Part III: Riemannian Geometry

2012 Third International Summer School on Geometric Methods

in Robotics, Mechanism Design and Manufacturing Research

Harald Löwe

TU Braunschweig, Institute Computational Mathematics

July 2012, Shanghai Jiao Tong University, China

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 1 / 61

 Part III: Riemannian Geometry The First Fundamental Form

General Assumptions

Throughout this part let

1 M ⊆ Rn a k–dimensional manifold,
2 p ∈ M a point,
3 x : U → Rk a chart with p ∈ U,
4 ϕ = x −1 the corresponding local parameterization.
Recall that we identify

∂ ∂x −1
and .
∂xi p ∂ui x(p)

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 2 / 61

 Part III: Riemannian Geometry The First Fundamental Form

The Problem

Consider a surface M ⊆ R3 .
warped torus in R3

Question
What does a flat mathematician living on M knows about M?

1 Can he measure distances and angles? (Indeed he can!)

2 Can he decide whether M is a sphere, a torus, a cylinder, or a plane?

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 3 / 61

 Part III: Riemannian Geometry The First Fundamental Form

First Fundamental Form

For (p, v ), (p, w ) ∈ Tp M we put

I ((p, v ), (p, w )) := hv | w i (euclidian scalar product)

Definition
I is called first fundamental form on M.
The intrinsic geometry of M is the geometry of I .

Mostly, we shall write v instead of (p, v ) for elements of Tp M.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 4 / 61

 Part III: Riemannian Geometry The First Fundamental Form

Example 1: The Sphere

Given two curves γ(t) and δ(t) on the sphere:

The angle between these curves at the intersection point γ(0) = δ(0) is
given by
I (γ̇(0), δ̇(0))
cos (∠(γ, δ)) = p q .
I (γ̇(0), γ̇(0)) · I (δ̇(0), δ̇(0))
Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 5 / 61
 Part III: Riemannian Geometry The First Fundamental Form

I given by a local parameterization

For i, j = 1, . . . , k we put
* +
∂x −1 ∂x −1
gij (p) := .
∂ui x(p) ∂uj x(p)

By identification, this means that

* +
∂ ∂
gij (p) := .
∂xi p ∂xj p
Observe that
k k k
X
i j
X ∂
i
X ∂
I (v , w ) = v gij (p)w for v = v and w = wj
∂xi p ∂xj p
i,j=1 i=1 j=1

Thus, the scalar product is given by the matrix Gxp = (gij (p))i,j :
I (v , w ) = v T · Gxp · w .
Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 6 / 61
 Part III: Riemannian Geometry The First Fundamental Form

Example 2: Polar Coordinates

x: chart of euclidian coordinates on R2 (i.e. ϕ = id)

y : chart of polar coordinates, i.e. y (p) = (|p|, Im(ln(p1 + ip2 )))
y −1 (ρ, α) = (ρ cos(α), ρ sin(α)).
For x, we obtain G = 1. For y , we have
∂y −1 ∂ p
= (cos(α), sin(α)) ⇒ =
∂ρ ∂y1 p |p|
−1
∂y ∂
= (−ρ sin(α), ρ cos(α)) ⇒ = (−p2, p1)
∂α ∂y2 p
   
1 0 1 0
G(p) = =
0 ρ2 0 |p|2

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 7 / 61

 Part III: Riemannian Geometry The First Fundamental Form

Changing the Chart

Let y be a second chart around p, and let

!
∂yi
J := (Jacobi matrix of the overlap map)
∂xj p
(i,j)

For v , w ∈ Tp M, we have that

I (v , w ) = vyT Gyp wy = (Jvx )T Gyp (Jwx ) = vxT (J T Gyp J)wx = vxT Gxp wx ,

whence !T !
∂yi ∂yi
Gxp = · Gyp · .
∂xj p ∂xj p

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 8 / 61

 Part III: Riemannian Geometry The First Fundamental Form

Example 2 (cont.): Polar Coordinates

In the case of polar coordinates y :

p1
! !
∂xi |p| −p2
J= = p2 .
∂yj p |p| p1
(i,j)

Therefore, we have
p1
!T  p1
!
−p2 −p2

|p| 1 |p|
Gyp = J T Gxp J = p2 · · p2
|p| p1 1 |p| p1
p1
!
p1 p2
−p2
   
|p| |p| |p| 1
= · p2 =
−p2 p1 |p| p1 |p|2

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 9 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Riemannian Metrics

We may use a Riemannian metric G on M instead of I :

Let M ⊆ Rn be a manifold. Consider TM as subset of T Rn = Rn × Rn .
Let G : M → Rn×n be a smooth map
such that Gp := G(p) is a positive definite symmetric matrix for every p.
For v , w ∈ Tp M ≤ {p} × Rn put

Gp (v , w ) := v T Gp w .

Then (M, G) is called a Riemannian manifold with Riemannian metric G.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 10 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Remarks:

1 A Riemannian metric is a “smooth choice” of a scalar product on every

Tp M.
2 For Gp ≡ 1, we obtain the first fundamental form.
3 There is also an abstract version of Riemannian manifolds (smooth
section in a “vector bundle” over M).

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 11 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Example 3: Ellipsoids

Choose numbers a, b, c > 0. For p ∈ S2 , we let

 2 
a
Gp :=  b2 
c 2

Then (S2 , G) is a Riemannian manifold.

Claim: There is an isometry between (S2 , G) and the ellipsoid

x2 y2 z2
M: + 2 + 2 = 1,
a2 b c
with its first fundamental form IM , i.e. a diffeomorphism f : S2 → M with

IM (Tp f (v ), Tp f (w )) = v T Gp w for all v , w ∈ Tp S2 ≤ {p} × R3 .

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 12 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Example 3 (cont.)

If f = F |S2 for some smooth function F : R3 → R3 , then

IM (Tp f (v ), Tp f (w )) = (Jp F (v ))T · (Jp F (w )) = v T ((Jp F )T · (Jp F ))w .

Thus, search a diffeomorphism F with

a2
 

(Jp F )T · (Jp F ) = Gp =  b2 ,
c2

and F (S2 ) = M, e.g.

F (x, y , z) = (ax, by , cz)

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 13 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Two Points Of View

F
←→
G I

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 14 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Example 4: The Sphere – Once Again

For p ∈ S2 put

1 + x2
 

Gp =  1 + y2 
1+ z2

Consider the following diffeomorphism on R√3 :

 √ √ 
1+x 2 arsinh(y )+y 1+y 2 1+z 2
F (x, y , z) := arsinh(x)+x
2 2
arsinh( z)+z
2

Then (Jp F )T (Jp F ) = Gp for every p, i.e. (S2 , G) looks like (F (S2 ), I ).

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 15 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Example 4 (cont.)

How a flat mathematician would see (S2 , G)

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 16 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Isometries

(M, GM ), (N, GN ): Riemannian manifolds

An isometry is a diffeomorphism F : (M, GM ) → (N, GN ) such that

GN (Tp F (v ), Tp F (w )) = GM (v , w ) for all p ∈ M, v , w ∈ Tp M.

Note: F : R → S1 ; t 7→ (cos(t), sin(t)) is a “local isometry”, but not an

isometry, because F is not bijective.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 17 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Example 5: Isometries of R2

Claim
If F : (R2 , I ) → (R2 , I ) is an isometry, and if det Jp F > 0 holds for at least
one p, then

F (x) = A · x + b with A ∈ SO(2) and b ∈ R2 ,

i.e. F is an orientation–preserving motion of R2 .

The map R2 → R \ {0}; p 7→ det Jp F is continuous.

Follows: det Jp F > 0 holds for all p.
Moreover, (Jp F )T (Jp F ) = 1, whence Jp F ∈ SO(2):
 
cos(α(x, y )) − sin(α(x, y ))
J(x,y ) F = , where α : R2 → R is smooth.
sin(α(x, y )) cos(α(x, y ))

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 18 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Example 5 (cont.)

With F = (F 1 , F 2 ) we have
 1
Fx Fy1
  
cos(α(x, y )) − sin(α(x, y ))
=
Fx2 Fy2 sin(α(x, y )) cos(α(x, y ))

By the theorem of Hermann Amandus Schwarz

1 1
−αy sin(α) = Fxy = Fyx = −αx cos(α)
2 2
αy cos(α) = Fxy = Fyx = −αx sin(α)

This leads to αx = αy = 0, i.e. α ≡ a. We infer that

    
cos(a) − sin(a) x b
F (x, y ) = + .
sin(a) cos(a) y c

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 19 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

G in local coordinates

Given a chart x of (M, G), compute

!T
∂x −1 ∂x −1
(Gxp )i,j := · Gp · .
∂ui x(p) ∂uj x(p)

With Gxp = ((Gxp )i,j )(i,j) ∈ Rk×k we obtain

 
k k
X ∂ X ∂
Gp  vi , wi  = v T · Gxp · w .
∂xi p ∂xj p
i=1 j=1

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 20 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Example 6: Kinetic Energy

S1l : r12 + r22 = l 2 (circle with radius l ).

Consider the configuration manifold M = S1l × R2 of the barbell.
The kinetic energy was given by

Ekin (s, r , ṡ, ṙ ) = m · (|ṙ |2 + |ṡ|2 )

We turn M into a Riemannian manifold by putting

G(s,r ) = m · 1.

For v = (s, r , ṡ, ṙ ) ∈ TM:

Ekin (v ) = G(v , v ).

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 21 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Example 6 (cont.)

Parameterization of M (length of rod: 2l ):

ϕ(q1 , q2 , q3 ) := ((l cos(q1 ), l sin(q1 )), (q2 , q3 )).

Kinetic energy:

∂ϕ ∂ϕ ∂ϕ
v = q̇1 + q̇2 + q̇3 ⇒ Gq (v , v ) = m (l 2 q̇12 + q̇22 + q̇32 )
∂q1 q ∂q2 q ∂q3 q

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 22 / 61

 Part III: Riemannian Geometry Riemannian Manifolds

Lenghts of Curves

(M, G) Riemannian manifold, γ : [t0 , t1 ] → M curve.

Lenght of γ:
Zt1 q
L(γ) := Gγ(t) (γ̇(t), γ̇(t)) dt.
t0

The infimum of all lengths of all curves joining p and q is called distance
d (p, q) between p and q.
(M, d ) is a metric space.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 23 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Derivatives of Vector Fields

M ⊆ Rn : k–dim. manifold, X vector field on M.

Y : M → Rn smooth map (a “vector field”, possibly not tangent to M).
Note that Y = (Y1 , . . . , Yn ), where Yj ∈ C ∞ (M).
DX Y : M → Rn is defined by

DX Y (p) := (X (Y1 )(p), . . . , X (Yn )(p))

In local coordinates:
k k
!
X ∂Y1 X ∂Yn
DP k ∂
X i ∂x
Y (p) = X i (p) , ..., X i (p)
i =1 i ∂xi p ∂xi p
i=1 i=1

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 24 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Covariant Derivatives

D(M): set of tangent vector fields on M.

Notice that X , Y ∈ D(M) 6⇒ DX Y ∈ D(M)!
Decompose DX Y into

DX Y = ∇X Y + hX Y , where ∇X Y ∈ D(M) and hX Y ⊥M.

(“⊥” means “perpendicular w.r.t. the usual scalar product”.)

Definition
∇X Y is called the covariant derivative of Y in direction of X .

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 25 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Computation Of ∇

Let x be a chart of M:
k
X ∂
X = Xi : operator on C ∞ (M)
∂xi
i=1
k k
X ∂ j
X ∂x −1
Y = Y : function p 7→ Y j (p)
∂xj ∂uj x(p)
j=1 j=1
 
k k k X k  
i ∂  j ∂  i ∂ j ∂
X X X
DX Y = X Y = X Y
∂xi ∂xj ∂xi ∂xj
i=1 j=1 i=1 j=1 | {z }
interesting part

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 26 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Computation Of ∇ (cont.)

!
∂x −1
 
∂ j ∂ ∂ j
Y (p0 ) = p 7→ Y (p)
∂xi ∂xj ∂xi p0 ∂uj x(p)
∂x −1
 
∂ j −1
= u 7→ Y (x (u))
∂ui x(p0 ) ∂uj u
∂(Y j ◦ x −1 ) ∂x −1 ∂ 2 x −1
= + Y j (p0 )
∂ui x(p0 ) ∂uj x(p0 ) ∂ui ∂uj x(p0 )
∂Y j ∂ ∂
= +Y j (p0 ) D ∂ (p0 )
∂xi p0 ∂xj
p0
∂xi ∂xj
| {z } | {z }
tangent to M interesting part

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 27 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Christoffel Symbols
We have that
k k 
∂Y j ∂

X
i
X
j ∂
DX Y = X · +Y ·D ∂
∂xi ∂xj ∂xi ∂xj
i=1 j=1
k k 
∂Y j ∂

X
i
X
j ∂
∇X Y = X · +Y ·∇ ∂
∂xi ∂xj ∂xi ∂xj
i=1 j=1

Definition
Let
k
∂ X ∂
∇ ∂ = Γrij .
∂xi ∂xj ∂xr
r =1

Then the functions Γrij are called Christoffel symbols.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 28 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Computation Of The Christoffel Symbols

∂ ∂
Recall that ∇ ∂
∂xj is the tangent part of D ∂
∂xj , i.e.
∂xi ∂xi

 
∂ ∂ ∂
I ,D∂ −∇ ∂ = 0 for all s.
∂xs ∂xi ∂xj ∂xi ∂xj

∂ Pk r ∂
With ∇ ∂
∂xj = r =1 Γij ∂xr we get
∂xi

  k  
∂ ∂ X ∂ ∂
I ,D∂ = Γrij ·I , .
∂xs ∂xi ∂xj ∂xs ∂xr
r =1 | {z }
=grs

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 29 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Computation Of The Christoffel Symbols (cont.)

We compute (ϕ := x −1 ):
!
∂ϕ T ∂ϕ
 
∂gβγ ∂
= ·
∂uα ∂uα ∂uβ ∂uγ
T
∂ϕ T
 2
∂2ϕ
 
∂ ϕ ∂ϕ
= · + ·
∂uα ∂uβ ∂uγ ∂uβ ∂uα ∂uγ

This implies that

T
∂2ϕ
    
∂ ∂ ∂ϕ 1 ∂gsj ∂gsi ∂gij
I ,D ∂ = · = · + −
∂xs ∂xi ∂xj ∂us ∂ui ∂uj 2 ∂ui ∂uj ∂us

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 30 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Computation Of The Christoffel Symbols (cont.)

Putting
(g rs )rs := (grs )−1
rs ,

we eventually obtain that

Christoffel Symbols In Local Coordinates

k  
X 1 ∂gsj ∂gsi ∂gij
Γrij = g rs
+ − .
2 ∂ui ∂uj ∂us
s=1

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 31 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Example 7: The Sphere

Consider the following parameterization of S2 :

ϕ(u1 , u2 ) = (cos(u1 ) cos(u2 ), cos(u1 ) sin(u2 ), sin(u1 ))

∂ϕ
ϕ1 (u1 , u2 ) = = (− sin(u1 ) cos(u1 ), − sin(u1 ) sin(u2 ), cos(u1 ))
∂u1
∂ϕ
ϕ2 (u1 , u2 ) = = (− cos(u1 ) sin(u2 ), cos(u1 ) cos(u2 ), 0)
∂u2
ϕ1 × ϕ2
ν(u1 , u2 ) = − = ϕ(u1 , u2 )
|ϕ1 × ϕ2 |
ν is the “normalfield” of S2
1 0
G = 2
0 cos (u1 )

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 32 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Example 7: The Sphere (cont.)

Compute

Γ111 = Γ211 = 0
∂2ϕ
= 0 · ϕ1 + 0 · ϕ2 − ν
∂u12
Γ112 = 0, Γ212 = − tan(u1 )
∂2ϕ
= 0 · ϕ1 − tan(u1 ) · ϕ2 + 0 · ν
∂u1 ∂u2
Γ122 = cos(u1 ) sin(u1 ), Γ222 = 0
∂2ϕ
= cos(u1 ) sin(u1 ) · ϕ1 + 0 · ϕ2 − cos2 (u1 ) · ν
∂u22

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 33 / 61

 Part III: Riemannian Geometry Covariant Derivatives

The Covariant Derivative For Riemannian Manifolds

Let (M, G) be a Riemannian manifold. If ϕ is a local parameterization,

then let  
∂ϕ ∂ϕ
gij = Gϕ(u) , .
∂ui u ∂uj u
Put
k  
X 1 ∂gsj ∂gsi ∂gij
Γrij = g rs
+ − .
2 ∂ui ∂uj ∂us
s=1

If X , Y are vector fields on M, then let

 
k k r k
X X ∂Y X ∂
∇X Y :=  Xi + X i Y j Γrij 
∂xi ∂xr
r =1 i=1 i,j=1

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 34 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Covariant Derivatives Along Curves

Let γ be a curve on M. Locally, γ(t) = x −1 (ρ1 (t), . . . , ρk (t)).

Suppose that Y is a curve in TM with Y (t) ∈ Tγ(t) M for all t,
i.e. Y “is a vector field along γ”.
 
k k
!
X ∂ X ∂
γ̇(t) = γ(t), ρ̇i (t) and Y (t) = γ(t), Y j (t) 
∂xi γ(t) ∂xj γ(t)
i=1 j=1

Locally, we may define

 
k k
X
Ẏ r (t) +
X ∂
∇γ̇ Y (t) := Γrij (γ(t))ρ̇i (t)Y j (t)
∂xr γ(t)
r =1 i,j=1

∇γ̇ Y is called covariant derivative of Y along γ.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 35 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Parallel Transport And Geodesics

Y is called parallel along γ iff ∇γ̇ Y ≡ 0, i.e.

k
X
r
Ẏ + Γrij ρ̇i Y j = 0 for r = 1, . . . , k.
i,j=1

A curve γ is called a geodesic iff γ̇ is parallel along γ, i.e.

k
X
ρ̈r + Γrij ρ̇i ρ̇j = 0 for r = 1, . . . , k.
i,j=1

Indeed, a geodesic is the “shortest path” between points – if it exists.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 36 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Example: Parallel Transport

The example shows the parallel transport on a piecewise geodesic curve.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 37 / 61
 Part III: Riemannian Geometry Covariant Derivatives

Useful Facts Concerning Geodesics

d
1 If γ is a geodesic, then dt Gγ(t) (γ̇(t), γ̇(t)) ≡ 0.
2 For p ∈ M and v ∈ Tp M, there exists a geodesic γ with γ(0) = p and
γ̇(0) = v . Locally at t = 0, γ and every other geodesic γ ∗ having
these initial values coincide.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 38 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Example 8: The Euclidian Space

Take Rn with its usual scalar product and the obvious chart.
Observe that Γrij ≡ 0 for all i, j, r ,
Thus, Y is parallel along γ iff Ẏ r ≡ 0, i.e. iff Y is a “constant” vector field.
This fact should explain the name “parallel”.

Moreover, γ is a geodesic iff γ̈ ≡ 0, i.e. if γ(t) = a + tb is a line.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 39 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Example 7 (cont.): The Sphere

ϕ(u, v ) = (cos(u) cos(v ), cos(u) sin(v ), sin(u))

Γvuv = − tan(u) = Γvvu
Γuvv = cos(u) sin(u)

Equations for a geodesics

0 = ü + cos(u) sin(u)v̇ 2
0 = v̈ − 2 tan(u)u̇ v̇ .

Difficult? Yes!

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 40 / 61

 Part III: Riemannian Geometry Covariant Derivatives

What is ∇γ̇ γ̇?

Let M ⊆ Rn be a k–dimensional manifold endowed with its first

fundamental form I . Decompose v ∈ Tp Rn into v = v > + v ⊥ , where
v > ∈ Tp M and v ⊥ ⊥M.
Let γ(t) = ϕ(ρ1 (t), . . . , ρk (t)) be a curve in M. Observe that
k
X ∂ϕ
γ̇(t) = ρ̇r (t)
∂ur ρ(t)
r =1
k k
X ∂ϕ X ∂2ϕ
γ̈(t) = ρ̈r (t) + ρ̇i (t)ρ̇j (t)
ρ(t)∂ur ∂ui ∂uj ρ(t)
r =1 i,j=1
 
k k
X X ∂ϕ
γ̈(t)> = ρ̈r (t) + Γrij (ρ(t))ρ̇i (t)ρ̇j (t) = ∇γ̇ γ̇(t).
∂ur ρ(t)
r =1 i,j=1

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 41 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Example 7 (cont.): The Sphere

Suppose that γ is a geodesic on S2 . Then

0 = ∇γ̇ γ̇ = γ̈ >
= γ̈ − I (γ̈, γ) · γ.
γ̈ = I (γ̈, γ) · γ.

Recall that a geodesic is uniquely determined by its initial conditions.

Geodesics On The Sphere

If p ∈ S2 and v ∈ Tp S2 , then
v
γ(t) = cos(|v | t) · p + sin(|v | t) ·
|v |

is the geodesic with γ(0) = 0 and γ̇(0) = v

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 42 / 61

 Part III: Riemannian Geometry Covariant Derivatives

Example 6 (cont.): The Barbell

Q = S1 × R2 configuration manifold of the barbell.

Kinetic energy by a Riemannian metric: Gp = m · 1.
In absence of forces: If γ describes the motion of a barbell, then γ̈ > = 0,
i.e. γ is a geodesic curve.
In generalized coordinates ϕ(q1 , q2 , q3 ) = (l cos(q1 ), l sin(q1 ), q2 , q3 ):

Gq1 ,q2 ,q3 ((q̇1 , q̇2 , q̇3 ), (q̇1 , q̇2 , q̇3 )) = m (l 2 q̇12 + q̇22 + q̇32 )

Observe that all Γrij vanish, whence the geodesics are lines.

Motions: γ(t) = (α0 + t α̇0 , ξ0 + t ξ˙0 , η0 + t η̇0 ).

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 43 / 61

 Part III: Riemannian Geometry Covariant Derivatives

The Energy Functional

(M, G) Riemannian manifold; γ : [a, b] → M curve on M.

Energy functional:

Zb
1
E (γ) := G(γ̇(t), γ̇(t)) dt.
2
a

(Proper) variation of γ: smooth map Φ : (−ε, ε) × [a, b] → M with

Φ(0, t) = γ(t) and Φ(s, a) = γ(a), Φ(s, b) = γ(b).
γ is a critical point of E iff, for every variation Φ of γ,

d
E (t 7→ Φ(t, s)) = 0.
ds s=0

This in turn holds iff γ is a geodesic curve.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 44 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

Orientable Manifolds

A n–dimensional submanifold M ⊆ Rn+1 is called a hypersurface.

M is orientable iff
M can be covered by charts (Mα , xα ) such that

det Jx(p) xβ ◦ xα−1 > 0 for all α, β.

Theorem
A hypersurface is orientable iff it admits a smooth normal field,
i.e. there is a smooth map N : M → Sn ⊆ Rn+1 such that N(p) is
perpendicular to M at p.
N is called the Gauss map of M.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 45 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

Remarks

1 N is unique up to its sign.

2 If M is the image of a regular parameterization ϕ : U → M,
then M is orientable:
For p ∈ M, there is a unique unit vector N(p) such that
 
∂ϕ ∂ϕ
det ∂u1 . . . ∂u n
N(ϕ(u)) > 0 for all u ∈ U.
u u

Then N(p) depends smoothly on p and is perpendicular to M at p.

3 Suppose that M is covered by two charts (V , x) and (V 0 , x 0 ).
Moreover, assume that V ∩ V 0 is connected. Then M is orientable.
4 Sn ⊆ Rn+1 is orientable: put N(p) := p.
5 The Möbius strip is not orientable.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 46 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

The Weingarten Map

M ⊆ Rn+1 orientable hypersurface, N the Gauss map.

We consider the derivate TN : TM → T Sn .
If (p, v ) ∈ Tp M, then Tp N(p, v ) = (N(p), w ) ∈ TN(p) Sn ,
whence w ⊥N(p), i.e (p, w ) ∈ Tp M.
Putting Lp (p, v ) := (p, −w ) yields a linear map Lp : Tp M → Tp M.
Lp is called the Weingarten map of M at p.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 47 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

Example 9: The Torus

The Gauss map of the torus is given in local coordinates by

N(x −1 (u, v )) = N((3 + cos(v )) cos(u), (3 + cos(v )) sin(u), sin(v ))

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

We shall compute Tp N : Tp T → Tp R3 . First, notice that

∂x −1
= (−(3 + cos(v )) sin(u), (3 + cos(v )) cos(u), 0)
∂u
∂x −1
= (− sin(v ) cos(u), − sin(v ) sin(u), cos(v ))
∂v
∂(N ◦ x −1 ) cos(v ) ∂x −1
= (− cos(v ) sin(u), cos(v ) cos(u), 0) =
∂u 3 + cos(v ) ∂u
∂(N ◦ x −1 ) ∂x −1
= (− sin(v ) cos(u), − sin(v ) sin(u), cos(v )) =
∂v ∂v

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 48 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

Example 9 (cont.)

For p = ((3 + cos(v )) cos(u), (3 + cos(v )) sin(u), sin(v )) we infer that

!
∂ cos(v ) ∂
Tp N =
∂xu p 3 + cos(v ) ∂xu p
!
∂ ∂
Tp N = .
∂xv p ∂xv p

∂ ∂
Thus, with respect to the basis ∂xu p , ∂xv p of Tp T, the Weingarten map
Lp is given by
! !
cos(v ) 3
− 3+cos(v ) p12 +p22
−1
Lp = =
−1 −1

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 49 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

The Second Fundamental Form

Theorem: The Weingarten map Lp : Tp M → Tp M is a self–adjoint map,

i.e.
hLp (v ) | w i = hv | Lp (w )i for all v , w ∈ Tp M.
The symmetric bilinear form

IIp (v , w ) := hLp (v ) | w i

on Tp M is called second fundamental form of M.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 50 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

Normal Curvature

Let γ be a curve on M with constant velocity |γ̇(t)| ≡ ν0 . Put p = γ(0).

Differentiating 0 = hγ̇(t) | N(γ(t))i yields

II (γ̇(0), γ̇(0)) = hLγ(0) γ̇(0) | γ̇(0)i

 
d
= − N(γ(t)) γ̇(0)
dt t=0
= hN(γ(0)) | γ̈(0)i = hN(p) | γ̈(0)i

The normal curvature of γ in p is given by

   
γ̈(0) γ̇(0) γ̇(0)
N(p) = II ,
ν02 ν0 ν0

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 51 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

Normal Curvature (cont.)

If v ∈ Tp M with I (v , v ) = 1,
then we call II (v , v ) the normal curvature of M in direction of v .
1 k1 , . . . , kn are called principal curvatures of M at p.
2 det Lp = k1 · . . . · kn is the Gaussian curvature of M in p.
1
3
2 tr Lp = 12 (k1 + . . . + kn ) is the mean curvature of M in p.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 52 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

Second Fundamental Form And Local Coordinates

∂(x −1 )
 
∂ −1
0 = N(x (u))
∂ui x(p) ∂uj u
* + * +
∂(N ◦ x −1 ) ∂(x −1 ) ∂ 2 x −1
= + N(x −1 (x(p)))
∂ui x(p) ∂uj x(p) ∂ui ∂uj x(p)
* ! + * +
∂ ∂ ∂ 2 x −1
= −Lp + N(p)
∂xi p ∂xj p ∂ui ∂uj x(p)

Therefore
! * +
∂ ∂ ∂ 2 x −1
hij := II , = N(p)
∂xi p ∂xj p ∂ui ∂uj x(p)

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 53 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

The Weingarten Map And Local Coordinates

Let (hij )ij be the matrix of the Weingarten map. Then

n
!
∂ ∂
hij
X
Lp = .
∂xi p ∂xj p
j=1

The coefficients hij are given by

n
hij :=
X
hik g kj ,
k=1

where (g kj ) is the inverse matrix of (gkj ).

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 54 / 61

 Part III: Riemannian Geometry The Second Fundamental Form

Example 9 (cont.): The Torus

Once again, we use the parameterization

ϕ(u1 , u2 ) = ((3 + cos(u2 )) cos(u1 ), (3 + cos(u2 )) sin(u1 ), sin(u2 )).

The normal vector field is given by

N(u, v ) = (cos(u2 ) cos(u1 ), cos(u2 ) sin(u1 ), sin(u2 )).

The matrix of the first fundamental form equals

(3 + cos(u2 ))2
   
∂ϕ ∂ϕ
(gij )ij = =
∂ui ∂uj ij
1

Moreover, the matrix of the second fundamental form equals

∂2ϕ
   
− cos(u2 ) (3 + cos(u2 ))
(hij )ij = N = .
∂ui ∂uj ij
−1

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 55 / 61

 Part III: Riemannian Geometry Local Theory

Equations Of Gauss And Weingarten

Let U ⊆ Rn be open and let g , h : U → sym(n) be smooth maps such that

g (u) is positive definite for all u. Put
n  
X 1 ∂gsj ∂gsi ∂gij
Γrij := g rs
+ − and (g ij (u))ij := g (u)−1 .
2 ∂ui ∂uj ∂us
s=1

If g , h belong to a hypersurface ϕ : U → Rn+1 with normal field N, then

n
∂2ϕ X ∂ϕ
= Γkij + hij · N (Gauss)
∂ui ∂uj ∂uk
k=1
n
∂N X
= − hij g jk . (Weingarten)
∂ui
j,k=1

Indeed, there are integrability conditions for these equations. . .

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 56 / 61

 Part III: Riemannian Geometry Local Theory

Conditions Of Gauss And Codazzi–Mainardi

For all i, j, k, s we have that

n
∂Γsij ∂Γsik X r s
Γij Γrk − Γrik Γsrj


− +
∂uk ∂uj
r =1
n
X
= (hij hkm − hik hjm ) g ms
m=1
(Condition of Gauss)

n
∂hij ∂hik X r
Γij hrk − Γrik hrj


0 = − +
∂uk ∂uj
r =1
(Condition of Codazzi–Mainardi)

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 57 / 61

 Part III: Riemannian Geometry Local Theory

Theorem Of O. Bonnet

Suppose that g , h : U → sym(n) (where g (u) always is positive definite)

satisfy the conditions of Gauss and of Codazzi–Mainardi.
Let u0 ∈ U, p0 ∈ Rn+1 , v1 , . . . , vn ∈ Tp Rn+1 with hvi | vj i = gij (u0 ) and
N0 ∈ Tp Rn+1 , |N0 | = 1, be perpendicular to all vi .
Then there exists a connected open subset U 0 ⊆ U with u0 ∈ U 0 and
precisely on parameterized hypersurface ϕ : U 0 → Rn+1 with Gauss map N
such that
∂ϕ
1 f (u0 ) = p0 . ∂ui u = vi , N(u0 ) = N0 , and
0
2 g (u) and h(u) are the first and second fundamental form of ϕ.
Without the initial conditions, ϕ is unique up to a motion of Rn+1 .

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 58 / 61

 Part III: Riemannian Geometry Local Theory

A Strange Corollary

Let n ≥ 3 and let ϕ : U → Rn+1 be a hypersurface.

Suppose that the Weingarten map Lp has rank ≥ 3 on a dense subset.
Then ϕ is rigid, i.e. every isometric deformation of M is a motion.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 59 / 61

 Part III: Riemannian Geometry Two Theorems

Two Theorems

Question
Are there abstractly defined manifolds / Riemannian manifolds which are
not submanifolds of some Rn ?

Theorem (Whitney, 1936)

Every smooth (abstractly defined) manifold admits a smooth embedding
into some Rn .

Theorem (Nash, 1956)

Every (abstractly defined) Riemannian manifold admits a metric–preserving
embedding into some Rn .

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 60 / 61

 Part III: Riemannian Geometry Two Theorems

Do We Need Abstract Manifolds?

Projective n–space: set RP n of all 1–dim. linear subspaces R v , v 6= 0, of

Rn+1 . Consider the bijective map

2
ϕ(R v ) := p p T − 1 ∈ sym(n + 1)
|p|2

Then ϕ(RP n ) is a Riemannian manifold when endowed with

1
G(A, B) := tr(AT B) for A, B ∈ sym(n + 1).
8
Neither the geometric structure nor the differential geometry of RP n is
sufficiently clear in this picture.

Löwe (TU Braunschweig) Part III: Riemannian Geometry July 2012 61 / 61