USAC Colloquium

Geometry of Bending Surfaces

Andrejs Treibergs

University of Utah

Wednesday, November 6, 2012

Figure: Bender
2. USAC Lecture: Geometry of Bending Surfaces

The URL for these Beamer Slides: “Geometry of Bending Surfaces”

http://www.math.utah.edu/~treiberg/BendingSlides.pdf
3. References

Manfredo doCarmo, Differential Geometry of Curves and Surfaces,

Prentice Hall, Englewood Cliffs, 1976.
Erwin Kreyszig, Differential Geometry, Dover, New York, 1991; orig.
U. Toronto Press, Toronto, 1959.
Andrew Pressley, Elementay Differential Geometry, Springer-Verlag
Limited, London, 2001.
Dirk Struik, Lectures on Classical Differential Geometry, Dover, New
York, 1988; orig. Addison-Wesley Pub., Reading, 1961.
4. Outline.

Surfaces
Examples. Definition. Moving Frame. Coframe. Metric.
Intrinsic Geometry is Preserved by Bending = Local Isometry
Length of Curve. Example of Torus. Rigid Motions Preserve Metric.
Extrinsic Geometry
Second Fundamental Form. Mean Curvature. Gaussian Curvature.
Examples: Sphere, Graph, Torus. Gauss Theorem.
Ruled Surfaces
Moving Frame along Space Curve
Gaussian Curvature of Ruled Surface.
Examples: Hyperbolic Paraboloid, Helicoid, Hyperboloid.
Tanget Developables are Isometric to the Plane.
What Can a Flat Surface be Bent Into?
Non-umbillic Flat Surfaces are Ruled.
Flat Ruled Surfaces are Plane, Cylinder, Cone or Tangent
Developable.
Global Geometry
Complete Flat surfaces are Planes or clinders
Compact, Closed Constant Gaussian Curvature Surface is Sphere.
5. Examples of Surfaces.

Some examples of what should be surfaces.

Graphs of functions

G2 = { x, y , f (x, y ) ∈ R3 : (x, y ) ∈ U}
where U ⊂ R2 is an open set.
Level sets, e.g.,
S2 = {(x, y , z) ∈ R3 : x 2 + y 2 + z 2 = 1}
This is the standard unit sphere.
Parameterized Surfaces, e.g.,
2
n
o
T = (a+b cos ψ) cos θ,(a+b cos ψ) sin θ,b sin ψ) :θ, ψ∈R
is the torus with radii a > b > 0 constructed as a
surface of revolution about the z-axis.
Figure: Surfaces
6. Local Coordinates.

A surface can locally be given by a curvilinear coordinate chart, also

called a parameterization. Let U ⊂ R2 be open. Let

X : U → R3

be a smooth function. Then we want M = X (U) to be a piece of a

surface. At each point P ∈ X (U) we can identify tangent vectors to the
surface. If P = X (a) some a ∈ U, then

∂X
Xi (a) = (a)
∂u i
for i = 1, 2 are vectors in R3 tangent to the coordinate curves. To avoid
singularities at P, we shall assume that all X1 (P)and X2 (P) are linearly
independent vectors. Then the tangent plane to the surface at P is

TP M = span{X1 (P), X2 (P)}.

7. Example of Local Coordinates.

For the graph G2 = {(x, y , z) ∈ R3 : z = f (x, y ) and (x, y ) ∈ U } one

coordinate chart covers the whole surface, X : U → G2 ∩ V = G2 , where

X (u 1 , u 2 ) = (u 1 , u 2 , f (u 1 , u 2 )).

where V = {(u 1 , u 2 , u 3 ) : (u 1 , u 2 ) ∈ U and u 3 ∈ R}.

The tangent vectors are thus
 
1 2 ∂f 1 2
X1 (u , u ) = 1, 0, 1 (u , u ) ,
∂u
  (1)
∂f
X2 (u 1 , u 2 ) = 0, 1, 2 (u 1 , u 2 )
∂u

which are linearly independent for every (u 1 , u 2 ) ∈ U.

8. Definition of a Surface.

Definition
A connected subset M ⊂ R3 is a regular surface if to each P ∈ M, there
is an open neighborhood P ∈ V ⊂ R2 , and a map

X :U →V ∩M

of an open set U ⊂ R2 onto V ∩ M such that

1 X is differentiable. (In fact, we shall assume X is smooth (C ∞ )
2 X is a homeomorphism (X is continuous and has a continuous
inverse)
3 The tangent vectors X1 (a) and X2 (a) are linearly independent for all
a ∈ U.
For simplicity, we shall assume our surfaces are also orientable.
9. Lengths of Curves.

The Euclidean structure of R3 , the usual dot product, gives a way to

measure lengths and angles of vectors. If V = (v1 , v2 , v3 ) then its length
q √
|V | = v12 + v22 + v32 = V • V

If W = (w1 , w2 , w3 ) then the angle α = ∠(V , W ) is given by

V •W
cos α = .
|V | |W |

If γ : [a, b] → M ⊂ R3 is a continuously differentiable curve, its length is

Z b
L(γ) = |γ̇(t)| dt.
a
10. Moving Frame.

It is convenient to find an orthonormal basis {e1 , e2 } for each tangent

space. One way, but not the only way to find an orthonormal basis is to
use the Gram-Schmidt algorithm:

X1 X2 − (X2 • e1 )e1
e1 = , e2 = . (2)
|X1 | |X2 − (X2 • e1 )e1 |

Then e1 and e2 vary smoothly point to point and span the tangent space.
We can also define a unit normal vector which is perpendicular to the
tangent plane by
X1 × X2
e3 = = e1 × e2 .
|X1 × X2 |
The resulting moving frame, {e1 , e2 , e3 } is orthonormal
(
1, if i = j,
ei • ej = δij =
0, otherwise.
11. Dual Coframe.

It is also convenient to introduce the moving coframe of one-forms

(linear functionals). For any V ∈ Tp M,

ω i (V ) = ei • V .

They satisfy duality equations

ω A (eB ) = δ A B .
One forms may be integrated along curves.
If the surface M is given in terms of local parameters X (u 1 , u 2 ) then the
one forms are expresses in terms of differentials du 1 and du 2 . These are
not orthonormal but are dual to coordinate directions

du 1 (X1 ) = 1, du 1 (X2 ) = 0, du 2 (X1 ) = 0, du 2 (X2 ) = 1.

Thus if ω i = p du 1 + q du 2 we get the coefficients from

p = ω i (X1 ), q = ω i (X2 ).
12. Metric or First Fundamental Form.

Then if the vector field is V = v 1 e1 + v 2 e2 then ω i (V ) = v i and the

length squared is

2
2
|V |2 = ω 1 (V ) + ω 2 (V ) .

For short we write the Riemannian Metric

2
2
ds 2 = ω 1 + ω 2

where s is for arclength. It also called the First Fundamental Form.

The area two form is
dA = ω 1 ∧ ω 2 .
Two forms may be integrated along surfaces.
13. Lengths of Curves.

If γ : [a, b] → M is a curve on the surface then we may factor through

the coordinate chart. There are continuously differentiable
u(t) = (u 1 (t), u 2 (t)) ∈ U so that
γ(t) = X (u 1 (t), u 2 (t)) for all t ∈ [a, b].
We write its velocity vector
γ̇(t) = X1 (u 1 (t), u 2 (t)) u̇ 1 (t) + X2 (u 1 (t), u 2 (t)) u̇ 2 (t)
= v 1 (t)e1 (γ(t)) + v 2 (t)e1 (γ(t))
so that the length of the curve
Z
L(γ) = ds
γ
Rbp
= a [ω 1 (γ̇(t))]2 + [ω 2 (γ̇(t))]2 dt.
Rb p
= a [v 1 (t)]2 + [v 2 (t)]2 dt.
Rb
= a |γ̇(t)| dt
14. Graph Example.

For the graph G2 = {(x, y , z) ∈ R3 : z = f (x, y ) and (x, y ) ∈ U } take

the patch X (u 1 , u 2 ) = (u 1 , u 2 , f (u 1 , u 2 )). Hence
   
1 0
X1 =  0  , X2 =  1  ,
f1 f2

Orthonormalizing {X1 , X2 } using (2),

   
1 −f1 f2
1  0  , e2 = q 1 1 + f12 
e1 = q q
2
1 + f1 f1 2 2
1 + f1 1 + f1 + f22
f2

so q
q
f1 f2 1 + f12 + f22
X1 = 1 + f12 e1 , X2 = q e1 + q e2 .
1 + f12 1 + f12
15. Graph Example. -

If ω i = ai du 1 + b i du 2 then ω i (X1 ) = ai and ω i (X2 ) = b i . It follows that

q
q
f1 f2 1 + f12 + f22
ω 1 = 1 + f12 du 1 + q du 2 , ω 2 = q du 2 .
1 + f1 2 1 + f1 2

and so the Riemannian metric is

ds 2 = (ω 1 )2 + (ω 2 )2
= (1 + f12 ) (du 1 )2 + 2f1 f2 du 1 du 2 + (1 + f22 ) (du 2 )2 ,

as expected. Also
q
dA = ω 1 ∧ ω 2 = 1 + f12 + f22 du 1 ∧ du 2 ,

also as expected. It gives the usual formula for area

Z q
A(X (D)) = 1 + f12 + f22 du1 du2 .
D
16. Intrinsic Geometry is Preserved by Bending.

Geometric quantities determined by the metric are called intrinsic. A

bending of one surface may into another is a local isometry, a
diffeomorphism that preserves lengths of curves, hence all intrinsic
quantities. Equivalently, the Riemannian metrics are preserved. Thus if

f : (M 2 , ds 2 ) → (M̃ 2 , ds̃ 2 )

is an isometry, then f : M n → M̃ n is a diffeomorphism and f ∗ ds̃ 2 = ds 2 .

This means that for at every point u ∈ M, and every tangent vector
V , W of M, the corresponding inner products are the same in M and M̃:

ds 2 (V (u), W (u))u = ds̃ 2 (dfu (V (u)), dfu (V (u)))ũ

where ũ = ũ(u) correspond under the map and dfu : Tu M → Tũ M̃ is the
differential. We have written the first fundamental form
ds 2 (V , W ) = v 1 w 1 + v 2 w 2 .
WARNING: funtional analysts and geometric group theorists define
“isometry” in a slightly different way.
17. Example of the Torus.

Lets work out the geometry of a torus T2 . Suppose that the torus has
radii 0 < b < a and is given parametrically as
 
cos θ(a + b cos ψ)
X (θ, ψ) =  sin θ(a + b cos ψ)  ,
b sin ψ
 
− sin θ
Xθ (θ, ψ) = (a + b cos ψ)  cos θ  = (a + b cos ψ)e1 ,
0
 
− cos θ sin ψ
Xψ (θ, ψ) = b  − sin θ sin ψ  = b e2 ,
cos ψ
 
cos θ cos ψ
Xθ × Xψ
=  sin θ cos ψ  = e3
|Xθ × Xψ |
sin ψ
18. Example of the Torus. +

By duality,

ω 1 = (a + b cos ψ)dθ, ω 2 = b dψ

Thus the metric of T2 is

ds 2 = (ω 1 )2 + (ω 2 )2 = (a + b cos ψ)2 dθ2 + b 2 dψ 2 .

19. Extrinsic Geometry.

Extrinsic Geometry deals with how M sits in its ambient space.

Near P ∈ M, the surface may be parameterized as the graph over its
tangent plane, where f (u1 , u2 ) is the “height” above the tangent plane

X (u1 , u2 ) = P + u1 e1 (P) + u2 e2 (P) + f (u1 , u2 )e3 (p). (3)

So f (0) = 0 and Df (0) = 0. The Hessian of f at 0 gives the shape

operator at P. It is also called the Second Fundamental Form.

∂2 f
hij (P) = (0)
∂ui ∂uj

The Mean Curvature and Gaussian Curvature at P are

1
H(P) = tr(hij (P)), K (P) = det(hij (P)).
2
20. Sphere Example.

The sphere about zero of radius r > 0 is an example

S2r = {(x, y , z) ∈ R3 : x 2 + y 2 + z 2 = r 2 }.
Let P = (0, 0, −r ) be the south pole. By a rotation (an isometry of R3 ),
any point of S2r can be moved to P with the surface coinciding. Thus the
computation of H(P) and K (P) will be the same at all points of S2r . If
e1 = (1, 0, 0), e2 = (0, 1, 0) and e3 = (0, 0, 1), the height function of (3)
near zero is given by
q
f (u1 , u2 ) = r − r 2 − u12 − u22 .
The Hessian is
r 2 −u22
 
−u1 u2
∂2f ( r 2 −u12 −u22
3/2
) (r 2 −u 2 −u 2 3/2 
1 2)
(u) =  −u1 u2 u22 r 2 −u12
∂ui uj

3/2 3/2
( r 2 −u12 −u22
) (r 2 −u12 −u22 )
Thus the second fundamental form at P is
1 
r 0 1 1
hij (P) = fij (0) = 1 so H(P) = and K (P) = 2
0 r r r
21. Graph Example.

If X (u1 , u2 ) = (u1 , u2 , f (u1 , u2 )), by correcting for the slope at different

points one finds for all (u1 , u2 ) ∈ U,

1 + f22 f11 − 2f1 f2 f12 + 1 + f12 f22

H(u1 , u2 ) =
3/2 ,
2 1 + f12 + f22

2
f11 f22 − f12
K (u1 , u2 ) =
2 .
1 + f12 + f22
22. 2nd Fundamental Form in Terms of Moving Frame.

Because eA · eA = 1, taking the directional derivative deA · eA = 0, so

that deA ⊥ eA and we may express the rate of rotation of the frame

deA = ωA B eB ,

where summation over repeated indices is assumed. ωA B are called

connection forms. Differentiation of eA · eB = δAB implies ωA B is skew
and satisfies
dω A = ω B ∧ ωB A . (4)
Also, differentiating the normal de3 = ω3 i ei , where lower case Roman
indices run over i, j, k, . . . = 1, 2. Moreover

ω3 i = −hij ω j

recovers the second fundamental form.

23. Both Expressions of 2nd Fundamental Form Coincide.

Indeed, for surfaces of the form (3),

X (u1 , u2 ) = P + u1 e1 (P) + u2 e2 (P) + f (u1 , u2 )e3 (p),

the normal vector equals

 
−f1
1 −f2 
e3 = q
1 + f12 + f22 1

so that at P where f (0) = f1 (0) = f2 (0) = 0, we get

2
X ∂2 f
de3 (P) = − (0) du i ej (P)
∂ui ∂uj
i,j=1
2
X
=− hij (P)ω i ej (P).
i,j=1
24. Compute the Second Fundamental Form of the Torus.

     
cos θ cos ψ − sin θ cos ψ − cos θ sin ψ
de3 = d  sin θ cos ψ  =  cos θ cos ψ  dθ +  − sin θ sin ψ  dψ
sin ψ 0 cos ψ
d
= cos ψ dθ e1 + e2 = ω3 1 e1 + ω1 2 e2
ψ
It follows that
cos ψ
ω3 1 = −h1j ω j = ω1
a + b cos ψ
1
ω3 2 = −h2j ω j = ω 2 .
b
Thus, the second fundamental form is diagonal too and
!
cos ψ
− a+b cos ψ 0
hij = .
0 − b1
25. Mean and Gauss Curvatures of Torus.

Since !
cos ψ
− a+b cos ψ 0
hij = .
0 − b1
it follows that
 
1 cos ψ
H(P) = 2 tr(hij (P)) = − 12 a+b cos ψ + 1
b
cos ψ
K (P) = det(hij (P)) = b(a+b cos ψ) .

So for the outside part of the torus |ψ| < π2 , the surface is on one side of
the tangent plane so K > 0.
On the inside of the torus π2 < |ψ| ≤ π, the surface is a saddle on both
sides of the tangent plane where K < 0.
On the circles on top and bottom ψ = ± π2 , the ψ direction leaves the
tangent plane but the θ directions stay in it, therefore there K = 0.
26. The Plane Can Be Bent into a Cylinder: They Are Isometric.

One imagines that one can roll up a piece of paper in R3 without

changing lengths of curves in the surface. Thus the plane P2 and the
cylinder Z2 are locally isometric. Let us check by computing the
Riemannian metrics at corresponding points. For any (u1 , u2 ) ∈ R2 the
plane is parameterized by X (u 1 , u 2 ) = (u 1 , u 2 , 0) so e1 = X1 = (1, 0, 0),
e2 = X2 = (0, 1, 0) and so ω 1 = du 1 , ω 2 = du 2 and thus
dsP2 = (du 1 )2 + (du 2 )2 .
Cylinder Z2 = {(x, y , z) : y 2 + z 2 = r 2 } of radius r is parameterized
Z (v 1 , v 2 ) = (v 1 , r cos v 2 , r sin v 2 ) so ẽ1 = Z1 = (1, 0, 0),
r ẽ2 = Z2 = (0, −r sin v 2 , r cos v 2 ) and so ω̃ 1 = dv 1 , ω̃ 2 = rdu 2 and thus

dsZ2 = (dv 1 )2 + r 2 (dv 2 )2 .

The map f : X (u1 , u2 ) 7→ Z (u1 , u2 /r ) is an isometry because it pulls
back the same metric and lengths of curves agree: v 2 = u 2 /r means
f ∗ (dv 2 ) = du 2 /r so
f ∗ (dsZ2 ) = f ∗ (dv 1 )2 + r 2 f ∗ (dv 2 )2 = (du 1 )2 + (du 2 )2 = dsP2 .
27. Caps of Spheres are Not Rigid.

By manipulating half of a rubber ball that has been cut through its
equator, one sees that the cap can be deformed into a football shape
without distorting intrinsic lengths of curves and angles of vectors. The
spherical cap is deformable through isometries: it is not rigid. (Rigid
means that any isometry has to be a rigid motion of R3 : composed of
rotations, translations or reflections.)
It turns out by Herglotz’s Theorem, all C 3 closed K > 0 surfaces (hence
surfaces of convex bodies which are simply connected) are rigid.
It is not true for nonconvex surfaces. The lid may be glued on up side
down or right side up to give locally isometric surfaces of revolution (they
can be bent into each other), but they are not congruent.
28. Gauss’s Excellent Theorem.

So far, the formula for the Gauss Curvature has been given in terms of
the second fundamental form and thus may depend on the extrinsic
geometry of the surface. However, Gauss discovered a formula that he
deemed excellent:
Theorem (Gauss’s Theorema Egregium 1828)
Let M 2 ⊂ R3 be a smooth regular surface. Then the Gauss Curvature
may be computed intrinsically from the metric and its first and second
derivatives.
In other words, the Gauss Curvature coincides at corresponding points of
isometric surfaces.
The word has the same Latin root as “egregious” or “gregarious.”
29. Proof of Theorema Egregium.

The proof depends on a fact about Euclidean Three Space, called the
Riemann Curvature Equation. For ANY moving frame {e1 , e2 , e3 }, the
connction forms satisfy
3
X
dωA C − ωA B ∧ ωB C = 0. (5)
B=1

It is a tensor: it is independent of choice of frame. It is easily seen to be

zero in the standard basis of E3 .
Applied to the ω1 2 , we see that the Gauss Curvature may be computed
from differentiation of ω1 2 . Indeed, by (5),
2
X
dω1 2 = ω1 3 ∧ ω3 2 = − h1i h2j ω i ∧ ω j
i,j=1

= −(h11 h22 − h12 h21 )ω 1 ∧ ω 2 = −K ω 1 ∧ ω 2 .

Now ω1 2 and therefore dω1 2 and K can be computed intrinsically.

30. How Gaussian Curvature can be Measured in the Surface.

It turns out that the Gaussian Curvature can be measured intrinsically in

another way.
The ball of radius r around a point in a surface P ∈ M 2 is the set

Br (P) = {x ∈ M 2 : dist(x, P) < r }.

If the area of the ball for small r is expanded in series, the Gaussian
curvature at P appears as a coefficient in the expansion. It can be
viewed as a correction to the Euclidean area growth, which is quadratic
on the nose.
π
Area(Br (P)) = πr 2 + K (P)r 4 + · · ·
12
31. Illustrate Gauss’s Theorem for the Torus.

Recall the coframe of T2 is

ω 1 = (a + b cos ψ)dθ, ω 2 = bdψ

We solve (4) for ω1 2 = −ω2 1 = p dθ + q dψ such that

bdψ ∧ (−p dθ − q dψ) = ω 2 ∧ ω2 1 = dω 1 = −b sin ψ dψ ∧ dθ,

(a + b cos ψ)dθ ∧ (p dθ + q dψ) = ω 1 ∧ ω1 2 = dω 2 = 0.

The second says q = 0 and the first p = sin ψ so ω1 2 = sin ψ dθ.

Differentiating,

−Kb(a + b cos ψ) dθ ∧ dψ = −K ω 1 ∧ ω 2 = dω1 2 = cos ψ dψ ∧ dθ.

Therefore, we recover the Gauss Curvature we found before extrinsically

cos ψ
K= .
b(a + b cos ψ)
32. Bending.

We imagine starting with a flexible and inextensible two dimensional

surface S ⊂ E3 and ask if it is possible to bend it into another surface
M ⊂ E3 ?
Let’s call f : S → M, the mapping of corresponding points a bending if it
preserves lengths along the surface. So f is a local isometry: it preserves
the metric, hence all intrinsic geometry, such as angles, areas, geodesic
curvatures of curves and the Gauss curvature of the surface. However it
may have different second fundamental forms at corresponding points.
Let f : S → M be a bending. We assume it is smooth: if X : U → S is a
local parameterization for S, then X̃ = f ◦ X : U → M is a local
parameterization of M. It is a local isometry: it preserves lengths of
vectors:
∂X ∂X ∂ X̃ ∂ X̃
gij = i
• j = • = g̃ij .
∂u ∂u ∂u i ∂u i
33. Rigid Motion Theorem.

There are trivial bendings, the rigid motions, consisting of rotations,

reflections and translations of space. As rigid motions preserve lengths
and corresponding moving frames, both the metrics and the second
fundamental forms are preserved.
Theorem (Invariance under Rigid Motions.)
Let S ∈ E3 be a surface and X : U → E3 be a local coordinate chart for
S. Let B : E3 → E3 be a rigid motion, i.e., B(x) = Rx + v where
R ∈ O(3) is an orthogonal matrix and v ∈ E3 a translation vector. Let
X̃ = B ◦ X . Then after an appropriate choice of normal vectors, the first
and second fundamental forms coincide at corresponding points

ds 2 = ds̃ 2 , hij = h̃ij .

Conversely, if there are two connected surfaces whose parameterizations

X , X̃ : U → E3 satisty ds 2 = ds̃ 2 and hij = h̃ij , then X̃ = B ◦ X for some
rigid motion B.
34. Ruled Surfaces, Cylinders, Cones and Tangent Developables.

We consider what surfaces are bendings of pieces of the plane. it turns

out they must be planes, cylinders, cones and tangent developables.
Imagine sweeping out a surface by moving a line. Such a surface is called
a ruled surface. It may be given by specifying a unit speed regular space
curve α(u 2 ) and a unit direction vector V (u 2 ). Locally it is
parameterized by

X (u 1 , u 2 ) = α(u 2 ) + u 1 V (u 2 ).

Examples are solutions of the the equations in E3 giving hyperbolic

paraboloids
z = xy
or hyperboloids of one sheet

x 2 + y 2 − z 2 = 1.
35. Generalized Cylinders.

When direction V = V (u 2 ) is constant,

then the ruling lines are parallel. If also α̇ is
not parallel to V then the surface C is a
Generalized Cylinder.
Let α(u 2 ) be a unit speed curve in the
plane perpendicular to V . Then X1 = V ,
X2 = α0 (u 2 ) and the metric is
Figure: Generalized Cylinder dsC2 = (du 1 )2 + (du 2 )2 .
X (u 1 , u 2 ) = α(u 2 ) + u 1 V .
If P is surface in the plane, it has metric
dsP2 = (dv 1 )2 + (dv 2 )2 .
The map F : X (u 1 , u 2 ) 7→ Y (u 1 , u 2 ) is an
isometry: if γ(t) = X (u 1 (t), u 2 (t)) is a
curve in C then F ◦ γ is a curve in P with
Figure: Unrolled Generalized the same length because
Cylinder in Plane
Y (v 1 , v 2 ) = (v 1 , v 2 , 0). |γ 0 (t)|C = |(F ◦ γ)0 |P .
36.Generalized Cones.

If V (u 2 ) is a curve on the unit sphere and

and α(u 2 ) = P is constant, all the ruling
lines pass through P ∈ E3 , then the surface
V is a Generalized Cone. If α0 is a unit
vector, then its metric is
dsV2 = (du 1 )2 + (u 1 )2 (du 2 )2 .
Figure: Generalized Cone If Q is surface Y in the plane, it has metric
X (u 1 , u 2 ) = P + u 1 V (v 2 ). (in polar coordinates)
dsQ2 = (dv 1 )2 + (v 1 )2 (dv 2 )2 .

The map F : X (u 1 , u 2 ) 7→ Y (u 1 , u 2 ) is an
isometry: if γ(t) = X (u 1 (t), u 2 (t)) is a
curve in V then F ◦ γ is a curve in Q with
the same length because
Figure: Unrolled Gen. Cone
|γ 0 (t)|V = |(F ◦ γ)0 |Q .
Y (v 1 , v 2 ) =
(v 1 cos v 2 , v 1 sin v 2 , 0)
37. Moving Frame along Space Curve.

Let α(u 2 ) : (a, b) → R2 be a smooth space curve. Assume that

T 0 (u 2 ) = α00 (u 2 ) 6= 0. Since T • T = 1 we have T • T 0 = 0 so that T 0 is
orthogonal to T . If

T0
N= , κ = |T 0 |, B =T ×N
|T 0 |

then {T , N, B} the orthonormal frame along the curve. Similarly by

differentiating N • N = 1, T • N = 0, T • B = 0, N • B = 0 and
B • B = 1, we find the Frenet Equations

T 0 = κN, N 0 = −κN + τ B, B 0 = −τ N.

for some functions κ(u 2 ) > 0 and τ (u 2 ) called the curvature and torsion
of a space curve.
If the function τ ≡ 0 then B is constant and the curve is in the plane B ⊥ .
If τ 6= 0 then the curve leaves its osculating plane and is called twisted.
38. Metric and Curvature of a Ruled Surface.

If α(u 2 ) is parametrized by arclength and V (u 2 ) unit, the ruled surface is

X (u 1 , u 2 ) = α(u 2 ) + u 1 V (u 2 ).

T = α0 and V • V 0 = 0. By solving dω12 = −K ω 1 ∧ ω 2 ,

X1 = V ; X2 = T + u 1 V 0
ω 1 = du 1 + T • V du 2 , ω 2 = |T + u 1 V 0 − (T • V )V |du 2
ds 2 = (du 1 )2 + 2V • T du 1 du 2 + |T + u 1 V 0 |2 (du 2 )2 .
T • V 0 + u 1 |V 0 |2
dω 1 = 0, dω 2 = du 1 ∧ du 2 ,
|T + u 1 V 0 − (T • V )V |
T • V 0 + u 1 |V 0 |2
ω12 = du 2 ,
|T + u 1 V 0 − (T • V )V |
(T • V )2 |V 0 |2 − |V 0 |2 + (T • V 0 )2 (T • V × V 0 )2
K= = −
|T + u 1 V 0 − (T • V )V |4 |T + u 1 V 0 − (T • V )V |4
If X is a generalized cylinder, V is constant, V 0 = 0 and K = 0.
If X is a tangent developable, V = T , V 0 = T 0 = κN so K = 0.
39. Hyperbolic Paraboloid is a Ruled Surface.

√
Set u 2 = x and u 1 = y 1 + x 2 ,
 2  
u 1 0
u
X (u 1 , u 2 ) =  0  + p  1 .
0 1 + (u 2 )2 u 2

T = (1, 0, 0), V = [1 + (u 2 )2 ]−1/2 (0, 1, u 2 ),

V 0 = [1 + (u 2 )2 ]−3/2 (0, −u 2 , 1) we get
T • V = 0, T • V × V 0 = [1 + (u 2 )2 ]−1 so

[1 + (u 2 )2 ]2
K =−
{[1 + (u 2 )2 ]2 + (u 1 )2 }2
Figure: Hyperbolic Paraboloid.
1
=− .
{1 + x 2 + y 2 }2
The surface z = xy is
obviously a ruled surface. Or use the formula for a graph.
40. Helicoid is Example of Ruled Surface..

If α(u 2 ) = (0, 0, u 2 ) the vertical line and V (u 2 ) = (cos u 2 , sin u 2 , 0)

rotates about a circle in the x-y plane, the resulting surface is a helicoid.
Plugging
 into2  the formula or solving
 1from 2scratch,

cos u −u sin u p
X1 =  sin u 2  = e1 , X2 =  u 1 cos u 2  = 1 + (u 1 )2 e2 .
0 1
p
so ω 1 = du 1 , ω 2 = 1 + (u 1 )2 du 2 . Thus

u1
ω12 = √ du 2 , dω12 = 1
du 1 du 2
1+(u 1 )2 {1+(u 1 )2 }3/2

1
and K = − .
[1+(u 1 )2 ]2

For different u 2 the rulings are skew lines in

space, whose closest points occur on the
Figure: Helicoid centerline α(u 2 ). This line is called the line
of stricture.
41. Hyperboloid is Example of Ruled Surface.

V (u 2 ) = √1 (− sin u 2 , cos u 2 , 1) the 45◦

2
vector in the α⊥ plane, the resulting surface
is a hyperboloid of one sheet. Then

 2
u1
x 2 + y 2 = cos u 2 − √ 2
sin u 2 +
 2
u1
sin u 2 + √ 2
cos u 2

(u 1 )2
Figure: Hyperboloid =1+ 2
2
=1+z .
If α(u 2 ) = (cos u 2 , sin u 2 , 0)
is the horizontal circle and
42. Hyperboloid is Example of Ruled Surface.

Then
− sin u 2 − sin u 2 − cos u 2
     
1 1
T = α0 =  cos u 2  , V = √  cos u 2  , V 0 = √  − sin u 2 
0 2 1 2 0
so
1 1 1
T • V = √ , T • V 0 = 0, |V 0 |2 = , T • V × V 0 = −
2 2 2
so by the curvature formula

|T • V × V 0 |2 1
K =− =− .
|T + u 1 V 0 − (T • V )V |4 [1 + (u 1 )2 ]2
43. Tangent Developable Surface.
Let’s consider a special ruled surface M, the
tangent surface. This time, let’s assume
that V (u 2 ) = α0 (u 2 ). Since T (u 2 ) = α0 (u 2 )
is a unit vector in the tangent direction, the
surface is swept out by lines tangent to a
space curve. In this case, we must assume
that T 0 (u 2 ) = α00 (u 2 ) 6= 0.
The surface is locally given by

X (u 1 , u 2 ) = α(u 2 ) + u 1 T (u 2 ).

Then the tangent vectors are

X1 = T (u 2 ), X2 = T (u 2 )+u 1 κ(u 2 ) N(u 2 )

so that e1 = N(u 2 ) and e2 = T (u 2 ) is

Figure: Tangent Developable moving frame along M for all u 1 > 0. Note
that the tangent vectors are independent iff
κ(u 2 ) > 0.
44. Tangent Developable Surface.

The fact that the e3 is constant along the generator for all u 1 > 0
distinguishes the developable surfaces among the ruled surfaces. Another
description is that this surface is the envelope of a family of planes in E3 .
Here the family is given for u 2 ∈ (a, b) by the osculating planes of α:

{Z ∈ E3 : (Z − α(u 2 )) • B(u 2 ) = 0}.

For X (u 1 , u 2 ) = α(u 2 ) + u 1 T (u 2 ), we have

X1 = T (u 2 ), X2 = T (u 2 ) + u 1 κ(u 2 ) N(u 2 ).

Since e1 = N(u 2 ) and e2 = T (u 2 ), the dual frames are

ω 1 = u 1 κ du 2 , ω 2 = du 1 + du 2 ;

and the metric is

ds 2 = (ω 1 )2 + (ω 2 )2 = (du 1 )2 + 2du 1 du 2 + (1 + (u 2 )2 κ2 )(du 2 )2 .

45. Metric of the Tangent Developable Surface.

The metric is

ds 2 = (ω 1 )2 + (ω 2 )2 = (du 1 )2 + 2du 1 du 2 + (1 + (u 2 )2 κ2 )(du 2 )2 .

Note that the metric DOES NOT DEPEND on τ . Thus if γ(u 2 ) is the
unit speed PLANE CURVE with the same curvature κ(u 2 ) (and torsion
zero) then
Y (u 1 , u 2 ) = γ(u 2 ) + u 1 γ 0 (u 2 )
is a parameterization of a piece of that plane and has the SAME
METRIC as X (u 1 , u 2 ) at corresponding points. This is called developing
the surface into the plane. Because Y (u 1 , u 2 ) is planar, its Gauss
Curvature is dead zero. Since it’s isometric, the curvature of X (u 1 , u 2 ) is
dead zero too. Thus a piece of the plane can be bent into a tangent
developable surface.
46. Classifying the Flat Surfaces.

Theorem
Let P ∈ M ∈ E3 be a point in a flat surface. Suppose that M is not
umbillic at P. Then M is a ruled surface in some neighborhood of P.

The hypotheses tell us that hij (P) 6= 0, so it is nonzero in a coordinate

neighborhood U ⊂ M. By rotating the orgthonormal frame at points of
U we may arrange that at all points of U,
 
0 0
hij = (6)
0 h22

where h22 6= 0 in U. For any P0 ∈ U, let γ(u 1 ) ∈ U be a curve such that

γ(0) = P0 and
γ 0 (u 1 ) = e1 (γ(u 1 ))
for all u 1 . We show that γ(s 1 ) is a straight line in E3 , since P0 is any, U
is foliated by generating lines. It suffices to show that

γ 00 (0) = De1 e1 = 0.
47. Classifying the Flat Surfaces. +

But
De1 e1 = ω1 2 (e1 )e2 + ω1 3 (e1 )e3 . (7)
Now ω1 3 = h11 ω 1 + h12 ω 2 = 0 by (6) so the second term vanishes. Also
h11 = h12 = 0 in U so

0 = dh11 = h111 ω 1 + h112 ω 2 + 2h12 ω1 2 ,

0 = dh12 = h121 ω 1 + h122 ω 2 + h22 ω1 2 + h12 ω2 1 .

The first tells us h111 = h112 = 0. By the Codazzi equations,

h112 = h121 = 0. So the second says

0 = e1 h12 = h121 ω 1 (e1 ) + h122 ω 2 (e1 ) + h22 ω1 2 (e1 ) + h12 ω2 1 (e1 )

= 0 + 0 + h22 ω1 2 (e1 ) + 0.

But h22 6= 0 so
ω1 2 (e1 ) = 0.
and the first term of (7) vanishes also.
48. Flat and Ruled Implies Cylinder, Cone or Developable.

Theorem
Let M be a flat ruled surface. Then M consists of pieces of planes,
generalized cylinders, generalized cones and tangent developables.

Locally, M is given by a unit speed space curve α(u 2 ) and a unit tangent
vector field V (u 2 ) with
X (u 1 , u 2 ) = α(u 2 ) + u 1 V (u 2 ).
From the computation for general ruled surfaces, curvature vanishes if
and only if at all points,
T • V × V 0 = 0, (8)
in other words T , V and V 0 are linearly dependent.
49. Flat and Ruled Implies Cylinder, Cone or Developable. +

In the first case, assume that V 0 = 0 for in a neighborhood U ⊂ M.

Then V is constant and U is a generalized cylinder.
Now assume V 0 6= 0 in a neighborhood. Since V • V 0 = 0, the vectors V
and V 0 are independent. By (8),

T (u 2 ) = f (u 2 )V (u 2 ) + g (u 2 )V 0 (u 2 ) (9)

for some smooth functions f and g .

In case f = g 0 in a neighborhood U,

T = α0 = g 0 V + gV 0 = (gV )0

Thus
α − gV = P
is constant, so that

X (u 1 , u 2 ) = α(u 2 ) + u 1 V (u 2 ) = P + (g (u 2 ) + u 1 )V (u 2 ).

Hence u is a generalized cone.

50. Flat and Ruled Implies Cylinder, Cone or Developable. + +

Now assume V 0 6= 0 and f 6= g 0 in a neighborhood of M. Define a new

curve and coordinate
u 1 + g (u 2 )
α̃(u 2 ) = α(u 2 ) − g (u 2 )V (u 2 ), ũ 1 = .
f (u 2 ) − g 0 (u 2 )

Hence, by (9)
α̃0 = T − g 0 V − gV 0 = (f − g 0 )V
thus

X (u 1 , u 2 ) = α + u 1 V = α̃ + gV + u 1 V = α̃ + ũ 1 (f − g 0 )V
= α̃(u 2 ) + ũ 1 α̃0 (u 2 ).

Hence the surface is a tangent developable in U.

It may be that none of the conditions considered are satisfied. In
different open pieces of M, the surface may be any of a cylinder, cone or
tangent developable, which meet along segments of generators.
51. Unrolling A Sheet of Paper with Circle Cut Out.
so that the circle becomes helical,

α(t) = (a cos t, a sin t, bt),

its curvature as a space curve must satisfy

a 1 2 1
+ b2 ≤


K≤ a2 +b 2
or a − 2K 4K 2
.
Figure: Circular Hole
That is why the farthest the circle can be
The curvature K of the hole unrolled is so the circle is the edge of
1
cannot be decreased in the regression , e.g., at pitch b = 2K with
1
paper. If the paper is bent radius a = 2K .
52. Bending Complete Flat Surfaces.

If we ask to bend the entire plane, then we must rule out surfaces whose
generating lines cross. This is an example of global differential geometry.
It was proved amazingly recently. “Complete” means that any intrinsic
straight line in S can be continued forever without running into a
boundary of S.

Theorem (P. Hartman & L. Nirenberg, 1959)

Let S be a complete, connected surface with zero Gaussian curvature.
Then S is a cylinder or plane.
53. Bending the Round Sphere.

There are theorems about the rigidity of other surfaces. In contrast to

zero curvature, the following theorem was proved much earlier.

Theorem (H. Liebmann 1899)

Let S be a compact connected, regular surface with constant Gaussian
curvature K . Then S is the round sphere.
54. Proof of Liebmann’s Theorem.

We outline Hilbert’s Proof which appeared in the second ever volume of

Proceedings of the American Mathematical Society, 1901.
The first thing that is proved is that if S is a compact surface then its
Gaussian curvature must be positive

As M ⊂ E3 is compact, there is a smallest

sphere S that contains M. Lets say the
radius is 1/k. Let P be a point in common
and let e3 be the inward normal. As M is
inside the sphere we can compare the
Hessian of M with the Hessian of S in the
e3 direction at P

hij ≥ kδik .
Figure: M and smallest
surrounding sphere S There K = det(hij ) ≥ k 2 > 0.
55. Proof of Liebmann’s Theorem. +

Next we √show that the√surface is umbillic: the eigenvalues k1 and k2 are

equal to K so hij = K δij . If not, then we may assume that
k1 (x) ≤ k2 (x) = K /k1 (x) √
are not constant, and a some point P ∈ M,
k1 (P) = minx∈M k1 (x) < K . At the same point k2 (x) is a global
maximum. Thus k1 − k2 < 0 at P. On the other hand, if we take a
moving frame near P so that e1 and e2 are eigendirections of hij for k1
and k2 respectively, then
 
k1 0
hij =
0 k2

in a neighborhood of P. By a computation of second (covariant)

derivatives of hij ,

h1122 − h2211 = h11 h22 (h11 − h22 ) < 0. (10)

On the other hand, at a minimum (h11 )22 ≥ 0 and at a maximum

(h22 )11 ≤ 0 so that at P, the left side of (10) is nonnegative, which is a
contradiction.
56. Proof of Liebmann’s Theorem. + +

Finally, we show that a connected

√ surface whose second fundamental
form is hij = kδij , where k = K is constant, is a piece of the sphere.
If we consider the vector function
1
Y =X+ e3
k
its derivatives are
2
1 1X p
D ei Y = D e i X − D ei e3 = ei + ω3 (ei )ep
k k
p=1
2 2
1 X 1 X
= ei − hpq ω q (ei )ep ei − kδpq δ q i ep
k k
p,q=1 p,q=1

= ei − ei = 0.
Thus Y is a fixed point and X is a fixed distance from Y :
1
|X − Y | = .
k
57.
Thanks!