You are on page 1of 46

GEOMETRY

D.Barden

January 9, 2017
2
Chapter 1

Spherical, Euclidean,
Conformal Geometry

1.1 Books and background


Rees is a useful book. However Wilson was written explicitly for this course
and it is well worth having either a personal version or making sure sure you
can get hold of the library copy - but let the others know you have it and give
them access please. You may also find it profitable to dip into various texts as
the need arises.

1.2 Introduction
Familiar geometry involves the measurement of distances and angles. The for-
mer requires a metric and the latter an inner product. Since an inner product
determines a norm and thence a metric, we shall generally be working in spaces
equipped with inner products. Usually this will be Rn with its standard inner
product.
At a more sophisticated level we have a metric space with distance function
d, and we look first for the isometries of the space, that is the mappings f such
that for all x, y, d(f (x), f (y)) = d(x, y). If the distance function is determined
by a norm then this implies that the norm is preserved by the mapping, and
if it is induced by an inner product, then that inner product is also preserved.
This means that isometries preserve not only distance but also, when they are
defined, angles.
More generally geometry may be defined as the study of the properties which
are invariant under any given group G of transformations of a set X. It might
or might not then be possible to endow X with a metric for which all the
transformations of G are isometries and, even if that is possible, it is unlikely
that G would be the group of all the isometries of X. In this course we shall
first study three metric geometries and one non-metric one, before introducing
the general concept of Riemannian metrics.

3
4 CHAPTER 1. SPHERICAL, EUCLIDEAN, CONFORMAL GEOMETRY

1.3 Spherical Geometry


This is geometry which takes place on the surface of a sphere. In our case
we confine attention to the 2-sphere, which is the surface of a ball in 3-space.
The radius of the sphere does not affect the geometric considerations so, for
convenience, we take the unit sphere, that is, the set of points of norm 1. The
inner product we use is simply that inherited from R3 .
As you have been told (I am not sure whether the course gets round to
proving it) the shortest distance along the surface of a sphere between two of its
points is along the shorter of the two great circular arcs joining them. A great
circle is the intersection with the sphere of a plane through the centre of the
sphere. Thus great circular arcs are the analogue for the sphere of straight lines
in the plane. So the analogues on the sphere of triangles, or other polygons, in
the plane would be the subsets delineated by great circular arcs. Unless stated
otherwise, when a triangle is specified by its vertices we take the sides to be
the shorter great circular arcs joining the pairs of vertices. Similarly we regard
the area delineated by the triangle (briefly its area) as the smaller of the two
possibilities.
Convention. For the spherical triangle = ABC it will be convenient to
denote by A, B, C the unit vectors in R3 which represent the vertices; by , ,
the angles at these vertices and by a, b, c the lengths of the sides opposite to
them.
Alternatively we may denote the vertices simply by x, y and z with the three
respective opposite edges having lengths a, b and c, etc. The angle between
the two edges of lengths b and c meeting at x is, of course, the angle between
the planes 0xz and 0xy in R3 .
Since we are working on the standard sphere of unit radius and centre the
origin in R3 the spherical distance, d, is given by

d(x, y) = cos1 (h x, y i),

where h x, y i denotes the standard inner product between x and y and the angle
is chosen in the range [0, ] in order to ensure that we take the shorter great
circular arc.

Throughout these notes, when there is the possibil-


ity of more than one metric we shall indicate which
is relevant by a suitable prefix.
Thus, in the present case, we shall denote the spherical distance by S-distance
and the Euclidean distance, which is also relevant on the sphere, by E-distance.
Definition A pole of the spherical line ` is either of the points that is S-distant
/2 from all the points of `.
In other words, they are what would be the poles if ` were regarded as the
equator. Given an S-triangle, the appropriate choice of three poles leads to the
following definition:
Definition Given the spherical triangle = ABC, the (spherical) triangle
0 = A0 B 0 C 0 defined by unit vectors in the directions B C, C A and A B,
respectively, is called the polar triangle of A.
1.3. SPHERICAL GEOMETRY 5

Note that these unit vectors are B C/ sin a, etc. Thus, for this definition,
the pole chosen for the vertex determined by A is where the directed normal to
BOC (oriented from B to C) meets S 2 .
Definition Let P be a point on the unit sphere S. For fixed r with 0 < r < ,
the spherical circle with centre P and (spherical) radius r is the set of
points {Q S | dS (P, Q) = r}.
Definition Spherical triangles 1 , 2 are said to be congruent if there is an
isometry of S (i.e. a bijection preserving spherical distances) taking 1 onto
2 .
Remark A similar definition applies in any metric space.

1.3.1 The cosine rule for spherical triangles


For a spherical triangle with vertices at x, y and z, and other notation as above,
we have h x z, x y i = sin b sin c cos , if you think about it, and expanding
the l.h.s. as a scalar triple product leads to the equation

cos a = cos b cos c + sin b sin c cos .

There is a second, dual cosine formula which you will meet in the examples:

cos + cos cos = sin sin cos a.

There is also a sine formula for spherical triangles which states that
sin a sin b sin c
= = .
sin sin sin

1.3.2 The area of spherical triangles


The surface of the unit sphere has area 4 and two central planes meeting at
an angle divide that surface into four lunes, two of angle and two of angle
. A lune of angle has area 2 , since it is /2 of the total area, and
the fact that the triangle and its antipodal triangle lie in each pair of lunes of
angle , and and all other points of the surface lie in just one of them leads
to the identity:

the area of the spherical triangle with angles , and is + + .

This is the Gauss-Bonnet theorem for the sphere. It is a special, indeed very
special, case of a much more general theorem.

1.3.3 The Euler Number for the Sphere


If we divide the surface of the sphere into spherical polygons, count the number
of polygons P , the number of (great circular arc) edges E and the number of
vertices where they intersect V , then we shall always find that P E + V = 2.
The number 2 is called the Euler number of the sphere and, once again, it is
a very special case of an important invariant.
The proof of this result is relatively easy for a triangulation that is, when
all the polygons are triangles. We then note that adding an internal edge join-
ing two existing vertices to subdivide a polygon also increases the number of
6 CHAPTER 1. SPHERICAL, EUCLIDEAN, CONFORMAL GEOMETRY

polygons by one, without changing the number of vertices. Hence it leaves the
Euler number constant.
For the triangulation case note that, counting face-edge incidences, 3P = 2E.
Then, if the ith triangle has angles i , i , i and area i , summing over all the
triangles we get
XP XP
(i + i + i ) = i = 4.
i=1 i=1
PP
But i=1 (i + i + i ) = 2V since the sum of the angles at each vertex is 2.
So 2V = (P + 4) and 4 = 2V P = 2V + 2P 3P = 2(P E + V ).

1.3.4 Some Questions


You are now able at least to understand the questions on spherical geometry.
However, in order to answer them, since the spherical geometry is induced by
the inner product geometry of R3 , you might also need some knowledge of the
isometries of R3 which preserve the unit sphere. That is, the orthogonal group
O(3) which also plays an important role in the next section.

1.4 Euclidean Geometry


This is the geometry of Euclidean n-space by which we mean Rn with its stan-
1
P 2
dard inner product h x, y i = i xi yi , the induced norm kxk = h x, x i / and
the induced metric
n
X 12
d(x, y) = kx yk = (xi yi )2 / .
i=1

To remind us of the extra structure we often denote the Euclidean n-space by


En rather than Rn . In fact we shall pay most attention to E2 and E3 .

1.4.1 Transformation Groups


Recall, from the Groups course, the facts about transformation groups: or-
bits, stabilisers (also now to be called isotropy groups) and the orbit-stabiliser
theorem, which arises from the fact that points in the orbit of x are bijec-
tive with the cosets of the stabiliser of x. Recall too that if y = g(x) then
stab(y) = (stab(x))g = g.stab(x).g 1 .

1.4.2 Isometries of Euclidean space


The basic result is that any isometry or rigid motion g of En is of the form

g : x 7 Ax + b

where x is the column vector of standard coordinates of a point X in En , A is


an orthogonal matrix (so that AAT = I and x 7 Ax determines an orthogonal
linear transformation X 7 (X) of Rn ) and b is the column vector of standard
coordinates of a point B in En . Thus B = g(0).
1.4. EUCLIDEAN GEOMETRY 7

Recall that a matrix A is orthogonal if, and only if, its columns form an
ortho-normal basis of En . The set of all orthogonal matrices forms the orthog-
onal group O(n). An orthogonal matrix has determinant +1 or 1, those with
determinant +1 form the subgroup SO(n) of special orthogonal transfor-
mations. In E2 and E3 all the special orthogonal transformations are rotations
(and conversely). In E2 the remaining orthogonal transformations are reflections
and in E3 they are either reflections or reflected rotations, the composite of
a reflection and a rotation (in either order).

1.4.3 The isometries of E2


Recall, again from Algebra and Geometry - or at least my presentation of it -
that an isometry of E2 , once it is identified with C1 the complex plane, can be
written as
z 7 az + b, or z 7 az + b, with mod a = 1.

Then the map z 7 az is the rotation through the angle arg a and z 7 az is the
reflection in the line at angle (arg a)/2 to the x-axis.
We may classify the isometries of E2 as follows:

(1) Direct isometries, that is those with det(A) = 1 or of the form z 7 az + b.

(a) Translations Tb : z 7 z + b
(b) Rotations Rot(a, ) about a point a through the angle . Writing
Rot() for Rot(0, ), we find that

Rot(a, ) = Ta Rot()(Ta )1 = (Rot())Ta : x 7 Rot()(xa)+a.

(2) Indirect, or orientation reversing, isometries, that is those with det(A) =


1 or of the form z 7 az + b.

(a) Reflections Refl (a, ) in a line through a point a in the plane at


angle to the x-axis. These are the transformations
 of the form
cos 2 sin 2
x 7 A(x a) + a where A is the matrix or of
sin 2 cos 2
the form z 7 (z a) + a where = exp(2i ). Again Refl (a, ) =
(Refl (0, ))Ta . The line of reflection is of course determined by the
eigenvector of A with eigenvalue +1: it is the line through a parallel
to this eigenvector.
(b) Glide reflections, where a glide reflection is a reflection followed by
a translation parallel to the line of reflection (the glide). That is, in
the above notation, Tb where b is a multiple of the eigenvector of A
with eigenvalue +1.

Remark. The fact that 2(a) and 2(b) are the only orientation reversing isome-
tries means that a reflection followed by a translation which is not parallel to
its invariant line must be re-expressible in terms of a different glide reflection.
You should be able to establish which.
8 CHAPTER 1. SPHERICAL, EUCLIDEAN, CONFORMAL GEOMETRY

1.4.4 The isometries of E3


We may classify the isometries of E3 similarly to those of E2 as follows:
(1) Direct isometries, that is those with det(A) = 1 or with A a rotation.
These all take the form of a screw. That is a rotation about some line,
followed by a translation along that line.
(2) Indirect, or orientation reversing, isometries, that is those with det(A) =
1, are of the following types:
(a) Reflections: a reflection may be in an affine plane not just in linear
subspaces.
(b) Rotated reflections: a (non-zero) rotation about some axis, fol-
lowed by a reflection in some plane not containing this axis.
(c) Glide reflections: reflection in some plane followed by (non-zero)
translation parallel to that plane.

1.4.5 The Platonic Solids


A regular polyhedron in R3 or Platonic solid is a convex 3-dimensional
subset bounded by 2-dimensional faces such that two faces are either disjoint
or meet in a common vertex or edge. We require the symmetry group to be
transitive on each of the sets of vertices, edges and faces, from which it follows
that each face is a regular n-gon, the same n for all faces, and that each vertex
lies on the same number of edges and the same number of faces.
Suppose each vertex has r regular n-gons meeting at it, where necessarily
n 3 and r 3. Then, since the interior angle of a regular n-gon is (n 2)/n
we must have r(n 2)/n < 2, so that r(n 2) < 2n. Here we have < rather
than since the surface will not be flat at a vertex.
It is then easily checked that the only possibilities are
(i) The regular tetrahedron with three equilateral triangles meeting at each
vertex.
(iia) The cube with three squares meeting at each vertex.
(iid) The regular octahedron with four equilateral triangles meeting at each
vertex.
(iiia) The regular dodecahedron with three regular pentagons meeting at each
vertex.
(iiid) The regular icosahedron with five equilateral triangles meeting at each
vertex.
These have, respectively, 4,6,8,12,20 faces, as indicated by their names - ex-
cept that we do not usually call the cube a regular hexahedron. They have,
respectively, 6,12,12,30,30 edges and 4,8,6,20,12 vertices. In fact there is a du-
ality obtained by starting from one solid, putting a vertex in the centre of
each face, joining these new vertices to form new edges whenever they are the
centres of contiguous faces of the original solid and spanning these new edges
by a new face whenever they correspond to consecutive pairs of faces around a
common vertex in the original solid. This duality maps (i)(i), (iia)(iid)
and (iiia)(iiid).
1.4. EUCLIDEAN GEOMETRY 9

1.4.6 Finite subgroups of SO(3)


Recall, and if necessary reprove, that any subgroup G of O(3) is either contained
in SO(3) or has GSO(3) a subgroup of index 2, called the rotation subgroup.
The full symmetry groups, and rotation subgroups of the platonic solids,
when their centroids are placed at the origin, are as follows. By duality (iia)
and (iid) have the same groups as do (iiia) and (iiid).

(i) S4 and A4 acting on the vertices or, by duality, on the faces.


(ii) S4 C2 and S4 acting on the body diagonals of the cube.
(iii) A5 C2 and A5 acting on the five inscribed cubes to the dodecahedron.

For a plane regular n-gon centred on the origin in R3 all its symmetries are
rotations and form a subgroup isomorphic with the dihedral group D2n . It is
then a fact that the only finite subgroups of SO(3) are (isomorphic with) the
(cyclic or dihedral) subgroups of the dihedral groups, or subgroups of A4 or S5 .
This takes some proving and is not explicitly mentioned in the syllabus, so you
are unlikely to be asked to prove it, even if the lecturer does. However you
should then observe the techniques used since a clever examiner could easily
embed them in a Tripos question.

1.4.7 Reflection Groups


I take this to mean groups generated by an explicit (usually finite) set of reflec-
tions, since that is analogous to general terminology common in the classification
of (again usually finite) groups. The facts are that any orthogonal transforma-
tion of En is a product of at most n reflections and a product of two reflections
is a rotation. (You should remember how to prove these facts.) So no subgroup
can be composed entirely of reflections. Again we concentrate on E2 and E3 .
Given a set S of reflections in lines l for E2 or planes for E3 (here
we mean affine lines or planes i.e. those not necessarily passing through the
origin), let W = W (S) be the subgroup of isometries generated by S. That is,
every w W is a composite w = 1 . . . k with each i in S and not necessarily
distinct from j when j 6= i.
Note first that the product of two reflections whose lines or planes of reflec-
tion meet at an angle is a rotation through the angle 2 . So this can only
have finite order if is a rational multiple of .
The main result for our purposes is the following which I just state for E3
using the notation introduced above. If S is finite and each pair of planes i , j
meet at an angle 2/k(ij) with k(ij) N or k(ij) = 0, meaning that the planes
are parallel; if some point p lies on no plane; if
\
C = { the component of the complement of i containing p}
i

and if w W, w 6= 1 then w(C) C = . Note that this result is not intended


to be obvious: the proof that I have occupies, with preliminary lemmas, over
three pages of notes.
This means that the images of C under all the elements of W tesselate three
space since, by conjugation, the group W contains the reflection in any image
10 CHAPTER 1. SPHERICAL, EUCLIDEAN, CONFORMAL GEOMETRY

under an element of W of a boundary face of C. Hence the union of the images


of C cannot be bounded.
Turning to the case of three elements in S the three planes of reflection will
meet in a common point, which we may take to be the origin, and the set C will
be a cone with vertex at the origin. Then, assuming the three angles between
pairs of planes are = /p, = /q and = /r, the cone C will meet the
unit sphere in a spherical triangle whose angles are , and . So, since the
area of this triangle is + + > 0 we must have
1 1 1
+ + > 1.
p q r

With p q r this inequality has only the solutions

(p, q, r) = (2, 2, n), (2, 3, 3), (2, 3, 4), (2, 3, 5).

The intersections with the unit sphere of C and its images under the elements
of W determine a triangulation of the sphere and, if we replace these spherical
triangles by the linear ones spanning the same vertices, we obtain the surface
of solid bodies which are respectively a double cone on a regular n gon, a
regular tetrahedron, a regular octohedron and a regular icosohedron having,
as we have seen, the full groups of symmetries D2n , S4 , S4 C2 and A5 C2
respectively. However each element of W is an isometry of the relevant solid
and it is not difficult to show that W is in fact the full group of its symmetries.
Hence we have in each case found a set of three reflections which generate the
group. Thus they are all three generator groups. This is true of most simple
groups or groups closely related to simple groups as are the above.

1.4.8 Other Topics


The lecturer might mention crystallographic groups. Note that the subgroup
T of all translations is a normal subgroup of the group of all isometries of
E3 , since (conjugacy principle) the conjugate of a translation is a translation
(invariant lines). A group of isometries is called crystallographic if L = T
is a lattice, that is, a subgroup of the form Zv1 + Zv2 + + Zvn where v1 vn
is a basis of R3 . (We here identify a translation with the vector by which it
translates.)
Note that the map from the group of isometries of En onto O(n) has kernel T
and its restriction to has kernel L. Moreover the orthogonal transformations
in the image must preserve L. This makes the image rather small and so enables
us to classify the crystallographic groups. See Burn chapter 23 for the case n = 2
leading to the so-called wall-paper groups. In this case the possible images
in O(2) are (conjugate and hence isomorphic to) the dihedral groups of orders
2, 4, 6, 8 or 12 or their cyclic subgroups.

1.5 Conformal Geometry


This is the (non-metric) geometry of a set X on which angles between curves
are defined and we consider the group G of conformal transformations of X,
that is transformations g such that if two curves 1 and 2 meet at P X at
1.5. CONFORMAL GEOMETRY 11

an angle then the image curves g 1 and g 2 meet at g(P ) at the same angle
. [g is the curve given by (g )(t) = g( (t)).]
We have already studied one conformal geometry: that of the complex plane.
More precisely we showed that Mobius transformations may be interpreted as
bijective transformations of the Riemann sphere, which is the complex plane
together with , with behaviour near z = interpreted as behaviour near 0
of = 1/z. Then not only are the Mobius transformations conformal on the
entire Riemann sphere but, although we have not proved it, they are the only
such transformations.
To understand this better, let
x + iy
N : S C {}; (x, y, z) 7
1z
denote the stereographic projection from the point N = (0, 0, 1), where the
centre of the unit sphere S is at the origin of C and let S : S C {}
denote the stereographic projection from the point S = (0, 0, 1), So that,
in particular, N (N ) = = S (S). Then if, for P S, N (P ) = z we
find that S (P ) = 1/z. Thus, if we think of our analysis as really taking
place on the sphere S but interpreted via the co-ordinates z(P ) = N (P ) or
(P ) = S (P ), then the relation between the two possible co-ordinate systems
is precisely z(P ) = 1/ (P ) whenever both are defined. That is, everywhere
except at the two points (0, 0, 1). We shall study the projection = N in
more detail in the exercises below. In spherical polar coordinates it takes the
form N (1, , ) = ei cot( /2).
12 CHAPTER 1. SPHERICAL, EUCLIDEAN, CONFORMAL GEOMETRY

1.6 Questions
1.6.1 Spherical Geometry
(1) Show that a spherical circle is a Euclidean circle and find the relation
between its spherical and Euclidean radii. Find a formula for the area of
the interior of a spherical circle, in terms of its spherical radius r.

(2) Given a spherical line l and a point P , not lying on l, on a sphere S, show
that there is a spherical line l0 passing though P and intersecting l at
right-angles. Prove that, as Q varies on l, the minimum distance d(P, Q)
is attained at one of the two points of intersection of l and l0 . Prove also
that if this minimum distance is less than /2 then l0 is unique.

(3) Prove that the polar triangle of the polar triangle of the spherical triangle
= ABC is itself.

(4) Prove that the sides and angles of the polar triangle of = ABC are
, , and a, b, c respectively. Deduce that

sin sin cos c = cos + cos cos .

(5) Prove that the group of isometries of the unit sphere (i.e. the ones that
preserve spherical distances) is isomorphic with the group of isometries of
R3 (those that preserve Euclidean distances) that fix the origin.

(6) Show that two spherical triangles are congruent if and only if they have
equal angles. What other conditions for congruence can you find?

(7) Show that in a spherical triangle = ABC, b = c if and only if =


. Show also that this occurs if and only if the spherical line AP is
perpendicular to BC, where P is the midpoint of BC.

(8) Show that, for any three positive numbers a, b and c less than , if the
spherical triangle = ABC has sides of length a, b, c then a + b + c < 2,
and also that a < b + c, b < c + a and c < a + b. Show conversely
that, for any three positive numbers a, b and c less than satisfying these
conditions, we have cos(b + c) < cos a < cos(b c), and that there is a
spherical triangle, unique up to congruence, with sides of lengths a, b and
c.

1.6.2 Euclidean Geometry


(9) Suppose that a Euclidean triangle in R2 has angles /p, /q and /r.
Find all possible values of p, q and r which are integers 2. In each case
draw a diagram illustrating a corresponding tesselation of R2 . Deduce the
existence of a tesselation of the plane by regular hexagons.

(10) Classify (with proof), up to isomorphism, all finite subgroups of O(2, R).
If G is a finite subgroup of GL(2, R), we define a pairing < , > on R2 by
X
< x, y >= (gx , gy),
gG
1.6. QUESTIONS 13

where ( , ) is the standard inner product on R2 . Show that < , > is also an
inner product on R2 . Hence, or otherwise, classify up to isomorphism the
finite subgroups of GL(2, R). Classify all the finite subgroups of GL(2, Z).

(11) Find the Euler number of the flat torus, that is the topological quotient
T of R2 by the integer lattice Z2 . Denote by Fn the number of faces with
precisely n edges, and by Vm the number of vertices where precisely m
edges meet. Assuming that each face has at least three edges, and at least
three edges meet at each vertex, exhibit a polygonal decomposition of T
with V3 = F3 = 0. Show that for the sphere V3 + F3 > 0.

1.6.3 Conformal Geometry


(12) Show that the group of rotations of the unit sphere S is isomorphic to
the group of Mobius transformations
  corresponding to the group SU (n)
a b
of matrices of the form where a, b C and |a|2 + |b|2 = 1.
b a

(13) Show that the projection of the unit sphere minus the north pole onto
the complex plane is a conformal map with the set of spherical circles on
the sphere mapping bijectively to the set of circles and straight lines in C.
(14) If points P, Q S correspond under to points u, v C, and d de-
notes the spherical distance between P and Q on S show that a certain
cross-ratio (to be specified) of the points u, v, 1/u, 1/v is equal to
tan2 (d/2).
(15) Show that the Mobius transformation corresponding, under , to a rota-
tion of the sphere S has exactly two fixed points zo and 1/z0 . Show
however that, if T is a Mobius transformation having fixed points z0 and
z0 then, either T corresponds to a rotation or one of the fixed points is
attractive.
14 CHAPTER 1. SPHERICAL, EUCLIDEAN, CONFORMAL GEOMETRY

1.6.4 Notes
(4) The relation between a spherical triangle and its polar triangle obtained
in question (3) is an example of a duality which is a common feature of
all geometries. It is a very powerful tool. Use it to the full in this and
subsequent questions.

(7) Use (6).

(8) Dont use the triangle inequality.

(10) Last part: consider trace and determinant.

(11) You may assume the polygons P on T are


P the bijective image of convex
polygons in R2 . Note that nFn = mVm = 2E for any polygonal
decomposition. Deduce that, on the given assumptions, V3 = 0 implies
that E > 2V with a similar consequence for F3 = 0.

(12) onwards. Recall the facts about Mobius transformations - yet again! You
will also find the conjugacy principle a good guide. For (12) use the fact,
which you should prove, that rotations of the sphere are generated by
all the rotations about one axis together with just one rotation about
a perpendicular axis through the angle /2. Note, and prove, also that
matrices of the type specified form a subgroup SU2 (C) of the group SL2 (C)
which maps 2-1 onto the Mobius group.

(14) Of course you know (and can prove!) what Mobius transformations do
to cross-ratios. This allows you to rotate the sphere to simplify the cal-
culation. I prefer the cross-ratio {z1 , z2 ; z3 , z4 } = T (z4 ) where T is the
(unique) Mobius transformation taking z1 , z2 , z3 to , 0, 1 respectively.
However you may choose your own, so long as you name it.

(15) Once again conjugation by a rotation will simplify matters. An attractive


fixed point is z0 such that, if z is not fixed, then T n (z) z0 as n .

1.7 Further Questions


These are questions that have appeared on past example sheets, but for which
I do not really have room - nor you probably the necessary time to answer. If
you do look at them we could discuss them in the supervision or, more likely,
in the surgery. The current 11 replaces numbers 2 and 3 while 2 and 3 in turn
replaced something akin to 11 and 4! Life is cyclic.
n+1
(1) In
PR with coordinates x0 , . . . , xn take the copy M of Rn defined by
xi = 0. Let standard basis vectors in Rn+1 and put i = 1, . . . , n. Let
i : M M be the in M that is orthogonal to ui . Show that hyperplanes
Hi are of the form 2/k for diagrams to indicate what happens in M for
the What group is generated by the i ?

(2) Let P, P 0 , P 00 be three regular pentagons with sides of length 1.

(a) Show that there is a unique way, up to isometry, of glueing them together
along common edges A00 , A, A0 , so that they share a vertex x.
1.7. FURTHER QUESTIONS 15

(b) In the resulting configuration,show that there exist vertices y P, y 0


P 0 , y 00 P 00 such that xy, xy0 , xy 00 form an orthogonal triple of vectors in
R3 each having length = ( 5 + 1)/2.

(c) Starting with a cube in R3 with sides of length , how many ways can
you glue the configuration obtained in (a) at a given corner x so that
xy, xy 0 , xy 00 form the edges of the cube incident with x. Deduce the
existence of the regular dodecahedron D.

(3) Given a regular dodecahedron D, show that there are five cubes in D all of
whose edges are diagonals of faces of D. Deduce that the rotation group
of D is isomorphic to the alternating group A5 and that the full symmetry
group of D is isomorphic to the direct product A5 C2 where C2 is the
cyclic group of order 2.
(4) (a) With the notation of Question 11, given a polygonal decomposition
of S 2 into convex spherical polygons, prove the identity
X X
(6 n)Fn = 12 + 2 (m 3)Vm
n m

(b) Assuming, as before, that V1 = V2 = F1 = F2 = 0, deduce that


3F3 + 2F4 + F5 > 12.
(c) Show how to construct (the surface of) a football composed of (con-
vex) spherical hexagons and pentagons, with precisely three faces
meeting at each vertex.

1.7.1 Comments
2(a) How does this relate to the existence of a spherical triangle with sides
3/5?
(b) Note that xy and xy 0 are parallel to certain edges of F and F 0 respectively.
Note also that two non-concurrent lines in R3 are orthogonal if, and only
if, one (and so each)) lies in a plane orthogonal to the other.
(c) Use the reflective symmetries of the cube which interchange two faces.
4(c) How many pentagons are there?
16 CHAPTER 1. SPHERICAL, EUCLIDEAN, CONFORMAL GEOMETRY
Chapter 2

Hyperbolic Geometry

2.1 Introduction
Euclidean geometry is the geometry of a space with zero curvature and spher-
ical geometry is that of a space with positive curvature. These two statements
correspond to the fact that the sum of the angles in a triangle is equal to and
exceeds respectively. Although we have only looked at the two dimensional
case, there are spaces of positive curvature of all dimensions, indeed the gener-
alised spheres are such: the standard n-dimensional sphere is the set of points
of unit modulus in Rn+1 . It has curvature +1 at each of its points. There are
also spaces of negative curvature in which the sum of the angles in a triangle is
less than . As for the positive case, we shall only consider the two-dimensional
space of constant curvature 1. This is generally called the hyperbolic plane
and its geometry is called hyperbolic geometry.
Unlike the Euclidean plane and the unit sphere, the hyperbolic plane has no
isometric embedding in Euclidean R3 : any attempt to embed it distorts the
distances between its points, in the same way that any attempt to represent
part of the surface of a sphere in the plane necessarily distorts the distances
between its points. It is possible to embed it if we replace the standard inner
(Euclidean) product on R3 by the indefinite but still non-degenerate bilinear
form
x1 y1
(x, y) = x2 , y2 = x1 y1 x2 y2 x3 y3 .
x3 y3
Then the hyperbolic plane is the upper sheet, x1 > 0, of the hyperboloid (hence
the terminology!) (x, x) = 1 with the metric d(x, y) = cosh1 ((x, y)). Note
the analogy with the definition of the unit sphere and indeed, using it, we
may develop a hyperbolic trigonometry analogous to the familiar spherical
trigonometry. However, as this requires some projective geometry, in this course
we shall emphasize two other models of hyperbolic space.
If we project a 2-sphere less its north pole onto its equatorial plane then we
induce a new distance function on the plane by defining the distance between
two points in the plane to be the spherical distance between the points which
project to them. [Of course one needs to check that this is a metric. However
for the purposes of this introduction you may take that for granted.] Similarly

17
18 CHAPTER 2. HYPERBOLIC GEOMETRY

one may represent the hyperbolic plane by suitably distorted metrics on suitable
subsets of the plane: namely the upper half-plane and the unit disc. However
before we describe these models it will be helpful to discuss the differential form
of a metric. This is the generalised metric mentioned in the introduction to
Chapter 1 that will be discussed in more detail in Chapter 3.

2.2 Riemannian Metrics


If the (variable) quadratic form a(x, y)u2 + 2b(x, y)uv + c(x, y)v 2 is positive
definite (in u, v) for all (x, y) in some domain U in R2 , [ i.e. a(x, y) > 0 and
a(x, y)c(x, y) b(x, y)2 > 0 for all (x, y) U ,] then the associated differential
form
g(x, y) = a(x, y)dx2 + 2b(x, y)dxdy + c(x, y)dy 2
is called a Riemannian metric on U . It is conventional to write g(x, y) =
ds2 formally although strictly ds does not exist except as a mnemonic for the
following application.
Recall that if (t) = (x(t), y(t)) is a curve with image in U and ds =
a(x, y)dx + b(x, y)dy is a one-form on U then we define
Z Z
ds to be {a(x(t), y(t))x0 (t) + b(x(t), y(t))y 0 (t)}dt.

Then, when ds2 is a Riemannian metric on U as above, and : [t1 , t2 ] U


is a curve
R in U we define the length of (with respect to the metric g) to be
Lg [ ] = ds which in turn we define to be
Z t2 p
a(x(t), y(t))(x0 (t))2 + 2b(x(t), y(t))x0 (t)y 0 (t) + c(x(t), y(t))(y 0 (t))2 dt.
t1

Note that the integrand is real since we assumed the quadratic form associated
with the Riemannian metric is positive definite everywhere on U . Note also that
we have implicitly assumed the curve is differentiable, i.e. that x0 (t) and y 0 (t)
exist for all t. However piece-wise differentiability will do provided the curve is
also continuous, and that is all we shall assume.
Then for points p, q in U we may define d(p, q) to be the infimum over all
such (continuous and piece-wise differentiable) curves : [t1 , t2 ] U such that
(t1 ) = p and (t2 ) = q of the lengths Lg [ ]. This infimum certainly exists as the
integrals representing lengths are all positive and we may show that it is a metric
in the familiar sense, that is, it satisfies the triangle inequality and, provided g
is continuous (i.e. its coefficient functions a, b and c are), d(x, y) = 0 x = y.
The infimum may or may not be attained but any curve which does attain it is
called a minimal geodesic from p to q: a geodesic is a curve that is a minimal
geodesic between any two of its points that are sufficiently close. Remark You
will meet other definitions of geodesics, but that one will suit us for now. It has
the advantage of matching our Euclidean intuition.
When we need to distinguish between the two forms of metric we refer to a
Riemannian metric or differential form of the metric in one case and to a distance
function or distance form of the metric in the other. Not all distance functions
arise from Riemannian metrics in the manner just described. However all those
with which we are concerned in this course do. For example the Euclidean
distance function so arises from the Riemannian metric g(x, y) = dx2 + dy 2 .
2.3. RIEMANNIAN METRICS AND SURFACE AREA 19

2.3 Riemannian Metrics and Surface Area


In the next chapter we shall study Riemannian metrics in more detail. However
I shall now cover as much as, or possibly slightly more than, our immediate
needs. If you wish to go further, entering the ground covered by the rest of
this course as well as the Part II Geometry courses, a good text is Elementary
Differential Geometry by Barrett ONeill, published by Academic Press.
The context we consider is that of surfaces, such as spheres, tori, etc., em-
bedded in Euclidean spaces. The theory generalises easily to higher dimensional
surfaces or manifolds, and is also valid for Euclidean spaces of any dimension.
However we shall restrict attention to familiar two-dimensional surfaces and,
when we need to be explicit, to embeddings in R3 .
The defining property of a surface is that it is locally homeomorphic with
Euclidean 2-space, R2 . That means that we can cover it with coordinate neigh-
bourhoods that are homeomorphic with open subsets of R2 . We then use the
natural coordinates in this copy of R2 , to find our way around the surface, just
as we do with maps of our locality on the surface of the earth. The terminology
tends to reflect this analogy.
Thus, given a surface M in R3 , we consider an open subset D of R2 and a
differentiable map x : D R3 such that x(D) M and x is a homeomorphism
of D with x(D) when the latter is given the topology that it inherits as a
subset of M . [This last condition is a technicality needed to avoid certain
potential problems. However you do not need to worry about it since you are
unlikely to wish to use a map that violates it.] The map x is referred to as a
parameterisation of M and its inverse, from the subset X(D) of the surface
to its coordinate space D R2 , as a chart on M .
We now consider the lengths of curves in M R3 . Let : [a, b] M
be such a curve with its image contained in the coordinate neighbourhood
x(D). Since M is covered by coordinate neighbourhooods - it has an atlas of
charts, we can always break a curve up into segments such that each segment
is contained in a coordinate neighbourhood. Then the conditions we placed on
our charts ensure that we can find a curve : [a, b] D such that = x .
Denoting the coordinates in R2 by (u, v) and those in R3 by (x1 , x2 , x3 ) we can
spell this out in painful detail as:

: [a, b] M : t 7 x1 (u(t), v(t)), x2 (u(t), v(t)), x3 (u(t), v(t)) ,

where : [a, b] D : t 7 (u(t), v(t)).


The chart itself gives rise to the obvious coordinate curves: for each point
(u0 , v0 ) of D, there is a u-parameter curve

v0 : t 7 x(t, v0 ),

given by u(t) = t and v(t) = v0 , and a v-parameter curve

u0 : t 7 x(u0 , t),

given by u(t) = u0 and v(t) = t.


The coordinate neighbourhood x(D) is covered by these two families of
curves, which are the images under x of the horizontal and vertical lines in
D, and one curve of each family goes through each point of D, though of course
20 CHAPTER 2. HYPERBOLIC GEOMETRY

we need to restrict the domain [a, b] on which each curve is defined to ensure
that it stays in D and, hence, that its image stays in x(D).
Of course these coordinate curves are curves in R3 and, as such, their length
Rb
is given by a k v0 0 (t)kdt, for example for the u-parameter curves. The tangent
vector, or velocity, v0 0 (t) is given by
x1
u (t, v0 )
v0 0 (t) = xu (t, v0 ) x2 (t,
u v0 ) ,
x3
u (t, v0 )

and k v0 0 (t)k2 is simply the inner product xu (t, v0 ) xu (t, v0 ).


Turning to the general curve above, the tangent vector 0 (t) has coordi-
nates

dxi xi du xi dv
(u(t), v(t)) = (u(t), v(t)) (t) + (v(t), v(t)) (t),
dt u dt u dt

so that
du dv
0 (t) = xu (u(t), v(t)) (t) + xv (u(t), v(t)) (t),
dt dt
and hence

du 2 du dv dv
k0 (t)k2 = E(u(t), v(t))(t) + 2F (u(t), v(t)) (t) (t) + G(u(t), v(t)) (t)2 ,
dt dt dt dt
()
where we have used the conventional notation, for the case of surfaces:

E = xu xu , F = xu xv = xv xu , G = xv xv ,

suppressing now the arguments (u(t), v(t)) throughout, since they would have
caused the line to overflow. If, instead we suppress all mention of the parameter
t, and write 0 (t) = ds
dt then () becomes

ds2 = E(u, v) du2 + 2F (u, v) du dv + G(u, v) dv 2 .

Indeed () is the true meaning of this latter expression, which we note is a Rie-
mannian metric in the sense of the previous section since, by Cauchy-Schwartz
and the fact that xu and xv are never parallel, E(u, v)G(u, v) > F (u, v)2 for
all u and v.
Having thus given some intrinsic meaning to the Riemannian metric we can
now turn to the main purpose of this section. Consider a small coordinate
square in D with opposite vertices (u, v) and (u + h, v + k). The image in M is
of course not a square, but has adjacent edges which are curves of approximate
length x(u+h, v)x(u, v) h x x
u and x(u, v +k)x(u, v) k v . Thus the area
of this image is approximately hkkxu xv k. Summing over all such squares in
D and taking the limit as
R their
size tends to zero we see that the area of x(D)
is given by the integral D EG F 2 dudv, since

kxu xv k2 = kxu k2 kxu k2 |xu xv |2


2.4. TWO PLANAR MODELS 21

2.4 Two planar models


Our two models for hyperbolic geometry are based on the (open) upper half-
plane, that is, excluding the real axis, and the (open) unit disc, that is, excluding
the unit circle. For convenience we think of both these domains as lying in the
complex plane and so write

H = {z | =z > 0} and U = {z | mod z < 1}.

Your lecturer will a.s. (almost surely) choose to develop one of these models
in some detail and leave you to deduce the other. I shall instead give you
the facts for both, largely without proof, and, in the first few of the following
exercises, indicate what you may assume as your starting point. However for
the remaining questions and also, unless clearly indicated otherwise, for Tripos
questions, you are free to choose whichever model leads most efficiently to the
required result. This is indeed one of the beauties of any geometry worthy
of the name - and hyperbolic geometry certainly is: there are always several
different points of view giving very different insights into its nature and the
interplay between these different aspects is a powerful tool for understanding
the geometry. The duality we met in the first chapter is another good example
of this.
The basic facts are as given in table 1, where is the hyperbolic distance in
either model.

Table 2.1: Properties of hyperbolic models

Model Upper Half-plane Unit Disc

Metric in differential form ds = |dz|


=z
2|dz|
ds = 1|z| 2

Metric in distance form tanh( /2) = | zw


zw |
zw
tanh( /2) = | 1z w |
(ip, iq) = | log(p/q)| 1+r
(0, r) = log 1r
Special case
R dxdy R  2 2
Hyperbolic area E y2 E 1|z|2
dxdy
Isometries h : z, z 7 az+b
cz+d g : z, z 7 az+c
cz+a
where a, b, c, d real, ad bc = 1 |a|2 |c|2 = 1
and khk2 = 2 cosh (i, h(i)) kgk2 = 2 cosh (0, g(0))

The distance (0, r) in the unit disc model may also be expressed as 2 tanh1 r.
Recalling that SL(2, R) is the group of 22 matrices with real entries and deter-
minant 1 and PSL(2, R), its quotient by the subgroup {I}, which is isomorphic
with a subgroup of the Mobius group, the full group of orientation preserving
isometries for the upper-half-plane model is isomorphic with PSL(2, R). How-
ever it also includes the orientation reversing reflection in the imaginary axis
given by : z 7 z. The map h : z 7 az+b
cz+d does not commute with the map
azb
. Indeed h : z 7 czd . This means that the full group of isometries is what
is known as a semi-direct product of SL(2, R) by the group < > of order
2, denoted SL(2, R)o < >
22 CHAPTER 2. HYPERBOLIC GEOMETRY

Analogous to SL(2, R) the  group


 U (1, 1; C) is the group of matrices with
a b
complex entries of the form with kak2 kbk2 6= 0. The C is usually
b a
omitted as it is implied by the U for unitary. Then SU (1, 1) is the subgroup
of matrices with determinant 1 and P SU (1, 1) is its quotient by the subgroup
of scalar multiples of the identity matrix. Once again the orientation-reversing
involution (transformation of order two) : z 7 z is an isometry that nor-
malises, but does not centralise P SU (1, 1) so that the full group of isometries
is P SU (1, 1)o < >.
Note that in each model the orientation preserving isometries form the full
group of Mobius transformation that preserve that model space. Moreover in
both models the hyperbolic geodesics or lines are the Euclidean circles and lines
which meet the appropriate (excluded) boundary at right angles. The mobius
transformation z 7 (z i)/(z + i), which maps the upper half plane onto the
unit disc, is in fact an isometry between the two models. The isometries of the
za
unit disc model can also be written in the equivalent form g : z, z 7 ei 1az ,
where a lies in the disc and is real.

2.5 Triangles in the Hyperbolic Plane


As for the spherical and Euclidean geometries there are sine and cosine for-
mulae for triangles in the hyperbolic plane. As in the spherical case we denote
the angles of the triangle by , , and the respective opposite sides by a, b, c.
Analogously to the spherical case, but allowing for the change in sign of the
curvature, the area of the triangle is ( + + ). The sine formula is

sinh a sinh b sinh c


= = ,
sin sin sin

and the (first) cosine formula is

cosh c = cosh a cosh b sinh a sinh b cos .

However, as for the spherical case, there is a second cosine formula obtained
here by interchanging the angles and sides and at the same time interchanging
the trigonometric and hyperbolic functions, as well as changing the sign on one
side of the equation. Thus in both formulae, as in the sign formula, one always
takes trigonometric functions of the angles and hyperbolic functions of the sides:

cos = cos cos + sin sin cosh c,

which we can re-write with cosh c as the subject:


cos cos + cos
cosh c = .
sin sin
Note that, since the sides are positive, this means that the angles of a hyperbolic
triangle determine its sides.
Note the following special case: in a triangle with angles , , /2 we have
cosh a cosh b = cosh c (c.f. Pythagoras!) and cosh b sin = cos . In particular
in a triangle with angles , 0, /2 we get cosh b sin = 1.
2.6. QUESTIONS 23

2.6 Questions
1(i) Use the differential form of the metric of the upper half-plane model of
hyperbolic space to show that the imaginary axis is a geodesic and, for
a, b real and a < b, (ia, ib) = log(b/a).

(ii) Show that the Mobius transformations in PSL(2, R) are isometries of the
upper half-plane model.

(iii) Show that all the orientation preserving isometries of the u.h.p. model
are in PSL(2, R). Identify the geodesics in this model.

2(i) Use the differential form of the metric of the disc model of hyperbolic
space to show that the real axis is a geodesic and, for a, b real and a < b,
(a, b) = log 1a 1+b 1
1+a 1b . In particular show that (0, b) = 2 tanh b.

(ii) Show that the Mobius transformations in PSU(1, 1) are isometries of the
disc model.

(iii) Show that all the orientation preserving isometries of the disc model are
in PSU(1, 1). Identify the geodesics in this model.

3(i) Show that any Mobius transformation taking the upper half-plane to the
unit disc is an isometry between the two models of hyperbolic space.

(ii) Prove that PSL(2, R) and PSU(1, 1) are isomorphic.

(iii) Prove that the hyperbolic distance


between two points u and v in the upper
1 uv
half plane model is 2 tanh uv and in the disc model is 2 tanh1 1uv
uv

(iv) If the H-geodesic from z1 to z2 in the upper half-plane meets the real axis
in z1 and z2 , show that the H-distance between z1 and z2 is log r where r
is a cross ratio of the four points.

4(i) Show that a hyperbolic circle C with H-centre v and H-radius r in the
upper half-plane model is also a Euclidean circle. Let L be the H-geodesic
through v perpendicular to the real axis. Show that L is orthogonal to C
and that there is a unique H-geodesic M that is orthogonal to both C and
L and that this M meets L at the H-centre of C.

(ii) Find the H-area and H-circumference of C. Find the maximum H-radius
of an H-circle that may be contained in an H-triangle.

(iii) Find the E-centre and E-radius of C in the case v = ic, c R+ .

5. Show that, in a hyperbolic triangle, the three hyperbolic lines that bisect
the angles are concurrent.

P hyperbolic n-gon with interior angles 1 n has


6. Prove that a convex
area (n 2) i . Show that for every n 3 and every with
0 < < (1 n2 ) there is a regular n-gon all of whose angles are .
24 CHAPTER 2. HYPERBOLIC GEOMETRY

7. Show that two hyperbolic lines have a common perpendicular if and only
if they are ultraparallel, and that in this case the perpendicular is unique.
Deduce that the composition of the H-reflections in the two lines has
infinite order.
8. Fix a point P on the boundary of D, the disc model of the hyperbolic plane.
Give, with proof, a description of the curves in D that are orthogonal to
every hyperbolic line that passes through P .

9. Let l be a hyperbolic line and P a point on l. Show that there is a unique


hyperbolic line l0 through P making a positive angle (0, ) with l.
If , are positive numbers with + < , show that there exists a
hyperbolic triangle (one vertex at infinity) with angles 0, and . For
any positive numbers , and with + + < , show that there
exists a hyperbolic triangle with these angles.

10. Assuming the second hyperbolic cosine formula,

cos cos + cos = sin sin cosh a ,

deduce that two hyperbolic triangles are congruent if and only if they have
the same angles. What other conditions for congruence are there?

11. For a circle centred on the real axis find the H-length, in the upper half-
plane model, of the arc subtended by radii at angles and to the axis.
Deduce that the set M of points on one side of, and at a constant H-
distance from, an H-geodesic L is not itself an H-geodesic.
Show, however, that the locus of points equally H-distant from two fixed
points is an H-geodesic.
Find the formula for the H-reflection in the H-geodesic in the u.h.p model
which is the semi-circle centred on a R with radius r.
2.7. A FEW HINTS 25

2.7 A Few Hints


For questions involving more than one metric use the adjectival prefixes E-, S-
and H-.

(1 and 2) The first two parts in each case should be proved from the
relevant Riemannian metric (the differential form). For the third part you
may use the first two.
Your lecturer will a.s. have proved a subset of these and deduced the rest
using, for example, 3(i). The idea is to fill in the proofs of the results (i)
and (ii) in each case directly from the metric. You may omit any that
were covered that way in lectures. Similarly for question 3.

(3) (i) should also be proved using the Riemannian metrics, but may use
1(ii) or 2(ii). (ii) should not need any explicit calculations. For (iii) use
the facts already established.

(5-9) Should now be quite short using the facts already established. (9)
may need a continuity argument.

(7) Two H-geodesics either meet in the hyperbolic plane, or meet on its
boundary (unit circle or real axis in our two models) when they are called
parallel, or they are ultraparallel and do not meet at all.

2.8 Further Questions


As you may have noted I have omitted any proofs of the hyperbolic sine and
cosine formulae. The following are some that have been set on example sheets,
but not, I believe in any Tripos.

(1) Deduce the second hyperbolic cosine formula,

cos cos + cos = sin sin cosh a ,

from the first.

(2) For z, w C, prove the identity

|1 zw|2 = |z w|2 + (1 |z|2 )(1 |w|2 ).

For z, w in the unit disc with H-distance , show that

1 |z w|2
sinh2 ( (z, w))) =
2 (1 |z|2 )(1 |w|2 )

(3) Use the Euclidean cosine rule and the previous result, deduce the first
hyperbolic cosine formula.

(4) Prove the hyperbolic sine rule from the first hyperbolic cosine formula.
26 CHAPTER 2. HYPERBOLIC GEOMETRY

2.8.1 Comments
(1) Set A = cosh a, B = cosh b and C = cosh c. Deduce from the first hyper-
bolic cosine formula that
(BC A) D
cos = and sin2 = ,
(B 2 1/2 2
1) (C 1)1/2 (B 1)(C 2 1)
2

where D = 1 + 2ABC (A2 + B 2 + C 2 ) 0 (which is symmetric in A, B


and C). Hence, show that

cos cos + cos (AB C)(AC B) + (BC A)(A2 1)


=
sin sin D
and simplify the right hand side.
(4) Reduce this to showing that (sinh a sinh b)2 (cosh a cosh b cosh c)2 is
symmetric in a, b and c.
Chapter 3

The geometry of Surfaces

3.1 Introduction
These notes should be read in conjunction with those of the lecturer, particularly
for the several proofs that I shall omit. The book by Wilson designed for the
course continues to be a good reference. On the other hand, if you wish to go
still further, entering the ground covered by the Part II differential Geometry
course, a good text is Elementary Differential Geometry by Barrett ONeill,
published by Academic Press.
The context we consider is that of surfaces, such as spheres, tori, etc., em-
bedded in Euclidean spaces. The theory is valid for Euclidean spaces of any
dimension and also generalises easily to higher dimensional surfaces or mani-
folds embedded in such spaces. However initially we shall restrict attention to
familiar two-dimensional surfaces explicitly embedded in R3 .

3.2 Surfaces in R3
The defining property of a surface is that it is locally homeomorphic with
Euclidean 2-space, R2 . That means that we can cover it with coordinate neigh-
bourhoods that are homeomorphic with open subsets of R2 . We then use the
natural coordinates in this copy of R2 , to find our way around the surface, just
as we do with maps of our locality on the surface of the earth. The terminology
tends to reflect this analogy.
Working with surfaces embedded in R3 we can short circuit the abstract
definitions by working directly with parameterisations, similar to the way that
we parameterised curves in order to perform computations on them. However,
whereas curves were parameterised using intervals of the real line, surfaces,
which are two-dimensional, need to be parameterised by subsets of the plane.
For simplicity we restrict attention to smooth surfaces, that is, those for which
the relevant functions are differentiable arbitrarily often.
Definition A subset S R3 is a smoothly embedded parametrised sur-
face or just a smooth surface, if each point x in S has an open neighbourhood
U for which there is an open subset V of R2 with a smooth homeomorphism
: V U such that for all q V the vectors u u (q) and v v (q) at
p = (q) are linearly independent, where (u, v) are the coordinates on R2 .

27
28 CHAPTER 3. THE GEOMETRY OF SURFACES

We should emphasise that U is to be open is the subspace topology on S.


That is, U = U 0 S where U 0 is open in R3 . Then the subspace of R3 spanned
by u and v is called the tangent space to S at p, which we shall denote by
p (S). The union of all the tangent spaces at the various points of S, denoted
(S), can be given the structure of a 4-dimensional manifold and, as such, it
is the total space of a vector bundle. We shall not make explicit use of the
latter, so I shall not spell out the details.
As indicated above, This is an ad hoc definition used here, and elsewhere,
for this particular situation. By a smooth homeomorphism we mean a smooth
map with a continuous inverse : U V called a chart on S: U is called
a coordinate neighbourhood and V the corresponding coordinate space.
Via the inverse function theorem the condition on u and v implies that there is
an open neighbourhood W of x in R3 and a smooth map : W V such that
= id|V . Then, on U W , = |U . If : V U and 0 : V 0 U 0 are
two parameterisations on S with U U 0 6= then 1 0 is a diffeomorphism,
a smooth homeomorphism with a smooth inverse, defined on 01 (U U 0 ), and
is called a coordinate transformation or transition function.
An abstract surface, S, that is, one that is not explicitly embedded in a
particular Eulidean space, is then a topological space with a family of charts,
called an atlas, whose coordinate neighbourhoods cover S and are such that
all the coordinate transformations are diffeomorphisms. However adapting this
definition would still require us to say what we mean by embedding such a
surface in a Euclidean space and then to obtain the data which we require for
practical calculations that is already explicit in the definition above.
The elements of the tangent space are, naturally, called tangent vectors.
We should think of the tangent space as a subspace of a copy of R3 with its
origin rooted at p = (q). To expand on that remark a little, tangent vectors
to R3 at a point p are the tangents at that point to curves through that point:
if : [a, b] R3 is a curve with (c) = p for some c (a, b), then 0 (c) is a
particular tangent vector to R3 at p. For example, if q has coordinates (u0 , v0 ),
the vector u is the tangent vector at p to the coordinate curve u 7 (u, v0 ).
Then the tangent space p (R3 ) to R3 at p is the set of all such tangent vectors
at p. it is obviously bijective with R3 itself and so can be given a natural linear
structure. You will recall that the derivative at q of a mapping from R2 to
R3 is a linear map D from R2 to R3 : more precisely it is the corresponding
linear map, commonly denoted d or , from q (R2 ) to p (R3 ). Then, when
is our parameterisation of the surface at p, the tangent space p (S) to S at p is
just the image space d (q (R2 )) p (R3 ).
Note that, while the tangent space at (u0 , v0 ) was defined as that spanned
by the tangent vectors to the coordinate curves u 7 (u, v0 ) and v 7 (u0 , v),
any other two curves whose images lie in S and whose tangent vectors are
linearly independent could have been used. Since u and v are tangent to the
surface, the cross product will be normal to it so that
u v
N=
ku v k
is a unit normal to S, which is independent of the parameterisation provided
the coordinate transformation between two parameterisations has jacobian ma-
trix with positive determinant, when it is said to be orientation preserving.
3.3. RIEMANNIAN METRICS 29

A surface which admits an atlas all of whose coordinate transformations are


orientation preserving is said to be orientable and such an atlas determines an
orientation by specifying the unit normals as above. Thus, although it is not
the intrinsic definition, we may think of an orientation as a smooth family of
unit normals one at each point of the surface. The Mobius band is the simplest
non-orientable surface; the sphere and torus (S 1 S 1 ) are both orientable.
Warning. The tangent spaces to Rn are closely related in a deceptively simple
way: that at the point p may be thought of as the parallel translation to p of
the basic copy of Rn and these parallel translations are unique and consistent.
However that is not usually true for a general surface or manifold: though it
is still true for the circle it already fails for the two dimensional sphere. This
latter result is essentially that known as the hairy ball theorem: you cannot
comb a hairy ball flat!

3.3 Riemannian Metrics


Recall from Chapter 2 that, if the (variable) quadratic form

a(u, v)x2 + 2b(u, v)xy + c(u, v)y 2

is positive definite (in x, y) for all (u, v) in some domain V in R2 ; that is,
a(u, v) > 0 and a(u, v)c(u, v)b(u, v)2 > 0 for all (u, v) V , then the associated
differential 2-form

g(u, v) = a(u, v)du2 + 2b(u, v)dudv + c(u, v)dv 2

is called a Riemannian metric on V . We now put a little more meat on those


bones.
More precisely, a Riemannian metric on a surface or more general subman-
ifold M Rn is a positive definite inner product on each tangent space to M
that varies smoothly over M : in the above, a, b and c are smooth functions,
as (u, v) varies in V . On Euclidean space Rn itself, we have the standard Rie-
mannian metric given, at each point p by the standard inner product on Rn
identified, by the parallel translation as above, with p (Rn )
Coming back down to earth in R3 , but still less trivially, on a surface S R3
we can define a Riemannian metric at p S just by taking the restriction
to p (S) of the standard metric of R3 . Since, for our parameterisation :
V, q U, p, where V R2 and U S, the derivative D (q) is an isomorphism
from the tangent space to V at q to the tangent space to S at p, we may pull
this metric on p (S) back to one, the induced metric, on q (V ). This will be
given, for vectors a, b tangent to V at q, by

h a, b iq = h D (q)(a), D (q)(b) ip .

Either of these metrics, the restriction of the standard metric on R3 at each


point of U = (V ) or the induced metric at each pont q of V , may be referred
to as the first fundamental form on S. If we denote the restricted metric on
U by g, then that on V should strictly be denoted (g), indicating that it has
been pulled back from U to V . However commonly it is also denoted g.
30 CHAPTER 3. THE GEOMETRY OF SURFACES
 
u
If, as we are assuming, q V has coordinates with respect to the
v
 
a1
standard basis and a q (V ) has coordinates with respect to the standard
a2
basis e1 , e2 of q (V ) so that a = a1 e1 + a2 e2 , then

D (q)(a) = a1 u +a2 v .

Then, the bilinearity of the inner product implies that

h a, b iq = h D (q)(a), D (q)(b) ip
= h a1 u +a2 v , b1 u +b2 v i
 
E F
= aT b,
F G

where E = u u , F = u v = v u and G = v v . If, as is commonly


done, we overwork our notation and write u for the coordinate function (u, v) 7
u on V , then the derivative Du is simply projection onto the first factor, so that,
wherever in V it is computed, Du(a) = a1 and, similarly, Dv(a) = a2 . Note
that Du is thus a linear map Du : q (V ) R and so an element of the dual
space q (V ) . Such an element, for each q in V , is called a 1-form on V and,
as such, is usually written du with lower case d. (It is by analogy with this that
general derivatives are often written with the lower case d.)
A two-tensor on V , more precisely a 2-covariant tensor, is a bilinear form
q (V )q (V ) R, again varying smoothly as q varies in V . Given two 1-forms
(or 1-tensors!) and on V , we can produce a 2-tensor, called their tensor
product and denoted by by defining

( )(a, b) = (a) (b).

Now
 
T E F
a b = a1 b1 E + (a1 b2 + a2 b1 )F + a2 b2 G
F G
= Edu(a)du(b) + F du(a)dv(b) + F dv(a)dv(b) + Gdv(a)dv(b)

= E du du + F (du dv + dv du) + Gdv dv (a, b).

So the metric is the 2-tensor g = E du du + F (du dv + dv du) + Gdv


dv. When applied to the diagonal (a, a) we get the associated quadratic form,
determining the norm of a vector:

kak2 = g(a, a) = E du du + 2F du dv + Gdv dv (a, a)




and it is conventional, though not strictly logical, to write the metric as

g = Edu2 + 2F du dv + Gdv 2 ,

suppressing the tensor symbol as well as the asymmetry du dv 6= dv du,


as we did at the start of this section. Thinking of the above interpretation of
the diagonal quadratic form we even write g(u, v) = ds2 informally, although
strictly ds does not exist except as a mnemonic for the following application.
3.4. LENGTHS OF CURVES 31

3.4 Lengths of Curves


Given a surface S in R3 , we consider a parameterisation on S, which is an open
subset V of R2 and a differentiable map : V R3 such that (V ) = U S
and is a homeomorphism of V with U when the latter is given the topology
that it inherits as a subset of R3 . This last condition is a technicality needed
to avoid certain potential problems. However you do not need to worry about
it since you are unlikely to wish to use a chart that violates it. We implicitly
assumed it in definition 3.2 since, by U being open on S we meant that it should
be open in the subspace topology and so the intersection U = W S of an open
subset W of R3 with S.
We now consider the lengths of curves in S R3 . Let : [a, b] S be such
a curve with its image contained in the coordinate neighbourhood U = (V ).
Since S is covered by coordinate neighbourhooods - it has an atlas of charts, we
can always break a curve up into segments such that each segment is contained
in a coordinate neighbourhood. Then the conditions we placed on our charts
ensure that we can find a curve : [a, b] S such that = . Denoting
the coordinates in R2 by (u, v) and those in R3 by (x1 , x2 , x3 ) we can spell this
out in painful detail as:

: [a, b] M : t 7 x1 (u(t), v(t)), x2 (u(t), v(t)), x3 (u(t), v(t)) ,

where : [a, b] V : t 7 (u(t), v(t)).


The chart itself gives rise to the obvious coordinate curves: for each point
(u0 , v0 ) of V , there is a u-parameter curve

v0 : t 7 (t, v0 ),

given by u(t) = t and v(t) = v0 , and a v-parameter curve

u0 : t 7 (u0 , t),

given by u(t) = u0 and v(t) = t. The tangent vector to v0 at q = (u0 , v0 ) is


just the tangent vector v0 0 (uo ) = u (q), etc., as stated above.
The coordinate neighbourhood (V ) is covered by these two families of
curves, which are the images under of the horizontal and vertical lines in
V , and one curve of each family goes through each point of U , though of course
we need to restrict the domain [a, b] on which each line is defined to ensure that
it stays in V and, hence, that its image stays in U .
Of course these coordinate curves are curves in R3 and, as such, their length
Rb
is given by a kv0 0 (t)kdt, for example for the u-parameter curves. The tangent
vector, or velocity, v0 0 (t) is given by
x1
u (t, v0 )
v0 0 (t) = u (t, v0 ) x
u (t, v0 )
2 ,
x3
u (t, v0 )

and kv0 0 (t)k2 is simply the inner product u (t, v0 ) u (t, v0 ).


Turning to the general curve above, the tangent vector 0 (t) has coordi-
nates
dxi xi du xi dv
(u(t), v(t)) = (u(t), v(t)) (t) + (v(t), v(t)) (t),
dt u dt u dt
32 CHAPTER 3. THE GEOMETRY OF SURFACES

so that
du dv
0 (t) = u (u(t), v(t)) (t) + v (u(t), v(t)) (t),
dt dt
and hence
du 2 du dv dv
k0 (t)k2 = E(u(t), v(t))(t) + 2F (u(t), v(t)) (t) (t) + G(u(t), v(t)) (t)2 ,
dt dt dt dt
()
where we have used the conventional notation, for the case of surfaces:

E = u u , F = u v = v u , G = v v ,

supressing now the arguments (u(t), v(t)) throughout, since they would have
caused the line to overflow. If, in addition, we suppress all mention of the
parameter t, and write 0 (t) = ds
dt then () becomes

ds2 = E(u, v) du2 + 2F (u, v) du dv + G(u, v) dv 2 .

Indeed () is the true meaning of this latter expression, which we note is a


Riemannian metric g on V in the sense of the previous section since, by Cauchy-
Schwartz and the fact that u and v are never parallel, E(u, v)G(u, v) >
F (u, v)2 for all u and v.
Now, recalling that is in fact a curve in R3 , its length measured in the
Rb
standard metric on R3 is a k0 (t)kdt which, by () is
Z b p
E(u(t), v(t))(u0 (t))2 + 2F (u(t), v(t))u0 (t)v 0 (t) + G(u(t), v(t))(v 0 (t))2 dt.
a
Rb
But, since (t) = (u(t), v(t)), this is just the length Lg [ ] = a k 0 (t)kg dt,
where k 0 (t)kg is the norm of 0 (t) as measured by the induced metric g: the
length of , as measured by the restriction to S of the standard metric on R3 ,
is the same as the length of its coordinate curve measured by the induced
metric g on the coordinate space V .
Remarks Note that the integrand above is real since we assumed the quadratic
form associated with the Riemannian metric is positive definite everywhere on
V . Note also that we have implicitly assumed the curve is differentiable, i.e.
that u0 (t) and v 0 (t) exist for all t. However piece-wise differentiability will do
provided the curve is also continuous, and that is all we shall assume. Similarly,
if our initial curve on S does not fit into a single coordinate neighbourhood,
we can integrate along subsections of the curve that do so fit and add the results
together.
The standard Riemannian metric on R2 is given by ds2 = du2 + dv 2 , that is,
E = 1 = G and F = 0. Such a metric on V is called flat or locally Euclidean
and the surface S that induced it is also called flat. In that case the equation
for the length of the coordinate curve reduces to the familiar one.

3.5 Geodesics
In view of the previous section and of the remarks at the end, we shall now tend
to describe concepts, such as the current geodesics, on the surface S itself while
3.5. GEODESICS 33

expressing all formulae and calculations in its coordinate spaces V in terms of


the induced metric g.
For points p, q in V we may define d(p, q) to be the infimum of the lengths
Lg [ ] of all, continuous and piece-wise differentiable, curves : [a, b] V such
that (a) = p and (b) = q. This infimum certainly exists as the integrals
representing lengths are all positive and we may show that it is a metric in
the familiar sense, that is, it satisfies the triangle inequality and, provided g is
continuous (i.e. its coefficient functions E, F and G are), d(x, y) = 0 x = y.
Beware however that this may not give the correct distance on S if the
coordinate neighbourhood U = (V ) is too large: there may be shorter routes
from (p) to (q) on S through other neighbourhoods. See below.
When we need to distinguish between the two forms of metric we refer to a
Riemannian metric or differential form of the metric in one case and to a distance
function or distance form of the metric in the other. Not all distance functions
arise from Riemannian metrics in the manner just described. However all those
with which we are concerned in this course do. For example the Euclidean
distance function so arises from the Riemannian metric g(u, v) = du2 + dv 2 .
The infimum may or may not be attained but any curve which does at-
tain it is called a geodesic from p to q. More precisely, since any monotonic
reparametrisation of a curve has the same length, we require geodesics to have
constant speed. Then, for each tangent vector v at q V , there is a geodesic,
v , unique if we require it to be maximal, defined on (a, b) where a < 0 < b
and a and/or b may be infinite, such that v (0) = q, v0 (0) = v and the speed,
kv0 (t)kremains equal to kvk for all t. It is a geodesic in the above sense in that
it does minimise the distance between any two of its points that are sufficiently
close. The qualification that they be sufficiently close is required since, for ex-
ample, on the cylinder x2 + y 2 = 1 in R3 the curve (t) = (cos t, sin t, 1) is a
geodesic but [t ,t ] does not minimise the distance between (t1 ) and (t2 ) if
1 2
|t1 t2 | > .
In fact the intrinsic definition of a geodesic is that its tangent vectors be
parallel. You may recognise this property for the straight lines in Euclidean
space that are the geodesics for the standard metric, but we are not in a position
to define what we mean by parallel vectors along a curve in general. So it may
not be obvious, but is nevertheless true that, for example, the tangent vectors to
a great circle, a geodesic, on the sphere are parallel, whereas those along other
circles are not.
On the other hand there is yet another characterisation of geodesics that
does not require us to look at sufficiently close points since, as for the parallel
property, it is already local in nature, and that also has the advantage of giving
us straightforward equations to solve in the coordinate spaces in order to identify
geodesics.
As well as the length of a curve we
R may define its energy, omitting the
rather awkward square root: Eg [ ] = ds2 which is
Z b
{E(u(t), v(t))(u0 (t))2 + 2F (u(t), v(t))u0 (t)v 0 (t) + G(u(t), v(t))(v 0 (t))2 } dt
a

Among the curves joining two given points a particular curve minimises
the energy if, and only if, it both has constant speed (k 0 k is constant) and
34 CHAPTER 3. THE GEOMETRY OF SURFACES

locally minimises the length. In other words it is a geodesic in the above sense.
The lecturer may accordingly use the following Theorem as his definition of a
geodesic.

Theorem 3.5.1. The curve in V , as above, is a geodesic if

d 1
Eu u2 + 2Fu uv + Gu v 2

(E u + F v) =
dt 2
d 1
Ev u2 + 2Fv uv + Gv v 2 .

(F u + Gv) =
dt 2

Here everything is evaluated at a point t of the domain of . Thus E u means


E(u(t), v(t)) u(t), et cetera. These are in fact the Euler-Lagrange equations for
a stationary point of the energy with respect to proper variations of curves
joining two given points. See the Summary section for more details.

3.6 Surface Area


Having thus given some intrinsic meaning to the Riemannian metric and used
it to measure the lengths of curves, we can now turn to areas of surfaces, or
subsets of them. Once again we can write any domain on our surface S as a
union of domains each of which is contained in a coordinate neighbourhood.
Thus it suffices to consider a domain D, or more precisely its closure, contained
in a coordinate space V that has a parameterisation : V U S. Consider,
then, a small coordinate square in D with opposite vertices (u, v) and (u +
h, v + k). The image in U is of course not a square, but has adjacent edges
which are curves of approximate length (u + h, v) (u, v) h
u and
(u, v +

k) (u, v) k v . Thus the area of this image is approximately hkk u v k.
Summing over all such squares in D and taking the limit as their size tends to
zero we see that
Z p
the area of (D) = EG F 2 dudv,
D

since k u v k2 = k u k2 k u k2 | u v |2 = EG F 2 .

3.7 Curvature of Surfaces


Recall that the parametrisation : V U S on a surface S gives rise to the
unit normal vector field
u v
N= .
ku v k
In particular N will be a vector valued function of q = (u, v) V with value
a tangent vector to R3 at p = (q) U that is normal to the surface S at p.
Since the parametrisation identifies the spaces V and U , and hence also their
tangent spaces, we may regard N as defined along S. Although N cannot be
uniquely defined outside the surface S, we can still define its partial derivatives
3.7. CURVATURE OF SURFACES 35

with respect to u and v. Looking at this more closely:

N 1
(u, v) = lim (N ((u + t, v)) N ((u, v)))
u t0 t
1
= lim (N ((u, v) + t u (u, v)) N ((u, v)))
t0 t
= DN ((u, v), u (u, v)),

which is my preferred notation for the directional derivative of

N at p = (q) (u, v) in the direction of u (q) = u (u, v).

We remind the reader that directional derivatives, even in all directions,


may exist without the map being differential with derivative a linear
map between the relevant tangent spaces. In the latter case it would be
denoted DN ((u, v)) u (u, v) and be referred to as the value that the
derivative DN of N , calculated at (u, v), takes on the vector u , if
maximum precision were required. The context should distinguish this
from an inner product between two vectors.
Since the latter is a tangent vector to S at p, the derivative of N in that direction
is well-defined and we may denote it Nu (p). More generally, it is defined for
any tangent vector w = u + v to S at p. By the linearity of the derivative,
the directional derivative Nw (p) is just Nu (p) + Nv (p). Note that, as for
Nu and Nv , if : (a, b) S is a curve with (0) = p and 0 (0) = w, then the
d
directional derivative Nw (p) is just dt N ( (t)) t=o . An alternative notation for
this directional derivative is w N (p), reflecting the fact that it is the covariant
derivative with respect to w determined by the standard (flat) metric on R3 .
Now we can define the shape operator or Weingarten map

W : p (S) p (S) : w 7 Nw = w N (p)

and the second fundamental form II(x, y) = W (x) y = h W (x), y i, the inner
product between the two tangent vectors W (x) and y in p (S). The artificial
minus sign is commonly chosen here in order to have to do so less frequently
later.

Proposition 3.7.1. The second fundamental form is well-defined and symmet-


ric (i.e. self-adjoint).

Proof. Note that any linearly independent tangent vectors x, y at p S can


arise as tangent vectors to coordinate lines at p in the same parametrisation, so
we may assume x = u and y = v . Then

N N = 1 = Nu N = 0 and Nv N = 0

so that W does map p (S) to itself.


Then N u = 0 and N v = 0 imply that

II(x, y) = Nu v = N vu = N uv = Nv u = II(y, x).


36 CHAPTER 3. THE GEOMETRY OF SURFACES

Definition The Gauss map or sphere map of the surface S is the map

G : S S 2 ; p 7 N (p),

where we have used, but not explicitly mentioned, the projection of the tangent
bundle U = U R3 onto the second factor R3 to reinterpret the unit vector
N (p) in p (U ) as a point of the unit sphere S 2 in R3 .
The Gauss map allows the following interpretation of the shape operator.

Proposition 3.7.2. For any tangent vector w to S at p, Nw (p) = DG(p) w.

Proof. Since Nw (p) is in p S and DG(p) w is in G(p) S 2 , what this is really


saying is that the two vectors have identical components. To clarify that, let
(t) be a curve, with image in U , such that (0) = p and 0 (0) = w, and write
G as
N 2
U (S) U R3 R 3
,
recovering the suppressed projection 2 onto the second factor, with G(U )
S 2 R3 . Then

DG(p) w = (G )0 (t) t=0 = D2 (N ( (t))) (N )0 (t) t=0


which, since 2 is linear and so D2 = 2 is the projection that we are sup-


pressing, is just (N )0 (t) t=0 = Nw (p).

The second fundamental form gives rise to a number of subsidiary definitions


at each point p of S as follows:
Definitions

(i) K(p) = det(W )(p) is the total or Gaussian curvature at p;

(ii) H(p) = trace(W )(p) is the mean curvature at p;

(iii) the eigenvalues of (W )(p) are the principal curvatures at p;

(iv) the eigenvectors of (W )(p) are the principal vectors or directions of


curvature at p;

(v) a curve (t) is a line of curvature if 0 (t) is a principal vector for all t;

(vi) p is an umbilic point if (W )(p) = p I;

(vii) p is a planar point if (W )(p) = 0;

(viii) (X) = II(X, X) = W (X) X; is the normal curvature of X, ( is the


quadratic form associated with the second fundamental form);

(ix) X is asymptotic if (X) = 0;

(x) the locus {X|(X) = 1} p (S) is the Dupin indicatrix at p.


3.8. ANALYSIS OF CURVATURE 37

Some authors define the mean curvature to be H(p)/2. Note that replacing
N by the equally valid normal field N or, equivalently, dropping the minus sign
in the definition of the second fundamental form, will not affect the Gaussian
curvature, but will change the sign of the mean curvature.
We note that, since W is self-adjoint, there are 2, possibly equal, principal
curvatures at p and p (S) has a basis of principal vectors. Moreover princi-
pal vectors with different eigenvalues are orthogonal and so we may find an
orthonormal basis of principal vectors.
Now p (S) is 2-dimensional and we have principal curvatures k1 , k2 at p.
The characteristic equation of W is

det(W I) det W tr W + 2 = 0.

So k1 k2 = det(W )(p) = K(p) and k1 + k2 = trace(W )(p) = H(p).


If X, Y are principal vectors forming an orthonormal basis of p (S), so that
W (X) = k1 X, W (Y ) = k2 Y then, for Z = aX + bY , (Z) = k1 a2 + k2 b2 . We
examine the indicatrix in various cases.

(i) If K(p) > 0, k1 and k2 are the same sign and we may choose our unit
normal field N so that they are both positive. Then the indicatrix is the
ellipse k1 a2 + k2 b2 = 1 and p is referred to as an elliptic point.

(ii) If K(p) < 0, k1 and k2 are of opposite signs. Then the indicatrix is a
hyperbola and p is referred to as a hyperbolic point.

(iii) If K(p) = 0 there are two cases. If k1 = 0 and, without loss of generality,
k2 >0, the indicatrix is k2 b2 = 1 which is the pair of parallel lines b =
1/ k2 , and we have a parabolic point. If k1 = k2 = 0, the indicatrix is
the single point (0, 0) of p (S), and we have a planar, flat or Euclidean
point.

(iv) If p is umbilic then k1 = k2 = k, all normal curvatures are equal to k,


the indicatrix is a circle (or point if k = 0), so that umbilic points are a
special case of elliptic points.

A surface for which K(p) 0 is called flat and a surface for which H(p) 0
is called minimal.

3.8 Analysis of Curvature


It is possible to analyse the curvature of a surface by means of curves lying in
that surface. In particular such a curve is a curve in R3 and we recall that we
may reparameterise : (a, b) R3 , by path length, to have unit speed. Thus,
if 0 as usual denotes differentiation with respect to the parameter t (a, b),
T = 0 has kT k = 1 and T 0 is the curvature field along with kT 0 k =
0, the curvature (function). Then T 0 is orthogonal to T and N = T 0 / is
called the principal normal to the curve and (we are in R3 ) B = T N its
binormal. Then B 0 = N , where is called the torsion of the curve and
N 0 = T + B. The orthonormal frame T, N, B along is called the Frenet
frame.
38 CHAPTER 3. THE GEOMETRY OF SURFACES

The proofs are straightforward. For example, N T = 0 follows from T T = 1


and then N 0 T = T 0 N = , by the definition of . The other parts of the
proof are similar.
Proposition 3.8.1. If is planar then T and N are tangent to its plane, and
bends away from the tangent line in the direction of T 0 : i.e. it lies on the
same side of T as does T 0 .
Proof. That T is tangent to the plane of is obvious. Let U be a unit normal
to the plane in which lies. Then T U = 0 implies 0 = (T U )0 = T 0 U + T U 0 .
But U is constant. (It has the same coordinates at each point of .) So U 0 = 0
and hence T 0 lies in the plane and so does N . That lies on the same side of T
as T 0 is clear from the special case of (t) = (cos t, sin t) for which T 0 = .
Proposition 3.8.2. Let be a unit speed curve in a surface S in R3 . Let
S have (local) unit normal field N and let , regarded as a curve in R3 , have
tangent field T , principal normal N and curvature . Then
(N N ) = T 0 N = (T ),
where (T ) is the normal curvature, W (T ) T , of T in S.
Proof. The first equality is from the definition of N and . For the second,
T N = 0 implies that
T 0 N + T N 0 = 0.
However N 0 = (T N =)NT = W (T ). So T 0 N = T W (T ) = (T ).

At a point p of the surface S in R3 , let be the plane determined by N (p)


together with a choice T (p) of a unit vector in p (S). The intersection, S, of
the plane with S near p must be a curve, by the implicit function theorem, since
the restriction to S of the normal distance from does not have 0 as a critical
value near p. (There exists a tangent vector to S at p which does not lie in
.) Let be a unit speed parameterisation of S then, by Proposition 3.8.1,
N = N . When N = +N , lies on the same side of T (p) as N , curving
towards N and, by Proposition 3.8.2, (T ) = < 0. When N = N , lies
on the opposite side of T (p) to N , curving away from N and (T ) = > 0.
The normal curvature of a principal vector is the corresponding principal
curvature so:
(i) At an elliptic point, choosing N such that k1 > 0 and k2 > 0 we have
the normal curvature (Z) of a general tangent vector Z = aX + bY is
(Z) = (aX + bY ) = k1 a2 + k2 b2 which is strictly positive for all Z. The
surface curves away from N in all planar sections through N . There are
no asymptotic directions at p.
(ii) At a hyperbolic point, assume k1 = (X) > 0 > k2 = (Y ), then the
surface curves away from N in the X-N plane and curves towards N
in the Y -N plane. The normal curvature of a general vector Z is now
(Z) = k1 a2 + k2 b2 which is zero if and only if
s
a |k2 |
= ,
b k1
giving two asymptotic directions at p.
3.9. COMPUTATION OF CURVATURE OF SURFACES 39

(iii) In the parabolic and planar cases there are respectively one or all direc-
tions for which the intersection curve has higher order contact with the
corresponding tangent vector. Unless the same is true for other neigh-
bouring points little more can be said.

3.9 Computation of Curvature of Surfaces


Just as it is conventional to write

E, F, G for the three distinct values u u , u v , v v

respectively of the first fundamental form, we also write l, m, n for the three
distinct values of the second fundamental form which, as we saw for m = Nu v
in the proof of Proposition 3.7.1, are N uu , N uv and N vv , respectively.
We use the lower case here so that the N that would otherwise result should
not be confused with the normal vector field to S.
 Then,  with respect to the basis u ,v , the 
first fundamental form has matrix
E F l m
and the second has matrix .
F G m n
Proposition 3.9.1.

ln m2 Gl + En 2F m
K= , H= .
EG F 2 EG F 2
Proof. Let X, Y be principal orthonormal vectors such that X Y = N , with
W (X) = k1 X, W (Y ) = k2 Y and let U = aX + bY, V = cX + dY . Then
W (U ) = k1 aX + k2 bY , et cetera, so that

W (U ) W (V ) = (ad bc)k1 k2 N = K(p)(U V )

Hence
(W (U ) W (V )) (U V ) (W (U ) U )(W (V ) V ) (W (U ) V )2
K(p) = 2
= .
kU V k (U U )(V V ) (U V )2

The computation of H(p) follows similarly from the identity

W (U ) V + U W (V ) = H(p)U V.

Example The surface z = f (x, y) has the parametrisation

: (u, v) 7 (u, v, f (u, v)).


f
Then u = (1, 0, fu ) = e1 + fu e3 and v = (0, 1, fv ), where fu = u , etc., and
N = (fu , fv , 1)/(1 + fu2 + fv2 )1/2 . We get l = fuu /(1 + fu2 + fv2 )1/2 , etc., and

fuu fvv fuv2 2fu fv fuv (1 + fu2 )fvv (1 + fv2 )fuu


K= , H=
(1 + fu2 + fv2 )2 2(1 + fu2 + fv2 )3/2 .

Applying these computations to the saddle surface, z = xy, and changing


to cylindrical polar coordinates r, , z, we get K = 1/(1 + r2 )2 and H =
40 CHAPTER 3. THE GEOMETRY OF SURFACES

z/(1 + r2 )3/2 . Thus, in general, K [1, 0) and, since l = n = 0, we see


that the u and v directions are asymptotic. At the origin K = k1 k2 = 1
and H = k1 + k2 = 0, so k1 , k2 = 1. The fact that m = 1 implies that
W (v ) = u and W (u ) = v . Hence W (u + v ) = (u + v ) and
W (u v ) = +(u v ), so that the principal vectors corresponding to k1 =
+1 and k2 = 1 are u v and u + v respectively.
Example If, for a general surface, we choose our axes such that the first two lie
in the tangent plane at p, with the corresponding unit orthonormal coordinate
vectors principal at p then, by the implicit function theorem, there is a function
f (x, y) such that (x, y, z) S if and only if z = f (x, y) on some neighbour-
hood of p. Using the notation introduced above, the fact that u = (1, 0, fu )
is tangent to the surface at p = (0, 0, 0) means that fu (0, 0) = 0. Similarly
fv (0, 0) = 0. Thus l(0, 0) = fuu (0, 0), m(0, 0) = fuv (0, 0), and n(0, 0) =
fvv (0, 0). The fact that u (0, 0) and v (0, 0) are principal and orthogonal implies
that m(0, 0) = h W (u ), v i = h k1 u , v i = 0. So that fuv (0, 0) = 0. Thus we
find K(0, 0) = fuu (0, 0)fvv (0, 0) and H(0, 0) = fuu (0, 0) + fvv (0, 0). Thus the
principal curvatures at p are k1 = fuu (0, 0) and k2 = fvv (0, 0). Moreover the
Taylor expansion for f at (0, 0) reduces to

1
la2 + nb2 + o(k(a, b)k2 ).

f (a, b) =
2
Thus the plane perpendicular to the z-axis with equation z =  meets M ap-
proximately where k1 a2 + k2 b2 = 2 . The limiting shape of this intersection as
 0 is the Dupin indicatrix.
Example The surface formed by revolving the curve (u) = (f (u), 0, g(u))
about the z-axis has parametrisation

(u, v) = f (u) cos v, f (u) sin v, g(u) ,
p
giving E = f 02 +g 02 , F = 0, G = f 2 and so dA = f f 02 + g 02 and the curvature
K(u, v) simplifies to
(f 0 g 00 g 0 f 00 )g 0
,
f (f 02 + g 02 )2
which further simplifies to f 00 /f when is parameterised with constant unit
speed: f 02 + g 02 = 1.
Example If the first fundamental form is du2 + G2 (u, v)dv 2 , for example in
geodesic polar coordinates, then K = Guu /G.

3.10 The Gauss-Bonnet Theorem


This
R states that if S is a closed surface S, that is it has no boundary, then
S
KdA = 2(S), where (S), the Euler number, is independent of the metric.
It is given for any triangulation, or more general cell decomposition, by V E +
F where V, E and F are the numbers of 0-, 1- and 2-dimensional simplices or
cells, respectively. For a 2-dimensional surface S with boundary S the theorem
says that Z Z
dA + g = 2(S),
S S
3.11. SUMMARY 41

where g is the geodesic curvature of the boundary. For a geodesic n-gon


the geodesic curvature is zero along the edges and the exterior angle (
the interior angle) at each vertex. Thus a triangle T in the unit sphere has
= +1 and (T ) = 1 (three vertices, three edges and one face), so that
area(T ) + (interior angles) = 2. So area(T ) = (interior angles).
Similarly a triangle in hyperbolic space has = 1 and still (T ) = +1 so that
area(T ) + (interior angles) = 2. So area(T ) = (interior angles).
Thus we obtain the previously stated formulae for the areas of triangles which
was independently calculated at least in the spherical case.

3.11 Summary
A smooth parameterisation of a surface takes the form of a differentiable
homeomorphism : V U ; (u, v) 7 (x(u, v), y(u, v), z(u, v)) between an
open subset V of R2 and an open subset U of the surface, where the tangent

vectors to the surface u = u and v are linearly independent. The induced
Riemannian metric on V is

g(u, v) = E(u, v)du2 + 2F (u, v)dudv + G(u, v)dv 2 where

E = u u , F = u v , G = v v .

A small rectangle R in V with opposite vertices (u, v) and (u + h, v + k) has


image (R) with approximate adjacent edges h u and k v and so approximate
area hkku v k. Since ku v k2 = EF F 2 , this gives the integrand for
calculating the area of (D) U corresponding to a domain DA in V , in
terms of the coordinates on V as

dA = (EG F 2 )du dv = ku v kdu dv.

Similarly, for a curve : [a, b] V , the integrand for calculating the length
of the curve = in U is

ds = k 0 (t)kg dt = (E 1 2 +2F 1 2 +G 2 2 ) dt,

since k 0 (t)kg = k0 (t)kstd .


In order to calculate the curvature of the surface we need the unit normal
to the surface which is N = (u v )/ku v k. Then defining

L = uu N = u Nu , M = uv N = u Nv , L = vv N = v Nv ,

the curvature is K = (LM N 2 )/(EG F 2 ).


The geodesic equations, arising from the Euler-Lagrange equations,
   
d I I d I I
= and = ,
dt u u dt v v

on the energy integrand

I(u, v, u, v) = g(u, v)(u, v) = E(u, v)u2 + 2F (u, v)uv + G(u, v)v 2 ,


42 CHAPTER 3. THE GEOMETRY OF SURFACES

are

d 2 2
2 (E 1 + F 2 ) = Eu 1 + 2Fu 1 2 + Gu 2 and
dt
d 2 2
2 (F 1 + G 2 ) = Ev 1 + 2Fv 1 2 + Gv 2 .
dt

There are (at least) three special forms of parameterisation where these
formulae simplify:

3.11.1 Graph parameterisation


If f : V R; (u, v)
7 f (u, v) is a function, then its graph is the surface
parameterised by (u, v) = (u, v, f (u, v)) and we get

u = (1, 0, fu ), v = (0, 1, fv ) so that

(fu , fv , 1)
E = 1 + fu2 , F = fu fv , G = 1 + fv2 and N = , and then
(1 + fu2 + fv2 )

(fuu fvv fuv2 )


dA = (1 + fu2 + fv2 )du dv and K =
(1 + fu2 + fv2 )2

3.11.2 Surface of revolution


Revolving the curve (u) = (f (u), 0, g(u)) in the x-z-plane about the z-axis
gives a surface that may be parameterised as

(u, v) = (f (u) cos v, f (u) sin v, g(u))

leading to

E = k 0 k2 = f 02 + g 02 , F = 0, G = f 2 and then

(g 0 cos v, g 0 sin v, f 0 ) 0 (f 0 g 00 g 0 f 00 )g 0
N= 02 , k u v k = |f |k k and K = .
(f + g 02 ) f k 0 k4

The curves u = constant are called parallels and the curves v = constant
meridians. [Lines of latitude and longitude respectively when the earth is
thought of as a surface of rotation about its NS axis.]
Assuming that the curve we are rotating has unit speed, the geodesic equa-
tions simplify and, if we also take them to have unit speed (it must be constant),
we get
d 2
u = f fu v 2 , (f v) = 0 and u2 + f (u)2 v 2 = 1
dt
where the last equation is for unit speed. It then follows that all meridians are
geodesics and a parallel u = u0 is a geodesic iff fu (u0 ) = 0.
3.11. SUMMARY 43

3.11.3 Geodesic polar coordinates


These, centred on P V , are defined by (, ) = () where is the unique
geodesic with (0) = P , with unit speed and with arg (0) = for some
continuous choice of arg on a suitable domain with a slit up to P .
The metric is then d 2 +G(, )2 d 2 and the Gaussian curvature K =
G /G. By definition the radial curves = constant are geodesics and
is the distance from P .
More generally K = Guu /G for a metric du2 + G(u, v)2 dv 2 even when u
and v are not the geodesic polar coordinates.
44 CHAPTER 3. THE GEOMETRY OF SURFACES

3.12 Questions
(1) Find the induced metrics on the coordinate spaces for the following para-
metrisations; in each case specify the relevant coordinate space and coordi-
nate neighbourhood and check that the stated map is indeed a parametri-
sation:

(i) spherical polar parameters on the sphere;


(ii) the inverse of stereographic projection of the sphere from the north
pole to the equatorial plane;
(iii) (u, v) = (f (u) cos v, f (u) sin v, g(u)), where f and g are smooth real-
valued functions such that f 0 (u)2 + g 0 (u)2 6= 0, f (u) > 0 and u 7
(f (u), g(u) is a homeomorphism of an interval (a, b) onto its image;
(iv) the case f (u) = 2 + cos u, g(u) = sin u of (iii).
(v) (u, v) = (cos u, sin u, v)
(vi) (u, v) = (u, v, F (u, v)) where F : R2 R is a smooth function.

(2) (i) Find the tangent space to the unit two-sphere at the point (x, y, z);
(ii) Describe and find the area of the surface parametrised by 1(iv);
(iii) Compare the cases

f (u) = u, g(u) = u a

and p
f (u) = u/ 1 + a, g(u) = u a/(1 + a)
of 1(iii). Describe the geodesics on the image, in particular showing
that they intersect themselves if, and only if, a > 3.
(iv) Find the geodesics in the surface parametrized by 1(v);
(v) Find the Gauss curvature of 1(iii) and confirm that it reduces to
f 00 (u)/f (u) when the curve (u) = (f (u), g(u)) has unit speed;
(vi) Find the Gauss curvature of the unit sphere and the hyperboloids
x2 + y 2 = z 2 1.
(vii) Find for 1(iv) where the Gauss curvature is positive, negative or zero.
Verify the global Gauss Bonnet theorem on the image.
(viii) Find the Gauss curvature of 1(vi).

(3) Find an atlas of charts on S2 for which each chart is area preserving and
the transition maps have derivatives with determinant +1.

(4) Let r2 = x2 +y 2 and the Riemannian metric be given by (dx2 +dy 2 )/h(r)2
for r < with h(r) > 0. Show the Gaussian curvature is given by

K = hh00 (h0 )2 + hh0 /r.

(5) Show that, for a compact, embedded, closed (empty boundary) surface S
in R3 the Gaussian curvature must be (strictly) positive at some point of
S.
3.13. FURTHER QUESTIONS 45

(6) Let S be the surface obtained by rotating the curve in the xz-plane
about the z-axis. Find such that S has curvature 1.
(7) Show that a genus two surface (two-holed torus) may be obtained by
identifying the sides of a regular octagon. Deduce that it may be given
a Riemannian metric which is locally isometric to the hyperbolic plane;
that is, it has curvature 1.
Generalise for arbitrary genus.

3.12.1 Comments and hints on the questions


I have, on a whim, separated the two components of most questions: the ap-
propriate parametrisation and the actual geometric question. This means that
these questions 1 and 2 cover 6-8 questions on a standard example sheet. So
dont panic if they seem to take a long time.
(1) Check precisely what it means to be a parametrisation before starting
this question. Note that the coordinate spaces and neighbourhoods are
not generally unique. Choose an appropriate one.
2(i) You may use a parametrisation, but it is more efficient to use the intrinsic
definition of tangent vectors.
2(iv) No need for explicit equations of the geodesics; just describe how to find
them.
(3) Use 1(i) and 1(v).
(5) Consider the smallest closed ball that contains S. It follows that the flat
torus (with locally Euclidean metric) cannot be isometrically embedded
in R3

3.13 Further Questions


Again a selection of questions from past example sheets that have some interest.

(1) Show that the embedded surface S R3 given by x2 + y 2 + c2 z 2 = 1


is homeomorphic with the standard unit sphere. Use the Gauss-Bonnet
theorem to deduce, for c > 0, that
Z 1
1
(1 + (c2 1)u2 )3/2 du =
0 c

(2) Prove the claim in the notes that a monotomic re-parameterisation (


f with f 0 (t) > 0 for all t) of a curve does not affect its length and that
it has a constant speed (k 0 (t)k = const.) parameterisation.
46 CHAPTER 3. THE GEOMETRY OF SURFACES

(3) Remove the North and South poles from the sphere S, and project the rest
to the cylinder C = {(x, y, z) : x2 + y 2 = 1} as follows. Given a point x in
S, let its image be the intersection with C of the horizontal line through
x that meets the z- axis. (This is Mercators Projection.)
(i) Show, with no calculation, that the image of a great circle in S is not
always a sine wave (when C is unwrapped).
(ii) Show in fact that it is never a sine wave, except in the trivial case
when the great circle is the equator.
(iii) Find parameterisations corresponding to Mercators Projection that
are conformal with meridians and parallels on the sphere being the
images of orthogonal sets of straight lines in the plane.
(4) For each of the following maps : U R3 , find the metric induced on
U and describe its image in R3 .

(i) U = {(u, v) R2 u > v}; (u, v) = (u + v, 2uv, u2 + v 2 )

(ii) U = {(r, t) R2 r > 0}; (r, t) = (r cos t, r sin t, t).

You might also like