Durham University Michaelmas 2014

Pavel Tumarkin

Riemannian Geometry IV, Term 1 (Sections 3–4)

3 Riemannian metric
Definition 3.1. Let M be a smooth manifold. A Riemannian metric gp (·, ·) or h·, ·ip is a family of real
inner products gp : Tp M × Tp M → R depending smoothly on p ∈ M . A smooth manifold M with a
Riemannian metric g is called a Riemannian manifold (M, g).
Examples 3.2–3.3. Euclidean metric on Rn , induced metric on M ⊂ Rn .
Definition 3.4. Let (M, g) be a Riemannian manifold. For v ∈ Tp M define the length of v by 0 ≤ kvkg =
p
gp (v, v).
Example 3.5. Three models of hyperbolic geometry:
model notation M g
{y ∈ Rn+1 | q(y,Py) = −1, yn+1 > 0}
Hyperboloid Wn gx (v, w) = q(v, w)
where q(x, y) = ni=1 xi yi − xn+1 yn+1
n
4
Bn {x ∈ Rn | kxk2 = x2i < 1}
P
Poincaré ball gx (v, w) = (1−kxk2 )2
hv, wi
i=1
1
Upper half-space Hn {x ∈ Rn | xn > 0} gx (v, w) = x2n
hv, wi

Definition 3.6. Given two vector spaces V1 , V2 with real inner products (Vi , h·, ·ii ), an isomorphism
T : V1 → V2 of vector spaces is a linear isometry if hT v, T wi2 = hv, wi1 for all v, w ∈ V1 .
This is equivalent to preserving the lengths of all vectors (since hv, wi = 21 (hv + w, v + wi − hv, vi −
hw, wi)).
Definition 3.7. A diffeomorphism f : (M, g) → (N, h) of two Riemannian manifolds is an isometry if
Df (p) : Tp M → Tf (p) N is a linear isometry for all p ∈ M .
Theorem 3.8 (Nash embedding theorem). For any Riemannian manifold (M m , g) the exists an isometric
embedding into Rk for some k ∈ N. If M is compact, there exists such k ≤ m(3m+1)
2 , and if M is not
m(m+1)(3m+1)
compact, there is such k ≤ 2 .
Definition 3.9. (M, g) is a Riemannian manifold, c : [a, b] → M is a smooth curve. The length L(c) of c
Rb 1/2
is defined by L(c) = a kc0 (t)kdt, where kc0 (t)k = hc0 (t), c0 (t)ic(t) . The length of a piecewise-smooth curve
is defined as the sum of lengths of its smooth pieces.
Theorem 3.10 (Reparametrization). Let ϕ : [c, d] → [a, b] be a strictly monotonic smooth function,
ϕ0 6= 0, and let γ : [a, b] → M be a smooth curve. Then for γ̃ = γ ◦ ϕ : [c, d] → M holds L(γ) = L(γ̃).
Definition 3.11. A smooth curve c : [a, b] → M is arc-length parametrized if kc0 (t)k ≡ 1.
Proposition 3.12 (evident). If a curve c : [a, b] → M is arc-length parametrized, then L(c) = b − a.
Proposition 3.13. Every curve has an arc-length parametrization.

1
Example 3.14. Length of vertical segments in H. Shortest paths between points on vertical rays.

Definition 3.15. Define a distance d : M × M → [0, ∞) on (M, g) by d(p, q) = inf γ {L(γ)}, where γ is a
piecewise smooth curve connecting p and q.

Remark. (M, d) is a metric space.

Example 3.16. Induced metric on S 1 ⊂ R2 .

Definition 3.17. If (M, g) is a Riemannian manifold, then any subset A ⊂ M is also a metric space
with the induced metric d|A×A : A × A → [0, ∞) defined by d(p, q) = inf γ {L(γ) | γ : [a, b] → A, γ(a) =
p, γ(b) = q}, where the length L(γ) is computed in M .
4
Example 3.18. Punctured Riemann sphere: Rn with metric gx (v, w) = (1+kxk2 )2
hv, wi.

4 Levi-Civita connection and parallel transport

4.1 Levi-Civita connection
∂
∈ X(Rn ) and a vector v ∈ Tp Rn define the
P
Example 4.1. Given a vector field X = ai (p) ∂x i

covariant derivative of X in direction v in Rn by ∇v (X) = lim X(p+tv)−X(p) ∂ 


∈ Tp Rn .
P
t = v(ai ) ∂x i p
t→0

Proposition 4.2. The covariant derivative ∇v X in Rn satisfies all the properties (a)–(e) listed below in
Definition 4.3 and Theorem 4.4.

Definition 4.3. Let M be a smooth manifold. A map ∇ : X(M ) × X(M ) → X(M ), (X, Y ) 7→ ∇X Y is
affine connection if for all X, Y, Z ∈ X(M ) and f, g ∈ C ∞ (M ) holds

(a) ∇X (Y + Z) = ∇X (Y ) + ∇X (Z)

(b) ∇X (f Y ) = X(f )Y (p) + f (p)∇X Y

(c) ∇f X+gY Z = f ∇X Z + g∇Y Z

Theorem 4.4 (Levi-Civita, Fundamental Theorem of Riemannian Geometry). Let (M, g) be a Rieman-
nian manifold. There exists a unique affine connection ∇ on M with the additional properties for all
X, Y, Z ∈ X(M ):

(d) v(hX, Y i) = h∇v X, Y i + hX, ∇v Y i (Riemannian property);

(e) ∇X Y − ∇Y X = [X, Y ] (∇ is torsion-free).

This connection is called Levi-Civita connection of (M, g).

Remark 4.5. Properties of Levi-Civita connection in Rn and in M ⊂ Rn with induced metric.

4.2 Christoffel symbols

Definition 4.6. Let ∇ be the Levi-Civita connection on (M, g), and let ϕ : U → V be a coordinate
chart with coordinates ϕ = (x1 , . . . , xn ). Since ∇ ∂ ∂x∂ j (p) ∈ Tp M , there exists a uniquely determined
∂xi

collection of functions Γkij ∈ C ∞ (U ) s.t. ∇ ∂ ∂x∂ j (p) = nk=1 Γkij (p) ∂x∂ k (p). These functions are called
P
∂xi
Christoffel symbols of ∇ with respect to the chart ϕ.

2
 n
∂b
bj ∂x∂ j = ai ∂xji ∂x∂ j + ai bj Γkij ∂x∂ k .
P P P
Remark. Christoffel symbols determine ∇ since ∇P
n
∂
ai ∂x j=1 i,j i,j,k
i=1 i

Proposition 4.7.
1 X km
Γkij = g (gim,j + gjm,i − gij,m ),
2 m

where gab,c = ∂x∂ c gab and (g ij ) = (gij )−1 , i.e. {g ij } are the elements of the matrix inverse to (gij ).
In particular, Γkij = Γkji .

Example 4.8. In Rn , Γkij ≡ 0 for all i, j, k. Computation of Γkij in S 2 ⊂ R3 with induced metric.

4.3 Parallel transport

Definition 4.9. Let c : (a, b) → M be a smooth curve. A smooth map X : (a, b) → T M with X(t) ∈
Tc(t) M is called a vector field along c. These fields form a vector space Xc (M ).

Example 4.10. c0 (t) ∈ Xc (M ).

Proposition 4.11. Let (M, g) be a Riemannian manifold, let ∇ be the Levi-Civita connection, c : (a, b) →
D
M be a smooth curve. There exists a unique map dt : Xc (M ) → Xc (M ) satisfying
D D D
(a) dt (αX + Y ) = α dt X+ dt Y for any α ∈ R.

(b) D
dt (f X) = f 0 (t)X + f dt
D
X for every f ∈ C ∞ (M ).
e ∈ X(M ) is a local extension of X
(c) If X 
(i.e. there exists t0 ∈ (a, b) and ε > 0 such that X(t) = X
e
c(t)
for all t ∈ (t0 − ε, t0 + ε))
D
then ( X)(t0 ) = ∇c0 (t ) X.
dt
e
0

D
This map dt : Xc (M ) → Xc (M ) is called the covariant derivative along the curve c.

Example 4.12. Covariant derivative in Rn .

D
Definition 4.13. Let X ∈ Xc (M ). If dt X = 0 then X is said to be parallel along c.

Example 4.14. A vector field X in Rn is parallel along a curve if and only if X is constant.

Theorem 4.15. Let c : [a, b] → M be a smooth curve, v ∈ Tc(a) M . There exists a unique vector field
X ∈ Xc (M ) parallel along c with X(a) = v.

Corollary 4.16. Parallel vector fields form a vector space of dimension n (where n is the dimension of
(M, g)).

Definition 4.17. Let c : [a, b] → M be a smooth curve. A linear map Pc : Tc(a) M → Tc(b) M defined by
Pc (v) = X(b), where X ∈ Xc (M ) is parallel along c with X(a) = v, is called a parallel transport along c.

Remark. The parallel transport Pc depends on the curve c (not only on its endpoints).

Proposition 4.18. The parallel transport Pc : Tc(a) M → Tc(b) M is a linear isometry between Tc(a) M and
Tc(b) M , i.e. gc(a) (v, w) = gc(b) (Pc v, Pc w).