Curves

JWR
January 27, 2014

These notes summarize the key points in the first chapter of Differential
Geometry of Curves and Surfaces by Manfredo P. do Carmo. I wrote them to
assure that the terminology and notation in my lecture agrees with that text.
All page references in these notes are to the Do Carmo text.
1. Definition. A parameterized smooth curve is a map α : I → Rn where
I ⊆ R is an interval. The set theoretic image

C = α(I) := {α(t) : t ∈ I}

is called the trace of α and α is called a parameterization of C. See do

Carmo page 2.
2. Remark. For do Carmo the words differentiable and smooth are synony-
mous. I prefer the word smooth. The adjective differentiable is often omitted
by do Carmo.
3. Remark. On page 2 do Carmo says that the interval I should be open but
on page 30 he extends the notion of smoothness to closed intervals. A function
defined on a closed interval [a, b] is said to be smooth iff it extends to an open
interval containing [a, b]. This means that the derivatives of the function are
defined at the end points a and b.
4. Definition. A reparameterization of α : I → Rn is a smooth map
β : J → Rn of form β = α ◦ σ where σ : J → I is a diffeomorphism. That σ
is a diffeomorphism means σ is one-to-one and onto (so there is an inverse map
σ −1 : J → I) and that σ ′ (t) 6= 0 for t ∈ I (so that the map σ −1 is also smooth).
5. Remark. If β is a reparameterization of α, then the maps α and β have
the same trace C. The idea of the definition is that we should think of α and
β as different ways of describing the same curve C. However do Carmo avoids
giving a precise definition of an (unparameterized) curve.
6. Example. The circle C = {(x, y) ∈ R2 : x2 + y 2 = 1} is the trace of the
parameterized curve α : R → R2 defined by



α(θ) = cos θ, sin θ = cos(θ + 2π), sin(θ + 2π) .

1
Define a map β : R → R2 by
1 − t2
 
2t
β(t) = , .
1 + t2 1 + t2
This map is a reparameterization of the restriction of α to the open interval
(−π, π) as follows:
cos ϕ − sin2 ϕ 2 sin ϕ cos ϕ
 2 


cos(2ϕ), sin(2ϕ) = ,
cos2 ϕ + sin2 ϕ cos2 ϕ + sin2 ϕ

1 − tan2 ϕ 2 tan ϕ
 
= , .
1 + tan2 ϕ 1 + tan2 ϕ
Take 2ϕ = θ, t = tan ϕ = tan(θ/2), and we get α = β ◦ σ where σ : (−π, π) → R
is defined by σ(θ) = tan(θ/2). The common trace of (the restriction of) α and
the map β is the punctured circle C \ (−1, 0). (This particular reparameteriza-
tion is called the Weierstrass substitution or half angle substitution. It
is one of the main techniques used to evaluate integrals in calculus.)
7. Definition. Let α : I → Rn be a smooth parameterized curve. The deriva-
tive α′ (t) is called velocity vector at t. The map α is called regular iff its
velocity vector never vanishes. The map α is said to be parameterized by
arc length iff its tangent vector always has length one.
8. Theorem. A smooth regular parameterized curve α has a reparameterization
by arc length, i.e. there is a reparameterization β : J → Rn of α such that
|β ′ (s)| = 1 for s ∈ J.

Proof: This is the content of Remark 2 in do Carmo page 21. The reparametriza-
tion is defined by β = α ◦ σ where σ is a solution of the differential equation
1
σ ′ (s) = .
|α′ (σ(s))|
By the chain rule β ′ (s) = α′ (σ(s))σ ′ (s) so |β ′ (s)| = 1.
9. Remark. The arc length
Z b
ℓ(C) = |α′ (t)| dt
a

of the trace C of a regular parameterized curve α : [a, b] → Rn is independent

of the parameterization α used to define it. This is an easy consequence of the
formula for changing variables in a definite integral: if σ : [a, b] → [c, d] is a
diffeomorphism, then
Z b Z d
′
|α (t)| dt = |α′ (σ(r))| |σ ′ (r)| dr.
a c

(The change of variables is t = σ(r) so dt = σ ′ (r) dr.) When α is parameterized

by arc length, ℓ(C) = |b − a|.

2
10. The reparameterization in Theorem 8 is unique in the following sense: If
β1 : J1 → Rn and β2 : J2 → Rn are two reparameterizations of the same map
α then β2 = β1 ◦ σ where σ : J2 → J1 has one of the two forms σ(s) = s + c or
σ(s) = −s + c. (This is because |σ ′ (s)| = 1.) On page 6 do Carmo says that
when σ(s) = −s + c the two curves β1 and β2 are said to differ by a change
of orientation.
This use of the word orientation can be viewed as a special case of the
definition of orientation of a vector space that do Carmo gives on pages 11
and 12. For a regular curve α the one dimensional vector space Rα′ (t) ⊆ Rn is
called the tangent space to the curve at the point α(t). The velocity vector
α′ (t) is a basis for this space. Changing the orientation of the curve changes the
sign of the velocity vector α′ (t) and thus reverses the orientation of the tangent
space.
11. Remark. Note the distinction between the tangent space and the tangent
line. The tangent line is the line containing the points α(t) and α(t) + α′ (t).
(See do Carmo page 5.) This line need not pass through the origin of Rn
and thus is not a vector subspace of the vector space Rn . This illustrates the
difference between points and vectors.
12. Remark. The two orientations of R3 correspond to the thumb, forefin-
ger, and middle finger of the right and left hands. (Recall the right hand rule
from calculus.) The two orientations of R2 correspond to clockwise and counter
clockwise. The two orientations of R = R1 correspond to the two directions
increasing and decreasing.
13. Definition. A map Φ : Rn → Rn is called an isometry iff it preserves
distance i.e. iff it satisfies
|Φ(p) − Φ(q)| = |p − q|
for p, q ∈ Rn . A map Rn → Rn is called a translation iff there is a vector
c ∈ Rn such that the map sends the point p ∈ Rn to the point p + c. A linear
transformation ρ : Rn → Rn is called orthogonal iff it satisfies (ρu)·(ρv) = u·v
for all vectors u, v ∈ Rn . A rigid motion of Rn is an isometry Φ such that the
corresponding orthogonal linear transformation ρ preserves orientation, i.e. has
positive determinant.
14. Theorem. A map Φ : Rn → Rn is an isometry if and only if it is the
composition of a translation and an orthogonal linear transformation.

Proof: For if see do Carmo page 23 Exercise 6 and do Carmo page 228 Exercise 7.
The converse is not very difficult but is not needed in the rest of these notes so
the proof is omitted.
15. Theorem. Let α : [a, b] → Rn be smooth, Φ : Rn → Rn be an isometry,
and β = Φ ◦ α :. Then the curves α and β have the same arc length. If α is
parameterized by arc length so is β.

Proof: This is because Φ preserves the length of vectors. The first part also
follows from Exercise 8 on page 10 of do Carmo.

3
16. Definition. Let α : I → Rn be parameterized by arc length. Then the
unit tangent vector is the vector valued function t : I → Rn defined by
d
t(s) = α′ (s) = α(s),
ds
the curvature vector is the vector valued function I → Rn
d d2
α′′ (s) = t(s) = 2 α(s),
ds ds
and the curvature is the length κ of the curvature vector, i.e.

κ(s) = |t′ (s)| = |α′′ (s)|.

The unit normal vector is the normalized curvature vector

t′
n= .
|t′ |

(The vector n is defined only where the curvature κ is not zero.) The binormal
vector is the vector product
b= t∧n
of the unit tangent vector t and the unit normal vector n. (The binormal vector
is defined only when n = 3.)
17. Theorem. Let α : I → Rn be parametrized by arc length, Φ : Rn → Rn
be an isometry, and β = Φ ◦ α : I → Rn . Then β is also parametrized by arc
length and α and β have the same curvature. If n = 3 and Φ is a rigid motion
they have the same torsion.

Proof: Exercise 6 page 23 of do Carmo.

18. Standing Assumption. Henceforth we assume that α : I → R3 is a
regular curve parameterized by arc length.
19. Theorem. Then the vectors t, n, b are orthonormal, i.e.

|t| = |n| = |b| = 1, t · n = t · b = n · b = 0.

The ordered orthonormal basis t, n, b is called the Frenet trihedron.

Proof: (See do Carmo pages 18-19.) The equations |t| = |n| = 1 hold by
definition. Since |t|2 = t · t is constant we get

d 2 d
0= |t| = t · t = 2 t · t′ = 2κ t · n
ds ds
so t · n = 0. Now b is the vector product of two orthogonal unit vectors t and
n so it is itself a unit vector and is orthogonal to both t and n.

4
20. Corollary. The derivative b′ of the binormal vector b is parallel to the
unit normal vector n, i.e. there is a real valued function τ such that
b′ = τ n, τ = b′ · n.
The function τ is called the torsion.

Proof: Since t′ ∧ n = κ n ∧ n = 0 we have

b′ = (t ∧ n)′ = t′ ∧ n + t ∧ n′ = t ∧ n′
so b′ · t = b′ · n′ = 0.
21. Frenet Formulas. The Frenet trihedron satisfies the differential equations

t′ = κn, n′ = −κt − τ b, b′ = τ n.

Proof: The first and last formulas hold by definition. For the middle formula
differentiate the identities n · t = n · b = 0 and n · n = 1 to get
0 = n′ · t + n · t′ = n′ · t + κ n · n = n′ · t + κ
0 = n′ · b + n · b′ = n′ · t + τ n · n = n′ · b + τ
0 = 2n′ · n
Since the Frenet trihedron is orthonormal
n′ = (n′ · t)t + (n′ · n)n + (n′ · b)b.
This proves the middle Frenet formula.
22. Remark. The Frenet formulas may be written in matrix form as
 ′    
t 0 κ 0 t
 n′  =  −κ 0 −τ   n 
b′ 0 τ 0 b
The coefficient matrix is skew symmetric. This is no coincidence. The two
triples
t(s), n(s), b(s), t(s0 ), n(s0 ), b(s0 )
are both bases for the vector space R3 so there is a unique change of basis matrix
U (s) satisfying    
t(s) t(s0 )
 n(s)  = U (s)  n(s0 )  .
b(s) b(s0
Since the two bases are both orthonormal, the matrix U (s) is orthogonal. Dif-
ferentiating with respect to s and evaluating at s = s0 gives the Frenet formula
(in matrix form) evaluated at s = s0 . But U (s0 ) is the identity matrix and
U (s)∗ = U (s)−1 so U ∗ (s)U (s) is the identity matrix so differentiating at s and
evaluating at s0 gives
U ′ (s0 )∗ + U ′ (s0 ) = 0.

5
23. Theorem. Reversing the orientation of α leaves the curvature κ and the
torsion τ unchanged, i.e. if β(s) = α(−s) the curves α and β have the same
curvature and torsion at s = 0.

Proof: By definition the curvature κ is nonnegative, the normal vector is only

defined at points where the curvature κ is not zero, reversing the orientation
of α reverses the sign of the unit tangent vector t and leaves the sign of the
curvature vector unchanged. Reversing the orientation of α reverses the sign of
t, preserves the sign of n, and therefore reverses the sign of b = t ∧ n. But
reversing the orientation of b reverses the sign of b′ so reversing the orientation
of α preserves the sign of b′ and hence (by the Frenet formula b′ = τ n) preserves
the sign of τ .
24. Fundamental Theorem. Let κ, τ : I → R be smooth functions defined on
an interval I. Assume κ > 0. Then
(Existence.) There is a curve α : I → R3 parameterized by arc length with
curvature κ and torsion τ .
(Uniqueness.) If α, β : I → R3 are two curves paramaterized by arc length both
having curvature κ and torsion τ , then there is a rigid motion Φ : R3 → R3
such that β = Φ ◦ α.

Proof: See do Carmo page 309.

25. Corollary. The curvature and torsion of the helix α(θ) = (a cos θ, a sin θ, bθ)
are both constant so for any two points p and q on the helix there is a rigid
motion carrying p to q and mapping the helix to itself.
26. Gauss curvature. In the case of a plane curve (n = 2) it is possible to
choose a normal vector even when the curvature is zero. In this case since t and
n are orthogonal unit vectors we can define n by rotating t clockwise through
90 degrees:
t = (ξ, η), n = (η, −ξ).
With this definition both t and n change sign when the orientation is reversed
so to maintain the equation t′ = κn it is necessary to allow κ to be negative.
For a plane curve α : I → R2 parameterized by arc length we can view the unit
normal vector as a map to the unit circle and define an angle θ = θ(s) by the
formula
n(s) = (cos θ(s), sin θ(s)).
We then define the signed curvature by the formula
dθ
κ= .
ds
The signed curvature κ for a plane curve C ⊆ R2 is analogous to the Gauss
curvature K of a surface S ⊆ R3 . (See do Carmo pages 146, 155, 167.) Note

6
that when α(s) = (cos s, sin s) is the counter clockwise parameterization of the
unit circle in R2 , the vector n defined by rotation of t as above is the outward
normal (=radius vector) to the circle and the curvature κ is identically one.
Thus the curvature compares the curve α to the unit circle.
27. Setup for local canonical form. Assume that α : I → R3 has positive
curvature and s0 ∈ I. The Taylor expansion
(s − s0 )2 ′′ (s − s0 )3 ′′′
α(s) = α(s0 ) + (s − s0 )α′ (s0 ) + α (s0 ) + α (s0 ) + · · ·
2 6
tells us what the trace C of α looks like near the point α(s0 ) ∈ C. Because any
reparameterization of C has the same trace we assume that α is parameterized
by arc length. Because the reparameterization defined by σ(s) = s − s0 is also a
parameterization by arc length, we assume that s0 = 0. Because the arc length,
curvature, and torsion are invariant under rigid motions, we assume that
α(0) = (0, 0, 0), t(0) = (1, 0, 0), n(0) = (0, 1, 0), b(0) = (0, 0, 1).
28. Local Canonical Form. In the notation of Setup 27 above, the Taylor
expansion of α(s) = (x(s), y(s), z(s)) is
κ(0)s3
x(s) = s− + Rx
6

κ(0)s2 κ′ (0)s3
y(s) = − + Ry
2 6

κ(0)τ (0)s3
z(s) = − + Rz
6
where Rx , Ry , Rz = o(s3 ).

Proof: There is no constant term in these formulas because α(0) = 0. By

definition
α′ = t, α′′ = κn.
Differentiating once more gives
α′′′ = κ′ n + κn′ = κ′ n + κ(−κt − τ b)
by the second Frenet formula. Now evaluate at s = s0 = 0.
29. Application. Recall (Remark 11 above and do Carmo page 5) that the
tangent line to the trace C of a regular curve α at a point p0 = α(s0 ) ∈ C is the
line containing the two points p0 and p0 + t0 where t0 = t(s0 ). The osculating
plane to C at p0 is the plane containing the three points p0 , p0 + t0 , p0 + n0
where n0 = n(s0 ). (See do Carmo pages 17, 29, 30. The definition assumes that
the curvature κ(s0 ) at p0 is positive.) Let p1 = α(s1 ) and p2 = α(s2 ) be two
other points on C distinct from p0 and each other. Then as s1 → s0 the limit of
the line through p0 and p1 is the tangent line at p0 and the limit as s1 , s2 → s0
of the plane through p0 , p1 , and p2 is the osculating plane.