1

Introductory example : a gravitational

catastrophe machine

“Tam afraid that I rather give myself away when 1

explain, said he. Results without causes are much
more impressive.” (The Stockbroker s Clerk)

It is well established that one should never begin a talk — or presumably a

book - with an apology. We apologize, therefore, for apologizing that
despite the title of this chapter our book is not primarily about catastrophe
theory. The reason for our beginning with a gravitational catastrophe
machine is that it exemplifies, in a vivid way, many of the ideas we shall
study in detail later, such as functions on a curve, envelopes, surfaces,
projections, evolutes and bifurcation sets. These ideas are merely touched
on in the present chapter: do not expect to understand all the details yet.

The gravitational catastrophe machine was invented by T. Poston and

is discussed in the well-known book on the subject (Poston and Stewart,
1978). Other introductions to catastrophe theory can be found in Zeeman
(1977), Poston and Stewart (1976), Saunders (1980).

Consider a parabola, cut off by a line (perpendicular to the axis say), as

in fig. 1.1. Imagine the region enclosed to be a lamina (thin sheet) that is

Fig. 1.1. Parabolic lamina in a vertical plane

2 Introductory example

constrained to move in a vertical plane, resting on a horizontal line; we

seek the position of stable equilibrium. We do not assume the lamina to
be of uniform density; in fact let its centre of gravity be at the point (a, b)
referred to axes x and y as shown, relative to which the equation of the
parabola is y =x?. (The point (a, b) is often said to be in the control space:
the values of a and b control the behaviour of the lamina.) Then we are
looking for positions that give (local or global) minima of the potential
energy of the lamina. The latter is easy to calculate: it is mgh, where m is
the mass, y the acceleration due to gravity and 7 the height of the centre of
gravity above some base line, which we take to be the line on which the
lamina rests. Thus, if the parabola makes contact with the line at (r, +*),
then A is the distance from (a, b) to the tangent y —2tx+ 1? =0. The poten-

Fig. 1.2. The surface of normals to a parabola

N.'

X fo, 16)

»
,
1
t
1
'
'
U
t
1

au

*,

(0, to, 48)

A gravitational catastrophe machine 3

tial energy is therefore

b—2ta+1?

1.1 0=ray MA,

and the mathematical problem is simply, for given a and », to find the
value or values of r which minimize V(t). We find that

(1+47)

1.2
2mg

V'(1)= UC), where V(1)=25> + 1(1 —2b) —a,

so that we want U(1)=0 fora turning point and U to change from negative
to positive at this t for a minimum. Notice that U depends on the given
a and b; as a and b vary we obtain a two-parameter family of functions.
Such families will occupy us a good deal later on.

Now 213 +1(1—-2y)-x=0 is precisely the equation of the normal

to the parabola at (t, 1”). Thus we obtain first the unsurprising fact that for
equilibrium (a, 6) must be vertically above the point of contact (t, 1?),
for the normal at (t, 1?) is the vertical line through that point. We shall
return to the question of minima shortly.

It is instructive to regard 213 +1(1—2b)—a=0 as defining a surface M

(called the catastrophe surface of V) in the three-dimensional space with
coordinates (t, a, b). With the r-axis vertical, t =10 is a horizontal plane
which meets M in the line £—to=2t8 +to(1—2b)-—a. This is a normal
line to the parabola in the plane /=to given by the equation b=a?, the
normal being at the point t =t0, a=to, b=t6. What this means geometric-
ally is that M is obtained by taking the normals to the parabola b=4? in
the (a, b)-plane and moving the normal at (to, 18) vertically to the height ro.
The normals are spread out to form a surface by using the r-direction*
(fig. 1.2).

The vertical plane b=b, meets M in the cubic curve 213 +1(1 —2b0)
—a=0. As b increases through bo =4 this curve acquires a maximum and
a minimum, thinking of a as a function of.

1.3 Exercise
Sketch the cubic curve in the plane b=b, for bo =0, 3 and 3.

* Itis possible to make a model of the surface M using cotton threads for
these lines, passing through holes drilled in perspex (or wooden) sides
b =const. Here, in centimetre units, are good dimensions to use. Try the
curve 326 =a? with planes b=0, b=24, taking r =0, +0.8, +1.6,..., +16.
The face b=0 can be about 36 in r-direction x 24 in a-direction ((t, a, b)
=(0, 0, 0) at the centre) and the face b =24 can be about 36 in r-direction
x 12 in a-direction ((0, 0, 24) at the centre).
4 Introductory example

We draw the folded surface M asin fig, 1.3.

For given a and b the vertical line through (0, a, b) meets M in points
(t, a, 6), which are solutions of U(1)=0 (see 1.2). The number of solutions
depends on (a, b); sometimes it is 1, sometimes 2 and sometimes 3. The
values of t correspond to turning points of the potential energy Y(t)
(see 1.1), and so to possible stable equilibrium positions for the parabolic
lamina.

M divides the (t, a, b)-space into two regions: that which is mostly
“above” M has U(t)>0 and the other has U(t)<0. Minima of V occur for
values of t at which U(t) changes from negative to positive. Thus when
there is one solution to V(t)=0 it is a minimum; when there are three the
middle one is a maximum and the other two are minima; when there are
two, so that the vertical line touches M (M has vertical tangent plane),
the point of contact is neither maximum nor minimum (U(1), hence Y (t)
remains the same sign) and the other solution is a minimum. The inner
part of the fold gives entirely maxima.

Let us find the set 2 in the (a, b)-plane consisting of points where there
are two solutions, ¡.e. the equation U(1)=0 has a repeated root. The condi-
tion for : to be a repeated root is U(1)= UV (t)=0, ie.

1.4 213 +1(1—2b)+a=0

6r? + (1-25) =0
and elimination of t gives
1.5 27a? =2(205—1).

Fig. 1.3. The catastrophe surface

U(1)>0
A gravitational catastrophe machine 5

Note that a=0, b =+is very special: this gives a triple root, for U(1)=1? in
this case, with triple root 0. The curve 2 given by 1.5 is a cuspidal cubic
in the (a, b) plane, .e. in the control space (fig. 1.4).

1.6 Exercises

(1) Show that, if r is the repeated root of U(t)=0, where (a, b) e 2, then
a=-— 41? and b=4(1+61?). This gives a parametrization of 2 (with param-
eter 1). The points (1, a, b) of M for which (a, b) € 2 therefore have the form
(1, —41, 401 +67).
This is the curve in M which projects to 2 in the (a, b)-plane. It is a space
curve and it has the regular parametrization given (this means that the
derivatives of the three component functions never vanish for the same
t - indeed here the derivative of the first component never vanishes at all).
Is the tangent line to the space curve ever vertical?

Sketch the potential function Y for various values of (a, b), such as (0, 0),
(0, 1), (1, 4). Compare with the rough sketches fig. 1.4. You may find it
helpful to use

V'(1)=201 +41?) 3213 + (1 —26)—a)

and

V"(1)=201 +41?) U(16b— 2)? + 12ar+1—2b).

The set 2, which is variously called the bifurcation set of Vor the
discriminant set of U, separates (a, b) giving one solution to U(1)=0
(outside Z, >in 1.5) from those giving three solutions (inside? 2, <in 1.5).
Looking at the surface M from above, a curve like 2 is what we 'see” of the
folded surface: it is the “apparent contour where the surface folds away
from us (likewise the apparent contour of a sphere held at arm's length is
a circle).

There is another way of regarding this. The first equation of 1.4 is the

Fig. 1.4. Number of solutions of U(1)=0, and sketches of Y for various

positions of (a, »).

a «+ “+
1 7 > 1 a >
N / " No N
7 18(0,4) N / N
, N / v V N
Ma 2 , q
/ 2 2 N / L/ V N
DA 313 N y / A w N y
Y
Yb W L
6 Introductory example

equation of the normal to the parabola b=a? at (t, r*): as r varies but a
and b remain fixed, it gives the family of all normals to the parabola.
Eliminating t between the equations of 1.4 amounts to finding the envelope
of this family of lines: a curve which is touched by each of the lines of the
family (sec 1.7 below). The envelope of the normals to a curve is also called
the evolute of the curve, so 2 is the evolute of the parabola b =a? (fig. 1.5).

1.7 Exercise
Show that the curve 1.5 consists precisely of points of the form
(—-41, 3(1 +61?) for r ER. Show that the tangent to 1.5 at the point given
by any value of t other than 0 coincides with the normal to the parabola
b=a? at the point (1, 1*). The missing normal, at (0, 0), is the line a=0. Is
this in some sense the tangent' to 1.5 at (0, 3)?

We could go on conjuring more geometry from the parabolic lamina,

but let us return to finding stable equilibria. For a given position (a, b) of
the centre of mass in the control space we now know (fig. 1.4) how many
solutions there are of U(t)=0, i.e. how many stationary points there are
of the potential Y, and also which are minima. Naturally for a physical
lamina the centre of gravity will be at some point of the lamina itself, ¡.e.
(a, b6) will satisfy a<b?. The exact values of t can be found by solving the
cubic equation 24? + 1(1 —25b)—a=0.

lis of particular interest to see what happens to the stable equilibria

if we move the centre of gravity (a, b) around in control space. We are not
concerned here with the dynamics of rolling, merely with the change in the
positions of stable equilibrium.

Fig. 1.5. Normals to a parabola and their envelope

A gravitational catastrophe machine 7

Suppose for example that the point (a, 6) moves steadily across %, along
the line 12345 in fig. 1.6. Corresponding to 1 there is only one stable equi-
librium, given by the point 1' of M. At 2 and 3 there are two equilibrium
positions available, but in practice the parabola tilts slowly so that posi-
tions given by ?' and 3' are taken. As (a, b) moves past 4 (which is on %),
however, no continuous change in the equilibrium position is possible:
the path on M above 345 has a break init at 4”, where the parabola sudden-
ly spins to the left, a large change in equilibrium position occurring from a
small shift in (a, b). This is called a catastrophe (fig. 1.7). 1t is well worth
making a model to illustrate this: as described in Poston and Stewart
(1978), one can be made from two parallel parabolas made out of thin
card, held apart by three or four polystyrene “pillars”. A magnet and disc
of metal on opposite sides of one card can be moved about to change
(a, b). Because of the weight of the card, the positions of the magnet at
which catastrophes occur will not coincide exactly with the bifurcation
set Z, but can be determined by experiment.

Fig. 1.6. Catastrophe surface and its projection to control space

A)

Fig. 1.7. Catastrophic change of equilibrium

Hb AS

Steady turning Sudden spin Steady turning

8 Introductory example

1.8 Exercises

(1) Describe the behaviour of the lamina as (a, b) moves (i) along the path
54321 in fig. 1.6; (ii) clockwise round a circle centred at the cusp in fig. 1.6;
(ii?) anticlockwise round the same path. In each case sketch the change in the
potential function Vas (a, b) traverses the path.

(2) How many normals to the parabola with equation b=a* pass through a
point (a, b) inside the parabola?

(3) A swimmer gets into difficulties in a parabolic cove, and needs to head for
the nearest point of land. How many choices does he have? This is similar
to, but not identical with, the above example. We should expect him to
choose the absolute minimum distance always, and the axis of the parabola
is now significant. How does the nearest point of shore change as the
swimmer's initial position moves across the cove? Does it move contin-
uously, or is there a catastrophe” The reader is recommended not to try a
practical experiment for this exercise. (This example was suggested to us by
Dr |. R. Porteous.)

One final point. Consider the potential function Y corresponding to the

cusp point (a, b) =(0, 3), namely

1+27
Y (1) +43 mg.

This has a degenerate minimum at t=0, ie. V'(0)=Y"(0)=7"(0)=0

but Y7*(0)40. Functions which have degenerate minima are to be
regarded as exceptional: most functions do not have them. However, we
do expect to find an occasional function with a degenerate minimum
amongst a two-parameter family of functions such as the potential
functions Y which depend on a and b. The occasional member of a general
family will be worse behaved than is a general function, and the bigger the
family, the worse the behaviour we can expect. This rather ominous maxim
will be made a little more precise in 6.21.

1.9 Project
An even more interesting catastrophe machine was invented by E. C.
Zeeman, and is described in various places, such as Zeeman (1977). Find
out about Zeeman's machine, its potential function, and the sudden changes
which occur as the control parameters (corresponding to a, b above) are
varied. (Why not build one too?)
2

Curves, and functions on them

*T think, Watson, that you have put on seven and a

half pounds since [ last saw you-

“Seven, 1 answered.

“Indeed, I should have thought a little more.

Just a trifle more, 1 fancy, Watson.

(A Scanda! in Bohemia)

Plane curves arise naturally in all sorts of situations and in many guises.
Solutions of Newton's laws of motion give the orbits of the planets as
ellipses with the Sun at a focus. A spot of paint on a train wheel describes
a cycloid as the wheel rolls. These are examples of curves parametrized by
time: for each time z a definite point on the curve is determined. If a solid
object (such as Dr Watson) is viewed from a distance its outline, also
called its apparent contour or profile, is essentially a plane curve (or a
curve on the retina), but this time it is not given dynamically as a moving
point (fig. 2.1). It is more reminiscent of curves given by equations f(x, y)
=0; these latter curves are one of the subjects of chapter 4. A curve may be
traced by a linkage of bars and gearwheels; the position of the pencil
drawing the curve perhaps depends on the angle of some controlling bar,
and so is parametrized by this angle. (Alas! We have no space for this
beautiful subject.) When the Sun's rays are reflected from the rounded
inner surface of a teacup they produce on the surface of the tea a bright
“caustic curve. The reflected rays are all tangents to this curve, which is
said to be the “envelope” of the rays. We study envelopes in chapter 5
and caustics at the end of chapter 7.

We are interested also in space curves (curves in R*) and to a limited

Fig. 2.1. Some apparent contours of a torus surface, hidden curves drawn
with broken lines

JJOO
10 Curves, and functions on them

extent in curves in R". For the present we look at parametrized curves and
coax the geometry from them by means of real-valued functions defined
on the curves. To see how this can be geometrically interesting and to set
the scene here is an example.

2.1 Example

Consider the ellipse x?+4y?=4. We can parametrize this by

(2 cost, sin 1): for each t this point is a definite point on the ellipse, and all
points on the ellipse have this form for some tr. We propose to measure,
at each point po =(2 cos to, sin £4), how round” the ellipse is, i.e. how closely
it approximates a circle. To do this take a general circle through po: let
its centre be (a, b). lts equation is C(x, y) =0 where

Clx, y) =(x—a)? +(y—b)?—4

and the constant 7 is chosen so that € vanishes at po. Thus the equation
g(t) =0 where

gt) = Cl? cos r, sin 1)=(2 cos r— a) + (sin 1-b)?—4

has one solution t=to0. But t=t, may be a repeated (double, triple, ...)
solution, which would indicate that several common points of circle and
ellipse have come into coincidence at t,. This phenomenon is measured by
the number of derivatives of y Which are zero at t=to. lf 9 (t0)=0 for
i=1,...,k—1 and +0 for i=k then £ solutions of g(t)=0 coincide at
t=t0; indeed it actually follows that g(t)=(1—to)91(t) for a smooth func-
tion 9, with g1(t0);-0 (compare 3.4). We call this k-point contact.

It is easy to verify that q'(t0)=0 if and only if —b cos to +2a sin t0—3
sin to cos rt, =0, Le. if and only if (a, 6) lies on the normal to the ellipse at
Po- For such (a, b) the circle and ellipse touch at po, having 'two coincident
points in common there. For most (a, b) it is impossible for y (14) to vanish
as well: we require (a, b) to satisfy also b sin to +2a cos to— 3 cos 214=0.
In fact for each rt; there is just one possible (a, b), namely

(3 cos? to, —3 sin? 10).

Thus (a, b) must itself lie on the curve traced out by this point as 1, varies.
This curve is called the evolute of the ellipse (fig. 2.2). It is the locus of
centres of circles having at least 3-point contact with the ellipse and these
circles are called osculating circles of the ellipse (fig. 2.3). The only points
(a, b) for which gy (t0) =9" (t0)=9"(t0) =0is possible are (+3, 0), (0, +3), and
sin 2r,=0 at the corresponding tq. These osculating circles have all four
of their possible intersections with the ellipse coincident at one point po,
and the only points po on the ellipse for which this can happen are the
*vertices” at the ends of the axes (10 =0, 37, 7, 37). The ellipse is 'rounder
there than anywhere else.
Curves, and functions on them 1

Fig. 2.2. Evolute of an ellipse

Fig. 2.3. Osculating circles of an ellipse

The function g, with the constant 4 omitted, is called the distance-

squared function on the ellipse:
f()=(2 cos 1—a)? +(sin 1—b)?
measures the square of the distance from (a, b) to the point with parameter
12 Curves, and functions on them

ron the ellipse. Of course, fand y have the same derivatives. We can put
all the distance-squared functions into one family

F(t, a, b) =(2 cos 1—a) + (sin 1 —bY.

Such families will be important for us later.

2.2 Exercises

(1) In example 2.1 replace the ellipse by the parabola y =x?, parametrised
(t, 17). Find the centres of the osculating circles of the parabola (circles
having at least 3-point contact; their centres trace out the evolute of the
parabola). Find the unique point (a, b) for which there is a circle with 4-
point contact. Note that the equation /(1)=0 (fdefined as in 2.1) has the
same solutions as Y (1)=0 (Vas in chapter 1).

In example 2.1 replace the ellipse by the ellipse x?/4?+y?/B?=1,

parametrised (A cos 1, B sin 1). Show that for each r there is exactly one point
(a, b) such that the circle centre (a, b) passing through (A cos r, B sin 1) has at
least 3-point contact there, namely (a, b)=((4? — B”) cos? 1/4, (B?— 4?)
sin? 1/B). Find the points (a, b) for which 4-point contact is possible. What
happens when 4 = 8?

Parametrized Curves

In order to talk about general curves we need a proper definition.

2.3 Definition
A (parametrized) curve in the real Euclidean space R" is a map
y: I>R"

where 7 is an open interval in R, where

AD =(71(1), Y 201); --- > 77(0)

and each function y; has derivatives of all orders, for all t € 7. Such a y;
is called smooth. The curve y is called regular provided there does not
exist t ET with y1(1)=74(1)=-:: =71(1)=0. The variable t is called a
parameter: the point y(t) has parameter value r.

Unless otherwise stated all curves will be regular in this book. The vector
(11), Y2(1), - - - 74(1)), which is written y'(r) or dy/dt, is called the velocity
vector of y at t.

The set of points y(1) in R”, ie. the image C=(7), is what we often think of
geometrically as “the curve”. Cis also referred to as the trace of y. The point
y(t) moves along C as r varies, and in fact never stops or turns round,
since y (1) is never zero.
2.4

Parametrized curves 13

Exercises
Unless otherwise stated, check that the following are regular curves.

(1) y(1)=(t, 1), 7=R: parabola in R?.

(2) y(1)=(cos 1, sin 1), 7 any open interval containing [0, 27]: circle in R?.

(3) 7()=(, 15), 7=R. The image y() is still the parabola of (1), but y is not
regular at r=0. What about y(£)=(?, 1*)?

(4) Y(1)=(A cos 1, B sin 1), 4>0, B>0, Tas in (2). The image here is an ellipse
in R?.

(5) y(1)=(1? —1, 3—1), 7=R. Here the image y(7) is a curve which crosses itself
(1=1,1=— | both give y(1)=(0, 0)) (fig. 2.4).

(6) (1) =(1, 17, 13), 1=R, gives a twisted cubic curve in R3.

(7) y(1) =(cos t, sin 1, 1), 1=R, gives a helix in R3.

(8) 7(1)=(2, 13), 7=R. This is not regular at 1 =0, and the image has a cusp at
(0, 0). (Compare fig. 1.4, where the cusp is at (0, 4).)

(9) The evolute of the ellipse in 2.1, parametrized y(1) = (3 cos? t, —3 sin? vis
regular except for 1 =0, 37, 7, 37 (and these plus multiples of 27, which give
the same points on the image of y). These non-regular points correspond
to the cusps on fig. 2.2. What is the corresponding result for the parabola
and ellipse of 2.2?

(10) Curves arise naturally as solutions of differential equations. For example,

the motion of a particle in a vertical plane under gravity is governed by the
equations y" = — 9, x" =0 where x and y are functions of the time t (and we
use ” for d/dr), and gis a positive constant. These have a solution
(x, y) =(ar+b, —3g1” + et+d)
for any fixed numbers a, », e, d. The formula y(t) =(x, y) gives a curve which
is regular for all r unless a =0, when it fails to be regular for t=c/g. What
does this irregularity correspond to physically? For a+0 the curve is
always a parabola. Note that in an example such as this one the parameter
t has a special significance (namely, it is time).

A slightly different kind of solution curve arises for equations such as

y =ky (k +0) (where ' means d/dx here) with solution y =ae** for any fixed

Fig. 2.4. The curve of exercise (5)

14 Curves, and functions on them

a. This curve could conveniently be parametrized by x (y(x)=(x, ae*”)) but

there is nothing wrong with say y(1)=(1 +1, ae**D) which has the same
image. In both cases we can take 7=R.

What is the condition on the constants a, b, c, d for the curve y(r)

=(acos1+bsin1,ccost +dsin 1) to be regular for all 1? Whatis the equation
of the curve when it is regular? What happens for other values of a, b, c, d?

Tangent vectors

We shall often use vector notation and draw pictures with vectors.
Given points q, r in R” the segment from q to r represents the vector r —q.
Thus the segment from 0 to p (0 being the origin) represents the vector p
and so does any segment parallel to this (fig. 2.5). Vectors are indicated by
arrowed segments in diagrams. The right-hand diagram above indicates
the well-known vector law of addition.

WARNING In chapter 4 we shall need to worry about the point where a

vector “starts from” — the vector is said to be based at that point. But until
then such matters will not concern us.

Let y: 7>R" be a (regular) curve. The vector v=Y(t +h)—y(1) corre-

sponds to the chord segment from y(t) to y(t +7).

YE + 4)-— (1)
h
has for its direction the limit of these chords, i.e. the tangent at (7) (fig.
2.6).

(71(1), - .-, 71(6)) =7()=lim

Fig. 2.5. Vectors

Fig. 2.6. Tangent vector

Tangent vectors 15

For any x=(xX1,..., X) ER" the length of xis |x]| =(x1+x3+--- +x2)*.
Thus ||y'(1)]) is never zero.

2.5 Definitions

The vector T(1)=7"(1()y" (1)]]) is called the unit tangent vector to y

at 1, or at y(r). The length | (1)| is called the speed of y at 1. We call y unit
speed if ||y'()] =1 for all r. The tangent line to y at t is the straight line
through Y(r), containing the direction 7(1).

Note that y(t,)=Y(t:) need not imply y (t,)=y(1:). For example consider
y()=(?—1, 3—1), 1,=—1, t,=+1 (fig. 2.4).
For n=2 the equation of the tangent line to y at ris easily verified to be

(x1—71(6)75(1)— (x2 —71(1))71(1)=0.

2.6 Recall that the scalar product or dot product of x=(x:,..., x1,) and
y=01....),) in R" is the real number
X'Y=X191+X292+ 1 + Xan)
We note the following facts:
(1) x-y=!xi| |y[] cos 9 where 0 is the angle between the vectors x and
y (fig. 2.7). In particular x- x= |x|? and x- y =0 if and only if x and
y are perpendicular. (The zero vector is perpendicular to every-
thing.)
(2) If the x; and y; are all (differentiable) functions of t then

,_d o
(xy) =76 y=xy+x"y

where x' =(x1,..., X,) etc. Also, |x]' =x-x/|x]|.

(3) If x-x=1 (Le. x is a unit vector) and the x; are all differentiable
functions of t, then x-x' =0 for all t so, for each 7, x and x' are
perpendicular vectors.

As a particular case of (3) the unit tangent vector T to a curve y satisfies

T-T'=0 so T' is a (possibly zero) vector perpendicular to 7: (We often
write T T' rather than 7(r), T'(t); shortness is sometimes better than
perfect precision.) Since |T| =1, 7” measures the rate at which the

Fig. 2.7.
16 Curves, and functions on them

tangent vector is turning: the longer 7” is the faster T is turning and,

roughly speaking, the more curved the curve is.

Contact

We shall shortly be considering functions defined on curves in

R? and R*. Here by way of motivation is a little more on contact, Which we
met previously in 2.1. Let 4 and p be points of R”, and consider the two
sets of points x of R” defined by equations Fx) =0 where
27 F]=|x-u|*-|u—p]*
2.8 FO)=x—p)u
For n=2 these are the equations, respectively, of the circle centre u
passing through p (provided u £ p) and (for v +0) of the straight line through
p perpendicular to the vector (fig. 2.8). For n=3 “circle” becomes sphere
and “straight line” becomes plane. (For general n the usual names are
sphere and hyperplane, respectively.)

In chapter 4 we shall begin to study solution sets of equations, that is

sets F-'(0)=(x eR": F(x)=0). Granted one condition —- that O isa
“regular value of F — we shall find that these sets are very nice, even
better than images y(7) of regular curves y. For the present we shall not
dwell on the technicalities since the above four examples will suffice,
but note that the function F used to define the set F-1(0) is important.
For example, the square of F will not do (though it defines the same set)
because O is no longer a regular value of F. Thus “F-1(0) in what follows
assumes that we are using the defining equation F(x)=0 (see 2.12(5)
for a slight relaxation of this).

Given a (regular) curve y: 7>R” we measure its contact with F-1(0)

as follows:

Fig. 2.8.

Y
..
Contact 17

2.9 Definition
We say that y and F-1(0) have k-point or k-fold contact for t=to
(or, more loosely, at p =(t0)) provided the function y defined by

9(1)= FY:(6), .... 7.(1))= FO)

satisfies g(to) =g(t0)=--- =9"= M0) =0, 9 t,) £0. We also say that the
order of contact is k. Dropping the condition 9 (t0) £0 we say that there is
at least k-point contact, or >k-point contact, or that the order of contact
is >k. (lt might be “infinite”.)
NOTE We shall always use functions F for which gis a smooth function.
Observe that (0, and not just p, must be specifie

The condition on y is simply the formulation, in terms of derivatives, that

to is a k-fold root of the equation g(t)=0. It actually implies (see 3.4) that
gt) =(t— t0)91(t) for a smooth function y; with y: (10) 0 when y (10) £0.

As a simple example consider n=2 (plane curves) and lines through

Y (tc). The line perpendicular to a unit vector 4 has equation (x — Y(t4)) 4 =0,
and the contact with y for t=tg is measured by g(t)=(y(t)—Y(t0))- 4.
Now g (t0)=0 if and only if 7T(t,)-u=0, ie. u is perpendicular to T(t0),
ie. the line is parallel to T(t,) and hence is the tangent line. So our defini-
tion does tell us that the tangent line at tg is the unique line having at least
2-point contact for 1 =t0. (Thatis certainly good news.) Sometimes tangent
lines have much higher contact. For example let n=2, y(t) =(t, 1"), 10 =0,
Fx, X2)=X2, so that F=0 is the x¡-axis in R?, Then AFy(1))=r" so the
curve has k-point contact with the x,-axis at the origin (fig. 2.9). For k=1
the curve (now a line) does not touch the axis at all (1-point contact =
crossing without touching) but as k increases, the curve y becomes increas-
ingly flat at the origin. Just as contact with circles (2.1) measures how
“round” the curve is, so contact with lines measures how “flat it is.

There are special names for points of plane curves which are especially
round or especially flat.

Fig. 2.9. k-point contact

18 Curves, and functions on them

2.10 Definition

An ordinary (respectively higher, or degenerate) vertex is a point

p=(t,) of a plane curve y, with parameter value to, for which there exists
a circle* having 4- (resp. at least 5-) point contact with the curve for t =to.
We say y has a vertex at t0, or at p.

2.11 Definition

An ordinary (resp. higher, or degenerate) inflexion is a point

p=(t0) of a plane curve y, with parameter value to, for which the tangent
line* at to has 3- (resp. at least 4-) point contact with the curve for 1 =to.
We say y has an inflexion at £5, or at p.

Thus the ellipses and parabola of 2.1 and 2.2 have no higher vertices, and
ordinary vertices at the ends of their axes. (The ellipse in 2.2 (2) with 4=B
is the exception: if y is a circle then naturally there exists a circle having
very high contact with y everywhere, namely the circle itself. Every point
of acircle is a higher vertex.) See figs. 2.10 and 2.12.

2.12 Exercises

(1) Let y(0) =(t, ), with k an integer > 1. Show that, for any 10 €0' the tangent
line to this curve at 1, has exactly 2-point contact with the curve for f =to.
Show that 1, =0 gives a vertex on the curve if and only ifk =2.

(2) Let y(1)=(£, Y (1)). Show that the tangent line to the curve y at 10 has k-
point

contact with the curve for £ =10 if and only if Y "(t0)=0 for ?2<i<k—1 and

Y (t))£0. In particular 1 =to gives an inflexion if and only if Y"(10)=0,

ordinary if and only if also Y "(1,)+0.

Let y(1)=(2—1, 3-1), to=1, 11 =—1, so that y(t0) =7(t1)=(0, 0). Show that

the tangent line at 1, has 2-point contact with y for t =1, and 1-point contact

Fig. 2.10. Inflexions and vertices

In this book we are careful to distinguish between contact with circles and contact
with ines. Thus we do not allow 4-point contact with the tangent line to count as
4-point contact with a circle “of infinite radius'. In some books a higher
inflexion
does count as a vertex.
Contact 19

with y for 1 =1,. (The curve has two “branches” at (0, 0), which cross, and the
tangent to one branch does not touch the other branch, but crosses it
(1-point contact).) Compare fig, 2.4.
(4) (This needs a little real analysis.) Show that the image of a plane curve y
crosses its tangent line at an ordinary inflexion.
(5) Let y: 7>R” be a (regular) curve. Let F and F, be functions R”>R such
that the functions F* y, F¡ * y are smooth. Let to € 7, p=y(t0) and suppose
F:(p)+0. Show that, for r =to, the curve y has the same contact with F-1(0)
as with (FF,) '(0). (You have to verify that g(:)=F(y(1)) and q91(1)
= FY()F(7(1)) have the same number of vanishing derivatives at to
(remember Leibniz rule). The result is certainly reasonable, since the set
Fx)FI(x)=0 in R" will be, near p, the same as the set F(x)=0. But note that
F, =Fis definitely not allowed.)
Let y: 7>R” be a (regular) curve and L: R">R” an invertible linear map.
Let G=F<c L, ó=L"! ey, where F:R">R is such that F* y issmooth, asin
2.9. Show that ó is regular and that G-1(0) and ó have the same order of
contact for t=t, as F-1(0) and y have for t =to. (Really this is a triviality.

(6

What it says is that contact is invariant under linear maps such as rotations
and reflexions. The same holds for translations L(x)=x+v for a fixed v.
HMustrate by taking say n=2, F(x1, x>)=x>, Y()=(1, 1?) and various maps L
such as L(x,, X2)=(X1, 4x1 + x1). Draw pictures so that you can see why
G has L and ó has L'.)

Once we have developed a little more in the way of technique we shall

return to these geometrical discussions. In practice it is slightly easier
to omit certain additive constants from the functions F(y(1)) where F is
2.7 or 2.8; of course this makes no difference to derivatives. Here are the
details. Let y: 7->R" be a (regular) curve.

2.13 Definition

Let w eR". The distance-squared function on y from u is the

function f;: 1>R defined by

Ja) =|700)—u|*=7(0)—1)- (7(1)—)

Provided is a unit vector, the height function on y in the direction
is the function f,: 7>R defined by

ht)=10-u
Note that y(r)- 1 is the distance of y(t) from the line (or plane) through 0
perpendicular to (fig. 2.11), ie. the “height of y(r) above this line (or plane).
We usually take u to be ofunit length in the height function: multiplication
by a non-zero constant does not affect which derivatives are zero.
The following are immediate from the definitions. Let y: 7>R? be a
(regular) plane curve.
20 Curves, and functions on them

Fig. 2.11. Height function

Y (7)

Y (1) -u

Unit circle

2.14 Proposition
y has k-point contact, for t=to, with the circle, centre u passing
through y(to), if and only if the distance-squared function f; on y from u
satisfies f P(t0)=0,i=1,...,k—1:f£ Mt) +0. In particular y has an ordinary
(resp. higher) vertex for t=to if and only if this holds for k =4 (resp. some
k>5, or f Mo) =0 for alli).
y has k-point contact, for t=to, with its tangent line, if and only if the
height function f, on y in the direction u perpendicular to T(to) satisfies

FPaY)=0, i=1,...,k—1: FPPAYEO. In particular y has an ordinary

(resp. higher) inflexion for 1 =to if and only if this holds for k =3 (resp. some
kK>4, or f (to) =0 for alli). 3

2.15 Exercises

(1) Let y(£)=(t, Y(1)) where Y(0)= Y (0)=0. Show that there is an inflexion
at (0, 0) (parameter value t =0) if and only if Y”(0)=0, ordinary if and only
ifalso Y "(0)+0. Show that there is a vertex at (0, 0) (parameter value ¿=0)
if and only if Y”(0)£0 and Y”(0)=0, ordinary if and only if also Y (0)
+3Y”(0).

(2) With y as in (1) and Y (0) +0 find the centre of the (unique) circle which has
at least 3-point contact with y, for 1=0. (This is the osculating circle or
circle of curvature there; compare 2.1, 2.27.)

(3) With y as in (1) show that if Y is an even function and Y”(0)70 then there
is a vertex at (0, 0) (: =0) andiif Y is odd then there is an inflexion.

(4) For a fixed real number 4 let y(1)=(??, 41+1*). Show that y is regular
provided 440. Show that 1 =0 always gives a vertex of y; are there any
values of 4 for which it gives a higher vertex? Investigate the distance-
squared functions when 4 =0.
Reparametrization 21

(5) The curve y(1)=(a cos t+cos 2t+1, a sin £+sin 21) (polar equation
r=2 cos r+a, where this is the same t and is the polar angle too), is called a
limagon (French for snail, but we cannot see the resemblance). Show that
inflexions (resp. vertices) correspond to the parameter values r for which
6a cos 1+ a? + 8 =0 (resp. sin '(a cos 1 +2)=0), (For the vertices consider the
distance-squared function f; from a general point (1, v) eR? and write the
first three derivatives of /, as linear combinations of 1, v and 1 (with coef-
ficients which are functions of r). Then eliminate by taking the determinant.)
For2<a<d4 there are both vertices and inflexions; for a=2 the curve is not
regular at 1 =7 (it has a cusp there); for O<a<2 there is a self-crossing and
no inflexions. For 2<a<4 there are two inflexions which coalesce for
a=4 into a single higher inflexion — see fig. 2.12. For a>4 there are no
inflexions. The coalescence and disappearance of inflexions (or indeed
vertices) is a common feature of families of curves. At the moment of
coalescence something “worse” happens — a higher inflexion or vertex.
(Note that according to our definition the curve for a =4 does not have a
vertex at £=7, but has a higher inflexion there. Compare the footnote on
p. 18.)

Reparametrization

There is a slightly unsatisfactory feature of our definition of

contact (now we admit it!). Consider y(t)=(t, 13), to =0, Fx, x:,)=x>2
so we are looking at contact with the x¡-axis, determined by F(t, 3)="*:
3-point contact. But (t)=(1— 21, (1 —21) gives the same curve” (image
y =image 6); it is simply that y(1 — 21) =6(t) so that for example y(0) =9(3)
and (0, 0) has parameter value $ in ó. Does ó have the same contact with
the x¡-axis for t=+ as y has for 1 =0? Mercifully it does, as you can check
easily, but there is clearly a general result we should prove. This changing
of parameter is something we can always do without affecting the geo-
metrical curve, and in fact changing parameter substantially shortens
some later calculations. So we pause here to give an account of these
matters.

The idea is that we make t in y(t) a function of a new parameter s, say

Fig. 2.12. Inflexions (x) and vertices (68) on limacons

22 Curves, and functions on them

t =h(s). As s traverses some interval J we want h(s) to traverse the interval 7

on which y is defined, exactly once, without stopping or turning round
(fig. 2.13). This just requires h to be monotonic (increasing or decreasing),
or equivalently dr/ds =7(s) to be never zero.

2.166 Definition
Let y: 7>R" be a (regular) curve. A change of parameter for y is a
map
h: J>1
where J is an open interval, satisfying
(1) his smooth (ie. all derivatives 45) exist for all s EJ),
(2) for all s € J, 4(5)F0,
(3) HJ)=1.
The curve ó: J>R” given by A(s) =y(h(s)) is said to be obtained from y by
the change of parameter h.

2.17 Remarks

(1) Is ó regular? Using the standard calculus rule on each component

Y: h of 6=y"h we have (y; * h)(5)=y(h(5))/(5); consequently
0'(5)=7 (h(5))/1(s). This is never zero for s € J, the first factor (a
vector) since y is regular, and the second (a number) by (2) above.
Of course y and ó give the same image in R”. They are different
ways of regarding the same set of points in R" as a curve.
It follows from 2.15 (1)-(3) that 7 is a bijection from J to 1, so
that the inverse h-!: 7+J exists. Furthermore 47! is smooth.
(This is not quite obvious. First one verifies directly that (4-')(r)
=1/h'(4-1(1)) for all r e 7, using the (crucial) fact that 7'(5) is
always non-zero. Then, knowing k-1 to have a derivative, we can

Dd

Fig. 2.13. Change of parameter

A 7 = h(5)

Y
Reparametrization 23

write down (4 'y(r) by standard calculus rules, and so on by

induction.) In fact the requirements (1)-(3) of 2.15 are equivalent
to (1), (3) and the following condition: 47! (which exists by (1)
and (3)) is smooth.
(3) A map has in 2.15 is called a diffeomorphism from J to 1 By the
remarks above, h-! is then a differomorphism from 7 to J. Also
the composite of two diffeomorphisms is a diffeomorphism.
Suppose that fand y are smooth functions 7>R (7 an open inter-
val, as usual). Then f +3, /y and cf (c constant) are all smooth.
Furthermore if g(t) is never zero then f/g and y!” (n odd) are
smooth. For n even one needs g(t) to be >0 rather than just non-
zero. All of these are easy to prove from standard calculus rules.

(4

2.18 Exercises

(1) Show that h(s)=s? defines a difeomorphism from (1, 2) to (1, 4), but not
from (—1, 2) to (1, 4).

(2) h(s)=s? does not define a diffeomorphism from (— 1, 1) to (—1, 1), even
though A? exists here.

(3) A(s)=as+b (a, b constants a +0) defines a diffeomorphism from 7 to h(7)

for any open interval 7.

(4) h(s)=e* defines a diffeomorphism from 7 to k(7) for any open interval 7.

(5) Let a, b, e, d, e be constants with d<e and let h(s)=as? +bs+c, 1=(d, e).
Find the conditions on a, b, e, d, e which ensure that his a diffeomorphism
from 7 to AD.

(6) Let y: 7>R? be a curve and ax +by + =0afixed line in R?. Define f: 7>R
by f(t)=the square of the distance from y(t) to the line. Verify that f is
smooth. Further, show that if a, b and c are smooth functions of t with a(t)
and A(t) never both zero for the same tr, then the square of the distance from
(1) to this variable line also gives a smooth function. What about the
distance function? The signed distance function? (Use the rules in 2.16 (4)
above.)

Recall from 2.9 that contact between a curve y and a “smooth hyper-
surface” F(x)=0 in R” is measured by the number of vanishing derivatives
of y, where gy(t)= F(y(1)). (In some special cases 2.7, 2.8, it is also measured
by the number of vanishing derivatives of distance-squared or height
functions.) Under a change of parameter h, y becomes y * h. Thus, the
following result shows in particular that y and any reparametrization
ó=y hol y have the same contact with F(x)=0.

2.19 Proposition

Suppose that g: 1>R is smooth and h: J>1is a diffeomorphism.

Let to E 1, so=h" "t0). Then the derivatives

gto), i=1, 2, ...,k, are all zero

24 Curves, and functions on them

if and only if the derivatives

(g 9 A) (so), i=1, 2... ,k, are all zero.

Proof We have (g * h)(s) =g'(h(s))'(5).

By repeatedly differentiating this formula we obtain

(y * hs) =9(MS)((5)” + terms having as factors

gs) for i=1,...,p—1.

H follows that if yM(h(s0)) =0 for i=1, ...,p, then (g * k)?(s,) =0. Applying

this for p=1, 2, ...,k gives the first result in one direction: the other
direction is obtained by writing 9 =(g*h)* kh"! and applying the same
method. 0

2.20 Remark

It also follows that (with y, h as in 2.19) y has a local maximum

(resp. minimum) at r if and only if y * h has a local maximum (resp. mini-
mum) at so. (Proof: 9 has a local maximum at ro ifand only if y'(t0)=0 and
g (1) changes from positive to negative as / increases through to. The result
then follows from the first line of the proof of 2.19 — note that if 4'(s,)<0
then g'(h(s)) changes from negative to positive as s increases through so, but
the factor /'(s) reverses the sign.)

Fora plane curve y: />R?, anda given to € 7, let /(t) be the arclength of the
curve y(7) from y(t4) to y(1). With y(1) =(X(1), Y(1)) we have

6-67
[Ca (a) Ta [

where we agree that /(t) should be >0 for r>t, and <O for t<to. The

intuitive idea is suggested in fig. 2.14; for more discussion see for example
Willmore (1959). It is the same for curves in R".

so that

7 (1) de

Fig. 2.14. Arclength

Curvature 25

2.21 Definition
The arc-length of a (regular) curve y: 7+R”, measured from
Y(to), where to e 7, is

10=| Wed «en

In particular if y is unit speed (ie. ||y(1)]|=1 for all 7), then 11)=r—to.
ie. apart from an additive constant the parameter + measures arc-length.
A unit speed curve is often said to be parametrized by arc-length.

Arc-lengths are more or less impossible to work out explicitly, except in

very special examples. In case the reader insists on working out an example,
here are a few possibilities.

n=2 y(1)=(r—sin t, 1 —cos 1) (cycloid)

(1) =(, log t) (graph of log function, 1 >0)

y(1)=(t, 1?) (parabola: this is hard)

Y(1)=(R cos t, R sin 1) (circle radius R: this is easy)

n=3 Y(1)=(cos 1, sin t, 1) (helix)

The importance of arc-length for us is the following, If we reparametrize

a given curve y using arc-length from a fixed £ as the new parameter, then
the curve will become unit speed.

Let y: 7>R" be a (regular) curve, and let 1, € 7. Define 4(r) as in 2.21.

Then / is smooth and ?(+)=|y'(1)|>0 for all 1 € 7, so / defines a diffeo-
morphism from 7 to an open interval J, namely J=(l(a), 1(5)) where
T=(a, b). Let 17* be the inverse of 7; then /- ' is also a diffeomorphism by
2.17 (2), ie. 17! is a change of parameter for the curve y.

2.22 Proposition
The curve a=y * 171: J>R" is unit speed.

Proof Write h=1-'. Then (y * h)(s)=7'(h(5)'(s) for all s e J.

However differentiating /(h(s)) =s gives 7 (h(s))'(5) =1 and 7'(h(s))=||y (A(s))|

By definition of /. Hence ||(y * AY(s)||=1, as required. a

Again only rather feeble examples can be written down explicitly.

The circle, radius R, centre (0, 0) has parametrization y(t) =(R cos 1, R sin +)

and unit speed parametrization a(s)=y(s/R)=(R cos (s/R), R sin (s/R)).

The other examples above can be treated similarly.

Curvature

Now that we know any regular curve can be made unit speed by a
mere reparametrization, we shall often assume our curves to be unit speed.
When it makes a difference to the formulae we shall say so. To make
26 Curves, and functions on them

matters (we hope) clearer, unit speed curves will be denoted by a below,
up to 2.29, and their parameter will be called s. Thus a'(s)= (s) for any
unit speed a (compare 2.5); for any y, T(1)=7(t|'(0)|.

2.23 Curvature and normal (1 > 3)

Let £: 7>R” be unit speed, and let n>3. The curvature of 2 at sis
defined to be x(s) = | T'(5)]| =((27(5))? + --- +(07(5))*)*. Thus x(5)>0 and
is a smooth function when restricted to the (open) subset of / where
K(s) +0. If x(s) 40 then the principal normal N(s)is the unit vector T'(/k(s).
Since T(s) is a unit vector, M(s) is perpendicular to 7(s) When x(s)=0,
Ns) is not defined.

For any arbitrary (regular) y, with arclength function / (relative to some

to € 1), we define x(1) to be the curvature of the unit speed curve a=y e/-!
at s=/() (see 2.22). This does not depend on the choice of 15, and we have

xO=|701/0)=|701/170)1-
(Proof Write T(1) for the unit tangent to y at rand T,(s) for the unit tangent
to % at s; then T(1)= T,(/(1)). Thus T'(1)=To(1(1))(t) and taking lengths
gives | (|| =x,(1(0))(t)= (1 (tr). Perhaps it is “clearer” in old-fashioned
notation: write s = (1); then

dT ds dT

ds dr di
and the length of the first term on the left is x_)

2.24 Curvature and normal (plane curves)

When n=2 we can usefully give the curvature a sign. Let

a: 1>R? be unit speed; then 7 is as before, but this time define the unit
normal N(s) to be obtained from 7(s) by rotating anticlockwise through
37. 1 as) =(X(3), Y(S)) then 71s)=(X (5), Y '(5)) and Ms)=(— Y (5), Y(5)).
Now T(s)is still perpendicular to 7(s) so (as we are in the plane) there is a
real number x(s) such that

T'(s) = (5) N(s).

We call k(s) the curvature of % at s. Note x(s) = + | T'(5)]. With 2 as above we
have, omitting s,

T=0"=(Y”, Y)=KN.
Dotting both sides with N=(— Y”, X') we have

K=YY— YY.

In particular K isa smooth function of sfor all s € 7. Remember this formula

assumes unit speed (see below).
Curvature 27

Note that x(s)>Oif 7(s) is turning towards the normal V(s) and x(s)<0
if 7(s) is turning away from Ms). See fig. 2.15.

Suppose that the angle between the tangent and the x-axis is /(s) (see
the right-hand figure), so that cos Y = Y, sin Y = Y”. (It is not too hard to
show that there is a smooth function y: I>R with these properties, which is
unique once we fix the angle for a particular so € 7 to be, say, in the range
O<Y(s0)<27). Then X" =(—sin YY, Y” =(cos YY so k(s)=Y(s). This
again assumes unit speed.

For an arbitrary (regular) y: 7>R?, y(1)=(Y(7), Y(1)) define x(1) as the

curvature of the unit speed curve y * /-1 at 7(t). We have

T(1)=(4(0, Y ()/(( (0)? + (Y()>*

MO=(— Y'(1), Y ()((X (6)? +((0)>)%
the denominators being of course /(t) (/=arclength from some t). As for
the curvature, we have y =%* / as in 2.22 with « unit speed. Thus
A= MU) ()= T(17(r)
y (1) =0"(41))7(0))? + € GE).
Now d is unit speed, so 2” (1(1)) is the curvature times the unit normal, and
(11) is the unit tangent. Dotting both sides of the last equation with
M1) (= Ny(1(1)) where N, is the normal for 4) we have
7 (1) M0) =x(07(1))*
and substitution gives, dropping t now,
YY YY
OY
The tangent angle formula becomes x(t)=(t/7(1).

2.25 — Serret-Frenet formulae for plane curves

These are formulae for 7”, N' in terms of T, N. For a unit speed
we already have, omitting s,

T=kN
(Le. T'(s)=x(s) N(s) for all s e 7) from 2.24. Now N' will be perpendicular

Fig. 2.15.

Y)

Ns) (5)

179)

Kk>0 K<O
28 Curves, and functions on them

to N (see 2.6 (3)) and so N'=4T for some real 4 (depending on s). Also
T- N=0, so TN4+ TN =0, which gives k +4 =0. Hence

N=—KT
For arbitrary plane curves we obtain
T'=KNI' =KNIy|

N=-kKTl=—KT
2.26 Exercises

(1) For (1), (2), (4), (5), (6), (7) and the regular points of (8), (9) in (2.4)
find
formulae for the unit tangent, unit normal (or principal normal) and cur-
vature as functions of. (For example, for the ellipse in (4) we get
T(1)=(— A sin t, B cos 1/(A? sin? 1 + B* cos? +
N(1)=(—B cos 1, — 4 sin 1)/(4? sin? 1+ B” cos? 1)?

K(1) =4B/(4? sin? 1+ B* cos? 1.)

(2) Let y(1)=(t, Y (1)). Show that x(1)= Y (1/(1 + (Y (+))>)*.

(3) The cubic curve y? =x— x* consists of an oval, lying in the region 0O<x<1,
and another part (fig. 2.16). Removing the points (0, 0) and (1, 0) the
oval splits into two pieces, parametrized by y(1)=(1, +(1—1?)*). Show
that, for these curves (0<r<1), k(1)= +2031*— 61? — 101 —413 —61? +
41 +1)%. Why do the two halves have opposite signs for K? Is x positive or
negative for the top half? (And is that the upper or the lower sign?)

Fig. 2.16. The curve of example (3)

Functions on plane curves 29

(4) Let y: />R" be a (regular) curve and ó=y ch where h: J—+1 is a change
of parameter. Writing K,, Ks for the curvatures, show K,(1)= + Ka(h”'(1))
for all te 7, + if A(s) is always >0, — ¡if A(s) is always <O. (This just
says that curvature is independent up to sign of parametrization, and
is a tiny extension of the definition. In fact y and ó can both be reparam-
etrized to the same unit speed curve a: K->R” or one to a and one to
—%, where — a has domain — K=(—u:u€ K) and formula — a( —)=a(1).)
Let y(1)=(1", 15) where t>0 and s>r>0. Show that y is regular and find
a formula for k(t). Show that, as 10, K(1)>0 when s>2r and K(1)+00
when s<2r. (The image of y for all (small) r is called an ordinary cusp when
r=2, s=3 (here k(t)>00) and a rhamphoid cusp when r=2, s=5 (here
K(1)—>0). More about cusps in chapter 6 — see 6.19.)

Du

Functions on plane curves

We can now use the formulae of 2.24 to discuss in detail the

distance-squared and height functions introduced in 2.13 and hence
contact of plane curves with circles and lines. Recall from 2.19 that re-
parametrization makes no difference to the number of vanishing deriva-
tives of a function on a curve, so there really is no loss of generality in
considering only unit speed curves here.

2.27 Distance-squared functions on plane curves

Let a: 7>R? be unit speed, and 4 =(a, b) eR?. We consider the
distance-squared function
SS) =|/2t5) —ul|* =(4s5) —1))- (45) —1).
Writing a for a(s), etc., we have
3 9)=(a—u) a =(a—u) T
3f"s)=a:T+(a—u) T=1+(4—u)N
3 (5)=2 KN +(a—u) KN +(4—u) KN
=(a—u) K N—(a—u)x*T
using 7- T=1, T: N=0. Itisnow a simple matter to write down conditions
for the first few derivatives to vanish.
We find
f(5)=05a—u=4N(7 eR)>1 is on the normal line to x at s
[=f(5)=0k +0 and a—u=— N/kex +0 andu=2+N/kats
N=f=f"=0atsorkkt0, u=2+ N/k and =0ats.
Working out /* we find that the extra condition for f Y to vanish at s
as well is that x"(s)=0.
Various deductions can be made from this and 2.14.
(i) Provided x(s) +0, there is a unique circle having at least 3-point
30 Curves, and functions on them

contact with x at s. The circle is called the osculating circle or circle of

curvature at sj its centre a(s) + N(s)/k(s) is called the centre of curvature of x
at s (or more loosely at a(s)); its radius p(s)= 1/x(s) the radius of curvature.
See fig. 2.3 for some circles of curvature of an ellipse. (The case k(s)=0
appears in 2.29.)

(ii) The unique circle of (i) has at least 4-point contact (resp. at least
3-point contact) iff x'(s) =0 (resp. K'(s) =x"(s) =0); hence this is the neces-
sary and sufficient condition (granted (s)£0) for a vertex* (resp. higher
vertex) of a.

(iii) An ordinary vertex occurs precisely at a simple maximum or

minimum of curvature, where k(s) +0.

For arbitrary (regular) plane curves y all of the above results hold.
(Reparametrize y as a=y /71 (2.22); this does not affect contact. As for
the conditions x'(s)=0, etc., recall x (1) =r,(1(t)) by definition 2.24, so that
Ky(1) =K(7(1)7 (1) and ri(t)=0 iff x,(1))=0, etc.)

2.28 Exercises
(1) Let « be a unit speed plane curve and let f be the distance-squared func-
tion as in 2.27. Suppose f (s)=0, so that u=a(s)+AN(s) for some 4 eR,
and also suppose x(s)> 0. Show, by investigating the sign of/(s), that f has
a maximum at s if 4> 1/x(s) and a minimum if 4< 1/(s). Thus the centre of
curvature at s separates points on the normal for which the distance to the
curve (or its square) is minimized at a(s) from those for which it is maxi-
mized. Can you say anything about 4 =1/k(s)?
Let 4 be a unit speed plane curve and define
9(5) =(4(s)— 4): N(5),
for a fixed u eR?, Show that g(s) is, up to sign, the distance from to the
tangent to £ at s. Show the following:
(1) 29(5)=— K(5)f (5) where f is as in 2.27,
(ii) y (5)=0ex(s)=0 or « lies on the normal at s,
(iii) y'(5)=9"(5)=0S(r(s) ='(5) =0) or (x(s) =0 and z lies on the normal
at s) or (4 is the centre of curvature at s).
What is the condition for g'(s)=g"(s5)=g9"(5)=0? What does it become if
K(s) +0? (Compare the Introductory Example in chapter 1.)
(3) Let a be a unit speed plane curve and define

9(s)=(a(s)—)- T(s).
What distance does g(s) represent? Find the conditions for the first xk
derivatives of y to vanish at s, for k=1, 2,3.

In some books a point where x= =0 counts as a vertex. See the footnote on

p. 18.
(4)

6)

(6)

Functions on plane curves 31

Let a be a unit speed plane curve, with x(0)=(0, 0), «'(0)= 7(0)=(1, 0),
as in fig. 2.17, Show that
(0) =(0, x(0))
07”(0)=(—x7(0), x1(0))
(0) =(— 3x(O)x(0), (0)— (1:(0))*).
Let a: /7>R? be unit speed and let F: R2>R? be a linear map with
matrix A (relative to standard bases). Let 8 = Fc a. Show that ['(s)= 4e(s)
(where we write a, $ as column vectors) and deduce that f is regular pro-
vided A is nonsingular, unit speed provided A is orthogonal. Deduce that,
if 4isorthogonal, then the curvatures ofx and $ are related by x(s) = + K,(s),
the sign being that of del(A). This shows that rotations preserve curvature,
reflections reverse it in sign. (Obviously, translations preserve curvature:
here, P(s) =a(s) + v for a fixed vector v.).
Let y be a plane curve with arclength function /, and x=ye/! be
unit speed as in 2.22. Show that x'(to)=--- =x))=02>K(s0)=--"-
=KP(s0)=0, where so =/(t0). Remember x(1)=K,(1(1)) by definition.
Consider curves parametrized y(1)=(1, Y(1)) for some Y with Y(0)= Y (0)
=0. Let a=yo/! (where 1(0)=0) be the unit speed curve correspond-
ing to y and r,(s) =x(/- 1(5)) be the curvature of a (see 2.24). Show that, for
any p>2, any one of the sets of numbers
(i) Y”(0), Y (0), ..., Y (0)

(ii) (0), (0), ..., x?-?Y0)

(iii) x,(0), x7(0), ..., x4-?10)

uniquely determines the the other two. (Repeated differentiation of
Y"=K(1+ Y > shows that Y? —xP-(1 + Y?) + (terms involving lower
derivatives of Y and x), and this shows that (i) and (ii) determine each other.
Repeated differentiation of /?=1+ Y? shows that Y”(0),..., Y?- (0)
determine /(0), . .., /?1(0) (note 7(0)=1, 7'(0)=0 automatically). Using this
and k,(1(1)) =x(1) shows that (i) and (ii), which determine cach other, also
determine (iii). Finally (iii) gives (ii) once /”(0), ..., /-?0) are known, and
these come inductively, determining them from Y”(0), ..., Y 30) and
these from x(0), ..., x?-350),)

Fig. 2.17.

o O
32 Curves, and functions on them

Deduce that y has >k-point contact with the x;-axis in R? (k>3) if

K(0)= - - + x-30)=0. Deduce also that any regular curve y: 7>R?

has >k-point contact with its tangent line at y(10) iM K(t0)= --- == 90)
=0.

(8) In the notation of (7), show that the first few formulae work out as follows:
K(0) =K,(0) = Y (0), x(0)=x:(0)= Y "(0), x"(0)= (0) = Y *(0)— 3(Y (0).
(Compare 2.15(1).)

2.29 Height functions on plane curves (see 2.13)

This proceeds much as for distance-squared functions. Let
a: 1>R? be unit speed, and consider, for a fixed unit vector u eR?,

f(5)=(s)-u

Thus f(s)=7(s)-u, /(s)= — K(s)N(s)-1, etc. We find

f(5)=0 > Ms) = +u (since N, 4 are unit vectors)
fI=f(5)=0 > Ms)= +4 and k(s)=0

f19)=f"(9)="()=06 Ms) = +1 and k(s)=1'(5)=0

If you go on long enough, you can re-prove (7) of the exercises above.
Note, at any rate, that the necessary and sufficient condition for x to have an
ordinary (resp. higher) inflexion for s=so is that (so) =0, and x(s,)*0
(resp. (so) =0). This result holds whether « is unit speed or not.

Now that we are not juxtaposing results for unit speed and other curves
we shall often use y, parameter t, for a unit speed curve, but always say
when it is assumed unit speed.

2.30 Exercises

(1) Pedal Show that the foot of the perpendicular from the origin to the tan-
gent to the plane curve y at y(') is given by (y(1): M(1))N(1) (fig. 2.18). As t
varies, this point moves on a curve called the peda! curve of y with respect to
0. Let us write

M1) =(y(1): NON), or just ó=(y- NIN

Fig. 2.18. Pedal curve

70

l AN
Y o Its) - NI

[1
Functions on plane curves 33

and now assume 7 is unit speed (5 won't be, in general). Show that

ó=— (7 TIN +6 NT).

Show that the second factor of 5'(1) is zero if and only if y(1)=0, which says
that, for this 1, the curve 7 passes through the origin. Assuming that y(1) is
never 0, deduce that d is a regular curve except at those points correspond-
ing to inflexions of y. Fig. 2.19 shows the pedal curve of a limagon - see
2.15(5), where we take a=2.3 and draw the pedal with respect to (1, 0).

(2) Parallels Let y be a unit speed plane curve and let d be a fixed real

number. The curve ó defined by (1) =y(1)+ 4 N(1) is called the para!lel to y
at distance d. Show that ó is a regular curve except for values of t where
K(1)+0 and d=1/x(1): also that for these irregular points ó(1) is the centre
of curvature of y at y(t). See figs. 2.21 and 5.10.

Evolute Given a unit speed plane curve y with k(t) never zero we can
consider the locus of centres of curvature of y, namely the curve

2(1)=y00)+ [1/4(1)] MO),

which is called the evolure of y.

(i) Show that the evolute is a regular curve except for those values of t
where x'(1)=0, ie. for points of e corresponding to vertices of y (Compare
2.27(ii)).

(ii) Assuming that x(r1)<0 on 7, so that the radius of curvature p is

increasing on 7, show that the arc-length on € from to to t,>to is P(t1)
— Plto).

(ii) With x<O again, show the unit tangent and normal to £ satisfy
T.=N, N.=— 7, and that the curvature of £ is xk, =— KK (ie. p,=pP").
Hence the evolute has no inflexions.

(iv) Deduce from (ii) thatifa piece ofstring is wrapped round the evolute,
one end being fastened at £(15) and the other end starting at y(10), then as the

Fig. 2.19. Limacon and its pedal curve

Limagon

*(1,0)

Pedal ——
(4

—-

Curves, and functions on them

string is unwrapped the ends are at e(1) and y(r) for all t>',. This says
Iy() — EI =|IY(t0)— Elto)|| + (arc-length on € from ro to 1.) See fig. 2.20.
Figure 2.21 shows some parallels and the evolute of a parabola: notice
that the cusps on the parallels all lie on the evolute compare (2) above —
and that the point of the evolute corresponding to the minimum of the
parabola, which is a vertex, is a cusp - compare (3) above.
Let r be a (differentiable) function of t and let X(1)=r(1) cos 1, Y (1)=r(1)
sin /. Then y(1)=(X(1), Y(1)) defines a plane curve by polar coordinates
(r, 1). What is the condition for y to be regular? To be unit speed? Show
that there are rather few functions r which make the curve unit speed.
Can you discover what the corresponding curves look like?

Fig. 2.20. String unwinding from the evolute

Nr)

El!
0) o)

El?)

Fig. 2.21. Parallels and evolute of a parabola

Functions on plane curves 35

2.31 Parametrization by x or y

Let y: 1>R?, y(1)=(X(1), Y (1)), be a (regular) curve in the plane.

Suppose that to € 7 and Y (t,)£0. Let 7, be a small enough open interval
containing t, so that, for all t € 7o, X(1)£0. Let X(70)=J: then X: Io->J
is a diffeomorphism. Using the change of parameter X-' applied to the
part of y defined by 7, we obtain

JE,
Calling this 8, we have 8(x)=( X- 1(x))=(X(X- '(x)), Y (X-1(x))=(x, f(x)
say where f is the smooth function Y * Y, Thus the reparametrization 2
of a part of y close to 1, displays this part of y as the graph of a function f,
with equation y =f (x). We say that y can be parametrized by x close to any
point where X (10)+0, ie. where the tangent line is not “vertical” (fig. 2.22).

Similarly if Y (10)+0 there is a reparametrization $ of the form 2(y)

=(9(y), y) for a smooth y, valid close to yo = Y (10).

Note that, if 4 is any reparametrization defined near ro, for which

y(h(1)) =(u, f(1)) for some £, then hy 1 = Y, for t close to to, so f =Y * Y.
Thus thef is uniquely determined by y and 15 — we should hardly expect the
image of y for t close to to to be the graph of two different functions.
Beware however, that we could have y(t0)="(£1) say, and then the pieces
of the image of y given by t close to t, and t close to 1, may well be graphs of
different functions.

2.32 Exercise
At (0,0) and (1, 0) the curve y? =x— x? in R? has vertical tangent. Assuming
that there exists a smooth function X(r), defined near +=0, such that
2 = X()—(X() (so that (Y(), 1) parametrizes the curve near (0, 0)), find
the curvature at the point with parameter r =0. Similarly for a parametriza-
tion of the same form near (1, 0). Comparing with 2.26(3), are your answers

Fig. 2.22. Parametrization by x

Ax) 10

EL
36 Curves, and functions on them

the limiting values of the x(t) there as 1>0, 1? What about the sign of xk?
(The assumption is true; compare 4.14.)

Space curves

We shall now take off from the plane into space and consider
space curves y: 1>R3. By 2.22 any (regular) space curve can be made unit
speed by a mere change of parameter, so we shall assume all our curves are
unit speed below. The necessary changes in the formulas for the general
case are easy to supply.

2.33 Serret-Frenet formulae for unit speed space curves

Let 2: 1>R* be unit speed, so that %(s)=T(s), the unit tangent

vector (2.5). Also the curvature is given by

k(s)=|7"(5)|
and provided x(s) +0 there is a unit principal normal Ms), perpendicular
to T(s), with (see 2.23)

T'(s)=K(S) Ms).
We shall continue to assume K(s) is non-zero in what follows. This is not as
artificial as it sounds, for in fact “most” space curves have nowhere zero
curvature. Compare 9.9.

Since T(s), NM(s) are perpendicular unit vectors there is a unique unit
vector B(s) called the binormal vector perpendicular to both and such
that 7(s), N(s), B(s) is right-handed (i.e. if the components are written as
the successive rows of a 3x3 matrix, then the determinant is +1 rather
than —1). The plane through a(s) spanned by M(s), B(s) is called the
normal plane at s (or more loosely, at 2(s)) and the plane spanned by 7(s),
M(s) is called the osculating plane there. The latter has higher contact
with a than do other planes; see 2.35 below.

For the purpose of calculation let us drop the s. We know MN is perpen-

dicular to N (see 2.6(3)), so NV =4T+1B for suitable 7, 7 eR (depending
on s, of course). But T: N=0 so T-N+T- N'=0, which gives k+4=0.
Hence

N(s)= — K(s)T(s) + 7(5)B(s),

for some real number 1(s). This number r(s) is called the torsion of the
curve at s, So long as k(s) +0, t is a smooth function of s.

Next, B' is perpendicular to B, say B' =uT+vXN for u, v eR. Using

B- T=0 we have B-T+8B- T'=0, so 4+0=0. Using B- N=0 we have
similarly v=—r. Hence

B(s)= — (5) Ns).

These three formulae, for 7”, N, 3, are the formulae we need.
Space curves 37

2.34 Distance-squared functions on a unit speed space curve

Let a: 7+>R* be unit speed, with k(s) never zero, and let u eR3.
Define f: 7>R by f(5)=|a(s) — ul? =(x(s)— 14): (a(5) —1). Dropping most
of the s's we have

3f()=(0-)-T

35 (5)=1+(4—u) EN

3 (5)=(4—u) KN +(2—1) K(—KT+1B)

Hence

f(5)=0 $ uis in the normal plane at s

f=f(5)=0 S u=a-K IN-+uB (at 5) for some y ER

f=f()=f"(5)=0 > y £0 and u=2+ N/K—(x'/K?1)B (at s)

or T=x =0 (at 5).

Now the distance-squared function from u measures contact with the

sphere, centre 1, through a given point of the curve, since it differs by a
constant from the function y of 2.9, where Fis given by 2.7. Compare the
plane case, 2.13. Thus the last equation above for u shows that, provided
K(s) F0 and 7(s) £0, there is a unique sphere having at least 4-point contact
with a for the parameter value s. This sphere is called the sphere of curvature
atsandits centre u the centre of spherical curvature at s.

2.35 Height functions on unit speed space curves.

With 2 as in 2.34, and a unit vector in R*, define f: 7>R by

fs) =a(s) 4. (Thus f(s) is the distance of «(s) from the plane through the
origin perpendicular to u.) We find:

f(=0 SH uwisa vector in the normal plane at s

f1=f"(5)=0 S u= + B(s)

fO=f()=f"()=0 S u= + Bs), 1(5)=0

f)=:::- =f)=0 $ u= + B(3), (5)=7(5)=0.

The verification of these is left as an exercise. It follows that the osculating

plane at s is the unique plane having at least 3-point contact with x for
that value of s, and that it has higher contact if 7(s) =0.

2.36 Exercises

(1) The vector product of a=(a, a, az) and b=(b,, ba, b3) ER? is axb
=(a:b;—azb,, azb,—arb3, ajb:—a;b) Show that B(s)=T(s)x NM)
(in the notation of 2.33).

(2) For the unit speed helix a(s)=(1/2Xcos s, sin s, 5), show that x(s)= 1/72
for all s. Find expressions for T(s), Ms) and show Bis) =(1//24sin s,
—cos s, 1) for all s. Find 7(s). Find the centre of spherical curvature at
S.
38

2.37

Curves, and functions on them

(3) Let f be the height function of 2.35. Show that x(s)=0 if and only if
[(5)=/(5)=0forallwin the normal plane at 5, Le. all w satisfying T(s) u=0.

(4) Let a: />R* be unit speed, with 2(0)=(0, 0, 0), 7(0)=(1, 0,0), N(0)=(0, 1,0).
Show that B(0)=(0, 0, 1) and

(0) =(0, x, 0) (x, 7 etc. are all at s=0)

a"(0)=(—x, K, KT)
AO) =(— 3KK, K” —-KI- KT, TIKT).

(5) Let a: 7>R? be unit speed, and so € 7. Define y: 1>R? by y(s) =(als)- T(s0),
als) N(so). (This is the plane curve obtained by projecting a to the osculat-
ing plane at so. Note that y need not be unit speed.) Show that, restrict-
ing y to a suitably small open interval containing so, it is a regular curve,
and that its curvature at so equals the curvature of % at so. Try projecting
to the planes spanned by 7(s0), Bso) or by N(so), Bso).

(6) Show, after the manner of 2.28(5), that curvature and torsion of space
curves are unaffected by rotations and translations of R*.

(7) Show that the difference between the length of a small arc of a space
curve and the length of the corresponding chord is of order at least s*
(parametrizing by arc-length 5). (One expects at least s”.)

(8) Let x be a unit speed space curve; write

as) =ay(s)T(s) + b.(5) NES) + c.(5)B(s)
for smooth functions a, b, c. Prove the reduction formulae
An+1=0)—Kby
Pus 1 =Dn+KAn— Tm

,
Cn+1=Ch+ Te

Project

There is a topic dear to the hearts of most authors of books on curves

which we have no! mentioned at all. To what extent does the curvature
function determine a plane curve, or the curvature and torsion functions
determine a space curve? Since the curvature of a unit speed plane curve a
has to do with the second derivative a” we should expect that x is fixed
by the curvature once we are given a point a(1,) and the tangent a(1,) there.
The same conclusion is suggested by the interpretation x(s)=d1//ds (see
2.24) where Y is the tangent angle. More geometrically we have 6 =xós so
one expects that, starting with a point and a tangent line, the curve can be
constructed 'step by step” from the function x (fig. 2.23.). It is important to
realize that xk must be given as a function of arc-length — or, to put it another
way, given a function +: />R, the curve constructed from x will have unit
speed.

Prove for yourselí, or look up in for example Thorpe (1979) or DoCarmo

(1976) the following facts.

(1) Let x: 7>R? and 2;: /->R? be unit speed, with 2 (10)=%2(t0),
Ai (19) =23(10) for some 19 € Tand k1(1)= (1) for all 7 € 7 (kx, =curvature of
%,) Then 21(1)=2:(1) for allt e 7.
Space curves 39

Fig. 2.23. Construction of the curve from x

ós
6

(2) Let 2,: /+R?* and a,: />R?* be unit speed, with 2(10)=%2(t0),
(0) =5(10), Xi (10) =3(15)£0 for some r € 7 and kx (1)=K,(), 71(4)=7,(1)
for all r e 7. Then 2 (1) =2:(1) for all r e 7.

What do these tell you about a plane curve with x(1)=0 for all 1? With
K(1) constant? A space curve with 1(1)=0 for all 7?