Lecture Note On Elementary Differential Geometry: Ling-Wei Luo

Lecture Note on Elementary Differential Geometry
Ling-Wei Luo*
Institute of Physics, Academia Sinica
July 20, 2019
Abstract
This is a note based on a course of elementary differential geometry as I gave the lectures in
the NCTU-Yau Journal Club: Interplay of Physics and Geometry at Department of Electrophysics
in National Chiao Tung University (NCTU) in Spring semester 2017. The contents of remarks,
supplements and examples are highlighted in the red, green and blue frame boxes respectively.
The supplements can be omitted at first reading. The basic knowledge of the differential forms
can be found in the lecture notes given by Dr. Sheng-Hong Lai (NCTU) and Prof. Jen-Chi Lee
(NCTU) on the website. The website address of Interplay of Physics and Geometry is http:
//web.it.nctu.edu.tw/~string/journalclub.htm or http://web.it.nctu.
edu.tw/~string/ipg/.
Contents
1 Curve on E2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Curve in E3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3 Surface theory in E3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4 Cartan’s moving frame and exterior differentiation methods . . . . . . . . . . . . . . 31
1 Curve on E2
We define n-dimensional Euclidean space En as
a n-dimensional real space Rn equipped a dot
product defined n-dimensional vector space.
Tangent vector In 2-dimensional Euclidean

( E plane,
2
space, an ) we parametrize a curve
p(t) = x(t), y(t) by one parameter t with re-
spect to a reference point o with a fixed Cartesian
coordinate frame. The( velocity ) vector at point p
is given by ṗ(t) = ẋ(t), ẏ(t) with the norm Figure 1: A curve.
√ √
|ṗ(t)| = ṗ · ṗ = ẋ2 + ẏ 2 , (1)
*
Electronic address: lwluo@gate.sinica.edu.tw
1
where ẋ := dx/dt. The arc length s in the interval [a, b] can be calculated by
∫ ∫ √ ∫ b√ ∫ b
s = ds = 2
(dx) + (dy) = 2 2 2
ẋ + ẏ dt = |ṗ(t)| dt . (2)
a a
The arc length can be a function of parameter t given by

∫ t
s(t) = |ṗ(t′ )| dt′ . (3)
a
From the fundamental theorem of calculus, we have

ds
̸= 0 =⇒ ṡ(t) = |ṗ(t)| > 0 . (4)
dt
According to the inverse

( function
) theorem, we have t = t(s). One can parametrize the( curve by arc)
length s as p(s) = x(s), y(s) . The corresponding velocity vector should be p (s) = x′ (s), y ′ (s) ,
′
where we have x′ := dx/ds. We can rewrite the derivatives of x and y with respect to s as


 ′ dx dx dt dt
x = = = ẋ ,
ds dt ds ds
(5)

 dt
 y = ẏ .
′
ds
Thus, the norm of the velocity vector parametrized by s can be calculated as
√ √ dt dt ds dt
|p′ (s)| = x′2 + y ′2 = ẋ2 + ẏ 2 = |ṗ| = = 1, (6)
ds ds dt ds
which implies that the velocity vector p′ (s) is a unit vector. We can define a unit tangent vector as a
velocity vector parametrized by s
T ≡ e1 := p′ (s) . (7)
Normal vector Due to e1 · e1 = p′ · p′ = 1, we have
e′1 · e1 + e1 · e′1 = 0 =⇒ e′1 · e1 = 0 =⇒ e′1 ⊥ e1 , (8)
it indicates that e′1 is a normal vector. The principle normal vector is defined by
e′1
N ≡ e2 := (9)
|e′1 |
as a unit normal vector at p(s). The curvature of a curve p(s) is given by κ(s) = |e′1 (s)| > 0, which
can be realized as a norm of the acceleration vector a := e′1 = p′′ . Therefore, we have a relation
e′1 = κ(s)e2 . (10)
Remark. If a vector V is an unit vector, |V | = 1, the corresponding derivative vector would be

perpendicular to itself, i.e.
V′ ⊥V . (11)
2
Osculating plane The plane is spanned by the vectors e1 and e2 is called osculating plane.
Newton’s second law In classical physics, we have a momentum vector p = mT = mp′ with mass
m. The force F is defined by Newton’s second law
dp dT
F = =m = ma = mp′′ (12)
ds ds
with respect to parameter s.
Frame A set of vector e1 , e2 equipped with a point p calls frame. In such of case, a frame at p is
denoted by (p; e1 , e2 ).
Frenet-Serret formula in 2D From the orthonormality condition ei · ej = δij (i, j = 1, 2), we have
e′i · ej + ei · e′j = 0 (13a)

=⇒ e′1 · e2 + e1 · e′2 = κ + e1 · e′2 = 0 (13b)
=⇒ e1 · e′2 = −κ (e′2 has component − κ along e1 direction) (13c)
=⇒ e2′ = −κe1 . (13d)
As a result, we have the following relations

 ′  ′  

 p = +e1 p 1 0 ( )
 ′   e1
e′1 = +κe2 =⇒ e1  =  0 κ (14)

e′ = −κe e2
2 1 e′2 −κ 0
called Frenet-Serret formula.
Example (Circle in E2 ). A circle with radius r can be parametrized by p(t) = (r cos t, r sin t)
with 0 ≤ t ≤ 2π.
Figure 2: A circle.
The tangent vector is
ṗ(t) = (−r sin t, r cos t) (15)
with norm
√
|ṗ| = r2 sin2 t, r2 cos2 t) := r . (16)
3
The arc length s(t) is
∫ t ∫ t
′ ′
s(t) = |p(t )|dt = r dt′ = rt . (17)
0 0
Therefore, the circumference is

∫ 2π ∫ 2π
′ ′
L= |p(t )|dt = r dt′ = 2πr . (18)
0 0
By t = s/r, the circle p(s) and its tangent vector are

( )
s s
p(s) = r cos , r sin (19a)
r r
and
( )
′ s s
p (s) = − sin , cos = e1 = T (19b)
r r
respectively. From (19b), we have

( )
1 s 1 s
e′1 (s) = − cos , − sin . (20)
r r r r
The curvature κ can be obtained by

√
1 s 1 s 1
κ = |e′1 | = 2
cos2 + 2 sin2 = , (21)
r r r r r
which is the inverse of the constant radius r. The normal vector can be calculated by
( ) ( )
e′1 1 s 1 s s s
e2 = ′ = r − cos − sin = − cos , sin . (22)
|e1 | r r r r r r
Gauss map Gauss map G is a mapping which

globally send all the points p of curve to a unit
circle S 1 (a Gauss circle) centered at c and send
the corresponding normal vector e2 to a radius
vector from c pointing to S 1 , which is shown
as Fig. 3. Therefore, e2 can be represented as
a point on S 1 .
Let’s consider two normal vectors e2 (s) and
e2 (s′ ) with respect to two infinitesimal points
p(s) and p(s′ ), where s′ = s + ∆s is infinitesi- Figure 3: The Gauss map G.
mal close to s. We can expand e2 (s′ ) at s:
e2 (s′ ) = e2 (s + ∆s)
≈ e2 (s) + e′2 (s)∆s
= e2 (s) + (−κ(s)e1 (s))∆s
= e2 (s) + (−κ(s)∆s)e1 (s) , (23)
4
which is the parametrization of a point under the Gauss map. Thus, we know the distance between
two infinitesimal point e2 (s) and e2 (s′ ) on Gauss circle given by
|e2 (s′ ) − e2 (s)| = |∆e2 | = κ(s)∆s . (24)
And we also have
( )
∆p ≡ p(s′ ) − p(s) ≈ p(s) + p′ (s)∆s − p(s) = p′ (s)∆s =⇒ |p(s′ ) − p(s)| = |∆p| = ∆s .
(25)
Therefore, in the local region, the ratio of the length between two points on the Gauss circle and curve,
i.e., |∆e2 |/|∆p| can be calculate by
|e2 (s′ ) − e2 (s)| |∆e2 | κ(s)∆s
= = = κ(s) , (26)
|p(s ) − p(s)|
′ |∆p| ∆s
which measure the curvature of a curve, κ(s), at the neighborhood of a local point p.
According to the example of circle, we assume a vector q = p+re2 = p+(1/κ)e2 . The derivative
of q is
1 1
q′ = p′ + e′2 = e1 + (−κe1 ) = 0 , (27)
κ κ
which means that q is fixed, i.e., q is the center of the osculating circle with radius 1/κ. By considering
the Gauss map of a circle. The radius vector should be e2 and the center c of Gauss circle corresponds
to the point q of the osculating circle which is the circle itself. Thus, the Gauss circle can be imaged
by rescaling the radius of osculating circle to unity.
( )
Example (Curvature of ellipse). An ellipse is described by p(t) = x(t), y(t) with the
parametrization of the coordinates x(t) = a cos t and y(t) = b sin t (a > b > 0), i.e.,
x2 y 2
2
+ 2 = cos2 t + sin2 t = 1 . (28)
a b
Figure 4: An ellipse.
The tangent vector is
( )
ṗ(t) = − a sin t, b cos t (29)
By changing the parameter to s, we have to calculate ds/dt first:

ds √ dt 1 1
= |ṗ| = a2 sin2 t + b2 cos2 t =⇒ =√ = . (30)
dt ds a2 sin2 t + b2 cos2 t ṡ
Therefore, the tangent vector parametrized by s is obtained by

dp dt
e1 = p′ =
( dt ds ) ( )
−a sin t b cos t −a sin t b cos t
= √ ,√ = , , (31)
a2 sin2 t + b2 cos2 t a2 sin2 t + b2 cos2 t ṡ ṡ
5
Subsequently, we have
( )
−b cos t −a sin t
e2 = , . (32)
ṡ ṡ
However
de1 de1 dt
e′1 = = = κe2 . (33)
ds dt ds
As a result, the curvature is
ab
κ(t) = 2 . (34)
(a2 sin t + b2 cos2 t)3/2
If we consider the particular case of a = b, an ellipse reduce to a circle with curvature κ = 1/a.
2 Curve in E3
( )
In E3 , a curve is parametrized as p(t) = x(t), y(t), z(t) and we have to look for an orthonormal
frame at p denoted by (p; e1 , e2 , e3 ) The vector e1 = p′ is uniquely defined by the same way. Due
to e′1 ⊥ e1 , vector e′1 should be proportional to e2 or e3 . Now we can fix e′1 = κe2 as the previous
section.
Binormal vector Now we define a unit vector orthogonal to T and N called binormal vector
B := T ∧ N
≡ e1 ∧ e2 := e3 , (35)
where ∧ is the exterior product or wedge product.
Remark. In 3-dimensional space, the exterior product ∧ is the same to the usual cross product ×
of two vectors.
By orthonormality condition ei · ej = δij (i, j = 1, 2, 3), we have
e′i · ej + ei · e′j = 0 , (36)
which implies:
(i) If i = j, we have e′i ⊥ ei , e′2 should be the combination of e1 and e3 .
(ii) If i ̸= j, we have
{
0 = e′1 · e2 + e1 · e′2 = (κe2 ) · e2 + e1 · e′2 = κ + e1 · e′2 (i = 1, j = 2) , (37a)
0 = e′1 · e3 + e1 · e′3 = (κe2 ) · e3 + e1 · e′3 = 0 + e1 · e′3 (i = 1, j = 3) . (37b)
Therefore, with the result (37a), we have to assume that
e′2 = −κ(s)e1 + τ (s)e3 . (38)
By comparing to (13d), it contains an additional term related to e3 . For i = 2, j = 3, we obtain
0 = e′2 · e3 + e2 · e′3 = (−κe1 + τ e3 ) · e3 + e2 · e′3 = τ + e2 · e′3 . (39)
6
Due to (i) and (37b), e′3 should be perpendicular to e1 and e3 . As a result, we obtain the unique solution
that
e′3 = −τ e2 , (40)
where τ (s) is called torsion of a curve p(s). The geometric meaning of torsion is that it make the
point of the curve leave for the osculating plane spanned by e1 and e2 .
Remark. Apparently, the torsion of a curve is always related to the binormal vector B ≡ e3 .
Frenet-Serret formula in 3D As a result, we have Frenet-Serret formula:

 ′  ′  
 p = +e1 p 1 0 0  


e′ =  ′   e1
1 +κe2 e1   0 κ 0  
′ =⇒  ′ =   e2  . (41)

 e2 = −κe1 +τ e3 e2   −κ 0 τ 

 ′ e3
e3 = −τ e2 e′3 0 −τ 0
Remark. If one defines B := N ∧ T , then one should assume e′2 = −κ(s)e1 − τ (s)e3 and obtain
e′3 = +τ e2 .
Parametrization of a curve in a neighborhood of s0 One can do the Taylor expansion of p(s) at s0 .
• First order:

dp
p(s) ≈ p(s0 ) + (s − s0 ) = p(s0 ) + e1 (s0 )(s − s0 ) . (42)
ds s=s0
• Second order:
1 ′′
p(s) ≈ p(s0 ) + p′ (s0 )(s − s0 ) + p (s0 )(s − s0 )2
2!
1
= p(s0 ) + e1 (s0 )(s − s0 ) + κ(s0 )e2 (s0 )(s − s0 )2 . (43)
2
• Third order:
1 ′′ 1
p(s) ≈ p(s0 ) + p′ (s0 )(s − s0 ) + p (s0 )(s − s0 )2 + p′′′ (s0 )(s − s0 )3
2! 3! | {z }
p′′′ =(p′′ )′ =κ′ e2 +κe′2 =κ′ e2 +κ(−κe1 +τ e3 )
1
= p(s0 ) + e1 (s0 )(s − s0 ) + κ(s0 )e2 (s0 )(s − s0 )2
2
1( )
+ − κ2 (s0 )e1 (s0 ) + κ′ (s0 )e2 (s0 ) + κ(s0 )τ (s0 )e3 (s0 ) (s − s0 )3 . (44)
6 | {z }
leading term
We only consider the leading term in the third order expansion, then we have
1 1( )
p(s) ≈ p(s0 ) + e1 (s0 )(s − s0 ) + κ(s)e2 (s0 )(s − s0 )2 + κ(s0 )τ (s0 )e3 (s0 ) (s − s0 )3 .
2 6
(45)
7
( )
Example (Helix in E3 ). A helix is parametrized as p = x(t), y(y), z(t) with


x(t) = a cos t ,
y(t) = a sin t , (46)


z(t) = bt .
The tangent vector and the corresponding norm are

( ) ( )
ṗ = ẋ, ẏ, ż = − a sin t, a cos t, b (47)
and
√ √
|ṗ| = a2 sin2 t + a2 cos2 t + b2 = a2 + b2 = ṡ . (48)
The relation of s and t can be obtained by

∫ t ∫ t√
ds ′ s
s(t) = dt = a2 + b2 dt′ := ct =⇒ t= . (49)
0 dt′ 0 c
Figure 5: A helix.
Subsequently, we have tangent vector
( )
′ a s a s b
e1 = p = − sin , cos , (50)
c c c c c
and
( )
a s a s
e′1 ′′
=p = − 2 cos , − 2 sin , 0 . (51)
c c c c
So the curvature is
a a
κ = |e′1 | = 2
= 2 . (52)
c a + b2
and the normal vector can be obtained by
( )
s s
e′1 = κe2 =⇒ e2 = − cos , − sin , 0 . (53)
c c
8
Finally, we have binormal vector
( )
b s b s a
e3 = e1 ∧ e2 = sin , − cos , . (54)
c c c c c
Due to
( )
b s b s
e′3 = cos , 2 sin , 0 , (55)
c2 c c c
we can calculate the torsion of p(s) form (40):
b b
τ= = . (56)
c2 a2 + b 2
3 Surface theory in E3
We consider a 2-dimensional
( surface M in E)3 , we parametrize the surface by two variables u and v
written as p(u, v) = x(u, v), y(u, v), z(u, v) .
Remark. If the point p(u, v) moves along u direction, i.e., parametrized by u only, we call the
trajectory u-curve. The infinitesimal vector along u is
∂p(u, v)
∆p|u = p(u + ∆u, v) − p(u, v) ≈ p(u, v) + ∆u − p(u, v) = pu ∆u . (57)
∂u
Similarly, we have v-curve along v direction and
∆p|v ≈ pv ∆v . (58)
Therefore, we have
∆p ≈ pu ∆u + pv ∆v . (59)
Tangent vector The differential of p is

dp = (dx, dy, dz) (60)
with

 ∂x ∂x

 dx = du + dv = xu du + xv dv ,

 ∂u ∂v


∂y ∂y
dy = du + dv = yu du + yv dv , (61)

 ∂u ∂v



 ∂z ∂z
dz = du + dv = zu du + zv dv ,
∂u ∂v Figure 6: A surface.
where xu := ∂x/∂u. Therefore, we can write dp
as
∂p ∂p
dp = du + dv := pu du + pv dv , (62)
∂u ∂v
9
where
{
pu := (xu , yu , zu )
(63)
pv := (xv , yv , zv )
are called velocity vectors along u and v respectively.
Remark. The vector p in E3 in the Cartesian coordinate system can be written as
p = x i + y j + z k := xa δa (a = 1, 2, 3) , (64)
where {δa } is a fixed reference frame of E3 . So that we have differential
dp = (dxa )δa + xa (dδa ) . (65)
Because δa is fixed, i.e. dδa = 0, it leads to the differential of p
dp = (dxa )δa = (dx, dy, dz) . (66)
The general situation for non-fixed frame in space Mn will be discussed in the Sec. 4 of moving
frame.
Tangent space We call a space spanned by pu and pv at point p a tangent space denoted by Tp M.
Supplement (Tangent bundle). In tangent space with dimension 2, a vector V has a generalized
coordinate transformation GL(2; R), which is uei = uei (u) and gives the transformation for vector
‹jp
V = V i pi = V ej . (67a)
The transformation of the basis and components are given by



 ∂ui
p e j (u
e) = p (u) (Pushforward) , (67b)
∂ uej i

 ‹ j (u ∂ui
V i (u) = V e) (Pullback) , (67c)
∂ uej
where
∂ui
(J)i j := (68)
∂ uej
is an element of the Jacobian matrix J of the general linear transformation GL(2; R). The map
pushforward (pullback) means that the covariant (contravariant) quantities expressed in new (old)
coordinate system under the generalized coordinate transformation from old (new) coordinate
system.
We can collect all pairs of the points p on M and their corresponding tangent space Tp M. A
tangent bundle TM is defined by the collection of Tp M, i.e.,
∪
TM = Tp M . (69)
p∈M
A tangent bundle TM is a vector bundle denoted by (E, M, π), which is a special fibre bundle
with
10
• base space B: M;
• standard (typical) fibre F over p (an object defined at p): Tp M;
• total space E: a collection of all Tp M;
• bundle projection π (an element u of bundle is projected by the fibre to the corresponding
point p): π(u) = p for u ∈ TM;
• structure group G: GL(2; R);
• transition function ti j : Jacobian matrix J of GL(2; R),
and we call Ep = π −1 (p) the fibre of E over point p.
First fundamental (quadratic) form we define
I := dp · dp (70a)
= pu · pu dudu + 2 pu · pv dudv + pv · pv dvdv
| {z } | {z } | {z }
E F G
= E dudu + 2F dudv + G dudv (70b)
( )( )
( ) E F du
= du dv . (70c)
F G dv
called the first fundamental form or metric tensor of surface M, which is a symmetric quadratic form
rather than an exterior 2-form.
Remark. In the case of F = 0, the first fundamental form is
I = E dudu + G dvdv (71a)

( )( )
( ) E 0 du
= du dv . (71b)
0 G dv
In such case, we call (u, v) an isothermal coordinates if E = G. Therefore, the component of

metric is
gij = E δij with E = pi · pi = |pi |2 > 0 , (72)
and we say that gij is conformally equivalent to δij , which preserved the angle between any two
vectors. Because δij gives the flat space, we say that gij is conformally flat.
( u and v)should be parametrized by one variable t,

We consider a curve on the surface, that means
i.e., u = u(t) and v = v(t). The curve p(t) = p u(t), v(t) . The tangent vector is obtained by
∂p du ∂p dv
ṗ = + = pu u̇ + pv v̇ , (73)
∂u dt ∂v dt
and the corresponding norm is
√
|ṗ| = E u̇2 + 2F u̇v̇ + Gv̇ 2 . (74)
11
We would like to calculate the arc length of a curve by
∫ ∫
s = ds = |ṗ| dt (75)
∫ √
= Edu2 + 2F dudv + Gdv 2 (76)
| √
{z }
2
∫ √ ds
∫
= Eu′2 + 2F u′ v ′ + Gv ′2 ds = |p′ | ds . (77)
| {z }
1
Therefore, we have
|p′ | = 1 . (78)
We would alwalys write the first fundamental form with u = u1 and v = u2 as
I ≡ ds2 ≡ g = gij dui duj (i, j = 1, 2) , (79)
where
( )
E F
gij = pi · pj −→ (80)
F G
is the metric tensor represented as a 2 × 2 matrix on the surface M. The inverse of gij is defined by
g ki gij = δjk . (81)
Remark. The first fundamental form describes the distance of two points on the surface M, which
gives the intrinsic structure of M.
Supplement (Induced metric). We can regard p as a set of functions defined on the surface M,
the differential of p is actually an infinitesimal tangent vector laid on M
( )
∂p ∂p ∂ ∂
dp = du + dv = du + dv p = (dui ∂i )p , (82)
∂u ∂v ∂u ∂u
which can be identified as differential operator dui ∂i act on a set of functions p. In abbreviated
notation, we have
dp = dui ∂i = dui ⊗ ∂i := ϑ (pi −→ ∂i ) , (83)
where we use ∂i to abbreviate the basis vector pi = ∂i p, i.e., the vector ∂i should be regarded as a
differential operator act on some functions. Here we call ϑ = dp the canonical 1-form or soldering
form, which is a vector-valued 1-form (1-form carries a vector). For any vector V = V k ∂k on
M, we apply ϑ on V and obtain
ϑ(V ) = V k dui (∂k )∂i = V k δki ∂i = V i ∂i = V . (84)
It is apparent that ϑ is an identity map for a vector.
12
Now we will define the general inner product for two basis vectors ∂i and ∂j instead of dot
product as
g(∂i , ∂j ) := gij . (85)
Therefore, for any two vectors V = V i ∂i and W = W j ∂j on M, we have

g(V , W ) = g(V i ∂i , W j ∂j ) = V i W j g(∂i , ∂j ) = V i W j gij = Vj W j = V i Wi . (86)
In general we also have
ḡ(∂a , ∂b ) := ḡab , (87)
where
∂
∂a := (a, b = 1, 2, 3) . (88)
∂xa
We call {xa } the Gauss normal coordinates or synchronous coordinates if ḡi3 = 0 and ḡ33 = 1,
i.e., ∂3 is an unit normal vector of M, which is proportional to n.
Furthermore, we can define a metric tensor
ḡ = ḡab dxa dxb = δab dxa dxb = ds2 (89)
as a line inteval of E3 , and it is clear that ḡi3 is one of the component of ḡ. If we assume that
x1 = x = x(u, v), x2 = y = y(u, v) and x3 = z = z(u, v) on M. We have basis vectors ∂u∂ i
spanned by ∂x∂ a as

 ∂ ∂x ∂ ∂y ∂ ∂z ∂ ∂xa ∂

 = + + = ,
∂u ∂u ∂x ∂u ∂y ∂u ∂z ∂u ∂xa ∂ ∂xa ∂ a ∂
=⇒ = := h i , (90)

 ∂ ∂x ∂ ∂y ∂ ∂z ∂ a
∂x ∂ ∂ui ∂ui ∂xa ∂xa
 = + + = ,
∂v ∂v ∂x ∂v ∂y ∂v ∂z ∂v ∂xa
where ha i is a projection operator of the vector in E3 and i, j = 1, 2. The component of metric
tensor g of M can be given by
( ) ( ) ( )
∂ ∂ a ∂ b ∂ ∂ ∂
gij = g , ≡ ḡ h i a , h j b = h i h j ḡ
a b
, = ha i hb j ḡab , (91)
∂ui ∂uj ∂x ∂x ∂xa ∂xb
which is the projection of ḡab of E3 onto M. We can define an projection operation P of differential
dxa in E3 onto M which is called the pullback (a map for contravariant quantities) of 1-form dxa :
∂xa i
P(dxa ) = du = ha i dui , (92)
∂ui

i.e., P(dxa ) can be spanned by dui on M. As a consequence, a line inteval ds2 M = P(ḡ) on M
is obtained by
( ) ( )
P(ḡ) = P(ḡab dxa dxb ) = ḡab ha i hb j dui duj = ḡab ha i hb j dui duj := gij dui duj = g = I . (93)

Therefore, the first fundamental form I = g = ds2 M of M can be regarded as a projection of
metric tensor ḡ with
∂xa ∂xb
gij := ḡab ha i hb j = ḡab . (94)
∂ui ∂uj
We called that g is a induced metric obtained by the pullback of ḡ.
13
Supplement (Interior product). We define an anti-derivation on exterior differential p-forms ω for
a vector X called interior product with respect to X. It sends an exterior p-form to an exterior
(p − 1)-form. We consider an 1-forms ω, the interior product of ω with respect to a vector X is
or
ιX ω = X⌋ω := ω(X) (95a)
= X i ωj duj (∂i ) = X i ωj δij = X j ωj . (95b)
Therefore, we have
ι∂i (duj ) = duj (∂i ) = δij . (96)
We note that:
• For 0-form f (a scalar), the interior product is vanished ιX f = 0 because of no (−1)-form.
• The second action of ι2X = 0. It can be shown that by considering an exterior 3-form ω =
(1/3!)ωijk dui ∧ duj ∧ duk , we have vanished second interior product by X
( )
1
2
ιX ω = ιX ιX ωijk du ∧ du ∧ du
i j k
3!
( (
1 k
= ιX X ωijk dui (∂l )duj ∧ duk
3!
))
+ (−1) du ∧ du (∂l )du + (−1) du ∧ du du (∂l )
1 i j k 2 i j k
( ))
1 l(
= ιX X ωljk du ∧ du − ωilk du ∧ du + ωijl du ∧ du
j k i k i j
3! |{z} |{z}
−ωlik +ωlij
( )
1 l
= ιX X ωljk duj ∧ duk
2
( ( ))
m l 1 j k 1 j k
=X X ωljk du (∂m )du + (−1) du du (∂m )
2
( ) symmetric in l,m
z }| {
1
= X mX l ωlmk duk − ωljm duj = X m X l ωlmk duk = 0 . (97)
2 |{z} |{z}
−ωlmj anti-symmetric in l,m
However, ιY ιX ̸= 0, e.g. the interior product of an exterior 3-form ω by X and Y should

be an 1-form Y m X l ωlmk duk ̸= 0.
Supplement (Isomorphism between tangent and cotangent space). The 1-form dui is defined on
the cotangent space which is dual to the basis ∂u∂ i ≡ ∂i on the tangent space. We can define a linear
map ψ : TM −→ T ∗M. For vectors X, Y ∈ TM and α ∈ T ∗M the 1-form corresponding to
vector X, then we define
⟨α, Y ⟩ := α(Y ) = ιY α = g(X, Y ) , (98)
where the ⟨•, •⟩ with two slots is a kind of inner product defined between the tangent and cotangent
space and
α := ψ(X) . (99)
14
Therefore, we can also write the corresponding 1-form α as
α(•) = g(X, •) , (100)
which can be recognized by

1 ( )
α := gij dui (X)duj + dui duj (X)
2
1 ( )
= gij X k dui (∂k )duj + dui duj (∂k )
2
1 ( )
= gij X k δki duj + dui δkj
2
= Xi dui . (101)
By choosing vector X = ∂i , we have an 1-form ψ(∂i ) = ψij duj , which leads to
gij = g(∂i , ∂j ) = ⟨ψ(∂i ), ∂j ⟩ = ψik ⟨duk , ∂j ⟩ = ψik δjk = ψij . (102)
Therefore, we have
ψ(∂i ) = gij duj := dui (103)
called the reciprocal basis of dui in T ∗M. It is apparent that gij transforms dui to its reciprocal
basis duj .
In addition, we have a linear inverse map ψ −1 : T ∗M −→ TM such that
X = ψ −1 (α) := φ(α) . (104)
For α = dui , the inverse map of dui can be written as ψ −1 (dui ) = φij ∂j . If we take α = dui and
Y = ∂j in (98), then
( ) ( )
δji = dui (∂j ) = g φik ∂k , ∂j = φik g ∂k , ∂j = φik gkj , (105)
therefore, φik = g ik and
ψ −1 (dui ) = g ij ∂j := ∂ i (106)
is the reciprocal basis of ∂i . We assume that β = ψ(Y ) and define the inner product in T ∗M
which is also denoted by g
( )
g(α, β) := g ψ −1 (α), ψ −1 (β) . (107)
The definition leads to the following relation by choosing α = dui and β = duj
( ) ( ) ( )
g(dui , duj ) = g ψ −1 (dui ), ψ −1 (duj ) = g ∂ k , ∂ j = g ik g jl g ∂k , ∂l = g ij . (108)
Now we can clearly express a vector X = X i ∂i as an inverse map ψ −1 of an 1-form α with
the help of (106):
X = X i ∂i = X i gij ∂ j = Xj ∂ j = Xj ψ −1 (duj ) = ψ −1 ( Xj duj ) = ψ −1 (α) . (109)
| {z } | {z }
a vector! an 1-form!
As a result, we conclude that the metric tensor g (not component gij or g ij ) turns a vector (1-form)
into a 1-form (vector). The component of metric tensor g ij (gij ) transforms the a vector ∂i (1-form
dui ) to its corresponding reciprocal basis ∂ i (dui ).
15
Normal vector of the surface We would like to look for an orthonormal frame (p; e1 , e2 , e3 ) of M.
Under the Gram-Schmit procedure, we can define
pu
e1 := (110)
|pu |
and
pv − (pv · e1 )e1
e2 := . (111)
|pv − (pv · e1 )e1 |
Therefore, we have
e3 = e1 ∧ e2 := n , (112)
which is an unit normal vector of M.

The unit normal vector n = n(u, v) can also be obtained by
p u ∧ pv
n(u, v) = . (113)
|pu ∧ pv |
The corresponding differential dn is

∂n ∂n
dn = du + dv = nu du + nv dv . (114)
∂u ∂v
However, we have:
(i) n ⊥ pu =⇒ n · pu = 0, which have the equations of the partial derivative with respect to u and
v are
{
∂u : nu · pu + n · puu = 0 =⇒ n · puu = −nu · pu := L , (115a)
∂v : nv · pu + n · puv = 0 =⇒ n · puv = −nv · pu := M . (115b)
(ii) n ⊥ pv =⇒ n · pv = 0, we obtain
{
∂u : nu · pv + n · pvu = 0 =⇒ n · pvu = −nu · pv := M , (116a)
∂v : nv · pv + n · pvv = 0 =⇒ n · pvv = −nv · pv := N . (116b)
Second fundamental (quadratic) form According to (62) and (114), we can define a quadratic
form
II := −dp · dn (117a)
= −(pu du + pv dv)(nu du + nv dv)
= −(pu · nu dudu + pu · nv dudv + pv · nu dudv + pv · nv dvdv)
= n · puu dudu + n · puv dudv + n · pvu dudv + n · pvv dvdv
| {z } | {z } | {z } | {z }
L M M N
= L dudu + 2M dudv + N dudv (117b)
( )( )
( ) L M du
= du dv (117c)
M N dv
called the second fundamental form of M. We define the second fundamental form as a tensor given
by
II = bij dui duj (118)
16
with the component of matrix form as
( )
L M
bij = n · pij = −pi · nj −→ . (119)
M N
Remark. The second fundamental form describes the shape of M and how the surface M embed-
ded in E3 . It is an extrinsic property of M and we call the component bij the extrinsic curvature.
Now we would like to discuss decomposition formulas of the derivative vector of frame (p; pu , pv , n).
We follow the principle:
• Any vector in the space can be spanned by the basis pu , pv and n.
Gauss formulas We take the partial derivative of pu and pv with respect to u and v:



∂
pu = (Γu )u u pu + (Γu )v u pv + (Γu )n u n

 puu :=

 ∂u



 = (Γu )u u pu + (Γu )v u pv + (puu · n)n , (120a)

 | {z }




L

 ∂

 puv := pu = (Γu )u v pu + (Γu )v v pv + (Γu )n v n

 ∂v



 = (Γu )u v pu + (Γu )v v pv + (puv · n)n , (120b)

 | {z }

M

 ∂

 pvu := p = (Γv )u u pu + (Γv )v u pv + (Γv )n u n

 ∂u v



 = (Γv )u u pu + (Γv )v u pv + (pvu · n)n , (120c)

 | {z }



 M




∂
pv = (Γv )u v pu + (Γv )v v pv + (Γv )n v n

 pvv :=

 ∂v

 = (Γv )u v pu + (Γv )v v pv + (pvv · n)n .

 (120d)
 | {z }
N
We call these set of equations the Gauss formulas. We identify the coefficients
(Γa )c b ≡ Γc ab , (121)
e.g., (Γn )u v = Γ1 n2 , then Gauss formulas (120) can be written as
pij = Γk ij pk + Γn ij n = Γk ij pk + bij n (i, j, k = 1, 2) , (122)
where the coefficients are obtained by
Γkij = Γl ij glk = Γl ij pl ·pk = pij · pk (123)

| {z }
pij −bij n
and
bij = Γn ij = n · pij . (124)
We note that Γkij and bij are symmetric in i, j.
17
Remark. The vectors dp and dpi in terms of the differential form are given by
dp = (dui ∂i )p = dui pi (125)
and
dpi = d(∂i p) = duj (∂j ∂i p) = duj (Γk ij pk + bij n) := Γk i pk + bi n = Γk i pk + Γn i n (126)
respectively, where Γk i := Γk ij duj is called connection form and bi := bij duj = Γn ij duj = Γn i .
Weingarten formulas Due to n · n = 1, we have

{
∂u : nu · n = 0 =⇒ nu ⊥ n , (127a)
∂v : nv · n = 0 =⇒ nv ⊥ n . (127b)
Therefore, nu and nv do not contain the component of n. We can assume that

{
nu = A pu + B pv = (Γn )uu pu + (Γn )vu pv , (128a)
nv = C pu + D pv = (Γn )uv pu + (Γn )vv pv , (128b)
which is called the Weingarten formulas. We calculate the inner product of nu · pu and nu · pv :
{
nu · pu = −L = EA + F B F M − GL F L − EM
=⇒ A= and B = . (129)
nu · pv = −M = F A + GB EG − F 2 EG − F 2
Similarly, we have
F N − GM F M − EN
C= and D= . (130)
EG − F 2 EG − F 2
The Weingarten formulas (128) can be written by
nj = Γk nj pk , (131)
where the coefficients can be calculated by

( )
−bij = pi · nj = pi · Γl nj pl = gil Γl nj = Γinj =⇒ Γk nj = g ki Γinj = −g ki bij := −bk j .
(132)
As a result, we obtain
nj = −bk j pk . (133)
Remark. The Weingarten formula written in the differential form is given by
dn = (−bk j duj )pk = −bk pk := Γk n pk (134)
18
Acceleration (curvature) vector We have an acceleration (curvature) vector p′′ (s) parametrized
by s, which can be decomposed by tangential and normal parts
p′′ = p′′t + p′′n := κg + κn , (135)
where the tangential part κg = κg t and normal part κn = κn n are called geodesic curvature and
normal curvature respectively.
Remark. We can identify p′′ := a the acceleration vector, therefore, (135) can be read as a =
aT + aN with the tangent acceleration vector aT := p′′t and normal acceleration vector aN := p′′n .
According to (11), we have p′′ · p′ = 0. We would like to discuss the geodesic curvature, we take
the inner product of p′′t with n and p′ respectively:
{
p′′t · n := 0 ,
( ) =⇒ p′′t ∝ t := n ∧ p′ , (136)
p′′t · p′ := p′′t + p′′n · p′ = p′′ · p′ = 0 ,
then we can have p′′t = κg t = κg (n ∧ p′ ).
Normal curvature of a curve We would like to discuss normal curvature first, and define κn =
κn n, so
κn = κn · n = (p′′ − κg ) · n = p′′ · n − κg · n . (137)

| {z }
0
However we also have
p′ · n = 0 =⇒ p′′ · n + p′ · n′ = 0 . (138)
Therefore, κn can be calculated by
κn = −p′ · n′
= −(pu u′ + pv v ′ ) · (nu u′ + nv v ′ )
= L u′ u′ + 2M u′ v ′ + N v ′ v ′ . (139)
However p′ = dp/ds and n′ = dn/ds, which leads to
dp dn −dp · dn II
κn = − · = 2
= . (140)
ds ds ds I
Remark. We have the norm of tangent vector

I
1 = |p′ | = Eu′ u′ + 2F u′ v ′ + Gv ′ v ′ = . (141)
ds2
According to (83)
dp dui
p′ = = ∂i = ui ′ ∂i , (142)
ds ds
19
we have
I(p′ , p′ ) = gkl duk dul (p′ , p′ )

= gkl ui ′ uj ′ duk (∂i ) dul (∂j )
= gkl ui ′ uj ′ δik δjl
= gij ui ′ uj ′ (143a)
= Eu′ u′ + 2F u′ v ′ + Gv ′ v ′ = 1 . (143b)
Similarly,
II(p′ , p′ ) = bij ui ′ uj ′ (144a)

= L u′ u′ + 2M u′ v ′ + N v ′ v ′ = κn . (144b)
We finally obtain
II(p′ , p′ )
κn = ′ ′
= II(p′ , p′ ) . (145)
I(p , p )
We assume that κn has value of λ, which gives the relation
II = λI . (146)
One can divide (146) by ds2 and then obtain

II I ( )
= λ =⇒ L u′ u′ + 2M u′ v ′ + N v ′ v ′ = λ Eu′ u′ + 2F u′ v ′ + Gv ′ v ′ , (147)
ds2 ds2
where λ can be recognized as the Lagrangian multiplier with constraint Eu′ u′ + 2F u′ v ′ + Gv ′ v ′ = 1.
By looking for the extrema λ of κn = II/ds2 , we take the partial derivative of (147) with respect to
ui ′ , which leads to the equation of matrix form
( ) ( ′) ( ) ( ′) ( ) ( ′)
L M u E F u L − λE M − λF u
′ =λ ′ =⇒ = 0, (148)
M N v F G v M − λF N − λG v′
which means
(bij − λgij )uj ′ = 0 . (149)
We have to look for the non-trivial solutions, i.e.,
det(bij − λgij ) = 0 ,
( ) ( )
=⇒ EG − F 2 λ2 − EN + GL − 2F M λ + LN − M 2 = 0 ,
( )
=⇒ gλ2 − EN + GL − 2F M λ + b = 0 , (150)
where we define {
g := det(gij ) = EG − F 2 , (151a)
b := det(bij ) = LN − M 2 . (151b)
As a result, we have the sum and product of two solutions λ1 and λ2
EN + GL − 2F M b
λ1 + λ2 = and λ1 λ2 = . (152)
g g
20
Gauss curvature We define the Gauss curvature (or called total curvature) as product of two cur-
vatures:
b
K := λ1 λ2 = . (153)
g
Mean curvature The mean curvature is defined by the mean value of sum of two curvatures:
1 EN + GL − 2F M
H := (λ1 + λ2 ) = . (154)
2 2g
Remark. The value λα (α = 1, 2) is called the principal curvature of κn . By substituting λα

′
into the equation (149), the corresponding solution of vector p′(α) = uj(α) pj or dp(α) = duj(α) pj is
called the principal direction.
Example (Cylindrical surface in E3 ). A surface parallel with z-axis can be described by

( ) ( ) ( ) ( )
p = x, y, z = x(u), y(u), v =⇒ dp = dx, dy, dz = x′ du, y ′ du, dv . (155)
A cylindrical surface need to have a constraint with

( )2 ( )2
dx dy
+ = x′ 2 + y ′ 2 = 1 . (156)
du du
Figure 7: A cylinder.
A cylinder is parametrized by
( ) ( )
p(u, v) = x(u), y(u), v =⇒ dp = pu du + pv dv = x′ du, y ′ du, dv , (157)
where
{ ( )
pu = x′ du, y ′ du, 0 , ( )
( ) =⇒ pu ∧ pv = y ′ , −x′ , 0 . (158)
pv = 0, 0, 1 .
and the normal vector is

p u ∧ pv 1 ( ′ ) ( )
n= = ′2 y , −x′ , 0 = y ′ , −x′ , 0 (159)
|pu ∧ pv | y +x ′2
21
and
( )
dn = nu du + nv dv = y ′′ du, −x′′ du, 0 . (160)
with
{ ( )
nu = y ′′ , −x′′ , 0 ,
( ) (161)
nv = 0, 0, 0 .
Then we have first fundamental form

( )
I = dp · dp = x′2 + y ′2 du2 + dv 2 = du2 + dv 2 , (162)
where
E = G = 1, F = 0. (163)
The second fundamental form can be obtained by
II = −dp · dn
= −(pu · nu dudu + pu · nv dudv + pv · nu dudv + pv · nv dvdv)
| {z } | {z } | {z }
( ′ ′′ ′ ′′
) 2 0 0 0
= − x y − y x du
( )
= y ′ x′′ − x′ y ′′ du2 , (164)
where
L = y ′ x′′ − x′ y ′′ , M = N = 0. (165)
So we have
g = 1, b = 0, EN + GL − 2F M = y ′ x′′ − x′ y ′′ . (166)
As a result, we obtain
Gauss curvature: K = λ1 λ2 = 0, (167)

1( ) 1( )
mean curvature: H = λ1 + λ2 = y ′ x′′ − x′ y ′′ . (168)
2 2
We can solve the above equations to obtain
λ1 = 0, and λ2 = y ′ x′′ − x′ y ′′ . (169)
Geodesic equations The tangent vector parametrized by s is
p′ (s) = pi ui ′ , (170)
22
it leads to the acceleration vector is given by
p′′ = p′i ui ′ + pi ui ′′
( )
= Γk ij pk + bij n ui ′ uj ′ + pi ui ′′
( )
= uk ′′ + Γk ij ui ′ uj ′ pk + bij ui ′ uj ′ n (171)
= p′′t + p′′n (172)
= κg + κn , (173)
where we have used p′i = pij (duj /ds) = (Γk ij pk + bij n)uj ′ . Now we call the curve p(s) geodesic if
p′′ = p′′n = κn , i.e., the tangential part is vanished
( )
κg = p′′t = uk ′′ + Γk ij ui ′ uj ′ pk = 0 , (174)
which means that
p only has the normal curvature κn .
Because pk s are linear independent, we obtain
uk ′′ + Γk ij ui ′ uj ′ = 0 , (175)
which is called geodesic equations.
Supplement (Connection and geodesic). We have differential of p′ given by
dp′ = dpi ui ′ + pi dui

( )
= Γk i pk + bi n ui ′ + pi dui ′
( )
= duk ′ + Γk i ui ′ pk + bi ui ′ n , (176)
i.e., the symbol d is a total or absolute differential with respect to frame (p; p1 , p2 , n) on E3 . The
total or absolute differentiation means that we have to differentiate not only the component ui ′ but
also the basis pi of a vector p′ . We assume that p′ can be written as
p′ = ui ′ ∂i := V i ∂i = V . (177)
Then (176) can be written as

( )
dV = dV k + Γk i V i ∂k + bi V i n . (178)
and the geodesic equation becomes as

( k )
dV + Γk i V i ∂k = 0 . (179)
We define the connection D = dui ⊗ Di on M, which act on the function V i and basis ∂i are
{
DV i = dV i = duj ⊗ ∂j V i , (180a)
D∂i = Γk i ⊗ ∂k = duj ⊗ Γk ij ∂k , (180b)
respectively. We note that D act on a function as a differential d on a function. The connection

act on a vector is given by
( ) ( ) ( )
DV = DV i ⊗ ∂i + V i D∂i = dV i ⊗ ∂i + V i Γk i ⊗ ∂k = dV k + Γk i V i ⊗ ∂k . (181)
| {z }
(DV )k
23
The resulting geodesic equation (179) can be read as
DV = 0 . (182)
Therefore, (178) can also written as
dV = dV ⊤ + dV ⊥ = DV + bi V i n , (183)
where ⊤ is the orthogonal projection onto the space spanned by {∂k } and ⊥ means the normal
component. If there is no normal space M⊥ of M, the differential
dV = DV (184)
would be the total or absolute differential of a vector V on surface M. We can multiply 1/duj to
the DV :
1 1 ( )
j
DV = j dV k + Γk i V i ∂k
du du
1 ( )
= j dV k + Γk il dul V i ∂k
du
( dul )
= ∂j V k + Γk il j V i ∂k
du
|{z}
δjl
( )
= ∂j V k + Γk ij V i ∂k
( )
:= ∇j V k ∂k , (185)
where we define the component

( )
DV j k := ∇j |{z}
V k = ∇∂j V k = ∂j V k + Γk ij V i (186)
component of V
called the covariant derivative of vector V k in ∂j direction. We note that the covariant derivatives
∇j act on the component of vector V k only.
It can also be recognized as a vector-valued 1-form DV act on a vector ∂j
(( )k i ) ( ) ( )
DV (∂j ) = DV i dx ⊗ ∂k (∂j ) = DV i k dxi (∂j ) ⊗∂k = DV j k ∂k . (187)
| {z }
δji
Now we will return to the discussion of the tangential part of acceleration (curvature) vector
( )⊤
′′ dV DV 1 duj 1 (185) ( )
aT := pt = := = DV = j
DV = V j
∇j V k
∂k := akT ∂k , (188)
ds ds ds ds du
which is the orthogonal projection of the acceleration vector p′′ onto the space spanned by {∂k }.
As a result, we have tangential acceleration with component
akT := V j ∇j V k , (189)
where
V j ∇j = V j ∇∂j = ∇V j ∂j = ∇V . (190)
24
So we can also write akT as
akT = ∇V V k , (191)
which is called the covariant derivative of V k along the direction of V . Then, we call
akT = ∇V V k = 0 or DV = 0 (192)
the parallel transport of tangent vector V , which is equivalent to the geodesic equation.
Remark. We would like to remind you the notation of the covariant dereivative in mathematics
and physics. Consider a vector V , the covariant dereivative (connection) (D or Dj ) of a full vector
V = V i ∂i is
( ) ( ) ( )
DV = duj ⊗ Dj V i ∂i = duj ⊗ ∂j V i +V k Γi kj ∂i := V;ji duj ⊗ ∂i = DV j i duj ⊗ ∂i .
|{z}
V,ji
(193)
We note that here Dj V i = ∂j V i = V,ji and Dj ∂i = ∂k Γk ij as shown in (180). In physics, we

always consider a vector represented by its component V i , the covariant dereivative (∇j ) of a
vector V i should be
( )
∇j V i = DV j i = V,ji + V k Γi kj ≡ V;ji . (194)
Christoffel symbols According to (123), we can calculate the coefficients Γkij . Now we would
like to derive the coefficients in terms of gij , the components of first fundamental form I. We take the
partial derivative of gij with respect to uk and then interchange the indices of the equations. Therefore,
we obtain 

 ∂

 gij = pik · pj + pi · pjk = Γjik + Γijk , (195a)

 ∂uk


∂
g = Γkji + Γjki , (195b)
 ∂ui jk




 ∂

 j gki = Γikj + Γkij . (195c)
∂u
Remark. Equations of (195) gives the metric compatibility, which can be written as the covarinat
deriavative of gij
∇k gij = ∂k gij − Γl ik glj − Γl jk gli = 0 . (196)
We can define the non-metricity
Qkij := −∇k gij (197)
and (196) would be equivalent to the vanished non-metricity Qkij = 0.
25
According to (344), which will be shown later that pi ̸= ∂i p in general, we have general case
that
( ) ( ) ( ) ( )
0 ̸= pij − pji = Dj pi + bij n − Di pj + bji n = Γk ij − Γk ji pk + bij − bji n . (198)
However, in (120), we have pij = ∂j pi = ∂j ∂i p due to the globally fixed frame in E3 , which will
be explained in the Sec. 4 by the reduction condition (345). Consequently, if we have
{
Γk ij = Γk ji , (199a)
pij − pji = 0 . =⇒
bij = bji , (199b)
which give the symmetric condition (torsion-free) for Γk ij and bij .
By computing (195c)+(195b)−(195a), we obtain coefficients

( )
1 ∂ ∂ ∂
Γkij = gki + i gjk − k gij (200)
2 ∂uj ∂u ∂u
and
( )
1 kl ∂ ∂ ∂
Γ k
ij
kl
= g Γlij = g gli + i gjl − l gij . (201)
2 ∂uj ∂u ∂u
Remark. According to (180b), the Γk ij is also called the connection coefficients, or simply the
connection. Due to metric compatibility (196) and torsion-free (304a), the connection (200) or
(201) is a function of the metric tensor gij , we also call this kind of connection the Levi-Civita
connection or Riemannian connection. The Levi-Civita connection can also be denoted as
{
Γkij (g) ≡ [ij, k] , (202a)
{ }
Γk ij (g) ≡ ijk , (202b)
which are called the Christoffel symbol of first and second kind respectively, in order to distinguish
the general connections.
Supplement (Torsion tensor). For general metric compatible connection, we always do not have
the symmetric property, i.e., Γk ij ̸= Γk ji . The connection would contain the symmetric and anti-
symmetric parts, which is shown as
1( k ) 1( k )
Γk ij = Γ ij + Γk ji + Γ ij − Γk ji
2 | {z } 2 | {z }
symmetric in i,j anti-symmetric in i,j
k k
=Γ (ij) +Γ [ij] . (203)
We define the torsion tensor
T k ji := Γk ij − Γk ji = 2 Γk [ij] (204)
26
{k}
as the anti-symmetric part of the connection. We have to note that Γk (ij) ̸= ij . The general
connection can be decomposed as
{k}
Γk ij = ij + K k ij , (205)
and it leads to the relation
T k ji = K k ij − K k ji , (206)
where K k ij is called the contorsion tensor. By permutatiing the indices of (206), it can be show
that the contorsion K k ij can be in terms of torsion tensor T k ij as
symmetric in i,j
1( ) 1( z }| { )
K k ij := − T k ij + T i jk − T j ki = − T k ij + T i jk + T j ik (207)
2 2 |{z}
anti-symmetric in i,j
or
symmetric in i,j
z }| {)
1(
K k ij := − T k ij −Ti k j − Tj k i (in pseudo-Riemannian geometry) . (208)
2 |{z}
anti-symmetric in i,j
So the torsion 2-form can be written as

1 (203) (205)
T k = T k ji duj ∧ dui = Γk ij duj ∧ dui = K k ij duj ∧ dui , (209)
2
where we define K k i := K k ij duj the contorsion 1-form. Therefore, we have the form equation
T k = K k i ∧ dui . (210)
Consequently, the symmetric and anti-symmetric parts of the connection are

 {k}
 k
Γ (ij) = ij + K (ij) ,
k
 1 1
 Γk [ij] = K k [ij] = − T k ij = + T k ji , (211a)
2 2
{k}
respectively, which shows that the symmetric part Γk (ij) contains Levi-Civita connection ij and
torsion T k ij .
Remark. We also note that if we identify
(Γa )c b ≡ Γc ba (212)
and the connection form is defined by Γk j = Γk ij dui , the torsion tensor will be denoted by
T k ij := Γk ij − Γk ji = 2 Γk [ij] . (213)
27
Christoffel symbols in the orthogonal coordinates If we consider the case in the orthogonal co-
ordinates, we have g12 = g21 = g 12 = g 21 = 0 and
1 1 1 1
g 11 = = , g 22 = = . (214)
g11 E g22 G
The component of Christoffel symbols becomes
1 ( )
Γk ij = ∂j gki + ∂i gjk − ∂k gij (no sum) , (215)
2gkk
and we have the following properties:
• For j = k, we have
1 ( ) 1 1 ( )
Γk ik = gki + ∂i gkk −
∂k g
∂kik = ∂i gkk = ∂i ln gkk (no sum) . (216)
2gkk 2gkk 2
• For i = j ̸= k, we have
1 ( ) 1
Γk ii = ∂i
gki + g
∂iik − ∂k gii = − ∂k gii (no sum) . (217)
2gkk 2gkk
• In the general case of dimension > 2, if i ̸= j ̸= k, we have
Γk ij = 0 (in orthogonal coordinates) . (218)
In dimension = 2, it is impossible that i, j, k are all distinct, so we have the same consequence
Γk ij = 0.
Therefore, for dimension = 2, the christoffel symbols in the orthogonal coordinates are given by
Eu Gv
Γ1 11 = , Γ2 22 = ,
2E 2G
Ev Gu
Γ1 12 = Γ1 21 = , Γ2 21 = Γ2 12 = ,
2E 2G
−Gu −Ev
Γ1 22 = , Γ2 11 = . (219)
2E 2G
Example (Polar coordinates). Consider the first fundamental form in orthogonal coordinates
ds2 = E du2 + 2F dudv + G dv 2 = dr2 + r2 dθ2 , (220)
where we have u = r, v = θ and


 {
E = 1 , Er = Eθ = Gθ = 0 ,
F = 0, =⇒ (221)

G = r 2 . Gr = 2r .
The Christoffel symbols are given by

2r 1 −2r
Γ2 21 = Γ2 12 = 2
= , Γ1 22 = = −r . (222)
2r r 2
The geodesic equations is given by (175). Therefore, each component of the geodesic equations
is obtained by 
 ′′ 2
θ + Γ2 12 r′ θ′ + Γ2 21 θ′ r′ = θ′′ + r′ θ′ = 0 , (223a)
r
 ′′
r + Γ1 22 θ′ θ′ = r′′ − rθ′ θ′ = 0 . (223b)
28
Gauss-Codazzi equation Now we have Gauss and Weingarten formulas
{
pik = Γl ik pl + bik n , (224a)
nj = −b j pl ,
l
(224b)
which correspond to the derivative vectors of tangent and normal vectors respectively. By taking the
partial derivative of pik with respect to uj , we have
∂j pik = ∂j Γl ik pl + Γl ik plj + ∂j bik n + bik nj
( )
= ∂j Γl ik pl + Γl ik Γm lj pm + blj n + ∂j bik n + bik (−bl j )pl
( ) ( )
= ∂j Γl ik + Γm ik Γl mj − bik bl j pl + Γl ik blj + ∂j bik n . (225)
By interchanging the indices j and k, we obtain
( ) ( )
∂k pij = ∂k Γl ij + Γm ij Γl mk − bij bl k pl + Γl ij blk + ∂k bij n . (226)
As a consequence, (225)−(226)= 0, which is
∂j pik − ∂k pij = ∂j ∂k pi − ∂k ∂j pi = 0 . (227)
We define the Riemann(-Christoffel) curvature tensor or simply the curvature tensor as
Rl ijk := ∂j Γl ik − ∂k Γl ij + Γl mj Γm ik − Γl mk Γm ij . (228)
and
Rlijk := glm Rm ijk (229)
is defined. Therefore, according to (227), we obtain a set of equations called Gauss-Codazzi equation,
which are given by
0 = ∂j pik − ∂k pij
( ) ( )
= Rl ijk − bik bl j + bij bl k pl + Γl ik blj − Γl ij blk + ∂j bik − ∂k bij n . (230)
| {z } | {z }
0 0
The first one is called Gauss equation

Rl ijk = bik bl j − bij bl k , (231)
and the second one is Codazzi equation
∂j bik − ∂k bij = Γl ij blk − Γl ik blj . (232)
We note that there are some symmetries of curvature tensor:
• Anti-symmetric in the indices j and k
Rl ijk = −Rl ikj . (233)
• Anti-symmetric in l and i (only for Levi-Civita connection)

Rlijk = −Riljk . (234)
• Symmetric in the pairs of li and jk (only for Levi-Civita connection)

Rlijk = +Rjkli . (235)
We also define the traced curvature tensor called Ricci tensor, which is given by
Rik = Rl ilk (236)
and the scalar curvature or Ricci scalar
R = g ik Rik . (237)
29
Theorem Egregium of Gauss The indices i, j, k, l = 1, 2, by the symmetries of Riemann curvature
tensor, the following components are vanished:
R11jk = R22jk = 0 , Rli11 = Rli22 = 0 , (238)
From the Gauss equation (231), the residual component can be given by
R1212 = b22 b11 − b12 b12
= NL − M2
= det(bij ) = b . (239)
Therefore, we can rewrite the Gauss curvature (153) as
b R1212
K= = , (240)
g g
which is a function of gij only, i.e., a 2-dimensional surface in E3 is totally determined by it’s intrinsic
structure. This is the famous intrinsic geometry of Gauss and we call this the theorem Egregium of
Gauss. From the Codazzi equation (232), we only need to consider the case of i = 1, 2 and j = 1 as
well as k = 2. This gives


 ∂b12 ∂b11
 1 − 2
= Γl 11 bl2 − Γl 12 bl1 , (241a)
∂u ∂u
(l = 1, 2) .


 ∂b22 − ∂b21 = Γl 21 bl2 − Γl 22 bl1 , (241b)
∂u1 ∂u2
Remark. In general n-dimensional space Mn , we can always choose a orthonormal frame, so that
g = 1. Then we call K = Rijij for i ̸= j the sectional curvature of the 2-dimensional surface in
Mn , where i, j labeled two components on the surface.
Third fundamental (quadratic) form In analogy we have a Gauss map for a surface, the Gauss
sphere S 2 . A normal vector n will be sent to be a radius vector of S 2 . Therefore, n on S 2 play the
same role as p on the surface M. As a result, we can calculate the first fundamental for of S 2 by
I|S 2 = dn · dn . (242)
Now we define the third fundamental form of M to be the first fundamental form of S 2
III|M := I|S 2 = dn · dn . (243)
From the first fundamental form, we can calculate the area element on the surface M and S 2 ,
which are denoted by ∆A|M and ∆A|S 2 respectively. The results can be obtained by
{
∆p ≈ pu ∆u + pv ∆v =⇒ ∆A|M = |pu ∆u ∧ pv ∆v| = |pu ∧ pv |∆u∆v , (244a)
∆n ≈ nu ∆u + nv ∆v =⇒ ∆A|S 2 = |nu ∆u ∧ nv ∆v| = |nu ∧ nv |∆u∆v . (244b)
However, the vectors nu and nv are given by the Weingarten formulas (128) with (129) and (130).
We can express nu ∧ nv in terms of pu and pv by
nu ∧ nv = AD(pu ∧ pv ) − BC(pu ∧ pv )
F M − GL F M − EN F L − EM F N − GM
= (pu ∧ pv ) − (p ∧ pv )
EG − F EG − F
2 2 EG − F 2 EG − F 2 u
LN − M 2 b
= (pu ∧ pv ) = (pu ∧ pv ) = K(pu ∧ pv ) . (245)
EG − F 2 g
30
Consequently, we can measure the absolute value of Gauss curvature by the ratio
∆A|S 2 |nu ∧ nv |∆u∆v |K||pu ∧ pv |∆u∆v
= = = |K| . (246)
∆A|M |pu ∧ pv |∆u∆v |pu ∧ pv |∆u∆v
4 Cartan’s moving frame and exterior differentiation methods

We would like to introduce a very useful lemma of Cartan first.
Lemma (Cartan’s lemma). Consider a set of linearly independent frame {ei } (or coframe {ϑi }) with
i = 1, . . . , p (p < n) in n-dimensional space M and {Ei } is another set of frame. If ei ∧ Ei =
e1 ∧ E1 + · · · + ep ∧ Ep = 0, then Ei = cij ej and cij = cji .
Proof. We set the linearly independent frame in M by extending to n-tuple from ei given by
e1 , e2 , . . . , ep , ep+1 , . . . , en (247)
| {z } | {z }
p n−p
with index α labeled components p + 1, p + 2, . . . , n. We assume that Ei is expanded by frame in M

as
Ei = cij ej + ciα eα . (248)
According to ei ∧ Ei = 0, we have
0 = ei ∧ Ei = cij ei ∧ ej + ciα ei ∧ eα
1( )
= cij − cji ei ∧ ej + ciα ei ∧ eα . (249)
2 | {z } |{z}
0 0
Therefore, we obtain the coefficients of Ei

{
cij = cji , (250a)
ciα = 0 , (250b)
which means that Ei = cij ej is constructed by {ej } only.
Orthonormal frame We have dp = pi dui with the basis pi , under the Gram-Schmit procedure,
we can obtain an orthonormal frame (p; e1̂ , e2̂ , e3̂ ) given by (110), (111) and (112), where we use the
hatted indices to label the component of orthonormal frame now. We can expand pi by eî shown as
( ) ( 1̂ )( )
ĵ p1 a 1 a2̂ 1 e1̂
pi = a i eĵ (î, ĵ = 1̂, 2̂) or = 1̂ 2̂
, (251)
p2 a2 a2 e2̂
where we call the expansion factor aĵ i the vielbein (vierbein or tetrad for 4-dimension), which can
be regarded as the GL(2, R) transformation of the frame on M. Therefore, we have to obtain the
differential of frame (p; e1̂ , e2̂ , e3̂ ). First we can rewrite dp spanned by frame {eâ } as
dp = p1 du1 + p2 du2
( ) ( )
= a 1 e1̂ + a 1 e2̂ du + a 2 e1̂ + a 2 e2̂ du2
1̂ 2̂ 1 1̂ 2̂
( ) ( )
1̂ 1 1̂ 2 2̂ 1
= a 1 du + a 2 du e1̂ + a 1 du + a 2 du e2̂ := ϑ1̂ e1̂ + ϑ2̂ e2̂
2̂ 2
(252)
31
with
ϑî := aî j duj (253)
Then we have to introduce the connection ω î ĵ for for eî . As a result, deî can be shown as

 1̂ 2̂ 3̂
de1̂ = ω 1̂ e1̂ + ω 1̂ e2̂ + ω 1̂ e3̂ ,
de2̂ = ω 1̂ 2̂ e1̂ + ω 2̂ 2̂ e2̂ + ω 3̂ 2̂ e3̂ , =⇒ deâ = ω b̂ â eb̂ (â, b̂ = 1̂, 2̂, 3̂) . (254)


de3̂ = ω 1̂ 3̂ e1̂ + ω 2̂ 3̂ e2̂ + ω 3̂ 3̂ e3̂ .
In particular, we call the connection form ω b̂ â = ω b̂ âĉ ϑĉ the linear connection form and the coefficient
ω b̂ âĉ the Ricci rotation coefficients in the orthonormal (non-coordinate) frame. However, we have
condition for ω b̂ â due to the orthogonality
eî · eĵ = δîĵ and eâ · e3̂ = δâ3̂ . (255)
We can differentiate the orthogonality condition eî · eĵ = δîĵ , thus we have
d(eî · eĵ ) = deî · eĵ + eî · deĵ

= ω k̂ î ek̂ · eĵ + eî · ω k̂ ĵ ek̂
= ω k̂ î δk̂ĵ + ω k̂ ĵ δîk̂
= ω ĵ î + ω î ĵ = 0 . (256)
Similarly, we have
d(eî · e3̂ ) = ω 3̂ î + ω î 3̂ = 0 (257)
from eî · e3̂ = δî3̂ . Therefore, all the components of ω b̂ â are anti-symmetric in the orthonormal frame,
we have the consequence:
• The metric compatibility gives the anti-symmetric property for linear connection form in the
orthonormal frame, i.e.,
∇ĉ gâb̂ = ∇ĉ δâb̂ = eĉ (δâb̂ ) −ω b̂ âĉ − ω â b̂ĉ = 0 =⇒ ω b̂ âĉ = −ω â b̂ĉ . (258)
| {z }
0
Remark. In pseudo-Riemannian space, we have
d(eâ · eb̂ ) = ω ĉ â ηĉb̂ + ω ĉ b̂ ηâĉ = ωb̂â + ωâb̂ = 0 , (259)
and the metric compatibility in pseudo-orthonormal frame should be read as

ˆ ˆ
∇ĉ gâb̂ = ∇ĉ ηâb̂ = eĉ (ηâb̂ ) −ω d âĉ ηdˆb̂ − ω d b̂ĉ ηâdˆ = 0 =⇒ ωb̂âĉ = −ωâb̂ĉ . (260)
| {z }
0
32
Finally the equation (254) is reduced to



2̂ 3̂
de1̂ = ω 1̂ e2̂ + ω 1̂ n , (261a)
de2̂ = ω 1̂ 2̂ e1̂ + ω 3̂ 2̂ n , (261b)


 1̂ 2̂
dn = ω 3̂ e1̂ + ω 3̂ e2̂ . (261c)
Now we can write down the first, second and third fundamental form in orthonormal frame, which
are given by  ( 1̂ )2 ( 2̂ )2

 I = dp · dp = ϑ + ϑ , (262a)

II = −dp · dn = −ϑ1̂ ω 1̂ 3̂ − ϑ2̂ ω 2̂ 3̂ = ω 3̂ 1̂ ϑ1̂ + ω 3̂ 2̂ ϑ2̂ , (262b)

 ( ) ( ) ( ) ( )
 2 2 2 2
III = dn · dn = ω 1̂ 3̂ + ω 2̂ 3̂ = ω 3̂ 1̂ + ω 3̂ 2̂ . (262c)
It can be shown that ω 3̂ 1̂ and ω 3̂ 2̂ are linear combinations of ϑî or dui given by (290) due to Cartan’s
first structure equation (be introduced later) and Cartan’s lemma
{ ( 3̂ ) ( ) ( 1̂ )
ω 3̂ 1̂ = b1̂1̂ ϑ1̂ + b1̂2̂ ϑ2̂ , ω 1̂ b1̂1̂ b1̂2̂ ϑ
=⇒ 3̂
= . (263)
3̂ 1̂
ω 2̂ = b2̂1̂ ϑ + b2̂2̂ ϑ . 2̂ ω 2̂ b2̂1̂ b2̂2̂ ϑ2̂
Consequently, the second fundamental form becomes

( )2 ( )2
II = b1̂1̂ ϑ1̂ + b1̂2̂ ϑ2̂ ϑ1̂ + b2̂1̂ ϑ1̂ ϑ2̂ + b2̂2̂ ϑ2̂ = bîĵ ϑî ϑĵ . (264)
We consider the following matrix representation for tensors
eî −→ e , ϑî −→ ϑ , bîĵ −→ B . (265)
Under the special orthogonal transformation SO(2, R) for frame and coframe, we have a new or-
thonormal frame {
e′ = P e
with P T = P −1 , (266)
ϑ′ = P T ϑ
where P ∈ SO(2, R) and T means the transpose operation for matrix. As a result, we can obtain the
diagonal matrix BD from B through P by
II = ϑT Bϑ = (ϑ′ )T P T
| {zBP} ϑ′ , (267)
BD
i.e.,
( )
κ1 0
B −→ BD = P BP = T
. (268)
0 κ2
Therefore, the Gauss curvature and mean curvature can be obtained easily by


K = det(BD ) = det(P ) det(B) det(P ) = det(B) = b1̂1̂ b2̂2̂ − b1̂2̂ b2̂1̂ ,
T
 (269a)

 1 1 1( )
H = tr BD = tr B = b1̂1̂ + b2̂2̂ , (269b)
2 2 2
respectively, where the trace of the matrix B is invariant under the SO(2, R) transformation.
33
Covariant exterior differentiation We define some notation for differential operators for function,
vector and 1-form. We use d, d and d∇ for differentiation, exterior differentiation and covariant
exterior differentiation respectively.
• For a function (0-form) f , the differential df which can also be regarded as the exterior differ-
entiation of 0-form f :
d∇ f = df = df . (270)
• For a vector eî , we have an absolute differential of vector deî which is described by Gauss
formulas in differential form formalism (126):
d∇ eî = deî = Deî + bî n , (271)
is a vector-valued 1-form. If there is no normal space M⊥ of M, i.e., there are no n vector and
bîĵ , the differential is actually equal to the orthogonal projection of vector eî on M
d∇ eî = deî = Deî . (272)
• For an 1-form ϑî , we only do the exterior differentiation on ϑî :
d∇ ϑî = dϑî . (273)
Remark. The covariant exterior differentiation d∇ is a combined operator, which do the exterior
differentiation and covariant derivative on an 1-form and vector respectively.
For a function f , we also note that the second differentiation is
∂ 2f ∂ 2f ∂ 2f ∂ 2f
d2 f (x, y) = dxdx + dxdy + dydx + dydy ̸= 0 , (274)
∂x∂x ∂y∂x ∂x∂y ∂y∂y
which should not be confused with the second exterior differentiation
( 2 )
∂ 2f ∂ 2f ∂ f ∂ 2f
2
d f (x, y) = ddf = dx ∧ dy + dy ∧ dx = − dx ∧ dy = 0 . (275)
∂y∂x ∂x∂y ∂y∂x ∂x∂y
In addition, d2∇ would not be vanished in general. Therefore, the second fundamental form is
II = −dp · dn = +d2 p · n = (pij dui duj ) · n = bij dui duj (276)
due to dp · n = 0 which has been shown in the last term in (120). We note that d2 p should be
realized as a second covariant derivatives of p in (280).
For coframe dui , the corresponding exterior differentiation is vanished, which is shown as
ddui = d2 ui = 0 . (277)
We call ∂u∂ i a holonomic frame and dui a holonomic coframe which is an exact form according to
the Poincaré lemma. For ϑî = aî j duj , its exterior differentiation is
dϑî = daî j ∧ duj + aî j d2 uj ̸= 0 , (278)
which is called an anholonomic coframe or a Pfaffian form dual to the anholonomic frame eî .
34
Supplement. We note that the exterior 2-form d2∇ eî will be introduced as the second structure
equation through the covariant exterior differentiation and related to the curvature 2-form of (293)
and structure constants of (315) later
1
d2∇ eî = d∇ (deî ) = Rl̂ îĵ k̂ ϑĵ ∧ ϑk̂ ⊗ el̂ (279a)
2
= d∇ (ϑ Dk̂ eî ) = d∇ ϑk̂ ⊗ Dk̂ eî −ϑk̂ ∧ ϑĵ ⊗ Dĵ Dk̂ eî
k̂
| {z } |{z}
(315)
Γl̂ îk̂ el̂
1 k̂ 1( )
= − c ĵ m̂ ϑ ∧ ϑ ⊗ Γ îk̂ el̂ + ϑ ∧ ϑ ⊗
ĵ m̂ l̂ ĵ k̂
Dĵ Dk̂ − Dk̂ Dĵ eî
2 ( 2
1 ( ) )
k̂↔m̂ ĵ
= ϑ ∧ϑ ⊗ k̂
− c ĵ k̂ Γ îm̂ el̂ + Dĵ Dk̂ − Dk̂ Dĵ eî
m̂ l̂
2 | {z }
Dm̂ eî
1( )
=ϑ ∧ϑ ⊗
ĵ k̂
− c ĵ k̂ Dm̂ + Dĵ Dk̂ − Dk̂ Dĵ eî .
m̂
(279b)
2
However, the second covariant derivatives of a vector eî should be
d2 eî = D2 eî = D(ϑk̂ ⊗ Dk̂ eî ) = ϑĵ ⊗ Dĵ ϑk̂ ⊗ Dk̂ eî +ϑĵ ⊗ ϑk̂ ⊗ Dĵ Dk̂ eî
| {z } |{z}
Γk̂ ĵ m̂ ϑm̂ Γl̂ îk̂ el̂
( )
k̂↔m̂
= ϑĵ ⊗ ϑk̂ ⊗ Γm̂ ĵ k̂ Γl̂ îm̂ el̂ +Dĵ Dk̂ eî
| {z }
Dm̂ eî
( )
= ϑĵ ⊗ ϑk̂ ⊗ Γm̂ ĵ k̂ Dm̂ + Dĵ Dk̂ eî . (280)
Therefor, we conclude that d2∇ eî ̸= d2 eî because deî is a vector-valued 1-form. We note that that
the wedge product is obtained by anti-symmetrizing the tensor product
1 1
A ∧ B = (A ⊗ B)A = (A ⊗ B − B ⊗ A) = (Aî Bĵ − Bĵ Aî )ϑî ∧ ϑĵ , (281)
2! 2
where A indicates the anti-symmetrization. The anti-symmetrization of Dĵ Dk̂ eî and Γm̂ ĵ k̂ will be
given later, which are shown in (318) and (327b) respectively. As a result, the anti-symmetrization
of d2 eî can be shown as
1( )
(d2 eî )A = (D2 eî )A = D ∧ Deî = ϑĵ ∧ ϑk̂ ⊗ − T m̂ ĵ k̂ Dm̂ + Rl̂ îĵ k̂ el̂ (cf. (279a) or (293)) .
2
(282)
Canonical 1-form In general case, {eâ } does not necessarily be chosen as orthonormal, i.e., the
metric tensor is eâ · eb̂ = gâb̂ ̸= δâb̂ . If {eâ } is an orthonormal frame, we have anti-symmetric
property of (256) and (257). We would discuss from the differential of frame (p; e1̂ , e2̂ , e3̂ ) and write
the equations by the covariant exterior differentiation as
{
d∇ p = ϑâ ⊗ eâ := ϑ , (283a)
d∇ eâ = ω b̂ â ⊗ eb̂ , (283b)
35
where we have defined ϑ := d∇ p = dp the canonical 1-form, which is a vector-valued 1-form. We
will show that the canonical 1-form ϑ is an identity map of vector in the frame eâ . Consider a vector
V = V b̂ eb̂ , the canonical 1-form act on V gives
ϑ(V ) = ϑâ ⊗ eâ (V b̂ eb̂ ) = V b̂ ϑâ (eb̂ )eâ = V b̂ δb̂â eâ = V b̂ eb̂ = V . (284)
Remark. If we consider a point p move on the surface M in E3 , the differential would be a vector
spanned by e1̂ and e2̂ only, which would be written as
d∇ p = ϑâ eâ = ϑî eî = ϑ1̂ e1̂ + ϑ2̂ e2̂ (285)
with ϑ3̂ = 0, it would be reduced to the equation given by (252).
Cartan’s first structure equation Now we do the covariant exterior differentiation on (283). The
covariant exterior differentiation of (283a) is
d∇ ϑ = d2∇ p = d∇ (ϑâ ⊗ eâ )

= dϑâ ⊗ eâ + (−1) ϑâ ∧ D̄eâ
= dϑâ eâ − ϑâ ∧ ω b̂ â eb̂
( )
= dϑâ + ω â b̂ ∧ ϑb̂ eâ
( )
= d∇ ϑ â eâ
:= T â eâ = T ̸= 0 , (286)
where D̄ is a connection with respect to eâ . Here we have defined T the vector-valued torsion 2-form
and the corresponding component the torsion 2-form as
( )
T â := d∇ ϑ â := D |{z}
ϑâ := dϑâ + ω â b̂ ∧ ϑb̂ , (287)
component of ϑ
where D can be identified as operation
D = d + ω∧ (288)
act on the differential form which is the component of the corresponding vector-valued form. The
equation (287) we obtained is called Cartan’s first structure equation.
Remark. Since p moves on the surface M in E3 , we have d∇ p = ϑ1̂ e1̂ + ϑ2̂ e2̂ and ϑ3̂ = 0. Follow
Cartan’s first structure equation (287), we have
0 = d∇ ϑ3̂ = −ω 3̂ 1̂ ∧ ϑ1̂ − ω 3̂ 2̂ ∧ ϑ2̂ = −ω 3̂ î ∧ ϑî . (289)
According to Cartan’s lemma, the connection form is obtained as
ω 3̂ î = bîĵ ϑĵ , (290)
which gives the equations (263).
36
(a) Torsion in space. (b) Torsion on surface.
Figure 8: Torsion is related to the translation.
We can consider an infinitesimal contour integral for d∇ p infinitesimally around a point as a

boundary ∂D of a small region D. By applying Stokes’ theorem to the contour integral of d∇ p over
∂D gives
I ∫ ∫
d∇ p = 2
d∇ p = T (291)
∂D D D
or equivalent to
I ∫ ∫
â
ϑ = Dϑ = â
T â . (292)
∂D D D
The integral result implies that the translation of a point or the displacement d∇ p is associated with
the torsion. If there is no displacement, i.e. d∇ p = 0, the space would not be twisted.
Cartan’s second structure equation Similarly, we do the covariant exterior differentiation on

(283b) and obtain
d2∇ eâ = d∇ (ω b̂ â ⊗ eb̂ )

= dω b̂ â ⊗ eb̂ + (−1) ω b̂ â ∧ D̄eb̂
= dω b̂ â eb̂ − ω b̂ â ∧ ω ĉ b̂ eĉ
( )
= dω b̂ â + ω b̂ ĉ ∧ ω ĉ â eb̂
( )
= d2∇ eâ b̂ eb̂
:= Rb̂ â eb̂ = Râ ̸= 0 . (293)
Therefore we have the vector-valued curvature 2-form Râ with the corresponding component curva-
ture 2-form given by
( )
Rb̂ â := d2∇ eâ b̂ := D ω b̂ â := dω b̂ â + ω b̂ ĉ ∧ ω ĉ â , (294)
|{z}
component of d∇ eâ
which is called Cartan’s second structure equation.
37
Figure 9: Curvature is related to the rotation.
The similar infinitesimal contour integral for d∇ eâ gives

I ∫ ∫
d∇ eâ = 2
d∇ eâ = Râ (295)
∂D D D
or equivalent to
I ∫ ∫
ω b̂
â = Dω b̂
â = Rb̂ â , (296)
∂D D D
which means that the rotation of a vector is associated with the curvature. If the vector does not
change the direction after moving around a contour, i.e. d∇ eâ = 0, the space would be flat.
First Bianchi identity The exterior differentiation of two structure equations can get more infor-
mation of torsion and curvature. The structure equations are
{
T â = dϑâ + ω â b̂ ∧ ϑb̂ , (297a)
Râ b̂ = dω â b̂ + ω â ĉ ∧ ω ĉ b̂ . (297b)
We take the exterior differentiation of the first structure equation shown by
dT â = d2 ϑâ + dω â b̂ ∧ ϑb̂ − ω â b̂ ∧ dϑb̂

( ) ( )
= Râ b̂ − ω â ĉ ∧ ω ĉ b̂ ∧ ϑb̂ − ω â b̂ ∧ T b̂ − ω b̂ ĉ ∧ ϑĉ

= Râ b̂ ∧ ϑb̂ −
ω â ∧
ĉ ω ĉ
b̂ ∧ ϑ − ω b̂ ∧ T +
b̂ â b̂
ω â ∧
b̂ ω b̂
ĉ ∧ ϑ ,
ĉ
(298)
then we obtain the first Bianchi identity
DT â = dT â + ω â b̂ ∧ T b̂ = Râ b̂ ∧ ϑb̂ . (299)
If we have torsion-free condition T â = 0, the first Bianchi identity becomes
0 = Râ b̂ ∧ ϑb̂
1 ˆ
= Râ b̂ĉdˆϑĉ ∧ ϑd ∧ ϑb̂
2
1( ) ĉ dˆ
= Râ b̂ĉdˆ + Râ ĉdˆb̂ + Râ dˆb̂ĉ − Râ b̂dĉ â â
ˆ b̂ ϑ ∧ ϑ ∧ ϑ
ˆ − R ĉb̂dˆ − R dĉ
b̂
2
( ) ˆ
= Râ b̂ĉdˆ + Râ ĉdˆb̂ + Râ dˆb̂ĉ ϑĉ ∧ ϑd ∧ ϑb̂ , (300)
38
resulting in
Râ b̂ĉdˆ + Râ ĉdˆb̂ + Râ dˆb̂ĉ = 0 . (301)
Second Bianchi identity Similarly, the exterior differentiation of the second structure equation is
dRâ b̂ = d2 ω â b̂ + dω â ĉ ∧ ω ĉ b̂ − ω â ĉ ∧ dω ĉ b̂
( ˆ ) ( ˆ )
= Râ ĉ − ω â dˆ ∧ ω d ĉ ∧ ω ĉ b̂ − ω â ĉ ∧ Rĉ b̂ − ω ĉ dˆ ∧ ω d b̂

ω dĉ ∧ ω ĉ b̂ − ω â ĉ ∧ Rĉ b̂ + ω ĉ
ˆ ˆ
= Râ ĉ ∧ ω ĉ b̂ −
ω â ∧
dˆ ω â ∧
ĉ dˆ ∧ ω b̂ ,
d
(302)
| {z }
+ω ĉ b̂ ∧Râ ĉ due to 2-form Râ ĉ
which leads to the second Bianchi identity
DRâ b̂ = dRâ b̂ − ω ĉ b̂ ∧ Râ ĉ + ω â ĉ ∧ Rĉ b̂ = 0 . (303)
Remark. It is essential to consider the geometric structure from Cartan’s viewpoint. The first
structure equations in Riemannian geometry is restricted to be torsion-free condition. Then the
structure equations are reduced to
{
dϑâ = −ω â b̂ ∧ ϑb̂ = +ϑb̂ ∧ ω â b̂ , (304a)
Râ b̂ = dω â b̂ + ω â ĉ ∧ ω ĉ b̂ . (304b)
Because of metric compatibility, we have ω â b̂ = −ω b̂ â (or ωâb̂ = −ωb̂â in pseudo-Riemannian

geometry), which gives
ω â b̂ = ω â b̂ĉ ϑĉ = −ω b̂ â = −ω b̂ âĉ ϑĉ =⇒ ω â b̂ĉ = −ω b̂ âĉ . (305)
Due to (304a), we obtain

1 ( â )
dϑâ = ϑb̂ ∧ ω â b̂ = ω â b̂ĉ ϑb̂ ∧ ϑĉ = ω b̂ĉ − ω â ĉb̂ ϑb̂ ∧ ϑĉ . (306)
2
However, dϑâ is a 2-form, it can be written as
1
dϑâ = aâ b̂ĉ ϑb̂ ∧ ϑĉ , (307)
2
which leads to
aâ b̂ĉ = ω â b̂ĉ − ω â ĉb̂ . (308)
By permutating the indices â, b̂ and ĉ, we have the equation
aâ b̂ĉ + ab̂ ĉâ − aĉ âb̂ = ω â b̂ĉ − ω b̂ âĉ = 2ω â b̂ĉ . (309)
The resulting connection coefficients are

1 ( â )
ω â b̂ĉ = a b̂ĉ + ab̂ ĉâ − aĉ âb̂ . (310)
2
39
It can be shown that
aâ b̂ĉ = −câ b̂ĉ , (311)
where câ b̂ĉ is called the structure constants or commutation coefficients, which is defined by the
commutation relation of the anholonomic frame
[ ]
a ∂ b ∂
[eâ , eb̂ ] = aâ ,a
∂xa b̂ ∂xb
( ) ( )
a ∂ b ∂ b ∂ a ∂
= aâ a − ab̂ aâ
∂xa b̂ ∂xb ∂xb ∂xa
( ) ( )
∂ ∂ ∂ ∂
= aâ a a ab̂ b b
− ab̂ b b aâ a
∂x ∂x ∂x ∂xa
( ) ( )
∂ ∂
= aâ a a ab̂ b aĉ b eĉ − ab̂ b b aâ a aĉ a eĉ
∂x ∂x
[ ( ) ( ) ]
a ∂ b ∂
= aâ a
ab̂ a b − ab̂
b ĉ
b
aâ a aĉ a eĉ
∂x ∂x
[ ( b ) ĉ ( a ) ĉ ]
= eâ ab̂ a b − eb̂ aâ a a eĉ := cĉ âb̂ eĉ . (312)
Here the anholonomic frame eâ is identified as the so-called Pfaffian derivative. As a result, we
obtain the the structure constants
( ) ( )
cĉ âb̂ := eâ ab̂ b aĉ b − eb̂ aâ a aĉ a (313)
We note that it is apparent that the commutator
[∂a , ∂b ] = 0 (314)
because two partial derivatives can be interchanged. As a result, we conclude that there is no
structure constants in holonomic frame. Therefore, the commutation coefficients can also be
called anholonomity which characterizes the property of the anholonomic frame. On the other
hand,
( )
dϑĉ = d aĉ b dxb
( )
= d aĉ b dxb
( )
∂ ĉ
= a b dxa ∧ dxb
∂xa
( )
1 ∂ ĉ ∂ ĉ ( a â ) ( b b̂ )
= a b − b a a aâ ϑ ∧ ab̂ ϑ
2 ∂xa ∂x
( )
1 b a ∂ a b ∂
= ab̂ aâ a
a b − aâ ab̂
ĉ
b
a a ϑâ ∧ ϑb̂
ĉ
2 ∂x ∂x
( )
1
= a eâ (a b ) − aâ eb̂ (a a ) ϑâ ∧ ϑb̂
b ĉ a ĉ
2 b̂
( )
1 ĉ
= − a b eâ (ab̂ ) − a a eb̂ (aâ ) ϑâ ∧ ϑb̂
b ĉ a
2
1
= − cĉ âb̂ ϑâ ∧ ϑb̂ , (315)
2
40
where we have used ab̂ b eâ (aĉ b ) = −aĉ b eâ (ab̂ b ) due to ab̂ b aĉ b = δb̂ĉ . The above result proves (311)
and finally we obtain the linear connection coefficients
1 ( â )
ω â b̂ĉ = − c b̂ĉ + cb̂ ĉâ − cĉ âb̂ (316)
2
or
1 ( â )
ω â b̂ĉ = − c b̂ĉ − cb̂ â ĉ − cĉ â b̂ (in pseudo-Riemannian geometry) . (317)
2
Supplement (Covariant derivative in anholonomic frame). We can do the calculation in both holo-
nomic and anholonomic frame. However, according to (251), we have
( )
Dĵ Dk̂ eî = aĵ j Dj ak̂ k Dk eî
= aĵ j ak̂ k Dj Dk eî + aĵ j (∂j ak̂ k )(Dk eî )
= aĵ j ak̂ k Dj Dk (aî i pi ) + al̂ k (eĵ ak̂ k )(Dl̂ eî )
( )
= aĵ j ak̂ k Dj aî i Dk pi + (∂k aî i )pi + al̂ k (eĵ ak̂ k )(Dl̂ eî )
( )
= aĵ j ak̂ k aî i Dj Dk pi + (∂j aî i )Dk pi + (∂k aî i )Dj pi + (∂j ∂k aî i )pi
+ al̂ k (eĵ ak̂ k )(Dl̂ eî ) . (318)

So we can find
( ) ( ) ( )
Dĵ Dk̂ − Dk̂ Dĵ eî = aî i aĵ j ak̂ k Dj Dk − Dk Dj pi + al̂ k (eĵ ak̂ k ) − al̂ k (ek̂ aĵ k ) (Dl̂ eî ) . (319)
Here we move the last term of (319) to the left-handed side and use the structure constants cl̂ ĵ k̂
defined by (313). Then, we also use (190) and (312) to obtain the equation
( )
aî i aĵ j ak̂ k (Dj Dk − Dk Dj )pi = Dĵ Dk̂ − Dk̂ Dĵ − cl̂ ĵ k̂ Dl̂ eî
( )
= Dĵ Dk̂ − Dk̂ Dĵ − Dcl̂ e eî
( ĵ k̂ l̂
)
= Dĵ Dk̂ − Dk̂ Dĵ − D[eĵ ,ek̂ ] eî . (320)
It gives the general formula of curvature tensor
 ( )

 R l
= D D − D D −
 Holonomic frame: ijk p l j k k j |{z} pi ,
0 (321a)
D[∂j ,∂k ] =0 which is vanished due to (314).

 ( )
Anholonomic frame: Rl̂ e = D D − D D − D[e ,e ] e . (321b)
îĵ k̂ l̂ ĵ k̂ k̂ ĵ ĵ k̂ î
Therefore, one can consider three vectors X = X ĵ eĵ , Y = Y k̂ ek̂ and Z = Z î eî , then it can be
shown that the frame independent formula of curvature tensor is
( )
X ĵ Y k̂ Z î Rl̂ îĵ k̂ el̂ = DX DY − DY DX − D[X,Y ] Z := R(X, Y )Z . (322)
After the calculation, (321) can be written in terms of the connections and structure constants


 Holonomic: Rl ijk = ∂j Γl ik − ∂k Γl ij + Γl mj Γm ik − Γl mk Γm ij , (323a)

Anholonomic: Rl̂ îĵ k̂ = eĵ ω l̂ îk̂ − ek̂ ω l̂ îĵ + ω l̂ m̂ĵ ω m̂ îk̂ − ω l̂ m̂k̂ ω m̂ îĵ − ω l̂ îm̂ cm̂ ĵ k̂ . (323b)
41
Finally (320) gives the transformation formula for curvature tensor by substituting pl = al̂ l el̂ to
the left-handed side of (320)
Rl̂ îĵ k̂ = al̂ l aî i aĵ j ak̂ k Rl ijk . (324)
Similarly, it can( be shown that )by computing the commutator of the covariant derivatives on
function f , i.e., Dĵ Dk̂ − Dk̂ Dĵ f , we obtain the following equations
{ ( )
Holonomic frame: T i jk Di f = T i jk ∂i f = Dj Dk − Dk Dj f , (325a)
( )
Anholonomic frame: T î ĵ k̂ Dî f = T î ĵ k̂ eî f = Dĵ Dk̂ − Dk̂ Dĵ − [eĵ , ek̂ ] f , (325b)
and the frame independent formula of torsion tensor is given by vectors X = X ĵ eĵ and Y = Y k̂ ek̂
with eî := Dî f of
X ĵ Y k̂ T î ĵ k̂ eî = DX Y − DY X − [X, Y ] := T (X, Y ) . (326)
In terms of the connections and structure constants, (325) can be written as



 Holonomic: T i jk = Γi kj − Γi jk , (327a)

Anholonomic: T î ĵ k̂ = ω î k̂ĵ − ω î ĵ k̂ − cî ĵ k̂ . (327b)
Here we note that a vector X act on a function f and a avector Y are respectively given by
X(f ) = X ĵ eĵ (f ) and X(Y ) = X ĵ eĵ (Y k̂ )ek̂ + X ĵ Y k̂ eĵ ek̂ . (328)
In addition, the transformation formula for torsion tensor should be
T î ĵ k̂ = aî i aĵ j ak̂ k T i jk . (329)
Non-fixed frame and gauge transformation For general space Mn , the vector V = V a Ea (a =
1, . . . , n) under local coordinate X a can be spanned by a non-fixed holonomic frame Ea := ∂X∂ a with
dEa ̸= 0. The vector Ea can be spanned by another set of anholonomic frame eb̂ given by a GL(n, R)
transformation
Ea = Ab̂ a eb̂ with Ab̂ a ∈ GL(n, R) , (330)
and we also have coframe
dX a = Ab̂ a ϑb̂ , (331)
where we have defined the inverse
Ab̂ a := (A−1 )b̂ a . (332)
Therefore,
dV = dV a Ea + V a dEa (333)
42
and we have to introduce the connection form Γb a and ω b̂ â for the frame Ea and eâ respectively, which
gives
{
dEa = Γb a Eb , (334a)
deâ = ω b̂ â eb̂ . (334b)
The differential of V can also be expressed as
( )
dV = dV a Ea + V a Γb a Eb = dV a + V b Γa b Ea := (D̄V )a Ea . (335)
Remark. We note that there is no normal space Mn⊥ of Mn , therefore, we obtain
dV = (dV )⊤ = D̄V , (336)
where the connection D̄ is defined with respect to basis Ea . As a result, the differetial operator d
also represents the covariant derivative D̄ on general space Mn .
On the other hand, the differetial

( )
dEa = dAb̂ a eb̂ + Ab̂ a deb̂ = dAb̂ a eb̂ + Ab̂ a ω ĉ b̂ eĉ = dAb̂ a + Aĉ a ω b̂ ĉ eb̂ . (337)
From (330) and (334a), it implies the relation between two connection forms
( ) ( )
Γc a Ec = Γc a Ab̂ c eb̂ = dAb a + Aĉ a ω b̂ ĉ eb̂ =⇒ Γc a = Ab̂ c dAb̂ a + ω b̂ ĉ Aĉ a , (338)
which is a frame transformation or GL(n, R) gauge transformation of the connection form.
Remark. We note that the relation (338) comes from the frame transformation or GL(n, R) gauge
transformation, rather than metric compatibility. This relation is sometimes called vielbein postu-
late. However, the equation is still valid even if the frame eâ is not orthonormal or the nonmetricity
Qabc = −∇a gbc is not vanished. In such case, the connection ω b̂ â contains the symmetric or trace
part, i.e., ω(âb̂) ̸= 0 or ω â â ̸= 0. So it is improper to call the relation postulate. People always
implicitly define a total connection D(Γ, ω) of tensor with respect to both the holonomic and an-
holonomic basis of Ea , dX a , eâ and ϑâ . By giving a tensor A := Ab̂ a eb̂ ⊗ dX a , the connection
D act on A is
( )
DA = D Ab̂ a eb̂ ⊗ dX a
( ) ( ) ( )
= dAb̂ â eb̂ ⊗ dX â + Ab̂ a Deb̂ ⊗ dX a + Ab̂ a eb̂ ⊗ DdX a
( ) ( ) ( )
= dAb̂ a eb̂ ⊗ dX a + Ab̂ a ω ĉ b̂ eĉ ⊗ dX a + Ab̂ a eb̂ ⊗ − Γa c dX c
( )
= dAb̂ a + Aĉ a ω b̂ ĉ − Ab̂ c Γc a eb̂ ⊗ dX a = 0 (339)
due to (338), which is independent of the metric compatibility. As a result, the component gives
the vielbein postulate
(DA)b̂ a = dAb̂ a + Aĉ a ω b̂ ĉ − Ab̂ c Γc a = 0 (340)
or
∇d Ab̂ a = ∂d Ab̂ a + Aĉ a ω b̂ ĉd − Ab̂ c Γc ad = 0 . (341)
43
Figure 10: Two points on the hypersurface Mn−1 in Mn .
Now we will discuss the connection on the hypersurface Mn−1 of M n . Consider a point p on
Mn−1 is identified by a vector V in M n . Simirlarly, a point q on Mn−1 is represented by V ′ in
M n . Here we only focus on the connection on Mn−1 , we have restricted our case that dV = q−p =
V ′ − V lays on the hypersurface Mn−1 only. So dV can be expanded not only by frame Ea = ∂X∂ a
(a = 1, . . . , n) on M n but also by frame ∂i = ∂u∂ i (i = 1, . . . , n − 1) on Mn−1 .
Remark. We can consider the case of n = 3 and two infinitesimal closed points q and p with the
spherical coordinate X a = (r, θ, ϕ) in M 3 and polar coordinate ui = (ρ, φ) in M2 . Therefore
Ea and ∂i are non-fixed frames.
The differential dV a can be given by

∂V a ∂V a
dV a = du + dv = ∂i V a dui , (342)
∂u ∂v
then all the n-dimensional vectors can be expanded by (n−1)-dimensional ones, the resulting equation
of (335) would be rewritten as
dV = (D̄V )a Ea
( )
= ∂i V a dui + V b Γa bc dX c Ea
( ∂X c i )
= ∂i V a dui + V b Γa bc i
du Ea
|∂u
{z }
hc i
( )
= ∂i V a + V b Γa bc hc i dui Ea := Via dui Ea . (343)
Here we define
( )
Vi := Via Ea = ∂i V a + V b Γa bc hc i Ea . (344)
Remark. The result of (66) in E3 can be reduced from (335) by



 M n −→ E3



 V a −→ pa = xa ,

Ea −→ δa , (345)



 Γa b −→ 0 ,


(D̄V )a −→ dxa .
44
Due to (334a) and (338), we have
0 = dδa = dAb̂ a eb̂ + Ab̂ a ω ĉ b̂ eĉ = (dAb̂ a + Aĉ a ω b̂ ĉ )eb̂ . (346)
As a consequence, the connection form ω b̂ ĉ is obtained by
ω b̂ ĉ = −Aĉ a dAb̂ a = +Ab̂ a dAĉ a . (347)
In addition, within (345), (343) becomes as

( ) ( )
dp = ∂i xa + xb Γa bc hc i dui δa = ∂i xa dui δa := ∂i pdui . (348)
|{z}
0
Therefore, (344) reduces to the derivative vector pi := ∂i p on the hypersurface M of E3 is
pi = ∂i p = ∂i xa δa = (∂i x, ∂i y, ∂i z) . (349)
Figure 11: Two vectors on the hypersurface Mn−1 in Mn .
Induced connection However, if we move the reference point o on the hypersurface Mn−1 , the
vector dV should be regarded as the difference between V ′ and V on Mn−1 and is equivalent to
DV with respect to the basis ∂u∂ i as shown in Fig. 11. Then, we have
( )
k ∂
dV = DV = D V
∂uk
( ) ∂
= ∂i V k + V l Γk li dui k . (350)
∂u
∂
By using chain rule to expand Ea in terms of ∂uk
∂ ∂uk ∂ ∂
Ea = a
= a k
= ha k k (351)
∂X |∂X
{z } ∂u ∂u
ha k
and substituting the relation into (343), we have

( )
( a b a c
) i k ∂
dV = ∂i V + V Γ bc h i du ha
∂uk
45
( ( a ) b
)
j ∂X k l ∂X a c k i ∂
= ∂i V j
ha + V l
Γ bc h i ha du
|∂u
{z } |∂u
{z } ∂uk
ha j hb
( (
l
))
j a k l a k b a c k ∂
= (∂i V ) h j ha +V (∂i h l )ha + h l Γ bc h i ha dui k
| {z } ∂u
δjk
( ( ))
k l a k b a c k ∂
= (∂i V ) + V (∂i h l )ha + h l Γ bc h i ha dui k . (352)
∂u
Now we compare (350) and (352), the induced connection Γk li on hypersurface Mn−1 can be obtained
from the connection Γa bc on M n through the projection ha k
Γk li = (∂i ha l )ha k + hb l Γa bc hc i ha k . (353)
We note that the discussion above can be generalized to the case for arbitrary frame (including fixed
and non-fixed frame).
Curvature and torsion in subspace Now we would like to consider an n-dimensional space M
embedded in a m-dimensional space M . We consider a so-called Darboux frame of M. We label
the components by indices â, b̂, ĉ = 1̂, . . . , m̂ on M the indices î, ĵ, k̂ = 1̂, . . . , n̂ on M, and the
indices p, q, r = n̂ + 1̂, . . . , m̂ on the normal space M⊥ of M in the orthonormal frame. We define
the geometric objects on M specified by barred symbols. The frame ēâ is extended by the vector eî
and ep̂
( )
( ) eî
ēâ = δâ eî + δâ ep̂ −→ ē := ēâ =
î p̂
. (354)
ep̂
Similarly, we have
( â ) ( )
â
ϑ̄ = δîâ î
ϑ + δp̂â p̂
ϑ −→ î
ϑ̄ := ϑ̄ = ϑ ϑ . p̂ (355)
Therefore, the components of {ēâ } and {ϑ̄â } are ēî = eî , ēp̂ = ep̂ , ϑ̄î = ϑî and ϑ̄p̂ = ϑp̂ . The metric
is defined by the inner product of two vectors, we have the following relations
ḡ(ēâ , ēb̂ ) = δâb̂ , g(eî , eĵ ) = δîĵ , g ⊥ (ep̂ , eq̂ ) = δp̂q̂ , ḡ(ēî , ēp̂ ) = 0 , (356)
where ḡ, g and g ⊥ are metrics of the manifolds M , M and M⊥ respectively. Due to the metric
compatible condition, we have ω b̂ â = −ω â b̂ and
ω b̂ â = δĵb̂ δâî ω ĵ î + δq̂b̂ δâî ω q̂ î + δĵb̂ δâp̂ ω ĵ p̂ + δq̂b̂ δâp̂ ω q̂ p̂ , (357)
which can also be recognized by the matrix as

( )
( b̂
) ω ĵ î ω q̂ î
ω := ω â = . (358)
ω ĵ p̂ ω q̂ p̂
Now we will discuss the dynamics on subspace M only. The differential of frame on M are given
by
{
d∇ p = ϑ̄â ēâ with ϑp̂ = 0 , (359a)
d∇ ēâ = ω̄ b̂ â ēb̂ . (359b)
46
or 

 d p = ϑî eî ,    î 

 ∇ d∇ p ϑ 0 ( ) (360a)
d∇ eî = ω ĵ î eĵ + ω q̂ î eq̂ , −→ d∇ eĵ  =  ĵ q̂  eî
 ω î ω î  . (360b)

 eq̂

d∇ ep̂ = ω ĵ p̂ e + ω q̂ p̂ eq̂ . d∇ ep̂ ĵ q̂
ω p̂ ω p̂
ĵ (360c)
In addition
ϑp̂ = 0 =⇒ ϑp̂ +ω p̂ î ∧ ϑî + ω p̂ q̂ ∧ |{z}
0 = Dϑp̂ = d |{z} ϑq̂ = T p̂ , (361)
0 0
Thus, we have
ω p̂ î ∧ ϑî = 0 . (362)
Applying the Cartan’s lemma, we obtain the connection form
ω p̂ î = −ω î p̂ = hp̂ îĵ ϑĵ , (363)

or
ωp̂î = hp̂îĵ ϑĵ and ωîp̂ = −ωp̂î = −hp̂îĵ ϑĵ = hîp̂ĵ ϑĵ (in pseudo-Riemannian geometry) .
(364)
Now we would like to calculate the differential of structure equations. The covarint exterior differ-
entiation of first structure equation is obtained by
d2∇ p = d∇ (ϑ̄â ēâ ) = d∇ (ϑî eî + ϑp̂ ep̂ )

= dϑî eî − ϑî ∧ d∇ eî + dϑp̂ ep̂ − ϑp̂ ∧ d∇ ep̂
( ) ( )
= dϑî eî − ϑî ∧ ω ĵ î eĵ + ω p̂ î ep̂ + dϑp̂ ep̂ − ϑp̂ ∧ ω î p eî + ω q̂ p̂ eq̂
= dϑi ei + ω ĵ î ∧ ϑî eĵ + ω p̂ î ∧ ϑî ep̂ + dϑp̂ ep̂ + ω î p̂ ∧ ϑp̂ eî + ω q̂ p̂ ∧ ϑp̂ eq̂
( ) ( )
= dϑî + ω î ĵ ∧ ϑî +ω î p̂ ∧ ϑp̂ eî + dϑp̂ + ω p̂ q̂ ∧ ϑq̂ +ω p̂ î ∧ ϑî ep̂
| {z } | {z }
T î T p̂
= T â ēâ = T î eî + T p̂ ep̂ . (365)

By using (361), we obtain
( )
d2∇ p = T î eî = T î + ω p̂ q̂ ∧ |{z}
ϑq̂ eî = T î eî , (366)
0
which leads to the equation
T î = T î . (367)
Remark. For case of M is embedded in M , we have consquence of
T î = T î . (368)
It means that there is no extrinsic torsion contribution in the equation of torsion in embedding
structure of geometry (cf. Gauss equation (372a)).
47
The covarint exterior differentiation of second structure equation is
d2∇ ēâ = Râ â ēâ = Rĵ â eĵ + Rp̂ â ep̂ , (369)

which can be calculated separately by d2∇ eî and d2∇ ep̂ . They are shown by
d2∇ eî = dω ĵ î eĵ − ω ĵ i ∧ d∇ eĵ + dω p̂ î ep̂ − ω p̂ î ∧ d∇ ep̂

( ) ( )
= dω ĵ î eĵ − ω ĵ î ∧ ω k̂ ĵ ek̂ + ω p̂ ĵ ep̂ + dω p̂ î ep̂ − ω p̂ î ∧ ω ĵ p̂ eĵ + ω q̂ p̂ eq̂
= dω ĵ î eĵ + ω k̂ ĵ ∧ ω ĵ î ek̂ + ω p̂ ĵ ∧ ω ĵ î ep̂ + dω p̂ î ep̂ + ω ĵ p̂ ∧ ω p̂ î eĵ + ω q̂ p̂ ∧ ω p̂ î eq̂
( ) ( )
= dω ĵ î + ω ĵ k̂ ∧ ω k̂ î +ω ĵ p̂ ∧ ω p̂ î eĵ + dω p̂ î + ω p̂ ĵ ∧ ω ĵ î + ω p̂ q̂ ∧ ω q̂ î ep̂
| {z }
Rĵ î
= Rĵ î eĵ + Rp̂ î ep̂ = Râ î ēâ (370)

and
d2∇ ep̂ = dω î p̂ eî − ω î p̂ ∧ d∇ eî + dω q̂ p̂ eq̂ − ω q̂ p̂ ∧ d∇ eq̂
( ) ( )
= dω î p̂ eî − ω î p̂ ∧ ω ĵ î eĵ + ω q̂ î eq̂ + dω q̂ p̂ eq̂ − ω q̂ p̂ ∧ ω î q̂ eî + ω r̂ q̂ er̂
= dω î p̂ eî + ω ĵ î ∧ ω î p̂ eĵ + ω q̂ î ∧ ω î p̂ eq̂ + dω q̂ p̂ eq̂ + ω î q̂ ∧ ω q̂ p̂ eî + ω r̂ q̂ ∧ ω q̂ p̂ er̂
( ) ( )
= dω î p̂ + ω î ĵ ∧ ω ĵ p̂ + ω î q̂ ∧ ω q̂ p̂ eî + dω q̂ p̂ + ω q̂ r̂ ∧ ω r̂ p̂ +ω q̂ î ∧ ω î p̂ eq̂
| {z }
Rq̂ p̂
= Rî p̂ eî + Rq̂ p̂ eq̂ = Râ p̂ ēâ , (371)

respectively. According to the results above, we have the following equations:


 Gauss equation: R î = R î + ω p̂ ∧ ω î ,
ĵ ĵ ĵ p̂
 (372a)
Codazzi equation: Rp̂ î = dω p̂ î + ω p̂ ĵ ∧ ω ĵ î + ω p̂ q̂ ∧ ω q̂ î , (372b)



Ricci equation: Rq̂ p̂ = Rq̂ p̂ + ω q̂ î ∧ ω î p̂ . (372c)
Subspace of Em We consider that a space M is embedded in the flat space Em . We can chose the
cartesian coordinate for Em , every component of the orthonormal frame {ēâ } is related to the fixed
cartesian frame by


 j j ∂
, and ϑ̄i = aî j dxj (i = 1, 2, . . . , n) ,
 ēî = aî δj = aî
∂xj
(373)


ēp̂ = ap̂ q δq = ap̂ q ∂ , and ϑ̄p̂ = ap̂ q dxq (p = n + 1, . . . , m) ,
∂xq
where aî j , aî j ∈ SO(m, R). Therefore, the differential of the frame on subspace M with ϑ̄p̂ = 0 is
given by
{
d∇ p = ϑî ⊗ eî = dxi ⊗ δi , (374a)
d∇ ēâ = ω b̂ â ⊗ ēb̂ = Γb a ⊗ δb , (374b)
and we can show that the torsion and curvature are vanished by using (374) in terms of cartesian
frame, which gives the following equations for frame with torsion-free and curvature-free on Em


 d∇ p = T eî = 0 ,
 2 î
(375a)
d2∇ eî = Râ î ēâ = 0 , (375b)


d2 e = Râ ē = 0 . (375c)
∇ p̂ p̂ â
48
The equations (367) and (372) turn out to be


 Torsion-free: T î = dϑî + ω î ĵ ∧ ϑĵ = 0 , (376a)



 Gauss equation: Rĵ = −ω ĵ ∧ ω p̂ = ω p̂ ∧ ω p̂ ,
î p̂ î ĵ î (376b)

 0 = dω î + ω ĵ ∧ ω î + ω q̂ ∧ ω q̂ î ,
p̂ p̂ ĵ p


Codazzi equation: (376c)


Ricci equation: Rq̂ p̂ = −ω q̂ î ∧ ω î p̂ = ω q̂ î ∧ ω p̂ î , (376d)
because all barred torsion and curvature 2-forms should be vanished in Em . According to (376a), we
have torsion-free, it leads us to have Ricci rotation coefficients written as (316). From the consequence
of (363), the Gauss equation (376b) is
1
Rĵ î = Rĵ îk̂l̂ ϑk̂ ∧ ϑl̂
2
1 ( p̂ p̂ )
= (hp̂ ĵ k̂ ϑk̂ ) ∧ (hp̂ îl̂ ϑl̂ ) = h ĵ k̂ h îl̂ − hp̂ ĵ l̂ hp̂ îk̂ ϑk̂ ∧ ϑl̂ , (377)
2
i.e.,
Rĵ îk̂l̂ = hp̂ ĵ k̂ hp̂ îl̂ − hp̂ ĵ l̂ hp̂ îk̂ , (378)
or
Rĵ îk̂l̂ = −hĵ p̂k̂ hp̂ îl̂ + hĵ p̂l̂ hp̂ îk̂ = hp̂ ĵ k̂ hp̂ îl̂ − hp̂ ĵ l̂ hp̂ îk̂ (in pseudo-Riemannian geometry) . (379)
The Codazzi equation (376c) becomes
0 = d(hp̂ îĵ ϑĵ ) + (hp̂ ĵ k̂ ϑk̂ ) ∧ ω ĵ î + ω p̂ q̂ ∧ (hq̂ îĵ ϑĵ )

= dhp̂ îĵ ∧ ϑĵ + hp̂ îĵ dϑĵ + hp̂ ĵ k̂ ϑk̂ ∧ ω ĵ î + hq̂ îĵ ω p̂ q̂ ∧ ϑĵ . (380)
The Ricci equation can be read as

1
Rq̂ p̂ = Rq̂ p̂k̂l̂ ϑk̂ ∧ ϑl̂
2
1 ( q̂ p̂ )
= (hq̂ îk̂ ϑk̂ ) ∧ (hp̂ îl̂ ϑl̂ ) = h îk̂ h îl̂ − hq̂ îl̂ hp̂ îk̂ ϑk̂ ∧ ϑl̂ , (381)
2
i.e.,
Rq̂ p̂k̂l̂ = hq̂ îk̂ hp̂ îl̂ − hq̂ îl̂ hp̂ îk̂ , (382)
or
Rq̂ p̂k̂l̂ = −hq̂ îk̂ hî p̂l̂ + hq̂ îl̂ hî p̂k̂ = hq̂ îk̂ hp̂ î l̂ − hq̂ îl̂ hp̂ î k̂ (in pseudo-Riemannian geometry) . (383)
Example (Hypersurface of E3 ). If we consider M and M to be M2 and E3 and ∂x∂ 3 is assume

to be aligned to the normal vector n of M, we have


 j j ∂
, and ϑ̄i = aî j dxj (i, j = 1, 2 and x1 = x, x2 = y) ,
 ēî = aî δj = aî j
∂x
(384a)

 ∂
ē3̂ = a3̂ δ3 = a3̂
3 3 3̂ 3̂ 3 p 3
, and ϑ̄ = a 3 dx (x = x = z) .
∂x3
49
Due to the fixed condition p = n + 1 = m = 3, it is impossible to have p ̸= q, which leads to the
results for hypersuface with ϑ̄3̂ = 0 of
Rq̂ p̂ = 0 (for hypersuface) , (385)
and
ω p̂ q̂ = 0 (for hypersuface) . (386)
We can identify h3̂ îĵ to be bîĵ which is the extrinsic curvature of M. The corresponding component
equations of (378) and (380) are

Rĵ = b b − b b , (387a)
îk̂l̂ ĵ k̂ îl̂ ĵ l̂ îk̂
 0 = dbîĵ ∧ ϑĵ + bîĵ dϑĵ + bĵ k̂ ω ĵ îl̂ ϑk̂ ∧ ϑl̂ . (387b)
If we use the holonomic frame with coordinate {ui } on M, we have
∂xj k ∂xj
ϑî = aî j dxj = aî j du = dui =⇒ eî k := aî j = δ î k (388)
∂uk ∂uk
such that dϑî = ddui = 0 (or cî ĵ k̂ = 0) and ω ĵ î = Γj i , therefore (387a) and (387b) becomes
Rj ikl = bjk bil − bjl bik (389)
and
( ) ( )
∂k bij duk ∧ duj + bjk Γj il duk ∧ dul = 0 =⇒ ∂k bij − ∂j bik + blk Γl ij − blj Γl ik = 0 ,
(390)
which have been given by (231) and (232) respectively.
Acknowledgments
Ling-Wei Luo (LWL) would like to thank Prof. Jen-Chi Lee (NCTU) who encurages LWL and gives
an opportunity for LWL to give a lecture on elementary differential geometry for both undergraduate
and dergraduate students in NCTU. LWL also would like to thans Mr. Shang-Yu Chien who is a PhD
student in National Tsing Hua University to check and discuss the contents of this lecture note. The
work was supported by Ministry of Science and Technology (MoST-107-2119-M-007-013-MY3) and
Academia Sinica Career Development Award Program (AS-CDA-105-M06).
References
[1] 小林昭七: 曲線と曲面の微分幾何, 裳華房, 1995.
[2] 小林昭七: 接続の微分幾何とゲージ理論, 裳華房, 1989.
[3] 陳維桓: 微分幾何, 北京大學出版社, 2006.
50
[4] Élie Cartan, On Manifolds with an Affine Connection and the Theory of General Relativity, Bib-
liopolis, Napoli, Italy, 1986.
[5] Élie Cartan, Riemannian Geometry in an Orthogonal Frame, World Scientific Publishing Co.,
Singapore, 2001.
[6] S. S. Chern, Weihuan Chen, K. S. Lam, Lectures on Differential Geometry, World Scientific
Publishing Co., Singapore, 1999.
[7] Manfredo P.do Carmo, Differential Forms and Applications, Springer, New York, 1998.
[8] Richard L. Faber, Differential Geometry and Relativity Theory: An Introduction, CRC Press,
Boca Raton, FL, 1983.
[9] Harley Flanders, Differential Forms with Applications to the Physical Sciences, Dover Publica-
tions, New York, 1989.
[10] Shoshichi Kobayashi, Katsumi Nomizu, Foundations of Differential Geometry, Vol. I, John Wi-
ley & Sons, New York, 1963.
[11] C.W. Misner, K.S. Thorne, J.A. Wheeler, Gravitation, Freeman, SanFrancisco, 1973.
[12] J. A. Schouten Ricci-Calculus: An Introduction to Tensor Analysis and Its Geometrical Appli-
cations, Springer, Berlin, 1954.
51

Lecture Note On Elementary Differential Geometry: Ling-Wei Luo

Uploaded by

Copyright:

Available Formats

Lecture Note On Elementary Differential Geometry: Ling-Wei Luo

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lecture Note On Elementary Differential Geometry: Ling-Wei Luo

Uploaded by

Copyright:

Available Formats

Lecture Note on Elementary Differential Geometry

July 20, 2019

4 Cartan’s moving frame and exterior differentiation methods . . . . . . . . . . . . . . 31

Tangent vector In 2-dimensional Euclidean

The arc length can be a function of parameter t given by

From the fundamental theorem of calculus, we have

According to the inverse

Normal vector Due to e1 · e1 = p′ · p′ = 1, we have

e′1 · e1 + e1 · e′1 = 0 =⇒ e′1 · e1 = 0 =⇒ e′1 ⊥ e1 , (8)

e′1 = κ(s)e2 . (10)

Remark. If a vector V is an unit vector, |V | = 1, the corresponding derivative vector would be

e′i · ej + ei · e′j = 0 (13a)

As a result, we have the following relations

called Frenet-Serret formula.

ṗ(t) = (−r sin t, r cos t) (15)

Therefore, the circumference is

By t = s/r, the circle p(s) and its tangent vector are

respectively. From (19b), we have

The curvature κ can be obtained by

Gauss map Gauss map G is a mapping which

By changing the parameter to s, we have to calculate ds/dt first:

Therefore, the tangent vector parametrized by s is obtained by

where ∧ is the exterior product or wedge product.

By orthonormality condition ei · ej = δij (i, j = 1, 2, 3), we have

e′i · ej + ei · e′j = 0 , (36)

Therefore, with the result (37a), we have to assume that

e′2 = −κ(s)e1 + τ (s)e3 . (38)

By comparing to (13d), it contains an additional term related to e3 . For i = 2, j = 3, we obtain

0 = e′2 · e3 + e2 · e′3 = (−κe1 + τ e3 ) · e3 + e2 · e′3 = τ + e2 · e′3 . (39)

Frenet-Serret formula in 3D As a result, we have Frenet-Serret formula:

Parametrization of a curve in a neighborhood of s0 One can do the Taylor expansion of p(s) at s0 .

The tangent vector and the corresponding norm are

The relation of s and t can be obtained by

we can calculate the torsion of p(s) form (40):

Tangent vector The differential of p is

are called velocity vectors along u and v respectively.

Remark. The vector p in E3 in the Cartesian coordinate system can be written as

where {δa } is a fixed reference frame of E3 . So that we have differential

dp = (dxa )δa + xa (dδa ) . (65)

Because δa is fixed, i.e. dδa = 0, it leads to the differential of p

dp = (dxa )δa = (dx, dy, dz) . (66)

The transformation of the basis and components are given by

First fundamental (quadratic) form we define

Remark. In the case of F = 0, the first fundamental form is

I = E dudu + G dvdv (71a)

In such case, we call (u, v) an isothermal coordinates if E = G. Therefore, the component of

gij = E δij with E = pi · pi = |pi |2 > 0 , (72)

( u and v)should be parametrized by one variable t,

We would alwalys write the first fundamental form with u = u1 and v = u2 as

I ≡ ds2 ≡ g = gij dui duj (i, j = 1, 2) , (79)

g ki gij = δjk . (81)

dp = dui ∂i = dui ⊗ ∂i := ϑ (pi −→ ∂i ) , (83)

ϑ(V ) = V k dui (∂k )∂i = V k δki ∂i = V i ∂i = V . (84)

It is apparent that ϑ is an identity map for a vector.

g(∂i , ∂j ) := gij . (85)

Therefore, for any two vectors V = V i ∂i and W = W j ∂j on M, we have

ḡ = ḡab dxa dxb = δab dxa dxb = ds2 (89)

ι∂i (duj ) = duj (∂i ) = δij . (96)