Differential Geometry

Lia Vas

The Second Fundamental Form. Geodesics. The Curvature Tensor. The

Fundamental Theorem of Surfaces. Manifolds

The Second Fundamental Form and the Christoffel symbols. Consider a surface x =
x(u, v). Following the reasoning that x1 and x2 denote the derivatives ∂x
∂u
and ∂x
∂v
respectively, we
denote the second derivatives
∂2x ∂2x ∂2x ∂2x
∂u2
by x11 , ∂v∂u
by x12 , ∂u∂v
by x21 , and ∂v 2
by x22 .

The terms xij , i, j = 1, 2 can be represented

as a linear combination of tangential and normal
component. Each of the vectors xij can be repre-
sented as a combination of the tangent component
(which itself is a combination of vectors x1 and
x2 ) and the normal component (which is a multi-
ple of the unit normal vector n). Let Γ1ij and Γ2ij
denote the coefficients of the tangent component
and Lij denote the coefficient with n of vector xij .
Thus,

xij = Γ1ij x1 + Γ2ij x2 + Lij n = Γkij xk + Lij n.

P
k

The formula above is called the Gauss formula.

The coefficients Γkij where i, j, k = 1, 2 are called Christoffel symbols and the coefficients Lij ,
i, j = 1, 2 are called the coefficients of the second fundamental form.
Einstein notation and tensors. The term “Einstein notation” refers to the certain summation
convention that appears often in differential geometry and its many applications. Consider a formula
can be written in terms of a sum over an P index that appears in subscript of one and superscript of
the other variable. For example, xij = k Γkij xk + Lij n. In cases like this the summation symbol is
omitted. Thus, the Gauss formula for xij in Einstein notation is written simply as

xij = Γkij xk + Lij n.

An important benefit of the use of Einstein notation can be seen when considering n-dimensional
manifolds – all the formulas we consider for surfaces generalize to formulas for n-dimensional man-
ifolds. For example, the formula xij = Γkij xk + Lij n remains true except that the indices i, j take
integer values ranging from 1 to n not just values 1 and 2.
If we consider the scalar components in certain formulas as arrays of scalar functions, we arrive
to the concept of a
tensor.

1
 For example, a 2 × 2 matrix with entries gij is
considered to be a tensor of rank 2. This matrix
is referred to as the metric tensor. The scalar
functions Γkij are considered to be the components
of a tensor Γ of rank 3 (or type (2,1)). The Ein-
stein notation is crucial for simplification of some
complicated tensors.
Another good example of the use of Einstein
notation is the matrix multiplication (students
who did not take Linear Algebra can skip this ex-
ample and the next several paragraphs that relate
to matrices). If A is an n × m matrix and v is a
m × 1 (column) vector, the product Av will be a n × 1 column vector. If we denote the elements of
A by aij where i = 1, . . . , n, j = 1, . . . P
, m and xj denote the entries of vector x, then the entries of
the product Ax are given by the sum j aij xj that can be denoted by aij xj using Einstein notation.
Note also that the entries of a column vector are denoted with indices in superscript and the
entries of row vectors with indices in subscript. This convention agrees with the fact that the entries
of the column vector aij xj depend just on the superscript i.
Another useful and frequently considered tensor is the Kronecker delta symbol. Recall that the
identity matrix I is a matrix with the ij-th entry 1 if i = j and 0 otherwise. Denote these entries by
δji . Thus,

i 1 i=j
δj =
0 i 6= j
In this notation, the equation Ix = x can be written as δji xj = xi .
Recall that the inverse of a matrix A is the matrix A−1 with the property that the products AA−1
and A−1 A are both equal to the identity matrix I. If aij denote the elements of the matrix A, aij
denote the elements of the inverse matrix Aij , the ij-th element of the product A−1 A in Einstein
notation is given by aik akj . Thus aik akj = δji .
In particular, let g ij denote the entries of the inverse matrix of [gij ] whose entries are the coeffi-
cients of the first fundamental form. The fact that the matrix and its inverse multiply to the identity
gives us the following formulas (all given in Einstein notation).

gik g kj = δij and g ik gkj = δji .

The coefficients g ij of the inverse matrix are given by the formulas

g22 −g12 g11
g 11 = , g 12 = , and g 22 =
g g g
where g is the determinant of the matrix [gij ].
Computing the second fundamental form and the Christoffel symbols. The formula
computing the Christoffel symbols can be obtained by multiplying the equation xij = Γkij xk + Lij n
by xl where k = 1, 2. Since n · xl = 0, and xk · xl = gkl , we obtain that

xij · xl = Γkij gkl

2
 To solve for Γkij , we have to get rid of the terms gkl from the left side. This can be done by using
the inverse matrix g ls .
(xij · xl )g ls = Γkij gkl g ls = Γkij δks = Γsij
Thus, we obtain that the Christoffel symbols can be computed by the formula

Γkij = (xij · xl )g lk .

To compute the coefficients of the second fundamental form, multiply the equation xij = Γkij xk +
Lij n by n. Since xl · n = 0, we have that xij · n = Lij n · n = Lij . Thus,

x1 ×x2
Lij = xij · n = xij · |x1 ×x2 |
.

While the first fundamental form determines the intrinsic geometry of the surface, the second fun-
damental form reflects the way how the surface embeds in the surrounding space and how it curves
relative to that space. Thus, the second fundamental form reflects the extrinsic geometry of the
surface. The presence of the normal vector n in the formula for Lij reflects this fact also since n
“sticks out” of the surface. In contrast, we shall see that the Christoffel symbols can be computed
using the first fundamental form only which shows that they are completely intrinsic.
Example 1. The Christoffel symbols and the second fundamental form of a plane.
Consider the xy-plane z = 0 for simplicity. Thus x(x, y) = (x, y, 0). Recall (or compute again) that
x1 = xx = (1, 0, 0), x2 = xy = (0, 1, 0), g11 = g22 = 1, and g12 = 0.
Since the first derivatives x1 and x2 are constant, the second derivatives x11 , x12 = x21 , x22 are
all zero. As a result, the second fundamental form Lij are zero.

Lij = xij · n = (0, 0, 0) · n = 0

 
0 0
Represented as a matrix, the second fundamental form is . This means that the surface does
0 0
not deviate from its tangent plane at all and this should not be surprising since the tangent plane of
a plane is that plane itself.
The Christoffel symbols are also zero.

Γkij = (xij · xl )g lk = ((0, 0, 0) · xl )g lk = 0.

We shall see later that this means that surfaces with this property are such that their shortest-distance
curves (geodesics) are really lines.
Example 2. The Christoffel symbols and the second fundamental form of a cylinder.
Consider the cylinder (a cos t, a sin t, z) with a circular base of radius a. Recall (or compute again)
that x1 = (−a sin t, a cos t, 0), x2 = (0, 0, 1) and so g11 = a2 , g12 = 0 and g22 = 1. Thus, g = a2 and so
2
g 11 = gg22 = a12 , g 12 = −gg12 = 0 and g 22 = gg11 = aa2 = 1 so the matrix inverse to the first fundamental
 1 
a 2 0
form is . The unit normal vector is n = (cos t, sin t, 0).
0 1

3
 The second derivatives are x11 = (−a cos t, −a sin t, 0), x12 = (0, 0, 0), and x22 = (0, 0, 0). Thus,
2 2
L11 = x11 · n = −a cos t − a sin t = −a, L12 = x12 · n = 0, and L22 = x22 · n = 0. So, the second
−a 0
fundamental form is . When comparing the second fundamental form of the cylinder to
0 0
that of the plane from the previous example, we can see that they differ only in the first coefficient.
Note also that these two matrices have the same determinant – we shall elaborate on this fact later.
Since x12 = x22 = (0, 0, 0), Γk12 = Γk21 = Γk22 = 0. The remaining two Christoffel symbols, Γ111 and
Γ211 can be computed as follows.

Γ111 = x11 · x1 g 11 + x11 · x2 g 21 = 0g 11 + 0(0) = 0 and

Γ211 = x11 · x1 g 12 + x11 · x2 g 22 = 0(0) + 0g 22 = 0.

Hence, all Christoffel symbols are zero.
Example 3. The Christoffel symbols and the second fundamental form of a sphere.
Consider the sphere x = (a cos θ cos φ, a sin θ cos φ, a sin φ) of radius a parametrized by geographic
coordinates. Recall (or compute again) that x1 = (−a sin θ cos φ, a cos θ cos φ, 0), x2 = (−a cos θ sin φ,
−a sin θ sin φ, a cos φ), g11 = a2 cos2 φ, g12 = 0, g22 = a2 , and n = (cos θ cos φ, sin θ cos φ, sin φ).
Compute that x11 = (−a cos θ, −a sin θ cos φ, 0), x12 = x21 = (a sin θ sin φ, −a cos θ sin φ, 0), and
x22 = (−a cos θ cos φ, −a sin θ cosφ, −a sin φ), so that  L11 = −a cos2 φ, L12 = 0 and L22 = −a and
−a cos2 φ 0
the second fundamental form is .
0 −a
1
Compute then that g 11 = a2 cos 2 φ, g
12
= 0, g 22 = a12 , so that the inverse of the first fundamental
1
 
2 2 0
form is a cos φ 1 . Lastly, compute the dot products of the second and the first derivatives to
0 a2
be x11 · x1 = 0, x11 · x2 = a2 sin φ cos φ, x12 · x1 = x21 · x1 = −a2 sin φ cos φ, x12 · x2 = x21 · x2 = 0,
x22 · x1 = 0, x22 · x2 = 0.
Thus, the Christoffel symbols are Γ111 = x11 · x1 g 11 + x11 · x2 g 21 = 0g 11 + x11 · x2 (0) = 0,
1
Γ211 = x11 · x1 g 12 + x11 · x2 g 22 = 0(0) + a2 sin φ cos φ = sin φ cos φ,
a2
1 − sin φ
Γ112 = Γ121 = x12 · x1 g 11 + x12 · x2 g 21 = −a2 sin φ cos φ + 0(0) =
a2 2
cos φ cos φ
and Γ212 = Γ221 = x12 · x1 g 12 + x12 · x2 g 22 = 0 + 0 = 0, Γ122 = x22 · x1 g 11 + x22 · x2 g 21 = 0 + 0 = 0, and
Γ222 = x22 · x1 g 12 + x22 · x2 g 22 = 0 + 0 = 0. Thus, all but three Christoffel symbols are zero. We shall
use the calculation above to compute the geodesics on the sphere.
Example 4. Find the coefficients of the second fundamental form of the surface z = z(x, y).
With x, y as parameters, we have that x(x, y) = (x, y, z(x, y)). Let us shorten the notation by
using z1 for zx , z2 for zy , and z11 = zxx , z12 = z21 = zxy and z22 = zyy . We have x1 = (1, 0, z1 ) and
x2 = (0, 1, z1 ) so that g11 = 1 + z12 , g12 = z1 z2 and g22 = 1 + z22 . Thus, g = 1 + z12 + z22 . One can
also compute that n = √1g (−z1 , −z2 , 1) and x11 = (0, 0, z11 ), x12 = (0, 0, z12 ), x22 = (0, 0, z22 ). Thus
z
Lij = √ijg .

4
 Normal and Geodesic curvature. Geodesics

The tangential and the normal component. Let us consider a curve γ = γ(t) on a surface
x so that γ(t) = x(u(t), v(t)).
The chain rule gives us that
γ 0 (t) = u0 x1 + v 0 x2 .
Using the notation xij , i, j = 1, 2, for the second partial derivatives of x, and differentiating the
equation for γ 0 with respect to t again, we obtain that

γ 00 = u00 x1 +u0 (u0 x11 +v 0 x12 )+v 00 x2 +v 0 (u0 x21 +v 0 x22 ) = u00 x1 +v 00 x2 +u02 x11 +u0 v 0 x12 +u0 v 0 x21 +v 02 x22 .

To be able to use Einstein notation, let us

denote u by u1 and v by u2 .
Thus the part u00 x1 + v 00 x2 can be written as (ui )00 xi and the part u0 v 0 x12 + u0 v 0 x21 + v 02 x22 as
(ui )0 (uj )0 xij . This gives us the short version of the formula above

γ 00 = (ui )00 xi + (ui )0 (uj )0 xij .

Substituting the Gauss formula xij = Γkij xk +

Lij n in the above formula, we obtain that
γ 00 = (ui )00 xi + (ui )0 (uj )0 (Γkij xk + Lij n) =
((uk )00 + Γkij (ui )0 (uj )0 )xk + (ui )0 (uj )0 Lij n.
The part ((uk )00 + Γkij (ui )0 (uj )0 )xk is the tan-
gential component, it is denoted by γ 00tan , and
it is in the tangent plane. The part (ui )0 (uj )0 Lij n
is the normal component and it is denoted by
γ 00nor . The normal component is colinear with n
and, thus, orthogonal to the tangent plane.

Up to a sign, the length of the tangential com-

ponent γ 00tan determines the geodesic curvature
κg and the length of the normal component γ 00nor
determines the normal curvature κn . Thus,
κg = ±|γ 00tan |, and κn = ±|γ 00nor |.
If γ is parametrized by the arc length, κ =
|γ 00 |. The figure on the right illustrates that the
curvatures κ, κg and κn are related by the formula

κ2 = κ2g + κ2n .
Since κ can be interpreted as the total extent of curving of γ, the formula κ2 = κ2g +κ2n means that
the two factors, κg and κn contribute to this extend of curving: κn indicates the extent of curving
coming from the curving of a surface and κg the extent of the interior curving. Thus, one can think
of κn as an external and κg as an internal curvature. The following two scenarios may be helpful for
understanding this.

5
 1. External curvature of the surface. If a
surface itself is curved relative to the sur-
rounding space in which it embeds, then a
curve on this surface will be forced to bend
as well. The level of this bending is mea-
sured by the normal curvature κn .
For example, the curving of any normal sec-
tion of a surface comes only from curving of Example with κn 6= 0, κg = 0
the surface itself. In particular, a horizon-
tal circle on the cylinder x2 + y 2 = a2 is
curved just because the cylinder is a rolled-
up plane. If we “un-roll” the cylinder back
to a plane, the circle becomes a straight line.
The horizontal circle has κg = 0 and κn 6= 0.
2. Internal curvature of the surface. Con-
sider a curve “meandering” in a plane. The
curvature of this curve comes only from the
“meandering”, not from any exterior curv-
ing of the plane since the plane is flat. This
level of bending is measured by the geodesic
curvature κg . In this case, κn = 0 and
κg 6= 0. Example with κn = 0, κg 6= 0

We now examine more closely the computation of the two curvatures. For κn , start from the
formula
γ 00 = γ 00tan + γ 00nor = ((uk )00 + Γkij (ui )0 (uj )0 )xk + (ui )0 (uj )0 Lij n
and dot it by n. Since xk · n = 0 and n · n = 1, we obtain that γ 00 · n = (ui )0 (uj )0 Lij . This last
expression computes the normal curvature. Thus

κn = γ 00 · n = (ui )0 (uj )0 Lij .

Next, we show that κg can be computed as

(n × γ 0 ) · γ 00 if γ is parametrized by the arc length.
To show this, start by noting that γ 00nor · γ 0 = 0
since γ 0 is in the tangent plane and γ 00nor is per-
pendicular to it.
Since γ is unit-speed, the length of γ 0 is con-
stant (and equal to 1) so γ 00 · γ 0 = 0 (recall the ar-
gument that differentiating the relation γ 0 ·γ 0 = 1
produces γ 00 · γ 0 + γ 0 · γ 00 = 2γ 00 · γ 0 = 0). Thus
0 = γ 00 · γ 0 = (γ 00tan + γ 00nor ) · γ 0 = γ 00tan · γ 0 + γ 00nor · γ 0 = γ 00tan · γ 0 + 0 = γ 00tan · γ 0 .
So, γ 00tan is orthogonal to γ 0 as well. Since γ 00tan is orthogonal to both γ 0 and n, it is colinear
with n × γ 0 . Hence, γ 00tan is a multiple of n × γ 0 . Since the length |γ 00tan | is ±κg and the length

6
|n × γ 0 | = |n||γ 0 | sin(± π2 ) = ±1 (recall that both n and γ 0 have length 1), we have that
γ 00tan = κg (n × γ 0 ).
Dotting the above identity by n × γ 0 , we obtain (n × γ 0 ) · γ 00tan = κg . But since n × γ 0 is perpendicular
to γ 00nor , the mixed product (n × γ 0 ) · γ 00tan is equal to (n × γ 0 ) · γ 00 . Thus κg = (n × γ 0 ) · γ 00 or, using
the bracket notation

κg = (n × γ 0 ) · γ 00 = [n, γ 0 , γ 00 ] = [n, T, T0 ].

Geodesics. A curve γ on a surface is said to be a geodesic if κg = 0 at every point of γ.

Thus, a geodesic curves only because of the curving of the surface – the extent of internal curving is
zero.
We show that the following conditions are equivalent. For some of the equivalences below, we
need to assume that γ is parametrized by the arc length.

1. γ is a geodesic. 2. [n, T, T0 ] = 0.
3. γ 00tan = 0 at every point of γ. 4. γ 00 = γ 00nor at every point of γ.
k 00 k i 0 j 0
5. (u ) + Γij (u ) (u ) = 0 for k = 1, 2. 6. κ = ±κn at every point of γ.
7. N is colinear with n (i.e. N = ±n).
Conditions 1 and 2 are equivalent since κg = [n, T, T0 ]. Conditions 3 and 4 are clearly equivalent.
Conditions 3 and 5 are equivalent since γ 00tan = ((uk )00 + Γkij (ui )0 (uj )0 )xk . Conditions 1 and 3 are
equivalent since γ 00tan = 0 ⇔ κg = |γ 00tan | = 0.
To see that conditions 1 and 6 are equivalent, recall the formula κ2n = κ2g + κ2n . Thus, if κg = 0
then κ2 = κ2n ⇒ κ = ±κn . Conversely, if κ = ±κn , then κ2 = κ2n ⇒ κ2g = 0 ⇒ κg = 0.
Finally, to show that 1 and 7 are equivalent, recall that γ 00 = T0 = κN if γ is parametrized by
the arc length. Assuming that γ is a geodesic, we have that γ 00 = γ 00nor = κn n. Thus, κN = κn n
and so the vectors N and n are colinear, in particular N = ±n since they both have unit length.
Conversely, if N and n are colinear, then γ 00 (always colinear with N if unit-speed parametrization is
used) is colinear with n as well. So γ 00 = γ nor and so condition 4 holds. Since we showed that 1 and
4 are equivalent, 1 holds as well. This concludes the proof that all seven conditions are equivalent.
Examples.

1. If γ is the normal section in the direction of

a vector v in the tangent plane (intersection
of the surface with a plane orthogonal to the
tangent plane), then the normal vector N has
the same direction as the unit normal vector n
and so N = ±n (the sign is positive if the ac-
celeration vector has the same direction as n).
So, every normal section is a geodesic.
2. A great circle on a sphere is the normal section and so, it is a geodesic. Having two points
on a sphere which are not antipodal (i.e. exactly opposite to one another with respect to the
center), there is a great circle on which the two points lie. Thus, the “straightest possible”
curve on a sphere that connects any two points is a great circle.

7
 Thus, κg of a great circle is 0 and its curva-
ture κ comes just from the normal curvature
κn (equal to a1 if the radius is a).
Any circle on a sphere which is not “great” (i.e.
whose center does not coincide with a center of
the sphere and the radius is smaller than a) is
not a geodesic. Any such “non-great“ circle is
an example of a curve on a surface whose nor-
mal vector N is not colinear with the normal
Just great circles are geodesics
vector of the sphere n.

Computing the geodesics. Consider the two equations (uk )00 + Γkij (ui )0 (uj )0 = 0 for k = 1, 2.
The expressions on the left side correspond to the coefficients of γ 00 with x1 and x2 . Two equations
considered together represent a system of two second order differential equations whose solutions
compute geodesics on a surface. So, this system of differential equations is a tool for explicitly
obtaining formulas of geodesics on a surface. This system is frequently being solved in everyday life,
for example when determining the shortest flight route for an airplane.
Consider, for example, the air traffic routes from Philadelphia to London, Moscow and Hong
Kong represented below. Each city being further from Philadelphia than the previous one, makes
the geodesic path appear more curved when represented on a flat plane. Still, all three routes are
determined as geodesics – as intersections of great circles on Earth which contain Philadelphia and
the destination city.

Example 1. Geodesics of a plane. For simplicity, let us consider the xy-plane, z = 0 again.
Thus x = (x, y, 0) so u = x and v = y. We computed the Christoffel symbols before and obtained
that they all vanish. Thus, the equations of geodesics are x00 = 0 and y 00 = 0. Integrating both
equations with respect to the unit-speed parameter s twice produces the following.

8
 x = as + b and y = cs + d

These equations are parametric equations of a line. Hence, geodesics are straight lines and an arc-
length curve γ(s) in xy-plane is a geodesic exactly if it is a straight line.
Example 2. Geodesics of a cylinder. Consider the cylinder x2 + y 2 = a2 parametrized by
x = (a cos t, a sin t, z) so that u = t and v = z. We computed the Christoffel symbols before and
obtained that they are all zero. Thus, the equations of geodesics are given by t00 = 0 and z 00 = 0.
These equations have solutions t = as + b and z = cs + d which are parametric equations of a line
in tz-plane. This shows that a curve on a cylinder is geodesic if and only if it is a straight line in
zt-plane. Thus, a unit-speed curve γ(s) on a cylinder is a geodesic exactly if it becomes a straight
line if the cylinder is “un-rolled” into a plane.
In particular, both meridians and parallels on the cylinder are geodesics. The meridians are
z-curves. They are parametrized by unit-speed since x2 = (0, 0, 1) has unit length. Since t = t0 is a
constant on a z-curve, t0 = t00 = 0 so the first equation holds. The second holds since z 0 = dz dz
=1
00
and so z = 0. Hence, both geodesic equations hold.
The parallels (or circles of latitude), are t-curves with z = z0 a constant. They are parametrized
by unit-speed for t = as . Thus, t0 = a1 and t00 = 0 and z 0 = z 00 = 0 so both geodesic equations hold.
Another way to see that the circles of latitude γ = (a cos t, a sin t, z0 ) = (a cos as , a sin as , z0 ) are
geodesics is to compute the second derivative (colinear with N) and to note that it is a multiple
of n (thus condition 4. holds). The first derivative is γ 0 = (− sin as , cos as , 0) and the second is
γ 00 = (− a1 cos as , − a1 sin as , 0). The second derivative is a multiple of n = (cos as , sin as , 0) (γ 00 = −1a
n)
and so γ is a geodesic.
Example 3. Meridians of a cone are geodesics. Consider the cone obtained by revolving
the line (3t, 4t) in rz-plane
q
about the z-axis. Note that the unit-speed parametrization of this
q
2 2
line is ( 35 s, 54 s) since 3 4 25


5
+ 5
= 25
= 1. Using this parametrization, the cone is given by
x = 51 (3s cos θ, 3s sin θ, 4s). Since the line has the unit-speed parametrization, the s-curves (the
meridians) on the cone have the unit-speed parametrization also.
We show that all the meridians are geodesics. Compute that x1 = 51 (3 cos θ, 3 sin θ, 4) and x2 =
1
5
(−3s sin θ, 3s cos θ, 0) so that g11 = x1 · x1 = 1 (this also tells you that the meridians have the unit-
 
9 2 1 0
speed parametrization), g12 = 0 and g22 = 25 s . So that the first fundamental form is 2 .
0 9s25
1 1
Then compute that x11 = (0,  = 5 (−3 sin θ, 3 cos θ, 0) and x22 = 5 (−3s cos θ, −3s sin θ, 0).
 0, 0), x12
1 0
The inverse matrix of [gij ] is 25 . Since x11 = 0, Γ111 = Γ211 = 0. Since x12 · x1 = 0 and g 21 = 0,
0 9s 2

Γ112 = Γ121 = 0. Since x22 · x2 = 0 and g 12 = 0, Γ222 = 0. The remaining symbols are
9s 25 1
Γ212 = Γ221 = x12 · x1 g 12 + x12 · x2 g 22 = 0 + 2
= and
25 9s s
−9s −9
Γ122 = x22 · x1 g 11 + x22 · x2 g 21 = +0= s.
25 25
Thus, two equations of geodesics are

s00 − 9
25
s(θ0 )2 = 0 and θ00 + 2s s0 θ0 = 0

9
for a unit-speed curve γ on the cone for which s and θ depend on a parameter t.
In particular, on a meridian s is the parameter (think that “t” is s here) and θ is a constant.
Thus θ0 = θ00 = 0 and s0 = 1, s00 = 0. So, both geodesic equations are satisfied and this shows that
the meridians on the cone are geodesics.
Example 4. Meridians of a sphere and the equator are geodesics. Recall that the
unit-speed parametrization of a circle of radius a is (a cos as , a sin as ). So, to obtain the unit-speed
parametrization of the sphere of radius a, one can consider the sphere as the surface of revolution of
a unit-speed circle in rz-plane. This produces the parametrization
s s s
x = (a cos θ cos , a sin θ cos , a sin ).
a a a
Compute x1 = xθ = (−a sin θ cos as , a cos θ cos as , 0), x2 = xs = (− cos θ sin as , − sin θ sin as , cos as )
and g11 = a2 cos2 as , g12 = 0, g22 simplifies to 1 (this also tells  you that the meridians have the
2 2 s
a cos a 0
unit-speed parametrization). The first fundamental form is . g = a2 cos2 as and
0 1
n = √1g (x1 × x2 ) = (− cos θ cos as , − sin θ cos as , − sin as ).
Compute that x11 = (−a cos θ cos as , −a sin θ cos as , 0), x12 = (sin θ sin as , − cos θ sin as , 0), and x22 =
 1 
1 s 1 s 1 s a2 cos2 as
0
(− a cos θ cos a , − a sin θ cos a , − a sin a ). The inverse of the first fundamental form is
0 1
2 s s 1 1 −1 s
and the Christoffel symbols are Γ11 = a cos a sin a , Γ12 = Γ21 = a tan a , and the rest are 0.
Thus, two equations of geodesics are

s00 + a cos as sin as (θ0 )2 = 0 and θ00 − a2 tan as s0 θ0 = 0

On a meridian, s is a parameter and θ is constant. Hence s0 = 1, s00 = 0 and θ0 = 0, θ00 = 0. Hence

both equations of the geodesics are satisfied.
On the equator, φ = as = 0 so s = 0. Hence the equation of the equator is (a cos θ, a sin θ, 0). Note
that this has the unit-speed parametrization if θ = aθ . Thus if θ is considered as a parameter, θ0 = a1 ,
θ00 = 0 and s0 = s00 = 0. Hence the two equations become

a cos as sin as a12 = 0 and 0 = 0

Using that s = 0, the first equation is 0 = 0 and so so both equations are satisfied.

Properties of geodesics. Since a geodesic curves solely because of curving of the surface, a
geodesic has the role of a straight line on a surface. Moreover, geodesics have the following properties
of straight lines.

1. If a curve γ(s) for a ≤ s ≤ b is the shortest route on the surface that connects the points γ(a)
and γ(b), then γ is a geodesic. One of the project topics focuses on the proof of this claim.
Note that the converse does not have to hold – if a curve is geodesic, it may not give the
shortest route between its two points. For example, a north pole and any other point on a
sphere but the south pole, determine two geodesics connecting them, just one of which will be
the shortest route.

10
 2. Every point P on a surface and a vector v in the tangent plane uniquely determine a geodesic
γ with γ(0) = P and γ 0 (0) = v.
As opposed to the straight lines, a geodesic connecting two points does not have to exist. For
example, consider the xy-plane without the origin. Then there is no geodesic connecting (1,0)
and (-1,0). Also, there can be infinitely many geodesics connecting two given points on a
surface (for example, take north and south poles on a sphere).

Two points do not determine a “line” There are many “lines” passing two points

The Gaussian Curvature

In this section, we obtain a simple formula computing the Gaussian K. Recall that the formula
for the normal curvature is given by κn = Lij (ui )0 (uj )0 . If the curve γ is not given by the arc-length
L (ui )0 (uj )0
parametrization, this formula becomes κn = ij |γ 0 |2 . Recall the formula for |γ 0 |2 from earlier
section

|γ 0 (t)|2 = g11 ((u1 )0 )2 + 2g12 (u1 )0 (u2 )0 + g22 ((u2 )0 )2 in Einstein notation = gij (ui )0 (uj )0

Thus, the normal curvature can be computed as

Lij (ui )0 (uj )0

κn =
gkl (uk )0 (ul )0

Differentiating this equation with respect to (ur )0 for r = 1, 2, and setting derivatives to zero in
order to get conditions for extreme values, we can obtain the conditions that (Lij − κn gij )(uj )0 = 0
for i = 1, 2. A nonzero vector ((u1 )0 , (u2 )0 ) can be a solution of these equations just if the determinant
of the system |Lij − κn gij | is zero.
This determinant is equal to (L11 − κn g11 )(L22 − κn g22 ) − (L12 − κn g12 )2 . Substituting that de-
terminant of [gij ] is g and denoting the determinant of [Lij ] by L, we obtain the following quadratic
equation in κn
gκ2n − (L11 g22 + L22 g11 − 2L12 g12 )κn + L = 0
The solutions of this quadratic equation are the principal curvatures κ1 and κ2 . The Gaussian
K is equal to the product κ1 κ2 and from the above quadratic equation this product is equal to the

11
quotient Lg (recall that the product of the solutions x1 and x2 of a quadratic equation ax2 + bx + c
is equal to ac ). Thus,

L
K= g

that is the Gaussian K is the quotient of the determinants of the coefficients of the second
and the first fundamental forms.
From the formula (Lij − κn gij )(uj )0 = 0 it follows that if L12 = L21 = g12 = g21 = 0, then the
principal curvatures are given by Lg11
11
and Lg22
22
and the principal directions are x1 and x2 . Conversely,
if directions x1 and x2 are principal, then L12 = L21 = g12 = g21 = 0. Using this observation, we can
conclude that the principal directions on a surface of revolution are determined by the meridian and
the circle of latitude through every point.
Note that from the equation (Lij − κn gij )(uj )0 = 0 also follows that the principal curvatures are
the eigenvalues of the operator determined by the first and the second fundamental form that can
be expressed as  
−1 L11 g22 − L12 g12 L12 g22 − L22 g12
S=g .
L12 g11 − L11 g12 L22 g11 − L12 g12
and is called the shape operator.
Examples. Gaussian of a plane, a cylinder, a sphere, and the surface z = z(x, y).

1. Recall that the first and the second fundamental forms of the xy-plane are the identity and the
zero matrix. So, g = 1 and L = 0 which readily gives you that K = 01 = 0.
 2 
2 2 2 a 0
2. Recall that the first and the second fundamental forms of the cylinder x +y = a are
  0 1
−a 0
and . Thus g = a2 , L = 0 and so K = a02 = 0. Recall that the fact that a plane
0 0
and a cylinder have the same Gaussian indicates that they are locally indistinguishable (locally
isometric).

3. 
The first and the
 second
 fundamental form
 of the sphere of radius a centered at the origin are
a2 cos2 φ 0 −a cos2 φ 0
and . Thus g = a4 cos2 φ and L = a2 cos2 φ. Hence
0 a2 0 −a

a2 cos2 φ 1
K= 4 2
= 2.
a cos φ a
This agrees with our earlier conclusion on K of a sphere.

4. 
We have computed  the first
 and the second
 fundamental form of the surface z = z(x, y) to be
1 + z12 z1 z2 z11 z12 z z −z 2
2 , and g1 . Thus g = 1 + z12 + z22 , L = 11 22g 12
z1 z2 1 + z2 z12 z22
2 2
z11 z22 − z12 z11 z22 − z12
K= = .
g2 (1 + z12 + z22 )2

12
 The Curvature Tensor. Theorema Egregium

Recall that Theorema Egregium states that the Gaussian curvature K can be calculated intrin-
sically, that is using the first fundamental form only. The formula K = Lg enables one to prove the
“Remarkable Theorem” by proving that the determinant L can be computed intrinsically only in
2
terms of gij (the determinant g = g11 g22 −g12 is already clearly intrinsic). Note that while Lij = xij ·n
are not intrinsic (note the presence of n in this formula), the determinant L, surprisingly, is intrinsic.
We show that L is intrinsic by showing the following steps.

1. The Christoffel symbols Γkij are intrinsic. Note that this implies that the geodesic curvature is
intrinsic also.
l
2. We introduce the Riemann curvature tensor Rijk and represent it via the Christoffel symbols.
l
By step 1, Rijk is intrinsic also.
l
3. We represent L via Rijk and gij .

The Christoffel symbols Γkij can be computed intrinsically. To prove this statement, start
by differentiating the equation gij = xi · xj with respect to uk . Get

∂gij
= xik · xj + xi · xjk
∂uk
In a similar manner, we obtain
∂gik ∂gjk
j
= xij · xk + xi · xkj and = xji · xk + xj · xki
∂u ∂ui
Note that the second equation can be obtained from the first by permuting the indices j and k
and the third equation can be obtained from the second by permuting the indices i and j. This is
called cyclic permutation of indices.
At this point, we require the second partial derivatives to be continuous as well. This condition
will guarantee that the partial derivatives xij and xji are equal. In this case, adding the second and
third equation and subtracting the first gives us
∂gik ∂gjk ∂gij
+ − k = xij · xk + xi · xkj + xji · xk + xj · xki − xik · xj − xi · xjk = 2xij · xk
∂uj ∂ui ∂u
Thus,
 
1 ∂gil ∂gjl ∂gij
Γkij = (xij · xl )g lk = 2 ∂uj
+ ∂ui
− ∂ul
g lk .

This shows that the Christoffel symbols Γkij can be computed just in terms of the metric coefficients gij
that can be determined by measurements within the surface. Since the geodesics can be computed
from the two differential equations which only feature the Christoffel symbols and the Christoffel
symbols can be computed only using gij , we showed the following theorem.
Theorem. The geodesic curvature is intrinsic.

13
 The Riemann curvature tensor. The coefficients of the Riemann curvature tensor (or
Riemann-Christoffel curvature tensor) are defined via the Christoffel symbols by
∂Γlik ∂Γlij
l
Rijk = − + Γpik Γlpj − Γpij Γlpk
∂uj ∂uk
The geometric meaning of this tensor cannot really be seen from this formula. Roughly speaking,
this tensor measures the extent of deviation of initial vector and the vector resulting when the initial
vector is parallel transported around a loop on a surface. For example, when a vector in space is
parallel transported around a loop in a plane, it will always return to its original position and the
Riemann curvature tensor directly measures the failure of this on a general surface. The extent of
this failure is known as the holonomy of the surface.
l
To represent the determinant L via Rijk and gij , we need a set of formulas known as the Gauss’s
equations. These equations are proven using Weingarten’s equations and as a byproduct of the proof,
we obtain a set of equations know as the Codazzi-Mainardi equations.
Proposition.
Weingarten’s equations nj = −Lij g ik xk

l
Gauss’s equations Rijk = Lik Ljp g pl − Lij Lkp g pl .

∂Lij ∂Lik
Codazzi-Mainardi equations ∂uk
− ∂uj
= Γlik Llj − Γlij Llk .

Proof. Let us prove Weingarten’s equations

first. Since n · n = 1, nj · n = 0 and so nj is
in tangent plane. Thus, it can be represented as
a linear combination of x1 and x2 . Let alj denote
the coefficients of nj with xl . Thus nj = alj xl .

Differentiate the equation n · xi = 0 with re-

spect to uj and obtain nj · xi + n · xij = 0. Recall
that Lij = n · xij .
Thus, 0 = nj · xi + Lij = alj xl · xi + Lij = alj gli + Lij and so alj gli = −Lij . To solve for alj , multiply
both sides by g ik and recall that gli g ik = δlk . Thus we have −Lij g ik = alj gli g ik = alj δlk = akj . This gives
us
nj = akj xk = −Lij g ik xk .
To prove the remaining two sets of equations, let us start by Gauss formulas for the second
derivatives
xij = Γlij xl + Lij n

14
 Differentiate with respect to uk and obtain
∂Γlij ∂L
xijk = ∂uk l
x + Γlij xlk + ∂uijk n + Lij nk
∂Γlij ∂L
= ∂uk l
x + Γlij (Γplk xp + Llk n) + ∂uijk n − Lij Lpk g pl xl (sub Gauss and Wein. eqs)
∂Γlij ∂L
= ∂uk l
x + Γlij Γplk xp − Lij Lpk g pl xl + Γlij Llk n + ∂uijk n (regroup the terms)
∂Γlij p l pl l ∂Lij
= k xl + Γij Γpk xl − Lij Lpk g xl + Γij Llk n + ∂uk n (make tangent comp via xl )
∂u∂Γ l
  
ij p l pl l ∂Lij
= ∂u k + Γ Γ
ij pk − L ij L pk g x l + Γ ij L lk + ∂u k n (factor xl and n)

Interchanging j and k we obtain

 l   
∂Γik p l pl l ∂Lik
xikj = + Γik Γpj − Lik Lpj g xl + Γik Llj + n
∂uj ∂uj
Since xijk = xikj , both the tangent and the normal components of xijk − xikj are zero. The
coefficient of the tangent component is
∂Γlij p l ∂Γlik
k
+ Γ Γ
ij pk − L ij L pk g pl
− j
− Γpik Γlpj + Lik Lpj g pl = Lik Lpj g pl − Lij Lpk g pl − Rijk
l
= 0.
∂u ∂u
This proves the Gauss’s equations.
The coefficient of the normal component is
∂Lij ∂Lik
Γlij Llk + k
− Γlik Llj − =0
∂u ∂uj
proving Codazzi-Mainardi equations. QED.
We can now prove Theorema Egregium.
Theorema Egregium. The Gaussian K is dependent solely on the coefficient of the first
fundamental form by
i
g1i R212
K= g
.

l
Proof. Multiplying Gauss’s equation Rijk = Lik Ljp g pl − Lij Lkp g pl by glm , we obtain Rijk
l
glm =
pl pl p
Lik Ljp g glm − Lij Lkp g glm = (Lik Ljp − Lij Lkp )δm = Lik Ljm − Lij Lkm . Taking i = k = 2, and
l
j = m = 1, we obtain L = L22 L11 − L21 L21 = R212 gl1 .
l
R212 gl1
L L11 L22 −L12 L21
From here we have that K = g = g
= g
. QED.
l
Once we obtain the relation Rijk glm = Lik Ljm − Lij Lkm in the above proof, we could also take
l
i = k = 1, and j = m = 2, and obtain R121 gl2 = L11 L22 − L12 L12 = L. Thus shows that
i i
R212 gi1 = R121 gi2
and so K can be computed both as

i
g2i R121 i
g1i R212
K= g
and as K = g
.

15
 Total curvature. The surface integral of the Gaussian curvature over some region of a surface
is called the total curvature. The total curvature directly corresponds to the deviation of the sum
of the angles of a geodesic triangle from 180 degrees. In particular,

• On a surface of total curvature zero, (such as a plane for example), the sum of the angles of a
triangle is precisely 180 degrees.

• On a surface of positive curvature, the sum of angles of a triangle exceeds 180 degrees. For
example, consider a triangle formed by the equator and two meridians on a sphere. Any
meridian intersects the equator by 90 degrees. However, if the angle between the two meridians
is θ > 0, then the sum of the angles in the triangle is 180 + θ degrees. In the figure on the
right, the angles add to 270 degrees.

• On a surface of negative curvature, the sum of the angles of a triangle is less than 180 degrees.

P P P
K=0⇒ angles = 180, K>0⇒ angles > 180, K<0⇒ angles < 180

The total curvature impact also the number of lines passing a given point, parallel to a given line.
Surfaces for which this number is not equal to one are models of non-Euclidean geometries.

Recall that the parallel postulate in Eu-

clidean geometry is stating that in a plane
there is exactly one line passing a given point that
does not intersect a given line, i.e. there is ex-
actly one line parallel to a given line passing a
given point.

In elliptic geometry the parallel postulate is replaced by the statement that there is no line
through a given point parallel to a given line. In other words, all lines intersect.
In hyperbolic geometry the parallel postulate is replaced by the statement that there are at
least two distinct lines through a given point that do not intersect a given line. As a consequence,
there are infinitely many lines parallel to a given line passing a given point.

16
 We present the projective plane which is a model of elliptic geometry and Poincaré half plane
and disc which are models of hyperbolic geometry.

The projective plane RP 2 is defined as the

image of the map that identifies antipodal points
of the sphere S 2 . More generally, n-projective
plane RP n is defined as the image of the map that
identifies antipodal points of the n-sphere S n .
While there are lines which do not intersect
(i.e. parallel lines) in a regular plane, every two
“lines” (great circles on the sphere with antipodal
points identified) in the projective plane intersect
in one and only one point. This is because every pair of great circles intersect in exactly two points
antipodal to each other. After the identification, the two antipodal points become a single point and
hence every two “lines” of the projective plane intersect in a single point. The standard metric on
the sphere gives rise to the metric on the projective plane. In this metric, the curvature K is positive.

The projective plane can also be represented

as the set of lines in R3 passing the origin. The
distance between two such elements of the pro-
jective plane is the angle between the two lines
in R3 . The “lines” in the projective planes are
the planes in R3 that pass the origin. Every two
such “lines” intersect at a point (since every two
planes in R3 that contain the origin intersect in a
line passing the origin).

All lines intersect

2
Poincaré half-plane.
 1Consider the upper half y > 0 of the plane R with metric given by the
2 0
first fundamental form y . In this metric, the geodesic (i.e. the “lines”) are circles with
0 y12
centers on x-axis and half-lines that are perpendicular to x-axis.

17
 Given one such line and a point, there is more than one line passing the point that does not
intersect the given line. In the given metric, the Gaussian curvature is negative.

Poincaré disc. Consider the disc x2 + y 2 < 1

in R2 with" metric given by the # first fundamen-
1
(1−x2 −y 2 )2
0
tal form 1 . In this met-
0 (1−x2 −y 2 )2
ric, the geodesic (i.e. the “lines”) are diameters
of the disc and the circular arcs that intersect the
boundary orthogonally. Given one such “line”
and a point in the disc, there is more than one
line passing the point that does not intersect the
given line. In the given metric, K is negative.

The Fundamental Theorem of Surfaces

The Fundamental Theorem of Surfaces states that a surface is uniquely determined by the
coefficients of the first and the second fundamental form. More specifically, if gij and Lij
are symmetric functions (i.e. gij = gji and Lij = Lji ) such that g11 > 0 and g > 0, and such that
both Gauss’s and Codazzi-Mainardi equations hold, there is a coordinate patch x such that gij and
Lij are coefficients of the first and the second fundamental form respectively. The patch x is unique
up to a rigid motion (i.e. rotations and translations in space).
l
Recall that the Gauss’s and Codazzi-Mainardi equations relate the coefficients Rijk , Γkij and gij
l k k
with Lij . Since Rijk can be expressed via Γij and Γij can be expressed via gij , the Gauss’s and Codazzi-
Mainardi equations can be viewed as equations connecting the coefficients of the first fundamental
form gij with the coefficients of the second fundamental form Lij .
The idea of the proof of the Fundamental Theorem of Surfaces is similar to the proof of the
Fundamental Theorem of Curves. Namely, note that the vectors x1 and x2 in tangent plane are
independent by assumption that the patch is proper. Moreover, the vector n is independent of x1

18
and x2 since it is not in the tangent plane. Thus the three vectors x1 , x2 and n represent a basis, or
a “moving frame”, of the surface analogous to the moving frame T, N and B of a curve.
Gauss formula and Weingarten’s equations represent (partial) differential equations relating the
derivatives of x1 , x2 and n in terms of the three vectors themselves
Gauss formula xij = Γkij xk + Lij n.

Weingarten’s equations nj = −Lij g ik xk

As opposed to a system of ordinary differential equations, there is no theorem that guarantees an

existence and uniqueness of a solution of a system of partial differential equations. However, in case
of the equations for x1 , x2 and n, the existence and uniqueness of solution follows from the fact that
both Gauss’s and Codazzi-Mainardi equations hold. Thus, the apparatus x1 , x2 , n, gij , Lij describes
a surface.
Example. Surfaces of revolution with constant Gaussian. It can be shown that a surface
of revolution obtained by revolving a unit-speed curve (r(s), z(s)) about z-axis, has the Gaussian
00
curvature K equal to −rr (try to fill in the blanks here. You can use that the curve has unit speed, so
that r02 +z 02 = 1). If K is constant, this yields a differential equation r00 +Kr = 0 whichRcan be solved
02 02 s√
for r. If r is obtained, z can be obtained from the condition z + r = 1 as z = ± 0 1 − r02 ds.
Consider the three cases as below.

• K = a2 > 0. In this case, the equation is r00 + a2 r = 0. Its characteristic equation has
two complex zeros ±ai so that the general solution is r(s) = c1 cos as + c2 sin as. Using some
trigonometric identities, this solution can be represented as r(s) = C1 cos(as + C2 ). Combining
this with the z equation produces a sphere, the outer part of a torus, and a part of one of these
two classes of surfaces.

• K = −a2 < 0. In this case, the equation is

r00 − a2 r = 0. Its characteristic equation has
two real zeros ±a and the general solution is
r(s) = c1 eas + c2 e−as = C1 cosh as + C2 sinh as.
A pseudo-sphere (see the figure on the right),
the inner part of a torus and a part of one of
these two classes of surfaces.

19
 • K = 0. In this case, r00 = 0 and so r(s) = c1 s + c2 . The z(s) equation is also a line. If the line
(r(s), z(s)) is horizontal, the surface is the xy-plane or a part of it. If (r(s), z(s)) is vertical,
the surface is a cylinder or a part of it. In general, the surface is a cone or a part of it.

Manifolds

The concept of an n-dimensional manifold generalizes that of a (2-dimensional) surface. In fact,

all the definitions we introduced for surfaces (starting from the definition of a surface, and including
definitions of the first and the second fundamental forms, the tangent and normal vectors, Gaussian
and the other curvatures) can be generalized from two dimensions to n-dimensions.
Intuitively, a surface is an object in the 3-dimensional space R3 which locally looks like the
2-dimensional space R2 . Generalizing this idea to n-dimensions, we arrive to the concept of an n-
dimensional manifold or an n-manifold for short. Intuitively, an n-manifold locally looks like
the space Rn . We can also generalize the concept of a coordinate patch so that the inverse of a
coordinate patch of an n-manifold is a mapping of a region on the manifold to Rn . The coordinate
patches are required to be continuous with continuous derivatives, one-to-one, to overlap smoothly
on the intersection of their domains, and such that the concept of the tangent plane at any point is
well-defined. The coordinate patches provide local coordinates on the n-manifold.
We note some advantages of considering manifolds instead of only surfaces.

1. Surfaces are 2-manifolds so this more general study of n-manifolds agrees with that of surfaces
for n = 2.

2. Formulas for surfaces we have considered involve the indices ranging from 1 to 2. All these
formulas remain true for n-dimensional manifolds if we let the indices range from 1 to n. In
particular, the proof of Theorema Egregium generalizes to an n-manifold.

3. The study of n-manifolds can be carried out without assuming the embedding into the space
Rn+1 . Thus, one can study surfaces without considering an embedding of it in the 3-dimensional
space R3 .
Although the n-manifolds for n > 2 may not be embedded in the physical, three-dimensional
space, the theory of n-manifolds is used in high energy physics, quantum mechanics and rel-
ativity theory and, as such, is relevant. The Einstein space-time manifold, for example, has
dimension four.

Coordinate Patches. Recall that a proper

coordinate patch of a surface is given by paramet-
ric equations x = (x(u, v), y(u, v), z(u, v)) such
that x, y, z are one-to-one continuous functions
with continuous inverses, continuous derivatives
and such that ∂x∂u
× ∂x
∂v
6= 0. Also, one requires
that two such patches overlap smoothly: patches
x and x̄ overlap smoothly provided that the composite functions x−1 ◦ x̄ and x̄−1 ◦ x are one-to-one
and onto continuous functions with continuous derivatives on the intersection of the domain of the
patches.

20
 We define coordinate patches on an n-
manifold analogously: a coordinate patch of a
nonempty set of points M is a one-to-one con-
tinuous mapping from an open region D in Rn
into M given by x(u1 , u2 , . . . , un ) such that the
image of a small enough part of D is indistin-
guishable from a small enough piece of Rn . The
n-tuple (u1 , u2 , . . . , un ) represents the local co-
ordinates on M . Two coordinate patches x and
x̄ overlap smoothly if the composite functions x−1 ◦x̄ and x̄−1 ◦x are one-to-one and onto continuous
functions with continuous derivatives up to order at least three on the intersection of the domains.
The condition that the partial derivatives up to order three are continuous guarantees the validity
of the formulas involving the second fundamental form and equations from the previous section. If
derivatives of any order are continuous, such patch is said to be smooth.
With these requirements, M is an n-manifold if there is a collection of coordinate patches such
that: (1) The coordinate patches cover every point of M and they overlap smoothly. (2) Every
two different points on M can be covered by two different patches. (3) The collection of patches is
maximal with respect to the conditions (1) and (2). A coordinate patch of an n-manifold is also
called a chart and a collection of coordinate patches is called an atlas.
Examples.

1. Euclidean space Rn . This space consists of all points of the form (x1 , x2 , . . . , xn ). Considering
x1 , . . . , xn as n parameters of a coordinate patch makes Rn into an n-manifold. In case n = 2
this is the xy-plane consisting of points (x, y). In case n = 3, this is the three-dimensional space
consisting of points (x, y, z).
The dot (or inner) product of two elements of Rn can be defined analogously to the dot
product in the case n = 3 as follows.

(x1 , x2 , . . . , xn ) · (y1 , y2 , . . . , yn ) = x1 y1 + x2 y2 + . . . + xn yn .

The dot product enables one to define the concept of an angle α between two vectors ~x and
·~
~y using the same formula as in the 3-dimensional case: cos α = |~x~x||~ y
y|
where the length of
~x = (x1 , x2 , . . . , xn ) is defined by
√ q
|~x| = ~x · ~x = x21 + x22 + . . . x2n .

Thus, two vectors are perpendicular exactly when their dot product is zero. One of your
project topics explores the generalization of the cross product for n-dimensional vectors.

2. Hypersurfaces. Let f be a function with continuous derivatives that maps Rn+1 into R. The
set of all vectors x = (x1 , . . . , xn+1 ) in Rn+1 such that f (x) = 0 defines an n-manifold usually
referred to as hypersurface. You can think of the equation f (x) = 0 as the relation which
relates the variables. If you can solve for xn+1 , for example and obtain a relation how xn+1
depends on the previous n-variables which can be then considered as parameters. The fact
that there is n parameters makes this manifold an n-manifold.

21
 For example, the n-plane can be defined as the set of vectors x = (x1 , . . . , xn+1 ) in Rn+1 such
that
a1 x1 + a2 x2 + . . . an+1 xn+1 = d
for some constant vector a = (a1 , a2 , . . . , an+1 ) and a constant d. Thus, the equation of an
n-plane containing a point b and which is perpendicular to a is given by a · (x − b) = 0 (in
this case the constant d is equal to a · b). Note that the vector x − b lies in the n-plane. The
equation a · (x − b) = 0 means that the vector a is perpendicular to the n-plane.
In the case when n = 2, a = (a, b, c), and b = (x0 , y0 , z0 ), this produces the familiar plane
equation
a(x − x0 ) + b(y − y0 ) + c(z − z0 ) = 0.

3. The n-sphere S n is another example of a hypersurface. It can be defined as the set of vectors
x = (x1 , . . . , xn+1 ) in Rn+1 such that

x21 + x22 + . . . + x2n+1 = a2

i.e. that |x|2 = a2 . So, this is a hypersurface defined by the relation |x|2 = a2 or x · x = a2 .
In the case when n = 2, this produces the familiar equation

x 2 + y 2 + z 2 = a2 .

4. The 2-torus T 2 in three-dimensional space can be defined as the set of coordinates (x, y, u, v)
where (x, y) is on one circle S 1 and (u, v) on the other S 1 . This is written as T 2 = S 1 × S 1 .
If the circles are x2 + y 2 = a2 and u2 + v 2 = b2 , we obtain another parametrization of the
torus from practice problem 3 of the previous handout. Generalizing this to more dimensions,
the n-torus T n is defined as the set of vectors x = (x1 , . . . , x2n ) in R2n such that x21 + x22 =
1, x23 + x24 = 1, . . . , x22n−1 + x22n = 1.

Partial Derivatives. Note that the domain of a coordinate patch is in the space Rn so the
concept of partial derivatives is well-defined there. When working with manifolds, we may want to
be able to differentiate on the range of the coordinate patch as well. This can be done by considering
derivative of a real-valued functions.
Let f be function that maps a neighborhood U of a point P on n-manifold M into a subset of
R. The function f is smooth if the composition f ◦ x is smooth where x is a coordinate patch that
contains P (thus meets U ). Note that f ◦ x is a function that maps domain D of x into R. Thus,
∂f ∂(f ◦x)
we can define the derivative of f with respect to coordinate ui as ∂u i = ∂ui
◦ (x−1 ) and use it to
define the partial derivative operator at point P as
∂ ∂f
i
(P )(f ) = (P ).
∂u ∂ui

Directional Derivative and Tangent Vectors. We can define tangent vectors using partial
derivatives. In the case of surfaces, tangent vectors are defined as velocity vectors of curves on
surfaces. However, the definition of velocity vector is not available if we want to avoid referring to a
specific embedding in R3 . We can still define tangent vectors using an alternate route - via directional

22
derivative. To understand the idea, consider a vector v in R3 given by (v 1 , v 2 , v 3 ). This defines a
directional derivative operator by
∂ ∂ ∂
Dv = v · ∇ = v 1 + v2 + v3 .
∂x ∂y ∂z
This operator is defined on the set of all real-valued functions f by Dv f = v · ∇f = v 1 ∂f ∂x
+ v 2 ∂f
∂y
+
3 ∂f
v ∂z . Thus, any linear combination of the partial derivatives can be considered as a directional
derivative.
The set of all tangent vectors corresponds exactly to the set of all directional derivatives. For
every curve γ(t) on the surface x(u1 , u2 ), the velocity vector γ 0 (t)(f ) = dtd (f ◦ γ) can be seen as an
operator γ 0 (t)(f ) = ∂(f∂u◦x)
i ∂f dui dui ∂
−1 0
i (x ◦ γ(t)) du
dt
= ∂u i γ(t) dt . Thus, γ(t) = dt ∂ui (γ(t)) is the directional

derivative D((u1 )0 ,(u2 )0 ) .

Having defined partial derivatives on an n-manifold M allows us to define the tangent vectors at
a point P as the set of all linear combinations of the partial derivatives ∂u∂ i (P ). Thus, v is a tangent
vector if v is a linear combination of partial derivatives ∂u∂ i (P ) i.e. v is of the form v i ∂u∂ i (P ).
The set of all vectors tangent to M at a point P is called the tangent space at P and is denoted
by TP M. This space represents the generalization of tangent plane of 2-manifolds. After showing
that the partial derivatives are linearly independent, the vectors ∂u∂ i (P ) can be viewed as the basis
of the tangent space TP M.
Hilbert space, inner product, and metric. The coefficients of the first fundamental form
are defined using the dot product of the basis vectors xi = ∂u∂ i . If an n-manifold with a patch x is
embedded in Rn+1 , the dot product of the partial derivatives x1 , x2 , . . . , xn generalizes precisely and

gij = xi · xj for i, j = 1, 2, . . . , n.

If a manifold is to be considered without any embedding in any Euclidean space, one has to be
more more flexible about the interpretation of a “vector” and the meaning of the inner product in
order to generalize the definition of gij to arbitrary manifold. This leads us spaces known as Hilbert
spaces.

A Hilbert space is a vector space with real

coefficients meaning that we can refer to its el-
ements as vectors, that we can add two vectors,
that we can multiply a vector with a real number,
and that the axioms of a vector space from Linear
Algebra hold. In addition, this space is equipped
with an inner product · which produces a real
number v · w given two vectors v and w, and
which satisfies the following properties of the dot
product in Rn .
David Hilbert
Below, u, v, and w are vectors in the Hilbert space and a and b are real numbers.
1. The inner product is linear: (a u + b v) · w = a u · w + b v · w.
2. The inner product is symmetric: v · w = w · v.

23
 3. The inner product is positive definite: v · v ≥ 0 and v · v = 0 if and only if v is the zero vector.
The inner product enables one to measures the length of a vector v by
√
|v| = v · v.

Thus, one can define a metric using the inner product. Lastly, one requires that a Hilbert space
is is complete with respect to this metric. This means that if the metric recognizes a sequence of
vectors vn as convergent, then such sequence has a limit in the space. More precisely, if the distance
|vm − vn | of vm and vn is small enough for large enough m and n, then there is a vector v in the
space such that the sequence vn converges to v.
One can also consider the coefficients to be complex instead of real numbers. In this case, the
definition is the same except that the requirement that the inner product is symmetric becomes
v · w = w · v where a + ib is the complex-conjugate a − ib.
The space Rn with the dot product is an example of a Hilbert space. For a more R b exotic example,
2
consider a “vector” to be a function f (t) defined on interval [a, b] and such that a f (t) dt is defined
(such a function is said to be square-integrable on [a, b]) and consider the inner product of two
square-integrable functions f and g to be
Z b
f ·g = f (t)g(t)dt.
a

You can check that this product satisfies the three properties defining an inner product above.
Riemannian manifold and Theorema Egregium. Going back to manifolds, if the tangent
space TP M at every point P of an n-manifold M is equipped with an inner product, we say that M
is a Riemannian manifold and the inner product is called a Riemannian metric. In this case,
the coefficients of the first fundamental form at any point can be defined as follows.
∂ ∂
gij = i
·
∂u ∂uj
If g ij denotes the matrix inverse to the matrix gij , then we can define Christoffel Symbols using
the same formula which holds for surfaces
 
k 1 ∂gil ∂gjl ∂gij
Γij = + − g lk .
2 ∂uj ∂ui ∂ul
With the inner product around, one can talk about the length between vectors and, hence, define
the concepts of a unit-speed curve on a Riemannian n-manifold. With Christoffel symbols around
also, we can also generalize the concept of a geodesic as follows. A curve γ on manifold M is
geodesic if in each coordinate system defined along γ the equation (uk )00 + Γkij (ui )0 (uj )0 = 0 holds for
k = 1, 2, . . . n. With this definition, every point P and every tangent vector v uniquely determine
a geodesic γ with γ(0) = P and γ 0 (0) = v. In addition, a curve with the shortest possible length
between two points is necessarily a geodesic connecting these points.
The Riemann curvature tensor can be defined via Christoffel symbols, using the same formula
which holds for surfaces
∂Γlik ∂Γlij
l
Rijk = j
− k
+ Γpik Γlpj − Γpij Γlpk
∂u ∂u

24
and the sectional curvature K, generalization of the Gaussian, at every point P of a manifold M
can be defined as
i
g1i R212
K= .
g
Since it can be expressed using the first fundamental form only, it can be computed completely
intrinsically – without any reference to an embedding of the manifold into any external space. It is
possible to define other types of curvatures: Ricci curvature and scalar curvature. These concepts
are used in physics, especially in relativity theory.
Besides generalized Theorema Egregium, there are others fascinating results in differential geom-
etry. We mention some of them below.

1. If the coefficients of the Riemann curvature tensor are equal to zero, then the n-manifold is
locally isometric to Rn . This generalizes the statement that if a surface has zero Gaussian, then
it is locally isometric to a plane.

2. If a connected and complete Riemannian manifold of even dimension has constant sectional
curvature a12 , then it is either a 2n-sphere of radius a or a projective space.

3. Since the sectional curvature corresponds to the Gaussian for 2-manifolds, the following general-
izes the description of the surfaces of revolution with constant Gaussian curvature. A complete,
connected and simply connected (every closed curve can be collapsed to a point) Riemannian
manifold of constant sectional curvature c is

• the sphere S n of radius √1

c
if c > 0,
• the space Rn if c = 0, and
• a hyperbolic space if c < 0. The hyperbolic space H n is the set of vectors in Rn of length
4δij
smaller than 1 with the metric coefficients gij given at point v by gij (v) = −c(1−|v| 2 )2 .

4. Poincaré conjecture (recently proven, see Wikipedia for more details): Every simply connected,
closed 3-manifold is homeomorphic to the 3-sphere. Informally phrased, this means that if
a 3-manifold “behaves” like a sphere in certain sense, then such manifold is topologically
indistinguishable from a 3-sphere.

25