Differential Geometry

James Emery

5/29/2015

Contents
1 Introduction 2

2 Curvature and Elementary Differential Geometry 2

3 Frame Vectors 6

4 The Serret-Frenet Formulas 6

5 The Reflection of a Ray From a Surface Tangent Plane 7

6 Curve Examples 8
6.1 The Circular Involute . . . . . . . . . . . . . . . . . . . . . . . 8

7 The Covariant Derivative in Rn 13

8 The Weingarten Map, The Shape Operator 14

9 Normal Curvature 16

10 Bilinear Forms 17

11 Quadratic Approximation 20

12 The Tangent Vector as a Derivation 20

13 Envelope Surfaces 23

1
14 The Covariant Derivative On a Surface 25

15 Christoffel Symbols 25

16 The Intrinsic Covariant Derivative 27

17 The Covariant Derivative in Riemmanian Geometry 28

18 The Bracket 30

19 Differential Manifolds 31

20 Riemannian Geometry 32

21 The Unique Riemannian Connexion Defined by the Riemannian

or semi-Riemannian Metric 33

22 Bibliography 33

1 Introduction
Differential Geometry is the study of curves and surfaces and their abstract
generalization: the differential manifold. More generally it is the study of
the calculus of curves and surfaces and involves definitions of curve tangents,
normals, and curvature. In higher dimensions we study surfaces and man-
ifolds and various kinds of curvature and so on. Some related subjects are
Vector Analysis, Algebraic Geometry, Tensor Analysis, Differential Topology,
Metrics and metric spaces, Riemanian Geometry, Differential Manifolds, and
General Manifold Theory, and so on. Areas of physics that use differential
geometry include General Relativity Theory, various subjects in Theoretical
Physics, Mechanics, Particle theory, and String Theory.

2 Curvature and Elementary Differential Geom-

etry
Curvature is the ratio of the change in turning to the distance traveled.
Consider the circular path. As a point on the circle moves through an angle

2
change , it moves a distance s = r. The ratio is a measure of the
curvature
1
= = .
s r r
The angle change of the tangent is here equal to the angle change , so
we can use the tangent angle in our definition of the curvature. So suppose
we are given a general curve in the plane

r = x(t)i + y(t)j.

Suppose dx/dt 6= 0, then the angle of the tangent is

" #
dy/dt
= tan1 .
dx/dt

Let us write
dx
= x,
dt
dy
= y,
dt
d2 x
= x,
dt2
d2 y
= y.
dt2
Then we have
d 1 y x yx
=
dt 1 + (y/x)2 x2
y x yx
= .
x2 + y 2
We have
ds q q
= (dx/dt)2 + (dy/dt)2 = x2 + y 2
dt
So the curvature is
d d/dt
= =
ds ds/dt
y x yx 1
= 2 2
2 .
x + y x + y 2

3
 y x yx
= .
(x2 + y 2 )3/2
We excluded the case where dx/dt = 0. However, we can just as well
define the tangent angle as
" #
1 dx/dt
= cot .
dy/dt
If we carry out a similure derivation to the one above we shall find that we
get the same formula for the curvature. Hence as long as at least one of
dx/dt or dy/dt is not zero, the formula holds.
Suppose we have a function y = f (x). This is a curve with t = x as the
parameter. Then x = 1, x = 0, and the curvature formula reduces to
d y
= = ,
ds (1 + y 2)3/2
where
d2 y
y = ,
dx2
and
dy
y = .
dx
Notice that in this formula the curvature can be either positive or negative. In
the case of a 1-dimensional function the curvature determines if the function
is concave up or down. In the more general definition in three space, the
curvature is always positive. So the new curvature will be defined to be the
absolute value of the 2d curvature defined above.
In three dimensional space there is no obvious tangent angle. So we must
define curvature using another approch. Let

r = x(t)i + y(t)j + z(t)k.

The velocity is
dr
v= = dx/dt(t)i + dy/dt(t)j + dz/dt(t)k.
dt
The magnitude of the velocity is
q ds
v= x2 + y 2 + z 2 = .
dt
4
The unit tangent vector T is defined as
v dr/dt dr
T= = = .
v ds/dt ds
T is a unit vector so
T T = 1.
We have
d(T T)
= dT/ds T + T dT/ds = 0,
ds
which implies that
2T dT/ds = 0.
So T and it derivative are orthogonal. Thus dT/ds is a vector normal to the
curve. The unit normal vector N is defined as
dT/ds
N= .
kdT/dsk
Thus
dT/ds = kdT/dskN.
We can define the curvature as

= kdT/dsk,

because we can show that for a two dimensional curve this agrees with the
two dimensional curvature.
Indeed, in two dimensions the unit tangent vector T can be written as

T = cos()i + sin()j,

where the tangent angle is a function of the arc length s. Then

dT/ds = sin()(d/ds)i + cos()(d/ds)j.

So q
kdT/dsk = |d/ds| sin2 () + cos2 () = |d/ds| = .
Notice that here the curvature is always nonnegative. We can therefore
write

dT/ds = N.

5
3 Frame Vectors
Let the curve R(s) in Euclidean 3-space be parameterized by arc length.
Define the tangent vector by
dR
T= ,
ds
and the curvature by  
 dT 
=   .
 
ds
The normal vector is defined to be
1 dT
N= ,
ds
and the binormal vector is defined by

B = T N.

The vectors T, N, and B, are called the frame vectors.

4 The Serret-Frenet Formulas

There are a set of formulas relating the frame vectors T, N, and B and their
derivatives.
The derivatives of the frame vectors with respect to arc length s are equal
to linear combinations of the frame vectors themselves.
These are called the Serret-Frenet formulas. The Serret-Frenet formulas
are derived from the facts that the frame vectors are mutually perpendicular,
and that they have unit length. The dot product of any pair of frame vectors
is zero. So the derivative of their dot product is also zero. Unit vectors are
perpendicular to their derivatives, and N is a unit vector. So dN/ds is per-
pendicular to N. Consequently dN/ds can be written as a linear combination
of T and B only. Thus
dN
= a1 T + a3 B.
ds
Because T is perpendicular to N,
dN dT
a1 = T = N .
ds ds

6
By definition the right hand expression is equal to . So we conclude that
a1 is equal to the curvature .
Define the torsion to be a3 . Thus
dN
= T + B.
ds
Let
dB
= b1 T + b2 N.
ds

dB dT
b1 = T = B = B N = 0.
ds ds
Then
dB dN
N = B = B (T + B) = .
ds ds
Thus
dB
= N.
ds
Therefore we have the Serret-Frenet transformation,

T0 0 0 T

N 0
=

0
N

B0 0 0 B

Notice that the matrix of the transformation is antisymmetric, and that the
top right element of the matrix is zero. An antisymmetric matrix has a zero
diagonal. The lower triangular part has negative elements. These facts might
aid one in remembering the formula.

5 The Reflection of a Ray From a Surface

Tangent Plane
Suppose the ray intersects the surface at a point p. Let n be a unit normal
to the tangent plane at point p. Then n is a second unit normal to the
plane. It does not matter which of these normals we choose. Let the ray be
represented by vector a. The component of a in the direction of n is

b = (n a)n.

7
 Notice that if we use the other normal we get the same vector b,

(n a)(n) = (n a)n = b.

The vector b is actually the projection of vector a to the line orthogonal to

the tangent plane.
We subtract this component from a, getting

c = a b,

which is the component of a orthogonal to n. So

a = b + c.

For the reflected ray, the component in the direction of n is the negative of the
original component, but the orthogonal component is in the same direction
as the original orthogonal component. Therefore the reflected ray is

d = b + c = c b.

That is
d = (a b) b = a 2b.
As a check we see that

n d = n a 2n b = n a.

6 Curve Examples
6.1 The Circular Involute
The involute of a curve C is a curve generated by unwinding a string wrapped
around C.
In the case of a circle of radius a centered at the origin, let t be the
angle at the string tangent point on the circle, as the string is unwrapped to
generate the involute. Then we find the parametric equation of the involute
to be

x = a(cos(t) + t sin(t))

y = a(sin(t) t cos(t)).

8
Figure 1: Two caustics (envelope curves of reflected rays) from the involute
curve. The two curves are formed from two sets of parallel rays, the first
making a zero angle with the horizontal, and the second making a twenty
degree angle with the horizontal. The smaller caustic comes from the zero
angle rays.

9
Figure 2: A caustic (envelope curve of reflected rays) from the involute curve.
This curve is produced from reflections from a set of parallel rays making an
angle of zero degrees with the horizontal axis. Notice that the cusp point
occurs at a height equal to the height of the circle.

10
Figure 3: Reflections of parallel rays from the right at zero degrees from an
involute curve.

11
 The position vector of the curve is
R(t) = xi + yj.
We have the derivatives
dx
x = = r( sin(t) + sin(t) + t cos(t)) = rt cos(t)
dt
dy
y = = r(cos(t) cos(t) + t sin(t)) = rt sin(t)
dt
d2 x
x = 2 = r(cos(t) t sin(t))
dt
d2 y
y = 2 = r(sin(t) + t cos(t)),
dt
and the derivative of the arclength
ds q 2
= x + y 2 = rt
dt
The tangent vector is
dR dR dt
T= =
ds dt ds
= cos(t)i + sin(t)j
The normal vector is
dT dT dt
N= =
ds dt ds
= sin(t)i + cos(t)j.
c from involute.ftn
c+ function fx definition
real*8 function fx(t)
implicit real*8(a-h,o-z)
a=1.
c=1.
fx=a*(cos(t) + c*t*sin(t))
return
end

c+ function fy definition
real*8 function fy(t)
implicit real*8(a-h,o-z)
a=1.
c=1.
fy=a*(sin(t) - c*t*cos(t))
return
end

12
7 The Covariant Derivative in Rn
Given any curve in Rn with P = (0) and
d
(0) = X,
dt
the covariant derivative of the vector field Y in the direction X at the point
P is
d
X Y (P ) = Y ()(0).
dt
The chain rule may be used to show that the definition does not depend on
the curve .
This definition may be applied to a two dimensional surface patch em-
bedded in R3 . Let be a surface patch mapping an open set in R2 to R3 .
Suppose is a curve in the domain of so that (t) = ((t)). Then
d
X Y (P ) = Y () =
dt
dY (((t)))
=
dt
Y ((+ )) Y (( ))
.
4t
This is a central difference approximation where
= (4t/2),
and
+ = (4t/2).
The tangent vector X is
d((t))
X=
dt
(+ ) ( )
.
4t
Richardson extrapolation can be applied to these central difference approx-
imations. An intrinsic covariant derivative may be defined on the patch,
which is derived from the covariant derivative in R3 . This is accomplished
by subtracting the normal component of the R3 covariant derivative (see
Hicks, p26).

13
8 The Weingarten Map, The Shape Operator
Let N be a unit normal vector field on the surface. The Weingarten map,
sometimes called the shape operator, is the mapping from the tangent space
at P to itself, defined by

W (X) = X N(P ),

where X is in the tangent space at P . The Weingarten map is linear and it is

symmetric. To prove that it is symmetric, let (u, v) be a coordinate patch.
Then u and v are linearly independent tangent vectors and so a basis for
the tangent space. We have

W (u ) v = u N() v

N()
= v
u
N()
=
u dv

= N() ( )
u dv

= N() ( )
v du
N()
=
v du
= W (v ) u .
We have used the fact that N is perpendicular to the tangent space and
hence to tangent vectors

,
v
and

.
u
We will present a method for determining W numerically when surface
patch values are available, but derivatives values are not. In addition we
suppose that the surface normal is available (although it could be computed
numerically from patch values).

14
 Let 1 be a curve in the uv coordinate space defined by

1 (t) = (u0 , v0 ) + tT1 ,

where T1 = (1, 0) is a tangent vector in the uv coordinate space. Let 1 (t)

be the curve (1 (t)). Let
d1 (0)
X1 = .
dt
Similarly define
d2 (0)
X2 = ,
dt
where the uv tangent vector is T2 = (0, 1). If has nonzero Jacobian then
X1 and X2 are independent vectors in the tangent space. X1 and X2 are not
necessarily perpendicular to one another.
Let Z1 and Z2 be orthogonal vectors in the tangent space, which means
that Z1 and Z2 are in R3 , and are perpendicular to N at P . Z1 and Z2 are
a basis of the tangent space, so there exists bij so that

X1 = b11 Z1 + b12 Z2

and
X2 = b21 Z1 + b22 Z2 .
We have
Xi Z j
bij = .
kZi k2
Then using the linearity of W we get the vector equations

W (X1 ) = b11 W (Z1 ) + b12 W (Z2)

and
W (X2 ) = b21 W (Z1 ) + b22 W (Z2 ).
We may solve these equations for the x, y and z components of W (Z1) and
W (Z2 ). Then let
W (Z1 ) = c11 Z1 + c12 Z2
and
W (Z2) = c21 Z1 + c22 Z2 .
Z1 and Z2 are perpendicular, so we have

15
 Zi W (Zj )
cij = .
kZi k2
The matrix representation of the Weingarten transformation in this basis is.
" #
c11 c21
c12 c22

9 Normal Curvature
Define the normal curvature of a surface in the direction of a curve , at a
point p, by
n = W (0 ) 0 ,
where a(0) = p, and 0 = 0 (0). This turns out to be the component of the
curvature of the curve in the direction of the surface normal, when is
parameterized by arc length. Let N be the surface normal restricted to .
Then N is a curve, and
dN()
W (0 ) = 0 N = = N 0.
ds
We have
0 N = 0.
So
00 N = 0 N 0 .
But
d0 dt
00 = = = n.
ds ds
So
n = N 0 0 = 00 N = n N = cos(),
where is the angle between the surface normal N and the curve normal n.
From the definition one sees that the normal curvature depends only upon
the direction of the unit vector t = 0 . Hence notice that given two curves
that pass through a given point p and have the same direction at that point
have the same surface normal component of their curvature vector. Also if
the curve lies in a plane which contains the surface normal, then the curve

16
normal coincides with the surface normal so the normal curvature is the same
as the curvature of the curve. If is an eigenvalue of W with eigenvector t
then the normal curvature is

n = W (t) t = t t = .

The direction of t is called a principal direction. W is a symmetric trans-

formation so the eigenvalues are real and the eigenvectors are orthogonal.
The principal directions are well defined except at an umbilic point where
the eigenvalues are equal and every vector is an eigenvector. The normal
curvature is a quadratic form and the maximum and minimum values of the
form occur at the eigenvalues of W . Thus the principal curvatures are the
maximum and the minimum normal curvatures.

10 Bilinear Forms
The classical fundamental bilinear forms defined on pairs of tangent vectors
include the first fundamental form

B1 (T1 , T2 ) = T1 T2 ,

and the second fundamental form

B2 (T1 , T2 ) = W (T1 ) T2 .

The quadratic form derived from the first fundamental form is

Q1 (T ) = T T.

This is the metric form.

In classical differential geometry a tangent vector is written as
n
X X
T = ds = dui.
i=1 ui

Then the first fundamental quadratic form would be written as

n X
n
gij duiduj .
X
Q1 (ds) = ds ds =
i=1 j=1

17
g is called the metric tensor. Suppose X(u, v) is a coordinate patch, then we
have
ds2 = Q1 (ds) = Xu Xu du2 + 2Xu Xv dudv + Xv Xv dv 2 .
= Edu2 + 2F dudv + Gdv 2.
As an example, consider a spherical patch of radius a,

X(u, v) = (a cos(v) sin(u), a sin(v) sin(u), a cos(u)).

Then
Xu (u, v) = (a cos(v) cos(u), a sin(v) cos(u), a sin(u)).
and
Xv (u, v) = (a sin(v) sin(u), a cos(v) sin(u), 0).
Thus
E = Xu Xu = a2
F = Xu Xv = 0
and
G = Xv Xv = a2 sin2 (u).
The second fundamental quadratic form is

Q2 (ds) ds = W (Xu ) Xu du2 + 2W (Xu ) Xv dudv + W (Xv ) Xv dv 2 .

= edu2 + 2f dudv + gdv 2.

Let N be a unit normal. Then

0 = Xu N = Xv N,

so
e = W (Xu ) Xu = Xu Nu = Xuu N,
f = W (Xv ) Xu = Xu Nv = Xuv N,
and
g = W (Xv ) Xv = Xv Nv = Xvv N.
The normal curvature in the direction of T is
W (T ) T Q2 (T )
n = W (T /kT k) T /kT k = = .
T T Q1 (T )

18
Proposition. The Gaussian curvature is given by

eg f 2
K(X) = ,
EG F 2
and the mean curvature by
Ge Eg 2F f
H(X) = ,
2(EG F 2 )

Proof. Let T1 and T2 be linearly independent tangent vectors. Let a be the

coefficient matrix of W with respect to this basis. Write

W (Ti ) = aji Tj

Then
W (T1 ) W (T2 ) = Det(a)T1 T2 = KT1 T2 .
and

W (T1 ) T2 + T1 W (T2 ) = 2Trace(a)T1 T2 = 2HT1 T2 .

These equations can be solved for the Gaussian curvature K and the mean
curvature H by dotting each equation by

T1 T2 .

We have
(W1 W2 ) (T1 T2 ) = K(T1 T2 ) (T1 T2 ).
The left side is

(W1 W2 ) (T1 T2 ) = W1 (W2 (T1 T2 ))

= W1 (T1 (W2 T2 ) T2 (W2 T1 ))

= (W1 T1 )(W2 T2 ) (W1 T2 )(W2 T1 )
= (W1 T1 )(W2 T2 ) (W1 T2 )2 .
We have used the BAC CAB rule. Similarly

(T1 T2 ) (T1 T2 ) = (T1 T1 )(T2 T2 ) (T1 T2 )2 .

19
Thus
(W1 T1 )(W2 T2 ) (W1 T2 )2
K=
(T1 T1 )(T2 T2 ) (T1 T2 )2
If
T1 = Xu
and
T2 = Xv ,
then
eg f 2
K(X) = .
EG F 2
The mean curvature equation is proved similarly.

11 Quadratic Approximation
Let p be a point on the surface. We may assume a coordinate system in
which p is the origin and the normal is in the z direction. Thus we may write
the surface as
z = f (x, y).
We expand f in a Taylor series. Keeping only second order terms we obtain
a quadratic approximation to the surface. Now we could fit local data to
a quadric of this form using least squares and then obtain approximations
to the principal curvatures. However, the least squares quadratic goes to
the quadratic approximation only as the data approaches p. In effect we are
approximating second derivatives, while the previous technique requires only
covariant derivatives of the first order.

12 The Tangent Vector as a Derivation

Let c be a curve with parameter t. A tangent to the curve is v = dc/dt. Let
f be a function. We define

df (c(t)) dc
Dv (f ) = = f = f v.
dt dt
This defines a functional Dv mapping functions to real numbers. It depends
only on the tangent vector v, not specifically on the curve c. It does not

20
depend on the coordinate system because

df (c(t))
dt
is coordinate independent. From the theory of differential equations, given a
tangent vector v at a point p, there exists a curve c so that

dc(0)
= v,
dt
and c(0) = p. Thus for every tangent vector there is a functional Dv . The
functional has the obvious properties

Dv (f + g) = Dv (f ) + Dv (g)

and
Dv (ag) = aDv (f ),
where a is a scalar. Dv is a linear functional. We have
df (c)g(c) dg(c) df (c)
Dv (f g) = = f (c) + g(c) =
dt dt dt
f (c)Dv (g) + Dv (f )g(c).
Functionals satisfying these properties are called derivations. Consider the
example of the x coordinate curve in Cartesian three space. We have c(t) =
tux where ux is the unit coordinate vector in the x direction. Then
dc
v= = ux
dt
and
f
Dv (f ) = f ux = .
x
We have shown that

Dv = .
x
Similarly if i is a local surface coordinate, then

,
i

21
is the derivation corresponding to its tangent vector. We have

(i ) = ji .
j
Hence

A={ : 1 i n}
i
is a linearly independent set of derivations. We shall show that tangent
vectors and derivations can be identified. So A is a basis of the tangent
space. Let v1 and v2 be tangent vectors ,then

Dv1 +v2 f = f (v1 + v2 ) =

f v1 + f v2 = Dv1 f + Dv2 f
If a is a scalar, then
Dav1 = aDv1 .
Therefore the mapping
v 7 Dv
is a homomorphism. Since the tangent space of an n-dimensional manifold
is n-dimensional, and there are tangent vectors mapping to A, the mapping
v 7 Dv is actually an isomorphism.
Now we no longer have to rely on a coordinate system in the embedding
space, in which the surface lies, to define a tangent vector. Given any curve
in the surface c, and a parameter point t0 , there is a derivation defined by
df (c)
Dc f = (t0 ).
dt
This is independent of the embedding. A local surface coordinate system can
be used to find the components of Dc with respect to the local tangent basis.
The covariant derivative X Y can be given as the action of the derivation
X on the vector field Y . Indeed let

X = x1 + x2 + x3
u1 u2 u3
and

Y = y1 + y2 + y3
u1 u2 u3

22
be tangent vectors in R3 and let X act as a derivation. Then

X Y = (X(y1 ), X(y2), X(y3)).

To prove this let c be a curve so that c0 = X. Then

Y (c(t))
X Y = =
dt
y1 (c(t)) y2 (c(t)) y3 (c(t))
( , , ) = (X(y1 ), X(y2 ), X(y3)).
dt dt dt
Reference Abraham Goetz, Introduction to Differential Geometry, Addison-
Wesley 1970.

13 Envelope Surfaces
Given a one parameter family of surfaces

f (x, y, z, t) = 0,

the envelope surface is defined by eliminating t from the equations

f (x, y, z, t) = 0

and
f (x, y, z, t)
=0
t
(see Goetz p125). To establish this, let the envelope surface have parametric
equation
r(u1 , u2 ) = (x(u1 , u2 ), y(u1, u2 ), z(u1 , u2 )).
Given the coordinates of a point u1 u2 of a point p on the envelope, there is
an element of the family of surfaces that is tangent to the envelope at this
point. Let the corresponding parameter of this tangent surface be t(u1 , u2 ).
Now f is normal to this tangent surface. Hence
dr
f = 0.
du1
But let

23
 g(u1, u2 ) = f ((x(u1 , u2 ), y(u1, u2 ), z(u1 , u2 ), t(u1 , u2)).
Then g = 0, so
dg dr f (x, y, z, t) dt
0= = f + .
du1 du1 t du1
Thus
f (x, y, z, t) dt
= 0.
t du1
Similarly
f (x, y, z, t) dt
= 0.
t du2
We shall assume that
dt
du1
and
dt
du2
do not vanish simultaneously. Thus
f (x, y, z, t)
= 0.
t
A canal surface is the envelope of a family of spheres. For example the
torus is a canal surface, and Dupin cyclides are also canal surfaces. A spheri-
cal Milling cutter produces a canal surface, other cutter shapes produce other
envelope surfaces. A rather general milling cutter may be considered the en-
velope of a family of flat disk like ellipsoids. Then a milled surface may be
considered the envelope of a two parameter family of surfaces
f (x, y, x, s, t) = 0.
The envelope surface is found by eliminating t and s from
f (x, y, z, s, t) = 0, ft (x, y, z, s, t) = 0, fs (x, y, z, s, t) = 0.
See Goetz.
The milled internal screw thread is an envelope of surfaces generated by
swept disks. It can also be considered a the envelope of a two parameter
family of algebraic surfaces and is thus algebraic.
An envelope of a one-parameter set of planes is called a developable sur-
face.

24
14 The Covariant Derivative On a Surface
Define the covariant derivative on the surface by

DX Y = X Y + < X N, Y > N.

We will show that this is the projection of the R3 covariant derivative to the
tangent space. For if we take the dot product with the normal N, we get

DX Y N =< X Y, N > + < X N, Y >= X < Y, N >= 0.

X, as a derivation, is simply a differential operator. Thus the expression on

the right is a differential operator operating on an inner product, and Y is
orthogonal to N, so it is zero. Then we have

< X N, Y >= < X , N > .

Then we have
DX Y = X Y < X Y, N > N,
which shows that DX Y is the projection of X Y to the tangent space.
The covariant derivative has an explicit coordinate representation, given
in terms of the soon to be introduced Christoffel symbols.

15 Christoffel Symbols
Let (u1, u2 ) be a surface patch (an inverse coordinate system). Write
2
ij = .
ui uj
We may express these derivatives in the moving basis

{1 , 2 , N}

We have
ij = kij k + ij N.
The kij are called the Christoffel symbols of the second kind. Taking inner
products we find that the ij are the coefficients bij of the second fundamental
form. Thus
ij =< ij , n >= bij , .

25
where
b11 = E, b12 = b21 = F, b22 = G.
Let us write the derivative of the normal as

ni = bki k .

Then we get
< ni , j >= bki gkj .
By differentiating
< n, j >= 0,
we get
< ni , j >= < n, ij >= bij
Thus
bij = bki gkj .
Then
bij g kj = bki gkj g kj = bki kl = bli ,
where
{g ij } = {gij }1 .
The Christoffel symbols are symmetric. In fact ij and bij are symmetric,
so
kij k = ij bij n = ji bji n = kji k .
Then
kij = kji,
because the i are linearly independent.
The Christoffel symbols of the first kind are defined by

ijl =< ij , l >=< kij k + bij n, l >= kij < k , l >= kij gkl .

They are symmetric in the first two indices because of the symmetry of ij .
It follows that
kij = ilj g lk .
Theorem
1 gjk gki gij
ijk = ( + ),
2 ui uj uk

26
and
g kl gjl gli gij
kij = ( + ).
2 ui uj ul
Proof. We have
gij < i , j >
k
= =< ik , j > + < i , jk >=
u uk
ikj + jki.
The permutation
i k, j i, k j
gives
gki
= kji + ijk = jki + ijk ,
uk
and the permutation
i j, j k, k i
gives
gjk
= jik + kij = ijk + ikj .
ui
Adding the last two equations and subtracting the first, we obtain the result
which was to be proved.

16 The Intrinsic Covariant Derivative

Let (u1 , u2) be a coordinate patch. Let X be the tangent to curve c(t) =
(u1 (t), u2(t)). Let
Y = y 1 1 + y 2 2 = y i i
Then the covariant derivative is
dY (c) dy i i duj
X Y = = i + y ij
dt dt dt
We may substitute the expressions involving the Christoffel symbols for the
second derivatives, obtaining a linear combination of the basis vectors

{1 , 2 , N}.

27
Taking the projection into the tangent space, (deleting terms involving N),
we get an expression for the surface covariant derivative

dy i dy j
DX Y = i + yi kij k =
dt dt
dy k i k duj
k + y ij k =
dt dt
dy k duj
( + y ikij )k
dt dt
which involves the basis vectors

{1 , 2 },

and the Christoffel symbols. The Christoffel symbols involve only the coef-
ficients of the first fundamental form and their derivatives. So the surface
covariant derivative is intrinsic, and depends only on the metric gij .
Let X = m and Y = n . Then

dy k
=0
dt
and
y i = ni ,
and
duj j
= m .
dt
Then
Dm n = kmn k .

17 The Covariant Derivative in Riemmanian

Geometry
In Hicks, the covariant derivative of vector field Y with respect to vector field
X is written as
DX Y.

28
We shall continue to write it as

X Y.

Above we introduced the derivation corresponding to vector v. A deriva-

tion D is a linear functional on a manifold that has the properties:

(1)D(f + g) = D(f ) + D(g)

(2)D(af ) = aD(f ),
(3)D(f g) = f (c)D(g) + D(f )g(c).
The derivations play the roll of tangent vectors. The set of derivations con-
stitute the tangent space. A derivation may also be defined as a parametric
derivative along a curve.
In the Euclidean case suppose we have a vector

X = (a1 , a2 , ...., an )

Then a directional derivative in the direction of this vector is a derivation

and the operation on a function f is given by
n
X f
Xf = ai .
i=1 ui

In this case it is easy to see that property (3) of the derivation follows from
the rule for diferentiating the product of functions. Further in this Euclidean
case one can show that every derivation is of this form. If Y is a vector field
in this Euclidean space with ith coordinate yi, then a covarient derivative is
given by
X Y = (Xy1, ...., Xyn ).
Covariant derivatives have the following four characteristic properties:

(1)X (Y + Z) = X Y + X Z

(2)X+W Y = X Y + W Z
(3)f (p)X Y = f (p)X Y
(4)X f (p)Y = (Xf )Y + f (p)X Y
Property (4) follows from the property (3) of derivations.

29
 Given a curve and a vector field Y defined along the curve, if T is the
curve tangent and
T Y = 0,
then the vector Y (0) is said to be parallel translated along .
If
T T = 0,
then the curve tangent is parallel translated along the curve and the
curve is a geodesic, a shortest path connecting the starting point and the
ending point.
Consider the case of translating a tangent vector around a path on a
sphere, say the earth. Say one starts at the north pole and suppose we
translate a vector along the zero degree meridian. Let the vector Y be per-
pendicular to the meridian as it is translated, that is it is always pointing
east, always making the same angle with the meridian, i.e. the geodesic
curve.
When we reach the equator the vector points east in the direction of the
equator, which is another geodesic. We keep it in the direction of the equator
until we get to say the 30thth degree meridian. The vector makes a 90 degree
angle with the 30th degree meridian as we translate it back up to the north
pole. When we reach the north pole we find that the translated vector is
pointing in a different direction from the starting direction. So on the earth
we have traveled along a spherical triagle path, always keeping our vector
pointing in the same direction with respect to what we consider locally as
straight lines. But when we return to our starting point we see that the
vector has been turned. So this tells us that our surface is curved. Because
on a flat surface this would not occur.
The geodesics are the paths of particles when they do not experience any
force. Their motion is the analogue of classical uniform motion.

18 The Bracket
If X and Y are C vector fields, their bracket is a C vector field (Hicks
P8) defined by
[X, Y ]f = X(Y f ) Y (Xf ).
The bracket has the following properties:
[X, Y ] = [Y, X].

30
 [X, X] = 0.
[X1 + X2 , Y ] = [X1 , Y ] + [X2 , Y ]
and
[f X, gY ] = f (Xg)Y g(Y f )X + f g[X, Y ].
At a fixed point p on the manifold,

[X, Y ]p f

is a functional, mapping f to a real number. For a fixed p and a fixed f ,

[X, Y ]p f

maps two vectors from the product tangent space to a real number. If it were
multilinear with respect to X and Y , then it would be a second rank tensor.
But property

[f X, gY ] = f (Xg)Y g(Y f )X + f g[X, Y ],

shows that this is not the case.

The bracket satisfies the Jacoby identity

[X, [X, Z]] + [Z, [X, Y ]] + [Y, [Z, X]] = 0.

The bracket comes from the Poisson bracket of mechanics, and was intro-
duced by Sophus Lie in his work on Lie Groups (originally called continuous
groups of transformations) and Lie Algebras in solving partial differential
equations.

19 Differential Manifolds
A differential manifold is a topological space M and a set of charts, where
each chart is a one to one mapping from an open neighborhood of M to
an open connected neighborhood in Rn . If 1 and 2 are overlapping charts,
then 1
1 2 is a C

map. M is covered by the domains of the set of charts.
The charts specify a set of local coordinates for the manifold. For example
on the spherical earth, the mapping of a point on the earth to its latitude
and longitude is a chart. It takes more than one chart to cover the earth.
Let us take the example of a 2d unit sphere embedded in R3 . The use of

31
spherical coordinates constitutes a chart. That is a point on the sphere is
mapped to the , coordinates. The inverse of this chart is the mapping

x = sin() cos()

y = sin() sin()
z = cos()
Tangent vectors in these coordinate directions are the functionals (deriva-
tions)

, and .

20 Riemannian Geometry
A Riemannian Space is a differential manifold with a Riemannian metric. A
Riemannian Metric is a function < X, Y >, where X and Y are elements of
a tangent space. This is the distance between X and Y . In classical notation
this is written as g(X,Y), and has tensor components gij . The Riemannian
metric is real valued, bilinear, symmetric, and positive definite. The metric is
a quadratic form and is represented by a symmetric matrix. Positive definite
means that if X is not zero, then g(X, X) > 0. The eigenvalues of the matrix
are all positive.
A Semi-Riemannian metric is real valued, bilinear, and symmetric. But
is not necessarily positive definite, but must be nonsingular. The eigenval-
ues must be nonzero, but not necessarily positive. This is the case for the
Minkowski metric of relativity theory. A positive definite matrix is clearly
nonsinglular, so a Riemannian metric is a Semi-Riemannian metric. Clearly,
there can be a nonsingular matrix that is not positive definite, namely a diag-
onal matrix. A Riemannian metric tensor is given classically as a symmetric
quadratic form
n X
n
ds2 =
X
gij dxi dxj .
i=1 j=1

where ds is the infinitesimal distance.

32
21 The Unique Riemannian Connexion De-
fined by the Riemannian or semi-Riemannian
Metric
The properties 1-6 of the standard connexion on Rn (Hicks pp19-20) are
satisfied by this Riemannian connection properties 5 and 6 are additional
properties not necessarily satisfied by a covariant derivative or connexion
that is not Riemannian.
There exists a unique Riemannian connexion (pp 69-70) on a Riemannian
or semi-Riemannian manifold. The Christoffel symbols are given in terms of
the metric tensor. We have by definition
n
ijk Xi .
X
Xk (Xj ) =
i=1

From the properties of the covariant derivative or connexion we have the

Christoffel symbols in terms of the metric
1 X 1 grj gri gij
kij = (g )kr ( + ).
2 r xi xj xr

The symbol g 1 is the inverse of the symmetric matrix of the metric quadratic
form. This is where the nonsingularity of the semi-Riemannian metric is
needed. This defines the connexion, because it has been defined for the basis
vector fields Xi .

22 Bibliography
[1] Auslander Louis, MacKenzie Robert E, Introduction to Differential
Manifolds, 1963, 1977 Dover.

[2] Auslander Louis, Differential Geometry, Harper and Roe, 1967.

[3] Belinfante Johan G. F., Kolman Bernard, A Survey of Lie Groups

and Lie Algebras with Applications and Computational Methods,
1972, SIAM.

33
[5] Bishop Richard L, Goldberg Samuel L, Tensor Analysis on Manifolds,
1968.

[6] Bradbury T. C., Theoretical Mechanics, sections 2.15 to 2.21, John

Wiley, 1968.

[7] Choquet-Bruhat Yvonne, Dewitt-Morette Cecile, Dillard-Bleick Margaret,

Analysis, Manifolds, and Physics, North-Holland, 1982.

[8] Eisenhart Luther Pfahler, A Treatise on the Differential Geometry

of Curves and Surfaces, 1909, 1960 Dover.

[9] Eisenhart Luther Pfahler, Continuous Groups of Transformations,

1933, 1961 Dover.

[10] Eisenhart Luther Pfahler, Riemannian Geometry, 1925, 5th printing

1964, Princeton.

[11] Emery J. D., Physics, JDEU Press, 1996, (See the chapter on Mechan-
ics.)

[12] Emery J. D., Tensor Analysis in Euclidean Space, 1996, Revised

2007.

[13] Flanders Harvey, Differential Forms With Applications to the

Physical Sciences, 1963,1989 Dover.

[14] Goetz Abraham, Introduction to Differential Geometry, Addison-

Wesley, 1970.

[15] Goldstein, Herbert Classical Mechanics,Addison-Wesley, 1965.

[16] Graustein William C, Differential Geometry, 1935,1962, Dover.

[17] Greub Werner H. Multilinear Algebra , Springer (January 1967).

[18] Guggenheimer Heinrich W, Differential Geometry, 1963,1977 Dover.

34
[19] Helgason Sigurdur, Differential Geometry, Lie Groups, and Sym-
metric Spaces, Academic Press, 1978.

[20] Hicks Noel J, Notes on Differentia Geometry, Van Nostrand, 1971.

[21] Kreyszig Erwin, Differential Geometry, University of Toronto Press,

1959.

[22] Lange Serge, Fundamentals of Differential Geometry, Springer-

Verlag, 1999.

[23] Lass Harry, Vector and tensor Analysis, Mcgraw-Hill, 1950.

[25] Lovelock David, Rund Hanno, Tensors, Differential Forms and vari-
ational Principles, 1977,1989 Dover.

[26] Marion Marion B, Classical Dynamics of Particles and Systems,

Academic Press, 1970.

[27] ONeil Barrett, Semi-Riemannian Geometry With Applications

to Relativity, Academic Press, 1983.

[28] ONeill Barrett, Elementary Differential Geometry, Academic Press,

1966.

[29] Schutz Bernard, Geometrical Methods of Mathematical Physics,

Cambridge 1980.

[30] Sokolnikoff I. S., Tensor Analysis, Theory and Applications, John

Wiley 1951.

[31] Sorani Giuliano , An Introduction to Real and Complex Mani-

folds, Gordon and Breach, 1969.

[32] Spivak Michael, A Comprehensive Introduction to Differential

Geometry, Volumes I and II, 2nd Edition, Publish or Perish, 1979.

35
[33] Spivak Michael, Calculus on Manifolds, W A Benjamin 1965.

[34] Steenrod Norman, The Topology of Fibre Bundles, Princeton 1951.

[35] Struik Dirk J, Lectures on Classical Differential geometry, 1950,

1961 Dover.

36