lecture06
lecture06
MATH 136
6.1. We look today at a theorem in two dimensions. It deals with signed curvature
′ ′′
κ = r|r×r ′ |3 using the cross product in 2 dimensions. 1. We do not assume that the
curve is Frenet. The curvature is allowed to become zero. We assume however that
the curve is closed and regular meaning that dr = r′ is never zero. In that case, there
is an arc length parametrization of the curve and |κ| = |r′′ | because r′ · r′ = 1 implies
r′′ is perpendicular to r′ . But we have a signed curvature!
6.2. Assume that the curve is parametrized on [a, b]. The rotation index is defined as
Rb Rb
1
2π a
κ(t) dt. If the closed curve is not arc length parametrized, this is a
κ(t)|r′ (t)| dt.
Theorem 1. The rotation index of a closed C 2 curve is in Z.
Proof. Using arc length parametrization, write
r′ (t) = [cos(α(t)), sin(α(t)]
then κ = α′ . Since the curve is closed, we have α(b) − α(a) = 2πn, where n is an
integer. □
6.3. The case r(t) = [cos(nt), sin(nt)] with t ∈ [0, 2π] shows that the rotation index can
take any integer value n. It is intuitively clear that if a curve has no self intersections,
then the index must be either 1 or −1. This is not so obvious however. We do not want
for example to refer to the Jordan curve theorem telling that a continuous simple closed
curve in the plane divides the plane into an inside and outside. Heinz Hopf found a
nice argument which proves this ”Umlaufsatz” in an elegant way using a deformation
picture:
Theorem 2 (Hopf Umlaufsatz). A simple closed regular C 2 curve has rotation index
1 or −1.
Proof. Arc length parametrization is not needed. We assume that r(t) is parametrized
on the interval [0, 1]. Define on the square Q = [0, 1] × [0, 1] the function f : Q →
T = R/(2πZ)} given by f (t, s) = arg((r(t) − r(s))/|r(t) − r(s)|) for t ̸= s and α(t) =
f (t, t) = arg(r′ (t)/|r′ (t)|) for t = s. Because r ∈ C 1 , the function f is continuous. Now
comes a homotopy argument. The index is [f (1, 1) − f (0, 0)]/(2π) and is an integer. If
we move along the diagonal and look at α(t) = f (t, t) we see a continuous curve which
1The n
cross product in n dimensions has = n(n − 1)/2 components. For n = 2 it is a scalar
2
Differential Geometry
moves on the circle T. If we deform the curve the total change remains the same. We
can continuously deform the curve so that we first deform from (0, 0) straight to (0, 1)
and then straight from (0, 1) to (1, 1). Choose a coordinate system so that is in y ≥ 0
just touching the x-axes. If r′ (0) = [a, 0] with positive a then f (t, s) ∈ [0, π] with
f (0, 0) = 0 and f (0, 1) = π and then f (1, 1) = 2π. If a < 0, then f (t, s) ∈ [−π, 0] with
f (0, 0) = π and f (0, 1) = 0 and then f (1, 1) = −π. In the former case, i = 1 in the
later i = −1. □
6.4. Remarks:
1) This is a Gauss-Bonnet type result for a 2 dimensional flat manifold with boundary.
2) The proof shows that this even works for C 1 curves as f (t, t) − f (s, s) is just the
angle change of the tangent. This works even if the curvature is not defined. In the
homework you even push it to polygons. Most texts assume C 2 .