2
y0 2x
= πy0 sinh +x
2 y0 −L/2

L L
= πy02 sinh + ,
y0 y0
where, to determine the last line, we have used the fact that sinh is an odd
function.
62
Acknowledgements
Grateful acknowledgement is made to the following sources:
Figure 9: c Soapbubble.dk. This file is licensed under the Creative
Commons Attribution-Noncommercial Licence
http://creativecommons.org/licenses/by-nc/3.0.
Figure 13: Brocken Inaglory. This file is licensed under the Creative
Commons Attribution-Share Alike Licence
http://creativecommons.org/licenses/by-sa/3.0.
Every effort has been made to contact copyright holders. If any have been
inadvertently overlooked, the publishers will be pleased to make the
necessary arrangements at the first opportunity.
63
Index
Index
action, 32 Leibniz, Gottfried Wilhelm, 17
angle of incidence, 28 Leibniz’s integral rule, 11
angle of reflection, 28 light ray, 28
Bernoulli, Johann, 16 mirage, 30

brachistochrone, 16, 34
broken extremal, 50 Newton, Isaac, 17
non-conservative system, 5
catenary, 18 non-deterministic system, 5
classical field theory, 50
conservative system, 5 optical action, 32
cosh, 43 optical Lagrangian, 32
cycloid, 16, 38
partial derivative, 19
Descartes, René, 39 Pascal, Blaise, 39
deterministic system, 5
distance along a curve, 8, 9 random system, 5
ray, 28
Essex, James, 39 refractive index, 29
Euler–Lagrange equations, 24
alternative forms, 26 sinh, 43
first integral, 26 Snellius, Willebrord, 31
Snell’s law, 31
Fermat, Pierre de, 29, 39 soap film, 17, 40, 49
Fermat’s principle, 28, 30 speed of light in different media, 29
first integral of the Euler–Lagrange equation, 26 stationary curve, 11
functional, 5, 9 stationary function, 11
stationary functional, 11
Galilei, Galileo, 16, 39 stationary path, 11
Goldschmidt solution, 46 classification of, 13
equation for, 11
hanging chain, 18 stochastic system, 5
Huygens, Christiaan, 39 surface of revolution, 17, 40
hyperbolic functions, 42
total derivative, 19
Lagrangian, 32
Lagrangian mechanics, 5 Wren, Christopher, 39
64

Introduction To The Calculus of Variations - ms327 PDF

Uploaded by

Copyright:

Available Formats

Introduction To The Calculus of Variations - ms327 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To The Calculus of Variations - ms327 PDF

Uploaded by

Copyright:

Available Formats

MS327

Deterministic and stochastic dynamics

The Open University, Walton Hall, Milton Keynes, MK7 6AA.

ISBN 978 1 8487 3962 8

Functionals share many of the same properties as functions. In particular,

1 Shortest distance between two

We begin in Subsection 1.1 by showing that the distance between two

1.1 Distance between two points on a given curve

Figure 1 Graph of the curve y = y(x) passing through points Pa and Pb

The curve is now approximated by a sequence of N straight-line segments.

Figure 3 Segment of the curve y = y(x) between xk and xk+1

The length of each segment can be determined using Pythagoras’ theorem:

A functional S[f ] is a map from functions f to the real numbers.

Distance along a curve

1.2 Stationary paths

We want to define something analogous to this for functionals. To do this,

Stationary path of a functional

Leibniz’s integral rule

Applying this rule gives

for all functions g(x) for which g(a) = g(b) = 0.

To solve this equation, we integrate by parts:

Stationary path of the curve-length functional

Classification of stationary paths

However, the first term on the right-hand side vanishes because

2 Some examples of functionals

Figure 7 A cycloid: the path of a point P on the rim of a wheel

History of the brachistochrone

In December 1696, Bernoulli extended the time limit to Easter 1697,

2.2 Minimal surfaces of revolution

Figure 8 Cylindrical-type shape produced when a curve y(x) is rotated

The area of this surface is shown in Subsection 5.3 to be

2.3 A problem in navigation

(0, 0) x 2.4 Catenary

which you encountered earlier, in equation (10).

It turns out that the shape of the catenary is given by y = c cosh(x/c) + d,

3 Reminder about partial and total

4.1 Derivation of the Euler–Lagrange equation

For fixed y and g, different values of ε give different values of

Deriving the differential equation

Equation (17) is known as the Euler–Lagrange equation, and is hugely

Let us apply the Euler–Lagrange equation in an example and some

4.2 Alternative forms of the Euler–Lagrange

After integrating with respect to x, we obtain

Special case of the Euler–Lagrange equation

First integral of the Euler–Lagrange equation

5 Applications of the Euler–Lagrange

5.1 Fermat’s principle

This observation about the two angles has an interesting consequence.

Consider now a general (possibly non-uniform) medium in the Cartesian

Figure 13 Sunset mirage

Snell’s law of refraction

Figure 14 A ray of light is refracted as it passes from water to air

In water the speed of light is approximately c2 = c1 /1.3, where c1 is

Figure 15 Path taken by a lifeguard to reach a distressed swimmer

Connection to Lagrangian mechanics

described by y+ and y− are shown in Figure 16. The observer

Figure 16 Graph of solutions

5.2 Revisiting the brachistochrone