A brief introduction to characteristic classes

from the differentiable viewpoint

Yang Zhang
LEPP, Cornell University
April 24, 2011

Contents
1 Introduction 2

2 Characteristic classes via the curvature forms 2

2.1 Invariant polynomial . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Chern-Weil homomorphism . . . . . . . . . . . . . . . . . . . 6

3 Chern Classes 10
3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Chern Number . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Splitting principle . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.4 Chern character and Todd class . . . . . . . . . . . . . . . . . 17
3.5 Application of Chern classes . . . . . . . . . . . . . . . . . . . 20

4 Pontryagin classes and Euler classes 26

A Lie-algebra-valued differential forms 28

B Connection in principal bundles 30

B.1 Right action and the vertical space . . . . . . . . . . . . . . . 30
B.2 Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
B.3 Local connection form . . . . . . . . . . . . . . . . . . . . . . 32
B.4 Horizontal lift and the holonomy group . . . . . . . . . . . . . 34

1
 B.5 Curvature form . . . . . . . . . . . . . . . . . . . . . . . . . . 36
B.6 Ambrose-Singer Theorem . . . . . . . . . . . . . . . . . . . . . 39

C Connection on associated fibre bundles 39

C.1 Associated fibre bundles . . . . . . . . . . . . . . . . . . . . . 39
C.2 Connection on associated fibre bundle . . . . . . . . . . . . . . 40
C.3 Covariant derivative . . . . . . . . . . . . . . . . . . . . . . . 42
C.4 Local form of the connection . . . . . . . . . . . . . . . . . . . 43
C.5 Local curvature form . . . . . . . . . . . . . . . . . . . . . . . 45
C.6 Riemann structure, revisited . . . . . . . . . . . . . . . . . . . 47
C.7 Holomorphic vector bundle and Hermitian structure . . . . . . 49
C.8 Complex geometry . . . . . . . . . . . . . . . . . . . . . . . . 52

1 Introduction
For the concepts of the associated fibre bundles, connection, curvature and
the holomorphic vector bundles, please read the appendix.

2 Characteristic classes via the curvature forms

The local curvature form F on the base space for a fibre bundle should con-
tain the information how the bundle is twisted. For example, for a trivial
bundle, we can define a connection with the everywhere-vanishing local cur-
vature form. To compare two bundles over the same base space M , we may
try to simply compare the corresponding curvature forms F1 and F2 on X.
However, the difficult is

• F is not globally defined on X. F defined on the patch Ui will differ

from that defined on Uj by an adjoint action of the structure group G
(gauge transformation). So it is hard to compare F1 and F2 by the
local formula.

• For one fibre bundle, the compatible connection and thus the curvature
form is not unique. So F contains redundant information about the
fibre bundle.

So we will construct the invariant polynomial in terms of F which,

2
 • is invariant under the adjoint action so it is not needed to think of the
gauge transformation,

• can be reduced easily to get the connection-independent information of

the fibre bundle.

2.1 Invariant polynomial

Recall that the local curvature form F is a Lie-algebra-valued (or Lie-algebra-
representation-valued, for the vector bundle case) two-form. So before the
discuss of invariant polynomial in terms of F, we need to discuss the invariant
polynomial of matrices.

Definition 2.1. If G is a Lie group with the Lie group g, a G-invariant

r-linear symmetric function P is defined to be a map
O
P : g→F (1)
r

such that,

1. For c1 , c2 ∈ C and Ai ∈ g, 1 ≤ i ≤ r,

P (A1 , ...c1 Ai1 + c2 Ai2 , ..., Ar )

= c1 P (A1 , ..., Ai1 , ..., Ar ) + c2 P (A1 , ..., Ai2 , ..., Ar ). (2)

2. For 1 ≤ i, j ≤ r, P (A1 , ...Ai , ..., Aj , ..., Ar ) = P (A1 , ...Aj , ..., Ai , ..., Ar ).

3. For g ∈ G, P (Adg (A1 ), ..., Adg (Ar )) = P (A1 , ..., Ar ).

The set of all such P is denoted as I r (G).

Example 2.2. Let G = GL(m, C) and consider its fundamental representa-

tion. For r m × m matrices Ai , 1 ≤ i ≤ r, we define the symmetric trace
as
1 X
str(A1 , ..., Ar ) = tr(Aσ(1) ...Aσ(r) ) (3)
r! σ∈S r

where the sum is over all the permutations of (1, ...r). It is clear that str ∈
I r (G).

3
 We can combine all the I r (G) into a graded algebra I ∗ (G) = r I r (G),
L
where the product is defined to be: for P ∈ I p (G), Q ∈ I q (G), Ai ∈ g and
1 ≤ i ≤ (p + q),

(P · Q)(A1 , ...Ap+q )
1 X
= P (Aσ(1) , ..., Aσ(p) )Q(Aσ(p+1) , Aσ(p+q) ) (4)
(p + q)! p+q
σ∈S

where the sum is over all the permutations of (1, ...p + q).

Definition 2.3. A homogeneous invariant polynomial P with the degree r

is a map P : g → F, for which ∃P̃ ∈ I r (G)

P (A) = P̃ (A, ..., A) (5)

An invariant polynomial is the sum of finite homogeneous invariant polyno-

mials with different degrees.

Example 2.4. If G has a k-dimensional representation, we can define

iA
P (A) = det(I + t ), A∈g (6)
2π
where the determinant is over the k × k matrices. P is invariant under the
adjoint representation. We can expend P in c,

P (A) = 1 + tP1 (A) + ... + tk Pk (A) (7)

then each Pi (A) is also G-invariant and a degree-i homogeneous invariant

polynomial. So P is an invariant polynomial. It is clear that P1 (A) = 2πi tr(A)
and Pk (A) = det( 2πi A).
Conversely, from a homogeneous invariant polynomial P we can find a
P̃ ∈ I r (G) which induces P

1 
P̃ (A1 , ..., Ar ) = P (t1 A1 + ... + tr Ar ) (8)
r! t1 ...tr

P̃ is called the polarization of P . As the previous example, we can check

that the homogeneous invariant polynomial P (A) = tr(Ar ) has polarization
str(A1 , ..., Ar ).

4
 For P̃ ∈ I r G, We can extend its domain to r g-valued differential forms,
as
P̃ (A1 η1 , ...Ar ηr ) ≡ η1 ∧ ...ηr P̃ (A1 , ...Ar ), (9)
and its linear extensions, where Ai ∈ g and ηi ∈ Ωpi (M ). Note that the η’s
may have different degrees. It is clear that,
P̃ (Adg Ω − 1, ...Adg Ωr ) = P̃ (Ω − 1, ...Ωr ) (10)
for g ∈ G and each Ωi is a g-valued pi -form.
Similarly, for a homogeneous invariant polynomial P with the degree r,
^

P (Aη) = η P (A). (11)
r

and its generalization for the non-homogeneous case is clear.

Consider the infinitesimal adjoint action Adexp tX , t → 0, we have,
Proposition 2.5. for X ∈ g, Ai ∈ g and P̃ ∈ I r (G),
r
X
P̃ (A1 , ..., [X, Ai ], ..., Ar ) = 0. (12)
i=1

Furthermore, let A be a g-valued p-form and each Ωi be a g-valued pi -form.

r
X
(−1)p(p1 +...+pi−1 ) P̃ (Ω1 , ..., [A, Ωi ], ..., Ωr ) = 0. (13)
i=1

Proof. The Lie algebra case (12) is self-evident. For (13), by the linearity of
P̃ , we just need to prove the case A = Xη, where X ∈ g and η is a p-form.
By the definition (141), we have
[X, Ωi ] = η ∧ [X, Ωi ], (14)
where [X, Ωi ] means the adX action on the Lie algebra component of Ωi . So,
(−1)p(p1 +...+pi−1 ) P̃ (Ω1 , ..., [A, Ωi ], ..., Ωr ) = η ∧ P̃ (Ω1 , ..., [X, Ωi ], ..., Ωr ). (15)
Therefore (13) holds because of the infinitesimal case of (10).
Proposition 2.6. Let P̃ ∈ I r (G) and each Ωi be a g-valued pi -form,
r
X
dP̃ (Ω1 , ...Ωr ) = (−1)p1 +...+pi−1 P̃ (Ω1 , ...dΩi , ...Ωr ). (16)
i=1

5
2.2 Chern-Weil homomorphism
Let F be a local curvature form. On the intersection of two patches Ui ∩ Uj ,

Fj = Adt−1
ij
Fi = t−1
ij Fi tij (17)

where the tij is the group element in G for the principal bundle case (183) or
its V -representation for the vector bundle case (212). So F is not globally
defined.
However, if P is a invariant polynomial, then P (F) is globally defined.

Theorem 2.7 (Chern-Weil). P (F) has the following good properties,

1. dP (F)=0.

2. If F and F 0 are the curvature forms of two connections of a fibre bundle

respectively, then P (F1 ) − P (F2 ) is exact.

Proof. 1. It is sufficient to prove that case when P is a homogeneous

invariant polynomial with the degree r. Let P̃ be the polarization of
P,

dP (F) = dP̃ (F, ..., F)

Xr
= P̃ (F, ...dF, ...F)
i=1
r
X
= P̃ (F, ...dF, ...F) + P (F, ...[A, F], ...F)
i=1
Xr
= P (F, ...DF, ...F) = 0. (18)
i=1

Because by the Bianchi identity DF = dF + [A, F] = 0, where A is

the local connection one-form. Here we used the propositions (13) and
(16).

2. Let {(Ui , φi )} be the local trivialization of the fibre bundle. Then on

each Ui , we have the connection one-forms Ai and A0i from the two
connections. We can define

Ait ≡ A + tθ, θ = A0 − A, 0≤t≤1 (19)

6
On the intersection Ui ∩ Uj , A and A0 satisfy the same transition law,
Aj = Adt−1
ij
Ai + t−1
ij dtij , A0j = Adt−1
ij
A0i + t−1
ij dtij (20)

So Ait and Ajt are compatible and we have a connection on the fibre
bundle for each 0 ≤ t ≤ 1. For the following calculate, we simply omit
the subscript for patches. The curvature form for At is,
Ft = dAt + At ∧ At = F + tDθ + t2 θ ∧ θ (21)
where Dθ = dθ + [A, θ] = dθ + A ∧ θ + θ ∧ A. Hence,
Z 1
0 d
P (F ) − P (F) = dt P (Ft )
0 dt
Z 1  
d
= r dtP̃ Ft , Ft ..., Ft
0 dt
Z 1     
= r dt P̃ Dθ, Ft ..., Ft + 2tP̃ θ ∧ θ, Ft ..., Ft
0
(22)
On the other hand,
dP̃ (θ, Ft , ..., Ft ) = P̃ (dθ, Ft , ..., Ft ) − (r − 1)P̃ (θ, dFt , ..., Ft )
= P̃ (Dθ, Ft , ..., Ft ) − (r − 1)P̃ (θ, DFt , ..., Ft )
(23)
where we used the proposition (13). Dt F is nonzero since D is defined
by the connection form A not At . So,
DFt = Dt Ft − [tθ, Ft ] = −t[θ, Ft ]. (24)
By using the proposition (13) again,
dP̃ (θ, Ft , ..., Ft ) = P̃ (Dθ, Ft , ..., Ft ) + (r − 1)tP̃ (θ, [θ, Ft ], ..., Ft )
= P̃ (Dθ, Ft , ..., Ft ) + 2tP̃ (θ ∧ θ, Ft , ..., Ft ) (25)
Hence,
Z 1
0
P (F ) − P (F) = r dtdP̃ (θ, Ft , ..., Ft )
0
Z 1 
= rd dtP̃ (θ, Ft , ..., Ft ) . (26)
0

7
 Note that θ transform between different patches by the adjoint action
of G.

Definition 2.8. For two connections on the same fibre bundle and an ho-
mogeneous invariant polynomial Pr , we define the transgression as,
Z 1
T Pr = r dtP̃ (A0 − A, Ft , ..., Ft ). (27)
0

T Pr is a globally defined (2r − 1)-form on M .

Corollary 2.9. Let the basis space M be a 2m-dimensional orientable com-
pact real manifold without boundary, Pm is a degree-m invariant polynomial,
then Z
Pm (F) (28)
M
is independent of the connection choice of the fibre bundle.
Proof. M Pm (F 0 ) − M Pm (F) = M d(T Pm ) = 0.
R R R

Example 2.10. Let M be a two-dimensional compact real surface without

boundary. If g is the Riemann metric of M, then we can locally choose the
orthonormal frame {e1 , e2 }. Then local connection one-form is a so(2)-valued
one form,  


0 A1 2
De1 De2 = e1 e2 (29)
A2 1 0
where A1 2 is a local one-form and A2 1 = −A1 2 . Let K = dA1 2 , then
curvature matrix is  
0 K
(30)
−K 0
From differential geometry, we know that K is the Gaussian curvature mul-
tiplied by the volume form, so K depends on the metric. However, because
G = SO(2) is an abelian group, an element of the curvature matrix is G-
invariant linear function. So by Chern-Weil theorem,
Z
K (31)
m

is independent of the metric choice. This is part of the classic Gauss-Bonnet

theorem.

8
 It is not easy to see this result if we do not use the orthonormal frame,
because in general the curvature form is gl(2, R)-valued local curvature form.
But in two-dimensional case, the only obvious invariant polynomials, tr(F) =
ρ
Rρµν = 0 and det(F) = 0 everywhere on M .
So given a fibre bundle E and invariant polynomial P , we can define a
de Rham class χE (P ) ≡ [P (F)] ∈ H ∗ (M ). Chern-Weil theorem ensures that
the change of the connection does not change the de Rham class. χE (P ) is
called a characteristic class.

Theorem 2.11. 1. χE : I ∗ (G) → H ∗ (G) is a ring homomorphism. (Weil

Homomorphism)

2. Let f : N → M be a differentiable map and f ∗ E be the pullback fibre

bundle of E. χE has the naturality property,

χf ∗ E (P ) = f ∗ χ(E) (32)

Proof. 1. It is clear that χE (P̃1 + P̃2 ) = χE (P̃1 ) + χE (P̃2 ) and χE (1) =

[1] ∈ H 0 (M ), where [1] is the class for the constant function f (p) ≡ 1,
∀p ∈ M . For the product, it is sufficient to prove the case when P̃r ∈
I r (G) and P̃s ∈ I s (G). Let F = F α Tα .

(P̃r P̃s )(F, ..., F) = F α1 ∧ ... ∧ F αr ∧ F αr+1 ∧ ... ∧ F αr+s ×

(P̃r P̃s )(Tα1 , ..., Tαr , Tαr+1 , ..., Tαr+s )
= F α1 ∧ ... ∧ F αr ∧ F αr+1 ∧ ... ∧ F αr+s ×
1 X
P̃r (Tασ(1) , ..., Tασ(r) )P̃s (Tασ(r+1) , ..., Tασ(r+s) )
(r + s)! σ∈S
r+s

1 X
= F ασ(1) ∧ ... ∧ F ασ(r) ∧ F ασ(r+1) ∧ ... ∧ F ασ(r+s) ×
(r + s)! σ∈S
r+s

P̃r (Tασ(1) , ..., Tασ(r) )P̃s (Tασ(r+1) , ..., Tασ(r+s) ) = Ps (F) ∧ Pr (F)

where we used the fact that all F α commute with each other.

2. If on a local patch Ui , E has the local connection form Fi , then on

f −1 (Ui ) , the curvature form is f ∗ Ωi .

9
Corollary 2.12. For a trivial fibre bundle E, χE maps all the invariant
polynomials to zero so all its characteristic classes are trivial.

Proof. For a trivial fibre bundle, no matter it is a principal bundle or not, we

can always choose a connection whose curvature form vanishes everywhere.
Then χE (P ) = [P (0)] = 0 ∈ H ∗ (M ).

3 Chern Classes
3.1 Definition
From the example (2.4), for A ∈ GL(k, C),
  X k
iA
det I + t = tr Pr (A) (33)
2π r=1

defined k degree-r invariant polynomials, Pr .

Definition 3.1. Let π : E → M be a complex vector bundle whose fibre is

Ck . Define the j-th Chern Class to be

cj (E) = Pj (F) ∈ H 2j (M ) (34)

and the total Chern class as

c(E) = c0 (E) + ... + ck (E) ∈ H ∗ (M ) (35)

Proposition 3.2. If a invariant polynomial Pr ∈ I r (GL(k, C)), can be writ-

ten as a polynomial of matrix elements,

Pr (A) = cα1 β1 ...αr βr Aα1 β1 ...Aαr βr , ∀A ∈ GL(k, C) (36)

then for a gL(k, C)-valued two-form F, P (F) equals the same polynomial in
two-form matrix elements,

Pr (F) = cα1 β1 ...αr βr F α1 β1 ∧ ... ∧ F αr βr (37)

10
Proof. Define the matrix Eαβ as the matrix with the element at (α, β) to be
1, and all the other elements vanished.

Pr (F) = Pr (F α β Eαβ ) = P̃r (F α1 β1 Eα1 β1 , ..., F αr βr Eαr βr )

= F α1 β1 ∧ ... ∧ F αr βr P̃r (Eα1 β1 , ..., Eαr βr )

α1 αr 1 
= F β1 ∧ ... ∧ F βr Pr (t1 Eα1 β1 + ... + tr Eαr βr )
r! t1 ...tr
1
= F α1 β1 ∧ ... ∧ F αr βr (cα1 β1 ...αr βr + permutations)
r!
= cα1 β1 ...αr βr F α1 β1 ∧ ... ∧ F αr βr (38)

where in the last line we used the fact that all the F α β commute.
By this proposition, Pr (F) can be realized as the operators on the two-
form-valued matrix. For example,
i i α
P1 (F) = tr(F) = F α, (39)
2π 2π
which is proportional to the Ricci form. And
 
iF
Pk (F) = det (40)
2π

where the multiplication between elements is the wedge product. Further-

more the total Chern class,
 
i
c(E) = [det I + F ], (41)
2π

and again, here the multiplication is the wedge product (or the product
between a number and a form).
By the definition of de Rham cohomology,

cj (E) = 0, if 2j > dimR M (42)

Proposition 3.3. If the complex vector bundle E is equipped with a Hermi-

tian metric h and its connection is Hermitian-compatible, then we can choose
a real representative for each Chern class.

11
Proof. From (247), if the Hermitian metric is h = (hαβ̄ ),

F = −h−1 F T h. (43)

Hence,
     
it it it −1 T
det I + F = det I − F = det I + h F h (44)
2π 2π 2π
 
it
= det I + F . (45)
2π
So the Chern classes are real forms.
Example 3.4. Let π : E → M be a complex line bundle, i.e. the fibre is
complex one-dimensional. Then the total Chern class is

c(E) = 1 + [F], (46)

where the curvature F is a real two-form.

Remark 3.5. Formally, we can use our “effective field theory” trick to calcu-
late the Chern classes. For A ∈ GL(m, C),
    
i i

det I + A = exp log det I + A
2π 2π
  
i

= exp tr log I + A
2π
 X ∞ 
−i
n n
= exp − tr(A )
n=1
2π
i 1
= 1+ tr(A) + 2 (tr(A2 ) − tr(A)2 ) + ..., (47)
2π 8π
so it is generated by tr(Ar ). 1 Then by the proposition (3.2), the same formal
works if A is replaced by F.
Theorem 3.6. Let πE : E → M be a complex vector bundle.
1
This series can be calculated by Mathematica,
Series [ Exp[−Sum[( −( I / ( 2 Pi ) ) ) ˆ i t ˆ i Array [ t r , 1 0 ] [ [ i ] ] / i , { i , 1 , n } ] ] , { t , 0 , n } ]
with n as the maximal degree of the invariant polynomial.

12
 • (Naturality) Let f : N → M be a smooth map. Then

c(f ∗ E) = f ∗ c(E) (48)

• Let πF F → m be another complex vector bundle. Then the total Chern

class for the Whitney sum bundle E ⊕ F is,

c(E ⊕ F ) = c(E) ∧ c(F ) (49)

Proof. • It directly follows from the theorem (2.11).

• By the structure of the Whitney sum2 , if E has the local trivialization

{Ui , φi } and F has the local trivialization {Vl , ψl }, we can choose the
local trivialization for E ⊕ F as,

{Ui ∩ Vl , φi ⊕ ψl }, (51)

as long as Ui ∩Vl 6= ∅. If on Ui , the connection of E reads ∇sα = sβ Aβ α ,

and similarly on Vl , F ’s connection reads ∇σa = σb Ab a , then the local
connection form of E ⊕ F on Ui ∩ Vl for the sections {sα , σb } is,
 α 
A β 0
(52)
0 Ab a

Hence, the local curvature form for E ⊕ F is,

 α 
F β 0
(53)
0 F ba

therefore direct calculation shows that c(E ⊕ F ) = c(E) ∧ c(F ) via

algebraic topology.

2
Formally, the Whitney sum is defined as the follows,
E × F

πE πF (50)
i
∨ ∨
X >X × X
E × F is a complex vector bundle over X × X. i is the diagonal map p → (p, p). Then
E ⊕ F is defined to be the pullback i∗ (E × F ).

13
3.2 Chern Number
Chern classes were originally defined to be elements in H ∗ (M, Z) via algebraic
topology. Chern proved it is equivalent to Chern classes defined by curvature
form, i.e. the j-th Chern classes cj (E) ∈ H ∗ (M, C) from the curvature form,
determined c0j (E) ∈ H ∗ (M, Z) from algebraic topology, via
Z
0
cj (E)(σ) ≡ cj (E) (54)
σ

for any 2j-dimensional singular cycle σ in M with integer coefficients. It

can be checked that, for the universal bundle the value of this integral is an
integer. Then by the naturality of Chern classes,
Z Z
σ f
∆2j > M > BG, cj (E) = cj (EG) ∈ Z (55)
σ f ·σ

so cj (E)0 ∈ Hom((H2j M, Z), Z). Z is PID, so H 2j (M, Z) = Hom(H2j (M, Z), Z)

and c0j (E) ∈ H ∗ (M, Z).
This equivalence implies that,

Proposition 3.7. For any singular cycle σ in M with integer coefficients,

the integral Z
cj1 (E) ∧ ... ∧ cjl (E) ∈ Z, (56)
σ
where if dim σ 6= j1 + .. + jl , the integral is simply set to zero.

The integers obtained from all elements in M ’s singular homology are

called Chern Numbers.

3.3 Splitting principle

From (49), if a complex vector bundle pi : E → M is the Whitney sum of
several complex line bundles,

E = L1 ⊕ ... ⊕ Ln (57)

then n
Y
c(E) = c(L1 ) ∧ ... ∧ c(Ln ) = (1 + xi ) (58)
i=1

14
  
iF
where xi is the first Chern class for Li . In this case, the matrix I + 2π
is
diagonalized.  
iF
However, in general I + 2π
cannot be diagonalized or decomposed
into the Jordan form, because it is not a complex-number matrix. Also E
cannot be decomposed as the Whitney sum of n complex line bundles either.
However, we have a strong claim that [5],

(The Splitting Principle) To prove a polynomial identity in the

Chern classes of complex vector bundles, it suffices to prove it
under the assumption that the vector bundles are the Whitney
sum of complex line bundles.

To see how to realize this principle, we need the following theorem,

Theorem 3.8. Let π : E → M be a complex vector bundle of rank n. There

exists a manifold F l(E), called the flag manifold associated with E and a
smooth map ξ : F l → M such that,

• the pullback of E to F l(E) splits into the Whitney sum of complex line
bundles,
ξ −1 (E) = L1 ⊕ ... ⊕ Ln (59)

• ξ ∗ : H ∗ (M ) → H ∗ (F L(E)) is injective.

Proof. See [5], section 21.

Corollary 3.9. In H ∗ (F l(E)), by the naturality of Chern classes,

n
Y n
Y
c(Li ) = (1 + c1 (Li )) = ξ ∗ (c(E)). (60)
i=1 i=1

More generally, suppose that there are several complex vector bundles E1 ,
.... Er over M . We can first introduce ξ1 : N1 → M which satisfies theorem
(3.8), and ξ1−1 E1 is splitting on N1 . Then we can introduce ξ2 : N2 → M by
theorem (3.8) and on N2 , ξ2−1 ξ1−1 E2 is splitting. Note that ξ2−1 ξ1−1 E1 is still
splitting. Repeat this process and finally we get an ξ : N → M , such that
all ξ −1 Ei are splitting and ξ ∗ is injective.

15
 Now we see how this theorem implies that the splitting principle. For
example, if we want to prove that for some polynomial P , complex vector
bundles E and F over M ,

P (c(E), c(F ), c(E ⊗ F )) = 0 (61)

Choose ξ : N → M such that both ξ −1 E and ξ F are splitting and ξ ∗ is

injective. The injective property means that it is sufficient to prove,

ξ ∗ P (c(E), c(F ), c(E ⊗ F )) = 0. (62)

by the naturality of Chern classes, we only need to prove on N ,

P (c(ξ −1 E), c(ξ −1 F ), c(ξ −1 (E) ⊗ ξ −1 (E)) = 0 (63)

Now all the vector bundles are splitting, so it suffices to prove the splitting
case only.
Proposition 3.10. Let P (x1 , ..., xn ) be a symmetric polynomial in (x1 , ..., xn ),
• there exists a unique element wP (E) ∈ H ∗ (M ), such that P (c1 (L1 ), ..., c1 (Ln )) =
ξ ∗ (w).

• (Naturality). Let f : N → M be a smooth map. Then wP (f −1 E) =

f ∗ (wP (E)).

Proof. • A symmetric polynomial is a polynomials in the elementary

symmetric polynomials Sj ,

P (x1 , ..., xn ) = Q(S1 (x1 , ..., xn ), ..., Sn (x1 , ..., xn )) (64)

Because ni=1 (1 + c1 (Li )) = ξ ∗ (c(E)),

Q

sj (c1 (Li ), ...cn (Li )) = ξ ∗ (cj (E)). (65)

Therefore,

P (c1 (Li ), ...cn (Li )) = Q(ξ ∗ (c1 (E)), ..., ξ ∗ (cn (E)))
= ξ ∗ Q(c1 (E), ...cn (E)) (66)

Now we define wp (E) = Q(c1 (E), ...cn (E)). The uniqueness comes from
the injective map ξ ∗ .

16
 • By the naturality of Chern classes,

f ∗ wp (E) = f ∗ Q(c1 (E), ...cn (E)) = Q(f ∗ c1 (E), ..., f ∗ c1 (E))

= Q(c1 (f −1 E), ..., c1 (f −1 E)) = wp (f −1 E) (67)

Here we used the uniqueness of wp (f −1 E).

This proposition provides a simple way to construct characteristic classes

in terms of Chern classes. In future, by abusing notations, we will denote
such w ∈ H ∗ (M ) defined by P , as wP (E) ≡ P (x1 , ..., xn ), where xi means
the first Chern class of i-th complex line bundle on F l(M ).

3.4 Chern character and Todd class

We can use Chern class to construct different characteristic classes,

Definition 3.11. The total Chern character is defined by

 
iF X 1  iF j
ch(E) ≡ [tr exp ]= [tr ] (68)
2π j=1
j! 2π

and the j − th Chern character is,

 j
1 iF
chj (E) = [tr ] (69)
j! 2π

Note that the exp series will truncate when 2j > dimR M , so ch(E) is a
polynomial of F in de Rham cohomology.

Theorem 3.12. Chern characters have the following properties,

1. (Naturality) For a smooth map f : N → M , ch(f −1 M ) = f ∗ ch(M ).

2. ch(F ⊕ E) = ch(F ) + ch(E).

3. ch(F ⊗ E) = ch(F ) ∧ ch(E).

Proof. 1. The proof is exactly the same as that for the Chern classes.

17
 2. We choose the local curvature form for F ⊕ E to be
 
FF 0
F= (70)
0 FE
so
exp( iF
 
iF 2π
F
) 0
tr exp( ) = tr = ch(F ) + ch(E) (71)
2π 0 exp(FE )

3. Locally, if that we have the frame {sα } for F and {σa } for E, then

∇F ∇F sα = sβ FFβ α , ∇E ∇E σa = σa FFa b . (72)

For the local frame {sα ⊗ σa }, we have3

∇(sα ⊗ σa ) = (∇F sα ) ⊗ σa + sα ⊗ (∇E σa ) (74)

∇∇(sα ⊗ σa ) = (sβ FFβ α ) ⊗ σa + sα ⊗ (sb FEb a ) (75)

The local curvature form for F ⊗ E is the Kronecker product FF ⊗ I +

I ⊗ FE . Note that FF ⊗ I and I ⊗ FE commute, so
 
i i
ch(E ⊗ F ) = tr exp( FF ⊗ I) exp( I ⊗ FE )
2π 2π
 
i i
= tr exp( FF ) ⊗ exp( FE )
2π 2π
   
i i
= tr exp( FF ) ∧ tr exp( FE )
2π 2π
= ch(E) ∧ ch(F ) (76)

Hence Chern character is a homomorphism from the ring of complex

vector bundles to H ∗ (M ).
3
We can extend the covariant derivative for the vector-valued forms as

∇(s ⊗ σ) = (∇F s) ⊗ σ + (−1)pq s ⊗ (∇E σ) (73)

for local sections s ∈ Γ(Ui , E) ⊗ Ωp (Ui ) and σ ∈ Γ(Ui , F ) ⊗ Ωq (Ui )

18
 Recall that we can define the flag manifold of E and a map ξ : F l(E) →
M , such that ξ −1 E = L1 ⊕ ... ⊕ Ln . By the naturality of the Chern classes
n
∗
Y i
ξ (c(E)) = (1 + xi ), xi ≡ [ Fi ] = c1 (Li ), i = 1, ..., n (77)
i=1
2π
∗
and ξ (cj (E)) is the j-th elementary symmetric polynomial in x1 , ..., xn .
For the Chern characters, again by the naturality,
n n
X i X
ξ ∗ (ch(E)) = [exp( Fi )] = exi (78)
i=1
2π i=1

Comparing (77) and (78), because ξ ∗ is injective, we have,

ch0 (E) = n, c0 (E) = 1
ch1 (E) = c1 (E)
1
ch2 (E) = c1 (E)2 − c2 (E)
2
... (79)
where by Newton’s identities, we can rewrite all the Chern characters in terms
of Chern classes.
Remark 3.13. We can prove the identity ch(F ⊗ E) = ch(F ) ∧ ch(E) again
by the splitting principle. Let ξ : N → M be the flag manifold such that
ξ −1 F = L1 ⊕ ... ⊕ Ln and ξ −1 F = l1 ⊕ ... ⊕ lk . We have
Xn X
k
ξ ∗ ch(F ⊗ E) = ch((ξ −1 F ) ⊗ (ξ −1 E)) = ch( Li ⊗ lk )
i=1 j=1
n X
X k
= ch(Li ⊗ lk ) (80)
i=1 j=1

It is clear that for complex line bundles, FLi ⊗lk = FLi + Flk , so
ch(Li ⊗ lk ) = ch(Li ) ∧ ch(lk ) (81)
Then,
n
X k
X
∗
ξ ch(F ⊗ E) = ch(Li ) ch(lk ) (82)
i=1 j=1
= ξ ch(F ) ∧ ξ ch(E) = ξ ∗ (ch(F ) ∧ ch(E))
∗ ∗
(83)

19
Because ξ ∗ is injective, this proved that ch(F ⊗ E) = ch(F ) ∧ ch(E). In
future, when we use the splitting principle, we can omit the notation ξ ∗ and
simply assume that each complex vector bundle is splitting.
Definition 3.14. For a complex vector bundle π : E → M , we define the
Todd class to be the unique element T d(E) in H ∗ (M ) such that,
Y xj
ξ ∗ (T d(E)) = (84)
j
1 − exj

where again ξ : N → M is the flag manifold and xi = c1 (Li )

Because the product above is symmetry in all the xj ’s. By the proposi-
tion (3.10), T d(E) is well-defined and natural. Todd class has the following
expansion,
Y 1

k−1 Bk
X
∗ 2k
ξ (T d(E)) = 1 + xj + (−1) x , (85)
j
2 k≥1
(2k)! j

where the Bk ’s are the Bernoulli numbers,

1 1 1 1 5
B1 = , B2 = , B3 = , B4 = , B5 = (86)
6 30 42 30 66
Theorem 3.15. T d(E ⊕ F ) = T d(E) ∧ T d(F ).
Proof. We use the splitting principle again. Here we omit ξ ∗ and simply put
E = L1 ⊕ ... ⊕ Ln and F = l1 ⊕ ... ⊕ lk . Let c1 (Li ) = xi and c1 (lj ) = xn+j

T d(E ⊕ F ) = T d(L1 ⊕ ... ⊕ Ln ⊕ l1 ⊕ ... ⊕ lk )

n+k
Y xi  n+k
Y xi   n+k 
Y xi
= xi
= xi
∧
i=1
1 − e i=1
1 − e i=n+1
1 − exi
= T d(E) ∧ T d(F ). (87)

3.5 Application of Chern classes

Example 3.16 (line bundles on CP 1 ). Here we show that several line bundles
on CP 1 are different. For line bundles, we just need to consider the first

20
Chern class c1 . CP 1 is defined to be the one-dimensional complex projective
space,
{C2 − (0, 0)}/{(z0 , z1 ) ∼ s(z0 , z1 ), s ∈ C∗ } (88)
The local coordinates patches for CP 1 are,
U0 = {C2 − (0, 0)|z0 =
6 0} , z ≡ z1 /z0
2
U1 = {C − (0, 0)|z1 = 6 0} , w ≡ z0 /z1 . (89)
On U0 ∩ U1 , the coordinates transit as w = 1/z. CP 1 has the S 2 topology,
we denote x to be the element in H 2 (CP 1 ), such that
Z
x = 1. (90)
CP 1

1. The trivial bundle CP 1 × C1 . In this case, c1 = 0.

2. The holomorphic tangle bundle (T CP 1 )+ . We know that (T CP 1 )+
allows the Fubini-Study metric, on U0 ,
1
dz ⊗ dz̄. (91)
(1 + |z|2 )2
On the intersection U0 ∩ U1 , the metric equals
   
1 1 1 1
d ⊗ d = dw ⊗ dw̄ (92)
(1 + |1/w|2 )2 w w (1 + |w|2 )2
which is non-singular at w = 0. So the metric is globally defined. The
curvature form on U0 is,
 
¯ 1 dz ∧ dz̄
F = −∂ ∂ log 2 2
=2 (93)
(1 + |z| ) (1 + |z|2 )2
So the first Chern class is,
 
dz ∧ dz̄
c1 = i . (94)
π(1 + |z|2 )2
The integral over CP 1 gives,
Z
c1 = 2, (95)
CP 1

so c1 = 2x.

21
 3. The holomorphic cotangent bundle (T CP 1 )(1,0) . The metric for the
cotangent bundle is just inverse of the metric for the tangent bundle.
So repeat the calculation, we get c1 = −2x.

4. The canonical line bundle γ 1 , which is defined to be

γ 1 = {[z0 , z1 ] ⊗ (u0 , u1 )|[z0 , z1 ] ∈ CP 1 , (u0 , u1 ) ∈ C2 , [u0 , u1 ] = [z0 , z1 ]}

(96)
1 1
and the bundle projection π : γ → CP is,

π([z0 , z1 ] ⊗ (u0 , u1 )) = [z0 , z1 ]. (97)

We have the following trivialization φ0 : U0 × C → π −1 (U0 ),

φ0 (z, c) = [1, z] ⊗ (c, cz) (98)

and φ0 : U0 × C → π −1 (U0 ),

φ0 (w, c0 ) = [w, 1] ⊗ (c0 w, c0 ) (99)

The transition relation is c0 = zc. We define the Hermitian metric on

π −1 (U0 ) as,
(1 + |z|2 )c1 c∗2 (100)
while on π −1 (U0 , U1 ) the metric reads,

(1 + |z|2 )c1 c∗2 = (1 + |z|2 )c01 c0∗ 2 2 0 0∗

2 /|z| = (1 + |w| )c1 c2 (101)

which is nonsingular at w = 0. So this metric is globally defined.

Repeat the calculation, we have c1 = −x.

Hence all the line bundles have different first Chern classes and so they are
different line bundles.
Example 3.17 (Dirac Monopole). Dirac Monopole is a point-like magnetic
charge of the U (1) gauge theory. Let the monopole has charge g, then U (1)
field strength is
gm p
Fij = ijk 3 xk , r = x2 + y 2 + z 2 . (102)
r

22
We can think that the Dirac monopole configuration is a complex line bun-
dle E over R3 − (0, 0, 0). By the convention of Lie-algebra in physics, the
curvature form 4 is F = ieF and,
eg k i
c1 (E) = −ijk x dx ∧ dxj . (103)
4πr3
Over a non-trivial singular cycle, S 2 : r2 = 1, we have
Z
c1 (E) = −2eg (104)
S2

By the proposition (3.7), the integral of a Chern class over a cycle must be
an integer, otherwise it is not a well-defined bundle. So

2eg ∈ Z (105)

which is the Dirac quantization condition.

Example 3.18 (Instanton in 4D). Consider the SU (2) Yang-Mills Theory in
Euclidean 4D spacetime R4 . We try to find field configurations corresponding
to non-trivial fibre bundles. However, because R4 is topologically trivial, it
seems that no trivial fibre bundle can exist.
The point is that we are only interested in finite action configuration,
Z
1
SE = 2 d4 xtr(Fµν F µν ) < ∞ (106)
2g
which implies that,
1
Fµν ∼ o(
), r → ∞. (107)
r2
So we may extend the fibre bundle to be over S 4 . The instanton configura-
tions correspond to non-trivial gauge bundles on S 4 .
Let SU (2) act on C2 . We use the Euclidean convention such that F = F ,
which anti-Hermitian-matrix valued. The first Chern classes vanished and
the second one,
 
iF 1
c2 (E) = det = 2 tr(F ∧ F ), (108)
2π 8π
4
Note that in physics, we use Dµ = ∂µ + ieAµ instead of Dµ = ∂µ + Aµ .

23
where we used the relation between Chern classes and Chern characters.
Again, By the proposition (3.7),
Z Z
1 1
2
tr(F ∧ F ) = 2
tr(F ∧ F ) ∈ Z (109)
S 4 8π R4 8π

which is the instanton twisting number. Note that c2 (E) is a closed form
in S 4 but may not be exact. In R4 , we can find a three-form K, such that
dK = c2 (E) but K cannot be extend to the S 4 .
Here we explicit construct the instanton solution, for the R4 − (0, 0, 0, 0)
patch of S 4 , we can define a map,
x4 + iσ · x
g : R4 − (0, 0, 0, 0) → SU (2), g(x1 , x2 , x3 , x4 ) = (110)
r
where σ = (σ1 , σ2 , σ3 ) are the Pauli matrices. Let w = g −1 dg be the Maurer-
Cartan form on SU(2), define a pure-gauge configuration on R4 − (0, 0, 0, 0)
as,
ω = g ∗ w. (111)
which has zero field strength,
dω + ω ∧ ω = 0, (112)
by the Maurer-Cartan equation. The field potential of an instanton would
approach ω at r → ∞, but vanishes at (0, 0, 0, 0) to smooth out the singu-
larity of ω at the origin,
A = f ω, (113)
where f is a smooth function depends only on r. f vanishes at the origin and
approach 1 for r → ∞. We use a different approach to show the twisting is
nontrivial:
The field strength is
F = df ∧ ω + f dω + f 2 ω ∧ ω = df ∧ ω + (f 2 − f )ω ∧ ω. (114)
And the second Chern class is
1
c2 (E) = tr(F ∧ F )
8π 2  
1 2 2
= tr 2(f − f )df ∧ ω ∧ ω ∧ ω + (f − f )ω ∧ ω ∧ ω ∧ ω
8π 2
 
1 2
= tr 2(f − f )df ∧ ω ∧ ω ∧ w (115)
8π 2

24
where ω ∧ ω ∧ ω ∧ ω = g ∗ (w ∧ w ∧ w ∧ w) = 0 because dim SU (2) = 2. Using
the Pauli matrix identities, we find that
df 1 1
tr(df ∧ ω ∧ ω ∧ ω) = −12 dx ∧ dx2 ∧ dx3 ∧ dx4 . (116)
dr r3
So the Chern number over S 4 , or the instanton twisting number is
Z Z
1 1
tr(F ∧ F ) = tr(F ∧ F )
8π 2 S 4 8π 2 R4
Z ∞
1 df 1
= 2
(2π 2 )r3 dr(−24)(f 2 − f )
8π 0 dr r3
Z 1
= −6 df (f 2 − f ) = 1. (117)
0
which has no independence on the detail of f . If we demand the instanton
has the size a, i.e.
r2
f (r) = 2 , (118)
r + a2
then on the other patch of S 4 , (y 1 , y 2 , y 3 , y 4 ) ≡ (x1 , x2 , x3 , x4 )/r2 , we can
check that c2 (E)µνλρ in the y-coordinates is finite at (y 1 , y 2 , y 3 , y 4 ) = (0, 0, 0, 0).
So finite-size instanton configuration is corresponding to a non-singular and
non-trivial fibre bundle.
Example 3.19 (D-brane action). In superstring theories, Dp-brane is a p + 1
extend object which couples to fields in string theory supersymmetrically.
The action for Dp-brane is rather complicated. Here we just consider the
bosonic part of the action.
Type IIA or IIB string theory contains the gravitational field Gµν , the
antisymmetric field Bµν , the dilaton Φ and the U (N ) gauge field Fµν , where
N is the number of Dp-branes. These fields coupled to the brane via the
Dirac-Born-Infeld action,
Z p
SDBI = −τp dp+1 ξe−Φ tr( det(gab + Bab + 2πα0 Fab )) (119)
Mp+1

where det acts on the Dp-branes coordinates indices while tr acts on the
gauge indices. Because Dp-brane is a BPS state which breaks half of the
supersymmetry, it must contain a BPS charge. So it seems that it would
couple to RR fields of the superstring theory as,
Z
SCS = µp Cp+1 (?) (120)
Mp+1

25
However, it is not the whole story since the actually SCS is much more
complicated than this. For example, we consider one D-1 brane in the 1 − 2
plane. Let x0 = ξ 0 , x1 = ξ 1 and x2 = x2 (ξ 1 ) be the embedding of the D-1
brane. The Chern-Simons term reads,
Z Z  
0 1 2
SCS = µ1 C2 = µ1 dξ dξ C01 + ∂1 X C02 (121)
M2

If we apply a T -duality along x2 -direction, the D-1 brane becomes a D-2

brane, the X 2 coordinate becomes the gauge potential component A2 , so the
action is now,
Z   Z  
0 1 2 0 0
SCS = µ2 dξ dξ dξ C012 + 2πα F12 C0 = µp C3 + 2πα F C1 .
Mp+1
(122)
The Chern-Simon action is changed so (120) is not complete.
The complete form is
Z
0
X
SCS = µp [ Cj+1 ] ∧ tre2πα F +B . (123)
Mp+1 j

0
where tre2πα F +B is the Chern character for the mixture of F and B.

4 Pontryagin classes and Euler classes

Now we consider the real vector bundle E. Let E C = E ⊕ iE be its complex-
ification.
Definition 4.1. We define the j-th and total Pontryagin class of E to be
X
pj (E) = (−1)j c2j (E C ) ∈ H 4j (M, Z), p(E) = pj (E) (124)
j

The local frame {eα } for E is also the local frame E C . Define that,

∇(ieα ) = (ieβ )(Aβ )α . (125)

So the local curvature form (F β )α for E C is the same as that for E. Then,
 
F 
p(E) = det I + (126)
2π even in F

26
 The new feature is, in cases we are interested in, F always comes from
the Riemann structure g on E. In this case, F is Riemann anti-symmetric,
(see the appendix),
F T = −gF g −1 (127)
or F T = −F if we use the orthonormal frame such that gαβ = δαβ . So
FT
       
F F −1 F
det I+ = det I+ = det g(I+ )g = det I− (128)
2π 2π 2π 2π
so the odd terms in F automatically vanished and we can define
 
F
p(E) ≡ det I + . (129)
2π
The first several Pontryagin classes are,
1
p1 (E) = − tr(F 2 ) (130)
8π 2
1
p2 (E) = [(trF 2 )2 − 2tr(F 4 )] (131)
128π 4
Example 4.2. For a Riemann manifold M . The local curvature form is
1
(F µ )ν = Rµ νλρ dxλ ∧ dxρ (132)
2
so the first Pontryagin class is,
1
p1 (T M ) = − Rµ νλ1 λ2 Rν µλ3 λ4 dxλ1 ∧ dxλ2 ∧ dxλ3 ∧ dxλ4 (133)
32π 2
Consider a real 2n × 2n antisymmetric matrix A. We have
det(A) = P f (A)2 (134)
where P f (A) is the Pfaffian of A,
(−1)l X
P f (A) = sgn(σ)Aσ1 σ2 ...Aσ2n−1 σ2n . (135)
2l l! σ

Definition 4.3. For a 2n-dimensional orientable real manifold. Choose the

orthonormal frames consistent with the orientation, we define the Euler class
as  
F
e(M ) = P f (136)
2π

27
A Lie-algebra-valued differential forms
Let M be a differential manifold and V is a linear space. A V -valued n-form
φ on M is defined to be an element in Ωn (M ) ⊗ V . φ at a point p ∈ M
naturally induced a linear map,

φp : Tp M ∧ ... ∧ Tp M → V. (137)

This map can be interpreted as the intrinsic definition of the a V -valued n-

form. In particular, if V = g is a Lie algebra, then φ is called a Lie-algebra-
valued n-form. Furthermore, if g is the Lie algebra of G, Adg φ = gφg −1 is
the adjoint action of g ∈ G on the Lie-algebra component of φ.
The exterior derivative on Ω(M ) ⊗ V is,
X
d( ηi ⊗ vi ) ≡ dηi ⊗ vi , (138)
i

where ηi ∈ Ωn (M ) and vi ∈ V .
If V itself is a R or C associate algebra, i.e., V has a product · structure
which is linear under R or C, we can define the exterior product of two
V -valued forms as,

(η ⊗ v) ∧ (η 0 ⊗ v 0 ) = (η ∧ η 0 ) ⊗ (v · v 0 ) (139)

and its linear extensions. Here η ∈ Ωn (M ), η 0 ∈ Ωm (M ) and v, v 0 ∈ V .

In particular, when V = g, the product · is defined by the product in the
universal enveloping algebra of g, i.e. the product of the matrices, not the
commutator.
Note that the usual commutation relation,

φ ∧ φ0 = (−1)mn φ0 ∧ φ (140)

does not hold for φ ∈ Ωm (M ) ⊗ V and φ0 ∈ Ωn (M ) ⊗ V . Hence We define

the commuatator of them as

[φ, φ0 ] = φ ∧ φ0 − (−1)mn φ0 ∧ φ. (141)

and it is still a Lie-algebra-valued form.

For g-valued m-from, φ = Ti ηi and n-forms, φ0 = Ti ηi0

[φ, φ0 ] = [Ti , Tj ] ⊗ (ηi ∧ ηj0 ) (142)

28
and in particular,
(
2φ ∧ φ = [Ti , Tj ] ⊗ (ηi ∧ ηj ) m is odd
[φ, φ] = (143)
0 m is even
Let G be a Lie group and g is its Lie algebra.
Definition A.1. The Maurer-Cartan form of G is the unique g-valued one-
form w such that for a vector X in Tg G,
w(X) = (Lg−1 )∗ X (144)
It is clearly the w is invariant under L∗g . If G is embedded in GL(n, R) and
its element g is written as a n × n matrices, then explicitly its Maurer-Cartan
form is g −1 dg. Conventionally, we may use g −1 dg as the general notation of
the Maurer-Cartan form.
Recall that for a one-form w, two vector fields X and Y ,
dω(X, Y ) = X[ω(Y )] − Y [ω(X)] − ω([X, Y ]) (145)
This relation also holds for V -valued one form. Let ω to be the Maurer-
Cartan form and X and Y to be left-invariant vector fields. At any point
g ∈ G, we have w(Y |g ) = Ye which is independent of g, so X[w(Y )] = 0.
Hence,
dw([X|g , Y |g ]) = −w([X, Y ]|g ) = −[X, Y ]|e
= −[w(X|g ), w(Y |g )] = −(w ∧ w)(X|g , Y |g ), (146)
where in the first we use the fact that the commutator of two left-invariant
fields is still left-invariant. Since the directions of X|g and Y |g are arbitrary,
we have
Theorem A.2 (Maurer-Cartan equation). dw + w ∧ w = dw + 21 [w, w] = 0.
It is useful to rewrite the Maurer-Cartan equation as the component form.
Let {Ti } be in g, we can decompose the Maurer-Cartan form as,
w = Ti ⊗ wi , (147)
where wi ’s are one-forms on G. It is clear that wi ’s are also left invariant. Let
T̃i be the corresponding left-invariant vector fields of Ti , and the structure
constants are,
[T̃i , T̃j ] = fijk T̃k (148)

29
We have,
wi (T̃j ) = δji , (149)
and Theorem A.2 becomes,
1
dwk + fijk wi ∧ wj = 0. (150)
2
The Maurer-Cartan equation plays a crucial role in the study of the fiber
bundle curvature.

B Connection in principal bundles

B.1 Right action and the vertical space
This is a short review of properties of connections in fiber bundles. The
notations follow [1]. Let P (M, G) be a principal bundle with the total space
P , base space M and the structure group G. G acts on P naturally by the
right action,
R: P × G → P
(151)
u g 7→ ug.
The infinitesimal right action linearly maps a Lie algebra element A ∈ g to
a vector field A# in the total space P : Let fu : g 7→ ug be the right action
restricted at the point u ∈ P , then A# at u is defined to be,


#
A  = fu∗ A. (152)
u


It is clear that A#  is in Vu P , the vertical subspace. Since fu∗ is injective,
u
# : g → Vu P is an isomorphism. A# is a smooth vector field on P and the
flow generated on P is u exp(At).
Proposition B.1. For A, B ∈ g, [A# , B # ] = [A, B]# .
Proof.
 
# # # 1 # #
[A , B ] = LA# B = lim Rexp(−tA)∗ B − B
t→0 t
 #
1
= lim Rexp(−tA)∗ ◦ Lexp(tA)∗ B − B = [A, B]# (153)
t→0 t

30
where in the second line, we use the fact that for u ∈ P and g1 , g2 ∈ G,
(ug1 )g2 = u(g1 g2 ).
Let Ti be the basis of g, then at each point u ∈ P , Ti# is a basis of Vu P . If
a vertical vector field X on P , i.e., X is vertical everywhere, then X = ai Ti#
and ai are smooth functions in P . For two vertical vector fields X = ai Ti# ,
Y = bi Ti# ,

[X, Y ] = [ai Ti# , bj Tj# ] = ai bj [Ti , Tj ]# + ai (Ti# bj )Tj# − bj (Tj# ai )Ti# . (154)

where (B.1) is used. Each term is vertical, therefore,

Proposition B.2. For two vertical vector fields X and Y , [X, Y ] is vertical.

B.2 Connection
In general, it is not straightforward to globally and smoothly define the
horizontal subspace, Hu P , the complement of Vu P in Tu P for each u ∈ P .
So we need to introduce the connection.

Definition B.3. A connection on P is a separation of the tangent space at

each point of P , into the vertical and horizontal space, Tu P = Vu P ⊕ Hu P ,
∀u, such that,

C1 (Smoothness) A smooth vector field X on P is separated as X = X V +

X H , such that X V ∈ Vu P , X H ∈ Hu P and both X V and X H are
smooth.

C2 (Right invariance) Hug P = Rg∗ Hu P ∀u ∈ P and g ∈ G.

The second property ensures that a horizontal lift of a curve is still hori-
zontal under the right group action.

Proposition B.4. For A ∈ g and a horizontal vector field Y , [A# , Y ] is

vertical. 5
5
Nakahara’s Lemma 10.2 [1] is not correct. Here is the correct version from [2]. Naka-
hara’s lemma claimed that for X ∈ Hu P and Y ∈ Vu P , [X, Y ] ∈ Hu P . If it is true, then
for a smooth function φ on P , φX is still vertical but [φX, Y ] = φ[X, Y ] − (Y φ)X whose
second term is not horizontal. The problem in Nakahara’s proof is that although there
always is a Ag such that A# |u = X, A# may not equal X everywhere.

31
Proof. A# generate the right action, so
1
[A# , Y ] = LA# Y = lim (Rexp(−At)∗ Y − Y ). (155)
t→0 t

Since the right action Rg∗ maps the horizontal space into another horizontal
space, both of the two terms are horizontal.
In practice, it is convenient to introduce the connection one-form to spec-
ify a connection on a principal bundle.
Definition B.5. A connection one-form is a Lie-algebra-valued one form
ω ∈ g ⊗ T ∗ P such that,
CF1 ω(A# ) = A, ∀A ∈ g.
CF2 Rg∗ ω = Adg−1 ω, where Adg−1 is the adjoint action on the Lie-algebra
component of the connection one-form.
Then we define the Hu P to be,

Hu P ≡ {X ∈ Tu P |ω(X) = 0}. (156)

It is clear that (CF1) guaranteed that Hu P ∩ Vu P = 0 and (CF2) guaranteed

that the right invariance (C2) for the connection is satisfied. The connection
one-form was strictly defined by Ehresmann.

B.3 Local connection form

The connection one-form ω is globally defined on the total space P . For
the simplicity, it would be convenient to define the “projection” of ω on the
base space M . Unfortunately, the global projection on M does not exist in
general. We need to define the projections on the open covering of M and
find the transition formula.
Let Ui be an open covering of M and σi be the local section defined on
Ui . The canonical local trivialization on Ui is φi : Ui × G → π −1 (Ui ), (p, g) 7→
σi (p)g.
φ−1
i φj
Ui ∩ Uj × G > Ui ∩ Uj × G
(157)
φj > < i
φ
−1
π (Ui ∩ Uj )

32
where φ−1i φj (p, g) = (p, tij (p)g). Here tij (p) ∈ G is the transition function
and σj (p) = σi (p)tij (p).
Remark B.6. It is a good place to see why it is necessary to use the connection
to define the horizontal space. Can we naively define the horizontal space
Hu P as (φj )∗ Tp M ? No. The problem is that this definition is not consistent
with the transition function. It is clear that for u = φj (p, g), X ∈ Tp M ,

(φj )∗ (X, 0) = (φi )∗ (X, (Rg tij )∗ X). (158)

Hence in general, (φj )∗ Tp M 6= (φi )∗ Tp M and Hu P cannot be defined in this

way.
We define the local connection one-form Ai on Ui as,

Ai ≡ σi∗ ω (159)

Ont the other hand, give a g-valued one-form on Ai on Ui , we can define the
corresponding connection one-form ωi on π −1 (U ), as

ωi = Adgi−1 (π ∗ Ai ) + gi−1 dgi (160)

where gi is the local coordinate for the total space such that φ(p, g) = u. So
gi : π −1 (U ) → G is a well-defined mapping and gi−1 dgi is the pullback of the
Maurer-Cartan form of G. It is straightforward to verify that,
• σi∗ ωi = Ai .
• ωi is a connection one-form on the bundle π −1 (Ui ), i.e., ωi satisfies the
conditions (CF1) and (CF2).
The compatibility condition for the local connection one-forms is ωi = ωj in
Ui ∩ Uj , ∀i, j. By the explicit construction (160), the compatibility condition
can be determined as,

Aj = t−1 −1
ij Ai tij + tij dtij . (161)

This relation is the gauge transformation in Yang-Mills theory.

A nontrivial fiber bundle does not have a global section, so there are
several patches Ui and each patch has a local connection one-forms Ai . Since
on each patch the bundle is trivial, Ai itself does not contain the global
information of the bundle. It is the transformation (161) that determines
the global structure.

33
B.4 Horizontal lift and the holonomy group
For a curve γ(t) in M , it is clear that there exist many curves γ̃(t) in P such
that π(γ̃(t)) = γ(t). We call γ̃(t) a lift of γ(t). In general γ̃(t) can have
“vertical-direction-motion”, We want to find a unique lift which is always in
the horizontal direction,
Definition B.7. For a curve γ : [0, 1] → M , its lift γ̃ : [0, 1] → P is γ’s
horizontal lift if and only if γ̃∗ (∂/∂t) is horizontal ∀t ∈ [0, 1].
We have the following theorem for the existence and uniqueness of hori-
zontal lift,
Theorem B.8. Let γ : [0, 1] → M be a C 1 (continuously differentiable)
curve in M and u ∈ π −1 (γ(0)). Then there exists a unique continuously
differentiable curve γ̃ such that π(γ̃) = γ and γ̃(0) = u.
Proof. For the proof without using the local sections, see [2]’s proposition
3.1. Here we sketch the proof with the local sections. Let Uα , σα to be
the local trivializations of M . By the compactness of [0, 1], we can divide
the interval as [t0 = 0, t1 ], ..., [tN −1 , tN ] such that each segment of the curve
γ : [tn−1 , tn ] → M is inside a Un . For t0 ≤ t ≤ t1 , we need to construct a
horizontal lift as γ̃(t) = σ1 (γ(t))g1 (t) in U1 , where σ1 (γ(0))g1 (0) = u. The
derivative of γ̃ is,
   #
dγ̃ dγ
−1 dg1
= Rg1 ∗ σ1∗ + (Lg1 )∗ , (162)
dt dt dt
By the horizontal condition,
 
dγ̃ dγ

dg1
0 = ω( ) = Adg1−1 ω σ1∗ + (Lg1 )−1
∗ (163)
dt dt dt
which is a differential equation for g1 (t),
   
dg1 dγ
dγ
= −ω σ1∗ g1 = −A1 g1 , (164)
dt dt dt
where in the second equality we used the local connection form A1 = σ1∗ ω.
By the fundamental theorem of ODE’s, the horizontal lift exist uniquely for
t ∈ [0, t1 ]. Repeat this process for N − 1 times, we get the unique horizontal
lift for t ∈ [0, 1].

34
 Formally, in each patch, the horizontal lift can be formally written as
the path-ordered form, i.e. to put late-time operator before the early-time
operator in the product.
 Z t 
dγ

gn (t) = P exp − An dt gn (tn−1 ) (165)
ti−1 dt
 Z t
dxi

= P exp − An,i dt gn (tn−1 ) (166)
ti−1 dt

where we used the local coordinate for Un in the second line. gn (tn−1 ) is
determined by the previous patch.
For a curve γ(t) in M and its horizontal lift γ̃(t) with γ̃(0) = u, we
denote the γ̃(1) as the parallel transport of u along γ. The parallel transport
induces an isomorphism Γγ between π −1 (γ(0)) and π −1 (γ(1)) because of the
uniqueness of the horizontal lift.
Because of the right-invariance of the connection form (CF2), we have
the following proposition:

Proposition B.9. If γ̃(t) is the horizontal lift of γ(t) with γ̃(0) = u, then If
γ̃(t)g is the horizontal lift of γ(t) with γ̃(0) = ug,

In the other word, if v ∈ P is the parallel transport of u along γ, then vg

is the parallel transport of ug along γ.
Now we consider the closed C 1 loop on M, γ(0) = γ(1) = π(u). The
parallel transport for u along γ is in the same fibre of u, so it can be written
as ugγ , where gγ ∈ G is the unique right action determined by γ. Consider
all the C 1 loops passing π(u), we define the holonomy group at u as,

Φu = {gγ |γ(t) is a C 1 curve in M , γ(0) = γ(1) = π(u)}. (167)

Φu is a subgroup of G because,

gγ1 gγ2 = gγ2 ∗γ1 , gγ−1 = gγ −1 . (168)

Note the order of the product of the two loops. Similarly, we can define the
restricted holonomy group as,

Φ0u = {gγ |γ(t) is a C 1 curve in M , γ(0) = γ(1) = π(u), [γ] = id. ∈ π(M, u)}.
(169)

35
Proposition B.10. The holonomy groups has the following properties,
1. For u ∈ P , Φug = Adg−1 (Φu ) and Φ0ug = Adg−1 (Φ0u ).

2. If u, v ∈ P can be connected by a horizontal line, then Φu = Φv and

Φ0u = Φ0v .
Proof. 1. Let v = ua be the parallel transport of u along γ. By (B.9),
vg = uag = ug(g −1 ag) is the parallel transport of ug along γ. So
Adg−1 is an isomorphism from Φu to Φug (and for the restrict holonomy
group).

2. Let α(t) to be the horizontal curve from u to v. If a loop γ, starting

and ending at π(v), induces the parallel transport from v to vgγ , then
the loop α ∗ γ̃ ∗ Rgγ (α−1 ) is horizontal curve starts from u and ends at
ugγ . This proved Φu = Φv . Furthermore if γ is trivial in π(M, π(v)),
the projection of α ∗ γ̃ ∗ Rgγ (α−1 ) is trivial in π(M, π(u). So Φ0u = Φ0v .
So for a connected base space we just need the holonomy group for an
arbitrary point u ∈ P .

B.5 Curvature form

The connection on a principal bundle P (M, G) separates Tu P = Hu P ⊕Vu P :
X = X H + X V . Let V be a linear space, for a V -valued n-form φ on P , we
define the covariant derivative of φ by,

Dφ(X1 , ..., Xr+1 ) = dφ(X1H , ..., Xr+1

H
). (170)

So Dφ is a V -valued (n + 1)-form. Because under Rg∗ , Hu P → Hug P and

Vu P → Vug P , we have (Rg∗ X)H = Rg∗ (X H ). Then it is straightforward to
check that,
Rg∗ ◦ D = D ◦ Rg∗ . (171)
The curvature two-form Ω is a g-valued two form defined as,

Ω ≡ Dω. (172)

Because of (171) and (CF2),

Rg∗ Ω = Adg−1 Ω. (173)

36
Theorem B.11 (Cartan structure equation).
1
Ω = dω + ω ∧ ω = dω + [ω, ω] (174)
2

Proof. It is sufficient to prove that for any X, Y ∈ Tu P .

Ω(X, Y ) ≡ dω(X H , Y H ) = dω(X, Y ) + [ω(X), ω(Y )] (175)

Every X in Tu P can be decomposed as X = X H + X V , so by the linearity,

we only need to consider three cases,
• X ∈ Hu P and Y ∈ Hu P . By the definition of the horizontal space,
ω(X) = ω(Y ) = 0. So Ω(X, Y ) = dω(X, Y ).

• X ∈ Vu P and Y ∈ Vu P . In this case, by definition the left hand side of

(175) is 0. Let A, B ∈ g such that A# |u = X and B # |u = Y . Then, 6

dω(X, Y ) = A# (ω(B # )) − B # (ω(A# )) − ω[A# , B # ]

= A# (B) − B # (A) − [A, B] = [A, B] (176)

where in the second line we used the proposition B.1. Hence dω(X, Y )+
[ω(X), ω(Y )] = [A, B] − [A, B] = 0 which equals the left hand side.

• X ∈ Vu P and Y ∈ Hu P . In this case, again the left hand side of

(175) is zero. Define Ag such that A# |u = X. We also extend Y to a
horizontal vector field, which is still called Y . Then,

dω(X, Y ) = −Y (ω(A# )) − ω([A# , Y ]) = 0, (177)

because [A# , Y ] is horizontal by proposition B.4. It is clear that [ω(X), ω(Y )] =

0. So again, both sides of (175) are zero.

Example B.12. For each local trivialization (Ui , σi ) such that u = σi (π(u))g,
∀u ∈ π −1 (Ui )), we can define the canonical flat connection ωi = g −1 dg = g ∗ w,
where w is the Maurer-Cartan form of G. It is clear that the horizontal space
6
This formula holds for any extension of the vectors X, Y . So it does not matter if
#
A 6= X outside the point u.

37
defined by ωi is σi∗ TP M for p = π(u). Therefore Ai = σi∗ ωi = 0. wi has zero
curvature, because
Ωi = dωi + ωi ∧ ωi = g ∗ (dw + w ∧ w) = 0, (178)
where we used the Maurer-Cartan equation (A.2). For a principal bundle P
with the global connection form ω, if we can find a set of local trivializations
(Ui , σi ) such that ω = ωi for all Ui , ω is called the flat connection. Such ω
has zero curvature everywhere.
Theorem B.13 (Bianchi’s identity). DΩ = 0
Proof. For vectors X, Y, Z ∈ Tu P , we have
DΩ(X, Y, Z) = dΩ(X H , Y H , Z H ) = (dω ∧ ω − ω ∧ dω)(X H , Y H , Z H ),
(179)
which vanishes since no matter which horizontal vector is combined with ω,
the result is always zero. Here we used the Cartan structure equation.
Definition B.14. We define the local curvature form on a local trivialization
(Ui , σi ) as, F = σ∗ Ω.
Pull back the Cartan structure equation (B.11) by σi∗ , we get its local
form
F = dA + A ∧ A (180)
which is the field strength expression in Yang-Mills theory.
Example B.15 (Pure gauge). The local connection A on Ui is called pure
gauge if A = g −1 dg, where g : Ui → G is a differentiable map. Pure guage
connection has zero local curvature. Let w to be the Maurer-Cartan form on
G. Then A = g ∗ w. By the Maurer-Cartan equation (A.2), dw + w ∧ w = 0
so F = dA + A ∧ A = g ∗ (dw + w ∧ w) = 0.
Proposition B.16. The local form of the Bianchi’s identity is,
0 = dF + [A, F] ≡ DF = 0. (181)
Proof. By Cartan structure equation (B.11), dΩ = dω ∧ ω − ω ∧ dω. The
pullback by σi∗ is
0 = dF − dA ∧ A + A ∧ dA = dF + [A, F]. (182)

38
 On the intersection of two branches Ui and Uj , the local curvature forms
are related by,
Fj = Adt−1
ij
Fi . (183)
which can be checked by the direct calculation. This is the gauge transfor-
mation for field strength in Yang-Mills theory.

B.6 Ambrose-Singer Theorem

The non-triviality of the horizontal lift for a principal bundle P is determined
by the curvature form Ω of P . Ambrose and Singer proved the holonomy
theorem for the relation between the restricted holonomy ground and the
curvature form.
Theorem B.17 (Ambrose-Singer). For u ∈ P , the Lie algebra of Φu is
spanned by {Ω(X, Y )|X, Y ∈ Hv P } for all v ∈ P such that v can be connected
with u by a horizontal curve.
One important conclusion from the holonomy theorem is that:
Proposition B.18. For a principal bundle P (M, G) with the connection
form ω and the curvature form Ω,
• The connection of P (M, G) is flat if and only if Ω vanishes everywhere.
• If Ω vanishes every where and M is simply connected, then P (M, G) is
isomorphic to the trivial bundle M × G and by this isomorphism ω is
mapped to the canonical flat connection form of M × G.
Proof. See [2], theorem 9.1.

C Connection on associated fibre bundles

C.1 Associated fibre bundles
Definition C.1. Let the structure group G act on a manifold V ,
G×V →V (184)
then from a principal bundle P (M, G) we can construct its associated bundle
E(M, V, G, P ) with the fibre V as,
E(M, V, G, P ) = P ×G V = P × V /{(u, v) ∼ (ug, g −1 v)}. (185)

39
 The fibre bundle projection is,

πE : E → M, [(u, v)] → π(u). (186)

In the following discussion we may simply use the notation π instead of πE .

The local trivialization of E is naturally determined by the P ’s local
sections (Ui , σi ). Define the local trivialization of E on Ui as φi : Ui × V →
π −1 (Ui ) : (p, v) 7→ [(σi (p), v)]. The transition relation is now,
φ−1
i φj
Ui ∩ Uj × V > Ui ∩ Uj × V
(187)
φj > < i
φ
−1
π (Ui ∩ Uj )

where φ−1i φj (p, v) = (p, tij (p)v). Recall that σj (p) = σi (p)tij (p). So for
the associated bundle, the transition function is the V -representation of the
transition function of the principal bundle.
Although G acts on V , it does not naturally act on πE−1 (p), for p in M .
We define the left action referred to u as, where u ∈ πP−1 (p),

Lu (g)([ua, v]) ≡ [(u, gav)], a ∈ G, g ∈ G (188)

Lu is a group homomorphism: G → Aut(πE−1 (p)). Note that Lua (g) =

Lu (a)Lu (g)Lu (a−1 ).

C.2 Connection on associated fibre bundle

For the principal bundle P (M, P ) with a connection, it is natural to define
the corresponding connection of the associated bundle E(M, V, G, P ).

Definition C.2 (Connection). For w = [(u, v)] ∈ E, we define the horizontal

space Hw E ⊂ Tw E to be
 
d
Hw E = [(γ̃(t), v)]t=0 |γ̃(t) is horizontal in P , γ̃(0) = u (189)
dt

This definition has no dependence of the particular choice (u, v) either.

We can use another representative w = [(ug, g −1 v], a ∈ G and a horizontal
curve γ̃ with γ̃(0) = ug. Then [(γ̃(t), g −1 v)] = [(γ̃(t)g −1 , v)] and γ̃(t)g −1

40
is also horizontal in P because of (C2). By the local trivialization, it is
straightforward to check that

Tw E = Hw E ⊕ Vw E. (190)

and πE : Hw E → Tp M is an isomorphism.
Again, a curve γ̃E (t) in E is a horizontal lift of a curve γ(t) in M , if
and only if πE (γ̃E (t)) = γ(t) and d/dt(γ̃E (t)) is horizontal in E, ∀t. If the
horizontal lift of γ(t) is known, γ̃E (t) can be determined as follows: Let
γ(0) = p, πE (w) = p, w = [(u, v)], γ̃(t) be the horizontally lift of γ(t) in P.
Then,
γ̃E (t) = [(γ̃(t), v)] (191)
is the horizontal lift of γ(t) in E with the starting point w. Again, the
horizontal lift in the associated bundle with specified starting point exists
and is unique.
Let γ be a smooth loop in M such that γ(0) = γ(1) = p. Again, The
horizontal lift on E evaluated at t = 1 gives a linear isomorphism,

γ : πE−1 (p) → πE−1 (p), (192)

which is called the parallel transport along γ. As before, consider all the
smooth loops or contractible loops, we define the holonomy group Holp and
restricted holonomy group Hol0p , as the subgroup of Aut(π −1 (p)).

Proposition C.3. Holp = Lu (Φu ) and Hol0p = Lu (Φ0u ), where Φu and Φ0u are
the holonomy groups at u for the principal bundle P .
Proof. For a loop γ in M such that γ(0) = γ(1) = p. Let γ̃ be the lift of γ
in P with γ̃(0) = u and γ̃(1) = ug. Then [(γ̃(t), v)] is a horizontal lift of γ
in E. The parallel transport of any [(u, v)] ∈ πE−1 (p) is,

[(γ̃(1), v)] = [(ug, v)] = [(u, gv)] = Lu (g)([(u, v)]). (193)

Therefore for any element in T ∈ Holp , there exists an element g ∈ G such

that Lu (g) = T .
Note that Φua = Ada−1 (Φu ), so

Lua (Φua ) = Lu (Ada (Φua )) = Lu (Φu ), (194)

which is consistent with the fact that Holp is independent of the choice of u.

41
C.3 Covariant derivative
In this section, we assume the V is a vector space with the field F = R or C.
The linear operators on E(M, V, P, G) are
• For w1 , w2 ∈ πE−1 (p), choose w1 = [(u, v1 )] and w2 = [(u, v2 )] and
w1 + w2 = [(u, v1 + v2 )] ∈ πE−1 (p).

• For w ∈ E and c ∈ F, choose w = [(u, v)] and cw = [(u, cv)].

For a vector bundle, the parallel transformation is linear, so Holp and Hol0p
are in Gl(π −1 p, F).
And we can naturally identify a vertical vector A ∈ Vw E defined by,
d 
A= [(u, v(t))]t=0 , v:R→V (195)
dt
as an element of E in the same fibre of w,

A ≡ [(u, v̇(0))]. (196)

Let s ∈ Γ(M, E) be a differentiable section on E, X be a vector Tp M .

Denote w = s(p). In general, the pushforward of X by s, s∗ X, may not be
horizontal in Tw E. We define the covariant derivate, ∇X s, to indicate the
discrepancy between s and the horizontal lift of X,
Definition C.4 (Covariant derivative).

∇X s = (s∗ X)V . (197)

∇X s is vertical and can be identified as an element in πE−1 (p). This

definition does not refer to the particular representatives of E.
Explicitly, ∇X s can be calculated as follows: extend X into a curve γ,
γ(0) = p with γ̇(0) = X. Let w = (u, v) and denote γ̃(t) as the horizontal
lift of γ(t) in P with γ̃(0) = u. Then γ̃E (t) = [(γ̃(t), v)] is the horizontal lift
in E. If s(γ(t)) = [(γ(t), η(t))], then

d 
s∗ X = γ̃E (t) + [(γ(0), η̇(0))] (198)
dt t=0

where the second term is the vertical component. Hence,

∇X s = [(γ(0), η̇(0))]. (199)

42
This formula implies that ∇X s is determined only by the values of s along
γ(t). Furthermore, if ∇γ̇(t) X = 0, ∀t, then s(γ(t)) is the horizontal lift of
γ(t) in E.
For a smooth vector field X, ∇X s is a global section of E with (∇X s)(p) =
∇Xp s. Formally, we define ∇s to be an element of Γ(M, E) ⊗ Ω1 (M ) such
that at each p ∈ M

X(∇s) ≡ ∇X s, ∀X ∈ Tp M. (200)

It straightforward to check that the covariant derivatives satisfy,

CD1 ∇(c1 s1 + c2 s2 ) = c1 ∇s1 + c2 ∇s2 ,

CD2 ∇(f s) = (df )s + f ∇s,

for c1 , c2 ∈ F and f ∈ C ∞ (M ). Alternatively we can define the covariant
derivate to be a map ∇ : Γ(M, E) → Γ(M, E) ⊗ Ω1 (M ) which satisfies
axioms (CD1) and (CD2). In this way we can recover the whole theory of
the connection.

C.4 Local form of the connection

Let ω be the connection form on P (M, G). It is not straightforward to write
down the corresponding global connection form on E(M, V, P, G). So we
try to find the local connection form of E(M, V, P, G) in terms of the local
connection form of P (M, G).
As before, (Ui , σi ) is the set of local trivializations of P (M, G). The local
connection form of P on Ui is Ai = σi∗ ω. Choose a basis {eα } for V . The
action G × V → V induces the Lie-algebra representation g × V → V ,

T eα = T β α eβ , T ∈g (201)

where T β α is the V -representation matrix. Note the {eα } is a basis not the
components, so the matrix T β α ’s first index contract with eβ .
On Ui , we can define the local canonical frame as,

sα (p) = [(σi (p), eα )]. (202)

It is clear that at each point p ∈ Ui , these sections form a basis for πE−1 (p). φi :
Ui × V → π −1 (Ui ), where φi (p, v) = [(σi (p), v)] is the canonical trivializations
of E.

43
Proposition C.5. ∇sα = (Ai )β α sβ .

Proof. Let X be a vector in Tp M and γ(t) is a curve in M such that γ̇(0) = X.

Define γ̃(t) = σi (p(t))gi (t) as the horizontal lift of γ(t), and gi (0) = e ∈ G.
Then the horizontal lift of γ(t) in E with the starting point sα (p) is,

γ̃E (t) = [(σi (p(t))gi (t), eα )] (203)

while si (p(t)) reads,

si (p(t)) = [(σi (p(t)), eα )] = [(σi (p(t))gi (t), gi (t)−1 eα )]. (204)

Therefore,
   
d −1
∇X eα = σi (p), gi (t) eα
dt
  t=0 
−1 d −1
= σi (p), −gi (t) gi (t) gi (t) eα
dt
= [(σi (p), Ai (X)eα )] (205)

where we used the ODE (164) for gi (t). Hence, in terms of the basis {sα },

∇X sα = Ai (X)sα ≡ (Ai (X))β α sβ (206)

The covariant derivative of sα ’s determines all sections’ covariant deriva-

tives, by (CD2).

Corollary C.6. Let ξ be a smooth section on Ui such that ξ = ξ α (p)sα .

∇X ξ = ξ α (Ai (X))β α sβ + (Xξ α )sα

∂ξ α
 
µ β α
= X ξ Aiµ β + µ sα (207)
∂x

In practice, we can choose an arbitrary frame {sα }. A frame provides a

local trivialization: φ : U × V → π −1 (U ), such that φ(p, eα ) 7→ sα (p). By
(207). We can define,
∇sα ≡ Aβ α s0β . (208)

44
For another frame s0α = M (p)β α sβ , M (p) ∈ GL(V, F), the new connection
matrix is,
β
A0i α = (M −1 )β γ (A)γ δ M δ α + (M −1 )β γ dM γ α (209)
In particular, if {sα } and {s0α } are the canonical sections on Ui and Uj re-
spectively, then
M β α = (tij )β α , (210)
and the transition (209) is the V -representation of the connection transition
law on the principal bundle.
Example C.7 (frame bundle and tangent bundle).

C.5 Local curvature form

Let F be the local curvature form on M for the principal bundle. The
local curvature form for the associated bundle is defined to be F α β , the
V -representation of F . In components,

F β α = dAβ α + Aβ γ ∧ Aγ α . (211)

If we change the sections as s0α = M (p)β α sβ , then,

β
F0 α = (M −1 )β γ (F)γ δ M δ α . (212)

where F 0 β α is calculated from A0 β α .

The local curvature form is determined by the covariant derivatives.
Proposition C.8. For two vector fields X, Y on M and a smooth section s
for E,
(∇X ∇Y − ∇Y ∇X )sα = F β α (X, Y )sβ + ∇[X,Y ] sα . (213)
Proof. By (C.5) and (CD2), we have

∇X ∇Y sα = ∇X (Aβ α (Y )sβ ) = Aγ β (X)Aβ α (Y )sγ + X(Aβ α (Y ))sβ ,

∇Y ∇X sα = ∇Y (Aβ α (X)sβ ) = Aγ β (Y )Aβ α (X)sγ + Y (Aβ α (X))s(214)
β.

Note that,

dAβ α (X, Y ) = X(Aβ α (Y )) − Y (Aβ α (X)) − Aβ α ([X, Y ]), (215)

so (213) holds.

45
 Formally, we can extend ∇ as a linear map Γ(M, E⊗Λp (M ∗ )) → Γ(M, E⊗
Λ (M ∗ )),
p+1

∇(s ⊗ η) ≡ (∇s) ∧ η + s ⊗ dη (216)

for s ∈ Γ(M, E), η ∈ Ωp (M ). Then it is straightforward to check that,

∇∇sα = sβ ⊗ F β α (217)

Unlike the covariant derivative, F(X, Y ) : π −1 (p) → π −1 (p) is well-defined,

F(X, Y )(cα sα (p)) ≡ sβ (p)F β α (X, Y )cα , (218)

and it is straightforward to check that it is independent of the frame choice.

If the vector bundle is associated with the principle bundles P , using the
canonical trivialization (Ui , σi ), we get

F(X, Y )([(σi (p), sα )]) = [(σi (p), sβ F β α (X, Y ))], (219)

where F β α is the V -representation matrix of Fi , the curvature form on the

principal bundle. Hence in Gl(π −1 (p), F),

F(X, Y ) = Lσi (p) (Fi (X, Y )) (220)

where we use the same notation Lσi (p) for the Lie algebra map.7
Theorem C.9 (Ambrose-Singer). Holp ’s Lie algebra is spanned by Γγ −1 ·
F(X, Y ) · Γγ , where γ is piecewise smooth curve in M with γ(0) = p and
γ(1) = q. X, Y ∈ Tq M .
Proof. We have following diagram,
Lφi (p)
G > Gl(π −1 (p), F)
(221)
Lφi (p)
Φσi (p) > Holp

where the section line is surjective. By the Ambrose-Singer theorem for

principal bundles, Φσi (p) ’s Lie algebra is spanned by {Ω(X, Y )|X, Y ∈ Hv P }
for all u0 ∈ P which can be connected to σi (p) by a horizontal curve. Without
7
Explicitly, we can verify that the definition is independent of the local section choice.
Lσj (p) (Fj ) = Lσi (p)tij (p) (Fj ) = Lσi (p) (tij (p))Lσi (p) (Fj )Lσi (p) (t−1
ij (p)) = Lσi (p) (Fi ).

46
loss of generality, we defined γ̃ be such a curve with γ̃(0) = σi (p) and γ̃(1) =
σj (q), q ∈ Uj ⊂ M . Let γ(t) = π(γ̃(t)). It is clear that,

Lσi (p) (Ω(X̃, Ỹ ))[(σi (p), v)] = [(σi (p), Ω(X̃, Ỹ )v)] (222)

for X̃, Ỹ ∈ Hσj (q) P . On the other hand, we can find vectors X, Y ∈ Tq M ,
such that,
Ω(X̃, Ỹ ) = Ω(σj∗ (X), σj∗ (Y )) = Fj (X, Y ). (223)
Furthermore, the parallel transport of γ induced a linear isomorphism πE−1 (p) →
πE−1 (q),
γ : [(σi (p), v)] 7→ [(σj (q), v)] (224)
Hence,

Lσi (p) (Ω(X̃, Ỹ ))[(σi (p), v)] = [(σi (p), Fj (X, Y )v)]
= γ −1 · Lσj (q) (Fj (X, Y )) · γ[(σi (p), v)]
= γ −1 · F(X, Y ) · γ[(σi (p), v)] (225)

Hence Lσi (p) (Ω(X̃, Ỹ )) = γ −1 · F(X, Y ) · γ.

Definition C.10. We define the Ricci form to be the trace of F,

R = F αα (226)

which is frame independent. A connection is called Ricci flat if R ≡ 0.

Corollary C.11. If the connection is Ricci flat, then ∀p ∈ Ui , Holp ⊂

sl(π −1 (p), F).

Proof. The Lie algebra of Holp is {F(X, Y )|X, Y ∈ Tp M }.

tr(F(X, Y )) = F α α (X, Y ) = 0. (227)

C.6 Riemann structure, revisited

In this section, we revisit our old friend, Riemann structure. Our discussion
work for any real vector bundle E → M , not restricted to the tangent bundle
T M.

47
 A Riemann structure on E means there is continuously defined inner
product for each fibre π −1 (p), p ∈ M .

gp : π −1 (p) × π −1 (p) → R (228)

A connection ∇ is called a metric connection if it preserves the inner product:

for two sections s1 , s2 and a vector X ∈ Tp M

X(g(s1 , s2 )) = g(∇X s1 , s2 ) + g(s1 , ∇X s2 ). (229)

In the other word, δX g = 0 if we extend the covariant derivative to the dual

space of V . The metric connection is restricted by the Riemann structure:
Let g(sα , sβ ) = gαβ , we have

dgαβ = gγβ Aγ α + gαγ Aγ β , dg = gA + AT g (230)

Take the exterior derivative, we have,

F β α = −gαγ F γ δ g δβ , F T = −gFg −1 (231)

Example C.12. If E = T M , we can choose Ui = {(x1 , ..., xn )} and the sections

sµ = ∂/∂xµ . In this case, (230) and the torsion free condition

Aµ νλ = Aµ λν , Aµ λ = Aµ νλ dxν (232)

uniquely determine the Levi-Civita connection, Γµνλ ≡ Aµ νλ . Furthermore,

the Riemann curvature tensor is defined by,

Rρ σµν ≡ (F ρ σ )µν = ∂µ Γρνσ − ∂ν Γρµσ + Γρµλ Γλνσ − Γρνλ Γλµσ . (233)

However, when E 6= T M , there may not exist a natural way to define the
torsion-free condition or the corresponding Levi-Civita connection.
Locally, we can always choose an orthonormal frame {êα } for E such that
g(êα , êβ ) = δαβ . The connection form for {êi } satisfy,

Aβ α = −Aα β (234)

which is so Lie-algebra-valued local connection one forms. This implies that

the structure group G = GL(R) can be reduced to O(R).

48
C.7 Holomorphic vector bundle and Hermitian struc-
ture
In this section, we follow both [1] and [4]. Let π : E → M be a complex
vector bundle. The fibre has complex dimension k. The manifold E is a
holomophic vector bundle if its a complex vector bundle and satisfy,
1. E and M are complex manifold and π is a holomorphic map.
2. The local trivialization φi : Ui × C → π −1 (Ui ) is a biholomorphism.
3. The transition function tij : Ui ∩ Uj → GL(k, C) is holomorphic.
Example C.13. Let M be a complex manifold with dimC = m. For p ∈ M ,
we choose p’s coordinate neighborhood Ui = {z 1 , ...z m }. The holomorphic
tangent space at p is
∂ ∂
T Mp+ ≡ T (1,0) M = spanC { 1
, ..., m }. (235)
∂z ∂z
By the complex structure of M , T Mp+ is independent of the coordinate choice.
We define the holomorphic tangent bundle as
[
TM = Tp M + , (236)
p∈M

which is a 2m-dimensional complex manifold with complex structure defined

by {z 1 , ..., z m , c1 , ..., cm }, where cµ is the complex coefficient of ∂/∂z µ . It
clear that for another coordinate neighborhood Uj = {w1 , ..., wm },
∂z µ
(tij )µ ν = , (237)
∂wν
which is a holomorphic map from M to GL(m, C). So T M + is a holomorphic
vector bundle.
¯ ∈ Γ(M, E ⊗ (T ∗ M )C ). For
For a section s ∈ Γ(M, E), we can define ∂s
k
∀p ∈ Ui , choose a map si : Ui → C such that,

φi (p, si (p)) = s (238)

Then for a vector X ∈ Tp M C ,

¯
∂s(X) ¯ i (X)).
= φi (p, ∂s (239)

49
¯ is well-defined because if we use a different trivialization,
∂s

φj (p, sj (p)) = s, sj (p) = tji (p)si (p) (240)

then
¯ j (X)) = φj (p, (∂t
φj (p, ∂s ¯ ji (p)si )(X)) = φj (p, tji (p)∂s
¯ i (X)), (241)
¯ ji = 0. Note that there is no natural way to define ∂s. Formally,
because ∂t
we can extend ∂¯ as linear map Γ(M, E ⊗ Λp (M ∗ )C ) → Γ(M, E ⊗ Λp+1 (M ∗ )C ),
¯ ⊗ η) ≡ (∂s)
∂(s ¯ ∧ η + s ⊗ ∂η
¯ (242)

for s ∈ Γ(M, E), η ∈ Ωp (M )C . Again,

Proposition C.14. ∂¯∂¯ = 0.

Proof. Introduce a local coordinate neighborhood [(z 1 , ..., z n )], and define
¯ ≡ ∂¯µ s ⊗ dz̄ µ . Each sµ is a local section in E.
∂s
¯ ⊗ η) = ∂(s
∂(s ¯ µ ⊗ (dz̄ µ ∧ η) + s ⊗ ∂η) ¯
= sµν (dz̄ ν ∧ dz̄ µ ∧ η) = 0. (243)
¯ µ ≡ sµν ⊗ dz̄ µ and used the fact sµν = sνµ , ∂¯∂η
where we defined ∂s ¯ = 0.

Definition C.15 (Hermitian structure). For a holomorphic vector bundle

E, a Hermitian structure is the Hermitian inner products h defined ∀p ∈ M ,
which satisfies,

1. hp (c1 u1 + c2 u2 , v) = c1 hp (u1 , v) + c2 hp (u2 , v), c1 , c2 ∈ C and u1 , u2 , v ∈

π −1 (p), 8

2. hp (u, v) = hp (v, u),

3. hp (u, u) ≥ 0, while the equality holds only if u = 0 ∈ π −1 (p),

4. h(s1 , s2 ) is a complex smooth function on M , if s1 , s2 ∈ Γ(M, E).

8
This convention follows [4] but is different from [1].

50
 For a frame sα , we define hαβ̄ = h(sα , sβ ). So the inner products for
s1 = f α sα and s2 = g α sα ,
h(s1 , s2 ) = hαβ̄ f α ḡ β . (244)
We can formally define the Hermitian connection on a holomorphic vector
bundle E,
Definition C.16. The Hermitian connection ∇ is a C-linear map Γ(M, E) →
Γ(M, E ⊗ (T ∗ M )C ), which satisfies:
HC1 ∇(f s) = (df )s + f ∇s, where f is a complex smooth function on M and
s ∈ Γ(M, E),
HC2 d[h(s1 , s2 )] = h(∇s1 , s2 ) + h(s1 , ∇s2 ),
HC3 ∇s = Ds + D̄s, where Ds ∈ Γ(M, E ⊗ T (1,0) M ) and D̄s ∈ Γ(M, E ⊗
¯
T (0,1) M ). We require that D̄s = ∂s.
The Hermitian connection uniquely exists for a Hermitian structure. Again,
we can extend ∇ and D as linear map Γ(M, E ⊗ Λp (M ∗ )C ) → Γ(M, E ⊗
Λp+1 (M ∗ )C ),
∇(s ⊗ η) ≡ (∇s) ∧ η + s ⊗ dη,
D(s ⊗ η) ≡ (Ds) ∧ η + s ⊗ ∂η, (245)
for s ∈ Γ(M, E), η ∈ Ωp (M )C .
As before, let ∇sα = sβ Aβ α . (HC2) can be written as,
dhαβ̄ = hγ β̄ Aγ α + hαγ̄ Aγ β , or dh = AT h + hA (246)
Take the exterior derivative, we have,
F α β = −hᾱγ F δ γ hδβ̄ , or F = −h−1 F T h (247)
where hᾱβ is the inverse of the metric matrix such that hᾱβ hβγ̄ = δ δ γ .
We can choose the local holomorphic frame sα for E. A local section s is
called holomorphic, if and only if, under local trivialization φi : Ui × Ck →
π −1 (Ui ),
φi (p, si (p)) = s (248)
si (p) is a holomorphic map Ui → Ck . This definition is independent of the
¯ = 0 for a holomorphic section. Therefore, Aα β
Ui choice. It is clear that ∂s
for the holomorphic frame is a (1, 0)-form. Furthermore, by (246), extract
the holomorphic part and we have

51
Theorem C.17. For the holomorphic frame Aα β = hγ̄α ∂hβγ̄ , or A = (h−1 )T ∂hT .
Hence the Hermitian connection for a Hermitian structure uniquely exists.

Corollary C.18. For the holomorphic frame, the curvature form F α β =

¯ α β and is a (1, 1)-form.
∂A

Proof.

F = dA + A ∧ A
= ¯ T + (h−1 )T ∂hT ∧ (h−1 )T ∂hT
−(h−1 dhh−1 )T ∧ ∂hT + (h−1 )T ∂ ∂h
= ¯ T ∧ (h−1 )T ∂hT + (h−1 )T ∂ ∂h
−(h−1 )T ∂h ¯ T
= ¯
∂A (249)

So F α β is a (1, 1)-form.
The trace of F is,
¯
trF = F α α = ∂∂(log det(h)). (250)

Alternatively, we can locally choose the orthonormal frame {êα } such

that E such that g(êα , êβ ) = δαβ . Let ∇êα = êβ Aβ α . Then, as before
F β α = dAβ α + Aβ γ ∧ Aγ α . From (HC2),

Aβ α = −Aα β , F β α = −F α β (251)

Proposition C.19. The curvature form F β α for an orthonormal frame {êα }

is a (1, 1)-form.

Proof.
∇∇êα = DDêα + D∂ê¯ α + ∂Dê
¯ α = F β α êβ (252)
¯ α and ∂De
By (245), DDeα is (2, 0)-form-valued section while D∂e ¯ α are (1, 1)-
β
form-valued sections. Hence F α has no (0, 2) component. However, by
(251), F β α has no (2, 0) component either.

C.8 Complex geometry

Following the analysis and example, we can quickly develop the basic con-
cepts about complex geometry.

52
 Suppose the holomorphic tangent bundle T M + of a complex manifold M
has a Hermitian metric h. dimC M = m. In the local coordinate (z1 , ..., zm ),

∂ ∂
h( µ
, ν ) ≡ hµν̄ (253)
∂z ∂z
hµν̄ is a positive-definite Hermitian matrix. In another coordinate patch
(w1 , ..., w2 ), the Hermitian metric reads,

∂z λ ∂z ρ
 
∂ ∂ 0
h( µ , ) ≡ hµν̄ = hλρ̄ µ (254)
∂w ∂wν ∂w ∂wν

The inverse of hµν̄ is hλ̄ρ such that hλ̄ρ hρν̄ = δ λ̄ ν̄ .

Simply, we define the conjugate Hermitian metric h̄ on the anti-holomorphic-
bundle as,
∂ ∂
h̄( µ , ν ) ≡ hµ̄ν = hµν̄ (255)
∂ z̄ ∂ z̄
all the analysis on h̄ is identical to h.
h on T M + and h̄ on T M − naturally defined a complex-bilinear function
g on the whole tangent space. For X, Y ∈ (Tp M )C , we decompose them to
the holomorphic and anti-holomorphic parts, X = X H + X A , Y = Y H + Y A ,

g(X, Y ) ≡ h(X H , Y A ) + h̄(X A , Y H ) (256)

such that g(X, Y ) = g(X, Y ), g(X, Y ) = g(Y, X). Locally, in terms of the
one-forms, g is

g = hµν̄ dz µ ⊗ dz̄ ν + hµ̄ν dz̄ µ ⊗ dz ν = hµν̄ (dz µ ⊗ dz̄ ν + dz̄ ν ⊗ dz µ ) (257)

Restrict g to the real tangent space directly, we get a Riemann metric on M ,

as a real manifold.

ds2R = 2Re(hµν̄ )(dxµ dxν + dy µ dy ν ) + 2Im(hµν̄ )(dxµ dy ν + dy ν dxµ ) (258)

The metric-compatible holomorphic connection for T M + uniquely exists,

which reads,
     
∂ λ ∂ ∂
∇µ = Γµν · , ∇µ̄ =0 (259)
∂z ν ∂z λ ∂z ν

53
Simply, we can define the metric-compatible anti-holomorphic connection for
T M −,      
∂ λ̄ ∂ ∂
∇µ̄ = Γµ̄ν̄ · , ∇µ =0 (260)
∂z ν̄ ∂ z̄ λ ∂z ν̄
So the only non-vanishing components of ΓA λ λ̄
BC are Γµν and Γµ̄ν̄ .
The connection matrix Γλµν can easily calculated by theorem (C.17) as
Γ = (h−1 )T ∂hT ,
Γλµν = hγ̄λ ∂µ hν γ̄ , (261)
and similarly,
Γλ̄µ̄ν̄ = Γλµν . (262)
¯
The curvature tensor is calculated by the simple formula (C.18), R = ∂Γ,

Rλ ν ρ̄µ = ∂ρ̄ Γλµν (263)

which is a matrix-valued (1, 1)-form in ρ̄, µ. So Rλ ν ρ̄µ = −Rλ νµρ̄ . Similarly,

Rλ̄ ν̄ρµ̄ = Rλ ν ρ̄µ (264)

All the other components of RA BCD vanish.

In particular, if m = 1, then the metric matrix is a positive function h
and the curvature tensor is a (1, 1) form
¯ −1 ∂h = ∂∂
R = ∂h ¯ log(h) = −∂ ∂¯ log(h) (265)

The Ricci tensor is defined to be the trace over the matrix indices. From
(250)
X
R = Rµρ̄ dz µ ∧ dz̄ ρ = ¯
Rλ λµρ̄ dz µ ∧ dz̄ ρ = ∂∂(log det(h)) (266)
λ

Then by the definition of the first Chern classes,

i
c1 (T M + ) = R (267)
2π
which is a real and closed 2-form.
So far, the torsion-free condition is not imposed and actually in general,

Γλµν 6= Γλνµ , (268)

54
There is a particular class of complex manifold, Kähler manifolds, on which
the torsion-free condition automatically holds,
Define the (1, 1)-form associated with h,

ω = i · hµν̄ dz µ ∧ dz̄ ν . (269)

Note that
ω = −ihµ̄ν dz̄ µ ∧ dz ν = ihν µ̄ dz̄ ν ∧ dz̄ µ = ω (270)
so ω is a real form.

Definition C.20. A complex manifold equipped with the Hermitian metric

h is a Kähler manifold, if and only if dω = 0.

Theorem C.21. Locally, there exists a smooth function K such that

hµν̄ = ∂µ ∂¯ν K (271)

K is called the Kähler potential.

Theorem C.22. A Kähler manifold is torsion-free, Γλµν = Γλνµ

The advantage of Kähler manifolds is that they are largely similar to

algebraic varieties.

References
[1] M. Nakahara, Geometry, Topology and Physics. Taylor and Francis 2003.

[2] S. Kobayashi, K. Nomizu, Foundations of differential geometry I. Inter-

science Publishers 1963

[3] S. Kobayashi, K. Nomizu, Foundations of differential geometry II. In-

terscience Publishers 1967

[4] S.S. Chern, W. H. Chen, K. S. Lam Lectures on Differential Geometry

World Scientific Publishing Company 1999

[5] Raoul Bott, Loring W. Tu Differential Forms in Algebraic Topology

World Scientific Publishing Company 1999

55