Cycloids, the Brachistochrone Problem and

Calculus of Variations

Maura Paterson
[email protected]

In this lecture we will be looking at problems that have a solution in the form of a curve
with speci c properties. Such questions arise naturally in mathematics, physics and
other applications, and over history their study has lead to the development of new
mathematical techniques. We will begin by considering some properties of a curve that
has been referred to as the \Helen of geometry" because of its attractiveness to mathe-
maticians over the centuries.

1 The Cycloid

Suppose a circle with radius r rolls along a straight line. If we mark a point on the cir-
cumference of the circle and trace out the path followed by the point as the circle rolls,
then the resulting curve is known as a cycloid. The name of this curve is due to Galileo,
who studied it in 1599.

The cycloid was an object of research for many famous mathematicians during the 17th
century. Pascal supposedly began thinking about it in order to distract himself from a
painful toothache. Miraculously the pain stopped; regarding this as a sign from God, he
spent the rest of the week studying the properties of the cycloid in detail.
Having seen a de nition for the cycloid, it would be useful to have an equation to de-
scribe it, to help us in illustrating some of its properties. Let P be the point on the cir-
cumference of the circle of radius a, and suppose it starts at the base of the circle. We

1
will choose our coordinate axes so that this initial starting point is at the origin. Now,
suppose the circle rolls along the x -axis. As it rolls, the point P rotates about the cen-
tre of the circle. We suppose that at some time the angle through which it has rotated
is t radians.

We would like to nd the coordinates of the point P as a function of t . First we con-

sider the x -coordinate. Since P started out at the base of the circle, the current dis-
tance of the base of the circle from the origin is equal to the length of the segment of
the circle that joins P to the current base. As the circle has radius a and this segment
subtends the angle t , it follows that this length is equal to at , hence the x -coordinate
of the base (and therefore also of the centre) of the circle is at . Now, the di erence in
x -coordinate between the point P and the centre of the circle is given by a sin t . Hence
we deduce that
Px = at a sin t:

To nd the y -coordinate of P , we observe that the centre of the circle has y -coordinate
a, and the di erence between the y -coordinate of the centre and that of P is a cos t .
Thus
Py =a a cos t:

In other words, the cycloid consists of the set of points

f(a(t sin t ); a(1 cos t ))jt 2 Rg:
(This is a parametric description of the cycloid.) Later in this talk, we will sometimes
wish to restrict our attention to the rst arc of the cycloid, which is given by
f(a(t sin t ); a(1 cos t ))j0  t  2g:
The cycloid has many nice geometric properties; we now consider a couple of these.

2
1.1 Area under one Arc of a Cycloid

Galileo attempted to nd an approximate value of the area under an arc of a cycloid by

weighing pieces of metal that were cut in the shape of a cycloid. He found that it was
around 3 times the area of the corresponding circle, and he guessed that the true value
was  times this area. Torricelli and Roberval independently found that the correct an-
swer was in fact exactly 3 times the area of the circle; Roberval subsequently accused
Torricelli of having stolen his proof, and he also fell out with Descartes, who claimed
that the problem was \a pretty one which I had not noticed before but which would
cause no diculty to any moderately skilled geometer." The use of modern calculus
techniques makes it easy to solve this problem:
Example 1.1. To nd the area under an arc of the cycloid we simply have to calculate
Z 2a
y dx ;
0

where y is given by
y = a(1 cos(t ))
and
x = a (t sin(t ));
so
dx
= a(1 cos(t )):
dt
Hence
Z 2a Z 2
y dx = a(1 cos(t ))a(1 cos(t ))dt
0 0
Z 2
=a 2
1 2 cos(t ) + cos2 t dt
0
2
1 1
Z
=a 2
1 2 cos(t ) + + cos(2t )dt
0 2 2
2 3 1
 2
=a t 2 sin(t ) + sin(2t )
2 4 0

= 3a :
2

1.2 Arclength of the Cycloid

In 1658, the architect Sir Christopher Wren was the rst person to compute the length
of an arc of the cycloid. We can perform this computation easily with the aid of calcu-
lus (although of course such methods were not available to Wren.)

3
Example 1.2 (Arclength of the Cycloid). A single arc of a cycloid can be described by
the function
f : [0; 2] ! R2 ;
t 7! (a(t sin(t )); a(1 cos(t ))):
As the parameter t ranges from 0 to 2, so the point f (t ) = (a(t sin(t )); a(1
cos(t ))) traces out a single arc of the cycloid. We can interpret the parameter t to be
representing time (so it takes time t to reach the point (a(t sin(t )); a(1 cos(t )))
when tracing out the cycloid). Suppose `(t ) represents the length covered as the cy-
cloid is traced from (0; 0) to the point f (t ). Then ddt` is the speed we are travelling
along the arc, and we can nd the total length `(2) of the arc of the cycloid by in-
tegrating the speed ddt` from 0 to 2.
At a given time t , the component of the velocity in the x -direction is given by ddxt where
x = a (t sin(t )), and the component of the velocity in the y direction is given by ddyt
where y = a(1 cos(t )). Given these values we can use Pythagoras' theorem to nd
the total speed ddt` :

d`
dt dy
dt

dx
dt

Now, the component of the velocity in the x -direction at time t is

dx d
= a(t sin(t ))
dt dt
= a(1 cos(t ))
and the component of the velocity in the y -direction at time t is
dy d
= a(1 cos(t ))
dt dt
= a sin(t ):
Hence the speed at time t is given by
s
dx dy
2  2 q
+ = a2 2a2 cos(t ) + a2 cos(t )2 + a2 sin2 (t )
dt dt
= a 2 2 cos(t )
p
q
= a 2 2(1 2 sin2 (t=2))
= 2a sin(t=2):

4
Thus we conclude that the length of a single arc of the cycloid is
Z 2
2a sin(t=2)dt = 2a [ 2 cos(t=2)]20
0

= 4a + 4a
= 8a:
In other words, the arclength of the cycloid is simply 8 times the radius of the circle
that generated the cycloid -a surprisingly simple result.

2 Fermat's Principle of Least Time

We have just considered the case where we start with a curve and nd some of its prop-
erties. However, there are many scenarios in which we are given some properties that
we want a curve to have, and we need to nd a curve that has the desired properties.
One such scenario is the following:
Example 2.1. Suppose Fred the lifeguard is at the beach, standing d1 metres from the
water's edge, and he sees a swimmer who is in trouble, at a distance of d2 meters from
the shoreline, and w meters further up the beach:

Fred
d1
shoreline

d2

swimmer
w

If Fred can run at a speed of Vrun on the land, and swim at a speed of Vswim < Vrun
through the water, what path should he take in order to reach the swimmer in as short
a time as possible?
If Fred's swimming speed were as fast as his running speed, then the answer would be a
straight line from Fred to the swimmer. However, since Fred's swimming is slower than
his running, it might be better overall to choose a path with a bit further to run if it
means that Fred does not have to swim as far.
In order to help us nd the exact answer, we rst observe that to reach the swimmer, it
will be necessary for Fred to enter the water at some point P . Suppose that the point
we choose is l meters along the shoreline from your current position. The fastest route
that passes through point P will be to run in a straight line to P , then to swim in a
straight line from P to the swimmer. Thus to nd the best path overall, we simply have
to determine what is the best point P to choose.

5
The distance from Fred to P is given byp l 2 + d12 , by Pythagoras' theorem. Similarly,
p

the distance from P to the swimmer is d22 + (w l )2 .

Fred
l 2 + d12
p
d1
i
P
shoreline
r
(w l )2 + d22
p
d2

swimmer
l w l

The total time taken to reach the swimmer when following this path is
l 2 + d12 (w l )2 + d22
p p
t = + :
Vrun Vswim

In order to nd the minimum time possible, we take the derivative with respect to l :
dt l w l
= ;
dl l2 + (w l )2 + d22
p p
Vrun d12 Vswim

which is equal to 0 when

l w l
= : (1)
+ (w l )2 + d22
p p
Vrun l 2
d12 Vswim

To simplify (1), we note that for the angles marked i and r on the diagram we have
l
sin(i ) = p 2
l + d12
and
w l
sin(r ) = p :
(w l )2 + d22

Hence (1) becomes

sin(i ) sin(r )
=
Vrun Vswim

or
sin(i ) Vrun
= : (2)
sin(r ) Vswim

6
Hence the path we require is precisely that for which the sines of the angle i at which
Fred approaches the shoreline and the angle r at which Fred swims away from the shore-
line are in the ratio de ned by the ratio of Fred's running speed and swimming speed
respectively.
For those who have studied any optics, (2) may look familiar. It is in fact Snell's law
that describes how light is refracted when it passes from one medium to another (this
phenomenon explains why the bottom of a swimming pool appears closer than it ac-
tually is, for example). The connection with the above example comes from Fermat's
principle: stated informally, this says that light follows a path that allows it to reach its
destination in the least possible amount of time. This is a principle with broad applica-
tions for solving problems in optics: in the above example, if we replace Fred by a beam
of light, and Fred's running and swimming speeds with the speed of light in the two dif-
ferent media, then we obtain Snell's law.

3 The Brachistochrone Problem

We now turn our attention to a famous problem that was posed in 1696 by Johann
Bernoulli1 :
Suppose you have two points A and B (with A higher than B ,) and you
wish to connect them with a curved wire so that a bead can slide down the
wire from point A to point B . What curve should you choose for the shape
of the wire to ensure that the bead travels from A to B in the smallest pos-
sible amount of time?
A

??

This is known as the brachistochrone problem, from the Greek words for shortest time.
Several mathematicians found the solution, including Johann Bernoulli, Jakob Bernoulli
(Johann's brother), Leibniz and Newton. (Note that we suppose the initial speed of the
bead is 0, and we ignore the e ects of friction.)
Johann Bernoulli devised an ingenious way of nding the solution, which we will explore
here.
1 Infact the problem had previously been considered by Gallileo, who thought (erroneously) that the
solution was a circle.

7
Example 3.1. Firstly, we will choose our coordinates so that A is at the origin. If af-
ter time t the bead has travelled along the curve to a point where its y -coordinate is
y (t ), then the change in gravitational potential energy from when it was at A is given
by mgy , where m is the mass of the bead, and g is the rate of acceleration due to
gravity near the Earth's surface (approximately 9:8m=s2 ).
Denote by `(t ) the total distance along the wire that the bead has travelled after time
t . Then its speed at time t is ddt` , and its kinetic energy at time t is given by 12 m ddt` .

2

Because the initial speed of the marble is 0, the law of conservation of energy tells us
that its kinetic energy at time t is equal to the change in potential energy from time 0,
hence we have
1 d`
 2
mgy = m (3)
2 dt
so
d` p
= 2gy : (4)
dt
(This is the only physics we will have to know in order to solve this problem!)
Bernoulli's trick for solving the brachistochrone problem was to suppose that the curve
is made up of a large number of very small straight-line segments, and to suppose that
the speed of the bead is e ectively constant on each of these segments.
We now consider two adjacent line segments on the curve:

Since we are seeking a curve over which the bead takes the shortest possible time, it
must be the case that the time taken for the bead cross these two segments is also as
small as possible. However, this is precisely the scenario to which Snell's law applies:
if the speed of the bead on the rst segment is Va and the speed of the bead on the
second segment is Vb , then for the time taken to traverse both segments to be as small
as possible we must have
sin(a ) sin(b )
= ;
Va Vb

where a and b are the angles the segments make with the vertical:

8
 a

b

If we suppose that we let the lengths of the line segments become in nitesimally small,
then we can conclude that as the bead travels down the brachistochrone curve then for
each time t we have
sin(t )
d`
=C (5)
dt

where C is some constant, and t is the angle that the tangent to the curve at the
point the bead has reached makes with the vertical. Combining equations (5) and (4)
we have
sin(t )
=C
2g ( y )
p

sin2 (t ) = C 2 2g ( y ): (6)

Since the slope of the curve at point (x ; y ) is given by dy
dx
, we can deduce that
1
sin(t ) = q
2 : (7)
1+ dy
dx

Finally, by combining equations (6) and (7), we deduce that

1

2 = C 2 2g ( y)
1+ dy
dx

and hence
dy
 2 !
( y) 1+ = 2a (8)
dx
for the constant a = (4C 2 g ) 1 .
The equation (8) is a di erential equation that we can solve to nd the brachistochrone
curve. It is easy to check that this equation is in fact satis ed by an upside-down cy-
cloid generated by a circle of radius a rolling along the underside of the y -axis. (The
parametric equations of this curve are x = a(t sin(t )) and y = a(1 cos(t )).)

9
Therefore, given points A and B , it suces to nd the circle of appropriate diameter
such that the upside-down cycloid it traces when starting at A also passes through B ,
and this cycloid is the answer to the brachistochrone problem for these points. It is in-
teresting to note that sometimes this results in a curve in which the bead has to travel
upwards for some time before reaching B . Nevertheless, this is still the fastest route
that the bead can take.

(This holds even in the extreme case where B has the same height as A: in the absence
of friction, a bead that takes a cycloidal path from A will indeed reach B , and in the
shortest possible time. We note that in this particular case, the path consisting of a
straight line between A and B would result in the bead never leaving A.)

10
4 Calculus of Variations

Both the problem of nding the quickest route to reach the struggling swimmer, and
the brachistochrone problem are problems that require us to nd a path with the prop-
erty that it can be traversed in the least amount of time possible. In each case we were
able to use ordinary di erential calculus to nd the path that solves our problem. How-
ever, in order to do so, we have had to make certain assumptions, such as the assump-
tion that the shortest distance between two points is a straight line. (This is a prop-
erty that we usually take for granted!) The work of Euler and Lagrange on the brachis-
tochrone and other related problems has given rise to a technique known as the calculus
of variations, that is more widely applicable and more rigorous than the approaches we
have been considering so far.
The sort of problems that can be tackled by this approach are those that involve nding
a path between two points P and Q that minimises (or in some circumstances max-
imises) a given quantity associated with paths from P to Q. In the examples we have
seen so far we wished to minimise the total time taken to traverse the path; for other
problems we might be interested in minimising the length of the path, or some other
quantity.
For the sake of simplicity, we will restrict our attention to problems involving paths in
R2. Let P , Q be points in R2. A path from P to Q can be given a parametric represen-
tation in terms of a continuous2 function f : [0; 1] ! R2 , with f (0) = P and f (1) = Q.
The basic type of problem that can be solved using calculus of variations is the follow-
ing: given a function S (f ) that takes a path as an argument and returns a real number,
we would like to nd the speci c path fmin for which S (fmin ) is as small as possible.
We note that this is somewhat di erent to an ordinary calculus problem. In standard
calculus we are given a function of a variable, and we nd maxima and minima of the
function by determining the values of the variable for which its rst derivative is zero.
These points at which the maxima/minima occur are stationary points, that is they are
points for which the value of the function only changes by a very small amount if we
deviate slightly from those points. In the case of calculus of variations, we are instead
given a function S that maps paths to real numbers, and we require the path for which
the value of S is minimised. However, the intuition is similar: if f is that path for which
the minimum value of S is attained, then for any path g that is close to f you would
expect that S (g ) would not vary much from S (f ).

4.1 Variations

We now need to introduce some notation to describe what we mean by \paths close to
f ". Suppose that f (t ) and g (t ) are two paths from P to Q. We can de ne a function
 : [0; 1] ! R2 by de ning  (t ) to be the vector given by  (t ) = f (t ) g (t ). Then
 (0) = f (0) g (0) = P P = (0; 0), and  (1) = f (1) g (1) = Q Q = (0; 0).
2 Technically, we will require the rst two derivatives of f to be continuous also.

11
 P

g (t )

 (t )
f (t )

We say that  is a variation since it describes how the path f (t ) varies from the path
g (t ). (Hence the name \calculus of variations"). Speci cally, we will de ne a variation
to be any continuous function  : [0; 1] ! R2 with continuous rst and second deriva-
tives, and which satis es (0) = (1) = (0; 0). Then for any s 2 R, and any path f
from P to Q we have that f + s is a path from P to Q. If we let s approach zero, then
the path f + s becomes close to the path f .

4.2 The Euler-Lagrange Equations

Now we wish to examine how the value of S changes for paths that vary only slightly
from the optimal path f . To do this, we consider what happens when we evaluate S on
the path f + s.
Suppose f is the path that minimises the value of S . We can consider the quantity
S (f + s ) to be a function of the variable s , which has a minimum at s = 0. Because
S (f + s ) is just a function of a real variable, we know from ordinary di erential calculus
that
d
S (f + s ) = 0:
ds 0

We note that this property holds for any variation . We will use it to derive the so-
called Euler-Lagrange equations, a fundamental tool used in the calculus of variations.
These allow us to obtain a di erential equation from a calculus-of-variations problem.
They can be expressed as follows:
Theorem 4.1 (Euler-Lagrange Equations). Let f : [0; 1] ! R2 be a path from P to Q,
with f (t ) = . For purposes of clarity will use the notation x_ = ddxt and y_ = ddyt .
x (t )


y (t )
Suppose S is a function of paths from P to Q de ned by
1
x x_
Z     
S (f ) = L ; ;t dt
0 y y_

for some smooth function L : R ! R.

5
If f is the path for which the value of S (f ) is a
minimum, then we have
@L d @L
=
@x dt @ x_

12
and
@L d @L
= :
@y dt @ y_
(We note that it is possible to generalise this to paths in a higher number of dimen-
sions, and also to consider functions L that also involve higher derivatives of x and y .)
Proof. (sketch.) We have seen that if f is the path for which f (S ) is minimised, then
d
S (f + s ) = 0 for any variation  (t ) = . To make use of this fact, we observe
1 (t )


ds 0
2 (t )

that
d d 1
+ s1 xx_ + s _1
Z     
S (f + s ) = L ; ; t dt
ds 0
ds 0 y + s2 y_ + s _y 0

d x + s1 x_ + s _1
Z 1     
= L ; ;t dt
0 ds y + s2 y_ + s _y 0
1
@L @L @L @L
Z
= 1 + 2 + _1 + _2 dt (multivariable chain rule)
0 @x @y @ x_ @ y_
1 1 1 1
@L @L @L @L
Z Z Z Z
= 1 dt + 2 dt + _1 dt + _2 dt
0 @x 0 @y 0 @ x_ 0 @ y_
d @L d @L
1 1
1 1
1 1
@L @L @L @L
Z Z  Z  Z
= 1 dt + 2 dt + 1 1 dt + 2 2 dt
0 @x 0 @y @ x_ 0 0 dt @ x_ @ y_ 0 0 dt @ y_
d @L d @L
Z 1 Z 1 Z 1 Z 1
@L @L
= 1 dt + 2 dt 1 dt 2 dt (9)
0 @x 0 @y 0 dt @ x_ 0 dt @ y_
(since 1 (0) = 2 (0) = 1 (1) = 2 (1) = 0, as  is a variation.)

@L @L
! 
1 d
1


Z 
= @x
@L
dt
d
@ x_
@L
dt: (10)
0 @y dt @ y_
2

It is a consequence of a result known as the Fundamental Lemma of the Calculus of

Variations that if equation (10) is true for all variations  then we have
@L @L
0
!
d  
@x dt @ x_
= ;
@L
@y
d
dt
@L
@ y_
0
which is precisely the Euler-Lagrange equations, as required.
To illustrate how the Euler-Lagrange equations are used, we will apply them to the
brachistochrone problem.
Example 4.2. Suppose we wish to solve the brachistochrone problem for points A and
B . As before, suppose that A is at the origin. We let f : [0; 1] ! R2 de ne a path from

13
 to B , with f (t ) = yx ((tt )) . We will use the notation `(t ) to describe the length of the


A
path from A along f up to the point f (t ).
The function that we wish to minimise to solve the brachistochrone problem is the total
time taken to traverse the path. As we did when considering Bernoulli's solution to the
problem, we will make use of the fact that when the bead reaches the point f (t ), its
speed ddt` is determined entirely by its y -coordinate, namely
d` p
= 2g ( y ): (11)
dt
Now, as in Example 1.2, we have ddt` = x_ 2 + y_ 2 . From this we can derive the fact that
p

d` d` dt
=
dx p dt dx
= 1 + y 02: (12)
(We are using the notation y 0 = ddyx .)
Then, the total time taken to traverse the path f can be found by integrating ddxt as x
varies from 0 to the x -coordinate of B , which we will denote by B1 . Thus we have
dt
Z B1
S (f ) = dx
0 dx
dt d`
Z B1
= dx
0 d` dx
1 + y 02
Z B1 p
= dx
2g ( y )
p
0

1 1 + y 02
Z B1 p
= p 2g 0
p y
dx : (13)
We can now apply the Euler-Lagrange equations in the form
@L d @L
=0
@y dx @y 0
p 2
to the function L = p1+yy . We have
0

@L d @L 1 d 0
0 = (1 + y 0 2 )1=2 ( y ) 3=2 (y (1 + y 0 2 ) 1=2 ( y ) 1=2 )
@y dx @y 2 dx
1
= (1 + y 0 2 )1=2 ( y ) 3=2
2
y 00 (1 + y 0 ) 1=2 ( y ) 1=2 + y 0 y 0 y 00 (1 + y 0 ) 3=2 ( y )1=2
2 2

1

+ (1 + y 0 2 ) 1=2 ( y ) 3=2 y 0


2
1
(1 + y 0 2 )2 y 00 (1 + y 0 2 )( y ) + y 0 2 y 00 ( y) 1
(1 + y 0 2 )y 0 2
= 2 2
(1 + y 0 2 )3=2 ( y )3=2
y y 00 + 21 (1 + y 0 )
2

= :
(1 + y 0 2 )3=2 ( y )3=2

14
Hence we obtain the following di erential equation:
2y y 00 + 1 + y 0 2 = 0: (14)
Thus we have succeeded in turning our variational problem into the problem of solving
an ordinary di erential equation. It is straightforward to check that (14) is satis ed by
the same cycloid described in Example 3.1.
Hence we now have a systematic approach for tackling variational problems. These
techniques can be applied to solve a wide range of problems, and have signi cant appli-
cations in physics. It is interesting that while the brachistochrone problem is a seemingly
recreational puzzle, it has inspired mathematicians to develop new branches of mathe-
matics with signi cant practical impact.

5 The Tautochrone Problem

Finally, we mention one further signi cant problem on curves, the so-called tautochrone
problem. First studied by Huygens, this can be expressed as follows
Suppose you wish to connect points A and B with a curved wire that has
the property that a bead released from any point on the wire takes the same
amount of time to slide to the bottom. What curve should you choose?
It is rather surprising that the answer to this problem is in fact the same as the answer
to the brachistochrone problem: namely, the desired curve is a cycloid. In the words of
Johann Bernoulli:
\Before I end I must voice once more the admiration I feel for the unex-
pected identity of Huygens' tautochrone and my brachistochrone... Nature
always tends to act in the simplest way, and so it here lets one curve serve
two di erent functions, while under any other hypothesis we should need two
curves ..."

Recommended Reading

[1] Feynman, R.P., Leighton, R.S. and Sands, M., The Feynman Lectures on
Physics, (vol. I and vol. II) Addison-Wesley (1964).
[2] Garner, M., Martin Gardner's Sixth Book of Mathematical Games from Scienti c
American, W.H. Freeman and Company (1997).
[3] MacTutor History of Mathematics Archive, Cycloid, http://www-history.mcs.
st-andrews.ac.uk/Curves/Cycloid.html, (accessed 7/11/2011).
[4] Stillwell, J., Mathematics and Its History, Springer (2001).
[5] Weisstein, Eric W., Cycloid, From MathWorld{A Wolfram Web Resource, http:
//mathworld.wolfram.com/Cycloid.html, (accessed 7/11/2011).

15