Elements of Differential Geometry

May 19, 2006

it is a draft of Lecture Notes of H.M. Khudaverdian.

Manchester, Spring 2006

Contents
1 Curves in En 3
1.1 Preliminary notes. n-dimensional Euclidean space. . . . . . . . 3
1.2 Curves in En . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Reparameterisation . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Tangent line and Osculating plane . . . . . . . . . . . . . . . . 7
1.5 Length of the curve . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 Natural parameterisation . . . . . . . . . . . . . . . . . . . . . 11
1.7 Invariants of curves. Curvature . . . . . . . . . . . . . . . . . 13
1.8 Curvature of curves in E3 and E2 . Signed curvature for plane
curves (curves in E2 ) . . . . . . . . . . . . . . . . . . . . . . . 17
1.9 Integral of curvature along the plane curve. . . . . . . . . . . . 22
1.10 Frenèt frame for the curves in the plane . . . . . . . . . . . . 24
1.11 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2 Surfaces in E3 . First and Second quadratic forms. Gaussian

and mean curvature. 27
2.1 Surfaces, tangent vectors . . . . . . . . . . . . . . . . . . . . . 27
2.2 Reparameterisation in condensed notations . . . . . . . . . . . 29
2.3 Internal and external coordinates of tangent vector . . . . . . 29
2.4 First Quadratic Form . . . . . . . . . . . . . . . . . . . . . . . 30
2.5 Second Quadratic Form . . . . . . . . . . . . . . . . . . . . . 32

1
 2.6 Second quadratic form and curvature of normal sections . . . . 34
2.7 Shape operator and Gaussian and Mean Curvatures . . . . . . 36
2.8 Calculations of Gaussian curvature and Mean curvatures . . . 39

3 Riemannian manifolds 48
3.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2 Volume element in Riemannian manifold . . . . . . . . . . . . 51
3.3 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4 Geodesics and isometries . . . . . . . . . . . . . . . . . . . . 57

4 Parallel transport; Gauss–Bonnet Theorem 59

4.1 Concept of parallel transport . . . . . . . . . . . . . . . . . . . 59
4.2 Parallel transport of vectors tangent to the sphere. . . . . . . 60
4.3 Parallel transport along a closed curve on arbitrary surface. . . 63
4.4 Gauss Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . 64

5 Levi Civita Connection on Riemannian manifold 66

5.1 Affine connection . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.2 Parallel transport II . . . . . . . . . . . . . . . . . . . . . . . 69
5.3 Levi-Civita connection . . . . . . . . . . . . . . . . . . . . . . 70

6 Gaussian map and Gauss Bonnet Theorem 72

6.1 Gaussian map . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.2 Gauß-Bonnet Theorem for convex surface . . . . . . . . . . . . 74

7 Appendices 77
7.1 Geodesics on the sphere and on Lobachevsky plane . . . . . . 77
7.2 Surfaces of constant Gaussian curvatures in E3 . . . . . . . . . 79
7.3 On one beautiful formula . . . . . . . . . . . . . . . . . . . . . 80
7.4 Tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.5 Levi Civita connection II . . . . . . . . . . . . . . . . . . . . . 82
7.6 Riemannian curvature . . . . . . . . . . . . . . . . . . . . . . 84
7.7 Scalar curvature. Gauss Theorema Egregium . . . . . . . . . . 85
7.8 A Tale on Differential Geometry . . . . . . . . . . . . . . 86

2
1 Curves in En
1.1 Preliminary notes. n-dimensional Euclidean space.
Rn is a real vector space of n-tuples of real numbers.
It is convenient to distinguish Rn from the space En —the point space of
n-tuples. Two points a, b ∈ En define a vector in Rn : if a = (a1 , . . . , an ),
b = (b1 , . . . , bn ), then the vector ab attached to the point a has coordinates
= (b1 − a1 , b2 − a2 . . . , bn − an ).
If we fix frame of reference in En then every vector defines a point.
En is Euclidean space: the distance between two points is Pythagorean:
q
||b − a|| = (b1 − a1 )2 + (b2 − a2 )2 + · · · + (bn − an )2 (1)

Coordinates (x1 , . . . , xn ) in En are called cartesian if the distance in these

coordinates is expressed by the formula (257).
Two different cartesian coordinates are related with each other by trans-
lation and orthogonal transformation (rotation+reflection). E.g. if (x, y) are
cartesian coordinates then
µ 0¶ µ ¶ µ ¶ µ 0¶ µ ¶
x a cos ϕ sin ϕ x a + x cos ϕ + y sin ϕ
= + = (2)
y0 b − sin ϕ cos ϕ y0 b − x sin ϕ + y cos ϕ

are cartesian coordinates too. Usually (by default) we will use cartesian
coordinates in En .
We recall also very important formula for scalar (inner) product: Let
x, y be two vectors in En with cartesian cooridinates x = (x1 , . . . , xn ),
y = (y 1 , . . . , y n ). Let ϕ be an angle between these vectors. Then scalar
product of these vectors is equal to

(x, y) = |x||y| cos ϕ = x1 y 1 + . . . xn y n (3)

In particular p
|x| = (x, x) (4)

1.2 Curves in En
A curve in En with parameter t ∈ (a, b) is a continuous map

γ: (a, b) → En r(t) = (x1 (t), . . . , xn (t)), a<t<b (5)

3
 For example consider in E2 the curve

γ: (0, 2π) → E2 r(t) = (R cos wt, R sin wt), 0 ≤ t < 2π (6)

The image of this curve is the circle of the radius R. It can be defined by
the equation:
x2 + y 2 = R 2 (7)
To distinguish between curve and its image we say that curve γ in (5)
is parameterised curve or path. We will call the image of the curve unpa-
rameterised curve or just curve (see for details the next subsection). It is
very useful to think about parameter t as a ”time” and consider parame-
terised curve like point moving along a curve. Unparameterised curve is the
trajectory of the moving point.
We consider only smooth curves, i.e. curves r(t) = (x1 (t), . . . , xn (t)) such
that all functions xi (t), (i = 1, 2, . . . , n) are smooth functions. (Function is
called smooth if it has derivatives of arbitrary order.)

Velocity and acceleration

Let γ : r = r(t) be a curve in En .
Velocity v(t) it is the vector

dr ¡ 1 ¢ ¡ ¢
v(t) = = ẋ (t), . . . , . . . ẋn (t) = v 1 (t), . . . , v n (t) (8)
dt
in En . Velocity vector is tangent vector to the curve.
We consider also acceleration vector:
µ 2 1 ¶
d2 r d x (t) d2 xn (t)
a(t) = 2 = ,..., (9)
dt dt2 dt2

The value of velocity vector is a speed:

p
|v| = (v 1 )2 + · · · + (v n )2 .

Example Consider the following curve in E2

(
x(t) = R cos wt
r(t) : (10)
y(t) = R sin wt

4
One can see that for this curve
µ ¶ µ ¶
−wR sin wt −w2 R cos wt
v(t) = , a(t) = = −w2 r(t) (11)
wR cos wt −w2 R sin wt

We come to the formula well-known from the school: If particle moves

with constant speed along circle then its acceleration (centripetal accelera-
tion) is orthogonal to the tangent vector.
The velocity vector changes only its value under changing of parameter-
isation, i.e. it is multiplied on a scalar coefficient. In general not only the
value of acceleration but its direction changes if we change a parameterisation
of the curve (See the next two subsection)

1.3 Reparameterisation
One can move along trajectory with different velocities, i.e. one can consider
different parameterisation. E.g. consider

( (
x(t) = t x(t) = sin t π
γ1 : ,0 < t < 1 γ2 : ,0 < t <
y(t) = t2 y(t) = sin2 t 2
(12)
Images of these two parameterised curves are the same. In both cases
point moves along a piece of the same parabola but with different velocities.
Definition
Two smooth curves
γ1 : r1 (t) : (a1 , b1 ) → En and
γ2 : r2 (t) : (a2 , b2 ) → En are called equivalent if there exists reparame-
terisation map:
ϕ : (a2 , b2 ) → (a1 , b1 ),
such that
r2 = r1 ◦ ϕ, r2 (t) = r1 (ϕ(t)) (13)
Reparameterisation ϕ is diffeomorphism, i.e. ϕ has derivatives of all orders
and first derivative ϕ0 (t) is not equal to zero.
E.g. curves in (12) are equivalent because a map ϕ(t) = sin t transforms
first curve to the second.

5
 Equivalence class of equivalent parameterised curves is called non-parameterised
curve.
Non-formally: Two curves are equivalent curves (belong to the same
equivalence class) if these parameterised curves ( paths) have the same im-
ages. We come to equivalent curves if we consider the movement along the
same trajectory with different speeds.
Non-parameterised curve—it is trajectory of point.
Or in other words: two equivalent curves have the same image. They de-
fine the same set of points in En . Different parameters correspond to moving
along curve with different velocity.
Example

( (
x = cos θ x=u
, 0 < θ < π, √ , −1 < u < 1, (14)
y = sin θ y = 1 − u2
(
x = tan t π π
√ ,− <t< (15)
y = 1 − tan2 t 4 4
These three parameterised curves,(paths) define the same non-parameterised
curve: the upper piece of the circle: x2 + y 2 = 1, y > 0. In the first case point
moves with constant speed |v(θ)| = 1 and acceleration is orthogonal to the
velocity and it is directed to the centre.
In the second and third case speed is not constant. Hence acceleration is
not orthogonal to the velocity. It has tangential component also.
Very practical observation:
• if the angle between velocity and acceleration vector is right (i.e. the
scalar product (v, a) of acceleration and velocity vectors is equal to
zero) then the speed is constant: velocity vector only may change its
direction.

• if the angle between velocity and acceleration vector is acute (i.e. the
scalar product (v, a) of acceleration and velocity vectors is a positive
number) then the speed is increasing

• if the angle between velocity and acceleration vector is obtuse (i.e. the
scalar product (v, a) of acceleration and velocity vectors is a negative
number) then the speed is decreasing

6
 (To see it: it is very useful to make exercise about moving of point along
ellipse: x = a cos t, y = b sin t (see Homework 2), )

1.4 Tangent line and Osculating plane

What happens with velocity and acceleration if we change parameterisation?
Velocity vector changes its value but does not change direction, in a
sense that new and former vectors are proportional (are on the one line)
Acceleration vector in general changes its direction. (E.g. if particle moves
along circle increasing its speed then acceleration vector possesses not only
orthogonal centripetal component but also tangential component)
Consider in En arbitrary smooth path (parameterised curve) r(t), a <
t < b. Change parameterisation: Suppose t = t(τ ), a0 < τ < b0 . Consider
paths r1 (t) = r(t) and r2 (τ ) = r(t(τ )). It is different parameterisation of the
same non-parameterised curve.
2
Compare velocity and acceleration vectors v(t) = dr(t) dt
, a(t) = d dtr(t)
2 with
dr2 (τ ) d2 r2 (τ )
velocity and acceleration vectors v(τ ) = dτ , a(τ ) = dτ 2 at the same
point r(t) = r(t(τ )).
Simple calculations using chain rule show that

dr2 (τ ) dr(t(τ )) dt(τ ) dr(t) ¯¯

v(τ ) = = = ·
dτ dτ dτ dt t=t(τ )
Hence ¯ ¯
v(τ )¯τ = tτ (τ )v(t)¯t=t(τ ) (16)

For the case of acceleration calculations are simple too. Just little bit
more boring:
µ ¶
d2 r2 (τ ) d dt(τ ) dr(t) ¯¯
a(τ ) = = · =
dτ 2 dτ dτ dt t=t(τ )
µ ¶2
dt(τ ) d2 r(t) ¯¯ d2 t(τ ) dr1 (t) ¯¯
· + ·
dτ dt2 t=t(τ ) dτ 2 dt t=t(τ )
Hence ¯ ¯ ¯
a(τ )¯τ = tτ τ (τ )v(t)¯t=ϕ(τ ) + t2τ (τ )a(t)¯t=ϕ(τ ) (17)
Combine together the formulae (16) and (17):

7
 µ ¶ µ ¶ µ ¶
v0 (τ ) tτ 0 v(t) ¯¯
= · (18)
a0 (τ ) tτ τ t2τ a(t) t=t(τ )

Now we can summarize formulae (16), (17) and (18) in the

Proposition-Definition
Under changing of parameterisation a velocity vector in a given point of
dt
the curve is multiplied on a scalar number dτ . Velocity vector does not change
1
its direction . It changes its value (speed). Velocity vector defines a line which
does not depend on parameterisation. This line is called a tangent line to
the given point of the curve.
Acceleration vector in a given point of the curve in new parameterisa-
tion becomes equal to linear combination of velocity and acceleration vectors
in former parameterisation. The plane formed by acceleration and velocity
vectors remains unchanged under reparameterisation. This plane is called
osculating plane.
Contrary to velocity vector acceleration vector changes it direction under
changing of parameterisation. But nevertheless under changing of parame-
terisation it does not quit osculating plane. 2
Formulae (16), (17) are very important for studying functionals on curves
which does not depend on parameterisation. (A typical problem of differential
geometry of curves)
Consider explicit formulae for tangent line and osculating plane:
Let r = r(t) be a curve in En . Let r0 = r(t0 ) be a point on this curve.

• tangent line: line spanned by velocity vector. If v0 is velocity vector at

the given point r0 = r(t0 ) then this line is defined by the equation:

dr(t) ¯¯
l(t) = r0 + v0 (t − t0 ) , v0 = (19)
dt t=t0

1
We understand direction in a wide sense: two vectors a and b have the same direction
if they are proportional. If a = µb with µ < 0 (this corresponds to the case where we
consider reparameterisation with tτ < 0) then in narrow sense α and b have opposite
directions
2
We do not consider degenerate case where acceleration vector belongs to tangent line
and osculating plane is not defined

8
 • osculating plane: plane spanned by acceleration and velocity vectors. If
v0 is velocity vector at the given point r0 = r(t0 ), and a0 is acceleration
vector at this point then this plane is defined by the equation:
L(ξ, η) = r0 + v0 ξ + a0 η, (ξ, η ∈ R) (20)

Geometrical meaning of tangent line and osculating plane.

Tangent line has first order contact (touching) with a curve at the touch-
ing point:
r(t) = r0 + v(t − t0 ) + o(t) (21)
Now we are ready for giving another definition of tangent line:
Consider two points on the curve the fixed point r(t0 ) and second point
r(t0 + δt) Consider lines which pass through this points. Tangent line at the
point r(t0 ) of the curve r(t) is a limit of these lines when δt → 0.
This definition intuitively is more understandable in spite of the fact that
it is not effective for working out formulae.
Now two words about geometrical meaning of osculating plane.
One can see that oscullating plane has second order contact (touching)
with a curve at the touching point. Indeed Let r(t) be an arbitrary point of
the curve in the vicinity of the given point r0 . Then the distance between
the point r(t) and the oscullating plane L(ξ, η) defined by (20) is less or
equal than distance between points r(t) and the point on the plane with
coordinates ξ = (t − t0 ), η = 21 (t − t0 )2 . Hence
1
d(r(t), L) ≤ |r(t) − r(t0 ) − v(t − t0 ) − (t − t0 )2 | = o((t − t0 )2 )
2
Remark Osculating plane is not defined at the points where acceleration
and velocity vectors are proportional each other, i.e. points where curvature
of the curve is equal to zero.
The following exercise is very instructive:
Exercise Find coefficients A, B, C, D such that plane Ax + By + Cz = D
is an osculating plane to the curve x = f (t), y = g(t), z = z(t)
(See solution in solutions of Homework 2.)

1.5 Length of the curve

Velocity vector measures the length of the curve. One can consider length δl
of the small arc of the curve as δl ≈ |v|δt.

9
 This leads to the definition of the length of the curve: If r(t), a ≤ t ≤ b
is a parameterisation of the curve L and v(t) velocity vector then length of
the curve is equal to the integral of of |v(t)| over curve:
Z b
Length of the curve L = |v(t)|dt = (22)
a
sµ ¶2 µ ¶2 µ ¶2
Z b
dx1 (t) dx2 (t) dxn (t)
+ + ··· + dt
a dt dt dt
Note that formula above is reparameterisation invariant. The length of
the image of the curve does not depend on parameterisation. Indeed con-
sider curve r1 = r1 (t), a1 ≤ t ≤ b1 . Let t = t(τ ), a2 < τ < b2 be another
parameterisation of the curve r = r(t), In other words we have two different
parameterised curves r1 = r1 (t), a1 ≤ t ≤ b1 and r2 = r1 (t(τ )), a2 ≤ τ ≤ b2
such that their images coincide (See (13)). Show that length of the curve
r2 (τ ) coincide with the length of the curve r1 (t). Note that under reparam-
eterisation velocity vector is multiplied on tτ (see (16)):

dr2 dt dr1
v2 = = = tτ (τ )v1 (t(τ ))
dτ dτ dt
Hence
Z b1 Z b2 Z b2 Z b2
dt(τ )
L1 = v1 (t)|dt = |v1 (t)| dτ = |tτ v1 (t)|dτ = |v2 (τ )|dτ = L2 .
a1 a2 dτ a2 a2
(23)
Remark In the formula above we suppose that tτ > 0. If it is not the case then
a2 > b2 and we have to put |tτ | instead tτ and change a sign of integral
Consider also simple examples:
1) interval of line in E2 which connects the points (a1 , b1 ) and (a2 , b2 )
x = a1 + t, y = b1 + kt, where k = ab22 −b 1
−a1
, 0 ≤ t ≤ a2 − a1 . (At t = 1,
x = a1 + (a2 − a1 ) = a2 , y = b1 + k(a2 − a1 ) = b2 )
The integral above gives that
Z a2 q Z a2 √ √ p
L= 2 2
vx + vy dt = 1 + k 2 dt = (a2 −a1 ) 1 + k 2 = (a2 − a1 )2 + (b2 − b1 )
a1 a1

It is just the length which is given by Pythagoras formula.

10
 2) Arc of the circle of the radius R with angle ϕ: x = R cos t, y = R sin t
0 ≤ t ≤ ϕ. Calculating integral
Z ϕq Z ϕq Z ϕp
L= vx2 + vy2 dt = vx2 + vy2 dt = R2 sin2 t + R2 sin2 tdt = Rϕ
0 0 0

we come to the formula which already know.

1.6 Natural parameterisation

Non-parameterised curve can be parameterised in many different ways. (See
e.g.(14)).
Is there any distinguished parameterisation?
Yes. there is.
Let γ : r(t) = (x1 (t), . . . , xn (t)), a < t < b be some curve in En .
Using (22) consider the following parameter on the curve γ:

s(t) =

{length of the arc of the curve for parameter less or equal to t} = (24)
s
Z t µ 1 ¶2 µ 2 ¶2 µ n ¶2 Z t
dx (τ ) dx (τ ) dx (τ )
= + + ··· + dτ = |v(τ )|dτ .
a dτ dτ dτ a
(25)
2
Examples Consider circle: x = R cos t, y = R sin t in E . Then we come
to the obvious answer

s(t) = {length of the arc of the circle for parameter less or equal to t} =
s
Z t µ ¶2 µ ¶2 Z tp Z t
dx(τ ) dy(τ ) 2 2 2 2
+ dτ = R sin τ + R cos τ dτ = Rdτ = Rt
0 dτ dτ 0 a

Another example:
Consider arc of the parabola x = t, y = t2 , 0 < t < 1:

s(t) = {length of the arc of the curve for parameter less or equal to t} =
s (26)
Z t µ ¶2 µ ¶2
dx(τ ) dy(τ )
+ dτ =
0 dτ dτ

11
 Z t√ √ ³ ´
t 1 + 4t2 1 √
2
1 + 4τ dτ = + log 2t + 1 + 4t 2
0 2 4
The first example was very simple. The second is harder to calculate 3 . In
general case natural parameter is not so easy to calculate. But it is very
important for studying properties of curves.
Natural parameterisation is distinguished. Later we will often use the
following very important property of natural parameterisation:
Proposition If a curve is given in natural parameterisation then the
speed is equal to 1 and acceleration is orthogonal to velocity:

v(s) · v(s) ≡ 1, i.e. |v(s)| ≡ 1 , (27)

v(s) · a(s) = 0 , i.e.acceleration is orthogonal to velocity:. (28)

The first relation is just definition of natural parameter and speed. The
second relation means that value of the speed does not change.
Formal proof: let s be a natural parameter for (non-parameterised) curve
r = r(s). Then by definition
Z s
s= |v(τ )|dτ
a

Differentiating by s we see that

ds
= length of velocity vector. It is equal to one:|v(s)| = 1 (29)
dτ
We come to (27). Differentiating the equation (27) by s we come to the
condition (28).
3
Rt√
Denote by I = 0
1 + 4τ 2 dτ . Then integrating by parts we come to:
p Z p Z
2
4τ 2 2
1
I = t 1 + 4t − √ dτ = t 1 + 4t − I + √ dτ .
1 + 4τ 2 1 + 4τ 2
Hence √ Z
t 1 + 4t2 1 1
I= + √ dτ .
2 2 1 + 4τ 2
and we come to the answer.

12
 Note Very useful formulae which follow immediately from definition of
natural parameter: Let t be an arbitrary parameter then it follows from
definition (25) that

ds(t) dt(s) 1
= |v(t)|, = (30)
dt ds |v(s)|

1.7 Invariants of curves. Curvature

How to find invariants of non-parameterised curve, i.e. magnitudes which
depend on the points of non-parameterised curve but which do not depend
on parameterisation?
Answer at the first sight looks very simple: Consider the distinguished
natural parameterisation r = r(s) of the curve. Then arbitrary functions
on xi (s) and its derivatives do not depend on parameterisation. But the
problem is that it is not easy to calculate natural parameter explitly (See e.g.
calculations of natural parameter for parabola in the previous subsection).
So it is preferable to know how to construct these magnitudes in arbitrary
i d2 xi
parameterisation, i.e. construct functions f ( dx
dt
, dt2 , . . . ) such that they do
not depend on parameterisation. We want to define curvature.

First formulate reasonable conditions on curvature: (31)

• it have to be a function of the points of the curve

• it does not depend on parameterisation (or at most only a sign depends

on parameterisation)

• curvature of the line must be equal to zero

• curvature of the circle with radius R must be equal to 1/R

We first give definition of curvature in natural parameterisation. Then

consider it for arbitrary parameterisation.
For a given non-parameterised curve consider natural parameterisation
r = r(s). We know already that velocity vector has length 1 and acceleration
vector is orthogonal to curve in natural parameterisation (see (27) and (28)).
Definition. The curvature of the curve in a given point is equal to the
modulus (length) of acceleration vector in natural parameterisation.

13
Namely, let r(s) be natural parameterisation of this curve. Then curvature
at every point r(s) of the curve is equal to the length of acceleration vector:

d2 r(s)
k = |a(s)|, a(s) = (32)
ds2
First check that it corresponds to our intuition (see reasonable conditions
(31)).
It does not depend on parameterisation by definition.
ItPis evident that for the line in normal parameterisation xi (t) = xi0 + bi t
with bi bi = 1 the acceleration is equal to zero.
Now check that the formula (32) gives a natural answer for circle: if radius
is equal to R then curvature k is equal to 1/R.
For circle of radius R in natural parameterisation
s s
r = r(s) = (x(s), y(s)), where x(s) = R cos , y(s) = R sin
R R
(length of the arc of the angle θ of the circle is equal to s = Rθ.) Then
µ ¶
dr2 (s) 1 s 1 s
a(s) = = − cos , − sin
ds2 R R R R
and for curvature
1
k = |a(s)| = (33)
R
we come to the answer which agrees with our intuition.
Geometrical meaning of curvature: We will consider this question
later in more detail. But even now it is easy to see from this example that k1
is just a radius of the circle which has second order touching to curve.
Example: parabola: y = ax2 .(a > 0) Calculate the curvature at the
point (0, 0).
The straightforward calculations are very long. Later we will return to
this example when we find a formula expressing curvature in arbitrary pa-
rameterisation. But try to guess an answer using geometrical meaning of
curvature that k1 is just a radius of the circle which has second order touch-
ing to curve.
Consider a circle given by equation

x2 + (y − R)2 = R2 (34)

14
It is evident that this circle touches parabola

y = ax2 (35)

at the point (0, 0). Choose radius R such that this will be a touching of
the second order. Opening brackets in (34) we come to quadratic√equation
y 2 − 2yR + x2 = 0 and expressing y via x we come to y = R ± R2 − x2 .
The lower part of circle touches to parabola. Hence:
Ãr !
√ x 2
y = R − R 2 − x2 = R − R 1− 2 = (36)
R
µ ¶
x2 2 x2
R−R 1− + o(x ) = + o(x2 ) (37)
2R2 2R
1 1
Comparing with (35) we see that a = 2R
and k = R
= 2a.

Formula (5) looks simple but it is hard to work with it (because we have
to consider a curve in a natural parameterisation).
Try to rewrite formula for curvature in arbitrary parameterisation.
Let [γ] be non-parameterised simple regular curve (equivalence class of
parameterised curves). Let r(t) be any parameterisation of this curve. Con-
sider arbitrary point r(t0 ) of this curve.
Straightforward attack. Instead considering explicitly natural param-
eterisation of the curve we just try to rewrite the formula in definition (32)
using chain rule and the relation (30): Using chain rule we calculate ac-
cording definition (32) the curvature. To avoid confusion we denote here
by v = v(t), a = a(t) velocity and acceleration vectors in a given parame-
terisation t. We denote by A(s) the acceleration vector in natural parame-
2
terisation. First using (30) we calculate the vector A(s) = dds2r (the vector
of acceleration in normal parameterisation). Then we calculate its length
(curvature):
µ ¶ µ ¶ µ ¶
d2 r(t(s)) d dt dr(t) d v(t) 1 d v(t)
A(s) = = = = =
ds2 ds ds dt ds |v(t)| |v(t)| dt |v(t)|
µ ¶
1 a v(a · v) av2 − v(av)
− = (38)
|v| |v| |v|3 v4

15
v2 = (v, v) = |v|2 , (v a) are inner (scalar) products.
Curvature is equal to
p
√ a2 v2 − (a · v)2
k(r(t)) = |A| = A · A = (39)
|v|3

Try to understand the geometrical meaning of the final answer. Show

that the numerator of the fraction in RHS of the last formula is nothing but
area of the parallelogram formed by the vectors v, a
Indeed recall the formula from elementary geometry: Area of the par-
allelogram is equal to the height multiplied on the length of the side, i.e.
area of the parallelogram formed by the vectors v, a is equal to the length of
the vector v multiplied on the length of the vector a and multiplied on the
sinus of the angle between these vectors. On the other hand inner product
of vectors v, a is equal to the product of their length on the cosines of the
angle between them: v · a = |v||a| cos ϕ. Hence

area of the parallelogram formed by the vectors v, a =

s µ ¶2
p v·a p
2
|v||a| sin ϕ = |v||a| 1 − cos ϕ = |v||a| 1 − = a2 v2 − (a · v)2
|v||a|
(40)
We come to the
Theorem Let [γ] be non-parameterised curve (equivalence class of pa-
rameterised curves). Let r(t) be any parameterisation of this curve. Then
curvature at the point r(t0 ) of the curve is equal to the area of parallelogram
formed by the vectors a, v at this point divided by the cube of the speed v:
¯ Area of parallelogram formed by the vectors v and a
k ¯r=r(t0 ) = =
Cube of the speed
q
v2 a2 − (v · a)2
(41)
|v|3
¯
dr ¯
where vectors α, v are considered at the point r(t0 ): v = dt t=t0
, a =
d r¯
2 ¯
dt2 .
t=t0
Previous calculations were in fact proof of this Theorem. But these cal-
culations were just straightforward calcualtions.

16
 One can give the proof of the Theorem independent on previous calcula-
tions which is beautiful and illuminating.
Proof: In natural parameterisation according to (27) acceleration is or-
thogonal to velocity and speed is equal to 1: (a, v) = 0, |v| = 1. Hence RHS
of (41) coincides with (32) in natural parameterisation. It remains to prove
that it is independent on parameterisation. Consider arbitrary reparameter-
isation. According to formulae (18) v → tτ v, a → t2τ a + tτ v. The area of
parallelogram will be multiplied on t3τ . (Transformation v → v, a → a + µv
does not change the area). The denominator of the fraction (41) will be
multiplied on the same number t3τ . Hence RHS of (18) is independent of
parameterisation.
Formula (41) is a workable definition of curvature.

1.8 Curvature of curves in E3 and E2 . Signed curvature

for plane curves (curves in E2 )
If r = r(t) is curve on E3 then one can express area of parallelogram via cross-
product. Recall that if a, b are two vectors in three-dimensional Euclidean
space. then one can consider their cross-product (vector product): vector a×
b (sometimes denoted by [a, b] which is defined by the following properties:

• its length is equal to the area of parallelogram formed by the vectors

a, b

• it is orthogonal to the plane spanned by the vectors a, b

• The vectors {a, b, a × b} form an basis such that it has the same
orientation that the basis {ex , ey , ez }.

One can see that

 
ex ey ez
a×b = det  ax ay az  = (ay bz −az by )ex +(az bx −ax bz )ey +(ax by −ay bx )ez ,
bx by bz
(42)
Its length is given by the formula
q
|a × b| = (ay bz − az by )2 + (az bx − ax bz )2 + (ax by − ay bx )2 = (43)

17
 q¡ ¢¡ ¢ p
b2x + by2 + b2z − (ax bx + ay by + az bz )2 = a2 b2 − (ab)
a2x + a2y + a2z
(44)
3
We see that for curve in E curvature (41) can be expressed via cross-
product in a very compact way:
|v × a|
k= (45)
|v|3
Now consider a case if curve belongs to the plane, is so called plane curve.
WLOG suppose that curve belongs to the plane OXY .
Recall basic formulae about area of parallelogram: area of parallelogram
formed by vectors a, b (a, b ∈ OXY is equal to the determinant of the matrix
µ ¶
ax ay
(46)
bx by

(Compare this formula with (42) in the case if az = bz = 0)

One have to distinguish between area of oriented parallelogram and non-
oriented parallelogram. Area of non-oriented parallelogram is non-negative
number. It is equal to module of the determinant (46).
The area of oriented parallelogram is defined by the orientation of pair of
vectors (a, b).
The determinant (46) may take positive or negative values. E.g. for the
vectors a = ex , b = ey determinant is equal to 1, if we change ordering and
consider a = ey , b = ex then we come to −1.
Assume that basic vectors ex , ey are oriented in standard way, i.e. rotation
from ex to ey is counter clock wise. Then it is easy to see that
Area of oriented parallelogram formed by vectors a, b is positive if ro-
tation from a to b is counter clock wise. Respectively the area of oriented
parallelogram formed by vectors a, b is negative if rotation from a to b is
clock wise.
We say that the ordered pair of vectors a, b on the plane OXY has
positive orientation if this ordered pair has the same orientations as the
ordered pair of basic vectors ex , ey , i.e. rotation from a to b is counter clock
wise and determinant (46) is positive. Respectively we say that the ordered
pair of vectors a, b on the plane OXY has negative orientation if this pair has
the same orientations as the ordered pair of basic vectors ey , ex , i.e. rotation
from a to b is clock wise and determinant (46) is negative.
Example The pairs {a, b} and {b, a} have opposite orientations. The pairs

18
{a, b} and {−a, b} have opposite orientations. The pairs {a, b} and {−b, a}
have the same orientations.

Return to curvature.
For curve in OXY r(t); x = x(t), y = y(t) curvature (length of the ac-
celeration vector in normal parameterisation) according to (41) and (46) is
equal to the area of non-oriented parallelogram divided by the cube if the
velocity:
µ ¶
xt y t
det
xtt ytt |xt ytt − xtt yt |
k=± = (47)
|v(t)| 3 (x2t + yt2 )3/2
Signed curvature for plane curve
For plane curve in oriented plane OXY one can consider so called signed
curvature (curvature with sign), i.e. curvature defined by the orientation of
ordered pair (v, a) which can be positive or negative. It is equal to the area
of oriented parallelogram formed by velocity and acceleration vectors divided
by the cube of the velocity:
µ ¶
xt y t
det
xtt ytt xt ytt − xtt yt
ksign = = 2 , k = |ksign | (48)
|v(t)| 3 (xt + yt2 )3/2

Signed curvature is defined by cross-product (area of oriented parallel-

ogram). Usual curvature is defined by modulo of cross-product (area of
non-oriented parallelogram).
Curvature coincides with signed curvature up to a sign.
Signed curvature is positive (coincides with curvature) if vectors v, a are
oriented as ordered pair of basic vectors ex , ey :

ksign = k, if rotation from v to a is counter clock wise. (49)

Signed curvature is negative (coincides with curvature up to a sign) if

vectors v, a are oriented as ordered pair of basic vectors ey , ex :

ksign = −k, if rotation from v to a is clock wise. (50)

We know that curvature does not depend on parameterisation. What about

signed curvature? If curvature is equal to k then signed curvature can be

19
equal to k or to −k. If we change parameterisation of curve t → −t, i.e.
roughly speaking move along the curve in opposite direction then signed
curvature changes the sign (the vectors (−v, a) have orientation opposite to
(v, a)). Changing t → −t it means changing orientation of curve.
Signed curvature does not depend on parameterisation up to a sign. It
changes a sign if we change orientation of the curve.
Before considering examples one additional remark:
Remark In the case if at the given point acceleration vector is orthogonal
to the velocity vector then area of parallelogram is just equal to |a||v| and
curvature at the given point is equal to the length of acceleration divided by
the square of velocity:
¯ |a|
k ¯r(t0 ) = in the case if a is orthogonal to v in the given point r(t0 )
|v|2
(51)
Note that the condition that acceleration is orthogonal to velocity at the
given point does not necessarily implies that it is orthogonal to the velocity
at the all points in the vicinity of this point (like for natural parameter (see
(28)). E.g. consider parabola in the parameterisation x = t, y = t2 . One can
see that at the point (0, 0) and only at this point acceleration is orthogonal
to velocity.
Example 1 Consider ellipse
x2 y 2
+ 2 =1 (52)
a2 b
(a, b > 0) Consider the following parameterisation of this ellipse:
2π
r = r(t) : x = a cos wt, y = b sin wt, 0 ≤ t < (53)
w
One can see that t is not natural parameter. So we cannot apply formula
from definition. Make calculations according to Theorem:
µ ¶ µ ¶
−aw sin wt, −aw2 cos wt,
v(t) = , a(t) = (54)
wb cos wt −bw2 sin wt
µ ¶
xt , y t
The area of oriented parallelogram is equal to S(v, a) = det =
xtt , ytt
abw3 . It is positive. Signed curvature coincides with curvature. (This cor-
responds to the fact that the point moves counter clock-wise in (53), ac-
celeration is directed in the interior of ellipse. The pair (v, a) has positive

20
orientation. Hence curvature is equal to the signed curvature according to
(49):
ab
k = ksign = p 2
(55)
(a sin wt + b2 cos2 wt)3
2

At the points x = 0 or y = 0 acceleration is orthogonal to velocity and

curvature is equal to b/a2 (x = 0), a/b2 (y = 0).
If we change parameterisation then curvature remains the same..
Signed curvature ksign remains the same under changing of parameteri-
sation only up to the sign. Change orientation of the curve, i.e. consider
parameterisation such that velocity vector moves clock wise, e.g. t → −t in
(53). Then signed curvature ksign will be negative. Point moves clock-wise,
acceleration again is directed in the interior of ellipse. The pair (v, a) has
negative orientation. Hence signed curvature becomes negative according to
(50):
ab
ksign = −k = − p 2
(56)
(a sin wt + b2 cos2 wt)3
2

Example 2 Consider parabola y = ax2 . Assume that a > 0. Consider

parameterisation x = t, y = at2 . t is not natural parameter (see footnote
in the subsection ”Natural parameterisation”). So we cannot apply formula
from definition (32). Make calculations according to Theorem:
µ ¶ µ ¶
1, 0,
v(t) = , a(t) = (57)
2at 2a
µ ¶ µ ¶
xt , y t 1, 2at
The area of parallelogram is equal to S(v, a) = det = det =
xtt , ytt 0, 2a
2a. It is positive since a > 0. Hence curvature is equal to signed curvature
and it is equal to
2a 2a
k= = 3 . (58)
|v|3 (1 + 4a2 t2 ) 2
In particular at the point (0, 0) where t = 0, k = 2a. Note that acceleration
is orthogonal to velocity at the point (0.0) and only at this point!. Thus we
can calculate curvature at the point (0.0) immediately by (51):
2a
k= = 2a .
1

21
One can see that curvature coincides with signed curvature because the pair
(v, a) is positive oriented:in standard parameterisation: x = t, y = at2 veloc-
ity is directed to the right, acceleration- up.
If a < 0 then signed curvature is negative.
(Compare these calculations with calculation of curvature via touching
circle in (37))
Example 3. Curve is a interval of line if and only if its curvature is
equal to zero. Intuitively it is almost evident. Prove it. If curve is an
interval of line then obviously one can consider parameterisation xi = ai +bi t.
Acceleration is equal to zero. Hence curvature is equal to zero. Prove inverse
implication. If curvature is equal to zero then by (41) acceleration vector is
parallel to velocity vector in arbitrary parameterisation. In particularly in
natural parameterisation acceleration vector is equal to zero. Hence velocity
i
vector is constant: dx ds
= ci . Hence xi (s) = xi0 + ci s. It is just equation of
the line.

1.9 Integral of curvature along the plane curve.

Why so much attention to signed curvature for plane curve?
The signed curvature has the following beautiful property:
Let γ : r = r(t) a ≤ t ≤ b be plane curve on OXY . Let ksign (t) be a
signed curvature for this curve.
Consider the following integral:
Z t2 Z t2
Iγ = ksign (t)ds(t)dt = ksign (t)|v(t)|dt (59)
t1 t1

One can easy to see that this integral is independent on reparameterisation

up to the sign.
It turns out that this integral possesses the following property:
The value of this integral does not change if we deform the curve in way
such that velocity vector in initial and final points remains unchanged.
Clarify our statement: Consider on the points of the curve r(t) the con-
tinuous function ϕ(t) such that it is equal (up to 2πk) to the angle between
velocity vector v(t) and OX axis:
yt
ϕ(t) : tan ϕ(t) = , t 1 ≤ t ≤ t2 (60)
xt

22
E.g.for ellipse (53) ϕ(t) = wt. It changes from 0 till 2π. For parabola x =
t, y = at2 , ϕ(t) = arctan 2at. It changes from − π2 = ϕ(−∞) till π2 = ϕ(∞)
Denote by ϕ1 = ϕ(t1 ), ϕ2 = ϕ(t2 ). Then the following remarkable identity
holds:
Z t2 Z t2
dϕ(t)
Iγ = ksign (t)ds(t)dt = dt = ϕ2 − ϕ1 (61)
t1 t1 dt
Note that straightforward calculation often is difficult.
Proof When the result is formulated it is evident. Calculate the derivative
of (60):
µ ¶
dϕ d tan ϕ dϕ d yt 1 xt ytt − xtt yt
= = 2 =
dt dt d tan ϕ dt xt 1 + t2 y x2t + yt2
x t

We see that it is equal to integrand ksign (t)|v(t)| in (61).

In the case if signed curvature ksign (t) does not change the sign (is positive
or negative for all t) then this formula calculates integral of curvature up to
the sign:
Z
k(s)ds = |Iγ | if ksign (t) does not change the sign (62)

Consider two examples.

Example 1. Ellipse. Then use formulae of previous subsection (see
(53)–(55)). Consider parameterisation ϕ = wt Then
we come to the formula:
Z 2π
ab
Ielipse = 2 dϕ = 2π
0 a sin ϕ + b2 cos2 ϕ
2

For ellipse signed curvature has the same sign for all the points. Hence
integral of curvature over ellipse is equal to Ielipse = 2π (Try to calculate this
integral in other way)
One can see that if γ is arbitrary closed convex curve, (i.e. its interior is
convex domain) then integral of curvature along this curve will be equal to
2π.
Geometrical meaning: k(s)∆s is equal to the rotation of the normal vec-
tor. We can interpret curvature as velocity of instantaneous rotation of nor-
mal vector (see in more detail the next subsection)

23
1.10 Frenèt frame for the curves in the plane
this subsection is compulsory only for MSC students
Let γ be plane curve– a curve in E2 . Let r(s) be its natural parameterisation
(0 ≤ s ≤ s0 ). Consider velocity vector v(s). Acceleration vector a(s) = dv ds is
orthogonal to velocity vector and its length is equal to the curvature k(s).
Suppose that curvature is not equal to zero for every s ∈ [0, s0 ]:
¯ ¯
k(s) = ¯a(s)¯ 6= 0 (63)

Then acceleration vector a(s) defines unit vector n(s) which is orthogonal 4 to v:

dv(s)
= k(s)n(s) . (64)
ds
Hence at every point of the curve where curvature k(s) 6= 0 we defined a basis
(frame) {v(s), n(s)} adjusted to the curve. This frame is orthonormal frame. It is
called Frenèt frame.
This frame moves along the curve and rotates in the plane. One can show that
dn(s)
= −k(s)v(s). (65)
ds
This follows from (64). Indeed consider expansion of R.H.S. of the equation
(65) with respect to the basis v(s), n(s) in the plane:

dn(s)
= α(s)v(s) + β(s)n. (66)
ds
Multiplying both parts of this equation on n(s) we come to:
dn 1 d 1 d
n· = (nṅ) = (1) = 0 = α(n · v) + β(n · n) = β
ds 2 ds 2 ds
because v is orthogonal to n: (v · n) = 0. To prove that α(s) = −k(s) differentiate
the identity (v · n) = 0 by s:
µ ¶ µ ¶
d dv dn dn
0= (vṅ) = n+ v· = k(s)(n · n) + v ·
ds ds ds ds
¡ dn ¢ ¡ dn ¢
It follows from (66) that v · ds = α(s). Hence α(s) = v · ds = −k(s) We come
to (65).
Equations (64), (65) are called Frenèt equations for Frenèt frame.
4
Note that there are two unit vectors which are orthogonal to v. The direction of unit
vector vector n in (64) is defined by the direction of acceleration a(s).

24
 Their geometrical meaning is that
curvature k(s) defines the speed of instantaneous rotation of Frenèt frame at the
point r(s)
To see it rewrite these equations in the following form:
µ ¶ µ ¶ µ ¶µ ¶ µ ¶µ ¶
d v(s) k(s)n(s) 0 k(s) v(s) 0 1 v(s)
= = = k(s)
ds n(s) −k(s)v(s) −k(s) 0 n(s) −1 0 n(s)
(67)
Consider Frenèt frame (v(s), n(s)) moving along the curve for 0 < s < a.
Denote by ϕ(s) be an angle of rotation: Then
µ ¶ µ ¶µ ¶
v(s) cos ϕ(s) sin ϕ(s) v
= , (68)
n(s) − sin ϕ(s) cos ϕ(s) n
µ ¶
v
where we denote by the Frenèt frame at the initial point r(0)
n
Differentiate formula above along s:
µ ¶ µ ¶µ ¶
d v(s) d cos ϕ(s) sin ϕ(s) v
= = (69)
ds n(s) ds − sin ϕ(s) cos ϕ(s) n
µ ¶µ ¶
−ϕ̇(s) sin ϕ(s) ϕ̇(s) cos ϕ(s) v
= (70)
−ϕ̇(s) cos ϕ(s) −ϕ̇(s) sin ϕ(s) n
µ ¶ µ ¶µ ¶ µ ¶µ ¶
0 ϕ̇(s) cos ϕ(s) sin ϕ(s) v 0 1 v(s)
· = ϕ̇(s) (71)
−ϕ̇(s) 0 − sin ϕ(s) cos ϕ(s) n −1 0 n(s)
Compare this for formula in (67). We see that:
µ ¶µ ¶ µ ¶µ ¶
0 1 v(s) 0 1 v(s)
ϕ̇(s) = k(s) (72)
−1 0 n(s) −1 0 n(s)

i.e.
ϕ̇(s) = k(s) (73)
Curvature measures velocity of instantaneous rotation of the frame (Compare
these considerations with considerations of previous subsection)
Comparing formulae above for Frenét frame rotation with formulae (59)–(61)
we see that in particularly for for convex curve (closed curve such that its interior
is convex domain) Frenèt frame makes rotation on the angle 2π.
It is easy to see that for an arbitrary closed oriented curve on the plane the
rotation angle is equal to 2πn, where n is equal to ”winding number”

25
 Remark In formulae above we assume that curvature is not equal zero at all
s (see footnote to the formula (64)). If k(s) = 0 at some point s = s0 then one
needs apriori definition of direction of orthogonal vector. In the case if curve is
in E2 then the direction of normal vector can be defined by orientation in E2
and orientation of curve: We choose n such that ordered pair (v, n) is positively
oriented. It is easy to see that relative ksign curvature considered in the subsections
above can be defined as a proportionality coefficient between acceleration vector
a(s) and normal unit vector n(s). If point moves along curve counter clockwise,
ksign (s) : a(s) = ksign (s)n(s) (74)
where n is chosen in the way that rotation from the vector v to vector n is counter
clock wise. (Compare this definition with definition above.)
Remark It is easy to see that one can consider instead curvature (usual or
signed) just a vector A(s) which is equal to acceleration vector in normal param-
eterisation. The vector A(s) is invariant (if we change s → a − s then A remains
unchanged.) It is special case of second quadratic form.

1.11 Torsion
Considering higher derivatives of the curve one can consider Frenet frame for the
curve in arbitrary n-dimensional Eucliden space En .
Consider very briefly the case of E3 .
Let r(s) be a curve in natural parameterisation in E3 . Suppose that curvature
is not equal to zero at all the points. In the same way as in(64) consider unit
vector n(s) such that a(s) = k(s)n(s), where a(s) is acceleration vector. Vectors
v(s), a(s) form orthonormal basis in osculating plane. Consider third unit vector
t(s) = v(s) × a(s). This vector is rothogonal to osculating plane. Three vectors
{t(s), v(s), a(s)} form orthonormal basis in E3 adjusted to the curve. It is Frenet
basis.
We have by definition of n(s)
d
v(s) = k(s)n(s). (75)
ds
In the same way like in (65) one can deduce that
d
n(s) = −k(s)v(s) + κ(s)t(s) . (76)
ds
Considering Frenet basis one can deduce the following analogue of equations (67):
   
v(s) k(s)n(s)
d 
n(s) = −k(s)v(s) + κ(s)t(s) (77)
ds
t(s) −κ(s)n(s)

26
 Definition Proportionality coefficient κ(s) in formulae (76), (77) is called a
torsion of the curve.
In the same way as curve belongs to the line if and only if its curvature is equal
to zero (see example in the subsection ”Curvature”), one can see that torsion is
equal to zero if and only if curve in the space belongs to plane.

2 Surfaces in E3. First and Second quadratic

forms. Gaussian and mean curvature.
2.1 Surfaces, tangent vectors
Let r(u, v) = (x(u, v), y(u, v), z(u, v) be a parameterisation of surface in E3 .
(x, y, z are cartesian coordinates in E3 ) One can consider tangent vectors:
   
∂x(u,v) ∂x(u,v)
∂r(u, v)  ∂y(u,v)
∂u
 ∂r(u, v)  ∂y(u,v)
∂v

=  ∂u  , =  ∂v  (78)
∂u ∂z(u,v) ∂v ∂z(u,v)
∂u ∂v

Later we often use shorter notations:

   
xu (u, v) xv (u, v)
∂r(u, v)  ∂r(u, v) 
ru = = yu (u, v)  , rv = = xv (u, v) (79)
∂u ∂v
zu (u, v) xv (u, v)

For example consider surface defined by the equation z − F (x, y) = 0. It

can be parameterised:
   
x(u, v) u
r(u, v) = y(u, v) =  v  (80)
z(u, v) F (u, v)
   
1 0
ru =  0  , rv =  1  (81)
Fu (u, v) Fv (u, v)
Example Consider sphere of radius R:
   
x(θ, ϕ) R sin θ cos ϕ
r(θ, ϕ) = y(θ, ϕ) =  R sin θ sin ϕ  . (82)
z(θ, ϕ) R cos θ

27
    
R cos θ cos ϕ −R sin θ sin ϕ
rθ =  R cos θ sin ϕ  , rϕ =  R sin θ cos ϕ  (83)
−R sin θ 0
Tangent plane
Let p be a given point of the surface M . Consider the plane formed by
the vectors which are adjusted to the point p and tangent to the surface M .
We call this plane plane tangent to M at the point p and denote it by Tp M .
Let r = r(t) be a curve belonging to the surface C, i.e. r(t) = r(u(t), v(t)).
Let p = r(t0 ) be any point on this curve. Then vector
dr dr(u(t), v(t))
rt = = (84)
dt dt
belongs to the tangent plane Tp M .
Basis in tangent plane
Let r = r(u, v) be a parameterisation of the surface M . Then for every
point p ∈ M one can consider a basis in the tangent plane Tp M adjusted to
the parameters u, v. Every vector X ∈ Tp M can be expanded over this basis:
X = Xu ru + Xv rv , (85)
where Xu , Xv are coefficients, components of the vector X.
The basis vector ru ∈ Tp M ,is velocity vector for the curve u = u0 + t, v =
v0 , where (u0 , v0 ) are coordinates of the point p. Respectively the basis vector
rv ∈ Tp M ,is velocity vector for the curve u = u0 , v = v0 + t, where (u0 , v0 )
are coordinates of the point p.
Note that for the vector (84) components Xu , Xv are equal to Xu =
ut , Xv = vt because
dr dr(u(t), v(t))
rt = = = ut ru + vt rv (86)
dt dt
We begin to use condensed notations. In condensed notation instead denoting
coordinates by (u, v) we often denote them by uα = (u1 , u2 ). Respectively
we denote by
dr
rα = α , ru = r1 , rv = r2
du
The formula (93) for tangent vector field will have the following appearance:
X = X α rα = X 1 r1 + X 2 r2 , (X 1 = Xu , X 2 = Xv ) (87)

28
The formula (84) will have the appearance:
dr duα
= uαt rα = rα
dt dt
When using condensed notationsP we usually omit explicit summation sym-
bols. E.g. we write u rα instead 2i=1 uα rα or u1 r1 + u2 r2
α

One can consider also differentials duα = (du1 , du2 ):

duα (rβ ) = δβα : du1 (r1 ) = du2 (r2 ) = 1, du1 (r2 ) = du2 (r1 ) = 0 (88)

2.2 Reparameterisation in condensed notations

Study how formulae above change if we change parameterisation:
Let uα = uα (ξ p ) (α = 1, 2, , p = 1, 2). It is condensed notation for changing of
parameters u1 , u2 on new parameters ξ 1 , ξ 2 : u1 = u1 (ξ 1 , ξ 2 ), u2 = u2 (ξ 1 .ξ 2 ).
Then for (88)
µ ¶ µ ¶
a ∂uα p 1 ∂u1 1 ∂u1 2 2 ∂u2 1 ∂u2 2
du = dξ , du = dξ + 2 dξ , du = dξ + 2 dξ (89)
∂ξ p ∂ξ 1 ∂ξ ∂ξ 1 ∂ξ
For basis vectors:
∂ξ p (u) ∂r
ra = ξpa rp = , i.e. (90)
∂uα ∂ξ p
∂r1 ∂ξ 1 (u) ∂r ∂ξ 2 (u) ∂r ∂r1 ∂ξ 1 (u) ∂r ∂ξ 2 (u) ∂r
r1 = 1
= 1 1
+ 1 2
, r2 = 2
= 2 1
+
∂u ∂u ∂ξ ∂u ∂ξ ∂u ∂u ∂ξ ∂u2 ∂ξ 2
and for tangent vectors:
0 0 0
X = X α rα = X α ξαp rp = (X α ξαp ) rα ⇒ X p = ξαp X α , e.g. X 1 = ξ11 X 1 + ξ21 X 2 (91)
For curve:
dr
= uαt rα = uαt ξαp rp (92)
dt

2.3 Internal and external coordinates of tangent vector

Look again on the formulae (93), (87). Denote Xu = a, Xv = b

     
xu (u, v) xv (u, v) axu (u, v) + bxv (u, v)
X = Xu ru +Xv rv = aru +brv = a  yu (u, v) +b xv (u, v) =  ayu (u, v) + byv (u, v) 
zu (u, v) xv (u, v) azu (u, v) + bzv (u, v)
(93)

29
(a, b) can be considered as internal coordinates of the tangent vector X. Co-
ordinates of the vector X in the ambient space

(axu (u, v) + bxv (u, v), ayu (u, v) + byv (u, v), azu (u, v) + bzv (u, v))

can be considered as external coordinates of the tangent vector X. Ant living

on the surface deals with the vector X in terms of coordinates (a, b). Exter-
nal observer which contemplates the surface embedded in three-dimensional
space deals with vector X as with vector with external coordinates
(axu (u, v) + bxv (u, v), ayu (u, v) + byv (u, v), azu (u, v) + bzv (u, v)).

2.4 First Quadratic Form

Definition
Let M : r = r(u, v) be a surface embedded in E3 .
First quadratic form defines length of the tangent vector to the surface in
internal coordinates and distance between points of the surface.
The first quadratic form at the point r = r(u, v) is defined by symmetric
matrix:
µ ¶ µ ¶
G11 G12 (ru , ru ) (ru , rv )
= , Gαβ = (rα , rβ ) (94)
G12 G22 (ru , rv ) (rv , rv )

where ( , ) is a scalar product:

G = Gαβ duα duβ = G11 du2 + 2G12 dudv + G22 dv 2 (95)

Consider a vector X = X α rα = aru + brv tangent to the surface M .

The square of the length |X| of this vector

|X|2 = (X, X) =< aru + brv , aru + brv >= a2 (ru , ru ) + 2ab(ru , rv ) + b2 (rv , rv )
(96)
It is just equal to the value of the first quadratic form on this tangent vector:
µ ¶ µ ¶
¡ ¢ G11 G12 a
G(X, X) = a, b · · = G11 a2 + 2G12 ab + G22 b2 (97)
G12 G22 b

External observer (person living in ambient space E3 ) calculate the length

of the tangent vector using formula (96). An ant living on the surface calcu-
late length of this vector in internal coordinates using formula (97). External

30
observer deals with external coordinates of the vector, ant on the surface with
internal coordinates.
If X, Y are two tangent vectors in the tangent plane Tp C then G(X, Y)
at the point p is equal to scalar product of vectors X, Y:
(X, Y) = (X 1 r1 + X 2 r2 , Y 1 r1 + Y 2 r2 ) = (98)
X 1 (r1 , r1 )Y 1 + X 1 (r1 , r2 )Y 2 + X 2 (r2 , r1 )Y 1 + X 2 (r2 , r2 )Y 2 =
X α (rα , rβ )Y β = X α Gαβ Y β = G(X, Y)
Remark We identify quadratic forms and corresponding symmetric bilinear
forms 5
First quadratic form and length of the curve
Let r(t) = r(u(t), v(t)) a ≤ t ≤ b be a curve on the surface.
The first quadratic form measures the length of velocity vector at every
point of this curve. Thus we come to the formula for length of the curve.
Velocity of this curve at the point r(u(t), v(t)) is equal to
dr(t)
v = X = ξru + ηrv where ξ = ut , η = vt : v= dt
= ut ru + vt rv .
The length of the curve is equal to
Z b Z bp Z bp
L= |v(t)|dt = (v(t), v(t))dt = (ut ru + vt rv , ut ru + vt rv )dt =
a a a
(100)
Z bq
(ru , ru )u2t + 2(ru , rv )ut vt + (rv , rv )vt2 dτ =
a
Z bp
G11 u2t + 2G12 ut vt + G22 vt2 dt (101)
a
An external observer will calculate the length of the curve using (22). An
ant living on the surface calculate length of the curve via first quadratic form
using (101): first quadratic form defines Riemannian metric on the surface:
ds2 = Gik dui duk = G11 du2 + 2G12 dudv + G22 dv 2 (102)
5
Bilinear symmetric form B(X, Y) = B(Y, X) defines quadratic form Q(X) =
B(X, X). Quadratic form satisfies the condition Q(λX) = λ2 Q(X) and so called par-
allelogram condition
Q(X + Y) + Q(X − Y) = 2Q(X) + 2Q(Y) (99)

31
 Invariance of first quadratic form
We give above an invariant definition of the first quadratic form. Double
check that it is reparameterisation invariant: Let ξ 1 , ξ 2 be new parameters:
uα = uα (ξ p ).
Let first quadratic form G is equal to duα Gαβ duβ in parameters (u1 , u2 )
and it is equal to dξ p Gpq
0
dξ q in new parameters (ξ 1 , ξ 2 ), where Gαβ = (rα , rβ ) =
∂r ∂r 0 ∂r ∂r
( ∂uα , ∂uβ ) and respectively Gpq = (rp , rq ) = ( ∂ξ p , ∂ξ q ). We have to check that

duα Gαβ duβ = dξ p Gpq 0

dξ q . Using (89) and (90) we see that

duα Gαβ duβ = duα (rα , rβ )duβ = dξ p (rp , rq )dξ q = dξ p Gpq

0
dξ q (103)

2.5 Second Quadratic Form

First quadratic form and corresponding symmetric bilinear form measure
length of tangent vector and scalar product of tangent vectors and length of
the curve.
Now we define the second quadratic form which measures curvature. For
curves in En we define curvature via acceleration and velocity vectors. For
different curves beginning at the giving point curvature is different. On the
other hand it has to depend on second derivatives.
We give a formal definition for second quadratic form and show that it is
reparameterisation invariant. Then we will reveal its geometrical meaning.
Definition–Proposition
Let M be a surface given by parameterisation r = r(u, v). Consider at
every point of surface the following form:

A = (n, ruu )du2 + 2(n, ruv )dudv + (n, rvv )dv 2 (104)

where n is a unit normal vector and

∂ 2r ∂2r ∂ 2r
ruu = , r uv = , r vv =
∂ 2u ∂u∂v ∂ 2v
This expression defines the second quadratic form and corresponding bilinear
form on tangent vectors:

A = Auu du2 + 2Auv dudv + Avv dv 2 (105)

A(X, X) = Auu a2 + 2Auv ab + Avv b2 if X = aru + brv (106)

32
In condensed notations:

A = duα Aαβ duβ = duα (n, rαβ )duα (107)

A(X, Y) = X α Aαβ Y β , where X = X α rα , Y = Y α rα (108)

µ ¶ µ ¶
A11 A12 (n, ruu ) (n, ruv )
Aαβ = (n, rαβ ) = =
A21 A22 (n, rvu ) (n, rvv )
We have to prove that expression (104) indeed defines quadratic and
corresponding bilinear form (108), i.e. (104) is invariant under changing of
parameterisation.
Remark The proof of reparameterisation invariance for first quadratic form
was double checking. First quadratic form was invariant apriori. In the case of
second quadratic form the invariance does not follow apriori from the definition
(104)
Let A = duα Aαβ duβ be an appearance of second quadratic form in coordi-
nates (u1 , u2 ). Let dξ p Apq 0 dξ q be an appearance of second quadratic form in new
2r
coordinates (ξ 1 , ξ 2 ). Here Aαβ = (n, rαβ ) = (n, ∂u∂α ∂u 0
β ) and respectively Apq =
2
(rp , rq ) = ( ∂ξ∂p ∂ξ
r ∂r α β p 0
q , ∂ξ q ). We have to check that A = du Aαβ du = dξ Apq dξ .
q

Let ξ 1 , ξ 2 be new parameters: uα = uα (ξ p ).

2
Using chain rule calculate Apq = (n, ∂ξ∂ p rξq ):
µ ¶
∂2r ∂ ∂uβ ∂r ∂ 2 uβ ∂r ∂uα ∂uβ ∂ 2 r
p q
= p = +
∂ξ ∂ξ ∂ξ ∂ξ q ∂uβ ∂ξ p ∂ξ q ∂uβ ∂ξ p ∂ξ q ∂uα ∂uβ

Now: simple but important observation: in the last formula the first term in the
2 a
RHS which possesses second derivatives of reparameterisation, ∂ξ∂ pu∂ξq is propor-
∂r
tional to tangent vector ∂u α . Hence its scalar product with normal vector n is

equal to zero:
µ ¶ µ ¶ µ ¶ β
∂2r ∂ 2 uβ ∂r ∂uα ∂2r ∂u
n, p q = p q n, β + p n, α β
∂ξ ∂ξ ∂ξ ∂ξ ∂u ∂ξ ∂u ∂u ∂ξ q
| {z }
vanishes
³ ´ ³ ´
Apq = (n, rpq ) = n, uαpq ra + uαp uβq rαβ = n, uαp uβq rαβ = uαp Aαβ uβq (109)
α 2 α
(We use notations: uap = ∂u α ∂ u
∂ξ p , upq = ∂ξ p ∂ξ q )
The formula above establishes the transformation of components of second
quadratic form under changing of parameters.

33
 Using (89) we see that

duα Aαβ duβ = dξ p uαp Aαβ uβq dξ q = dξ p Apq

0 β
uq dξ q (110)

We came above to the notion of the first quadratic form calculating length
of vectors tangent to the surface and length of the curve on the surface.
What about to calculate acceleration and curvature. For curves accel-
eration defines curvature, at least in normal parameterisation. In arbitrary
parameterisation curvature is defined by velocity and acceleration vectors
(see (41))).
Our task is to define a curvature on the points of surface.
Note that different curves starting at this point have different curvatures.
Sure curvature depends on direction of the curve: (consider e.g. cylinder.)
But it is not the end of the story. Consider curves starting at the given
point which are tangent to the same vector X:

dr(t) ¯¯
r(t) = r(u(t), v(t)) = r(u0 , v0 ) + tX + . . . v(u0 , v0 ) = (111)
dt t=0
Curvature depends on second derivatives. Even fixing tangent vector we do
not fix curvature of the curve.

2.6 Second quadratic form and curvature of normal

sections
Let p be an arbitrary point of the surface M . Let n be normal vector to the
surface M at the point p, i.e. n is orthogonal to the surface at the point p
and the length of n is equal to 1:
ru × rv
n(u, v) = (112)
|ru × rv |

Definition
A plane which passes through the point p and possesses the vector n is called
normal plane at the point p.
An intersection of normal plane with surface gives a curve. This curve is
called normal section at the point p
Let n be a normal vector at the point p and X be a tangent vector at this
point. Consider plane spanned by vectors n, X. This will be normal plane

34
which possesses vector X. Intersection of the normal plane with surface will
be normal section. This normal section we denote by lX (p). X is tangent
vector of normal section lX (p) at the point p.
We will calculate now the curvature of normal section via second and first
quadratic forms.
Proposition Second quadratic form measures curvature of normal sec-
tions. The curvature k(lX ) of normal section lX (p) at the point p is given up
to the sign by the following formula:
A(X, X)
k(lX ) = , (113)
G(X, X)
where A is the the second quadratic form at the point p and G is the first
quadratic form at the point p.
In particularly if one chooses parameterisation such that |X| = 1 then
k(lX ) = A(X, X), (114)

Remark Here we again can consider signed curvature as in subsection

1.7. As it was mentioned before unit vector n is decided up to a sign. Hence
second quadratic form A(X, X) is defined up to the sign too.
Fixing the direction of n (one can do it using orientation or in other way)
fixes the sign of curvature for a normal section.
Proof.
Before going to calculations note that right hand side of the formula
in proposition does not depend on the length of the vector X, i.e. it is
invariant with respect to different parameterisation of normal section: we
change X → aX then numerator and denominator are multiplied on a2 .
Let r = r(u(t), v(t) be parametrisation of normal section lX at the point
p.
Let v be velocity vector and a be acceleration vector of the normal section
lX at the point initial point p. Velocity vector is proportional to tangent
vector X. If r = r(u(t), v(t)) is parametric equation of normal curve then
v = X = ru ut + rv vt at the initial point p. Without loss of generality suppose
that v = X. (If v 6= X, we change parameter t → at, then v → av)
Calculate acceleration at the point p. Normal section belongs to the
normal plane formed by normal vectors n and velocity vector v = X. Hence
acceleration vector is linear combination of these vectors:
a = aperpendic + aparallel = Ln + bX (115)

35
 Calculate curvature at the point p. Curvature k is equal to the modulus
of the vector v × a divided by the cube of the modulus of the velocity vector
k = |v×a|
|v|3
(see (45)). On the other hand according the formula above

v × a = v × Ln + bX = LX × n, and |v × a| = L|X| .

Hence
|v × a| |aperpendic | L
k= = =
|v|3 |X|3 |X|2
It remains to calculate coefficient L in the expansion (115) of acceleration
vector. Show that it is equal to the value of second quadratic form on the
velocity vector X. Take a scalar product of a on the unit vector n: (X, n) = 0,
hence (a, n) = (Ln + bX, n) = L. We come to
µ 2 ¶ µ ¶
dr d
L = (a, n) = ,n = (ru ut + rv vt ) , n =
dt2 dt
 
 ¡ ¢ 
 (r u + r v ) + ruu (ut )2 + 2ruv ut vt + rvv (vt )2 , n =
| u tt {z v tt}
vector tangent to the surface
¡¡ ¢ ¢
ruu (ut )2 + 2ruv ut vt + rvv (vt )2 , n = (ruu , n) u2t +2 (ruv , n) ut vt +(rvv , n) vt2 =
Auu u2t + Auv ut vt + Avv vt2 = A(X, X) .
Hence we see that curvature of normal section is equal to
L A(X, X)
k= 2
=
|X| G(X, X)
because |X|2 = G(X, X), This is just (113).

2.7 Shape operator and Gaussian and Mean Curva-

tures
Consider first and second quadratic forms G(X, X), A(X, X) for arbitrary
point r(u, v) of the surface and arbitrary tangent vector X. We proved (see
Proposition (113), (114)) that
A(X, X)
= k(lX )(curvature of the normal curve lX at the initial point)
G(X, X)

36
It depends on the direction of vector X and does not depend on its value.
Considering a circle |X| = 1 we see that in a general case there are two
directions such that in one direction curvature kmax is maximal and in the
other direction curvature kmin is minimal:

kmin ≤ k(lX ) ≤ kmax (116)

Definition Product of maximal and minimal curvatures is called Gaus-

sian curvature. Their sum is called Mean curvature:
1 1 1 1
K = kmax kmin = · , H = kmax + kmin = + (117)
r1 r2 r1 r2
Note that here we consider signed curvature (see remark after the for-
mula (114). If we change the direction of normal vector n then all curvatures
kmin , klX , kmax change the sign K Gaussian curvature kmin kmax remains un-
changed. Mean curvature change the sign.
Consider shape operator S(X) defined by the following way:

S(X) : A(X, X) = G(X, S(X)) (118)

For corresponding bilinear forms

S(X) : A(Y, X) = G(Y, S(X)), for an arbitrary tangent vector Y (119)

S = G−1 · A
Theorem Eigenvectors of shape operator define directions in which cur-
vature is maximal and minimal. Eigenvalues of shape operator are maximal
and minimal curvatures.: If vector X1 defines the direction in which curva-
ture of normal section is maximal: k(lX1 ) = kmax and vector X2 defines the
direction in which curvature of normal section is minimal: k(lX2 ) = kmin then

klX1 = kmax , klX2 = kmin , SX1 = kmax X1 , SX2 = kmax X2 (120)

Vectors X1 , X2 are orthogonal (if k1 6= k2 )

Gaussian curvature is a product of maximal and minimal curvatures. It
is equal to the determinant of Shape operator, i.e.to the ratio of the determi-
nants of second and first quadratic forms:
det A
K = kmax kmin = det S = det(G−1 · A) = (121)
det G

37
 Mean curvature is a sum of maximal and minimal curvatures. It is equal
to the Trace of the Shape operator:

K = kmax + kmin = TrS = Tr(G−1 · A) (122)

Give a short
µ proof¶of this Theorem.
amax
Let X1 = be a vector such that curvature of normal section lX1 is
bmax
µ ¶
amin
maximal. Respectively let X2 == be a vector such that curvature of
bmin
normal section lX2 is minimum. We consider components of the vector in the basis
ru , rv (e.g. Xmax = amax ru µ
+ bmax rv ) ¶ µ ¶
Auu Auv Guu Guv
Consider matrices A = and G = of second and
Avu Avv Gvu Gvv
first quadratic forms.
Then

µ ¶µ ¶ µ ¶µ ¶
Auu Auv amax Guu Guv amax
AX1 = kmax GX1 , i.e. = kmax
Avu Avv bmax Gvu Gvv bmax
(123)
and
µ ¶µ ¶ µ ¶µ ¶
Auu Auv amin Guu Guv amin
AX2 = kmin GX1 , i.e. = kmin
Avu Avv bmin Gvu Gvv bmin
(124)
Indeed kmax (kmin ) is maximum (minimum) value of the function

A(a, b) = A(X, X) = Auu a2 + 2Auv ab + Avv b2

subject to the condition that

G(a, b) = Guu a2 + 2Guv ab + Gvv b2 ≡ 1

∂A
Standard Lagrange multipliers consderation gives: ∂a = λ ∂G
∂a ,
∂A
∂b = λ ∂G
∂b , i.e.

Auu a + Auv b = λ (Guu a + Guv b) , Auv a + Avv b = λ (Guv a + Gvv b)

It is just formulae (123), (124) above.

It follows from these relations that that X1 , X2 are eigenvectors of shape op-
erator G−1 A:
µ ¶−1 µ ¶µ ¶ µ ¶
Guu Guv Auu Auv amax amax
· = kmax
Gvu Gvv Avu Avv bmax bmax

38
and respectively
µ ¶−1 µ ¶µ ¶ µ ¶
Guu Guv Auu Auv amin amin
· = kmin
Gvu Gvv Avu Avv bmin bmin

Orthogonality of eigenvectors X1 , X2 (in the case if kmax 6= kmin ) follows immedi-

ately from the relation:

kmax (X1 , X2 ) = G(SX1 , X2 ) = A(X1 , X2 ) = A(X2 , X1 ) = G(SX2 , X1 ) = kmin (X2 , X1 )

(It is general property of eigenvectors of symmetric operators.)

Note It is instructive to consider the function det(1 + zS), where z is formal

parameter. This function is quadratic polynomial in z and coefficients are just
mean and Gaussian curvature:

det(1 + zS) = 1 + Hz + Kz 2 (125)

If M is a compact surface in E3 then one can consider remarkable polynomial:

Z
√
PM (z) = det(1 + zS) gd2 x (126)
M

Remarkable formulae are related with this object (see Appendix ”Tubes”)

2.8 Calculations of Gaussian curvature and Mean cur-

vatures
First of all calculate curvatures for the general case when a surface C is
defined by the equation z − F (x, y) = 0. One can consider the following
parameterisation of this surface:


x = u
r(u, v) : y=v (127)


z = F (u, v)

1. Calculation of first quadratic form

   
1 0
ru =  0  rv =  1  (128)
Fu Fv

39
,
(ru , ru ) = 1 + Fu2 , (ru , rv ) = Fu Fv , (rv , rv ) = 1 + Fv2
and first quadratic form (95) is equal to
µ ¶ µ ¶
G11 G12 (ru , ru ) (ru , rv )
G= = (129)
G12 G22 (ru , rv ) (rv , rv )

µ ¶
1 + Fu2 Fu Fv
G= , ds2 = (1 + Fu2 )du2 + 2Fu Fv dudv + (1 + Fv2 )dv 2
Fu Fv 1 + Fv2
(130)
and the length of the curve r(t) = r(u(t), v(t)) on C (a ≤ t ≤ b) can be
calculated by the formula:
Z Z bq
L= (1 + Fu2 )u2t + 2Fu Fv ut vt + (1 + Fv )2 vt2 dt (131)
a

1. Calculation of second quadratic form

First of all calculate normal unit vector n(u, v)
One can do it in two different ways:
using cross-product:
ru × rv
n(u, v) = , (132)
|ru × rv |
 
ex ey ez
ru × rv = det  1 0 Fu  = −Fu ex − Fv ey + ez
0 1 Fv
Hence according to (132)
 
−Fu
1 −Fv 
n= p (133)
2 2
1 + Fu + Fv 1

(n is defined up to the sign)

Sometimes it is more easy to calculate n using that it is proportional to
the gradient of equation defining surface:

n is proportional to grad (z − F (x, y)) = (−Fx , −Fy , 1) (134)

40
 Now for calculating second quadratic form it remains to calculate
     
0 0 0
ruu =  0  , ruv =  0  , rvv =  0 
Fuu Fuv Fvv

and
Fuu Fuv Fvv
(ruu , n) = p , (ruv , n) = p , (rvv , n) = p ,
1 + Fu2 + Fv2 + 1 + Fu2 + Fv2 1 + Fu2 + Fv2

Hence using expression (133) for unit normal vector n and (??) we come to:
µ ¶ µ ¶ µ ¶
A11 A12 (ruu , n) (ruv , n) 1 Fuu Fuv
= =p (135)
A12 A22 (ruv , n) (rvv , n) 1 + Fu2 + Fv2 Fuv Fvv

Now calculate shape operator:

µ ¶
−1 1 1 + Fv2 −Fu Fv
G =
1 + Fu2 + Fv2 −Fu Fv 1 + Fu2

Hence shape operator:

µ ¶ µ ¶
−1 1 1 + Fv2 −Fu Fv Fuu Fuv
S=G A= ·
(1 + Fu + Fv2 )3/2
2 −Fu Fv 1 + Fu2 Fuv Fvv

Calculation of Gaussian and Mean curvatures:

Gaussian curvature according previous considerations (121) is equal to
the determinant of the shape operator:
det A
K = det S = =
det G
2
Fuu Fvv −Fuv
2
(1+Fu2 +Fv2 ) Fuu Fvv − Fuv
= (136)
(1 + Fu2 )(1 + Fv2 )2 − Fu Fv (1 + Fu2 + Fv2 )2
Mean curvature according to (122) is equal to

Fuu + Fvv + Fv2 Fuu − 2Fu Fv Fuv + Fu2 Fvv

H = TrH = (137)
(1 + Fu2 + Fv2 )3/2

41
 Calculate now Gaussian and mean curvature for cylinder, cone, sphere
and saddle. Of course we can use general formulae obtained above. But
why not to calculate independently? (It seems to be more interesting and
sometimes much easy)
Cylinder
Cylinder is given by the equation x2 + y 2 = a2 . One can consider the
following parameterisation of this surface:


x = a cos ϕ
r(h, ϕ) : y = a sin ϕ (138)


z=h

Calculation of first quadratic form for cylinder

   
0 −a sin ϕ
rh = 0 rϕ =  a cos ϕ  (139)
1 0
,
(rh , rh ) = 1, (rh , rϕ ) = 0, (rϕ , rϕ ) = a2
and first quadratic form (95) is equal to
µ ¶
(ru , ru ) (ru , rv )
G= = (140)
(ru , rv ) (rv , rv )
µ ¶
1 0
, ds2 = dh2 + a2 dϕ2 (141)
0 a2
and the length of the curve r(t) = r(h(t), ϕ(t)) on the cylinder (a ≤ t ≤ b)
can be calculated by the formula:
Z bp
L= h2t + a2 ϕt dt (142)
a

Calculation of second quadratic form for cylinder

It is evident without any calculations that normal unit vector n(u, v) to
the cylinder (138) is defined by the following formula
 
cos ϕ
n =  sin ϕ  (143)
0

42
(unit vector n is defined up to the sign)
But you can calculate using general formulae above:
Using cross-product:
 
ex ey ez
rh × rϕ = det  0 0 1  = − cos ϕex − sin ϕey
−a sin ϕ a cos ϕ 0

and we come (up to a sign) to the answer (143). (Normal unit vector is
defined up to direction n → −n)
One can calculate n using that it is proportional to the gradient of equa-
tion defining surface:
¡ ¢
n is proportional to grad x2 + y 2 − a2 ) = (2x, 2y, 0) (144)

and come to the same answer (143).

Now calculate second quadratic form:
     
0 0 −a cos ϕ
rhh = 0 , rhϕ = 0 , rϕϕ =  −a sin ϕ 
0 0 0

and
µ ¶ µ ¶
(rhh , n) (rhϕ , n) 0 0
A= = (145)
(rhϕ , n) (rϕϕ , n) 0 −a
Calculation of Shape operator for cylinder:
µ ¶
−1 1 0
G =
0 1/a2

Hence shape operator:

µ ¶ µ ¶ µ ¶
−1 1 0 0 0 0 0
S=G A= · =
0 1/a2 0 −a 0 −1/a

Gaussian and Mean curvatures for cylinder:

det A 1
K = det S = = 0, H = −
det G a

43
 Cone
Cone is given by the equation x2 + y 2 − kz 2 = 0. One can consider the
following parameterisation of this surface:


x = kh cos ϕ
r(h, ϕ) : y = kh sin ϕ (146)


z=h

Calculation of first quadratic form for cone

   
k cos ϕ −kh sin ϕ
rh =  k sin ϕ  rϕ =  kh cos ϕ  (147)
1 0
,
(rh , rh ) = 1 + k 2 , (rh , rϕ ) = 0, (rϕ , rϕ ) = k 2 h2
and first quadratic form (95) is equal to
µ ¶
(ru , ru ) (ru , rv )
G= = (148)
(ru , rv ) (rv , rv )
µ ¶
1 + k2 0
, ds2 = (1 + k 2 )dh2 + k 2 h2 dϕ2 (149)
0 k 2 h2
and the length of the curve r(t) = r(h(t), ϕ(t)) on the cone (a ≤ t ≤ b) can
be calculated by the formula:
Z bq
L= (1 + k 2 )h2t + k 2 h2 ϕ2t dt (150)
a

Calculation of second quadratic form for cone

In this case easiest way to calculate a unit vector n is to note that it is
proportional to the gradient of equation defining surface:
¡ ¢
n is proportional to grad x2 + y 2 − k 2 z 2 ) = (2x, 2y, −2zk 2 ) (151)

Hence  
kh cos ϕ
n = λ  kh sin ϕ 
−k 2 h

44
Find λ such that |n| = 1:
 
cos ϕ
1  sin ϕ 
n= √
1 + k2 −k

Now calculate second quadratic form:

     
0 −k sin ϕ −kh cos ϕ
rhh = 0 , rhϕ =  k cos ϕ  rϕϕ =  −kh sin ϕ 
0 0 0

and
µ ¶ µ ¶
(rhh , n) (rhϕ , n) 0 0
A= = kh (152)
(rhϕ , n) (rϕϕ , n) 0 − √1+k 2

Shape operator for cone:

µ 1
¶
−1 1+k2
0
G = 1
0 k2 h2

Hence shape operator:

µ 1 ¶ µ ¶ µ ¶
−1 0 0 0 0 0
S = G A = 1+k2 · = =
0 1
k2 h2
kh
0 − √1+k 2 0 − kh√11+k2

Gaussian and Mean curvatures for cone:

det A 1
K = det S = = 0, H= √
det G kh 1 + k 2

Sphere
Sphere is given by the equation x2 + y 2 + z 2 = a2 . Consider the following
(standard ) parameterisation of this surface:


x = a sin θ cos ϕ
r(θ, ϕ) : y = a sin θ sin ϕ (153)


z = a cos θ

45
 Calculation of first quadratic form for sphere
   
a cos θ cos ϕ −a sin θ sin ϕ
rθ =  a cos θ sin ϕ  rϕ =  a sin θ cos ϕ  (154)
−a sin θ 0
,
(rθ , rθ ) = a2 , (rh , rϕ ) = 0, (rϕ , rϕ ) = a2 sin2 θ
and first quadratic form (95) is equal to
µ ¶
(ru , ru ) (ru , rv )
G= = (155)
(ru , rv ) (rv , rv )
µ 2 ¶
a 0
, ds2 = a2 dθ2 + a2 sin2 θdϕ2 (156)
0 a2 sin2 θ
and the length of the curve r(t) = r(θ(t), ϕ(t)) on the sphere of the radius a
(a ≤ t ≤ b) can be calculated by the formula:
Z b q
L= a θt2 + sin2 θ · ϕ2t dt (157)
a

Calculation of second quadratic form for sphere

It is obvious that unit vector n for the sphere is just
 
sin θ cos ϕ
r
n = =  sin θ sin ϕ 
r
cos θ
Second quadratic form:
     
−a sin θ cos ϕ −a cos θ sin ϕ −a sin θ cos ϕ
rθθ =  −a sin θ sin ϕ  , rθϕ =  a cos θ cos ϕ  rϕϕ =  −a sin θ sin ϕ 
−a cos θ 0 0
and µ ¶ µ ¶
(rθθ , n) (rθϕ , n) −a 0
A= = (158)
(rθϕ , n) (rϕϕ , n) 0 −a sin2 θ
Calculation of Shape operator for sphere:
µ1 ¶
−1 2 0
G = a
1
0 a2 sin 2θ

46
Hence for Shape operator:
µ1 ¶ µ ¶ µ 1 ¶
−1 2 0 −a 0 −a 0
S=G A= a · = =
1
0 a2 sin2θ 0 −a sin2 θ 0 − a1

Gaussian and Mean curvatures for Sphere:

det A 1 2
K = det S = = 2, H=
det G a a

Saddle
Saddle is given by the equation z − xy = 0. (This surface contains
horizontal and vertical lines...)
Consider the following (standard ) parameterisation of this surface:


x = u
r(u, v) : y=v (159)


z = uv

Calculation of first quadratic form for saddle

   
1 0

ru = 0  rv = 1 
 (160)
v u
,
(rθ , rθ ) = 1, (rh , rϕ ) = 0, (rϕ , rϕ ) = a2 sin2 θ
and first quadratic form (95) is equal to
µ ¶
(ru , ru ) (ru , rv )
G= = (161)
(ru , rv ) (rv , rv )
µ ¶
1 + v2 uv
, ds2 = (1 + v 2 )du2 + 2uvdudv + (1 + u2 )dv 2 (162)
uv 1 + u2
and the length of the curve r(t) = r(u(t), v(t)) on the sphere of the radius a
(a ≤ t ≤ b) can be calculated by the formula:
Z b q
L= a (1 + v 2 )u2t + 2uvut vt + (1 + u2 )vt2 dt (163)
a

47
 Calculation of second quadratic form for saddle
Calculate n. grad(z − xy) = (−y, −x, 1). Hence
 
−v
1 
n= √ −u
1 + u2 + v 2 1

Second quadratic form:

     
0 0 0
ruu = 0 , ruv = 0 , rvv = 0
0 1 0

and µ ¶ µ ¶
(ruu , n) (ruv , n) 1 0 1
A= =√ (164)
(ruv ) (rvv , n) 1 + u2 + v 2 1 0
Calculation of Shape operator for saddle:
 
µ 2
¶ 0 1
1 1+u −uv
S = G−1 A = 2 ·  1 0
(1 + u2 + v 2 ) 2 −uv 1 + v
3
=
µ ¶
1 −uv 1 + u2
2
(1 + u2 + v 2 ) 2 1 + v −uv
3

Gaussian and Mean curvatures for Saddle:

det A 1 2uv
K = det S = =− , H=− 3
det G (1 + u + v 2 )2
2
(1 + u2 + v 2 ) 2

3 Riemannian manifolds
3.1 Definitions
The Riemannian metric on the manifold M defines the length of the tangent
vectors and the length of the curves.
Riemannian metric
Definition

48
 Riemannian metric on n-dimensional manifold M n defines for every point
P the scalar product of tangent vectors in the tangent space Tp M smoothly
depending on the point P .
It means that in every coordinate system (x1 , . . . , xn ) a metric G is defined
by a matrix gik (x) such that

• gik (x) = gki (x) (Metric is defined by symmetric tensor of second rank)

• gik (x)ui uk ≥ 0, gik (x)ui uk = 0 iff u1 = · · · = un = 0 (it is positively

defined)

• gik (x) are smooth functions

For any two vectors

   1
A1 B
 ·   · 
   
A=
 ·  ,B =  · 
  (165)
 ·   · 
n
A Bn
the scalar product is equal to:
 
  B1
g11 . . . g1n  
¡ ¢ ... ... . . .  · 
G(A, B) = Ai Gik B k = A1 . . . An   ·  (166)
gn1 . . . gnn   
 · 
Bn

For any two coordinate systems (x1 , . . . , xn ), (y 1 , . . . , y n ), xi = xi (y p )

matrices gik (x), gi0 k0 (x) are related by the relation:

∂xi ∂xk
g̃pq (y) = gik (x(y)) (167)
∂y p ∂y q
It is convenient to write metric:

G = gik (x)dxi dxk (168)

If (y 1 , . . . , y n ) are new coordinates then

G = gik (x)dxi dxk = G = dxi gik (x(y)) dxk =

49
 µ ¶ µ ¶
∂xi ∂xk
dy p p gik (x(y)) dy k = dy p g̃pq (y)dy q (169)
∂y ∂y q
We come to the formula (167). (We use here condensed notations)
Length of the curve. Let γ : (x1 (t), . . . , xn (t)) (a ≤ t ≤ b) be a curve
on the Riemannian manifold (M, G). At the every point of the curve the
velocity vector (tangent vector) is defined:
 1 
ẋ (t)
 · 
 
v(t) =  · 
 (170)
 · 
ẋn (t)

Then the length of the curve is defined by the integral of the length of
velocity vector: Z bp Z bp
Lγ = hv, vi = gik (x)ẋi ẋk (171)
a a

Bearing in mind that metric (168) defines the length we often write metric
in the following form
ds2 = gik dxi dxk (172)

For example consider 2-dimensional Riemannian manifold with metric

gik (u, v) (i, k = 1, 2). Then

ds2 = gik dxi dxk = g11 (u, v)du2 + 2g12 (u, v)dudv + g22 (u, v)dv 2

The length of the curve γ : u = u(t), v = v(t), where t0 ≤ t ≤ t1 according to

(171) is equal to
Z t1 p Z t1 p
Lγ = hv, vi = gik (x)ẋi ẋk = (173)
t0 t0
Z t1 q
g11 (u (t) , v (t)) u2t + 2g12 (u (t) , v (t)) ut vt + g22 (u (t) , v (t)) vt2 dt
t0
(174)
If metric has diagonal form:

ds2 = gik dxi dxk = a(u, v)du2 + b(u, v)dv 2 , (a = g11 , b = g22 )

50
then
Z t1 p Z t1 q
Lγ = hv, vi = a (u (t) , v (t)) u2t + b (u (t) , v (t)) vt2 dt (175)
t0 t0

First quadratic form is an example of Riemannian metric on the two-

dimensional surface. The first quadratic form is defined by the position of
surface in three-dimensional ambient space E3 . In the case of Riemannian
metric we input it apriori.
Consider two-dimensional manifold with metric (1 + k 2 )du2 + k 2 u2 dv 2 .
The length of the curve u = u(t), v = v(t) on this surface is equal to
Z t1 q
L= (1 + k 2 )u2t + k 2 vt2 dt
t0

One can compare this metric with metric defined by the first quadratic form
on the cone x2 + y 2 − k 2 z 2 = 0

Example Consider upper half plane with Riemannian metric

dx2 + dy 2
G= (176)
y2
Calculate the length of the vertical line x = a, 0 < t0 ≤ y ≤ t1 .

3.2 Volume element in Riemannian manifold

The volume element in n-dimensional Riemannian manifold with metric G =
gik dxi dxk is defined by the formula
p
det gik dx1 dx2 . . . dxn (177)
If D is a domain in the n-dimensional Riemannian manifold with metric
G = gik dxi then its volume is equal to to the integral of volume element over
this domain. Z p
V (D) = det gik dx1 dx2 . . . dxn (178)
D
Remark Students who know the concept of exterior forms can read the
volume element as p
det gik dx1 ∧ dx2 ∧ · · · ∧ dxn (179)

51
 Note that in the case of n = 1 volume is just the length, in the case if
n = 2 it is area.
Note that the formula (177) gives volume of n-dimensional parallelepiped.
In cartesian coordinates we come to standard formula for domain.
Invariance of volume element under changing of coordinates
Prove that volume element is invariant under coordinate transformations, i.e.
if y1 , . . . , yn
are new coordinates: x1 = x1 (y 1 , . . . , y n ), x2 = x2 (y 1 , . . . , y n )...,

xi = xi (y p ), i = 1, . . . , n , p = 1, . . . , n

and g̃pq (y) matrix of the metric in new coordinates:

∂xi ∂xk
g̃pq (y) = gik (x(y)) (180)
∂y p ∂y q
(See formulae (167) and(169)) then
p q
det gik (x) dx1 dx2 . . . dxn = det g̃pq (y) dy 1 dy 2 . . . dy n (181)

This follows from (180). Namely

s µ ¶
p 1 2 n ∂xi ∂xk
det gik (y) dy dy . . . dy = det gik (x(y)) dy 1 dy 2 . . . dy n
∂y p ∂y q
³ i´ ³ k´
∂x ∂x 6
Using the fact that det(ABC) = det A · det B · det C and det ∂y p = det ∂y q
we see that from the formula above follows:
s µ i ¶
p 1 2 n ∂x ∂xk
det gik (y) dy dy . . . dy = det gik (x(y)) q dy 1 dy 2 . . . dy n =
∂y p ∂y
sµ µ i ¶¶2
∂x p
det det gik (x(y))dy 1 dy 2 . . . dy n =
∂y p
µ i¶
p ∂x
det gik (x(y)) det dy 1 dy 2 . . . dy n = (182)
∂y p
Now note that
µ ¶
∂xi
det dy 1 dy 2 . . . dy n = dx1 . . . dxn
∂y p
6
determinant of matrix does not change if we change the matrix on the adjoint, i.e.
change columns on rows.

52
according to the formula for changing coordinates in n-dimensional integral 7 .
Hence
µ i¶
p ∂x 1 2 n
p
det gik (x(y)) det dy dy . . . dy = det gik (x(y))dx1 dx2 . . . dxn (183)
∂y p

Thus we come to (181).

Example
Consider first very simple example: Volume element of plane in cartesian
coordinates, metric g = dx2 + dy 2 . Volume element is equal to
s µ ¶
p 1 0
det gdxdy = det dxdy = dxdy
0 1

Volume of the domain D is equal to

Z p Z
V (D) = det gdxdy = dxdy
D D

If we go to polar coordinates:

x = r cos ϕ, y = r sin ϕ (184)

Then we have for metric:

G = dr2 + r2 dϕ2
because

dx2 + dy 2 = (dr cos ϕ − r sin ϕdϕ)2 + (dr sin ϕ + r cos ϕdϕ)2 = dr2 + r2 dϕ2
(185)
Volume element in polar coordinates is equal to
s µ ¶
p 1 0
det gdrdϕ = det drdϕ = drdϕ
0 r2

Example. Volume element of the metric of Lobachesvky plane.

³ i´
7 ∂x
Determinant of the matrix ∂y p of changing of coordinates is called sometimes Ja-
cobian. Here we consider the case if Jacobian is positive. If Jacobian is negative then
formulae above remain valid just the symbol of modulus appears.

53
 2 2
In coordinates x, y (y > 0) metric G = dx y+dy
2 , the corresponding matrix
µ 2 ¶
1/y 0 √
G= 2 . Volume element is equal to det gdxdy = dxdy y2
0 1/y
Example Consider the two dimensional plane with Riemannian metrics

du2 + dv 2
G= (186)
(1 + u2 + v 2 )2

(It is indeed the sphere in stereographic coordinates)

Calculate its volume element and volume. It is easy to see that:
à !
1
(1+u2 +v 2 )2
0 1
G= 1 det g = (187)
0 (1+u2 +v 2 )2 (1 + u2 + v 2 )4
√
and volume element is equal to det gdudv = (1+ududv
2 +v 2 )2

The volume (area) of plane will be:

Z Z ∞ Z ∞ Z
dudv du π ∞ du
2 2 2
= du 2 2 2
= =π
(1 + u + v ) −∞ −∞ (1 + u + v ) 2 −∞ (1 + u)3/2

We see that in coordinates (u, v) calculation of the integral is not very easy.
One can consider volume form in polar coordinates u = r cos ϕ, v =
r sin ϕ. Then it is easy to see that according to (185) we have for the metric
du2 +dv 2 dr2 +r2 dϕ2 √ rdrdϕ
G = (1+u 2 +v 2 )2 = (1+r2 )2
and volume form is equal to det gdrdϕ = (1+r 2 )2

Now calculation of integral becomes easy:

Z Z ∞ Z ∞
rdrdϕ rdr du
V = 2 2
= 2π 2 2
=π =π
(1 + r ) 0 (1 + r ) 0 (1 + u)2
Example Volume element of the segment of the sphere.
Consider sphere of the radius a in Euclidean space with standard Riema-
nian metric
a2 dθ2 + a2 sin2 θdϕ2
This metric is nothing but first quadratic form on the sphere (see (156)).
The volume element is
s µ 2 ¶
p a 0
det gdθdϕ = det dθdϕ = a2 sin θdθdϕ
0 a2 sin θ

54
Now calculate the volume of the segment of the sphere between two parallel
planes, i.e. domain restricted by parallels θ1 ≤ θ ≤ θ0 : Denote by h be the
height of this segment. One can see that

h = a cos θ0 − a cos θ1 = a(cos θ0 − a cos θ1 )

There is remarkable formula which express the area of segment via the height
h: Z Z θ1 µZ 2π ¶
¡ 2 ¢ ¡ 2 ¢
V = a sin θ dθdϕ = a sin θ dϕ dθ =
θ1 ≤θ≤θ0 θ0 0
Z θ0
2πa2 sin θdθ = 2πa2 (cos θ0 − cos θ1 ) = 2πa(a cos θ0 − acosθ1 ) = 2πah
θ1
(188)
2
E.g. for all the sphere h = 2a. We come to S = 4πa . It is remarkable
formula: area of the segment is a polynomial function of radius of the sphere
and height (Compare with formula for length of the arc of the circle)

3.3 Geodesics
Let A, B are two points on Riemannian manifold (M n , G). Consider the
length of the shortest curve which connects these points More formally con-
sider the set CAB of the curves which start at the point A and end at the
point B. Then the length of the shortest curve (if it exists8 ) is equal to

d(A, B) = inf L (γ) (189)

γ∈CAB

Let {ui (t)} be local coordinates which are defined in the vicinity of the
points A and B. If metric is equal to G = gik dui duk in these coordinates
then length of arbitrary curve γ : ui (t) which starts at A and ends at B is
equal to
Z t1 r
dui (t) dv i (t)
Lγ = gik (u) , ui (t0 ) = ui0 , ui (t1 ) = ui1 (190)
t0 dt dt

where ui0 are coordinates of the initial point A and ui1 are coordinates of the
final point B and the shortest distance is just the inferior of this functional
by all the curves beginning at A and ending at B.
8
we do not consider existence problem and suppose that the shortest curve exist

55
 Example 1
Consider two points in E2 with cartesian coordinates (x, y) (metric G =
dx + dy 2 ): A = (x0 , y0 ), B = (x1 , y1 ).
2

Consider an arbitrary curve γAB x(t), y(t) (such that x(t0 ) = x0 , y(t0 ) =
t0 , x(t1 ) = x1 , y(t1 ) = t1 ) and consider the line
lAB x(t) = x0 + t(x1 − x0 ), y(t) = y0 + t(y1 − y0 ) (191)
It is easy to see that
Z t1 p p
LγAB = x2t + yt2 ≥ LlAB = (x1 − x0 )2 + (y1 − y0 )2
t0

(Technically it is very easy to see in coordinates such that y0 = y1 )

We come to the well-known fact: segment of the line is the shortest
distance between two points in Euclidean space.
Generalising the concept of the line as shortest distance between two
points for Rieamnnian manifodls we come to geodesic
One of definitions of geodesics is following:
Definition The curve γ : xi (t) is called geodesic if of for arbitrary two
(enough closed) points A = xi (t0 ), B = xi (t1 ) of this curve the following
condition holds: The length of the arc of the curve γ between the points
A, B is the shortest, i.e. the length of the arbitrary curve which connects
these two points is bigger or equal to the length of this arc.
How to find geodesics?
In general case one have to consider the corresponding
Rp variational problem
(The Euler-Lagrange equations for functional gik ẋ ẋk ). In some cases
i

one can come to the answer by elementary methods. For example one can
easy to show that geodesics on sphere are great circles. Consider standard
Riemannian metrics on the sphere in E3 with the radius a: Coordinates θ, ϕ,
metrics (first quadratic form):
G = a2 (dθ2 + sin2 θdϕ2 ) (192)
Consider two arbitrary points A and B on the sphere. Let (θ0 , ϕ0 ) be coor-
dinates of the point A and (θ1 , ϕ1 ) be coordinates of the point B
Let γ be a curve which connects these points: γ : θ(t), ϕ(t) such that
θ(t0 ) = θ0 , θ(t1 ) = θ1 , ϕ(t0 ) = θ0 , θ(t1 ) = θ1 then:
Z q
LγAB = a θt2 + sin2 θ(t)ϕ2t dt (193)

56
 Without loss of generality suppose that they have the same latitude, i.e.
if (θ0 , ϕ0 ) are coordinates of the point A and (θ1 , ϕ1 ) are coordinates of the
point B then ϕ0 = ϕ1 (if it is not the fact then we can come to this condition
rotating the sphere)
Now one can see that the meridian ϕ = ϕ0 is geodesics: Indeed consider an
arbitrary curve θ(t), ϕ(t) which connects the points A, B: θ(t0 ) = θ0 , θ(t1 ) =
t1 , ϕ(t0 ) = ϕ(t1 = ϕ0 . Compare its length with the length of the meridian
which connects the points A, B:
Z t1 q Z t1 p Z t1
2 2 2 2
a θt + sin θϕt dt ≥ a θt dt = a θt dt = a(θ1 − θ0 ) (194)
t0 t0 t0

the big circles on sphere are geodesics. It corresponds to geometrical intu-

ition: The geodesics on the sphere are the circles of intersection of the sphere
with the plane which crosses the centre.
Remark In the integral (194) we considered the smallest arc of the great
circle between points A, B.

3.4 Geodesics and isometries

Consider triangles on the Earth.
Let A, B, C be three points on the Earth, e,g, A = P aris, B = M anchester,
C = Berlin. Draw the lines connecting these points. They will be arcs of great
circles. This triangle is called spheric triangle. The sum of angles of this triangle
will not be equal to π. One can prove that if S is a area of spherical triangle then:

S
sum of the angles of spheric triangle − π = = KS (195)
R2
where K is Gaussian curvature of the sphere. (In the general case the formula
above holds only for small triangles.)
We did not notice this phenomenon in ordinary life because radius of earth is
equal to 6400 km.
The fact that sum of the angles is not equal to π is very important property
of the sphere. In principal one can guess that Earth is round just drawing big
triangles. (See the tale on Aunts in Appendix.)
What happened if surface is cone? with Riemannian metric

(1 + k)2 du2 + k 2 dv 2

57
The sum of angle of triangle will be again different from ∂ or no.
We know empirically that plane can be bended to the cone. What it means
exactly:
Definition The diffeomorphism of Riemannian manifolds which preserve the
metrics is called isometry.
In other words isometry is transformation such that preserves distance between
the points.
E.g. standard cylinder and standard cone are to the domain in the euclidean
plane.
More exactly consider cylinder with punctured line:


x = a cos ϕ
r(h, ϕ) : : y = a sin ϕ , 0 < ϕ < 2π, −∞ < h < ∞


z=h

Then it is isometric to the domain D in the plane (x, y) where x ∈ (0, 2π),
−inf ty < y < ∞. This isometry has the following form:

x = aϕ, y = h ⇒ dx2 + dy 2 = a2 dϕ2 + dh2 (196)

It is easy to see that a2 dϕ2 + dh2 is just first quadratic form (Riemannian metric)
on the cylinder. The equation (196) corresponds to unfolding of cylinder.
The same with cone: Consider the upper cone with punctured ray:


x = kh cos ϕ
r(h, ϕ) : : y = kh sin ϕ , 0 < ϕ < 2π, 0 < h < ∞


z=h

Its Riemannian metric (first quadratic form) is equal (show it!)

ds2 = (1 + k 2 )dh2 + k 2 h2 dϕ2

Establish the isometry with domain in plane e.g. for the case k = 1:
(
√ √ x = r cos Ψ,
r = 2h, Ψ = ϕ/ 2, where
y = r sin Ψ

One can see that the map above transforms he metric dx2 + dy 2 into metric on
cone. Hence it is isometry. This map corresponds to unfolding of the cone.
Empirically it is evident. One can prove it formally

58
4 Parallel transport; Gauss–Bonnet Theorem
4.1 Concept of parallel transport
Parallel transport of the vectors is one of the fundamental concept of differ-
ential geometry. Here we just give some preliminary ideas and formulate the
concept of parallel transport for surfaces embedded in Euclidean space. The
detailed approach is founded on the conception of connection and covariant
derivative (see the next section).
Let C be a surface r = r(u, v) in E3 and γ(t), t1 ≤ t ≤ t2 a curve on this
surface γ(t) : r = r(t) = r(u(t), v(t))).
Let X1 be a vector tangent to the surface at the initial point p = γ(t1 )
of the curve γ(t) on the surface: X1 ∈ Tp C (p = γ(t1 )). We define parallel
transport of the vector along the curve:
Definition Let γ(t) be a curve on the surface C. Let X(t) be a family
of vectors depending on the parameter t (t1 ≤ t ≤ t2 ) such that following
conditions hold

• For every t ∈ [t1 , t2 ] vector X(t) is a vector tangent to the surface C

attached to the point r(t) = r(u(t), v(t)) of the curve γ(t).

• X(t) = X1 for t = t1
dX(t)
• dt
is orthogonal to the surface, i.e.

dX(t) dX(t)
is parallel to the normal vector n(t), = λ(t)n(t) (197)
dt dt
Recall that normal vector n(t) is a vector attached to the point r(t) of
the curve γ(t) which is orthogonal to the surface C. It can be calculated
by the formula:
N
n= , where N = [ru × rv ]
|N|
The condition (197) means that only orthogonal component of vector
could be changed.

We say that a family X(t) is a parallel transport of the vector X1 along a

curve γ(t) on the surface C. The final vector X2 = X(t2 ) is the image of the
vector X1 under the parallel transport along the curve γ(t).

59
 Using the relation (197) it is easy to see that the scalar product of two
vectors remains invariant under parallel transport. In particularly it means
that length of the vector does not change. If X(t), Y(t) are parallel transports
of vectors X1 , Y1 then
µ ¶ µ ¶
d dX(t) dY(t)
(X(t), Y(t)) = , Y(t) + X(t), =0
dt dt dt

because vector dX(t)

dt
is orthogonal to the vector Y(t) and vector dY(t) dt
is
orthogonal to the vector X(t). In particularly length does not change:

d d dX(t)
|X(t)|2 = (X(t), X(t)) = 2( , X(t)) = 2(λ(t)n(t), X(t)) = 0
dt dt dt
(198)
Remark The relation (197) shows how the surface is engaged in the
parallel transport. Note that it is non-sense to put the right hand side of
the equation (197) equal to zero: In general a tangent vector ceased to be
tangent to the surface if it is not changed! (E.g. consider the vector which
transports along the great circle on the sphere)

4.2 Parallel transport of vectors tangent to the sphere.

1. In the case if surface is a plane then everything is easy. If vector X1 is
tangent to the plane at the given point, it is tangent at all the points. Vector
does not change under parallel transport X(t) ≡ X.
Consider a case of parallel transport along curves on the sphere.
Consider on the sphere tangent vectors:
   
a cos θ cos ϕ −a sin θ sin ϕ
rθ =  a cos θ sin ϕ  rϕ =  a sin θ cos ϕ  (199)
−a sin θ 0
 
a sin θ cos ϕ
attached at the point r(θ, ϕ) =  a sin θ sin ϕ . One can see that
a cos θ

(rθ , rθ ) = a, (rh , rϕ ) = 0, (rϕ , rϕ ) = a2 sin2 θ

60
It is convenient to introduce vectors which are parallel to these vectors but
have unit length:
rθ rϕ
eθ = , eϕ = (eθ , eθ ) = 1, (eθ , eϕ ) = 0, (eϕ , eϕ ) = 1 . (200)
a a sin θ
How these vectors change if we move along parallel (i.e. what is the value
of ∂e
∂ϕ
θ
, ∂e
∂ϕ
ϕ
); how these vectors change if we move along meridians (i.e. what
is the value of ∂e
∂θ
θ
, ∂eϕ
∂θ
). First of all recall that unit normal vector to the
sphere at the point θ, ϕ is equal to r(θ,ϕ)
a
:
 
sin θ cos ϕ
n(θ, ϕ) =  sin θ sin ϕ 
cos θ
Now calculate:
   
cos θ cos ϕ − sin θ cos ϕ
∂eθ ∂ 
= cos θ sin ϕ  =  − sin θ sin ϕ  = −n (201)
∂θ ∂θ
− sin θ − cos θ
,    
cos θ cos ϕ − cos θ sin ϕ
∂eθ ∂ 
= cos θ sin ϕ  =  cos θ cos ϕ  = cos θeϕ , (202)
∂ϕ ∂ϕ
− sin θ 0
,    
− sin ϕ − cos θ sin ϕ
∂eϕ ∂ 
= cos ϕ  =  cos θ cos ϕ  = 0, (203)
∂θ ∂θ
0 0
   
− sin ϕ − cos ϕ
∂eϕ ∂ 
= cos ϕ  =  − sin ϕ  = − sin θn − cos θeθ , (204)
∂ϕ ∂ϕ
0 0
Some of these formulaes are intuitively evident: For example formula
(201) which means that family of the vectors eθ (θ) is just parallel transport
along meridian, because its derivation is equal to −n.
Another intuitively evident example: consider the meridian θ(t) = t,
ϕ(t) = ϕ0 , 0 ≤ t ≤ π. It is easy to see that the vector field
 
cos θ(t) cos ϕ0
X(t) = eθ (θ(t), ϕ0 ) =  cos θ(t) sin ϕ0 
− sin θ(t)

61
attached at the point (θ(t), ϕ0 ) is a parallel transport because for family of
vectors X(t) all the conditions of parallel transport are satisfied. In particular
according to (201)
 
cos θ cos ϕ
dX(t) dθ(t) ∂ 
= cos θ sin ϕ  = −n(θ(t), ϕ0 )
dt dt ∂θ
− sin θ
Now consider an example which is intuitively not-evident.
Example. Calculate parallel transport of the vector eϕ along the parallel.
On the sphere of the radius a consider the parallel

θ(t) = θ0 , ϕ(t) = t, 0 ≤ t ≤ 2π (205)

In cartesian coordinates equation of parallel will be:

   
a sin θ(t) cos ϕ(t) a sin θ0 cos t
r(t) =  a sin θ(t) sin ϕ(t)  = a sin θ(t) sin t , 0 ≤ t ≤ 2π (206)
−a cos θ(t) −a cos θ0

It is easy to see that the family of the vectors eϕ (θ0 , ϕ(t)) on parallel, is
not parallel transport! because deϕ (θdt0 ,ϕ(t)) = deϕdϕ
(θ0 ,ϕ)
is not equal to zero (see
(204) above). Let a family of vectors X(t) be a parallel transport of the vector
eϕ along the parallel (205): X(t) = a(t)eθ (t) + b(t)eϕ (t) where a(t), b(t) are
components of the tangent vector X(t) with respect to the basis eθ , eϕ at
the point θ = θ0 , ϕ = t on the sphere. Initial conditions for coefficients are
a(t)|t=0 = 0, b(t)|t=0 = 1 According to the definition of parallel transport and
formulae (201)—(204) we have:
µ ¶
dX(t) d (a(t)eθ (t) + b(t)eϕ (t)) da(t) db(t)
= = eθ + a(t) cos θ0 eϕ + eϕ +
dt dt dt dt
b(t) (− sin θ0 n − cos θeθ ) =
µ ¶ µ ¶
da(t) db(t)
= − b(t) cos θ0 eθ + + a(t) cos θ0 eϕ − b(t) sin θ0 n (207)
dt dt
Under parallel transport only orthogonal component of the vector changes.
Hence we come to differential equations
(
da(t)
dt
− wb(t) = 0
db(t) a(0) = 0, b(0) = 0, w = cos θ0 (208)
dt
+ wa(t)

62
The solution of these equations is a(t) = sin wt, b(t) = cos wt. We come to
the following answer: parallel transport along parallel θ = θ0 of the initial
vector eϕ is the family
X(t) = sin wt eθ + cos wt eϕ , w = cos θ0 (209)
During traveling along the parallel θ = θ0 the eθ component becomes non-
zero At the end of the traveling the initial vector X(t)|t=0 = eϕ becomes
X(t)|t=2π = sin 2πweθ + cos 2πweϕ : the vector eϕ after woldtrip travel-
ing along the parallel θ = θ0 transforms to the vector sin(2π cos θ0 )eθ +
cos(2π cos θ0 )eϕ . In particularly this means that the vector eϕ after
parallel transport will rotate on the angle
angle of rotation = 2π cos θ0
Compare the angle of rotation with the area of the segment of the sphere
above the parallel θ = θ0 . According to the formula (188) area of this segment
is equal to S = 2πah = 2πa2 (1 − cos θ0 ). On the other hand Gaussian
curvature of the sphere is equal to a12 . Hence we see that up to the sign angle
of rotation is equal to area of the seqment divided on the Gaussian curvature:
S
∆ϕ = ± = ±2π cos θ0 (210)
K

4.3 Parallel transport along a closed curve on arbi-

trary surface.
The formula above for the parallel transport along parallel on the sphere
keeps in the general case.
Theorem Let M be a surface in E3 . Let γ(t) : r(t), t1 ≤ t ≤ t2 , r(t1 ) =
r(t2 ) be a closed curve on the surface M such that it is a boundary of do-
main D of the surface M . (We suppose that the domain D is bounded an
orientate.) Let X(t) be a parallel transport of the arbitrary tangent vector
along this closed curve, i.e. for every t ∈ [t1 , t2 ] X(t) is a vector tangent to
the surface M attached at the point r(t) of the curve γ(t) such that dX(t) dt
is the vector orthogonal to the surface. Consider initial and final vectors
X(t1 ), X(t2 ). They have the same length. The angle ∆ϕ between these
vectors is equal to the integral of Gaussian curvature over the domain D:
Z p
δϕ = ± K det gdudv (211)
D

63
 The calculations above for traveling along the parallel are just example
of this Theorem. The integral of Gaussian curvature over the domain above
parallel θ = θ0 is equal to K · 2πa(1 − cos θ0 )= a12 · 2πa2 (1 − cos θ0 ) = 2π(1 −
cos θ0 ). This is equal to the angle of rotation 2π cos θ0 (up to a sign and
modulo 2π). Another simple
Example. Consider on the sphere x2 + y 2 + z 2 = a2 points A = (0, 0, 1),
B = (1, 0, 0) and C = (0, 1, 0). Consider arcs of great circles which connect
these points. Consider the vector ex attached at the point A. This vector is
tangent to the sphere. It is easy to see that under parallel transport along
the arc AB it will transform at the point B to the vector −ez . The vector
−ez under parallel transport along the arc BC will remain the same vector
−ez . And finally under parallel transport along the arc CA the vector −ez
will transform at the point A to the vector −ey . We see that under traveling
along the curvilinear triangle ABC vector ex becomes the vector −ey , i.e. it
rotates on the angle π2 . It is just the integral of the curvature a12 over the
2
triangle ABC: K · S = a12 · 4πa8
= π2 .

4.4 Gauss Bonnet Theorem

Consider the integral of curvature over whole closed surface M . According
to the Theorem above the answer has to be equal to 0 (modulo 2π), i.e. 2πN
where N is an integer, because this integral is a limit when we consider very
small curve. We come to the formula:
Z Z p
Kdv = ± K det gdudv = 2πN
D D

What is the value of integer N ?

We present now one remarkable Theorem which answers this question.
For more detail see the section 6.
Let M be a closed orientable surface.9 All these surfaces can be classi-
fied up to a diffeomorphism. Namely arbitrary closed oriented surface M is
diffeomorphic either to sphere (zero holes), or torus (one hole), or pretzel
(two holes),... ”Number k” of holes is intuitively evident characteristic of the
9
Closed means compact surface without boundaries. Intuitively orientability means
that one can define out and inner side of the surface. In terms of normal vectors ori-
entability means that one can define the continuous field of normal vectors at all the
points of M . The direction of normal vectors at any point defines outward direction.
Orientable surface is called oriented if the direction of normal vector is chosen.

64
surface. It is related with very important characteristic—Euler characteristic
κ(M ) by the following formula:
κ(M ) = 2(1 − g(M )), where g is number of holes (212)
Remark What we have called here ”holes” in a surface is often referred
to as ”handles” attached o the sphere, so that the sphere itself does not have
any handles, the torus has one handle, the pretzel has two handles and so
on. The number of handles is also called genus.
Euler characteristic appears in many different way. The simplest appear-
ance is the following:
Consider on the surface M an arbitrary set of points (vertices) connected
with edges (graph on the surface) such that surface is divided on polygons
with (curvilinear sides)—plaquets. (”Map of world”)
Denote by P number of plaquets (countries of the map)
Denote by E number of edges (boundaries between countries)
Denote by V number of vertices.
Then it turns out that
P − E + V = κ(M ) (213)
It does not depend on the graph, it depends only on how much holes has
surface.
E.g. for every graph on M , P − E + V = 2 if M is diffeomorphic to
sphere. For every graph on M P − E + V = 0 if M is diffeomorphic to torus.
Now we formulate Gauß -Bonnet Theorem.
Let M be closed oriented surface in E3 .
Let g = gik dui duk d be induced Riemanian metric on this surface, i.e. first
quadratic form and K(p) Gaussian curvature at any point p of this surface.
Recall that sign of Gaussian curvature does not depend on the orienta-
tion. If we change direction of normal vector n → −n then both principal
curvatures change the sign and Gaussian curvature K = det A/ det G does
not change the sign 10 .
10
For an arbitrary point p of the surface M one can always choose cartesian coordinates
(x, y, z) such that surface in a vicinity of this spoint is defined by the equation z =
ax2 + bx2 + . . . , where dots means terms of the order higher than 2. Then Gaussian
curvature at this point will be equal to ab. If a, b have the same sign then a surfaces looks
as paraboloid in the vicinity of the point p. If If a, b have different signs then a surfaces
looks as saddle in the vicinity of the point p. Gaussian curvature is positive if ab > 0 (case
of paraboloid) and negative if ab < 0 saddle

65
 Theorem (Gauß -Bonnet) The integral of Gaussian curvature over the
closed compact oriented surface M is equal to 2π multiplied by the Euler
characteristic of the surface M
Z p
1
K det g dudv = κ(M ) = 2(1 − number of holes) (214)
2π M

In particular for the surface M diffeomorphic to the sphere κ(M ) = 2,

for the surface diffeomorphic to the torus it is equal to 0.
The value of the integral does not change under continuous deformations
of surface! It is integer number (up to the factor π) which characterises
topology of the surface.
E.g. consider surface M which is diffeomorphic to the sphere. If it is
sphere of the radius R then curvature is equal to R12 , area of the sphere is
4π
equal to 4πR2 and left hand side is equal to 2π = 2.
If surface M is an arbitrary surface diffeomorphic to M then metrics and
curvature depend from point to the point, Gauß -Bonnet states that integral
nevertheless remains unchanged.
Very simple but impressive corollary:
Let M be surface diffeomorphic to sphere in E3 . Then there exists at least
one point where Gaussian curvature is positive.
R √
Proof: Suppose it is not right. Then M K det gdudv ≤ 0. On the other
hand according to the Theorem it is equal to 4π. Contradiction.
In the first section in the subsection ”Integrals of curvature along the
plane curve” we proved that the integral of curvature over closed convex
curve is equal to 2π. This Theorem seems to be ”ancestor” of Gauß-Bonnet
Theorem11 .

5 Levi Civita Connection on Riemannian man-

ifold
5.1 Affine connection
How to differentiate functions, vector fields on a (smooth )manifold M ? along vector
fields.
11
Note that there is a following deep difference: Gaussian curvature is internal property
of the surface: it does not depend on isometries of surface. Curvature of curve depends
on the position of the curve in ambient space.

66
 First of all consider differentiation of functions along vector fields.
∂
Let X = Xi (x)ei (x) be a vector field on M (ei (x) = ∂x i ). Recall that vector field
12
i i i i
X = X ei defines at the every point x0 an infinitezimal curve: x (t) = x0 + tX .
∂
Let f be an arbitrary (smooth) function on M and X = X i ∂x i . Then derivative of
i ∂
function f along vector field X = X ∂xi is equal to

∂f
∇X f = X i
∂xi
The geometrical meaning of this definition is following: If X is a velocity vector of the
curve xi (t) at the point xi0 = xi (t) at the ”time” t = 0 then the value of the derivative
∇X f at the point xi0 = xi (0) is equal just to the derivative by t of the function f (xi (t))
at the ”time” t = 0:
¯ dxi (t) ¯¯ ¯ d ¡ ¢¯
if X i (x)¯x = , then ∇X f ¯xi =xi (0) = f xi (t) ¯t=0 (215)
0 =x(0) dt t=0 dt
One can see that the operation ∇X satisfies the following conditions:

• ∇X (af + bg) = a∇X f + b∇X g where λ ∈ R (linearity over numbers )

• ∇hX+gY (f ) = h∇X (f ) + g∇Y (f ) (linearity over functions)
• ∇X (λf g) = f ∇X (λg) + g∇X (λf ) (Leibnitz rule)

(216)

How to define differentiation of vector fields along vector fields.

The formula (215) cannot be generalized, because vectors at the point x0 and x0 + tX
are vectors from different vector spaces. (We cannot substract the vector from one vector
space from the vector from the another vector space, because apriori we cannot compare
vectors from different vector space)
One have to define an operation of transport of vectors from the space Tx0 M to the
point Tx0 +tX M 13 .
Try to define the operation ∇ on vector fields such that conditions (216) above be
satisfied.
Definition Affine connection on M is the operation ∇ which assigns to every vector
field X a linear map, (but not C(M )-linear map!) (i.e. a map which is linear over numbers
not over functions) ∇X on the space O(M ) of vector fields:

∇X (aY + bZ) = a∇X Y + b∇X Z, for every constants a, b (217)

(Compare the first condition in (216)).

which satisfies the following conditions:
12
here like always we suppose
Pn by default the summation over repeated indices. E.g.X =
X ei is nothing but X = i=1 X i ei
i
13
one can define this transport depending on the path from the point Tx0 M to the point
Tx0 +tX M

67
 • for arbitrary functions f, g on M

∇f X+gY (Z) = f ∇X (Z) + g∇Y (Z) (C(M )-linearity) (218)

(compare with second condition in (216))

• for arbitrary function f

∇X (f Y) = (∇X f ) Y + f ∇X (Y) (Leibnitz rule) (219)

(Compare with Leibnitz rule in (216)).
The vector field ∇X Y is called covariant derivative of vector field Y along the vector
field X.

Write down explicit formulae.

Using properties above one can see that
¡ ¡ ¢¢
∇ X Y = ∇ X i e i Y k ek = X i ∇ i Y k ek , where ∇i = ∇ei (220)

Then according to (218)

¡ ¢ ¡ ¢
∇i Y k ek = ∇i Y k ek + Y k ∇i ek

The vector field ∇i ek can be decomposed by basis:

∇ i ek = Γ m
ik em

and
¡ ¢ ∂Y k (x)
∇ i Y k ek = ek + Y k Γm ik em , (221)
∂xi
∂Y m (x)
∇X Y = X i em + X i Y k Γ m
ik em , (222)
∂xi
In components µ ¶
m i ∂Y m (x)
(∇X Y) =X + Y k Γm
ik
∂xi
Coefficients {Γm
ik } are called Christoffel symbols.

Example of affine connection

It follows from the properties of connection that it is suffice to define connection at
vector fields which form basis at the every point.
For example consider n-dimensional euclidean space En . Consider n vector fields
r1 (x), . . . , rn (x) such that

they are linearly independent at any point of En (223)

In other words they form basis at every point.

Define connection such that it obeys the condition:

∇rk (x) rq (x) = 0 for every k, q = 1, 2, 3, . . . n (224)

68
 In the case if {rk (x)} is just the standard basis on En then we come to the standard
connection. We consider arbitrary vector fields satisfying the condition (223). In the case
if {rk (x)} is just the standard basis on En then we come to the standard connection in
En .
The relation (224) defines covariant derivative ∇X Y for arbitrary vector fields X, Y.
Indeed expand vector fields with respect to the basis {rk (x)}:

X = X k (x)rk (x), Y = X = X q (x)rq (x)

(here as always in condensed notations we have summation over indices k, q).

Using relations (217)—(219) we see that in the same way as in (221)

∇X Y = ∇X k (x)rk (Y q (x)rq ) = (225)

 
X k (x) (∇rk Y q (x)) +X k (x) (∇rk rk ) Y q  rq
| {z } | {z }
I II

Second term vanishes according (224). The first term is just derivative of function along
vector field (see (215)).
Remark Of course we can define the connection on the basis taking the right hand
side in (224) not zero, but arbitrary Γi km. In this case second term in the last relation
will not vanish.

5.2 Parallel transport II

Let M be a manifold with affine connection ∇ on it.
Let γ : xi (t), a ≤ t ≤ b be a curve on M Let v(t) = dx(t)
dt be a velocity vector of the
curve γ. Let Y(x) be an arbitrary vector field on M . Then one can consider the vector
field covariant derivative ∇v Y defined just at the points x(t) of the curve γ:
µ ¶
¯ ∂Y m (x)
¯ i m i
∇v Y t = v (t)∇i (Y em )|t = v (t) m k
|x(t) + Γik Y |x(t) em (226)
∂xi

Note that vector field ∇v Y is well-defined at the points of the curve even if the vector
field Y is defined only at the points of this curve, because

∂Y m (x) dY m (x(t))
v i (t) i
|x(t) = |x(t)
∂x dt
Hence for every vector field Y(t) attached at the points x(t) of the curve (Y(t) ∈ Tx(t) M )
the connection ∇ defines at the points of this curve a vector field ∇v Y:
¯ dY m (x(t))
∇v Y ¯t = |x(t) + v i (t)Γm k
ik Y |x(t) (227)
dt
It is covariant derivative of Y along the curve.
Now we are able to define parallel transport of vector field.

69
 Definition
Let γ : xi (t), a ≤ t ≤ b be a curve on M .
The family of vectors Y(t), where the vector Y(t) is a vector attached at the point
x(t) (a ≤ t ≤ b, Y(t) ∈ Tx(t) M ) is called a parallel transport of the initial vector Y(t0 )
along a curve γ : xi (t) if covariant derivative ∇v Y ≡ 0 at all the points of curve, i.e.

dY i (x(t)) dxi (t)

+ v k (x(t))Y m (x(t))Γikm (x(t)) = 0, where v i (t) = (228)
dt dt
In the case if Γikm (x(t)) = 0 then parallel transport of the vector means just preserv-
ing the components of the vector. We consider later really interesting examples of parallel
transport, when the final vector depends on the curve, i.e. if γ1 , γ2 are two curves con-
necting the points A, B of the manifold, then parallel transport of initial vector Y from
the point A to the point B is different for curves γ1 , γ2 .

5.3 Levi-Civita connection

Let M be a Riemannian manifold with metrics G.
Recall that metrics defines scalar product of vector fields:

< X, Y >= gik X i Y k

Let ∇ be a connection on M :
∂Y k
∇X Y = X i + X i Y k Γm
ik
∂xi
We say that this connection is symmetric connection if Christophel symbol Γm
ik satisfies
the condition 14 :
Γm m
ik = Γki

Definition Symmetric connection ∇ on the Riemanian manifold (M, G) is called

Levi-Civita connection if it preserves the scalar product It means the following:
Let γ : x(t) be arbitrary curve on M and X, Y arbitrary vectors attached at the initial
point of this curve. Let X(t), Y(t) be parallel transport of these vectors. Then scalar
product (X(t), Y(t)) of these vectors does not depend on t. In particular it means that
under parallel transport length of the vector does not change: (consider X = Y we see
that (X(t), Y(t)) = |X|2 is preserved) and the angle between vectors does not change:
((X(t), Y(t)) = |X||Y| cos ϕ).
14
In a more invariant way one can define define a symmetric connection ∇ as a connection
which satisfies the condition:

∇X Y − ∇Y X − [X, Y] = 0

The left hand side of the formula above defines the torsion of the connection.

70
 One can show that Levi-Civita connection on Riemanian manifold can be uniquely
defined by the Riemanian metric G = Gik dxi dxk (See the subsection in Appendix)
Here we consider only the special case of two-dimensional surface, in E3 when Riema-
nian metric is defined by first quadratic form, i.e. it is induced by the metric on E3 .
In this case vectors tangent to surface can be viewed as vectors in E3 and their length
is just the standard length of the vector in E3 .
The Levi-Civita connection defines the parallel transport of an arbitrary vector along
an arbitrary curve.
Using the fact that connection have to be symmetric one can prove the following
Proposition:
Proposition
Consider surface M in E3 with induced Riemanian metric, i.e. with metric defined
by the first quadratic form.
Let γ : r(t), a ≤ t ≤ b be arbitrary curve on this surface. Let Y(t) be parallel transport
of the vector Y along this curve with respect to Levi-Civita connection on the surface M .
Then for vector field Y(t) the following conditions hold:
• Y(t)|t=a = Y (initial condition)
• Y(t) is always tangent to the surface
(Y(t), n(t)) = 0, where n(t) is normal vector (229)
dY(t)
• only normal component of Y changes, i.e. derivative dt is proportional to the
normal vector:
dY(t)
= λ(t)n(t) (230)
dt
These conditions uniquely define parallel transport15 .
In particular it follows from these conditions that the length of the vector Y(t) is
preserved:
µ ¶
d d dY(t)
|Y(t)|2 = (Y(t), Y(t)) = 2 , Y(t) = (λ(t)n(t), Y(t)) = 0
dt dt dt
The statement of this Proposition is very useful criterium for constructing parallel
transport of vector fields along curves in surfaces. Consider
Example Consider a sphere x2 + y 2 + z 2 = 1 in E3 with induced Riemanian met-
ric=first quadratic form. Consider the vector Y = ey + ez attached at the point A =
(1, 0, 0). It is evident that this vector is tangent to the sphere at the point A. Consider
the arc of the great circle x = cos t, y = sin t, z = 0, 0 ≤ t ≤ π/2 beginning at the point A.
Find parallel transport of the vector Y along this curve. In this simple case it is very easy
to guess an answer, and then to check is it right or no. Vector ez at all the points of the
circle x = cos t, y = sin t, z = 0 remains tangent vector. Vector ey has to be transformed
to remain tangent. (it has to be rotated: ey → ey cos t − ex sin t) Consider vector field
Y(t) = ez + ey cos t − ex sin t
15
it is just a special case of parallel transport considered in the section 4 above

71
.
Check that conditions
(229), (230) of proposition are satisfied.
1. Y(t)|t=0 = ez + ey = Y
Normal vector n(t) is equal to r = ex cos t + ex sin t

(Y(t), n(t)) = (ez + ey cos t − ex sin t, ex cos t + ey sin t) = 0,

Hence condition (229) is satisfied.

dY(t) d
= (ez + ey cos t − ex sin t) = −ex cos t − ey sin t = −n(t)
dt dt
Condition (230) is satisfied too. Hence it is indeed parallel transport.
Remark One can see that if we will do the parallel transport along a closed curve
r(t), r(a) = r(b) then initial vector and the final vector will be different. E.g. consider close
curved triangle ABC formed by three arcs of great circles: AB is the arc x = sin t, y =
0, z = cos t, 0 ≤ t ≤ π2 , BC is the arc x = cos t, y = sin t, z = 0, 0 ≤ t ≤ π2 and CA is the
arc x = 0, y = cos t, z = sin t, 0 ≤ t ≤ π2
If the initial vector at the point A is ey , then its parallel transport at the point B
will be again ey , parallel transport at the point C will be the vector −ex and finally the
returned vector at the point A will be the vector − − ex . The vector ey transforms to the
vector −ex . It is indication of the fact that there is no flat metric on the sphere 16

6 Gaussian map and Gauss Bonnet Theorem

We try to give some ideas which lead to understanding and proof of the Gauss Bonnet
Theorem (see the previous section).

6.1 Gaussian map

Let M : r = r(u, v) be an oriented surface in E3 and n(u, v) normal vector filed. (Recall
that at the every point r(u, v), the normal vector n(u, v) is orthogonal to the surface:
(n, ru ) = (n, rv ) = 0) and the length of the vector n is equal to 1). Unit vector n =
(nx , ny , nz ) can be considered as a point on the unit sphere x2 + y 2 + z 2 = 1. Thus
16
One can prove the following very beatiful formula: let γ be a closed curve which is a
boundary of the domain on the surface M . Let Y(t1 ) is parallel transport of the initial
vector Y after travelling along the closed curve. Then the angle δϕ between these angles
is equal to the integral of the Gaussian curvature of the surface over domain D:
Z p
δϕ = K det gdudv (231)
D

In particular for the sphere of radius R δϕ = AreaR2of D

72
we define a map which assigns to every point p = r(u, v) ∈ M the point on the unit
sphere=unit vector n(u, v): This map is called Gaussian map.
Definition Gaussian map maps every point p of the oriented surface M to the normal
unit vector n(p)—point of the unit sphere:

Gaussian map n : M → S 2 M 3 r(u, v) 7→ n(u, v) ∈ S 2 (232)

The direction of normal vector is defined by the orientation of the surface.

Examples
1.Plane Ax + By + Cz = 1.
For all points of the plane unit normal vector n is the same vector. It is easy to see
that  
A
1  B 
n= √
A2 + B 2 + C 2 C
³ ´
A B C
The image of Gaussian map is just the point √A2 +B 2 +C 2
, √A2 +B 2 +C 2
, √A2 +B 2 +C 2
on
the unit sphere.
2. Cylinder x2 +y 2 = a2 . Normal vector n(h, ϕ) at the point r(h, ϕ) = (a cos ϕ, a sin ϕ, h)
is equal to  
cos ϕ
n(h, ϕ) =  sin ϕ 
0
Gaussian map maps the cylinder on the equator points (cos ϕ, sin ϕ, 0) of the sphere.
3. Upper part of the Cone: x2 + y 2 − k 2 z 2 = 0, z ≥ 0.
The unit normal vector (see (160)) at the point (kh cos ϕ, kh sin ϕ, h) is equal to
 
cos ϕ
1  sin ϕ 
n= √
1 + k2 −k

Gaussian map maps the upper part of the cone on the circle
k 1
(sin θ cos ϕ, sin θ sin ϕ, cos θ) on unit sphere where cos θ = − √1+k 2
, sin θ = √1+k 2
. (An-
other part of the cone (z < 0) maps under Gaussian map to the circle (sin θ cos ϕ, sin θ sin ϕ, − cos θ))
4.Sphere x2 +y 2 +z 2 = R2 . If r is the point on the sphere then unit vector is just equal
to r/R. Every point r of the sphere maps to the point r/R of the unit sphere. Gaussian
map is one-one map.
Sphere is convex surface. It is a boundary of the ball which is convex body.
Consider arbitrary convex surface. (We call the closed surface convex if it is a boundary
of convex domain. Domain D in E3 is called convex if for arbitrary two points a, b ∈ D
all the points of the interval [a, b] belong to D).

73
 One can see that Gaussian map establishes one-one correspondence between points of
the surface M and points of unit sphere17
5. Torus: (c) Consider the torus M in E3 given by parameterisation:


x = (a + b cos ϕ1 ) cos ϕ2
r(ϕ1 , ϕ2 ) : y = (a + b cos ϕ1 ) sin ϕ2 ,


z = b sin ϕ1

where 0 ≤ ϕ1 < 2π, 0 ≤ ϕ2 < 2π and a, b are constants such that 0 < b < a.
One can prove that normal unit vector at the point r(ϕ1 , ϕ2 ) is equal to
 
cos ϕ1 cos ϕ2
n(ϕ1 , ϕ2 ) =  cos ϕ1 sin ϕ2  ,
sin ϕ1

To prove it calculate the length of the vector n and prove that it is orthogonal to tangent
vectors rϕ1 , rϕ2 . (Do it!)
Torus is not convex surface. Image of Gaussian map is whole unit sphere, but the
map is not one-one correspondence. Every unit vector n (point of the unit sphere) has
two pre images. E.g. consider n = (0, 1, 0). Then it follows from the previous formula
that cos ϕ1 cos ϕ2 = 0, cos ϕ1 sin ϕ2 = 1, sin ϕ1 = 0. It implies two cases:
cos ϕ1 = 1, sin ϕ1 = 0, cos ϕ2 = 0, sin ϕ2 = 1, i.e. point on he torus (0, a + b, 0)
or cos ϕ1 = −1, sin ϕ1 = 0, cos ϕ2 = 0, sin ϕ2 = −1, i.e. point on he torus (0, −(a −
b), 0)
In the first three cases (plane,cylinder, cone) Gaussian curvature of surfaces is equal
to zero and image of Gaussian map is point (for plane) or curve (for cylinder and conus)
In the case of convex surface and torus image of Gaussian map is whole sphere.
In the case of convex surface the Gaussian map is one-one-correspondence. The Gaus-
sian curvature at all the points is positive (see the footnote before (214)) and according to
the Theorem the integral of curvature is equal to 4π. In the case of torus Gaussian map
is not one-one-correspondence. The Gaussian curvature is positive , negative or equal to
zero depending on the points of torus. Gauss-Bonnet Theorem tells that not only for torus
but for every surface diffeomorphic to torus integral of gaussian curvature over the surface
is equal to zero.

6.2 Gauß-Bonnet Theorem for convex surface

Here we give basic ideas to prove Gauß -Bonnet Theorem for convex surfaces. As it was
noted in the example 4 the Gaussian map for these surfaces establishes one-one correspon-
17
To prove this construct the map inverse to gaussian map on the unit sphere. Let n be
an arbitrary unit vector. Consider a plane l(n) tangent to this vector. It can be proved
that there exists unique plane which is parallel to the plane l(n) and which touches the
surface M at same point p. It is easy to see that n = n(p).

74
dence between points of surface and points of unit sphere.We prove that in this case:
Z p
1
K det g dudv = 2 (233)
2π M
It is just Gauß-Bonnet Theorem for convex surfaces, because evidently convex surfaces
are diffeomorphic to sphere (The convex domains are diffeomorphic to balls)
The proof of (233) in the case if Gaussian map establishes one-one correspondence
between points of surface and points of unit sphere follows from
Proposition
Let M be an oriented surface in E3 . Consider at arbitrary point p of this surface tan-
gent vectors a, b. Under the action of differential of Gaussian map vectors a, b transform
to the vectors a0 , b0 . Let K(p) be gaussian curvature of the surface M at the point p. Then

K(p)S = S 0 (234)

where S is an area of the parallelogram formed by the vectors a, b, S 0 is an area of paral-

lelogram formed by the vectors a0 , b0
Remark We consider the signed area of parallelogram. The modulus of area of
parallelogram is the length of the cross product of vectors a, b: |S| = |a × b|. The signed
area is defined by the direction of normal vector. It is equal to

S = (n, a × b) (235)

If n → −n, S → −S. The direction of normal vector is defined by the orientation.

Show that this Proposition leads to the proof of (233).
Consider the covering of closed surface M by the collection {Πk } of infinitesimal
parallelograms. Every infinitesimal parallelogram Πk is attached to the point pk of the
surface and is formed by the tangent vectors ak , bk , i.e. it has sides εak , εbk .
Consider Gaussian map of the surface M into S 2 . Gaussian map establishes one-
one correspondence. Hence under this map the covering of surface M by the collection
{Πk } of infinitesimal parallelograms transforms onto the covering of unit sphere S 2 by the
collection {Π0k } of infinitesimal parallelograms. According to (234)
X X
Area of Π0k ≈ K(pk )Area of Π0k

These sums tend to corresponding integrals. Left hand side of thisR relation
√ tends to the
area of unit sphere. The right hand side of this relation tends to M K(p) det g dudv. We
come in the limit to the relation
Z p
4π = area of unit sphere = K(p) det g dudv
M

It is just (233)
Now we give a
Proof of the Proposition (234)
The proof of the Proposition follows from the following Lemma:

75
 Lemma In the vicinity of the given point p of the surface M consider unit normal
vector field n(u, v), i.e. Gaussian map (232). Then the action of differential dn on
arbitrary tangent vector a is equal up to the sign to the action of the shape operator at this
vector
dn(a) = −Sa (236)
(Shape operator at the point p (more precisely acting on the tangent vectors attached at
the point p) is equal to S = G−1 A, where G is first quadratic form at the point p and A
is second quadratic form at the point p, (see in detail Section 2))
The proof of this Lemma follows from definition of shape operator18
Show that Proposition follows from the Lemma.
Let vectors a, b are attached to the point p. Let under the action of differential of
Gaussian map the vector a transforms to the vector a0 and vector b transforms to the
vector b0 :
dn(a) = a0 , dn(b) = b
According to the Lemma:

a0 = −S(a), b0 = −S(b) (237)

where S = G−1 A is shape operator acting on the tangent vectors at the point p. Write
down in components these relations. We will write vectors a, b, a0 , b as columns 2 × 2
matrix:
µ 1 1 ¶ µ 01 01 ¶ µ 1 ¶
a b 0 0 a b s1 s12
(a, b) → , (a , b ) → , shape operator S = ,
a2 b2 a02 b02 s21 s22

Then relations (237) will have an appearance:

µ 1 ¶ µ 1 1 ¶ µ 01 ¶
s1 s12 a b a b01
= (238)
s21 s22 a2 b2 a02 b02

Take determinant of this relation and use the fact that determinant is multiplicative:
det(AB) = det ·A det B. We come to the relation
µ 1 ¶ µ 1 1 ¶ µ 01 01 ¶
s1 s12 a b a b
det · det = det (239)
s21 s22 a2 b2 a02 b02
| {z } | {z } | {z }
I II III

Remembering the definition of Gaussian curvature we see that the first term is equal just
to the Gaussian curvature at the point p: K = det S = det(G−1 A) = det A/ det G.
Consider parallelogram Π formed by the vectors a, b and parallelogram Π0 formed
by the vectors a0 , b0 . Second and third determinants (up to a sign) are just areas of
18
Let a0 = dn(a) = aα ∂α ni . This vector is a vector tangent to M , because it is or-
thogonal to n. Hence a0 = a0α rα = aα ∂α ni . Multiplying both parts by rβ we come to
a0α (rα , rβ ) = a0α gαβ = aα (∂α ni , rβ ). But (∂α ni , rβ ) = −(ni , rαβ ) because (ni , rβ ) = 0.
Hence a0α gαβ = −aα Aαβ . This leads to (236).

76
parallelograms Π, Π0 . If we change the direction of normal vectors in (235) they both
change a sign19 . Hence the last relations is just:

K(p) · Area of the parallelogram Π = Area of the parallelogram Π (240)

It is just the statement of Proposition.

7 Appendices
7.1 Geodesics on the sphere and on Lobachevsky plane
In two-dimensional case the following lemma helps to find geodeics:
Lemma Consider metric which has the following appearance in the local coordinates
u, v: a(u)du2 + b(u, v)dv 2 where a, b > 0. Then for every curve u(t), v(t), t0 ≤ t ≤ t1 the
following inequality holds
Z t1 q Z t1 q
2 2
a(u)ut + b(u, v)vt dt ≥ a(u)u2t dt = (241)
t0 t0
Z t1 p Z u1 p
a(u)ut dt = a(u)du
t0 u0

where u0 = u(t0 ), u1 = u(t1 ).

The proof of the lemma is obvious.
From this lemma it follows immediately that the lines v = consta are geodesics of the
metric a(u)du2 + b(u, v)dv 2 .
We can use this lemma to find geodesics of sphere and Lobachevsky plane.

Geodesics of sphere (See the subsection 3.3)

Consider riemannian metrics on the sphere in E3 with the radius a: Coordinates θ, ϕ,
metrics (first quadratic form):

G = a2 (dθ2 + sin2 θdϕ2 ) (242)

Consider two arbitrary points A and B on the sphere. Let (θ0 , ϕ0 ) be coordinates of the
point A and (θ1 , ϕ1 ) be coordinates of the point B
Let γ be a curve which connects these points: γ : θ(t), ϕ(t) such that θ(t0 ) = θ0 , θ(t1 ) =
θ1 , ϕ(t0 ) = θ0 , θ(t1 ) = θ1 then:
Z q
LγAB = a θt2 + sin2 θ(t)ϕ2t dt (243)
µ ¶
19 a1 b1
to see that (n, [a, b]) = ± det we note that left hand side and right hand
a2 b2
side of this expression both are bilinear antisymmetric forms which coincide (up to a sign)
on the vectors a = (1, 0), b = (0, 1)

77
 Without loss of generalisity suppose that they have the same latitude, i.e. if (θ0 , ϕ0 )
are coordinates of the point A and (θ1 , ϕ1 ) are coordinates of the point B then ϕ0 = ϕ1
(if it is not the fact then we can come to this condition rotating the sphere)
Now it is easy to see from the lemma that ϕ = ϕ0 is geodesics: Indeed consider an
arbitrary curve θ(t), ϕ(t) which connects the points A, B: θ(t0 ) = θ0 , θ(t1 ) = t1 , ϕ(t0 ) =
ϕ(t1 = ϕ0 . Compare its length with the length of the meridian which connects the points
A, B: Z t1 q Z t1 q Z t1
a θt2 + sin2 θϕ2t dt ≥ a θt2 dt = a θt dt = a(θ1 − θ0 ) (244)
t0 t0 t0

the big circles on sphere are geodesics. It corresponds to geometrical intuition: The
geodesics on the sphere are the circles of intersection of the sphere with the plane which
crosses the centre.

Lobachevsky plane and its geodesics

One of the model of Lobachevsky geometry is following: Consider upper plane of E2 :
(x, y) with y ≥ 0 with metric
dx2 + dy 2
ds2 = (245)
y2
The length of the curve γ : x = x(t, y = y(t)) is equal to
Z s 2
xt + yt2
L= dt
y 2 (t)

In particularly the length of the vertical interval [1, ε] tends to infinity if ε → 0:

Z s 2 Z 1r
xt + yt2 1 1
L= dt = dt = log
y 2 (t) ε t 2 ε

One can see that the distance from every point to the line y = 0 is equal to infinity. This
motivates the fact that the line y = 0 is called absolute.
It is easy to see from lemma that vertical lines are geodesics of Lobachevsky plane.
Find geodesics which connects two points A, B. Consider semicircle which passes these
two points such that its centre is on the absolute.
We prove that it is a geodesic.
Proof Let coordinates of the centre of the circle are (a, 0). Then consider polar coor-
dinates (r, ϕ):
x = a + r cos ϕ, y = r sin ϕ (246)
In these polar coordinates r-coordinate of the semicircle is constant.
Find Lobachevsky metric in these coordinates: dx = −r sin ϕdϕ + cos ϕdr, dy =
r cos ϕdϕ + sin ϕdr, dx2 + dy 2 = dr2 + r2 dϕ2 . Hence:

dx2 + dy 2 dr2 + r2 dϕ2 dϕ2 dr2

ds2 = = 2 == + (247)
y2 r2 sin ϕ sin ϕ r2 sin2 ϕ
2

78
We see that the length of the arbitrary curve which connects points A, B is greater or
equal to the length of the arc of the circle:
Z t1 s 2 Z t1 s 2
ϕt rt2 ϕt
LAB = 2 + 2 2 dt ≥ dt = (248)
t0 sin ϕ r sin ϕ t0 sin2 ϕ
Z t1 Z ϕ1
ϕt dϕ tan ϕ1
dt = = log
t0 sin ϕ ϕ0 sin ϕ tan ϕ1
The proof is finished.

Why so much attention to Lobachevsky plane?

Lobachevsky plane appears as a first model of non-Euclidean geometry 20
Why so much attention to Lobachevsky plane?
We already know Euclidean plane and sphere. They are two dimensional Rieman-
nian manifolds with 3 isometries.(More exactly three-parametric group of isometries).
Lobachevsky plane has three isometries too. And it is all!
In the class of two-dimensional Riemannian manifolds there are only these three cases
with maximal group of symmetries. (It is easy to show that number of isometries cannot be
more than nn + 1/2 for n-dimensional case.) So in some sense there are only three possi-
bilities for geometry of two-dimensional manifold: usual euclidean, spheric and hyperbolic
(geometry of Lobachevsky plane)

7.2 Surfaces of constant Gaussian curvatures in E3

We want to consider examples of surfaces of constant gaussian curvatures in E3 . If K > 0
the best known example is sphere. Of course we can make a hole in sphere and .... it.
(One can show that globally sphere cannot be ....)
If K = 0 there is again plenty trivial examples: plane, cylinder, cone,...
How to construct a surfaces with K ≡ −1. In other words how to realize Lobachevsky
plane in E3 .
We find the solution to this problem in the class of surfaces:


x = f (h) cos ϕ
r(h, ϕ) : y = f (h) sin ϕ (249)


z=h

Calculate derivatives of r and normal vector n:

     
fh cos ϕ −f sin ϕ cos ϕ
1
rh =  fh sin ϕ  , rϕ =  f cos ϕ  , n= p  sin ϕ  , (250)
1 0 1 + fh2 −f
h

20
Lobachevsky plane has the distinguished role from the point of view of fifth Euclid
axiom.

79
      
fhh cos ϕ −fh sin ϕ −f cos ϕ
rhh =  fhh sin ϕ  , rhϕ =  fh cos ϕ  , rϕϕ =  −f sin ϕ  , (251)
0 0 0
First quadratic form: metric is equal to:
µ ¶ µ ¶
(rh , rh ) (rh , rϕ ) 1 + fh2 0
G= = , G = (1 + fh2 )dh2 + f 2 (h)dϕ2 (252)
(rh , rϕ ) (rϕ , rϕ ) 0 f2

and second quadratic form is equal to

µ ¶ µ ¶
(rhh , n) (rhϕ , n) 1 fhh 0
A= =p (253)
(rhϕ , n) (rϕϕ , n) 2
1 + fh 0 −f

Gaussian and Mean curvature are equal to:

det A −fhh
K = det(G−1 A) = = (254)
det G f (1 + fh2 )2
µ ¶
−1 1 fhh 1
H = Tr(G A) = p − (255)
1 + fh2 1 + fh2 f
To find revolution surfaces with constant (gaussian) curvatures we have to solve dif-
ferential equation:
−fhh
=K (256)
f (1 + fh2 )2
It is evident √from geometrical considerations that solution to this equation at K > 0 is
sphere (f = 1 − h2 ) and its isometries.
In the case K < 0

7.3 On one beautiful formula

Let C be a surface in E3 C : r = r(u, v). let a surface Cw be on the distance w from
this surface, i.e.
rw (u, v) = r(u, v) + wn(u, v), (257)
where n(u, v) is a unit vectro onrthogonal to the surface C at the point r(u, v). (One can
see that this vector will be orthogonal to the surface Cw too). There is a beautiful formula
related Gaussian curvature of Cw with Gaussian and mean curvatures of C: If K, H be
Gaussian curvatures of surface C at the point r0 = r(u0 , v0 ) then Gaussian curvatture of
the surface Cw at the point rw (u0 , v0 ) = r(u0 , v0 ) + wn(u0 , v0 ) is equal to

K
K= (258)
1 − wH + w2 K
In particularly
if the surface C has the constant mean curvature H ≡ h then the surface Cw which is
in the distance w = h1 from the surface C has a constant Gaussian curvature equal to h.

80
 The proof is founded on the formulae for gaussian curvature and very elementary use
of Cayley-Hamilton identities.
Namely if xi = xi (uα ) + wni (ua lpha) (it is conveninet to denote by xi components
of vector r, (i = 1, 2, 3), by uα (α = 1, 2) parameteres u, v). Denote by g(w)αβ A(w)α, β
tensors of metric (first quadratic form) and second quadratic form. Note that vector n is
orthogonal to the surface C as well: (rα + wnα , n) = 0 because

1 ∂ ¡ i i¢
(na , n) = niα ni = n n = 0. (259)
2 ∂uα
Then:

gαβ (w) = (rα +wn, rβ +wn) = (xiα +wniα )(xiβ +wniβ ), Aαβ (w) = (rαβ +wnαβ , n) (260)

It is easy to see from definition of n and (259) that vector na is tangent to the surface
C and the following relations hold:
¡ ¢
(rα , nβ ) = xiα ni β = −xiαβ ni = −Aαβ , (na , nβ ) = A · g −1 · A αβ (261)

Now everything is ready to calculate K(w):

³ ´
det A(w) det Aαβ + wniαβ ni
K(w) = = ³ ´=
det g(w) det gαβ + 2wxiα niβ + w2 niα niβ
³ ´
¡ ¢
det Aαβ − wniα niβ det A − wAg −1 A
³ ´=
det gαβ − 2wxiαβ ni + w2 niα niβ det (g − 2wA + w2 Ag −1 A)

Remember that
det A
K = det S = , H = Tr S, where S = g −1 A
det G
Also it is useful to use the following identity for 2 × 2 matrices:

det(1 + A) = 1 + Tr A + det A (262)

It is the elementariest of Cayley-Hamilton identities.
Now using this we can calculate K(w):

¡ ¢ ¡ ¢
det A − wAg −1 A det A det 1 − wg −1 A
K(w) = = (263)
det (g − 2wA + w2 Ag −1 A) det g det (1 − 2wg −1 A + w2 g −1 Ag −1 A)
¡ ¢
det A det 1 − wg −1 A det A K
= = = (264)
det g det2 (1 − wg −1 A) det g det (1 − wS) 1 − wH + w2 K

81
7.4 Tubes
The ideas of the previous Appendix can be developed.
Consider the function det(1 + zS), where z is formal parameter. This function is
quadratic polynomial in z and coefficients are just mean and Gaussian curvature:

det(1 + zS) = 1 + Hz + Kz 2 (265)

because for operators in 2-dimensional space det(1 + A) = 1 + TrA + detA

If M is a compact surface in E3 then one can consider remarkable polynomial:
Z
√
PM (z) = det(1 + zS) gd2 x (266)
M
√
Here gd2 x is volume form defined by the first quadratic form. On one hand monoms
of this polynomial defines invariants of the surface M , on the other hand for small z this
polynomial defines the area of the surface Mz which is on the distance z from M :
Z Z Z
√ 2 √ 2 √
S(Mz ) = 1 · gd x +z H gd x + K gd2 x
| M {z } | M {z } | M {z }
S(M0 ) S(M )·averaged mean curvature 2π·(Euler number of M )
(267)
One cna prove thisformula in the following way: Consider in tubular neighbourhood
of M coordinates (ξ 1 , ξ 2 , ρ) such that ρ measures he distance from the points till M One
can see that This formula has many applications.
In particular it allows to calculate an average mean and gaussian curvature for singular
surfaces: parallelepiped. It is evident that if M is parallelepiped with sides a, b, c,. Then

S(Mz ) = 2(ab + ac + bc) + πz(a + b + c) + 4πz 2 (268)

Comparing these formulae we see that averaged mean curvature of parallelepiped is equal
π(a+b+c)
to 2(ab+ac+bc)
These formulaes can be easy generalised for hypersurfaces in En

7.5 Levi Civita connection II

Let M be a Riemannian manifold with metrics G.
Recall that metrics defines scalar product of vector fields:

< X, Y >= gik X i Y k

Let ∇ be a connection on M :
∂Y k
∇X Y = X i + X i Y k Γm
ik
∂xi

82
We say that this connection is symmetric connection if Christophel symbol Γm
ik satisfies
the condition 21 :
Γm m
ik = Γki

We say that symmetric connection ∇ is Levi Civita connection if it preserves scalar,

i.e. if for arbitrary vectors Y, Z attached at the arbitrary point the

∇X < Y, Z >=< ∇X (Y) , Z > + < Y, ∇X (Z) > (269)

Theorem On the Riemannian manifold (M, G) there exists uniquely defined Levi
Civita connection. In local coordinates Christoffel symbols of this connection have the
following appearance:
µ ¶
1 mn ∂gin (x) ∂gkn (x) ∂gik (x)
Γm
ik (x) = g (x) + − (270)
2 ∂xk ∂xi ∂xn

Prove this Theorem. Suppose that there exist symmetric connection ∇ satisfying
condition with Christoffel symbols Γm
ik in local coordinates. We show that these coefficients
are defined uniquely by the condition (269).
Rewrite the condition (269) in components for X = em :
¡ ¢ ¡ ¢ ¡ ¢
∂m gik Y i Z k = gik ∂m Y i + Γimr Y r Z k + gik Y i ∂m Z k + Γkmr Z r (271)

Comparing left and right hand sides of this expression for arbitrary vectors Y, Z we see
that
∂m gik = Γk;mi + Γi;mk (272)
where we denote by
Γk;mi = gkr Γrmi
Now using the symmetricity condition Γrmi = Γrim we obtain that

Γi;mk = ∂m gik − Γk;im = ∂m gik − (∂i gkm − Γm;ik ) = ∂m gik − ∂i gkm + ∂k gim − Γi;mk

Hence
1
Γi;mk = (∂m gik + ∂k gim − ∂i gkm ) , Γimk = g ij Γj;mk (273)
2
and we come to (270).
One can see that Levi Civita connection is well defined by Christophel symbols (270).
Example Consider two-dimensional surface with Riemannian metrics
µ ¶ µ ¶
2 2 g11 g12 a(u, v) 0
G = a(u, v)du + b(u, v)dv , G= =
g21 g22 0 b(u, v)
21
In a more invariant way one can define define a symmetric connection ∇ as a connection
which satisfies the condition:

∇X Y − ∇Y X − [X, Y] = 0

The left hand side of the formula above defines the torsion of the connection.

83
Calculate Christoffel symbols of Levi Civita connection.
Using (273) we see that:
1
Γ1;11 = 2 (∂1 g11 + ∂1 g11 − ∂1 g11 ) = 21 ∂1 g11 = 1
2 au

1
Γ1;21 = Γ1;12 = 2 (∂1 g12 + ∂2 g11 − ∂1 g12 ) = 21 ∂2 g11 = 1
2 av

1
Γ1;22 = 2 (∂2 g12 + ∂2 g12 − ∂1 g22 ) = − 12 ∂1 g22 = − 1
2 bu
(274)
1
Γ2;11 = 2 (∂1 g12 + ∂1 g12 − ∂2 g11 ) = − 12 ∂2 g11 = − 1
2 av

1
Γ2;12 = Γ2;21 = 2 (∂2 g21 + ∂1 g22 − ∂2 g21 ) = 21 ∂1 g22 = 1
2 bu

1
Γ2;22 = 2 (∂2 g22 + ∂2 g22 − ∂2 g22 ) = 21 ∂2 g22 = 1
2 bv

To calculate Γikm = g ir Γr;km note that for the metric a(u, v)du2 + b(u, v)dv 2
µ 11 ¶ Ã 1 !
−1 g g 12 a(u,v) 0
G = = 1
g 21 g 22 0 b(u,v)

Hence
au av −bu
Γ111 = g 11 Γ1;11 = 2a , Γ121 = Γ112 = g 11 Γ1;12 = 2a , Γ122 = g 11 Γ1;22 = 2a

−av bu bv
Γ211 = g 22 Γ2;11 = 2b , Γ221 = Γ212 = g 22 Γ2;12 = 2b , Γ222 = g 22 Γ2;22 = 2b
(275)
Example Sphere.
On the sphere first quadratic form (Riemannian metric) G = R2 dθ2 + R2 sin2 θdϕ2
Hence we use calculations from previous example with a(θ, ϕ) = R2 , b(θ, ϕ) = R2 sin2 θ
(u = θ, v = ϕ). Note that aθ = aϕ = bϕ = 0. Hence only non-trivial components of Γ will
be: µ ¶
−bθ − sin 2θ −R2 sin 2θ
Γ122 = = , Γ1;22 = , (276)
2a 2 2
µ ¶
bθ cos θ R2 sin 2θ
Γ212 = = Γ2;12 = (277)
2b sin θ 2
All other components are equal to zero:

Γ111 = Γ112 = Γ121 = Γ211 = Γ222 = 0

7.6 Riemannian curvature

Let M be manifold with connection ∇.
Then one can consider the curvature of this connection defined by the relation:

R(X, Y) = ∇X ∇Y − ∇Y ∇X − ∇[X,Y] (278)

84
 R(X, Y) defines operation on vector field.
i
(R(X, Y)Z) = Z k Rkpq
i
X pY q (279)

where
i
Rkpq = ∂p Γiqk − ∂q Γipk + Γips Γsqk − Γiqs Γspk (280)
is (1.3) tensor.
Christoffel symbol is not a tensor. The object defined above is a tensor.
i
In particularly if Rkpq ≡ 0 in given local coordinates then it is equal to zero in every
local coordinates. This gives constructive answer to the question:
Question: Consider the Riemannian manifold with metric G = gik (u)dui dv k . How
to know do there exis new coordinates such that in these new coordinates metric is flat:
∂ui (x) ∂uk (x) a b
G = gik (u)dui dv k = gik (u(x)) dx dx = (dx1 )2 + · · · + (dxn )2
dxa dxb
To answer this question one have to calculate Levi-Civita connection (270) of the metric
G then the Riemann tensor (280) of this connection. Then:
Theorem Riemann tensor of Levi-Civita connection Γ(g) is equal to zero (in a vicinity
of the point) iff there are local coordinates (in a vicinity of this point) such that metric in
these local coordinates is cartesian.
We will partly to discuss this for two-dimensional manifolds in the next subsections.

7.7 Scalar curvature. Gauss Theorema Egregium

Let M be Riemannian manifold with metric G. Let ∇ be Levi-Civita connection of this
i
metric Rkpq be Riemann curvature tensor.
j
Consider tensor Rikpq = gij Rkpq . One can show that tensor Rikpq obeys the following
properties:
Rikpq = −Rkipq , Rikpq = −Rikqp , Rikpq = Rpqki (281)
One can consider scalar curvature:

R = g ip g kq Rikpq
It is a scalar which is called scalar curvature.
In the case of 2-dimensional space formulae are extremely simple: Tensor Rikpq has
only one non-trivial components which we will denote by p:

R1212 = −R2112 = R2121 = P

Indeed R1112 = R2212 = R1122 = R1211 = · · · = 0 by the condition (281). Scalar curvature
in this case is equal to:

R = g ip g kq Rikpq = g 11 g 22 R1212 + g 12 g 21 R1221 + g 21 g 12 R2112 + g 22 g 11 R2121 =

¡ ¢
P g 11 g 22 − g 12 g 21 − g 21 g 12 + g 22 g 11 = 2P (g 11 g 22 − g 12 g 21 ) =

85
 2p 2R1212
= 2
det G g11 g22 − g12

On two dimensional surface embedded in E3 one can consider Riemannian metric

defined by first quadratic form: G = g11 du2 +2g12 dudv+g22 dv 2 , where g11 = (ru , ru ), g12 =
(ru , rv ), g22 = (rv , rv ), Levi-Civita connection ∇ with Christopher symbols defined by
(270) and curvature defined by (280). Emphasize again that in the case of two-dimensional
surface there exist only one non-trivial component of riemannian tensor.
Now we formulate

Theorema Egregium Let C be two-dimensional surface in E3 . Let G be Riemannian

metric on this surface defined by the first quadratic form. Let R be scalar curvature defined
by the Riemann curvature tensor and K Gaussian curvature. Then
R R1212
= 2 =K (282)
2 g11 g22 − g12
In the right hand side of this formula stands Gaussian curvature. It is defined by
the way of how surface is embedded in E3 . External observer calculates second and first
quadratic form and obtains Gaussian curvature (See the Section 2). In the left hand side
of this formula stands Riemannian scalar curvature. It is defined by the metric on the
surface. Internal observer, aunt on the surface, takes the metric 22 and calculates curvature
without any knowledge of the second quadratic form. The Theorem states that the two
answers will coincide
In particular if we bend the surfaces (i.e. transform it without changing the metric,
then r.h.s. will be the same. Hence Gaussian curvature will be the same.
The Theorem above explains why it is not possible to consider on the sphere coordi-
nates u, v such that g = du2 + dv 2 , i.e. it is not possible to bend the plane list to the
sphere. Indeed suppose there exist u, v : θ = θ(u, v), ϕ = ϕ(u, v) such that

R2 dθ2 + R2 sin2 θdϕ2 = du2 + dv 2 (283)

Then scalar curvature R is equal to zero. But R.H.S. of (282) is Gaussian curvature of
the sphere. It is equal to R12 . Contradiction.
This formula in particlularly enables to calculate Riemannian curvature for two-
dimensional surfaces in E3 . E.g. for sphere K = R12 , det g = R4 sin2 θ. Hence R1212 =
K det g = R2 sin2 θ. On the other hand the same answer follows from the straightforward
calculations of R1212 via formulae (269) and (280).

7.8 A Tale on Differential Geometry

Once upon a time there was an ant living on a sphere of radius R. One day he asked
himself some questions: What is the structure of the Universe (surface) where he lives? Is
22
which of course is defined by the external observer because metric is defined by the
first quadratic form. But external observer send information to aunt only about metric
not about second quadratic form!

86
it a sphere? Is it a torus? Or may be something more sophisticated, e.g. pretzel (a surface
with two holes)
Three-dimensional human beings do not need to be mathematicians to distinguish
between a sphere torus or pretzel. They just have to look on the surface. But the ant
living on two-dimensional surface cannot fly. He cannot look on the surface from outside.
How can he judge about what surface he lives on 23 ?
Our ant loved mathematics and in particular Differential Geometry. He liked to draw
various triangles, calculate their angles α, β, γ, area S(∆). He knew from geometry books
that the sum of the angles of a triangle equals π, but for triangles which he drew it was
not right!!!!
Finally he understood that the following formula is true: For every triangle
(α + β + γ − π)
=c (1)
S(∆)
A constant in the right hand side depended neither on size of triangle nor the triangles
location. After hard research he came to conclusion that its Universe can be considered as
a sphere embedded in three-dimensional Euclidean space and a constant c is related with
radius of this sphere by the relation
1
c= 2 (2)
R
...Centuries passed. Men have deformed the sphere of our old ant. They smashed it. It
seized to be round, but the ant civilisation survived. Moreover old books survived. New
ant mathematicians try to understand the structure of their Universe. They see that
formula (1) of the Ancient Ant mathematician is not true. For triangles at different places
the right hand side of the formula above is different. Why? If ants could fly and look on the
surface from the cosmos they could see how much the sphere has been damaged by humans
beings, how much it has been deformed, But the ants cannot fly. On the other hand they
adore mathematics and in particular Differential Geometry. One day considering for every
point very small triangles they introduce so called curvature for every point P as a limit
of right hand side of the formula (1) for small triangles:
(α + β + γ − π)
K(P ) = lim
S(∆)→0 S(∆)
Ants realise that curvature which can be calculated in every point gives a way to decide
where they live on sphere, torus, pretzel... They come to following formula 24 : integral
of curvature over the whole Universe (the sphere) has to equal 4π, for torus it must equal
0, for pretzel it equalts −4π...
Z
1
K(P )dP = 2 (1 − number of holes)
2π
23
This is not very far from reality: For us human beings it is impossible to have a global
look on three-dimensional manifold. We need to develop local methods to understand
global properties of our Universe. Differential Geometry allows to study global properties
of manifold with local tools.
24
In human civilisation this formula is called Gauß -Bonet formula. The right hand side
of this formula is called Euler characteristics of the surface.

87