Classical differential geometry of

two-dimensional surfaces
1 Basic definitions
This section gives an overview of the basic notions of differential geometry for two-
dimensional surfaces. It follows mainly Kreyszig [Kre91] in its discussion.

Definition of a surface
Let us consider the vector function X(ξ 1 , ξ 2 ) ∈ R3 with

X : R2 ⊃ Ξ 3 (ξ 1 , ξ 2 ) 7→ X(ξ 1 , ξ 2 ) ∈ U ⊂ R3 , (1)

where Ξ is an open subset of R2 . Let X(ξ 1 , ξ 2 ) be of class r ≥ 1 in Ξ, which means

that one of its component functions Xi (i ∈ {x, y, z}) is of class r and the other
x ,Xy ,Xz )
ones are at least of this class.1 Let furthermore the Jacobian matrix ∂(X∂(ξ 1 ,ξ 2 ) be
of rank 2 in Ξ which implies that the vectors
∂X
ea := = ∂a X , a ∈ {1, 2} , (2)
∂ξ a

are linearly independent. The mapping (1) then defines a smooth two-dimensional
surface patch U embedded in three-dimensional Euclidean space R3 with coordi-
nates ξ 1 and ξ 2 (see Fig. 1). A union Σ of surface patches is called a surface
if two arbitrary patches U and U 0 of Σ can be joined by finitely many patches
U = U1 , U2 , . . . , Un−1 , Un = U 0 in such a way that the intersection of two subsequent
patches is again a surface patch [Kre91, p. 76]. To simplify the following let us
restrict ourselves to a surface that can be covered by one patch U only.
The vectors ea , defined in Eqn. (2), are the tangent vectors of the surface. They are
not normalized in general. Together with the unit normal
e1 × e2
n := , (3)
|e1 × e2 |

they form a local basis (local frame) in R3 (see Fig. 2):

ea · n = 0 , and n · n = 1. (4)

1
A function of one or several variables is called a function of class r if it possesses continous
partial derivatives up to order r.

1
 z

y
X ξ2

x
ξ1

Figure 1: Parametrization of a surface

n
e2
γ
e1

z X

y
x

Figure 2: Local frame on the surface

2
The metric tensor (first fundamental form)
With the tangent vectors ea , one can define the metric tensor (also called the first
fundamental form)
gab := ea · eb . (5)
This covariant second rank tensor is symmetric (gab = gba ) and positive definite
[Kre91, p. 86]. It helps to determine the infinitesimal Euclidean distance in terms
of the coordinate differentials [Kre91, p. 82]
ds2 = [X(ξ 1 + dξ 1 , ξ 2 + dξ 2 ) − X(ξ 1 , ξ 2 )]2 = (e1 dξ 1 + e2 dξ 2 )2
= (ea dξ a )2 = (ea · eb ) dξ a dξ b
= gab dξ a dξ b , (6)
where the sum convention is used in the last two lines (see App. ??). The con-
travariant dual tensor of the metric may be defined via
(
1, if a = b
gac g cb := δab := , (7)
0, if a 6= b
where δab is the Kronecker symbol. The metric and its inverse can be used to raise
and lower indices in tensor equations. Consider for instance the second rank tensor
tab :
Raising: tac g cb = tab , and lowering: tac gcb = tab . (8)
The determinant of the metric2
g := det g = |gab | = g11 g22 − g12 g21 (9)
can be exploited to calculate the infinitesimal area element dA: let γ be the angle
between e1 and e2 (see Fig. 2). Then
|e1 × e2 |2 = |e1 |2 |e2 |2 sin2 γ = g11 g22 (1 − cos2 γ) = g11 g22 − (e1 · e2 )2
= g11 g22 − g12 g12 = g , (10)
and thus
√
dA = |e1 × e2 | dξ 1 dξ 2 = g d2 ξ . (11)

The covariant derivative

The partial derivative ∂a is itself not a tensor. One therefore defines the covariant
derivative ∇a on a tensor tab11ba22...b
...an
m

∇c tab11ba22...b
...an
m
= ∂c tab11ba22...b
...an
m

+ tdb1ab22...a a1 a1 d...an a2 a1 a2 ...d

...bm Γdc + tb1 b2 ...bm Γdc + . . . + tb1 b2 ...bm Γdc
n an

− tad1ba22...b
...an
m
Γb1dc − tab11d...b
a2 ...an
m
Γb2dc − . . . − tab11ba22...d
...an
Γbmdc , (12)
2
Note that g is the matrix consisting of the metric tensor components g ab .

3
where the Γabc are the Christoffel symbols of the second kind with
Γabc = (∂a eb ) · ec , (13)
and ∇a is now a tensor. For the covariant differentiation of sums and products of
tensors the usual rules of differential calculus hold. The metric-compatible Laplacian
∆ can be defined as ∆ := ∇a ∇a .
Note in particular that
∇a eb = ∂a eb − Γabc ec , and (14)
bc
∇a gbc = ∇a g = ∇a g = 0 . (15)
Equation (15) is also called the Lemma of Ricci. It implies that raising and lowering
of indices commutes with the process of covariant differentiation.

Orientable surfaces
The orientation of the normal vector n in one point S of the surface depends on
the choice of the coordinate system [Kre91, p. 108]: exchanging, for instance, ξ 1
and ξ 2 also flips n by 180 degrees. A surface is called orientable if no closed curve
C through any point S of the surface exists which causes the sense of n to change
when displacing n continuously from S along C back to S. An example of a surface
that is not orientable is the Möbius strip.

The extrinsic curvature tensor (second fundamental form)

Two surfaces may have the same metric tensor gab but different curvature properties
in R3 . In order to describe such properties let us consider a surface Σ of class3 r ≥ 2
and a curve C of the same class on Σ with the parametrization X(ξ 1 (s), ξ 2 (s)) on
Σ, where s is the arc length of the curve (see Fig. 3).
At every point of the curve where its curvature k > 0, one may define a moving
trihedron {t, p, b} where t = Ẋ is the unit tangent vector, p = ṫ/|ṫ| = ṫ/k is the
unit principal normal vector, and b = t × p is the unit binormal vector of the curve. 4
Furthermore, let η be the angle between the unit normal vector n of the surface and
the unit principal normal vector p of the curve with cos η = p · n (see again Fig. 3).
The curvature k of the curve can then be decomposed into a part which is due to
the fact that the surface is curved in R3 and a part due to the fact that the curve
itself is curved. The former will be called the normal curvature Kn , the latter the
geodesic curvature Kg . One defines:
Kn := −ṫ · n = −k (p · n) = −k cos η , and (16)
Kg := t · (ṫ × n) = k t · (p × n) = k sin η sign (n · b) . (17)
3
This means that its parametrization X(ξ 1 , ξ 2 ) is of class r ≥ 2.
4
The dot denotes the derivative with respect to the arc length s.

4
 Σ n
S
η t
C p
b

Figure 3: Curve on a surface

Here, we are interested in the curvature properties of the surface. Therefore, the
normal curvature Kn is the relevant quantity that has to be studied a bit further.5
The vector ṫ may be written as
d
ṫ = Ẍ = (ea ξ˙a ) = (∂b ea ) ξ˙a ξ˙b + ea ξ¨a . (18)
ds
Thus, Eqn. (16) turns into

Kn = −k cos η = (−n · ∂a eb ) ξ˙a ξ˙b , (19)

where it has been exploited that ∂a eb = ∂b ea . The expression in brackets is the

extrinsic curvature tensor or second fundamental form

Kab := −n · ∂a eb = ea · ∂b n . (20)

It is a symmetric covariant second rank tensor such as the metric. The second
relation in Eqn. (20) follows if one differentiates the first equation of (4) with respect
to ξ a .
The extrinsic curvature can be written covariantly:

Kab := −n · ∇a eb . (21)

This is possible because ∂a eb differs from ∇a eb only by terms proportional to the

tangent vectors ec , which vanish when multiplied by n (see Eqn. (14)).
5
The minus sign in the definition of Kn , Eqn. (16), is unfortunately a matter of convention and is
here chosen in accordance to the literature where the surface stress tensor for fluid membranes
has been introduced [CG02, Guv04]. A sphere with outward pointing unit normal has a positive
normal curvature then. Note that this differs from Ref. [Kre91].

5
One can easily see from Eqn. (20) that Kab has got something to do with curvature:
at every point of the surface it measures the change of the normal vector in R3 for
an infinitesimal displacement in the direction of a coordinate curve.
To learn more about the normal curvature let us consider a reparametrization of the
curve C with the new parameter t. One gets
dξ a dt ξa0
ξ˙a = = 0 , (22)
dt ds s
where 0 denotes the derivative with respect to t. Equation (19) thus takes the form
0 0
˙a ˙b
Kab ξ a 0 ξ b (6) Kab ξ a 0 ξ b Kab dξ a dξ b
Kn = Kab ξ ξ = = = . (23)
(s0 )2 gab ξ a 0 ξ b 0 gab dξ a dξ b
For a fixed point S, Kab and gab are fixed as well. The value of Kn then only depends
on the direction of the tangent vector t of the curve. One may search for extremal
values of Kn at S by rewriting Eqn. (23):
(Kab − Kn gab ) ξ˙a ξ˙b = 0 . (24)
A differentiation with respect to ξ˙c yields the result
(Kac − Kn gac ) ξ˙a = 0 , (25)
because dKn = 0 is necessary for Kn to be extremal. Through the raising of one
index, Eqn. (25) becomes an eigenvalue problem for Kab . Its eigenvectors are the
tangent directions along which the normal curvature is extremal. They are called
principal directions and are orthogonal to each other [Kre91, p. 129]. The eigenvalues
will be called the principal curvatures k1 and k2 of the surface in point S. All other
values of Kn in S in any direction can be calculated via Euler’s theorem [Kre91,
p. 132]. If the curve follows a principal direction at every point, it is also called a
line of curvature.
For an arbitrary curve on the surface the symbol Kk denotes the normal curvature
belonging to the direction the curve is following, whereas K⊥ denotes the normal
curvature belonging to the direction perpendicular to the curve in every point.
It is useful to define the following two notions: the total curvature
K := g ab Kab = Kaa = k1 + k2 , (26)
and the Gaussian curvature
KG := |Kab | = k1 k2 . (27)
The quantities |K| and KG are invariant under surface reparametrizations because
they only involve the eigenvalues of the extrinsic curvature tensor. They occur, for
instance, in the surface Hamiltonian of a fluid membrane. Note that one can rewrite
KG
K11 K22 − K12 K21
KG = |Kab | = |Kac g cb | = |Kac | |g cb | = . (28)
g

6
The equations of Gauss and Weingarten
With the help of the extrinsic curvature it is also possible to find relations for the
partial derivatives of the local frame vectors: the normal vector n is a unit vector
(see Eqn. (4)) and therefore
n · ∂a n = 0 . (29)
Thus, ∂a n is a linear combination of the tangent vectors ea . We know that ∂a n·ea =
Kab (see Eqn. (20)), which yields the Weingarten equations

∂a n = ∇a n = Kab eb . (30)

For the tangent vector ea a decomposition yields

(20),(13)
∂a eb = (n · ∂a eb )n + (ec · ∂a eb )ec = −Kab n + Γabc ec . (31)

These are the Gauss equations, which can be rewritten covariantly:

(14)
∇a eb = −Kab n . (32)

Intrinsic curvature and integrability conditions

Do the partial differential Eqns. (30) and (32) have solutions for any chosen gab and
Kab ? The answer is no; certain integrability conditions have to be satisfied. We
require the embedding functions X to be of class r ≥ 3 and

∂a ∂b e c = ∂ b ∂a e c . (33)

From this follows [Kre91, p. 142 et seq.]

Rabcd = Kbd Kca − Kbc Kda , and (34)

∇a Kbc = ∇b Kac , (35)

where
Rabcd := ∂c Γbda − ∂d Γbca + Γbde Γeca − Γbce Γeda , (36)
is called the mixed Riemann curvature tensor. It is intrinsic because it does not
depend on the normal vector n. Expression (35) is also referred to as the equation
of Mainardi-Codazzi.
The Ricci tensor is defined as the contraction of the Riemann tensor with respect
to its first and third index:
Rab := Rc acb . (37)
A further contraction of the Ricci tensor yields the intrinsic scalar curvature of the
surface (Ricci scalar)
R := g ab Rab . (38)

7
From Eqn. (34) one then obtains

Rab = KKab − Kac Kbc , and (39)

R = K 2 − K ab Kab . (40)

Combining Eqn. (28) with the completely covariant form of Eqn. (34), one gets after
a few calculations:

Rab = KG gab , and (41)

R = 2 KG . (42)

These equations confirm Gauss’ Theorema Egregium, which states that the Gaussian
curvature, even though originally defined in an extrinsic way, in fact only depends on
the first fundamental form [Kre91, p. 145] and is thus an intrinsic surface property.

2 Gauss-Bonnet theorem
The Gauss-Bonnet theorem for simply connected surfaces
The Gauss-Bonnet theorem states the following [Kre91, p. 169]: Let Σ0 be a simply
connected surface patch of class rΣ0 ≥ 3 with simple closed boundary ∂Σ0 of class
r∂Σ0 ≥ 3. Furthermore, let X(ξ 1 (s), ξ 2 (s)) be the parametrization of the boundary
curve, where s is the arc length. Then
Z Z
ds Kg + dA KG = 2π , (43)
∂Σ0 Σ0

where dA is the infinitesimal area element, Kg is the geodesic curvature of ∂Σ0 , and
KG is the Gaussian curvature of Σ0 . Note that the integration along the boundary
curve has to be carried out in such a sense that the right-hand rule is satisfied: take
your right thumb and point it in the direction of the normal vector n. If you then
curl your fingers, the tips indicate the direction of integration.
One can check the consistency of Eqn. (43) easily by considering a flat circle with
radius a: Its Gaussian curvature is zero and therefore also the integral over it.
The geodesic curvature, however, is equal to 1/a in every point of the boundary.
Thus, the integral over Kg yields 2πa × 1/a, which is equal to the right-hand side
of Eqn. (43).

Generalization to multiply connected surfaces

A generalization of this theorem to multiply connected surfaces is also possible
[Kre91, p. 172]: One can cut multiply connected surfaces into simply connected
ones. Take, for instance, a surface as in Fig. 4. The path of integration along the

8
 

























































































































C


1





















































































































































































































































































































































































































C





















































































































2 


































































































































































































































































































































































































































































C





3 


























































































































































































































































































































































































































































































































































































































































Figure 4: Integration contour for multiply connected surface patches

boundary may be chosen as depicted by the arrows. The sections are passed twice in
opposite directions; their contributions therefore cancel each other.
R The end points
of every section, however, add a term of π each to the integral ds Kg . This is due
to the rotation the tangent makes at each of these points. Every section therefore
contributes 2π to the integral. For the case of Fig. 4 we thus have an extra term of
4π.

Application to closed surfaces

It is also possible to apply the Gauss-Bonnet theorem to closed surfaces [Kre91,
p. 172]. Topologically, any closed orientable surface is homeomorphic6 to a sphere
with p attached “handles”. This number p is also called genus of the surface.
Consequently, a sphere has genus 0, a torus genus 1, etc. One then obtains for
any closed orientable surface Σ of genus p [Kre91, p. 172]:
Z
dA KG = 4π(1 − p) . (44)
Σ

This implies that the integral over the Gaussian curvature is a topological invariant
for any closed surface with fixed genus p.

3 Monge parametrization
For surfaces with no “overhangs”, it is sufficient to describe their position in terms of
a height h(x, y) above the underlying reference plane as a function of the orthonormal
coordinates x and y. The direction of the basis vectors {x, y, z} ∈ R3 is chosen as
depicted in Fig. 5.

6
This means that the mapping and its inverse are continuous and bijective.

9
 z

h(x,y)

Figure 5: Monge parametrization

The tangent vectors on the surface can then be expressed as ex = (1, 0, hx )T and
ey = (0, 1, hy )T , where hi = ∂i h (i, j ∈ {x, y}). The metric is equal to
gij = δij + hi hj , (45)
where δij is the Kronecker symbol. We also define ∇ = (∂x , ∂y )T . The metric
determinant and the infinitesimal surface element can then be written as
g = |gij | = 1 + (∇h)2 and (46)
√
dA = g dx dy . (47)
The inverse metric is given by
hi hj
g ij = δij − . (48)
g
Note that Eqns. (45) and (48) are not tensor equations. The right-hand side gives
merely numerical values for the components of the covariant tensors gij and g ij . The
unit normal vector is equal to
µ ¶
1 −∇h
n= √ . (49)
g 1

With the help of Eqn. (20) the extrinsic curvature tensor can be calculated:
hij
Kij = − √ , (50)
g
where hij = ∂i ∂j h. Note that Eqn. (50) again is not a tensor equation and gives
only numerical values for the components of Kij .

10
Finally, it is also possible to write the total curvature K in Monge parametrization:
³ ∇h ´
K = −∇ · √ . (51)
g

One is often interested in surfaces that deviate only weakly from a flat plane. In
this situation the gradients hi are small. Therefore, it is enough to consider only
the lowest nontrivial order of a small gradient expansion. K and dA can then be
written as

K = −∇2 h + O[(∇h)2 ] , (52)

n 1 o
dA = 1 + (∇h)2 + O[(∇h)4 ] dx dy . (53)
2

References
[CG02] R. Capovilla and J. Guven. Stresses in lipid membranes. J. Phys. A: Math.
Gen., 35: pp. 6233–6247, 2002.

[Guv04] J. Guven. Membrane geometry with auxiliary variables and quadratic con-
straints. J. Phys. A: Math. Gen., 37: L313–L319, 2004.

[Kre91] E. Kreyszig. Differential geometry. Dover, Mineola, NY, 1991.

11