DIFFERENTIAL GEOMETRY OF CURVES AND SURFACES

3. Regular Surfaces
3.1. The definition of a regular surface. Examples. The notion of surface we are
going to deal with in our course can be intuitively understood as the object obtained by a
potter full of phantasy who takes several pieces of clay, flatten them on a table, then models
of each of them arbitrarily strange looking pots, and finally glue them together and flatten
the possible edges and corners. The resulting object is, formally speaking, a subset of the
three dimensional space R3 . Here is the rigorous definition of a surface.
Definition 3.1.1. A subset S ⊂ R3 is a1 regular surface if for any point P in S one can find
an open subspace U ⊂ R2 and a map ϕ : U → S of the form
ϕ(u, v) = (x(u, v), y(u, v), z(u, v)),
for (u, v) ∈ U with the following properties:
1. The function ϕ is differentiable, in the sense that the components x(u, v), y(u, v) and
z(u, v) have partial derivatives of any order.
2. For any Q in U, the differential map d(ϕ)Q : R2 → R3 is (linear and) injective.
3. There exists an open subspace V ⊂ R3 which contains P such that ϕ(U) = V ∩ S
and moreover, ϕ : U → V ∩ S is a homeomorphism2.
The pair (U, ϕ) is called a local parametrization, or a chart, or a local coordinate system
around P . Conditions 1 and 2 say that (U, ϕ) is a regular patch. See also Figure 1.

P V

∩
S

Figure 1. The pair (U, ϕ) is a local parametrization

around P on the surface S.

Now let’s try to understand the definition. In a less precise, but (hopefully!) more intuitive
formulation, condition 3 is saying that any point P is contained in one of the strange looking
pots made by the potter to be glued together; the map ϕ is just the “continuous” deformation
(done by the potter’s hands) of the flat piece of clay U into V ∩ S. Condition 1 is just meant
to let us do differential geometry, that is, study these objects by using derivatives (in the
1Terminology: in this course, whenever we say surface we actually mean regular surface.
2That is, it is continuous, bijective, and its inverse ϕ−1 : V ∩ S → U is continuous as well (here V ∩ S is
equipped with the topology of subspace of R3 ).
1
2

same way as we only considered differentiable curves in the previous two chapters of the
course).
But what’s hiding behind condition 2? Answer: it guarantees that there is a tangent
plane to the surface at any point. Here are the detailed explanations. First let us recall from
Vector Calculus that the differential of ϕ : U → R3 ,
ϕ(u, v) = (x(u, v), y(u, v), z(u, v))
at a point Q is the linear map d(ϕ)Q : R2 → R3 whose matrix is the Jacobi matrix
 ′ 
xu (Q) x′v (Q)
(Jϕ)Q :=  yu′ (Q) yv′ (Q)  .
zu′ (Q) zv′ (Q)
Here the entries of the matrix from above are
∂x ′ ∂x
x′u = , xv = . . . etc.
∂u ∂v
that is, the partial derivatives of the components x, y, z of ϕ. A result from Linear Algebra
says that the map d(ϕ)Q is injective if and only if the rank of its matrix is equal to 2. In turn,
this is equivalent to the fact that the two columns of the matrix are linearly independent
vectors in R3 . These two vectors have a natural geometric interpretation, as follows. Let
the coordinates of Q be (u0 , v0 ). We consider the coordinate curve u 7→ ϕ(u, v0 ), which goes
through ϕ(Q). The tangent vector to this curve at u0 is
 
d ∂ϕ ∂x ∂y ∂z
|u=u0 ϕ(u, v0 ) = (Q) = (Q), (Q), (Q)
du ∂u ∂u ∂u ∂u
which is the first column of J(ϕ)Q . Similarly, the tangent vector to the other coordinate
curve, namely v 7→ ϕ(u0, v), at v0 is the other column of J(ϕ)Q .
In conclusion, condition 2 in the definition is saying that for any Q in U, the tangent vectors
ϕ′u (Q), ϕ′v (Q) to the coordinate curves (which are the same as the columns of the Jacobi
matrix J(ϕ)Q ) are linearly independent.
In this way, we can define the tangent plane to S at ϕ(Q) as the plane through ϕ(Q) which
is parallel to ϕ′u (Q) and ϕ′v (Q). We will get back to the notion of tangent plane later on.
Examples of surfaces. 1. Let S be the horizontal coordinate plane, which consists of all
points (x, y, 0) with x, y in R. We claim that S is a surface, in the sense of Definition 3.1.1.
To see that, we take U to be the whole R2 and ϕ : R2 → S given by
ϕ(u, v) = (u, v, 0),
for all (u, v) in R2 . It s obvious that any point P = (x, y, 0) of S is in the image of ϕ (we
could say that (U, ϕ) is a global parametrization). Now let’s check the conditions 1, 2, and
3 above. First, al three components of ϕ are differentiable (i.e. they have partial derivatives
of any order). Second, the Jacobi matrix of ϕ at a point Q = (u, v) is
 
1 0
(Jϕ)Q :=  0 1  .
0 0
Its columns are linearly independent. Third, we take V = R3 , we note that V ∩ S = S,
and we see that ϕ : R2 → S is a homeomorphism (it is obviously continuous, bijective, and
its inverse is given by ϕ−1 (x, y, 0) = (x, y), which is also continuous, as restriction of the
projection map R3 → R2 , (x, y, z) 7→ (x, y) to S). See also Figure 2.
 3

( x , y , 0 )

Figure 2. The horizontal coordinate plane is an exam-

ple of a surface (probably the simplest one).

2. In the same way as before, if U is any open subset of the horizontal coordinate plane
(for instance the inside of a circle), then U is a surface. See Figure 3.

Figure 3. A thin piece of clay on a table, not touched

by the potter’s hands.

3. The usual notation for the sphere in R3 with centre at O and radius 1 is S 2 . It consists
of points (x, y, z) in R3 with distance to the origin O equal to 1. Equivalently, they must
satisfy
x2 + y 2 + z 2 = 1.
4

We will show that S 2 is a regular surface. To this end we construct some local parametriza-
tions whose images cover S 2 .

Figure 4. The sphere is covered by six semi-spheres,

each of them being the image of a local parametrization
of the form (U, ϕ).

+
Consider the semi-sphere Sxy which consists of points (x, y, z) in S 2 with z > 0. Let π1
+
denote the orthogonal projection of Sxy to the horizontal coordinate plane. The image of π1
is the open disk Dxy bounded by the circle of radius 1 and centre 0; it consists of all points
(x, y) in the horizontal plane with
x2 + y 2 < 1.
One can easily see that π1 is bijective and its inverse, call it ϕ1 , is
p
ϕ1 (x, y) = (x, y, 1 − x2 − y 2).
+
Exercise. Show that ϕ1 : Dxy → Sxy is a local parametrization, that is, it satisfies conditions
1, 2, and 3 in Definition 3.1.1.
Similarly one can consider five other local parametrizations (see Figure 4). Finally, we
only need to note that each point P in S 2 belongs to one of the six semi-spheres. So S 2 is a
regular surface.
Remark 1. The following question concerning Definition 3.1.1 can be raised: is condition
2 not implied by conditions 1 and 3? Here are two examples which show that the answer is
“no, it’s not implied”. Probably the simplest one is ϕ : R2 → R3 ,
ϕ(u, v) = (u3 , v 3 , 0).
This is another “parametrization” of the horizontal coordinate plane (see Example 1 above).
It satisfies conditions 1 and 2 (easy to check this!), but not condition 3. Indeed, we have
ϕ′u (u, v) = (3u2 , 0, 0) and ϕ′v (u, v) = (0, 3v 2, 0),
 5

for any (u, v) ∈ R2 . This implies that at (0, 0) we have

ϕ′u (0, 0) = ϕ′v (0, 0) = (0, 0, 0)
so these two vectors cannot be linearly independent. Note that in this case, the inconvenience
can be fixed: we can so to say reparametrize ϕ(R2 ) and make a surface out of it (like we did
in Example 1).
A slightly more complicated (but more instructive) example3 is ϕ : R2 → R3 , ϕ(u, v) =
(u , v 3 , uv). The image of ϕ (that is, the potential surface parametrized by ϕ) is represented
3

in Figure 5. Again, it’s simple to check that ϕ satisfies conditions 1 and 3, and again we
have
ϕ′u (0, 0) = ϕ′v (0, 0) = (0, 0, 0),
that is, condition 2 is not satisfied. But this time we have no chance to reparametrize ϕ(R2 )
and make a regular surface out of it: the figure shows that we cannot remove the inconvenient
of not having a tangent plane at the point (0, 0, 0).

Figure 5. The image of ϕ(u, v) = (u3 , v 3 , uv), which is

not a regular surface.

Remark 2. In the same spirit as above, one can ask if the requirement that ϕ−1 : V ∩S → U
is continuous at point 3 does not follow from the other conditions. The following example is
showing that this is not the case, so we always have to check this, too. Take
ϕ(u, v) = (sin u, sin 2u, v)
where 0 < u < 2π and 0 < v < 1 (otherwise expressed, (u, v) is in (0, 2π) × (0, 1)). Its
image, call it S, can be seen in Figure 6. One can check that all conditions in Definition
3.1.1 are satisfied, except that ϕ−1 : S → (0, 2π) × (0, 1) is not continuous (at points of the
form (0, 0, v)). One needs to be a a bit advanced in point set topology to understand the last
thing. However, the lesson is that the “open” ribbon bent like in the figure is not a regular
surface.
3See [Gr-Abb-Sa], page 291.
6

Figure 6. The image of ϕ(u, v) = (sin u, sin 2u, v),

where 0 < u < 2π, 0 < v < 1, which is not a regular
surface.

Whoever took Vector Calculus knows that “graphs of functions of two variables are sur-
faces”. In fact those graphs are just a special type of regular level surfaces of functions of
three variables, which are discussed below.
Theorem 3.1.2. Let W be an open subset of R3 , g : W → R a differentiable function and
c a number. We denote by g −1(c) the set4 of all points P in W with g(P ) = c. Assume
that for any P in g −1 (c) at least one of the partial derivatives gx′ (P ), gy′ (P ), and gz′ (P ) is
different from5 0. Then g −1(c) is a regular surface.
Note. If the assumptions above are satisfied, we say that g −1 (c) is a regular level of g.
Proof. Without loss of generality we may assume that c = 0 (try to figure out why). We
take P in g −1 (0) and we check Definition 3.1.1. In other words, we construct U, ϕ and V .
By the hypothesis of the theorem, we know that gx′ (P ) 6= 0, or gy′ (P ) 6= 0 or gz′ (P ) 6= 0. Let’s
say, for instance, that gz′ (P ) 6= 0. We consider the function F : W → R3 given by
F (x, y, z) = (x, y, g(x, y, z))
for all (x, y, z) in W . The Jacobi matrix of F at P is
 
1 0 0
(JF )P =  0 1 0 .
gx (P ) gy (P ) gz (P )
′ ′ ′

Its determinant is gz′ (P ), which is different from 0. From the Inverse Function Theorem we
deduce that there exists an open subspace V of W , which contains P , and an open subspace
V1 of R3 , which contains F (P ), such that F restricted to V is a diffeomorphism from V to
V1 . By this we mean that F , regarded as a map from V to V1 , is differentiable, bijective,
4This is called the pre-image or the level of c.
5Equivalently, the gradient of g at P is e nonzero vector.
 7

and its inverse F −1 from V1 to V is differentiable. We note that F (V ∩ g −1 (0)) consists of

points in V1 which are of the form (x, y, 0); we identify the set of points of the latter form
with R2 and so we have
F (V ∩ g −1(0)) = V1 ∩ R2 = U
where U is open in R2 . The last equation implies
F −1 (U) = V ∩ g −1(0).
The pair consisting of U and F −1 restricted to U is a local parametrization of g −1(0) around
P (you may want to check carefully conditions 1,2, and 3 in Definition 3.1.1). 
Corollary 3.1.3. If U is an open subset of R2 and h : U → R is a differentiable function,
then the set S of R3 which consists of all points (x, y, h(x, y)) with (x, y) in U is a surface.
Note. The set S is called the graph of h.
Proof. Let U × R denote the set of all triples (x, y, z) with (x, y) in U and z in R. We
consider the function g : U × R → R given by
g(x, y, z) = z − f (x, y).
One can easily see that
S = g −1 (0).
We want to apply Theorem 3.1.3. To this end, we show that at any P in g −1 (0) at least one
of the partial derivatives of g is non-zero. But one can see that
gz′ (x, y, z) = 1 6= 0
at any point (x, y, z). By Theorem 3.1.3, the graph S is a surface. 
2 3
Examples. 1. The sphere S is, as we saw already, the set of all points (x, y, z) in R with
x2 + y 2 + z 2 = 1.
Take g : R3 → R, g(x, y, z) = x2 + y 2 + z 2 . Then S 2 is just g −1 (1).
Exercise. Show that for any P = (x, y, z) in S 2 , at least one of the numbers gx′ (P ), gy′ (P )
and gz′ (P ) is different from 0.
From Theorem 3.1.2 we deduce that S 2 is a regular surface (more precisely, we reprove
this fact, see example 3 above).
2. More generally, we can take the ellipsoid, which is the set of all points in R3 with
x2 y 2 z 2
+ 2 + 2 = 1.
a2 b c
The best way to visualize it is by noting that its intersections with each of the coordinated
2 2
planes is an ellipse (for instance, the intersection with the xy plane has equation xa2 + yb2 = 1).
We can prove that the ellipsoid is a regular surface in exactly the same way as above for the
sphere S 2 . See also Figure 7.
3. The hyperboloid with one sheet has equation
x2 y 2 z 2
+ 2 − 2 = 1.
a2 b c
We can also prove that it is a regular surface in the same way as the in the previous two
examples. See also Figure 8 (try to understand the intersections with the coordinate planes).
4. The torus is obtained by rotating a circle around a straight line which is contained in
the plane of the circle and has no intersection point with the circle. It is not difficult to
8

Figure 7. The ellipsoid (figure out where the coordinate

axes are).

Figure 8. The hyperboloid with one sheet (again, figure

out where the coordinate axes are).

figure out what the resulting surface looks like (a doughnut, see also Figure 9). Denote by
T the resulting set.

Figure 9. The torus.

Let’s show that it is a regular surface. Intuitively, it looks very likely that Definition 3.1.1,
is satisfied. However, it’s simpler to use Theorem 3.1.2.
Exercise. Choose the coordinate system such that the rotation line is the z axis and the
rotated circle is in the yz plane (see Figure 10).
 9

(a) Show that a point P = (x, y, z) is on T if and only if

p
(1) ( x2 + y 2 − a)2 + z 2 = r 2 .
Hint. Let (0, y0 , z0 ) be the point on the original (vertical) circle which gives rise to P
by rotation. Show that we have x2 + y 2 + (z − z0 )2 = y02 and z = z0 . Use the fact that
(0, y0 , z0 ) is on the original circle to deduce an equation involving y0 and z0 . Combine the
three equations you have obtained to deduce equation 1.
(b) Use Theorem 3.1.2 to deduce that T is a regular surface.

( 0 , y , z )

o o

( 0 , 0 , z )

( 0 , a , 0 )

( x , y , z )
P =

Figure 10. The torus which arises by rotating the “orig-

inal” circle in the yz plane about the z axis.

3.2. Differentiable maps and functions.6 Let S be a regular surface. Very often in this
course we will deal with functions f from S to R. A simple example is when we fix a point
P0 in R3 and we define the function f which assigns to any P on S the number f (P ) given
by the squared distance from P0 and P . We will actually need to deal with derivatives of
such functions, so we would like to trace a general framework where to refer whenever we
use these. This is the context of differentiable functions. Here is the definition.
Definition 3.2.1. We say that a function f : S → R is differentiable if for any P in S there
exists a local parametrization (U, ϕ) with P in ϕ(U) such that the function f ◦ ϕ : U → R
given by
(f ◦ ϕ)(u, v) = f (ϕ(u, v)), for all (u, v) ∈ U
is differentiable (see also Figure 3).
It is important to note that the function f ◦ ϕ in the previous definition is defined on U,
which is a subset of R2 ; in other words, f ◦ ϕ is a function of two variables, and we know
exactly what differentiability of such functions means (they must have partial derivatives
with respect to u and v).
If we look carefully at the definition, we see that one requests the existence of one local
coordinate system around each given point. On the other hand, around a given point there
exists in general more than one parametrization (think of the sphere, see Figure 4). It is
natural to raise the following question: is it not possible to have a parametrization (U, ϕ) with
f ◦ϕ differentiable and another parametrization (U ′ , ϕ′) with f ◦ϕ′ not differentiable? In other
words, does the definition of differentiability depend on the choice of the parametrization?
The answer is “no, it doesn’t depend”. To be able to prove this, we need the following
theorem (see also Figure 12).
6The following concerns terminology: a function takes values in R whereas the range of a map is not R.
10

P V

∩
S

Figure 11. The dotted arrows are: ϕ, f and the com-

position f ◦ ϕ.

Theorem 3.2.2. (Change of Parameters) On the regular surface S one considers two local
parametrizations (U, ϕ) and (Ũ , ϕ̃) with the property that the intersection
W := ϕ(U) ∩ ϕ̃(Ũ )
is non-empty. Then the map
ϕ−1 ◦ ϕ̃ : ϕ̃−1 (W ) → ϕ−1 (W )
is differentiable.
Proof’s Idea. We take Q in ϕ̃−1 (W ) and show that ϕ−1 ◦ ϕ̃ is differentiable on an open
neighborhood of Q in ϕ̃−1 (W ). The main point of the proof is that there exists a differentiable
function F : U × R → R3 such that:
• F restricted to U × {0} is ϕ
• F is a local diffeomorphism at (ϕ−1 (ϕ̃(Q)), 0)
We deduce that on an open neighborhood of Q in ϕ̃−1 (W ) we have ϕ−1 ◦ ϕ̃ = F −1 ◦ ϕ̃, which
is differentiable. The details can be found in [dC, page 70] (see also [Gr-Abb-Sa, page 300]).
Note. The map ϕ−1 ◦ ϕ̃ : ϕ̃−1 (W ) → ϕ−1 (W ) is called change of coordinates. It is a map
from an open subset in R2 to an open subset in R2 , and we know what its differentiability
means (both its components have partial derivatives). If we interchange ϕ and ϕ̃, we deduce
that the map
ϕ̃−1 ◦ ϕ : ϕ−1 (W ) → ϕ̃−1 (W )
is differentiable as well. But this is just the inverse of ϕ−1 ◦ ϕ̃, thus the latter map is
actually a diffeomorphism (which means that it is differentiable, bijective, and its inverse is
differentiable).
We are now ready to return to Definition 3.2.1 and the discussion following it. The
following is a corollary of Theorem 3.2.2.
Corollary 3.2.3. If f : S → R is differentiable, then for any local parametrization (U, ϕ)
of S the function f ◦ ϕ : U → R is differentiable.
Proof. We take Q an arbitrary point in U and show that there exists an open subset N in
R2 which contains Q and is contained in U such that the function f ◦ ϕ restricted to N is
 11

~ ~

U )

φ (

' 1
φ

' 1 o

φ φ

U )

φ (

Figure 12. The shaded regions are as follows: the one

on the surface is the intersection ϕ(U) ∩ ϕ̃(Ũ ), denoted
W ; the one in U is ϕ−1 (W ); the one in Ũ is ϕ̃−1 (W ).

differentiable. To this end, we consider the point

P := ϕ(Q)
in S and we use Definition 3.2.1: there exists a local parametrization (Ũ , ϕ̃) such that P is
in ϕ̃(Ũ) and the map
f ◦ ϕ̃ : Ũ → R
is differentiable. The intersection
W := ϕ(U) ∩ ϕ̃(Ũ )
is nonempty, since P is an element of it. We take N := ϕ−1 (W ). On N the function f ◦ ϕ
can be written as
f ◦ ϕ = f ◦ ϕ̃ ◦ ϕ̃−1 ◦ ϕ.
Since both ϕ̃−1 ◦ ϕ and f ◦ ϕ̃ are differentiable, their composition is differentiable as well.
Hence f ◦ ϕ restricted to N is differentiable, QED. 
Examples of differentiable functions on surfaces. 1. Let F : R3 → R be an arbitrary
differentiable function and let S be a regular surface. Then the function f : S → R which is
the restriction of F to S is differentiable (in the sense of Definition 3.2.1). Indeed, if (U, ϕ)
is any parametrization of S, then the function f ◦ ϕ : U → R is the same as F ◦ ϕ, which is
differentiable (as composition of two differentiable functions).
In the following we take two standard examples of differentiable functions from R3 to
R: the height and the distance function. Restricted to any surface, they give differentiable
functions.
2. Height functions. We consider a plane through the origin. The height of the point P
with respect to the plane is, up to a possible minus sign, the length of the line segment P Q,
which is perpendicular to the plane. We choose v a unit vector perpendicular to the plane
(see Figure 13).
12

Figure 13. The height of P with respect to the plane

in the figure is v · P .

Then the height of P with respect to the plane is

kP Qk = kOP k cos θ = P~Q · v.
~ , we can describe the height function hv : R3 →
If we identify the point P with the vector OP
R by the formula
hv (P ) = P · v,
for any P ∈ R3 . If we take only P in the surface S, we obtain the height function hv : S → R.
For instance, the height function with respect to e3 restricted to the sphere S 2 is differentiable
and has a minimum at the South pole S = (0, 0, −1) and a maximum at the North pole
N = (0, 0, 1).
3. Distance functions. We fix a point P0 and we consider the function
d(P ) = kP − P0 k2 ,
for all P ∈ R3 . This function is differentiable. Its restriction to any surface S is a differen-
tiable function on S.
Very often we will be interested in maps from a surface to another and their differentials.
For instance, the antipodal map a from S 2 to S 2 , given by a(x, y, z) = (−x, −y, −z) or the
map f from S 2 to the ellipse x2 /a2 + y 2 /b2 + z 2 /c2 = 1 given by
f (x, y, z) = (ax, by, cz).

Definition 3.2.4. Let S1 and S2 be two surfaces and f a map from S1 to S2 . We say that
f is differentiable if for any P in S1 there exists local parametrizations (U1 , ϕ1 ) on S1 and
(U2 , ϕ2 ) on S2 with P in ϕ1 (U1 ) and f (P ) in ϕ2 (U2 ) such that the map
ϕ−1 −1
2 ◦ f ◦ ϕ1 : ϕ1 (ϕ1 (U1 ) ∩ f
−1
(ϕ2 (U2 ))) → U2
 13

is differentiable.
The following is another corollary of Theorem 3.2.2.
Corollary 3.2.5. If f : S1 → S2 is a differentiable map, then for any point P in S1 and
any two local parametrizations (U1 , ϕ1 ), (U2 , ϕ2 ) on S1 , respectively S2 with P in ϕ1 (U1 ) and
f (P ) in ϕ2 (U2 ), the map
ϕ−1 −1
2 ◦ f ◦ ϕ1 : ϕ1 (ϕ1 (U1 ) ∩ f
−1
(ϕ2 (U2 ))) → U2
is differentiable.
Examples of differentiable maps between surfaces.
1. If f : S1 → S2 and g : S2 → S3 are differentiable, then the composition g ◦ f : S1 → S3
is differentiable.
2. If S is a surface and (U, ϕ) a local parametrization, then both ϕ : U → S and
ϕ−1 : ϕ(U) → U are differentiable (recall that U is a surface, see Section 3.1, Example 2).
3. If F : R3 → R3 is a differentiable map and S1 , S2 are two surfaces with F (S1 ) ⊂ S2 ,
then the restriction of F to S1 is a differentiable map between S1 and S2 (the proof is a
consequence of 1 and 2 above).
4. The antipodal map a : S 2 → S 2 , a(x, y, z) = (−x, −y, −z) is differentiable (from 3).
5. The map f from S 2 to the ellipse x2 /a2 + y 2 /b2 + z 2 /c2 = 1, given by f (x, y, z) =
(ax, by, cz), is differentiable (again from 3).
3.3. Tangent vectors, the tangent plane, and the differential of a map.
One possibility to define the tangent plane to a surface at a point P has already been
mentioned in Section 3.1: one could choose a local parametrization (U, ϕ) with P = ϕ(Q),
where Q is in U and then consider the plane determined by ϕ′u (Q) and ϕ′v (Q). The incon-
venient of this definition is that it depends on the choice of the parametrization around P
(we recall that in general there exists more than one parametrization around a point on a
surface). This is why we prefer to use a quite different idea. First we define the tangent
vectors at P , as follows.
Definition 3.3.1. (i) A tangent vector to the regular surface S at its point P is a vector of
the form α′ (0), where α is a curve α : (−ǫ, ǫ) → R3 such that
• α(t) is in S, for all t
• α(0) = P .
(ii) The tangent plane to S at P is the set of all tangent vectors to S at P . It is
denoted TP (S).
We will prove now that TP (S) is really a plane, that is, a two dimensional vector subspace
of R2 . Also that for any local parametrization (U, ϕ) around P the space TP S coincides with
the plane determined by ϕ′u (Q) and ϕ′v (Q) (where ϕ(Q) = P ).
Theorem 3.3.2. Let S be a regular surface, (U, ϕ) a local parametrization, Q ∈ U, and
P := ϕ(Q). Then the space of all tangent vectors at P to S coincides with the image of the
injective linear map
d(ϕ)Q : R2 → R3 .
This space is the same as the two dimensional vector space spanned by the vectors ϕ′u (Q)
and ϕ′v (Q).
Proof. We have to show that
TP S = d(ϕ)Q (R2 ).
14

TP S ⊂ d(ϕ)Q (R2 ): Let α be a curve like in Definition 3.3.1. The vector α′ (0) is in TP S. We
need to show that it is in the image of d(ϕ)Q . We consider the curve β in U given by
β(t) = ϕ−1 (α(t)),
where t in (−ǫ, ǫ), like in Figure 14. The curve β is differentiable (to justify this we use
the argument in the proof of Theorem 3.2.2: we can extend ϕ to a differentiable map
F : U × R → R3 which is a local diffeomorphism at Q, so we have ϕ−1 (α(t)) = F −1 (α(t)),
for all t sufficiently close to 0). We have α(t) = ϕ(β(t)) for all t, which implies that
d d
α′ (0) = |0 ϕ(β(t)) = d(ϕ)β(0) ( |0 β(t)) = d(ϕ)Q (β ′ (0))
dt dt
which is in the image of d(ϕ)Q .

5 ε ε

( 0

α '

U )

φ (

β )

( 0

'

Figure 14. The curves α and β are related by α(t) = ϕ(β(t)).

d(ϕ)Q (R2 ) ⊂ TP S: Let a be a vector in R2 , and take w := d(ϕ)Q (a). We need to show that
w is in TP S. To this end, we take the curve
α(t) = ϕ(Q + ta),
where t is in an interval of the form (−ǫ, ǫ), with ǫ sufficiently small. We have
α′ (0) = d(ϕ)Q (a) = w.
The proof is finished. 
The following lemma will be needed later:
Lemma 3.3.3. Let (U, ϕ) be a local parametrization on the surface S, and α′ (0) a vector in
TP S like in Definition 3.3.1, where P = ϕ(Q), with Q ∈ U. Write
ϕ−1 (α(t)) = (u(t), v(t)),
for all t in (−ǫ, ǫ). Then the coordinates of α′ (0) with respect to the basis ϕ′u (Q) and ϕ′v (Q)
are u′(0) and v ′ (0).
Proof. We have
α(t) = ϕ(u(t), v(t))
which implies
d ∂ϕ d ∂ϕ d
α′ (0) = |0 ϕ(u(t), v(t)) = (Q) |0 u(t) + (Q) |0 v(t) = u′ (0)ϕ′u (Q) + v ′ (0)ϕ′v (Q).
dt ∂u dt ∂v dt
Here we have used the chain rule for functions of two variables. 
 15

The next notion we will define is the differential of a map between two surfaces.
Definition 3.3.4. Let S1 and S2 are two surfaces, f : S1 → S2 a differentiable map (in the
sense of Definition 3.2.4), and P a point in S1 . The differential of f at P is the map
d(f )P : TP S1 → Tf (P ) S2
defined as follows: take w in TP S1 and α a curve like in Definition 3.3.1 with w = α′ (0);
then
d
(2) d(f )P (w) = |0 (f (α(t)).
dt
The problem with the definition is that there exists several curves on S1 through P with the
same tangent vector. More specifically, if w = α1′ (0) = α2′ (0), are the vectors d/dt|0 f (α1 (t))
and d/dt|0(α2 (t)) equal? The positive answer is given by the following theorem.
Theorem 3.3.5. (a) The vector d/dt|0f (α(t)) in Definition 3.3.4 is the same for any curve
α in S1 with α′ (0) = a. Thus the map d(f )P : TP S1 → Tf (P ) S2 is well defined.
(b) The map d(f )P : TP S1 → Tf (P ) S2 is linear.
Proof. Let us consider local parametrizations (U, ϕ) and (Ũ , ϕ̃) of S1 around P , respectively
of S2 around f (P ). We know that the map
ϕ̃−1 ◦ f ◦ ϕ : ϕ−1 (ϕ(U) ∩ f −1 (ϕ̃(Ũ)) → Ũ
is a differentiable map between two open subsets in R2 . Let us denote its components by h1
and h2 , that is
ϕ̃−1 ◦ f ◦ ϕ(u, v) = (h1 (u, v), h2(u, v)),
for all (u, v) in the domain.
Like in Lemma 3.3.3, we denote
ϕ−1 (α(t)) = (u(t), v(t))
and deduce that we have
w = α′ (0) = u′ (0)ϕ′u (Q) + v ′ (0)ϕ′v (Q),
where Q = ϕ−1 (P ). We consider the curve γ given by
γ(t) := f (α(t))
and we are interested in the tangent vector γ ′ (0). We are planning to use again Lemma
3.3.3. To this end we will need to find the components of ϕ̃−1 (γ(t)). We can write
ϕ̃−1 (γ(t)) = ϕ̃−1 ◦γ(t) = ϕ̃−1 ◦f ◦α(t) = ϕ̃−1 ◦f ◦ϕ◦ϕ−1 ◦α(t) = (h1 (u(t), v(t)), h2(u(t), v(t)).
Take Q̃ := ϕ̃−1 (f (P )). The coordinates of γ ′ (0) with respect to the basis (ϕ̃′u (Q̃), ϕ̃′v (Q̃)) are
d
|0 h1 (u(t), v(t)) = (h1 )′u (Q)u′ (0) + (h1 )′v (Q)v ′ (0)
dt
and
d
|0 h2 (u(t), v(t)) = (h2 )′u (Q)u′ (0) + (h2 )′v (Q)v ′ (0).
dt
In other words, we have
γ ′ (0) = [(h1 )′u (Q)u′ (0) + (h1 )′v (Q)v ′ (0)]ϕ̃′u (Q̃) + [(h2 )′u (Q)u′ (0) + (h2 )′v (Q)v ′ (0)]ϕ̃′u (Q̃).
The last equation is decisive for the proof, as it shows as follows:
16

• The vector γ ′ (0) depends only on w, that is, on its coordinates with respect to the
basis (ϕ′u (Q), ϕ′v (Q)) (so it does not depend on the curve α). This shows that the
map d(f )Q is well defined.
• The map d(f )P : TP S1 → Tf (P ) S2 is linear. Its matrix with respect to the bases
(ϕ′u (Q), ϕ′v (Q)) and (ϕ̃′u (Q̃), ϕ̃′v (Q̃)) is
 
(h1 )′u (Q) (h1 )′v (Q)
(h2 )′u (Q) (h2 )′v (Q)
which is just the Jacobi matrix J(ϕ̃−1 ◦ f ◦ ϕ)Q .

We can do similar things for differentiable functions f : S → R, as follows.
Definition 3.3.6. Let S be a regular surface, f : S → R be a differentiable function (in the
sense of Definition 3.2.1) and P a point in S1 . The differential of f at P is the function
d(f )P : TP S → R
defined as follows: take w in TP S and α a curve like in Definition 3.3.3 with w = α′ (0); then
d
(3) d(f )P (w) = |0 (f (α(t)).
dt
Again we can show that d(f )P (w) is independent of the choice of the curve α with α′ (0) =
w, so the function d(f )P is well defined. Moreover, the function is linear.
Remarks. (a) One can easily show that if F : R3 → R is differentiable, S is a surface, and
f : S → R is the restriction of F to S, then the function d(f )P : TP S → R is the restriction
of d(F )P : R3 → R to TP S.
(b) Similarly, take S1 , S2 two surfaces, G : R3 → R3 a differentiable map with G(S1 ) ⊂ S2
and g : S1 → S2 the restriction of G to S1 . If P is a point in S1 , then d(G)P (TP S1 ) ⊂ Tg(P ) S2
and the differential map d(g)P : TP S1 → Tg(P ) S2 is the same as the restriction of d(G)P :
R3 → R3 to TP (S1 ).

References
[dC] M. P. do Carmo, Differential Geometry of Curves and Surfaces, Prentice Hall 1976
[Gr-Abb-Sa] A. Gray, E. Abbena, and S. Salamon, Modern Differential Geometry with MATHEMATICA,
3rd edition