Lecture Notes 12: Riemannian Metrics
Lecture Notes 12: Riemannian Metrics
Lecture Notes 12: Riemannian Metrics
Differential Geometry I
Fall 2006, Georgia Tech
Lecture Notes 12
Riemannian Metrics
0.1 Definition
If M is a smooth manifold then by a Riemannian metric g on M we mean a smooth
assignment of an innerproduct to each tangent space of M . This means that, for
each p ∈ M , gp : Tp M × Tp M → R is a symmetric, positive definite, bilinear map,
and furthermore the assignment p 7→ gp is smooth, i.e., for any smooth vector fields
X and Y on M , p 7→ gp (Xp , Yp ) is a smooth function. The pair (M, g) then will be
called a Riemannian manifold. We say that a diffeomorphism f : M → N between
a pair of Riemannian manifolds (M, g) and (N, h) is an isometry provided that
0.2 Examples
0.2.1 The Euclidean Space
The simplest example of a Riemannian manifold is Rn with its standard Euclidean
innerproduct, g(X, Y ) := hX, Y i.
1
0.2.3 Quotient of a Riemannian manifold by a group of isometries
Note that the set of isometries f : M → M forms a group. Another source of exam-
ples of Riemannian manifolds are generated by taking the quotient of a Riemannian
manifold (M, g) by a subgroup G of its isometries which acts properly discontinu-
ously on M . Recall that if G acts properly discontinuously, then M/G is indeed
a manifold. Then we may define a metric h on M/G by setting h[p] := gp . More
precisely recall that the projections π : M → M/G, given by π(p) := [p] is a local
diffeomorphism, i.e., for any q ∈ [p] there exists an open neighborhood U of p in M
and an open neighborhood V of [p] in M/G such that π : U → V is a diffeomorphism.
Then we may define
One can immediately check that h does not depend on the choice of q ∈ [p] and is
thus well defined.A specific example of proper discontinuous action of isometries is
given by translations fz : Rn → Rn given by fz (p) := p + z where z ∈ Zn . Recall
that Rn /Zn is the torus T n , which may now be equipped with the metric induced by
this group action. Similarly RPn admits a canonical metric, since RPn = Sn /{±1},
and reflections of a sphere are isometries.
Exercise 0.2.1. Show that the inversion i : Rn − {o} → Rn given by i(x) := x/kxk
is a conformal transformation.
hX, Y i
gp (X, Y ) := ,
(pn )2
2
where pn denotes the nth coordinate of p. Another description of the hyperbolic
space may be given by taking the open unit ball if Rn and defining
hX, Y i
gp (X, Y ) := .
(1 − kpk2 )2
This is known as Poincare’s ball model.
Exercise 0.2.3. Show that the the Poincare half-plane and the half-disk are iso-
metric (Hint: identify the Poincare half-plane with the region y > 1 in R2 and do
an inversion).
Thus in any local coordinate (U, φ) a metric is completely determined by the func-
tions gij which may be regarded as the coefficients of a positive definite matrix.
To obtain a concrete example, note that if M ⊂ Rn is a submanifold, with the
induce metric from Rn , and (φ, U ) is a local chart of M , then if we set f := φ−1 ,
f : φ(U ) → Rn is a parametrization for U , and d(f )(ei ) = Di f . Consequently,
gij (p) = hDi f (f −1 (p)), Dj f (f −1 (p))i.
For instance, note that a surface of revolution in R3 which is given by rotating the
curve (r(t), z(t)) in the xz-plane about the z axis can be parametrized by
f (t, θ) = (r(t) cos θ, r(t) sin θ, z(t)).
So
D1 f (t, θ) = (r0 (t) cos θ, r0 (t) sin θ, z 0 (t)) and D2 f (t, θ) = (−r(t) sin θ, r(t) cos θ, 0),
and consequently gij (f (t, θ)) is given by
0 2
(r ) + (z 0 )2 0
.
0 r2
Note that if we assume that the curve in the xz-plane is parametrized by arclength,
then (r0 )2 + (z 0 )2 = 1, so the above matrix becomes more simple to work with.
3
Exercise 0.3.1. Compute the metric of S2 in terms of spherical coordinates θ and
φ.
Exercise 0.3.2. Compute the metric of the surface given by the graph of a function
f : Ω ⊂ R2 → R.
where
c0 (t) := dct (1).
Note that the definition for the length of curves here is a generalization of the
Euclidean case where we integrate the speed of the curve. Indeed the last formula
above coincides with the regular notion of derivative when M is just Rn . To see this,
recall that dct (1) = (c ◦ γ)0 (0) where γ : (, ) → [a, b] is a curve with γ(0) = t and
γ 0 (0) = 1, e.g., γ(u) = t+u. Thus by the chain rule (c◦γ)0 (0) = c0 (γ(0))γ 0 (0) = c0 (t).
Exercise 0.4.1. Compute the length of the radius of the Poincare-disk (with respect
to the Poincare metric).
4
and then length of a curve was defined as the integral of ds, which was called “the
element of arclength”, along that curve:
Z
Length[c] = ds.
c
0.6 Distance
For any pairs of points p, q ∈ M , let C(p, q) denote the space piecewise smooth
curves c : [a, b] → M with c(a) = p and c(b) = q. Then, if M is connected, we may
define the distance between p and q as
So the distance between a pair of points is defined as the greatest lower bound
of the lengths of curves which connect those points. First we show that this is a
generalization of the standard notion of distance in Rn .
Lemma 0.6.1. For all continuous maps f : (a, b) → Rn
Z b
Z b
f (t)dt
≤ kf (t)kdt.
a a
So dg (p, q) ≤ kp − qk. It remains then to show that dg (p, q) ≥ kp − qk. The later
inequality holds because for all curves c : [a, b] → Rn
Z b
Z b
0 0
kc (t)k dt ≥
c (t)dt
= kc(b) − c(a)k.
a a
5
The previous result shows that (M, dg ) is a metric space when M is the Euclidean
space Rn and g, which induces d, is the standard innerproduct. Next we show that
this is the case for all Riemannian manifolds. To this end we first need a local
lemma:
n
Lemma 0.6.3. Let (B, g) be a Riemannian manifold, where B := B r (o) ⊂ Rn .
Then there exists m > 0 such that for any piecewise C 1 curve c : [a, b] → B with
c(a) = o and c(b) ∈ ∂B we have Length[c] > m.
Proof. Define f : Sn−1 × B → R by f (u, p) := gp (u, u). Note that, since g is positive
definite, f > 0. Thus since f is continuous and Sn−1 × B is compact f ≥ λ2 > 0.
Consequently, bilinearity of g yields that
gp (v, v) ≥ λ2 kvk2 .
6
Proof. It is immediate that d is symmetric and satisfies the triangle inequality.
Furthermore it is clear that d is always nonnegative. Showing that d is positive
definite, however, requires more work. Specifically, we need to show that when
p 6= q, then d(p, q) > 0. Suppose p 6= q. Then, since M is Hausdorff, there exists
an open neighborhood V of p such that q 6∈ V . Let (U, φ) be a local chart centered
at p. Choose r so small that B r (o) ⊂ φ(V ∩ U ), and set W := φ−1 (B r (o)). Then
φ : W → B r (o) is a diffeomorphism, and we may equip B r (o) with the push forward
metric dφ(g) which will turn φ into an isometry. Now let c : [a, b] → M be any
piecewise C 1 curve with c(a) = p and c(b) = q. Then there exist a ≤ b0 ≤ b
◦
such that c[a, b0 ] ⊂ W and c(b0 ) ∈ ∂W (to find b0 let W := φ−1 (Br (o)) be the
◦
interior of W , then c−1 (W ) is an open subset of [a, b] which contains a, and we
◦
may let b0 be the upperbound of the component of c−1 (W ) which contains a.) Let
c : [a, b0 ] → W be the restriction of c. Then obviously Length[c] ≥ Length[c]. But
Length[c] = Length[φ ◦ c] since φ is an isometry, and by the previous lemma then
length of any curve in (Brn (o), dφ(g)) which begins at the center of the ball and ends
at its boundary is bounded below by a positive constant.
Now recall that any metric space has a natural topology. In particular (M, dg )
is a topological space. Next we show that this topological space is identical to the
original M .
Proof. Define f : Sn−1 × M → R by f (u, p) := gp1 (u, u)/gp2 (u, u). Note that, since
g is positive definite, f > 0. Thus since f is continuous and Sn−1 × M is compact
f ≥ λ2 > 0. Consequently, bilinearity of g yields that
for any curve c : [a, b] → M . In particular the above inequalities hold for all curves
c : [a, b] → M with c(a) = p and c(b) = q.
Proposition 0.6.7. The metric space (M, dg ), endowed with its metric topology, is
homeomorphic to M with its standard topology.
7
Proof. There are two parts to this argument:
Part I: We have to show that every open neighborhood U of M is open in its
metric topology, i.e., for every p ∈ U there exists an r > 0 such that Brg (p) ⊂ U ,
where
Brg (p) := {q ∈ M | dg (p, q) < r}.
To see this first note that, as we showed in the proof of the previous proposition,
there exists an open neighborhood V of p with V ⊂ U such that there exists a home-
n
omorphism φ : V → B 1 (o). Now, much as in the proof of the previous proposition,
n n
if we endow B 1 (o) with the push forward metric induced by φ then (B 1 (o), dφ(g))
becomes isometric to (V , g). But recall that, as we showed in the earlier proposition,
n
the distance of any point in the boundary ∂B1n (o) = Sn of B 1 (o) from the origin o
was bigger than some constant, say λ. Thus the same is true of the distance of ∂V
from p. In particular, if we choose r < λ, then Brg (p) ⊂ V ⊂ U .
Part II: We have to show that every metric ball Brg (p) is open in M , i.e., at every
q ∈ Brg (p) we can find a open neighborhood U of q in M such that U ⊂ Brg (p). To
see this let V be an open neighborhood of p such that there exists a homeomorphism
n n
ψ : V → B 1 (o), and endow B 1 (o) with the push forward metric dψ(g). Then the
g
distance of ψ(V ∩ Br (p)) from o is equal to r, with respect to the metric dψ(g). So,
by the previous proposition, this distance, with respect to the Euclidean metric on
n
B 1 (o) must be at least λr > 0. Thus if we choose r0 < λr , then the Euclidean ball
Brn0 (o) ⊂ ψ(V ). Consequently, U := ψ −1 (Brn0 (o)) ⊂ V , and U is open in M , since
Brn0 (o) is open.