Surface Curvatures∗

(Com S 477/577 Notes)

Yan-Bin Jia

Oct 23, 2017

1 Curve on a Surface: Normal and Geodesic Curvatures

One way to examine how much a surface bends is to look at the curvature of curves on the surface.
Let γ(t) = σ(u(t), v(t)) be a unit-speed curve in a surface patch σ. Thus, γ̇ is a unit tangent
vector to σ, and it is perpendicular to the surface normal n̂ at the same point. The three vectors
γ̇, n̂ × γ̇, and n̂ form a local coordinate frame by the right-hand rule.
Differentiating γ̇ · γ̇ yields that γ̈ is orthogonal to γ̇. Hence γ̈ is a linear combination of n̂ and
n̂ × γ̇:
γ̈ = κn n̂ + κg n̂ × γ̇ (1)
Here κn is called the normal curvature and κg is the geodesic curvature of γ.

n̂

γ̈
γ̇
n̂ × γ̇ φ
κn
κg
γ

Since n̂ and n̂ × γ̇ are orthogonal to each other, (1) implies that

κn = γ̈ · n̂ and κg = γ̈ · (n̂ × γ̇).
Since γ is unit-speed, its curvature is
κ = kγ̈k
= kκn n̂ + κg n̂ × γ̇k
q
= κ2n + κ2g . (2)
∗
The material is adapted from the book Elementary Differential Geometry by Andrew Pressley, Springer-Verlag,
2001.

1
Let ψ be the angle between the principal normal n̂γ of the curve, in the direction of γ̈, and the
surface normal n̂. We have
κn = κn̂γ · n̂ = κ cos ψ. (3)
Therefore, equation (2) implies
κg = ±κ sin ψ.
If γ is regular but arbitrary-speed, the normal and geodesic curvatures of γ are defined to be
those of a unit-speed reparametrization of the same curve.
A unit-speed curve γ is a geodesic if κg = 0. By (1), its acceleration γ̈ is always normal to the
surface. Geodesics have many applications that we will devote one lecture to the topic later on.

2 Darboux Frame on a Curve

On the unit-speed surface curve γ, the frame formed by the unit vectors γ̇, n̂ × γ̇, and n̂ is called
the Darboux frame on the curve. This frame is different from the Frenet frame on the curve defined
by γ̇, the principal normal γ̈/kγ̈k, and the binormal γ̇ × γ̈/kγ̈k.
Let us rename the three unit vectors γ̇, n̂ × γ̇, n̂ as T , V , U , respectively. Their derivatives
must be respectively orthogonal to themselves. Differentiation of U · T = 0 yields

U′ · T = −U · T ′
= −U · (κn U + κg V ) (by (1))
= −κn .

The vector U ′ has a second component — along V . The geodesic torsion, defined to be τg = −U ′ ·V ,
describes the negative rate of change of U in the direction of V . With τg we characterize this change
rate completely as below:
U ′ = −κg T − τg V.
Similarly, we express V ′ in terms of T, U, V , and combine it with the above equation and (1) into
the following compact form (where the vectors are viewed as “scalars”):
 ′   
T 0 κg κn T
 V  =  −κg 0 τg   V  (4)
U −κn −τg 0 U

The formulas (4) describe the geometry of a curve at a point in a local frame suiting the curve as
well as a surface on which it lies, whereas the Frenet formulas describe its geometry at the point
in a local frame best suiting the curve alone.
The curve γ is asymptotic provided its tangent γ̇ always points in a direction in which κn = 0.
In some sense, the surface bends less along γ than it does along a general curve. In (4), T ′ = κg V .
Thus, κ = κg and V is aligned with the principal normal T ′ /kT ′ k (assuming κg 6= 0). Consequently,
the Darboux frame T -V -U coincides with the Frenet frame everywhere.

2
3 The Second Fundamental Form
Let σ be a surface patch in R3 with standard unit normal
σu × σv σu × σv
n̂ = =√ . (5)
kσ u × σ v k EG − F 2
With an increase (∆u, ∆v) in the parameter values, the movement of the point is described by
Taylor’s series below:

σ(u + ∆u, v + ∆v) − σ(u, v)

1  
= σ u ∆u + σ v ∆v + σ uu (∆u)2 + 2σ uv ∆u∆v + σ vv (∆v)2 ) + O (∆u + ∆v)3 .
2
The first order terms are tangent to the surface, hence perpendicular to n̂. The terms of order
higher than two tend to zero as (∆u)2 + (∆v)2 does so.
The deviation of σ from the tangent plane is determined by the dot product of the second order
term with the surface normal n̂, namely, it is
1 
L(∆u)2 + 2M ∆u∆v + N (∆v)2 , (6)
2
where

L = σ uu · n̂
σ uu · (σ u × σv )
= by (5)
kσ u × σ v k
det(σ uu σu σv )
= √ ,
EG − F 2
det(σ uv σu σ v )
M = σ uv · n̂ = √ , (7)
EG − F 2
det(σ vv σu σv )
N = σ vv · n̂ = √ .
EG − F 2
Recall that a unit-speed space curve α(s) can be approximated around s = 0 up to the second
order as α(0) + st̂(0) + 21 κ(0)s2 n̂(0), where t̂(0) and n̂(0) are the tangent and principal normal,
respectively, and κ(0) the curvature. The expression (6) for the surface is analogous to the curvature
term 12 κ(0)s2 n̂(0) for a curve. In particular, the expression

Ldu2 + 2M dudv + N dv 2

is the second fundamental form of σ.

While the first fundamental form permits the calculation of metric properties such as length
and area on a surface patch, the second fundamental form captures how ‘curved’ a surface patch is.
The roles of the two fundamental forms on describing the local geometry are analogous to those of
speed and acceleration for a parametric curve. Just as a unit-speed space curve is determined up
to a rigid motion by its curvature and torsion, a surface patch is determined up to a rigid motion
by its first and second fundamental forms.

3
Example 1. Consider a surface of revolution
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)),
where f (u) > 0 always holds. The figure below plots a catenoid where
u
f (u) = 2 cosh and g(u) = u
2
with (u, v) ∈ [−2, 2] × [0, 2π].

Assume the profile curve u → (f (u), 0, g(u)) is unit-speed, i.e., f˙2 + ġ 2 = 1, where a dot denotes
differentiation with respect to u. We perform the following calculations:
σu = (f˙ cos v, f˙ sin v, ġ),
σv = (−f sin v, f cos v, 0),
E = σ u · σ u = f˙2 + ġ 2 = 1,
F = σ u · σ v = 0,
G = σv · σv = f 2,
σu × σv = (−f ġ cos v, −f ġ sin v, f f˙),
kσu × σ v k = f,
σu × σv
n̂ = = (−ġ cos v, −ġ sin v, f˙),
kσu × σ v k
σ uu = (f¨ cos v, f¨ sin v, g̈),
σ uv = (−f˙ sin v, f˙ cos v, 0),
σ vv = (−f cos v, −f sin v, 0),
L = σ uu · n̂ = f˙g̈ − f¨ġ,
M = σ uv · n̂ = 0,
N = σ vv · n̂ = f ġ.
Thus, the second fundamental form is
(f˙g̈ − f¨ġ)du2 + f ġdv 2 .

4
4 Principal Curvatures
Let γ(t) = σ(u(t), v(t)) be a unit-speed curve on a surface patch σ with the standard unit normal n̂.
Below we obtain the normal curvature of the curve:

κn = γ̈ · n̂
d
= γ̇ · n̂
dt
d
= n̂ · (σ u u̇ + σ v v̇)
dt 
= n̂ · σ u ü + σ v v̈ + (σ uu u̇ + σ uv v̇)u̇ + (σ uv u̇ + σvv v̇)v̇
= Lu̇2 + 2M u̇v̇ + N v̇ 2 , (8)

where L, M, N are coefficients of the second fundamental form defined (7). In the last step above,
we used the fact that n̂ is normal to both σ u and σ v , i.e., n̂ · σ u = n̂ · σ v = 0.
Equation (8) states that the normal curvature of γ(t) depends on u, v, u̇, and v̇. The first two
quantities specify the location of the point on the surface σ, and are thus curve independent. Let û
be the unit tangent vector so that γ̇ = û. Take the dot products of the equation û = σu u̇ + σ v v̇
with σ u and σ v separately, yielding

E u̇ + F v̇ = û · σ u ,
F u̇ + Gv̇ = û · σ v ,

where E, F, G are the coefficients of the first fundamental form of the surface patch. Since σ u
and σ v are linearly independent, the coefficient matrix in the above linear system in u̇ and v̇ is
non-singular. We solve the system to obtain
   −1  
u̇ E F û · σ u
=
v̇ F G û · σv

This implies that u̇ and v̇ are independent of the parametrization of γ. By (8), we conclude that
any two unit-speed curves passing through the same point in the same direction û must have the
same normal curvature at this point.
Subsequently, we refer to κn as the normal curvature of the
surface σ at the point p = σ(u, v) in the tangent direction of û. It
measures the curving of the surface in that direction. Generally,
the surface bends at different rates in different tangent directions.
n̂
The tangent û and the normal n̂ at p defines a plane that cuts a
û curve α out of the patch. This curve is called the normal section
of σ in the û direction. Since the principal normal n̂γ of the
α normal section is related to the surface normal n̂ by n̂γ = ±n̂.
Equation (3) implies that the curvature of the normal section is
σ the normal curvature κn at the point or its opposite, depending
on the choice of the surface normal n̂ at the point.

5
 To analyze the normal curvature κn further, we make use of the first and second fundamental
forms:
Edu2 + 2F dudv + Gdv 2 and Ldu2 + 2M dudv + N dv 2 .
For convenience, we introduce two symmetric matrices
   
E F L M
F1 = and F2 =
F G M N

The tangent vector of the unit-speed curve γ(t) = σ(u(t), v(t)) is γ̇ = u̇σ u + v̇σv . Introducing
T = (u̇, v̇)t , where t denotes the transpose operator, equation (8) can be rewritten as

κn = T t F2 T. (9)

Since σ u and σ v span the tangent plane, consider two tangent vectors:

t̂1 = ξ1 σ u + η1 σv and t̂2 = ξ2 σ u + η2 σ v .

We obtain their inner product:

t̂1 · t̂2 = (ξ1 σ u + η1 σ v ) · (ξ2 σu + η2 σ v )

= Eξ1 ξ2 + F (ξ1 η2 + ξ2 η1 ) + Gη1 η2
 
t E F
= T1 T2 , (10)
F G

where    
ξ1 ξ2
T1 = and T2 = .
η1 η2
The principal curvatures of the surface patch σ are the roots of the equation

L − κE M − κF
det(F2 − κF1 ) = = 0. (11)
M − κF N − κG

From the linear independence of σ u and σv , it is easy to show that matrix F1 is always invertible.
Equation (11) essentially states that the principal curvatures are the eigenvalues of F1−1 F2 .
Let κ be a principal curvature of F1−1 F2 , and T = (ξ, η)T the corresponding eigenvector. That
−1
(F1 F2 )T = κT implies
(F2 − κF1 )T = 0. (12)
The unit tangent vector t̂ in the direction of ξσ u + ησv is called the principal vector corresponding
to the principal curvature κ.

Theorem 1 Let κ1 and κ2 be the principal curvatures at a point p of a surface patch σ. Then

(i) κ1 , κ2 ∈ R;

(ii) if κ1 = κ2 = κ, then F2 = κF1 and every tangent vector at p is a principal vector.

(iii) if κ1 6= κ2 , then the two corresponding principal vectors are perpendicular to each other

6
 For a proof of the theorem, we refer the reader to [2, 133–135]. Intuitively, the principal vectors
give the directions of maximum and minimum bending of the surface at the point, and the principal
curvatures measure the bending rates. In case (ii), the point is umbilic. In this case, the surface
bends the same amount in all directions at p (thus all directions are principal).

Example 2. A sphere bends the same amount in every direction. Take the unit sphere in Example 9 in the
notes “Surfaces”, for instance, with the parametrization

σ(θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ),

We found previously that

E = 1, F = 0, G = cos2 θ.
Since a sphere is a surface of revolution, we can plug in the result from Example 1, with f (θ) = cos θ and
g(u) = sin θ:

L = f˙g̈ − f¨ġ
= − sin θ(− sin θ) − (− cos θ) cos θ
= 1,
M = 0,
N = f ġ
= cos2 θ.

Hence the principal curvatures are the roots of

1−κ 0
det(F2 − kF1 ) = = 0.
0 cos2 θ − κ cos2 θ

Hence κ = 1. And every tangent direction is a principal vector.

Example 3. Consider a cylinder with the z-axis as its axis and circular cross sections of unit radius. The
parametrization is given as
σ(u, v) = (cos v, sin v, u).

t̂1
n̂ t̂2

The coefficients of the first and second fundamental forms can be computed as

E = 1, F = 0, G = 1, L = 0, M = 0, N = 1.

The principal curvatures are roots of

0−κ 0
0 1−κ

7
So we obtain κ1 = 0 and κ2 = 1. The eigenvectors Ti = (ξi , ηi ), i = 1, 2 of F1−1 F2 are found from solving
the equation
(F2 − κi F1 )Ti = 0.
The results are T1 = λ1 (1, 0)t and T2 = λ2 (0, 1)t for any non-zero λ1 , λ2 ∈ R. Hence the principal vector
t̂1 is along the direction of 1σu + 0σ v , i.e., t̂1 = (0, 0, 1). The principal vector t̂2 is along the direction of
0σ u + 1σ v = (− sin v, cos v, 0), i.e., t̂2 = (− sin v, cos v, 0).

A curve γ on the surface σ is a principal curve if its velocity γ ′ always points in a principal
direction, that is, the direction of a principal vector. At every point on a principal curve, the normal
curvature is a maximum or minimum. The next figure shows some principal curves on the ellipsoid
x2 y 2
+ + z 2 = 1.
12 5

5 Euler’s Formula
Suppose the two principal curvatures κ1 6= κ2 at p on the surface σ. Then by Theorem 1(iii), the
two corresponding principal vectors t̂1 = ξ1 σ u + η1 σ v and t̂2 = ξ2 σ u + η2 σ v must be orthogonal
to each other. Denote by T1 = (ξ1 , η1 )t and T2 = (ξ2 , η2 )t . Replace the T in (12) with Tj , j = 1, 2,
multiply both sides of the equation by Tit to the left, and move the second resulting term to the
right hand side of the equation. This yields

Tit F2 Tj = κj Tit F1 Tj , i, j = 1, 2.

Meanwhile, the orthogonality of the two principal vectors implies that

t̂1 · t̂2 = T1t F1 T2 = 0, from (10).

8
To summarize, we have 
κi , if i = j,
Tit F2 Tj = κi Tit F1 Tj = (13)
0, otherwise
With the principal curvatures and vectors at p, we can evaluate the normal curvature in any
direction.
Theorem 2 Let κ1 , κ2 be the principal curvatures, and t̂1 , t̂2 the two corresponding principal vec-
tors of a patch σ at p. The normal curvature of σ in the direction û = cos θ t̂1 + sin θ t̂2 is

κn = κ1 cos2 θ + κ2 sin2 θ.

Proof Let û = ξσ u + ησ v and T = (ξ, η)t . We first look at the special case κ1 = κ2 = κ. By
Theorem 1(ii), û = ξσ u + ησ v is a principal vector. The normal curvature in the direction û is

κn = T t F2 T (by (9))
t
= κT F1 T (by (12))
= κû · û = κ. (14)

Meanwhile, we have
κ1 cos2 θ + κ2 sin2 θ = κ(cos2 θ + sin2 θ) = κ.
So the theorem holds when the point is umbilic.
Assume κ1 6= κ2 . Therefore by Theorem 1(iii), t̂1 and t̂2 are perpendicular to each other. Let

t̂i = ξi σu + ηi σv , and Ti = (ξi , ηi )t .

Thus,

û = cos θ(ξ1 σ u + η1 σv ) + sin θ(ξ2 σ u + η2 σv )

= (ξ1 cos θ + ξ2 sin θ)σ u + (η1 cos θ + η2 sin θ)σ v .

So we have û = ξσu + ησ v , where

ξ = ξ1 cos θ + ξ2 sin θ,
η = η1 cos θ + η2 sin θ,

The above is written succinctly as T = cos θT1 + sin θT2 . By equation (9) the normal curvature in
the û direction is

κn = (cos θT1t + sin θT2t )F2 (cos θT1 + sin θT2 )

= cos2 θ T1t F2 T1 + cos θ sin θ (T1t F2 T2 + T2t F2 T1 ) + sin2 θ T2t F2 T2
= κ1 cos2 θ + κ2 sin2 θ.

The last step above followed from the equation (13).

Theorem 2 implies that κ1 and κ2 are the maximum and minimum of any normal curvatures
at the point. Equivalently, among all tangent directions at the point, the geometry varies the most
in one principal direction while the least in the other.

9
References
[1] B. O’Neill. Elementary Differential Geometry. Academic Press, Inc., 1966.

[2] A. Pressley. Elementary Differential Geometry. Springer-Verlag London, 2001.

10