Differential Geometry Chapter 3

Curves
Given a curve α : I → Rn , we will misuse notation, calling both the
function and the image of the function the curve.

Definition 1 The speed function of α is

n
!1/2
X
v(t) = kα0 (t)k = α0i (t)2 ,
i=1

and Z b
kα0 (t)k dt
a
is the arc-length of α from a to b.

Example 2 For the helix α(t) = (a cos t, a sin t, bt), with a > 0 and b 6= 0,
we have
α0 (t) = (−a sin t, a cos t, b)α(t) .
Then
√
v(t) = a2 + b 2
√
and the arc-length from t = t0 to t1 is a2 + b2 (t1 − t0 ) .

Definition 3 Let I, J be open intervals of R and α : I → Rn a curve. If

h : J → I is a differentiable function, then the reparametrization of α by
h is the curve β = α ◦ h : J → Rn .

We stated the Chain Rule earlier for scalar-valued functions. Thus the
following result follows immediately for the component functions β i = αi ◦ h.
The stated result simply combines all these results.

Lemma 4
β 0 (t) = α0 (h(t)) h0 (t) .

Definition 5 Let α : I → Rn be a curve. Then

i. α is a regular curve if α0 (t) 6= 0 for all t ∈ I,
ii. α is a unit speed curve if v(t) = 1 for all t ∈ I.

1
Theorem 6 If α is a regular curve in Rn then there exists a reparametriza-
tion β of α such that β has unit speed.

Proof Choose a ∈ I and let

Z t
s(t) = kα0 (u)k du,
a

the arc-length function. Since α is differentiable (i.e. in C ∞ ) the integrand

kα0 (u)k is a continuous. So the Fundamental Theorem of Calculus says that
s(t) is differentiable (and thus continuous) with s0 (t) = kα0 (t)k for all t ∈ I.
Since kα0 (u)k > 0 the integral s(t) is a strictly increasing differentiable
function. Thus the inverse function theorem states that Im s is an open
interval in R, J say, and s has an inverse, i.e. there exists a differentiable,
strictly increasing f : J → I such that f (s(t)) = t so all t ∈ I.
Let β(x) = α(f (x)) for x ∈ J. Then β 0 (x) = α0 (f (x)) f 0 (x) and so

kβ 0 (x)k = kα0 (f (x))k f 0 (x) since f 0 > 0

= s0 (f (x)) f 0 (x)
d
= s (f (x)) by the Chain Rule
dx
d
= x = 1.
dx

1/2
Example Helix α(t) = (a cos t, a sin t, bt), t ∈ R, when v(t) = (a2 + b2 ) ,
constant but not necessarily 1. Since 0 ∈ R we can start the integral at 0 in
Rt 1/2 −1/2
s(t) = 0 v(u) du = (a2 + b2 ) t. Let f (x) = x (a2 + b2 ) . Then the unit
speed curve is

β(s) = (a cos (s/c) , a sin (s/c) , bs/c) , s

1/2
s ∈ R, where c = (a2 + b2 ) .
Note that for a unit speed curve α(t) we have v(t) = 1 for all t, so s(t) =
Rt 0
a
kα (u)k du = t − a. So we can replace t by s, write α(s) for the curve and
say that it is parametrized by the arc-length.

2
Frenet Formula
Let β : I → Rn be a unit speed curve so kβ 0 (s)k = 1 for all s ∈ I.

Definition 7 T (s) = β 0 (s) is the unit tangent vector field on β,

T 0 (s) is the curvature vector field of β,
k(s) = kT 0 (s)k is the curvature function of β.

Always κ(s) ≥ 0 and the larger κ is, the greater the rate of change of β
in the direction of T 0 .
Note that
2
T • T = β 0 • β 0 = kβ 0 (s)k = 1
since the curve is of unit speed. On differentiating T • T 0 = 0, i.e. T 0 is
orthogonal to T .
Assume κ(s) > 0 for all s ∈ I.

Definition 8 The principal normal vector field of β is

T 0 (s) β 00 (s)
N = N (s) = =
κ(s) κ(s)
and
B = B(s) = T (s) × N (s)
is the binormal vector field of β.

By definition both T and N are of unit length while T • T 0 = 0 means

T • N = 0 so T and N are orthogonal. thus by an earlier result {T, N, B} is
a frame at each point of β.

Definition 9 {T, N, B} is the Frenet frame field on β.

Example 10 Let α(t) = (4 (cos t) /5, 1 − sin t, −3 (cos t) /5) for t ∈ R.

Then α0 (t) = (−4 (sin t) /5, − cos t, 3 (sin t) /5)α(t) for which kα0 (t)k = 1
and so we have a unit speed curve. Thus T (t) = α0 (t) .
Next α00 (t) = (−4 (cos t) /5, sin t, 3 (cos t) /5)α(t) and again kα00 (t)k = 1.
Thus κ(t) = 1 for all t and N (t) = α00 (t).

3
 Finally

B(t) = T (t) × N (t)

 
3 2 3 2 12 12 4 2 4 2
= − cos t − sin t, − cos t sin t + sin t cos t, − sin t − cos t
2 5 5 5 5 5 α(t)
 
3 4
= − , 0, − .
5 5 α(t)

Note that in this example the binormal vector is independent of the point
of intersection. This is not such a surprise. The curve in question is the
intersection of the sphere (y − 1)2 + x2 + z 2 = 1 with the plane 3x + 4z =
0. The binormal vector will be orthogonal to this plane, as is the vector
(−3, 0, −4) /5.
Question How does the Frenet Frame {T (s) , N (s) , B(s)} change as s changes?
Consider first B 0 (s). Since B is of unit length B • B = 1 and so, on
differentiating, B 0 • B = 0.
Also, since {T, N, B} is a frame we have B • T = 0, and so, on again
differentiating, B 0 • T + B • T 0 = 0. But B • T 0 = κB • N = 0 hence
B 0 • T = 0.
Thus, since {T, N, B} is a frame,

B 0 = (B 0 • T ) T + (B 0 • N ) N + (B 0 • B) B = (B 0 • N ) N.

Definition 11 Define τ : I → R by B 0 (s) = −τ (s) N (s) , the torsion func-

tion of β . Note the −ve sign.

Example As noted before β(s) = (a cos (s/c) , a sin (s/c) , bs/c), s ∈ R,

1/2
where c = (a2 + b2 ) is a unit speed curve. Assume a > 0.

 s b
0 a s a
β (s) = − sin , cos , = T (s) ,
c c c c c β(s)
 a s a s 
T 0 (s) = − 2 cos , − 2 sin ,0
c c c c β(s)

a  s s 
= 2 − cos , − sin ,0 .
c c c β(s)

4
So κ(s) = a/c2 . So N (s) = (− cos (s/c) , − sin (s/c) , 0)β(s) and then

B(s) = T (s) × N (s)

 s b
a s a  s s 
= − sin , cos , × − cos , − sin ,0
c c c c c β(s) c c β(s)

 s a
b s b
= sin , − cos , .
c c c c c β(s)

Finally,
 s 
0 b s b b
B (s) = 2
cos , 2 sin ,0 = − 2 N (s) .
c c c c β(s) c

Therefore τ (s) = b/c2 . 

The important observation to take away from this example is that for a
helix both the curvature and torsion are constant.

Theorem 12 Frenet Formula Let β be a unit speed curve with κ(s) > 0
for all s ∈ I. Then

T 0 (s) = κ(s) N (s) ,

0
N (s) = −κ(s) T (s) +τ (s) B(s) ,
B 0 (s) = −τ (s) N (s) .

Proof Only the second result here is new. Again since {T, N, B} is a frame,

N 0 = (N 0 • T ) T + (N 0 • N ) N + (N 0 • B) B.

From N • N = 1 we have N 0 • N = 0.
From N • T = 0 we have N 0 • T + N • T 0 = 0, i.e. N 0 • T + N • (κN ) = 0.
Thus N 0 • T = −κ.
Similarly, from N • B = 0 we have N 0 • B + N • B 0 = 0, i.e. N 0 • B − N •
(τ N ) = 0. Thus N 0 • B = τ .
Combining we get the required result. 
The plane containing T & B is the rectifying plane, that containing N
& B the normal plane and that containing T & N the osculating plane.

5
Question what do κ and τ represent?
Consider the unit speed curve β (s). Taylor’s expansion states that, for s
sufficiently small,
s2 s3
β(s) = β(0) + β 0 (0) s + β 00 (0) + β 000 (0) + ....
2 3!
Here β 0 (0) = T (0) and β 00 (0) = T 0 (0) = κ(0) N (0). Also

000 d
= κ0 (0) N (0) + κ(0) N 0 (0)

β (0) = (κ(s) N (s))
ds s=0

= κ0 (0) N (0) + κ(0) (−κ(0) T (0) + τ (0) B(0))

Substituting back,
κ2 (0) 3 s2 s3 s3
   
0
β(s) ≈ β(0)+ s − s T (0)+ κ(0) + κ (0) N (0)+κ(0) τ (0) B(0) .
6 2 6 6
(1)
So a first approximation to β(s) is the tangent line β(0) + sT (0). The
second is the parabola
s2
β(0) + sT (0) + κ(0) N (0) . (2)
2
Thus κ(0) controls how fast the curve diverges from the straight line in the
direction of N (0) (how much it bends). Note that as s varies, the curve (2)
lies in the plane β(0) + span {T (0) , N (0)}, the osculating plane mentioned
before. We say that the osculating plane is the best approximating plane to
β at β(0).
If we had more time we would talk about the osculating circle, and the
evolute and involute curves. But we don’t!
The third approximation is the cubic (1). Hence τ (0) controls how fast
the curve leaves the {T (0) , N (0)} plane (or how much the curve twists.).
Question If τ (s) = 0 for all s does the curve remain in the {T (0) , N (0)}
plane? (If so we say, unsurprisingly, that the curve is planar.)

Lemma 13 Let β be a unit speed curve with κ(s) > 0 for all s ∈ I. Then β
is planar iff τ (s) = 0 for all s ∈ I.

6
Proof (=⇒) If β is planar then there exist points p and normal vector n,
such that (β(s) − p) • n = 0. Differentiating two times

β 0 (s) • n = β 00 (s) • n = 0, i.e. T (s) • n = κ(s) N (s) • n = 0

for all s. This means that n is orthogonal to both T (s) and N (s) for all s.
Yet B(s) is also orthogonal to both T (s) and N (s) and so B(s) = ±n/ knk
for all s. (This steps uses the fact that we have only 3 dimensions.) Therefore
B 0 (s) = 0, i.e. τ (s) = 0 for all s.
(⇐=) Assuming τ (s) = 0 for all s means B 0 (s) = 0, i.e. B(s) is constant for
all s. Claim (β(s) − β(0)) • B(s) = 0, i.e. β is planar.
Let f (s) = (β(s) − β(0))•B(s). Then f´0 (s) = β 0 (s)•B(s) = T (s)•B(s) =
0. So f (s) is constant. Yet f (0) = 0 so f (s) = 0 for all s as claimed.


Example 14 In the earlier example of α(t) = (4 (cos t) /5, 1 − sin t, −3 (cos t) /5)
for t ∈ R we found B(t) = − 53 , 0, − 45 α(t) . Thus B 0 (t) = (0, 0, 0)α(t) in

which case τ (t) = 0 for all t and the curve is planar.

Further, from the proof of the lemma, the curve lies in the plane (x − α (0))•
B (t) = 0. That is,
   
4 3 3 4
x − , y − 1, z + • − , 0, − = 0,
5 5 5 5
or 3x + 4z = 0.
In this example we also found that κ(t) = 1 for all t. This is a special
case of

Lemma 15 If τ ≡ 0 and κ(s) is constant then β is part of a circle.

Proof By Lemma 13, τ ≡ 0 means that β is planar. Consider the curve

1
γ(s) = β(s) + N (s) , (3)
κ

7
where κ = κ(s). Then, since κ is constant,
1
γ 0 (s) = β 0 (s) + N 0 (s)
k
1
= T (s) + (−κ(s) T (s) + τ (s) B(s))
κ
= 0.

Thus γ (s) is constant, i.e. equal to some p ∈ Rn . Then, rearranging (3) and
taking norms,
1 1
kβ(s) − pk = kN (s)k = ,
κ κ
i.e. β(s) lies on a circle, centre p, radius 1/κ. 

Example 16 In the earlier example of α(t) = (4 (cos t) /5, 1 − sin t, −3 (cos t) /5)
for t ∈ R we found that N (t) = (−4 (cos t) /5, sin t, 3 (cos t) /5)α(t) and
κ(t) = 1 for all t. Thus α(t) lies on the circle or radius 1, centre
   
1 4 3 4 3
α(0) + N (0) = , 1, − + − , 0, = (0, 1, 0) .
κ 5 5 5 5

Instead of lying in a circle what if the curve lies in the surface of a sphere?

Lemma 17 If the image of the unit speed α : I → Rn lies within the surface
of a sphere, then κ(t) 6= 0 and
2
ρ2 + (ρ0 σ) = r2 ,

where r is the radius of the sphere, ρ (t) = 1/κ(t) and σ(t) = 1/τ (t) .

Proof That α lies on the surface of a sphere means there is a point c ∈ Rn

and radius r > 0 such that

(α(t) − c) • (α(t) − c) = r2

for all t ∈ I. For simplicity I drop the dependency on t from my expressions.

The first differentiation gives α0 • (α − c) = 0, i.e.

T • (α − c) = 0. (4)

8
 Differentiate again, T 0 • (α − c) + T • T = 0 i.e. κN • (α − c) = −1. Thus
κ 6= 0 and
N • (α − c) = −ρ. (5)
Differentiate again, N 0 • (α − c) + N • α0 = −ρ0 . But N • α0 = N • T = 0
while N 0 = −κT + τ B. Thus (−κT + τ B) • (α − c) = −ρ0 . Yet, from above,
T • (α − c) = 0, so τ B • (α − c) = −ρ0 , i.e.

B • (α − c) = −σρ0 . (6)

Since {T, N, B} is a frame,

α − c = ((α − c) • T ) T + ((α − c) • N ) N + ((α − c) • B) B

= −ρN − σρ0 B.

by (4) , (5) and (6). Returning to the definition of a sphere

r2 = (−ρN − σρ0 B) • (−ρN − σρ0 B)

2
= ρ2 N • N + (σρ0 ) B • B
2
= ρ2 + (σρ0 ) .

Why are κ and τ of interest?

There are perhaps many answers to this question but I’m interested in the
fact that a “unit speed curve is uniquely determined by the pair of functions
(κ(t) , τ (t)) , up to position in R3 ”.
The map between the same object in different positions is the following.

Definition 18 A map F : R3 → R3 is an isometry if kF (x) − F (y)k =

kx − yk for all x, y ∈ R3 .

There are four basic isometries, reflection; glide reflection; rotation and
translation Isometries are given by an affine map x 7→ a + Ax, with a ∈ R3
and 3×3 orthogonal matrix A, so AT A = I3 . The uniqueness result is

9
Theorem 19 If α, β : I → R3 are unit speed curves with (κα (t) , τ α (t)) =
(κβ (t) , τ β (t)) for all t ∈ I then there exists an isometry F : R3 → R3 such
that β = F ◦ α.

Proof not given, but the isometry is constructed by mapping the Frenet
Frame on α at time t to the Frenet Frame on β at the same time. 
Example If a unit speed curve α : I → R3 has constant curvature and
torsion then it is a helix of the form β(s) = (a cos (s/c) , a sin (s/c) , bs/c) for
some a, b ∈ R.
Another fundamental result is one of existence .

Theorem 20 Given differentiable functions κ(s) > 0 and τ (s), s ∈ I, there

exists a regular curve α : I → R3 such that s is the arc-length, κ(s) is the
curvature and τ (s) the torsion of α.

Proof not given, but involves the existence and uniqueness of ordinary dif-
ferential equations. 

Arbitrary Speed Curves

If α(t) is not of unit speed we can find the arc-length parameter s(t) and
β(s) a unit speed curve satisfying β(s(t)) = α(t) .
Calculate Tβ (s) , Nβ (s) , Bβ (s) , κβ (s) and τ β (s) for β (s). Then {Tβ (s) , Nβ (s) , Bβ (s)}
is a frame for β for all s.
Write T (t) = Tβ (s(t)) , N (t) = Nβ (s(t)) , B(t) = Bβ (s(t)) , κ(t) =
κβ (s(t)) and τ (t) = τ β (s(t)). Then {T (t) , N (t) , B(t)} is a frame for α all t.
To see how this frame for α transforms as t varies,
d d d
T (t) = Tβ (s) s(t) = κβ (s) Nβ (s) v(t) = κ(t) N (t) v(t) ,
dt ds dt
having used the Frenet formula for unit speed curves, Theorem 12. And
d d d
N (t) = Nβ (s) s(t) = (−κβ (s) Tβ (s) + τ β (s) Bβ (s)) v(t)
dt ds dt
= (−κ(t) T (t) + τ (t) B(t)) v(t) .

10
Finally
d
B(t) − τ (t) N (t) v(t) .
dt
So, for an arbitrary speed curve the Frenet formula become

T 0 (t) = κ(t) v(t) N (t) ,

N 0 (t) = −κ(t) v(t) T (t) +τ (t) v(t) B(t) ,
0
B (t) = −τ (t) v(t) N (t) .

Calculations

When it comes to calculations we can quickly differentiate the given α(t)

but how to use these derivatives to calculate the T, N, B, κ and τ ?
First
d d
α0 (t) = β(s) s(t) = Tβ (s(t)) v(t) = T (t) v(t) .
ds dt
Yet T is of unit length so T = α0 (t) / kα0 (t)k.
Continue differentiating,
d
α00 (t) = Tβ (s(t)) v(t) + T (t) v 0 (t)
dt
= κβ (s(t)) Nβ (s(t)) v 2 (t) + T (t) v 0 (t)

= κ(t) N (t) v 2 (t) + T (t) v 0 (t) .

This shows that the acceleration of α(t) has a tangential component, the
T (t) v 0 (t) term, and a normal component proportional to the square of ve-
locity and to the curvature of the curve..
Next, again dropping the dependency on t for ease of notation,

α0 ×α00 = (T v)× κN v 2 + T v 0 = κv 3 B,

since B = T × N and T × T = 0. Yet B is of unit length so B = α0 ×

α00 / kα0 ×α00 k . And for the same reason, κv 3 = kα0 ×α00 k, and thus

kα0 ×α00 k
κ= .
kα0 k3

Find N from N = B ×T = (α0 ×α00 )×α0 / k(α0 ×α00 )×α0 k.

11
 For the final derivative we have

0
α000 = κv 2 N + κv 2 N 0 + v 00 T + v 0 T 0

0
= κv 2 N + κv 2 (−κvT + τ vB) + v 00 T + v 0 κvN

0 
= v 00 − κ2 v 3 T + κv 2 + v 0 κv N + κτ v 3 B.

We only need to know the coefficient of B here, since

(α0 ×α00 ) • α000 = κ2 τ v 6 .

Hence
k(α0 ×α00 ) • α000 k
τ= ,
kα0 ×α00 k2
since, from earlier, κv 3 = kα0 ×α00 k .

Example 21 Find T, N, B, κ and τ for the curve

α (t) = (a cos t, a sin t, d sin t) ,

t ∈ R.
1/2
Solution First α0 (t) = (−a sin t, a cos t, d cos t)α(t) so kα0 (t)k = (a2 + d2 cos2 t) .
Continuing,

α00 (t) = (−a cos t, −a sin t, −d sin t)α(t)

α000 (t) = (a sin t, −a cos t, −d cos t)α(t) .

1/2
Then α0 ×α00 = (0, −ad, a2 )α(t) and kα0 ×α00 k = a (a2 + d2 ) .
For N we need

(α0 ×α00 )×α0 = 0, −ad, a2

×(−a sin t, a cos t, d cos t)α(t)
α(t)

−ad2 − a 3
cos t, −a3 sin t, −a2 d sin t α(t) .



=

Then
2
2
k(α0 ×α00 )×α0 k ad2 + a3 cos2 t + a6 + a4 d2 sin2 t


=

= a2 a2 + d 2 a2 + d2 cos2 t + a2 sin2 t

= a2 a2 + d2 a2 + d2 cos2 t .

12
 Next, (α0 ×α00 ) • α000 = a2 d cos t − a2 d cos t = 0.
Putting these results together,

α0 (t) (−a sin t, a cos t, d cos t)α(t)

T = = ,
kα0 (t)k (a2 + d2 cos2 t)1/2

(α0 ×α00 )×α0 ((−ad2 − a3 ) cos t, −a3 sin t, −a2 d sin t)α(t)
N= = ,
k(α0 ×α00 )×α0 k a (a2 + d2 )1/2 (a2 + d2 cos2 t)1/2

α0 ×α00 (0, −ad, a2 )α(t)

B= = ,
kα0 ×α00 k a (a2 + d2 )1/2
1/2
kα0 ×α00 k a (a2 + d2 )
κ= = ,
kα0 k3 (a2 + d2 cos2 t)3/2
and
k(α0 ×α00 ) • α000 k
τ= = 0,
kα0 ×α00 k2
i.e. the curve is planar. (By observation it lies in the plane dy − az = 0.)
Finally

13