(Lahiri) Lecture Notes On Differential Geometry PDF

c Amitabha Lahiri: Lecture Notes on Differential Geometry for Physicists 2011
Chapter 1
Topology
We start by defining a topological space.

• A topological space is a set S together with a collection O of
subsets called open sets such that the following are true:
i) the empty set ∅ and S are open, ∅, S ∈ O
ii) the intersection of a finite number of open sets is open; if

U1 , U2 ∈ O, then U1 ∩ U2 ∈ O
the union of any number of open sets is open, if Ui ∈ O, then

iii) S
Ui ∈ O irrespective of the range of i. 2
i
It is the pair {S, O} which is, precisely speaking, a topological
space, or a space with topology. But it is common to refer to S as
a topological space which has been given a topology by specifying
O.
Example: S = R, the real line, with the open sets being open
intervals ]a, b[ , i.e. the sets {x ∈ R | a < x < b} and their unions,
plus ∅ and R itself. Then (i) above is true by definition.
For two such open sets U1 = ]a1 , b1 [ and U2 = ]a2 , b2 [ , we can
suppose a1 < a2 . Then if b1 6 a2 , the intersection U1 ∩ U2 = ∅ ∈ O .
Otherwise U1 ∩ U2 = ]a2 , b1 [ which is an open interval and thus
U1 ∩ U2 ∈ O. So (ii) is true.
And (iii) is also true by definition. 2
Similarly Rn can be given a topology via open rectangles, i.e.

via the sets {(x1 , · · · , xn ) ∈ Rn | ai < xi < bi }. This is called the

standard or usual topology of Rn .
• The trivial topology on S consists of O = {∅, S}. 2
1
2 Chapter 1. Topology
• The discrete topology on a set S is defined by O = {A | A ⊂ S},

i.e., O consists of all subsets of S. 2
• A set A is closed if its complement in S, also written S\A or
as A{ , is open. 2
Closed rectangles in Rn are closed sets as are closed balls and
single point sets.
A set can be neither open nor closed, or both open and closed.
In a discrete topology, every set A ⊂ S is both open and closed,
whereas in a trivial topology, any set A 6= ∅ or S is neither open nor
closed.
The collection C of closed sets in a topological space S satisfy
the following:
i) the empty set ∅ and S are open, ∅, S ∈ C
ii) the union of a finite number of open sets is open; if A1 , A2 ∈ C ,

then A1 ∪ A2 ∈ C
iii) the intersection of any number of open sets is open, if Ai ∈ C ,

then Ai ∈ C irrespective of the range of i.
T
i
Closed sets can also be used to define a topology. Given a set S
with a collection C of subsets satisfying the above three properties of
closed sets, we can always define a topology, since the complements
of closed sets are open. (Exercise!)
• An open neighbourhood of a point P in a topological space
S is an open set containing P . A neighbourhood of P is a set
containing an open neighbourhood of P . Neighbourhoods can be
defined for sets as well in a similar fashion. 2
Examples: For a point x ∈ R, and for any > 0,
]x − , x + [ is an open neighbourhood of x,
[x − , x + [ is a neighbourhood of x,
{x − 6 y < ∞} is a neighbourhood of x,
[x, x + [ is not a neighbourhood of x. 2
• A topological space is Hausdorff if two distinct points have
disjoint neighbourhoods. 2
Topology is useful to us in defining continuity of maps.
• A map f : S1 → S2 is continuous if given any open set U ⊂ S2

its inverse image (or pre-image, what it is an image of) f −1 (U ) ⊂ S1

is open. 2
m n
When this definition is applied to functions from R to R , it is
3
the same as the usual − δ definition of continuity, which says that

• f : Rm → Rn is continuous at x0 if given > 0, we can always
find a δ > 0 such that |f (x) − f (x0 )| < whenever |x − x0 | < δ. 2
For the case of functions from a topological space to Rn , this
definition says that
• f : S → Rn is continuous at s0 ∈ S if given > 0, we can
always find an open neighbourhoood U of s0 such that |f (s)−f (s0 )| <
whenever s ∈ U. 2
• If a map f : S1 → S2 is one-to-one and onto, i.e. a bijection,
and both f and f −1 are continuous, f is called a homeomorphism
and we say that S1 and cs2 are homeomorphic. 2
Proposition: The composition of two continuous maps is a con-
tinuous map.
Proof: If f : S1 → S2 and g : S2 → S3 are continuous maps, and
U is some open set in S3 , then its pre-image g −1 (U ) is open in S2 .
So f −1 (g −1 (U )), which is the pre-image of that, is open in S1 . Thus
(g ◦ f )−1 (U ) = f −1 (g −1 (U )) is open in S1 . Thus g ◦ f is continuous.
2

Chapter 2
Manifolds
Now that we have the notions of open sets and continuity, we are
ready to define the fundamental object that will hold our attention
during this course.
• A manifold is a topological space which is locally like Rn . 2
That is, every point of a manifold has an open neighbourhood
with a one-to-one map onto some open set of Rn .
• More precisely, a topological space M is a smooth n-
dimensional manifold if the following are true:
i) We can cover the space with open sets Uα , i.e. every point of
M lies within some Uα .
ii) ∃ a map ϕα : Uα → Rn , where ϕα is one-to-one and onto
some open set of Rn . ϕα is continuous, ϕα −1 is continuous, i.e.
ϕα → Vα ∈ Rn is a homeomorphism for Vα .
(Uα , ϕα ) is called a chart (Uα is called the domain of the
chart). The collection of charts is called an atlas.
iii) In any intersection Uα ∩ Uβ , the maps ϕα ◦ ϕβ −1 , which are
called transition functions and take open sets of Rn to open
sets of Rn , i.e. ϕα ◦ ϕβ −1 : ϕβ (Uα ∩ Uβ ) → ϕα (Uα ∩ Uβ ), are
smooth maps. 2
• n is called the dimension of M. 2
We have defined smooth manifolds. A more general definition is
that of a C k manifold, in which the transition functions are C k , i.e. k

times differentiable. Smooth means k is large enough for the purpose

at hand. In practice, k is taken to be as large as necessary, up to
C ∞ . We get real analytic manifolds when the transition functions
4
5
are real analytic, i.e. have a Taylor expansion at each point, which
converges. Smoothness of a manifold is useful because then we can
say unambiguously if a function on the manifold is smooth as we will
see below.
• A complex analytic manifold is defined similarly by replac-
ing Rn with Cn and assuming the transitions functions ϕα ◦ ϕβ −1 to
be holomorphic (complex analytic). 2
• Given a chart (Uα , ϕα ) for a neighbourhood of some point P, the
image (x1 , · · · , xn ) ∈ Rn of P is called the coordinates of P in the
chart (Uα , ϕα ). A chart is also called a local coordinate system.2
In this language, a manifold is a space on which a local coordi-
nate system can be defined, and coordinate transformations between
different local coordinate systems are smooth. Often we will suppress
U and write only ϕ for a chart around some point in a manifold. We
will always mean a smooth manifold when we mention a manifold.
Examples: Rn (with the usual topology) is a manifold. 2
The typical example of a manifold is the sphere. Consider the
sphere S n as a subset of Rn+1 :
(x1 )2 + · · · + (xn+1 )2 = 1 (2.1)
It is not possible to cover the sphere by a single chart, but it is

possible to do so by two charts.1
For the two charts, we will construct what is called the stereo-
graphic projection. It is most convenient to draw this for a circle
in the plane, i.e. S 1 in R2 , for which the equatorial ‘plane’ is simply
an infinite straight line. Of course the construction works for any
S n . consider the ‘equatorial plane’ defined as the x1 = 0, i.e. the
set {(0, x2 , · · · , xn+1 )}, which is simply Rn when we ignore the first
zero. We will find homeomorphisms from open sets on S n to open
sets on this Rn . Let us start with the north pole N , defined as the
point (1, 0, · · · , 0).
We draw a straight line from N to any point on the sphere. If
that point is in the upper hemisphere (x1 > 0) the line is extended till
it hits the equatorial plane. The point where it hits the plane is the

1
The reason that it is not possible to cover the sphere with a single chart
is that the sphere is a compact space, and the image of a compact space un-
der a continuous map is compact. Since Rn is non-compact, there cannot be a
homeomorphism between S n and Rn .
6 Chapter 2. Manifolds
image of the point on the sphere which the line has passed through.
For points on the lower hemisphere, the line first passes through the
equatorial plane (image point) before reaching the sphere (source
point). Then using similarity of triangles we find (Exercise!) that
the coordinates on the equatorial plane Rn of the image of a point
on S n \{N } is given by
x2 xn+1

1 2 n+1

ϕN : x , x , · · · , x 7→ ,··· , . (2.2)
1 − x1 1 − x1
Similarly, the stereographic projection from the south pole is
ϕS : S n \{S} → Rn ,
x2 xn+1

1 2 n+1

x ,x ,··· ,x 7→ , · · · , . (2.3)
1 + x1 1 + x1
If we write
x2 xn+1

z= , · · · , , (2.4)
1 − x1 1 − x1
we find that
2 n+1 2
x2 1 − (x1 )2 1 + x1

2 x
|z| ≡ + · · · + = = (2.5)
1 − x1 1 − x1 (1 − x1 )2 1 − x1
The overlap between the two sets is the sphere without the poles.
Then the transition function between the two projections is
z
ϕS ◦ ϕN : Rn \{0} → Rn \{0}, z 7→ . (2.6)
|z|2
These are differentiable functions of z in Rn \{0}. This shows that
the sphere is an n-dimensional differentiable manifold. 2
• A Lie group is a group G which is also a smooth (real analytic
for the cases we will consider) manifold such that group composition
written as a map (x, y) 7→ xy −1 is smooth. 2
Another way of defining a Lie group is to start with an n-
parameter continuous group G which is a group that can be
parametrized by n (and only n) real continuous variables. n is called

the dimension of the group, n = dim G. (This is a different defi-

nition of the dimension. The parameters are global, but do not in
general form a global coordinate system.)
7
Then any element of the group can be written as g(a) where

a = (a1 , · · · , an ) . Since the composition of two elements of G must
be another element of G, we can write g(a)g(b) = g(φ(a, b)) where
φ = (φ1 , · · · , φn ) are n functions of a and b. Then for a Lie group,
the functions φ are smooth (real analytic) functions of a and b.
These definitions of a Lie group are equivalent, i.e. define the
same objects, if we are talking about finite dimensional Lie groups.
Further, it is sufficient to define them as smooth manifolds if we are
interested only in finite dimensions, because all such groups are also
real analytic manifolds. Apparently there is another definition of
a Lie group as a topological group (like n-parameter continuous
group, but without an a priori restriction on n, in which the compo-
sition map (x, y) 7→ xy −1 is continuous) in which it is always possible
to find an open neighbourhood of the identity which does not contain
a subgroup.
Any of these definitions makes a Lie group a smooth manifold,
an n-dimensional Lie group is an n-dimensional manifold. 2
The phase space of N particles is a 6N -dimensional manifold, 3N
coordinates and 3N momenta. 2
The Möbius strip is a 2-dimensional manifold. 2
The space of functions with some specified properties is of-
ten a manifold. For example, linear combinations of solutions of
Schrödinger equation which vanish outside some region form a man-
ifold. 2
Finite dimensional vector spaces are manifolds. 2
Infinite dimensional vector spaces with finite norm (e.g. Hilbert
spaces) are manifolds. 2
• A connected manifold cannot be written as the disjoint union
of open sets. Alternatively, the only subsets of a connected manifold
which are both open and closed are ∅ and the manifold itself. 2
SO(3), the group of rotations in three dimensions, is a 3-
dimensional connected manifold. O(3), the group of rotations plus
reflections in three dimensions, is also a 3-dimensional manifold,
but is not connected since it can be written as the disjoint union
SO(3)∪PSO(3) where P is reflection. 2
L↑+ , the group of proper (no space reflection) orthochronous (no

time reflection) Lorentz transformations, is a 6-dimensional con-

nected manifold. The full Lorentz group is a 6-dimensional manifold,
8 Chapter 2. Manifolds
not connected. 2
Rotations in three dimensions can be represented by 3 × 3 real
orthogonal matrices R satisfying RT R = I. Reflection is represented
by the matrix P = −I. The space of 3 × 3 real orthogonal matrices
is a connected manifold. 2
The space of all n× real non-singular matrices is called GL(n, R).
This is an n2 -dimensional Lie group and connected manifold. 2

Chapter 3
Tangent vectors
Vectors on a manifold are to be thought in terms tangents to the

manifold, which is a generalization of tangents to curves and sur-
faces, and will be defined shortly. But a tangent to a curve is like
the velocity of a particle at that point, which of course comes from
motion along the curve, which is its trajectory. And motion means
comparing things at nearby points along the trajectory. And com-
paring functions at nearby points leads to differentiation. So in order
to get to vectors, let us first start with the definitions of these things.
• A function f : M → R is differentiable at a point P ∈ M if
in a chart ϕ at P, the function f ◦ ϕ−1 : Rn → R is differentiable at
ϕ(P ). 2
This definition does not depend on the chart. If f ◦ ϕα −1 is
differentiable at ϕα (P ) in a chart (Uα , ϕα ) at P , the f ◦ ϕβ −1 is
differentiable at ϕβ (P ) for any chart (Uβ , ϕβ ) because
f ◦ ϕβ −1 = (f ◦ ϕα −1 ) ◦ (ϕα ◦ ϕβ −1 ) (3.1)
and the transition functions (ϕα ◦ ϕβ −1 ) are differentiable.

This should be thought of as a special case of functions from one
manifold to another. Consider two manifolds M and N of dimension
m and n, and a mapping f : M → N , P 7→ Q. Consider local charts
(U, ϕ) around P and (W, ψ) around Q. Then ψ ◦ f ◦ ϕ−1 is a map
from Rm → Rn and represents f in these local charts.
• f is differentiable at P if ψ ◦ f ◦ ϕ−1 is differentiable at ϕ(P ).

In other words, f is differentiable at P if the coordinates y i = f i (xµ )

of Q are differentiable functions of the coordinates xµ of P . 2
• −1
If f is a bijection (i.e. one-to-one and onto) and f and f are
9
10 Chapter 3. Tangent vectors
both differentiable, we say that f is a diffeomorphism and that M

and N are diffeomorphic. 2
k
In all of these definitions, differentiable can be replaced by C or
smooth.
• Two Lie groups are isomorphic if there is a diffeomorphism
between them which is also a group homomorphism. 2
• A curve in a manifold M is a map γ of a closed interval R
to M. (This definition can be given also when M is a topological
space.) 2
We will take this interval to be I = [0, 1] ⊂ R. Then a curve is a
map γ : I → M. If γ(0) = P and γ(1) = P 0 , for some γ, we say that
γ joins P and P 0 .
• A manifold M is connected (actually arcwise connected)1 if
any two points in it can be joined by a continuous curve in M. 2
As for any map, a curve γ is called smooth iff its image in a chart
is smooth in Rn , i.e., iff ϕ ◦ γ : I → Rn is smooth in Rn .
Note that the definition of a curve implies that it is parametrized.
So the same collection of points in M can stand for two different
curves if they have different parametrizations.
We are now ready to define tangent vectors and the tangent space
to a manifold. There are different ways of defining tangent vectors.
i) Coordinate approach: Vectors are defined to be objects satis-
fying certain transformation rules under a change of chart, i.e.
coordinate transformation, (Uα , ϕα ) → (Uβ , ϕβ ).
ii) Derivation approach: A vector is defined as a derivation of func-

tions on the manifold. This is thinking of a vector as defining
a “directional derivative”.
iii) Curves approach: A vector tangent to a manifold is tangent to

a curve on the manifold.
The approaches are equivalent in the sense that they end up defin-
ing the same objects and the same space. We will follow the third
approach, or perhaps a mix of the second and the third approaches.
Later we will briefly look at the derivation approach more carefully
and compare it with the way we have defined tangent vectors.

Consider a smooth function f : M → R. Given a curve γ :

I → M, the map f ◦ γ : I → R is well-defined, with a well-defined
1
It can be shown that an arcwise connected space is connected.
11
df
derivative. The rate of change of f along γ is written as .
dt
Suppose another curve another curve µ(s) meets γ(t) at some
point P , where s = s0 and t = t0 , such that
d d
(f ◦ γ) = (f ◦ µ) ∀f ∈ C ∞ (M) (3.2)

dt P ds P
That is, we are considering a situation where two curves are tangent
to each other in geometric and parametric sense. Let us introduce a
convenient notation. In any chart ϕ containing the point P, let us
write ϕ(P ) = (x1 , · · · , xn ). Let us write f ◦ γ = (f ◦ ϕ−1 ) ◦ (ϕ ◦ γ),
so that the maps are
f ◦ ϕ−1 : Rn → R, x 7→ f (x) or f (xi ) (3.3)
n i
ϕ◦γ : I →R , t 7→ {x (γ(t))}. (3.4)
The last are the coordinates of the curve in Rn .
Using the chain rule for differentiation, we find
d d ∂f dxi (γ(t))
(f ◦ γ) = f (x(γ(t))) = . (3.5)
dt dt ∂xi dt
Similarly, for the curve µ we find
d d ∂f dxi (µ(s))
(f ◦ µ) = f (x(µ(s))) = . (3.6)
ds ds ∂xi ds
Since f is arbitrary, we can say that two curves γ, µ have the same
tangent vector at the point P ∈ M (where t = t0 and s = s0 ) iff
dxi (γ(t)) dxi (µ(s))

= . (3.7)
dt
t=t0 ds
s=s0
We can say that these numbers completely determine the rate of
change of any function along the curve γ or µ at P. So we can define
the tangent to the curve.
• The tangent vector to a curve γ at a point P on it is defined
as the map
d
γ̇P : C ∞ (M) → R, f 7→ γ̇P (f ) ≡ (f ◦ γ)|P . (3.8)
dt
As we have already seen, in a chart with coordinates {xi } we can

write using chain rule

dxi (γ(t)) ∂f

γ̇P (f ) = (3.9)
dt ∂xi ϕ(P )
12 Chapter 3. Tangent vectors
dxi (γ(t))

The numbers are thus the components of γ̇P . We will
dt
ϕ(P )
often write a tangent vector at P as vP without referring to the curve
it is tangent to.
We note here that there is another description of tangent vectors
based on curves. Let us write γ ∼ µ if γ and µ are tangent to each
other at the point P . It is easy to see, using Eq. (3.7) for example,
that this relation ∼ is transitive, reflexive, and symmetric. In other
words, ∼ is an equivalence relation, for which the equivalence class
[γ] contains all curves tangent to γ (as well as to one another) at P .
• A tangent vector at P ∈ M is an equivalence class of curves
under the above equivalence relation. 2
The earlier definition is related to this by saying that if a vector
vP is tangent to some curve γ at P , i.e. if vP = γ̇P , we can write
vP = [γ].

Chapter 4
Tangent Space
• The set of all tangent vectors (to all curves) at some point P ∈
M is the tangent space TP M at P. 2
Proposition: TP M is a vector space with the same dimension-
ality n as the manifold M.
Proof: We need to show that TP M is a vector space, i.e.
XP + YP ∈ TP M , (4.1)
aXP ∈ TP M , (4.2)
∀XP , YP ∈ TP M, a ∈ R.
That is, given curves γ, µ passing through P such that XP =
γ̇P , YP = µ̇P , we need a curve λ passing through P such that
λ̇P (f ) = XP (f ) + YP (f )∀f ∈ C ∞ (M).
Define λ : I → Rn in some chart ϕ around P by λ = ϕ ◦ γ + ϕ ◦
µ − ϕ(P ). Then λ is a curve in Rn , and
λ=ϕ◦λ:I →M (4.3)
is a curve with the desired property. 2
Note: we cannot define λ = γ + µ − P because addition does not
make sense on the right hand side.
The proof of the other part works similarly. (Exercise!)
To see that TP M has n basis vectors, we consider a chart ϕ with
coordinates xi . Then take n curves λk such that

ϕ ◦ λk (t) = x1 (P ), · · · , xk (P ) + t, · · · , xn (P ) ,

(4.4)
i.e., only the k-th coordinate varies along t. So λk is like the axis of
the k-th coordinate (but only in some open neighbourhood of P ).
13
14 Chapter 4. Tangent Space

∂
Now denote the tangent vector to λk at P by , i.e.,
∂xk P

∂ d
f = λ̇ (f ) = (f ◦ λ ) . (4.5)

k k
∂xk P dt

P
P
This notation makes sense when we remember Eq. (3.9). Using it we

can write

∂f
λ̇k (f ) = ∀f ∈ C ∞ (M). (4.6)

P ∂xk P

∂
Note that is notation. We should understand this as
∂xk P

∂ ∂ −1
∂f
f= f ◦ϕ ≡ (4.7)
∂xk P ∂xk ϕ(P ) ∂xk ϕ(P )

∂
in a chart around P . The are defined only when this chart
∂xk P
is given, but these are vectors on the manifold at P , not on Rn .
Let us now show that the tangent space at P has λ̇k |P as a basis.
Take any vector vP ∈ TP M , which is the tangent vector to some
curve γ at P . (We may sometimes refer to P as γ(0) or as t = 0.)
Then

d
vP (f ) = (f ◦ γ) (4.8)
dt t=0

d −1

= ((f ◦ ϕ ) ◦ (ϕ ◦ γ)) . (4.9)

dt ϕ(P )
t=0
Note that ϕ ◦ γ : I → Rn , t 7→ (x1 (γ(t)), · · · , xn (γ(t))) are the coor-

dinates of the curve γ, so we can use the chain rule of differentiation
to write

∂ −1
d i
vP (f ) = i
(f ◦ ϕ ) (x ◦ γ) (4.10)
∂x ϕ(P ) dt t=0

∂ −1
vP (xi ) .

= i
(f ◦ ϕ ) (4.11)
∂x ϕ(P )

The first factor is exactly as shown in Eq. (4.7), so we can write

∂
vP (f ) = f v (xi ) ∀f ∈ C ∞ (M) (4.12)
∂xk P P
15
i.e., we can write

i ∂
vP = vP ∀vP ∈ TP M (4.13)
∂xk P
where vPi = vP (xi ). Thus the vectors ∂x∂ k P span TP M . These are

to be thought of as tangents to the coordinate curves inϕ. These

can be shown to be linearly independent as well, so ∂x∂ k P form a
basis of TP M and vPi are the components of vP in that basis.
∂
n The o∂xk P are called coordinate basis vectors and the set
∂

∂xk P
is called the coordinate basis.
It can be shown quite easily that for any smooth (actually C 1 )
function f a vector vP defines a derivation f 7→ vP (f ) , i.e., satisfies
linearity and Leibniz rule,
vP (f + αg) = vP (f ) + αvP (g) (4.14)

vP (f g) = vP (f )g(P ) + f (P )vP (g) (4.15)
1
∀f, g ∈ C (M) and α ∈ R (4.16)

Chapter 5
Dual space
• The dual space TP∗ M of TP M is the space of linear mappings

ω : TP M → R. 2
We will write the action of ω on vP ∈ TP M as ω(vP ) or sometimes
as hω | vP i .
Linearity of the mapping ω means
ω(uP + avP ) = ω(uP ) + aω(vP ) , (5.1)

∀uP , vP ∈ TP M and a ∈ R .
The dual space is a vector space under the operations of vector ad-
dition and scalar multiplication defined by
a1 ω1 + a2 ω2 : vP 7→ a1 ω1 (vP ) + a2 ω2 (vP ) . (5.2)
• The elements of TP∗ M are called dual vectors, covectors,

cotangent vectors etc. 2
A dual space can be defined for any vector space V as the space of
linear mappings V → R (or V → C if V is a complex vector space).
Example:
Vector Dual vector

column vectors row vector
kets |ψi bras hφ|
functions linear functionals, etc.2

• Given a function on a manifold f : M → R , every vector at

P produces a number, vP (f ) ∈ R ∀vP ∈ TP M . Thus f defines a
16
17
covector df , given by df (vP ) = vP (f ) called the differential or

gradient of f . 2
Since vP is linear, so is df ,
df (vP + awP ) = (vP + awP )(f )
= vP (f ) + awP (f ) (5.3)
∀vP , wP ∈ TP M, a ∈ R .
Thus df ∈ TP∗ M .
Proposition: TP∗ M is also n-dimensional.
Proof: Consider a chart ϕ with coordinate functions xi . Each xi
is a smooth function xi : M → R . then the differentials dxi satisfy

i ∂ ∂ i ∂ i −1
= δji . (5.4)

dx = (x ) = x ◦ϕ
∂xj P ∂xj P ∂xj ϕ(P )
The differentials dxi are covectors, as we already know. So we

have constructed n covectors in TP∗ M . Next consider a linear combi-
nation of these covectors, ω = ωi dxi . If this vanishes, it must vanish
on every one of the basis vectors. In other words,

∂
ω=0 ⇒ ω =0
∂xj P

i ∂
⇒ ωi dx =0
∂xj P
⇒ ωi δji = 0 i.e. ωj = 0 . (5.5)
So the dxi are linearly independent.
Finally, given any covector ω , consider the covector λ = ω −
∂ i

ω ∂xi P dx . Then letting this act on a coordinate basis vector, we
get

∂
λ
∂xj P

∂ ∂ i ∂
=ω −ω dx
∂xj P ∂xi P ∂xj P

∂ ∂
=ω −ω δ i = 0∀j (5.6)
j
∂x P ∂xi P j

∂
So λ vanishes on all vectors, since the form a basis. Thus
∂xj P
the dxi span TP∗ M , so TP∗ M is n-dimensional.
18 Chapter 5. Dual space
Also, as we have just seen, any covector ω ∈ TP∗ M can be written

as

i ∂
ω = ωi dx where ωi = ω , (5.7)
∂xi P
so in particular for ω = df , we get

∂ ∂f
ωi ≡ (df )i = df = (5.8)
∂xi P ∂xi ϕ(P )
This justifies the name gradient.

It is straightforward to calculate the effect of switching to another
overlapping chart, i.e. a coordinate transformation. In a new chart
ϕ0 where the coordinates are y i (and the transition functions are thus
y i (x)) we can use Eq. (5.8) to write the gradient of y i as
i
i ∂y
dy = dxj (5.9)
∂xj P
This is the result of coordinate transformations on a basis of covec-
tors.
∂
Since i
is the dual basis in TP M to {dxi }, in order
∂x
P
∂
for to be the dual basis to {dy i } we must have
∂y i P
j
∂ ∂x ∂
= (5.10)
∂y i P ∂y i P ∂xj P
These formulae can be generalized to arbitrary bases.
Given a vector v, it is not meaningful to talk about its dual, but
given a basis {ea }, we can define its dual basis {ω a } by ω a (eb ) = δba .
We can make a change of bases by a linear transformation,
ω a 7→ ω 0a = Aab ω b , ea 7→ e0a = (A−1 )ba eb , (5.11)
with A a non-singular matrix, so that ω 0a (e0b ) = δba .

Given a 1-form λ we can write it in both bases,

λ = λa ω a = λ0a ω 0a = λ0a Aab ω a , (5.12)
from which it follows that λ0a = (A−1 )ba λb .

19
Similarly, if v is a vector, we can write
v = v a ea = v 0a e0a = v 0a (A−1 )ba eb , (5.13)
and it follows that v a = Aab v b .

• Quantities which transform like λa are called covariant, while
those transforming like v a are called contravariant. 2

Chapter 6
Vector fields
• Consider the (disjoint) union of tangent spaces at all points,

[
TM = TP M . (6.1)
P ∈M
This is called the tangent bundle of M. 2

• A vector field v chooses an element of TP M for every P , i.e.
v : P 7→ v(P ) ≡ vP ∈ TP M. 2
We will often write v(f )|P = vP (f ).
Given a chart, v has components v i in the chart,

i ∂
vP = v , (v i )P = vP (xi ) . (6.2)
∂xi P
• The vector field v is smooth if the functions v i = v(xi ) are
smooth for any chart (and thus for all charts). 2
• ∗
A rule that selects a covector from TP M for each P is called a
one−form (often written as a 1−form). 2
• Given a smooth vector field v (actually C 1 is sufficient) we can
define an integral curve of v, which is a curve γ in M such that
γ̇(t)|P = vP at every P ∈ γ. (One curve need not pass through all
P ∈ M.) 2
Suppose γ is an integral curve of a given vector field v, with
γ(0) = P. Then in a chart containing P , we can write
d i
γ̇(t) = v ⇒ x (γ(t)) = v i (x(t)) , (6.3)

dt
with initial condition xi (0) = xi |P . This is a set of ordinary first order
differential equations. If v i are smooth, the theory of differential
20
21
equations guarantees, at least for small t (i.e. locally), the existence

of exactly one solution. The uniqueness of these solutions implies
that the integral curves of a vector field do not cross.
One use of integral curves is that they can be thought of as co-
ordinate lines. Given a smooth vector field v such that v|P 6= 0, it
is possible to define a coordinate system {xi } in a neighbourhood
∂
around P such that v = .
∂xi
• A vector field v is said to be complete if at every point P ∈ M
the integral curve the integral curve γ(t) of v passing through P can
be extended to all t ∈ R . 2
The tangent bundle T M is a product manifold, i.e., a point in
T M is an ordered pair (P, v) where P ∈ M and v ∈ TP M. The topo-
logical structure and differential structure are given appropriately.
• The map π : T M → M, (P, v) 7→ P (where v ∈ TP M) is called
the canonical projection (or simply projection). 2
• For each P ∈ M, the pre-image π −1 (P ) is TP M. It is called the
fiber over P . Then a vector field can be thought of as a section of
the tangent bundle. 2
Given a smooth vector field v, we can define an integral curve γ
through any point P by γ̇(t) = v , i.e.,
d i
x (γ(t)) = v i (γ(t)) ≡ v(xi (γ(t))) , (6.4)
dt
γ(0) = P . (6.5)
We could also choose γ(t0 ) = P.

Then in any neighbourhood U of P we also have γQ , the integral
curve through Q. So we can define a map φ : I × U → M given by
φ(t, Q) = γQ (t) where γQ (t) satisfies
d i
x (γQ (t)) = v(xi γQ (t)) ,

(6.6)
dt
γQ (0) = Q . (6.7)
• This φ defines a map φt : U → M at each t by φt (Q) = φ(t, Q) =

γQ (t) , i.e. for given t, φt takes a point by a parameter distance t along

the curve γQ (t). This φt is called the local flow of v. 2

The local flow has the following properties:
i) φ0 is the identity map of U ;
22 Chapter 6. Vector fields
ii) φs ◦ φt = φs+t for all s, t, s + t ∈ U ;
iii) each flow is a diffeomorphism with φt −1 = φ−t .

The first property is obvious, while the second property follows
from the uniqueness of integral curves, i.e. of solutions to first order
differential equations. Then the integral curve passing through the
point γQ (s) is the same as the integral curve passing through Q, so
that moving a parameter distance t from γQ (s) finds the same point
on M as by moving a parameter distance s + t from γQ (0) ≡ Q .
A vector field can also be thought of as a map from the space
of differentiable functions to itself v : C ∞ (M) → C ∞ (M), f 7→
v(f ) , with v(f ) : M → R, P 7→ vP (f ) . Often v(f ) is called the Lie
derivative of f along v and denoted £v f .
The map v : f 7→ v(f ) has the following properties:
v(f + αg) = v(f ) + αv(g) (6.8)

v(f g) = f v(g) + v(f )g (6.9)
∞
∀f, g ∈ C (M), α∈R
The set of all (real) vector fields V (M) on a manifold M has the
structure of a (real) vector space under vector addition defined by
(u + αv)(f ) = u(f ) + αv(f ), u, v ∈ V (M), α ∈ R.

(6.10)
It is possible to replace α by some function on C ∞ (M). If u, v

are vector fields on M and α is now a smooth function on M, define
u + αv by
(u + αv)P (f ) = uP (f ) + α(P )vP (f ) ∀f ∈ C ∞ (M), P ∈ M.

(6.11)
This looks like a vector space but actually it is what is called a

module.
• A ring R is a set or space with addition and multiplication
defined on it, satisfying (xy)z = x(yz) , x(y + z) = xy + xz , (x +
y)z = xz + yz , and two special elements 0 and 1, the additive and

multiplicative identity elements, 0 + x = x + 0 = x , 1x = x1 =

x . A module X is an Abelian group under addition, with scalar
multiplication by elements of a ring defined on it.
23
• A module becomes a vector space when this ring is a commu

tative division ring, i.e. when the ring multiplication is commu-
tative, xy = yx, and an inverse exists for every element except 0.
Given a smooth function α, in general α−1 ∈ / C ∞ (M), so the space
of vector fields on M is in general a module, not a vector space.
Given a vector field v, in an open neighbourhood of some P ∈ M
and in a chart, and for any f ∈ C ∞ (M) , we have

i ∂f
v(f ) = vP (f ) = vP , where vPi = vP (xi ) . (6.12)

∂x i
P
P
Thus we can write

∂
v = vi with v i = v(xi ) , (6.13)
∂xi
as an obvious generalization of vector space expansion to the module
V (M).
∂
The v i are now the components of the vector field v, and ∂x i are
now vector fields, which we will call the coordinate vector fields.
Note that this is correct only in some open neighbourhood on which a
chart can be defined. In particular, it may not be possible in general
to define the coordinate vector fields globally, i.e. everywhere on M,
and thus the components v i may not be defined globally either.

Chapter 7
Pull back and push forward
Two important concepts are those of pull back (or pull-back or pull-
back) and push forward (or push-forward or pushforward) of maps
between manifolds.
• Given manifolds M1 , M2 , M3 and maps f : M1 → M2 , g :
M2 → M3 , the pullback of g under f is the map f ∗ g : M1 → M3
defined by
f ∗g = g ◦ f . (7.1)
2 So in particular, if M1 and M2 are two manifolds with a map

f : M1 → M2 and g : M2 → R is a function on M2 , the pullback
of g under f is a function on M1 ,
f ∗g = g ◦ f . (7.2)
While this looks utterly trivial at this point, this concept will become
increasingly useful later on.
• Given two manifolds M1 and M2 with a smooth map f : M1 →
M2 , P 7→ Q the pushforward of a vector v ∈ TP M1 is a vector
f∗ v ∈ TQ M2 defined by
f∗ v(g) = v(g ◦ f ) (7.3)
for all smooth functions g : M2 → R . 2

Thus we can write
f∗ v(g) = v(f ∗ g) . (7.4)
24
25
The pushforward is linear,
f∗ (v1 + v2 ) = f∗ v1 + f∗ v2 (7.5)
f∗ (λv) = λf∗ v . (7.6)
And if M1 , M2 , M3 are manifolds with maps f : M1 → M2 , g :

M2 → M3 , it follows that
(g ◦ f )∗ = g∗ f∗ , i.e.
(g ◦ f )∗ v = g∗ f∗ v ∀v ∈ TP M1 . (7.7)
Remember that we can think of a vector v as an equivalence class of

curves [γ]. The pushforward of an equivalence class of curves is
f∗ v = f∗ [γ] = [f ◦ γ] (7.8)
Note that for this pushforward to be defined, we do not need the

original maps to be 1-1 or onto. In particular, the two manifolds may
have different dimensions.
Suppose M1 and M2 are two manifolds with dimension m and n
respectively. So in the respective tangent spaces TP M1 and TQ M2
are also of dimension m and n respectively. So for a map f : M1 →
M2 , P 7→ Q , the pushforward f∗ will not have an inverse if m 6= n .
Let us find the components of the pushforward f∗ v in terms of
the components of v for any vector v. Let us in fact consider, given
charts ϕ : P 7→ (x1 , · · · , xm ) , ψ : Q 7→ (y 1 , · · · , y n ) the pushforward
of the basis vectors.
∂ ∂

For the basis vector ∂x i P , we want the pushforward f∗ ∂xi P ,

∂
which is a vector in TQ M2 , so we can expand it in the basis ∂y i
,
Q
µ
∂ ∂ ∂
f∗ = f∗ (7.9)
∂xi P ∂xi P ∂y µ Q
In any coordinate basis, the components of a vector are given by the
action of the vector on the coordinates as in Chap. 4,
vPµ = vP (y µ ) (7.10)

Thus we can write

µ
∂ ∂
f∗ = f∗ (y µ ) (7.11)
∂xi P ∂xi P
26 Chapter 7. Pull back and push forward
But
f∗ v(g) = v(g ◦ f ) , (7.12)
so

∂ µ ∂
f∗ (y ) = (y µ ◦ f ) . (7.13)
∂xi P ∂xi P
But y µ ◦f are the coordinate functions of the map f , i.e., coordinates

around the point f (P ) = Q . So we can write y µ ◦ f as y µ (x) , which
is what we understand by this. Thus
µ
∂y µ (x)

∂ ∂ µ
f∗ = (y ◦ f ) = . (7.14)
∂xi P ∂xi P ∂xi P
Because we are talking about derivatives of coordinates, these are
actually done in charts around P and Q = f (P ) , so the chart maps
are hidden in this equation.
• The right hand side is called the Jacobian matrix (of y µ (x) =
y µ ◦ f with respect to xi ). Note that since m and n may be unequal,
this matrix need not be invertible and a determinant may not be
defined for it. 2
For the basis vectors, we can then write
∂y µ (x)

∂ ∂
f∗ = (7.15)
∂xi P ∂xi P ∂y µ f (P )
Since f∗ is linear, we can use this to find the components of (f∗ v)Q
for any vector vP ,

i ∂
f∗ vP = f∗ vP
∂xi P

i ∂
= vP f∗
∂xi P
µ

i ∂y (x) ∂

= vP (7.16)
∂xi P ∂y µ f (P )
µ

µ i ∂y (x)

⇒ (f∗ vP ) = vP . (7.17)

i
∂x P
Note that since f∗ is linear, we know that the components of f∗ v
should be linear combinations of the components of v , so we can
27
already guess that (f∗ vP )µ = Aµi vPi for some matrix Aµi . The matrix
is made of first derivatives because vectors are first derivatives.
Another example of the pushforward map is the following. Re-
member that tangent vectors are derivatives along curves. Suppose
vP ∈ TP M is the derivative along γ . Since γ : I → M is a map,
we can consider pushforwards under γ , of derivatives on I . Thus for
γ : I → M , t 7→ γ(t) = P , and for some g : M → R ,

d d
γ∗ g = (g ◦ γ)|t=0
dt t=0 dt
= γ̇P (g)|t=0 = vP (g) , (7.18)
so
d
γ∗ = vP (7.19)
dt t=0
• We can use this to give another definition of integral curves.
Suppose we have a vector field v on M . Then the integral curve of v
passing through P ∈ M is a curve γ : t 7→ γ(t) such that γ(0) = P
and

d
γ∗ = v|γ(t) (7.20)
dt t
for all t in some interval containing P . 2
Even though in order to define the pushforward of a vector v
under a map f , we do not need f to be invertible, the pushforward
of a vector field can be defined only if f is both one-to-one and onto.
If f is not one-to-one, different points P and P 0 may have the
same image, f (P ) = Q = f (P 0 ) . Then for the same vector field v we
must have
f∗ v|Q = f∗ (vP ) = f∗ (vP 0 ) , (7.21)
which may not be true. And if f : M → N is not onto, f∗ v will be

meaningless outside some region f (M) , so f∗ v will not be a vector
field on N .
If f is one-to-one and onto, it is a diffeomorphism, in which case
vector fields can be pushed forward, by the rule

(f∗ v)f (P ) = f∗ (vP ) . (7.22)

Chapter 8
Lie brackets
A vector field v is a linear map C ∞ (M) → C ∞ (M) since it is basi-

cally a derivation at each point, v : f 7→ v(f ) . In other words, given
a smooth function f , v(f ) is a smooth function on M . Suppose we
consider two vector fields u , v . Then u(v(f )) is also a smooth func-
tion, linear in f . But is uv ≡ u ◦ v a vector field? To find out, we
consider
u(v(f g)) = u(f v(g) + v(f )g)

= u(f )v(g) + f u(v(g)) + u(v(f ))g + v(f )u(g) . (8.1)
We reorder the terms to write this as
uv(f g) = f uv(g) + uv(f )g + u(f )v(g) + v(f )u(g) , (8.2)
so Leibniz rule is not satisfied by uv . But if we also consider the

combination vu , we get
vu(f g) = f (vu(g) + vu(f )g + v(f )u(g) + u(f )v(g) . (8.3)
Thus
(uv − vu)(f g) = f (uv − vu)(g) + (uv − vu)(f )g , (8.4)
which means that the combination
[u , v] := uv − vu (8.5)

is a vector field on M , with the product uv signifying successive

operation on any smooth function on M .
28
29
• This combination is called the commutator or Lie bracket of

the vector fields u and v . 2
In any chart around the point P ∈ M , we can write a vector
field in local coordinates
∂f
v(f ) = v i , (8.6)
∂xi
so that

∂j i ∂f
u(v(f )) = u v
∂xj ∂xi
∂v i ∂f ∂2f
= uj j i + uj v i j i ,
∂x ∂x ∂x x
i
∂u ∂f ∂2f
v(u(f )) = v j j i + uj v i j i . (8.7)
∂x ∂x ∂x x
Subtracting, we get
∂v i ∂f i
j ∂u ∂f
u(v(f )) − v(u(f )) = uj − v , (8.8)
∂xj ∂xi ∂xj ∂xi
from which we can read off the components of the commutator,
∂v i j ∂u
i
[u , v]i = uj − v (8.9)
∂xj ∂xj
• The commutator is antisymmetric, [u , v] = −[v , u] , and satisfies
the Jacobi identity
[[u , v] , w] + [[v , w] , u] + [[w , u] , v] = 0 . (8.10)
The commutator is useful

for the following reason: Once we have a
∂
chart, we can use as a basis for vector fields in a neighbour-
∂xi
hood.
Any set of n linearly independent vector fields may be chosen as
a basis, but they need not form a coordinate system. In a coordinate
system,
∂ ∂
, = 0, (8.11)
∂xi ∂xj

because partial derivatives commute. So n vector fields will form a

coordinate system only if they commute, i.e., have vanishing commu-
tators with one another. Then the coordinate lines are the integral
30 Chapter 8. Lie brackets
curves of the vector fields. For analytic manifolds, this condition is

sufficient as well.
A simple example is the polar coordinate system in R2 . The unit
vectors are
er = ex cos θ + ey sin θ
eθ = −ex sin θ + ey cos θ , (8.12)
∂ ∂
with ex = and ey = being the Cartesian coordinate basis
∂x ∂y
vectors, and
x y p
cos θ = , sin θ = , r = x2 + y 2 (8.13)
r r
Using these expressions, it is easy to show that [er , eθ ] 6= 0 , so
{er , eθ } do not form a coordinate basis.

Chapter 9
Lie algebra
• An (real) algebra is a (real) vector space equipped with a bi-

linear operation (product) under which the algebra is closed, i.e., for
an algebra A
i) x • y ∈ A ∀x, y ∈ A
ii) (λx + µy) • z = λx • z + µy • z

x • (λy + µz) = λx • y + µx • z ∀x, y, z ∈ A , λ, µ ∈ R .
If λ, µ are complex numbers and A is a complex vector space, we
get a complex algebra. 2
• A Lie algebra is an algebra in which the operation is
i) antisymmetric, x • y = −y • x , and
ii) satisfies the Jacobi identity ,
(x • y) • z + (y • z) • x + (z • x) • y = 0 . (9.1)
2
The Jacobi identity is not really an identity — it does not hold
for an arbitrary algebra — but it must be satisfied by an algebra for
it to be called a Lie algebra.
Example:
i) The space Mn = {all n × n matrices} under matrix multipli-
cation, A • B = AB . This is an associative algebra since

matrix multiplication is associative, (AB)C = A(BC) .
ii) The same space Mn of all n × n matrices as above, but now
31
32 Chapter 9. Lie algebra
with matrix commutator as the product,
A • B = [A , B] = AB − BA . (9.2)
This product is antisymmetric and satisfies Jacobi identity, so

Mn with this product is a Lie algebra.
iii) The angular momentum algebra in quantum mechanics.

If Li are the angular momentum operators with [Li , Lj ] =
iijk Lk , we can write the elements of this algebra as
( )
X
L= a= ζi Li |ζi ∈ C (9.3)
i
P P
If a = ai Li and b = bi Li , their product is
X X
a • b ≡ [a , b] = ai bj [Li , Lj ] = i ijk ai bj Lk . (9.4)
This is a Lie algebra because it [a , a] = 0 and the Jacobi iden-

tity is satisfied.
iv) The Poisson bracket algebra of a classical dynamical sys-

tem consists of functions on the phase space, with the product
defined by the Poisson bracket,
f • g = [f , g]P.B. . (9.5)
This is a Lie algebra. As a vector space it is infinite-

dimensional.
v) Vector fields on a manifold form a real Lie algebra under the

commutator bracket, since the Jacobi identity is a genuine iden-
tity, i.e. automatically satisfied, as we have seen in the previ-
ous chapter. This algebra is infinite-dimensional. (It can be
thought of as the Lie algebra of the group of diffeomorphisms,
Dif f (M)) .

Chapter 10
Local flows
We met local flows and integral curves in Chapter 6. Given a vector

field v , write its local flow as φt .
• The collection φt for t < (for some > 0, or alternatively for
t < 1) is a one−parameter group of local diffeomorphisms . 2
Consider the vector field in a neighbourhood U of a point Q ∈
M . Since φt : U → M, Q 7→ γQ (t) is local diffeomorphism , i.e.
diffeomorphism for sufficiently small values of t , we can use φt to push
forward vector fields. At some point P we have the curve φt (P ) . We
push forward a vector field at t = to t = 0 and compare with the
vector field at t = 0 .
We recall that for a map ϕ : M1 → M2 the pullback of a function
f ∈ C ∞ (M2 ) is defined as
ϕ∗ f = f ◦ ϕ : M1 → R , (10.1)
and ϕ∗ f ∈ C ∞ (M1 ) if ϕ is C ∞ .
The pushforward of a vector vP is defined by
ϕ∗ vP (f ) = vP (f ◦ ϕ) = vP (ϕ∗ f ) (10.2)
vP ∈ TP M1 , ϕ∗ vP ∈ Tϕ(P ) M2 . (10.3)
If ϕ is a diffeomorphism, we can define the pushforward of a vector

field v by

ϕ∗ v(f )|ϕ(P ) = v (f ◦ ϕ)|P

i.e. ϕ∗ v(f )|Q = v (f ◦ ϕ)|ϕ−1 Q
= v (ϕ∗ f )|ϕ−1 Q . (10.4)
33
34 Chapter 10. Local flows
We can rewrite this definition in several different ways,
(ϕ∗ v)(f ) = v (f ◦ ϕ) ◦ ϕ−1

= (ϕ−1 )∗ (v (f ◦ ϕ))
= (ϕ−1 )∗ (v (ϕ∗ f )) . (10.5)
• If ϕ : M1 → M2 is not invertible, ϕ∗ v is not a vector field on

M2 . If ϕ−1 exists but is not differentiable, ϕ∗ v is not differentiable.
But there are some ϕ and some v such that ϕ∗ v is a differentiable
vector field, even if ϕ is not invertible or ϕ−1 is not differentiable.
Then v and ϕ∗ v are said to be ϕ−related. 2
Proposition: Given a diffeomorphism ϕ : M1 → M2 (say both
C ∞ manifolds) the pushforward ϕ∗ is an isomorphism on the Lie
algebra of vector fields, i.e.
ϕ∗ [u , v] = [ϕ∗ u , ϕ∗ v] . (10.6)
Proof:
ϕ∗ [u , v](f ) = [u , v] (f ◦ ϕ) ◦ ϕ−1
= u (v (f ◦ ϕ)) ◦ ϕ−1 − u ↔ v , (10.7)
while [ϕ∗ u , ϕ∗ v](f ) = ϕ∗ u (ϕ∗ v (f )) − u ↔ v
= u (ϕ∗ v (f ) ◦ ϕ) ◦ ϕ−1 − u ↔ v
= u v (f ◦ ϕ) ◦ ϕ−1 ◦ ϕ ◦ ϕ−1 − u ↔ v

= u (v (f ◦ ϕ)) ◦ ϕ−1 − u ↔ v . (10.8)
2
• A vector field v is said to be invariant under a diffeomorphism
ϕ : M → M if ϕ∗ v = v , i.e. if ϕ∗ (vP ) = vϕ(P ) for all P ∈ M . 2
We can write for any f ∈ C ∞ (M)
∗
(ϕ∗ v) (f ) = ϕ−1 (v (ϕ∗ f ))
⇒ ϕ∗ ((ϕ∗ v) (f )) = v (ϕ∗ f ) ,
⇒ ϕ∗ ◦ ϕ∗ v = v ◦ ϕ∗ . (10.9)

So if v is an invariant vector field, we can write
ϕ∗ ◦ v = v ◦ ϕ∗ . (10.10)
35
This expresses invariance under ϕ , and is satisfied by all differential

operators invariant under ϕ .
Consider a vector field u , and the local flow (or one-parameter
diffeomorphism group) φt corresponding to u ,
φt (Q) = γQ (t) , γ̇Q (t) = u(γQ (t)) . (10.11)
But for any f ∈ C ∞ (M) ,

d
γ̇Q (f ) = f ◦ γQ (t)
dt
d
= (f ◦ φt (Q))
dt
d ∗
= (φt (f )) = uγQ (t) (f ) ≡ u(f ) (10.12)

dt γQ (t)
At t = 0 we get the equation

d ∗
(φt (f )) = u(f ) (10.13)

dt t=0 Q
We can also write

d ∗
(φ f ) (Q) = u(f ) (φt (Q)) = φ∗t u(f )(Q) . (10.14)
dt t
This formula can be used to solve linear partial differential equations
of the form n
∂ X ∂
f (x, t) = v i (x) i f (x, t) (10.15)
∂t ∂x
i=1
with initial condition f (x, 0) = g(x) and everything smooth. This is
an equation on Rn+1 , so it can be on a chart for a manifold as well.
We can treat v i (x) as components of a vector field v . Then a
solution to this equation is
f (x, t) = φ∗t g(x)

≡ g (φt (x)) ≡ g ◦ φt (x) , (10.16)
where φt is the flow of v .

Proof:
∂ d ∗ ∂f
f (x, t) = (φt g) = v(f ) ≡ v i i , (10.17)
∂t dt ∂x
36 Chapter 10. Local flows
using Eq. (10.13) . 2

Thus the partial differential equation can be solved by finding the
integral curves of v (the flow of v) and then by pushing (also called
dragging) g along those curves. It can be shown, using well-known
theorems about the uniqueness of solutions to first order partial dif-
ferential equations, that this solution is also unique.
Example: Consider the equation in 2+1 dimensions

∂ ∂f ∂f
f (x, t) = (x − y) − (10.18)
∂t ∂x ∂y
with initial condition f (x, 0) = x2 + y 2 . The corresponding vector

field is v(x) = (x − y, −x + y) . The integral curve passing through
the point P = (x0 , y0 ) is given by the coordinates
γ(t) = (vx (P )t + x0 , vy (P )t + y0 ) , (10.19)
so the integral curve passing through (x, y) in our example is given

by
γ(t) = ((x − y)t + x, (−x + y)t + y) (10.20)

= Φt (x, y) ,
the flow of v. So the solution is
f (x, t) = Φ∗t f (x, 0) = f (x, 0) ◦ Φt (x, y)

= [(x − y)t + x]2 + [(−x + y)t + y]2
= (x − y)2 t2 + x2 + 2(x − y)xt + (x − y)2 t2 + y 2 − 2(x − y)yt
= 2(x − y)2 t2 + (x2 + y 2 )(1 + 2t) − 4xyt . (10.21)

Chapter 11
Lie derivative
Given some diffeomorphism ϕ, we have Eq. (10.5) for pushforwards

and pullbacks, ∗
(ϕ∗ v)f = ϕ−1 v (ϕ∗ (f )) . (11.1)
We will apply this to the flow φt of a vector field u , defined by
d ∗
(φt f ) = u(f ) . (11.2)

dt t=0 Q
Applying this at −t , we get

∗
φ−t∗ v(f ) = φ−1 v φ∗−t (f )

−t
= φ∗t v φ∗−t (f ) ,

(11.3)
where we have used the relation φ−1 t = φ−t . Let us differentiate this
equation with t ,
d d
(φ−t∗ v)(f ) = φ∗t v φ∗−t (f ) (11.4)

dt t=0 dt t=0
On the right hand side, φ∗t acts linearly on vectors and v acts linearly
on functions, so we can imagine At = φ∗t v as a kind of linear operator
acting on the function ft = φ∗−t f . Then the right hand side is of
the form

d d d
At ft = At ft + At ft
dt t=0 dt dt t=0
t=0
d ∗ d ∗
= φt v ft + At φ−t (f )
dt t=0 dt t=0

= u (v(f )) − v (u(f ))

t=0 t=0
= [u , v](f ) . (11.5)

t=0
37
38 Chapter 11. Lie derivative
The things in the numerator are numbers, so they can be compared

at different points, unlike vectors which may be compared only on
the same space. We can also write this as
φt∗ vφt (P ) − vP
lim = [u , v] . (11.6)
t→0 t
• This has the look of a derivative, and it can be shown to have
the properties of a derivation on the module of vector fields, appro-
priately defined. So the Lie bracket is also called the Lie derivative,
and written as
£u v = [u , v] . (11.7)
The derivation on functions by a vector field u : C ∞ (M) →
C ∞ (M) , f 7→ u(f ) , can be defined similarly as
φ∗t f − f
u(f ) = lim . (11.8)
t→0 t
• So this can also be called the Lie derivative of f with respect
to u , and written as £u f . 2
Then it is easy to see that
£u (f g) = (£u f ) g + f (£u g) ,
and £u (f + ag) = £u f + a£u g . (11.9)
So £u is a derivation on the space C ∞ (M) . Also,
£u (v + aw) = £u v + a£u w ,
and £u (f v) = (£u f ) v + f £u v ∀f ∈ C ∞ (M)
(11.10)
.
So £u is a derivation on the module of vector fields. Also, using

Jacobi identity, we see that
£u (v • w) = (£u v) • w + v • (£u w) , (11.11)
where v • w = [v , w] , so £u is a derivation on the Lie algebra of

vector fields.
Lie derivatives are useful in physics because they describe invari-

ances. For functions, £u f = 0 means φ∗t f = f , so the function does

not change along the flow of u . So the flow of u preserves f , or leaves
f invariant.
39
If there are two vector fields u and v which leave f invariant,

£u f = 0 = £v f . But we know from the Eq. (11.8), which defines
the Lie derivative of a function that
£u+av f = £u f + a£v = 0 ∀a ∈ R
and [£u , £v ]f = £[u ,v] f = 0 . (11.12)
So the vector fields which preserve f form a Lie algebra.

Similarly, a vector field is invariant under a diffeomorphism ϕ if
ϕ∗ v = v , as mentioned earlier. Using the flow of u , we find that a
vector field v is invariant under the flow of u if
φ−t v = v
⇒ £u v = v . (11.13)
So if a vector field w is invariant under the flows of u and v , i.e. if

£u w = 0 = £v w , we find that
0 = £u £v w − £v £u w = £[u ,v] w . (11.14)
Thus again the vector fields leaving w invariant form a Lie algebra.
• Let us also define the corresponding operations for 1-forms. As
we mentioned in Chap. 6, a 1−form is a section of the cotangent
bundle [
T ∗M = TP∗ M . (11.15)
P
Alternatively, a 1-form is a smooth linear map from the space of
vector fields on M to the space of smooth functions on M ,
ω : v 7→ ω(v) ∈ C ∞ (M), ω(u + av) = ω(U ) + aω(v) . (11.16)
A 1-form is a rule that (smoothly) selects a cotangent vector at each

point. 2
• Given a smooth map ϕM1 → M2 (say a diffeomorphism, for
convenience), the pullback ϕ∗ ω is defined by
(ϕ∗ ω) (v) = ω (ϕ∗ ω) . (11.17)
• We have already seen the gradient 1-form for a function f :

M → R , which is a linear map from the space of vector fields to

functions,
df (u + av) = u(f ) + av(f ) , (11.18)
and which can be written as

∂f i
df = dx (11.19)
∂xi
in some chart. 2
For an arbitrary 1-form ω , we can write in a chart and for any
vector field v ,
∂
ω = ωi dxi , v = vi , ω(v) = ωi v i . (11.20)
∂xi
All the components ωi , v i are smooth functions, so is ωI v i . The
space of 1-forms is a module. Since the function ω(v) is chart-
independent, we can find the components ωi0 of ω in a new chart
by noting that
0
ω(v) = ωi v i = ω 0i v i . (11.21)
Note that the notation is somewhat ambiguous here – i0 also runs
from 1 to n , and the prime actually distinguished the chart, or the
coordinate system, rather than the index i .
If the components of v in the new chart are related to those in
0 0
the old one by v i = Aij v j , it follows that
0 0
ωi0 Aij v j = ωj v j ⇒ ωi0 Aij = ωj (11.22)
Since coordinate transformations are invertible, we can multiply both
sides of the last equation by A−1 and write
j
ωi0 = A−1 i0 ωj . (11.23)
0
For coordinate transformations from a chart {xi } to a chart {x0i } ,
0
∂x0i
0 j ∂xj
Aij = , A−1 i0 = (11.24)
∂x j ∂x0i0
0
0 ∂x0i j ∂xj
so vi = v , ωi0 = ωj . (11.25)
∂x j ∂x0i0
We can define the Lie derivative of a 1-form very conveniently by
going to a chart, and treating the components of 1-forms and vector
fields as functions,
∂

£u ω(v) = £u ωi v i = uj j ωi v i

∂x
∂ωi ∂v i
= uj j v i + uj ωi j . (11.26)
∂x ∂x
41
But we want to define things such that
£u ω(v) = (£u ω) (v) + ω (£u v) . (11.27)
We already know the left hand side of this equation from Eq. (11.26),
and the right hand side can be calculated in a chart as
(£u ω) (v) + ω (£u v) = (£u ω)i v i + ωi (£u v)i

= (£u ω)i v i + ωi [u , v]i
i i

i j ∂v j ∂u
= (£u ω)i v + ωi u −v . (11.28)
∂xj ∂xj
Equating the right hand side of this with the right hand side of
Eq. (11.26), we can write
∂ωi ∂uj
(£u ω)i = uj + ωj . (11.29)
∂xj ∂xi
These are the components of £u ω in a given chart {xi } .
For the sake of convenience, let us write down the Lie deriva-
tives of the coordinate basis vector fields and basis 1-forms. The
coordinate basis vector corresponding to the i-th coordinate is
∂
v= ⇒ v j = δij . (11.30)
∂xi
Putting this into the formula for Lie derivatives, we get
∂ ∂
£u = [u , v]j j
∂xi ∂x
j j

k ∂v k ∂u ∂
= u k
−v k
∂x ∂x ∂xj
∂uj

∂
= 0 − δik k
∂x ∂xj
j
∂u ∂
=− . (11.31)
∂x ∂xj
i
Similarly, the 1-form corresponding to the i-th basis coordinate is
dxi = δji dxj , dxi j = δji .

i.e. (11.32)

Using this in the formula Eq. (11.29) we get

∂uk j ∂ui j
£u dxi = δki dx = dx . (11.33)
∂xj ∂xj
There is also a geometric description of the Lie derivative of 1-

forms,
1h ∗ i
£u ω|P = lim φt ω|φ (P ) − ωP
t→0 t t

d ∗
= φ ω . (11.34)
dt t P
We will not discuss this in detail, but only mention that it leads to
the same Leibniz rule as in Eq. (11.27), and the same description in
terms of components as in Eq. (11.29).

Chapter 12
Tensors
So far, we have defined tangent vectors, cotangent vectors, and also

vector fields and 1-forms. We will now define tensors. We will do
this by starting with the example of a specific type of tensor.
• A (1, 2) tensor AP at P ∈ M is a map
AP : TP M × TP M × TP∗ M → R (12.1)
which is linear in every argument. 2
So given two vectors uP , vP and a covector ωP ,
AP : (uP , vP , ωP ) 7→ AP (uP , vP ; ωP ) ∈ R . (12.2)
Suppose {ea }, {λa } are bases for TP M, TP∗ M . Write
Acab = AP (ea , eb ; λc ) . (12.3)
Then for arbitrary vectors uP = ua ea , vP = v a ea , and covector ωP =
ωa λa we get using linearity of the tensor map,

AP (uP , vP ; ωP ) = AP ua ea , v b eb ; ωc λc
= ua v b ωc Acab . (12.4)
It is a matter of convention whether A as written above should
be called a (1, 2) tensor or a (2, 1) tensor, and the convention varies
between books. So it is best to specify the tensor by writing indices
as there is no confusion about Acab .
A tensor of type (p, q) can be defined in the same way,

∗ ∗
Ap,q
P
: TP M × · · · × TP M × TP M × · · · × TP M → R
| {z } | {z }
q times p times (12.5)
43
44 Chapter 12. Tensors
in such a way that the map is linear in every argument.

• Alternatively, AP is an element of the tensor product space
∗ ∗
AP ∈ TP M ⊗ · · · ⊗ TP M ⊗ TP M ⊗ · · · ⊗ TP M
| {z } | {z }
p times q times (12.6)
We can define the components of this tensor in the same way that
we did for the (1, 2) tensor. Then a (p, q) tensor has components
which can be written as
a ···a
Ab11···bqp .
• Some special types of (p, q) tensors have special names. A (1, 0)
tensor is a linear map AP : TP∗ M → R , so it is a tangent vector. A
(0, 1) tensor is a cotangent vector. A (p, 0) tensor has components
with p upper indices. It is called a contravariant p−tensor. A
(0, q) tensor has components with q lower indices. It is called a
covariant q−tensor. 2
It is possible to add tensors of the same type, but not of different
types,
a ···a a ···a a ···a
Ab11···bqp + Bb11···bqp = (A + B)b11···bqp . (12.7)
• A tensor field is a rule giving a tensor at each point. 2
We can now define the Lie derivative of a tensor field by using
Leibniz rule in a chart. Let us first consider the components of a
tensor field in a chart. For a (1, 2) tensor field A , the components
in a chart are
∂ ∂
Akij = A( i , j ; dxk ) . (12.8)
∂x ∂x
The components are functions of x in a chart. Thus we can write
this tensor field as
∂
A = Akij dxi ⊗ dxj ⊗ , (12.9)
∂xk
where the × indicates a ‘product’, in the sense that its action on two
vectors and a 1-form is a product of the respective components,

∂
dx ⊗ dx ⊗ k (u, v; ω) = ui v j ωk .
i j
(12.10)
∂x

Thus we find, in agreement with the earlier definition,
A(u, v; ω) = Akij ui v j ωk . (12.11)

45
0
Under a change of charts, i.e. coordinate system xi → x0i , the
components of the tensor field change according to
∂ 0 0 0 ∂
A = Akij dxi ⊗ dxj ⊗ = Aki0 j 0 dx0i ⊗ dx0j ⊗ 0k0 (12.12)
∂xk ∂x
Since 0
0 ∂x0i i ∂ ∂xi ∂
dx0i = dx , = (12.13)
∂xi ∂x0i0 ∂x0i0 ∂xi
0
(i and i are not equal in general), we get
0 0
∂ 0 ∂x0i i ∂x0j ∂xk ∂
Akij dxi ⊗ dxj ⊗ k
= Aki0 j 0 i
dx ⊗ j
dxj ⊗ 0k0 k .
∂x ∂x ∂x ∂x ∂x
(12.14)
Equating components, we can write
0 0
0 ∂x0i ∂x0j ∂xk
Akij = Aki0 j 0 (12.15)
∂xi ∂xj ∂x0k0
0
0 ∂xi ∂xj ∂x0k
Aki0 j 0 = Akij . (12.16)
∂x0i0 ∂x0j 0 ∂xk
∂ ∂f
From now on, we will use the notation ∂i for and ∂i f for
∂xi ∂xi
unless there is a possibility of confusion. This will save some space
and make the formulae more readable.
We can calculate the Lie derivative of a tensor field (with respect
to a vector field u, say) by using the fact that £u is a derivative on
the modules of vector fields and 1-forms, and by assuming Leibniz
rule for tensor products. Consider a tensor field
m···n
T = Ta···b ∂m ⊗ · · · ⊗ ∂n ⊗ dxa ⊗ · · · ⊗ dxb . (12.17)
Then
m···n
£u T = (£u Ta···b ) ∂m ⊗ · · · ⊗ ∂n ⊗ dxa ⊗ · · · ⊗ dxb
m···n
+Ta···b (£u ∂m ) ⊗ · · · ⊗ ∂n ⊗ dxa ⊗ · · · ⊗ dxb + · · ·
m···n
+Ta···b ∂m ⊗ · · · ⊗ ∂n ⊗ (£u dxa ) ⊗ · · · ⊗ dxb , + · · ·
(12.18)
where the dots stand for the terms involving all the remaining up-

per and lower indices. Since the components of a tensor field are
functions on the manifold, we have
m···n
£u Ta···b = ui ∂i Ta···b
m···n
, (12.19)
46 Chapter 12. Tensors
and we also know that

∂ui ∂ua i
£u ∂m = − ∂i , £u dxa = dx . (12.20)
∂xm ∂xi
Putting these into the expression for the Lie derivative for T and
relabeling the dummy indices, we find the components of the Lie
derivative,
(£u T )m···n i m···n

a···b = u ∂i Ta···b
i···n
− Ta···b ∂i um − · · · − Ta···b
m···i
∂i un
m···n
+ Ti···b ∂a ui + · · · + Ta···i
m···n
∂b ui . (12.21)

Chapter 13
Differential forms
There is a special class of tensor fields, which is so useful as to have

a separate treatment. There are called differential p−forms or
p−forms for short.
• A p−form is a (0, p) tensor which is completely antisymmetric,
i.e., given vector fields v1 , · · · , vp ,
ω (v1 , · · · , vi , · · · , vj , · · · , vp ) = −ω (v1 , · · · , vj , · · · , vi , · · · , vp )
(13.1)
for any pair i, j . 2
A 0-form is defined to be a function, i.e. an element of C (M) , ∞
and a 1-form is as defined earlier.

The antisymmetry of any p-form implies that it will give a non-
zero result only when the p vectors are linearly independent. On the
other hand, no more than n vectors can be linearly independent in
an n-dimensional manifold. So p 6 n .
Consider a 2-form A . Given any two vector fields v1 , v2 , we have
A(v1 , v2 ) = −A(v2 , v1 ) . Then the components of A in a chart are
Aij = A (∂i , ∂j ) = −Aji . (13.2)
Similarly, for a p-form ω , the components are ωi1 ···ip , and compo-
nents are multiplied by (−1) whenever any two indices are inter-
changed.
n
It follows that a p-form has independent components in n-
p

dimensions.
Any 1-form produces a function when acting on a vector field. So
given a pair of 1-forms A, B, it is possible to construct a 2-form ω
47
48 Chapter 13. Differential forms
by defining
ω(u, v) = A(u)B(v) − B(u)A(v), ∀u, v . (13.3)
• This is usually written as ω = A ⊗ B − B ⊗ A , where ⊗ is called

the outer product. 2
• Then the above construction defines a product written as
ω = A ∧ B = −B ∧ A , (13.4)
and called the wedge product . Clearly, ω is a 2-form. 2

Let us work in a coordinate basis, but the results we find can be
generalized to any basis. The coordinate bases for the vector fields,
{∂i } , and 1-forms, {dxi } , satisfy dxi (∂j ) = δji . A 1-form A can be
written as A = Ai dxi , and a vector field v can be written as v = v i ∂i ,
so that A(v) = Ai v i . Then for the ω defined above and for any pair
of vector fields u, v,
ω(u, v) = A(u)B(v) − B(u)A(v)

= Ai ui Bj v j − Bi ui Aj v j
= (Ai Bj − Bi Aj ) ui v j . (13.5)
The components of ω are ωij = ω(∂i , ∂j ) , so that
ω(u, v) = ω(ui ∂i , v j ∂j ) = ωij ui v j . (13.6)
Then ωij = Ai Bj − Bi Aj for the 2-form defined above. We can now

construct a basis for 2-forms, which we write as dxi ∧ dxj ,
dxi ∧ dxj = dxi ⊗ dxj − dxj ⊗ dxi . (13.7)
Then a 2-form can be expanded in this basis as

1
ω= ωij dxi ∧ dxj , (13.8)
2!
because then

1
ωij dxi ⊗ dxj − dxj ⊗ dxi (u, v)

ω(u, v) =
2!
1
= ωij ui v j − uj v i = ωij ui v j .

(13.9)
2!
49
Similarly, a basis for p−forms is
dxi1 ∧ · · · ∧ dxip = dx[i1 ⊗ · · · ⊗ dxip ] , (13.10)
where the square brackets stand for total antisymmetrization: all

even permutations of the indices are added and all the odd permu-
tations are subtracted. (Caution: some books define the ‘square
brackets’ as antisymmetrization with a factor 1/p! .) For example,
for a 3-form, a basis is
dxi ∧ dxj ∧ dxk = dxi ⊗ dxj ⊗ dxk − dxj ⊗ dxi ⊗ dxk

+dxj ⊗ dxk ⊗ dxi − dxk ⊗ dxj ⊗ dxi
+dxk ⊗ dxi ⊗ dxj − dxi ⊗ dxk ⊗ dxj . (13.11)
Then an arbitrary 3-form Ω can be written as

1
Ω= Ωijk dxi ∧ dxj ∧ dxk . (13.12)
3!
Note that there is a sum over indices, so that the factorial goes away if
we write each basis 3-form up to permutations, i.e. treating different
permutations as equivalent. Thus a p−form α can be written in
terms of its components as
1
α= αi ···i dxi1 ∧ · · · ∧ dxip . (13.13)
p! 1 p
Examples: A 2-form in two dimensions can be written as
1
ω = ωij dxi ∧ dxj
2!
1
ω12 dx1 ∧ dx2 + ω21 dx2 ∧ dx1

=
2!
1
= (ω12 − ω21 ) dx1 ∧ dx2
2!
= ω12 dx1 ∧ dx2 . (13.14)
2
A 2-form in three dimensions can be written as

1
ω = ωij dxi ∧ dxj
2!
= ω12 dx1 ∧ dx2 + ω23 dx2 ∧ dx3 + ω31 dx3 ∧ dx1 (13.15)
2
In three dimensions, consider two 1-forms α = αi dxi , β = βi dxi .
Then
1
α∧β = (αi βj − αj βi ) dxi ∧ dxj
2!
= αi βj dxi ∧ dxj
= (α1 β2 − α2 β1 ) dx1 ∧ dx2
+ (α2 β3 − α3 β2 ) dx2 ∧ dx3
+ (α3 β1 − α1 β3 ) dx3 ∧ dx1 . (13.16)
The components are like the cross product of vectors in three dimen-
sions. So we can think of the wedge product as a generalization of
the cross product.
• We can also define the wedge product of a p−form α and a
q−form β as a (p + q)−form satisfying, for any p + q vector fields
v1 , · · · , vp+q ,
1 X
α ∧ β (v1 , · · · , vp+q ) = (−1)deg P α ⊗ β (P (v1 , · · · , vp+q )) .
p!q!
P
(13.17)
Here P stands for a permutation of the vector fields, and deg P is 0 or
1 for even and odd permutations, respectively. In the outer product
on the right hand side, α acts on the first p vector fields in a given
permutation P , and β acts on the remaining q vector fields. 2
The wedge product above can also be defined in terms of the
components of α and β in a chart as follows.
1
α= αi ···i dxi1 ∧ · · · ∧ dxip
p! 1 p
1
β = βj ···j dxj1 ∧ · · · ∧ dxjq
q! 1 q
1
αi1 ···ip βj1 ···jq dxi1 ∧ · · · ∧ dxip ∧ dxj1 ∧ · · · ∧ dxjq .

α∧β =
p!q!
(13.18)
Note that α ∧ β = 0 if p + q > n , and that a term in which some i
is equal to some j must vanish because of the antisymmetry of the
wedge product.

It can be shown by explicit calculation that wedge products are

associative,
α ∧ (β ∧ γ) = (α ∧ β) ∧ γ . (13.19)
51
Cross-products are not associative, so there is a distinction between

cross-products and wedge products. In fact, for 1-forms in three
dimensions, the above equation is analogous to the identity for the
triple product of vectors,
a × (b × c) = (a × b) × c . (13.20)
For a p-form α and q-form β , we find
α ∧ β = (−1)pq β ∧ α . (13.21)
Proof: Consider the wedge product written in terms of the com-

ponents. We can ignore the parentheses separating the basis forms
since the wedge product is associative. Then we exchange the basis
1-forms. One exchange gives a factor of −1 ,
dxip ∧ dxj1 = −dxj1 ∧ dxip . (13.22)
Continuing this process, we get
dxi1 ∧ · · · ∧ dxip ∧ dxj1 ∧ · · · ∧ dxjq

= (−1)p dxj1 ∧ dxi1 ∧ · · · ∧ dxip ∧ dxj2 ∧ · · · ∧ dxjq
= ···
= (−1)pq dxj1 ∧ · · · ∧ dxjq ∧ dxi1 ∧ · · · ∧ dxip . (13.23)
Putting back the components, we find
α ∧ β = (−1)pq β ∧ α (13.24)
as wanted. 2
• The wedge product defines an algebra on the space of differential
forms. It is called a graded commutative algebra . 2
• Given a vector field v , we can define its contraction with a
p-form by
ιv ω = ω(v, · · · ) (13.25)
with p − 1 empty slots. This is a (p − 1)-form. Note that the position
of v only affects the sign of the contracted form. 2
Example: Consider a 2-form made of the wedge product of two

1-forms, ω = λ ∧ µ = λ ⊗ µ − µ ⊗ λ . Then contraction by v gives
ιv ω = ω(v, • ) = λ(v)µ − µ(v)λ = −ω( • , v) . (13.26)

If we have a p-form ω = p!1 ωi1 ···ip dxi1 ∧ · · · ∧ dxip , its contraction

with a vector field v = v i ∂i is
1
ιv ω = ωii ···i v i dxi2 ∧ · · · ∧ dxip . (13.27)
(p − 1)! 2 p
1
Note the sum over indices. To see how the factor becomes (p−1)! , we
write the contraction as
1
ιv ω = ωi1 ···ip dxi1 ∧ · · · ∧ dxip v i ∂i .

(13.28)
p!
Since the contraction is done in the first slot, so we consider the
action of each basis 1-form dxik on ∂i by carrying dxik to the first
position and then writing a δiik . This gives a factor of (−1) for each
exchange, but we get the same factor by rearranging the indices of
ω , thus getting a +1 for each index. This leads to an overall factor
of p .
• given a diffeomorphism ϕ : M1 → M2 , the pullback of a 1-
form λ (on M2 ) is ϕ∗ λ , defined by
ϕ∗ λ(v) = λ(ϕ∗ v) (13.29)
for any vector field v on M1 . 2
∗ i
Then we can consider the pullback ϕ dx of a basis 1-form dx . i
For a general 1-form λ = λi dxi , we have ϕ∗ λ = ϕ∗ (λi dxi ) . But

ϕ∗ λ(v) = λ(ϕ∗ v) = λi dxi (ϕ∗ v) . (13.30)
Now, dxi (ϕ∗ v) = ϕ∗ dxi (v) and the thing on the right hand side is a
function on M1 , so we can write this as
ϕ∗ λ(v) = (ϕ∗ λi )ϕ∗ dxi (v) , (13.31)
where ϕ∗ λi are now functions on M1 , i.e.
(ϕ∗ λi )|P = λi |ϕ(P ) (13.32)
So we can write ϕ∗ λ = (ϕ∗ λi ) ϕ∗ dxi . For the wedge product of two

1-forms,
ϕ∗ (λ ∧ µ)(u, v) = (λ ∧ µ)(ϕ∗ u , ϕ∗ v)
= λ ⊗ µ(ϕ∗ u , ϕ∗ v) − µ ⊗ λ(ϕ∗ u , ϕ∗ v)

= λ(ϕ∗ u)µ(ϕ∗ v) − µ(ϕ∗ u)λ(ϕ∗ v)

= ϕ∗ λ(u)ϕ∗ µ(v) − ϕ∗ µ(u)ϕ∗ λ(v)
= (ϕ∗ λ ∧ ϕ∗ µ)(u , v) . (13.33)
53
Since u, v are arbitrary vector fields it follows that
ϕ∗ (λ ∧ µ) = ϕ∗ λ ∧ ϕ∗ µ
ϕ∗ (dxi ∧ dxj ) = ϕ∗ dxi ∧ ϕdxj . (13.34)
Since the wedge product is associative, we can write (by assuming

an obvious generalization of the above formula)

ϕ∗ dxi ∧ dxj ∧ dxk = ϕ∗ dxi ∧ dxj ∧ dxk

= ϕ∗ dxi ∧ dxj ∧ ϕ∗ dxk

= ϕ∗ dxi ∧ ϕ∗ dxj ∧ ϕ∗ dxk , (13.35)
and we can continue this for any number of basis 1-forms. So for any
p-form ω , let us define the pullback ϕ∗ ω by
ϕ∗ ω(v1 , · · · , vp ) = ω (ϕ∗ v1 , · · · , ϕ∗ vp ) , (13.36)
and in terms of components, by

1
ϕ∗ ω = ϕ∗ ωi1 ···ip ϕ∗ dxi1 ∧ · · · ∧ dxip .

(13.37)
p!
We assumed above that the pullback of the wedge product of a
2-form and a 1-form is the wedge product of the pullbacks of the
respective forms, but it is not necessary to make that assumption –
it can be shown explicitly by taking three vector fields and following
the arguments used earlier for the wedge product of two 1-forms.
Then for any p-form α and q-form β we can calculate from this
that
ϕ∗ (α ∧ β) = ϕ∗ α ∧ ϕ∗ β . (13.38)
Thus pullbacks commute with (are distributive over) wedge products.

Chapter 14
Exterior derivative
The exterior derivative is a generalization of the gradient of a func-

tion. It is a map from p-forms to (p + 1)-forms. This should be a
derivation, so it should be linear,
d(α + ω) = dα + dω ∀p-forms α , ω . (14.1)
This should also satisfy Leibniz rule, but the algebra of p-forms is
not a commutative algebra but a graded commutator algebra, i.e.,
involves a factor of (−1)pq for exchanges. So we need
d(α ∧ β) = dα ∧ β + (−1)pq dβ ∧ α , (14.2)
or alternatively,
d(α ∧ β) = dα ∧ β + (−1)p α ∧ dβ . (14.3)
This will be the Leibniz rule for wedge products. Note that it gives
the correct result when one or both of α, β are 0-forms, i.e., functions.
The two formulas are identical by virtue of the fact that dβ is a
(q + 1)-form, so that
α ∧ dβ = (−1)p(q+1) dβ ∧ α . (14.4)
We will try to define the exterior derivative in a way such that it has
these properties.

Let us define the exterior derivative of a p-form ω in a chart as

1
dω = ∂i ωi1 ···ip dxi1 ∧ · · · ∧ dxip (14.5)
p!
54
55
This clearly has the first property of linearity. To check the (graded)
Leibniz rule, let us write α ∧ β in components. Then
1
∂i αi1 ···ip βj1 ···jq dxi ∧ dxi1 ∧ · · · ∧ dxjq

d(α ∧ β) =
p!q!
1
∂i αi1 ···ip βj1 ···jq + αi1 ···ip ∂i βj1 ···jq dxi ∧ dxi1 ∧ · · · ∧ dxjq

=
p!q!
1
∂i αi1 ···ip βj1 ···jq dxi ∧ dxi1 ∧ · · · ∧ dxip ∧ dxj1 ∧ · · · dxjq

=
p!q!
1
(−1)p αi1 ···ip ∂i βj1 ···jq dxi1 ∧ · · · ∧ dxip ∧ dxi ∧ dxj1 ∧ · · · dxjq

+
p!q!
= dα ∧ β + (−1)p α ∧ dβ . (14.6)
A third property of the exterior derivative immediately follows

from here,
d2 = 0 . (14.7)
To see this, we write
1
d ∂i ωi1 ···ip dxi ∧ dxi1 ∧ · · · dxip

d(dω) =
p!
1
= ∂j ∂i ωi1 ···ip dxj ∧ dxi ∧ dxi1 ∧ · · · dxip . (14.8)
p!
But the wedge product is antisymmetric, dxj ∧ dxi = −dxi ∧ dxj ,

and the indices are summed over, so the above object must be anti-
symmetric in ∂j , ∂i . But that vanishes. So d2 = 0 on all forms.
Note that we can also write
1
dωi1 ···ip ∧ dxi1 ∧ · · · dxip ,

dω = (14.9)
p!
where the object in parentheses is a gradient 1-form corresponding
to the gradient of the component.
Consider a 1-form A = Aµ dxµ where Aµ are smooth functions on
M . Then using this definition we can write
dA = (dAν ) ∧ dxν
= ∂µ Aν dxµ ∧ dxν

1
= (∂µ Aν − ∂ν Aµ ) dxµ ∧ dxν
2
⇒ (dA)µν = ∂ µ Aν − ∂ ν Aµ . (14.10)
56 Chapter 14. Exterior derivative
We can generalize this result to write for a p-form,

1
α= αµ ···µ dxµ1 ∧ · · · ∧ dxµp (14.11)
p! 1 p
1
∂µ αµ1 ···µp dxµ1 ∧ · · · ∧ dxµp

dα =
p!
1
= ∂ α dxµ ∧ dxµ1 ∧ · · · ∧ dxµp
(p + 1)! [µ µ1 ···µp ]
⇒ (dα)µµ1 ···µp = ∂[µ αµ1 ···µp ] (14.12)
Example: For p = 1 i.e. for a 1-form A we get from this formula

(dA)µν = ∂µ Aν − ∂ν Aµ , in agreement with our previous calculation.
For p = 2 we have a 2-form, call it α. Then using this formula
we get
(dα)µνλ = ∂[µ ανλ]

= ∂µ ανλ − ∂ν αµλ + ∂ν αλµ − ∂λ ανµ + ∂λ αµν − ∂µ αλν .
(14.13)
Note that d is not defined on arbitrary tensors, but only on forms.

2
By definition, d2 = 0 on any p-form. So if α = dβ , it follows that
dα = 0 . But given a p-form α for which dα = 0 , can we say that
there must be some (p − 1)-form β such that α = dβ ?
• This is a good place to introduce some terminology. Any form
ω such that dω = 0 is called closed, whereas any form α such that
α = dβ is called exact. 2
So every exact form is closed. Is every closed form exact? The
answer is yes, in a sufficiently small neighbourhood. We say that
every closed form is locally exact. Note that if a p-form α = dβ , we
cannot uniquely specify the (p − 1)-form β since for any (p − 2)-form
γ , we can always write α = dβ 0 , where β 0 = β + dγ .
Thus a more precise statement is that given any p-form α such
that dα = 0 in a neighbourhood of some point P , there is some
neighbourhood of this point and some (p−1)-form β such that α = dβ
in that neighbourhood. But this may not be true globally. This
statement is known as the Poincaré lemma. 2

Example: In R2 remove the origin. Consider the 1-form

xdy − ydx
α= . (14.14)
x2 + y 2
57
Then
2x2 2y 2

1 1
dα = − dx ∧ dy − − dy ∧ dx
x2 + y 2 (x2 + y 2 )2 x2 + y 2 (x2 + y 2 )2
2 x2 + y 2
= 2 dx ∧ dy − 2 dx ∧ dy = 0 . (14.15)
x + y2 (x2 + y 2 )2
Introduce polar coordinates r, θ with x = r cos θ , y = r sin θ .
Then
dx = dr cos θ − r sin θdθ dy = dr sin θ + r cos θdθ
r cos θ (sin θdr + r cos θdθ) r sin θ (cos θdr − r sin θdθ)
α= −
r2 r2
2 2 2
r cos θ + sin θ dθ
= = dθ . (14.16)
r2
Thus α is exact, but θ is multivalued so there is no function f
such that α = df everywhere. In other words, α = dθ is exact only
in a neighbourhood small enough that θ remains single-valued.

Chapter 15
Volume form

n
The space of p-forms in n dimensions is dimensional. So the
p
space of n-forms in n dimensions is 1-dimensiona, i.e., there is only
one independent component, and all n-forms are scalar multiples of
one another.
Choose an n-form field. Call it ω. Suppose ω 6= 0 at some point
P . Then given any basis {eµ } of TP M , we have ω(e1 , · · · en ) 6= 0
since ω 6= 0 . Thus all vector bases at P fall into two classes, one for
which ω(e1 , · · · en ) > 0 and the other for which it is < 0 .
Once we have identified these two classes, they are independent of
ω . That is, if ω 0 is another n-form which is non-zero at P , there must
be some function f 6= 0 such that ω 0 = f ω . Two bases which gave
positive numbers under ω will give the same sign — both positive or
both negative — under ω 0 and therefore will be in the same class.
• So every basis (set of n linearly independent vectors) is a mem-
ber of one of the two classes. These are called righthanded and
lefthanded. 2
• A manifold is called orientable if it is possible to define a con-
tinuous n-form field ω which is non-zero everywhere on the manifold.
Then it is possible to choose a basis with the same handedness ev-
erywhere on the manifold continuously. 2
Euclidean space is orientable, the Möbius band is not.
• An orientable manifold is called oriented once an orientation
has been chosen, i.e. once we have decided to choose basis vectors

with the same handedness everywhere on the manifold. 2

• It is necessary to choose an oriented manifold when we discuss
the integration of forms. On an n-dimensional manifold, a set of n
58
59
linearly independent vectors define an n-dimensional parallelepiped.

If we define an n-form ω 6= 0 we can think of the value of these vectors
as the volume of this parallelepiped. This ω is called a volume form.
2
Once a volume form has been chosen, any set of n linearly inde-
pendent vectors will define a positive or negative volume.
The integral of a function f on Rn is the sum of the values of f ,
multiplied by infinitesimal volumes of coordinate elements. Similarly,
we define the integral of a function f on an oriented manifold as the
sum of the values of f , multiplied by infinitesimal volumes. The way
to do that is the following.
Given a function f , define an n-form in a chart by ω = f dx1 ∧· · ·∧
dxn . To integrate over an open set U , divide it up into infinitesimal
‘cells’, spanned by vectors

1∂ ∂ ∂
∆x 1
, ∆x2 2 , · · · , ∆xn n ,
∂x ∂x ∂x
where the ∆xi are small numbers.

Then the integral of f over one such cell is approximately
f ∆x1 ∆x2 · · · ∆xn = f dx1 ∧ · · · ∧ dxn ∆x1 ∂1 , · · · , ∆xn ∂n

= ω(cell) . (15.1)
Adding up the contributions from all cells and taking the limit of cell
size going to zero, we find
Z Z
ω= f dn x . (15.2)
U ϕ(U )
The right hand side is the usual integration in calculus of n variables,

and the left hand side is our notation which we are defining.
The right hand side can be seen to be independent of the choice
of coordinate system. If we choose a different coordinate system, we

get a Jacobian, but also a redefinition of the region ϕ(U ) . Let us

check that the left hand side is also invariant of the choice of the
coordinates. We will do this in two dimensions with ω = f dx1 ∧ dx2 .
60 Chapter 15. Volume form
In another coordinate system (y 1 , y 2 ) corresponding to ϕ0 (U )
∂x1 1 ∂x1 2
dx1 = dy + 2 dy
∂y 1 ∂y
∂x 2 ∂x2 2
dx2 = dy 1
+ dy
∂y 1 ∂y 2
1 2
∂x1 ∂x2

1 2 ∂x ∂x
⇒ dx ∧ dx = − 2 2 dy 1 ∧ dy 2
∂y 1 ∂y 1 ∂y ∂y
1 2
= Jdy ∧ dy , (15.3)
and J is the Jacobian.

So what we have here is
Z Z
ω = f (x1 , x2 )dx1 ∧ dx2
U U
Z
= f (y 1 , y 2 )Jdy 1 ∧ dy 2
U
Z
= f (y 1 , y 2 )Jd2 y , (15.4)
ϕ0 (U )
so we get the same result both ways.

Given the same f , if we choose a basis with the opposite orien-
tation, the integral of ω will have the opposite sign. This is why the
choice of orientation has to be made before integration.
Manifolds become even more interesting if we define a metric.

Chapter 16
Metric tensor
• A metric on a vector space V is a function g : V × V → R

which is
i) bilinear:
g(av1 + v2 , w) = ag(v1 , w) + g(v2 , w)

g(v, w1 + aw2 ) = g(v, w1 ) + ag(v, w2 ) , (16.1)
i.e., g is a (0,2) tensor;
ii) symmetric:
g(v, w) = g(w, v); (16.2)
iii) non-degenerate:
g(v, w) = 0 ∀w ⇒ v = 0. (16.3)
2
• If for some v, w 6= 0 , we find that g(v, w) = 0 , we say that v, w
are orthogonal. 2
• Given a metric g on V , we can always find an orthonormal
basis {eµ } such that g(eµ , eν ) = 0 if µ 6= ν and ±1 if µ = ν . 2
• If the number of (+1)’s is p and the number of (−1)’s is q , we
say that the metric has signature (p, q) .
We have defined a metric for a vector space. We can generalize
this definition to a manifold M by the following.

• A metric g on a manifold M is a (0, 2) tensor field such that if

(v, w) are smooth vector fields, g(v, w) is a smooth function on M ,
and has the properties (16.1), (16.2) and (16.3) mentioned earlier. 2
61
62 Chapter 16. Metric tensor
It is possible to show that smoothness implies that the signature

is constant on any connected component of M , and we will assume
that it is constant on all of M .
A vector space becomes related to its dual space by the metric.
Given a vector space V with metric g , and vector v defines a linear
map g(v, ·) : V → R , w 7→ g(v, w) ∈ R . Thus g(v, ·) ∈ V ∗ where V ∗
is the dual space of V . But g(v, ·) is itself linear in v , so the map
V → V ∗ defined by g(v, ·) is linear. Since g is non-degenerate, this
map is an isomorphism. It then follows that on a manifold we can
use the metric to define a linear isomorphism between vectors and
1-forms.
In a basis, the components of the metric are gµν = g(eµ , eν ) . This
is an n × n matrix in an n-dimensional manifold. We can thus write
g(v, w) = gµν v µ wν in terms of the components. Non-degeneracy
implies that this matrix is invertible. Let g µν denote the inverse
matrix. Then, by definition of an inverse matrix, we have
gµν g νλ = δµλ = g λν gµν . (16.4)
Then the linear isomorphism takes the following form.

i) If v = v µ eµ is a vector field in a chart, and {λµ } is the dual
basis to {eµ } ,
g(v, ·) = vµ λµ , (16.5)
where vµ = gµν v ν .
ii) If A = Aµ λµ is a 1-form written in a basis {λµ } , the corre-

sponding vector field is Aµ eµ , where Aµ = g µν Aν .
This is the isomorphism between vector fields and 1-forms. (We
could of course define a similar isomorphism between vectors and
covectors without referring to a manifold.) A similar isomorphism
holds for tensors, e.g. in terms of components,
T µν ←→ T µν ←→ Tµ ν ←→ Tµν (16.6)
µνρ ···
T ←→ T µν ρ ··· ←→ T µν
ρ ··· ←→ Tµνρ ···
←→ · · · (16.7)

These correspondences are not equalities — the components are not

equal. What it means is that, if we know one set of components, say
T µνρ ··· , and the metric, we also know every other set of components.
63
• Using the fact that a non-degenerate metric defines a 1-1 linear

map between vectors and 1-forms, we can define an inner product
of 1−forms, by
hA | Bi = g µν Aµ Bν (16.8)
for 1-forms A, B . This result is independent of the choice of basis, i.e.
independent of the coordinate system, just like the inner product
of vector fields,
hv | wi = g(v, w) = gµν v µ wν . (16.9)
2
Given a manifold with metric, there is a canonical volume form
dV (sometimes written as vol) , which in a coordinate chart reads
q
dV = | det gµν |dx1 ∧ · · · ∧ dxn . (16.10)
Note that despite the notation, this is not a 1-form, nor the gradient
of some function V . This is clearly a volume form because it is an
n-form which is non-zero everywhere, as gµν is non-degenerate.
We need to show that this definition is independent of the chart.
Take an overlapping chart. Then in the new chart, the corresponding
volume form is
q
dV 0 = | det gµν0 |dx01 ∧ · · · ∧ dx0n . (16.11)
We wish to show that dV 0 = dV . In the overlap,

∂x0µ ν
dx0µ = dx = Aµν dxν (say) (16.12)
∂xν
Then dx01 ∧ · · · ∧ dx0n = (det A)dx1 ∧ · · · ∧ dxn .
On the other hand, if we look at the components of the metric
tensor in the new chart,
0
gµν = g(∂µ0 , ∂ν0 )
α
∂xβ

∂x
= ∂ α , ∂ β
∂x0µ ∂x0ν

α β
= g A−1 µ ∂α , A−1 ν ∂β
α β
= A−1 µ A−1 ν gαβ . (16.13)
64 Chapter 16. Metric tensor
Taking determinants, we find

0
det gµν = (det A)−2 (det gµν ) . (16.14)
Thus q q
0 −1
| det gµν | = |det A| | det gµν | , (16.15)
and so dV 0 = dV .
• This is called the metric volume form and written as
p
dV = |g|dx1 ∧ · · · ∧ dxn (16.16)
in a chart. 2
When we write dV , sometimes p we mean the n-form as defined
above, and sometimes we mean |g|dn x , the measure for the usual
integral. Another way of writing the volume form in a chart is in
terms of its components,
p
|g|
dV = µ1 ···µn dxµ1 ∧ · · · ∧ dxµn (16.17)
n!
where is the totally
p antisymmetric Levi-Civita symbol, with
12···n = +1 . Thus |g| µ1 ···µn are the components of the volume
form.

Chapter 17
Hodge duality
We will next define the Hodge star operator. We will defineit in a

chart rather than abstractly.
• The Hodge star operator, denoted ? in an n-dimensional
manifold is a map from p-forms to (n − p)-forms given by
p
|g|
(?ω)µ1 ···µn−p ≡ µ1 ···µn g µn−p+1 ν1 · · · g µn νp ων1 ···νp , (17.1)
p!
where ω is a p-form. 2
The ? operator acts on forms, not on components.
Example: Consider R3 with metric +++, i.e. gµν =
diag(1, 1, 1) . Then |g| ≡ g = 1 , g µν diag(1, 1, 1) . Write the coordi-
nate basis 1-forms as dx, dy, dz . Their components are clearly
(dx)i = δi1 , (dy)i = δi2 , (dz)i = δi3 , (17.2)
the δ’s on the right hand sides are Kroenecker deltas. So
(?dx)ij = ijk g kl (dx)l = ijk g kl δl1 = ijk g k1

1 1
⇒ ?dx = (?dx)ij dxi ∧ dxj = ijk g k1 dxi ∧ dxj
2! 2!
g k1 = 1 for k = 1 , 0 otherwise
1
dx2 ∧ dx3 − dx3 ∧ dx2 = dx2 ∧ dx3 = dy ∧ dz .

⇒ ?dx =
2!

(17.3)
Similarly, ?dy = dz ∧ dx , ?dz = dx ∧ dy . 2
65
66 Chapter 17. Hodge duality
Example: Consider p = 0 (scalar), i.e. a 0-form ω in n dimen-

sions.
p
(?ω)µ1 ···µn = |g|µ1 ···µn ω
p
⇒ (?1)µ1 ···µn = |g|µ1 ···µn
p
|g|
⇒ (?1) = µ1 ···µn dxµ1 ∧ · · · ∧ dxµn
n!
= dV (17.4)
2
Example: p = n . Then
p
|g|
(?ω) = µ1 ···µn g µ1 ν1 · · · g µn νn ων1 ···νn . (17.5)
n!
For the volume form,
p
|g|
dV = µ1 ···µn dxµ1 ∧ · · · ∧ dxµn
p n!
(dV )ν1 ···νn = |g|ν1 ···νn
|g|
(?dV ) = µ ···µ g µ1 ν1 · · · g µn νn ν1 ···νn
n! 1 n
|g| |g| n!
= n!(det g)−1 = = sign(g) = (−1)s ,(17.6)
n! n! g
where s is the number of (−1) in gµν . 2
So we find that
?(?1) = ?dV = (−1)s , (17.7)
and
?(?dV ) = (−1)s (?1) = (−1)s dV , (17.8)
i.e., (?)2 = (−1)s on 0-forms and n-forms.
In general, on a p-form in an n-dimensional manifold with signa-
ture (s, n − s) , it can be shown in the same way that
(?)2 = (−1)p(n−p)+s . (17.9)

In particular, in four dimensional Minkowski space, s = 1, n = 4 , so
(?)2 = (−1)p(4−p)+1 . (17.10)

67
It is useful to work out the Hodge dual of basis p-forms. Suppose

we have a basis p-form dxI1 ∧ · · · ∧ dxIp , where the indices are ar-
ranged in increasing order Ip > · · · > I1 . Then its components are
I
p!δµI11 · · · δµpp . So
? dxI1 ∧ · · · ∧ dxIp ν1 ···νn−p

p
|g| 0 0 I
= ν1 ···νn−p µ1 ···µp g µ1 µ1 · · · g µp µp p! δµI10 · · · δµp0
p! 1 p
p µ1 I1 µp Ip
= |g| ν1 ···νn−p µ1 ···µp g ···g . (17.11)
We will use this to calculate ?ω ∧ ω .

For a p-form ω , we have
1
ω = ωµ ···µ dxµ1 ∧ · · · ∧ dxµp
p! 1 p
X
= ωI1 ···Ip dxI1 ∧ · · · ∧ dxIp (17.12)
I
where the sum over I means a sum over all possible index sets I =
I1 < · · · < Ip , but there is no sum over the indices {I1 , · · · , Ip }
themselves, in a given index set the Ik are fixed. Using the dual of
basis p−forms, and Eq. (13.13), we get
X
?ω = ωI1 ···Ip ? (dxI1 ∧ · · · ∧ dxIp )
I
p
X |g|
= ν ···ν µ ···µ g µ1 I1 · · · g µp Ip ωI1 ···Ip dxν1 ∧ · · · ∧ dxνn−p .
(n − p)! 1 n−p 1 p
I
(17.13)
The sum over I is a sum over different index sets as before, and
the Greek indices are summed over as usual. Thus we calculate
p
|g| X
?ω ∧ ω = ν1 ···νn−p µ1 ···µp g µ1 I1 · · · g µp Ip ωI1 ···Ip ×
(n − p)!
I,J
dxν1 ∧ · · · ∧ dxνn−p ∧ ωJ1 ···Jp dxJ1 ∧ · · · ∧ dxJp

p
|g| X

= ν1 ···νn−p µ1 ···µp g µ1 I1 · · · g µp Ip ωI1 ···Ip ωJ1 ···Jp ×

(n − p)!
I,J
dxν1 ∧ · · · ∧ dxνn−p ∧ dxJ1 ∧ · · · ∧ dxJp (17.14)
68 Chapter 17. Hodge duality
We see that the set {ν1 , · · · , νn−p } cannot have any overlap with the
set J = {J1 , · · · , Jp }, because of the wedge product. On the other
hand, {ν1 , · · · , νn−p } cannot have any overlap with {µ1 , · · · , µp } be-
cause is totally antisymmetric in its indices. So the set {µ1 , · · · , µp }
must have the same elements as the set J = {J1 , · · · , Jp } , but they
may not be in the same order.
Now consider the case where the basis is orthogonal, i.e. g µν is
diagonal. Then g µk Ik = g Ik Ik etc. and we can write
p
|g| X
?ω ∧ ω = ν1 ···νn−p I1 ···Ip g I1 I1 · · · g Ip Ip ωI1 ···Ip ωJ1 ···Jp ×
(n − p)!
I,J
dxν1 ∧ · · · ∧ dxνn−p ∧ dxJ1 ∧ · · · ∧ dxJp . (17.15)
We see that in each term of the sum, the indices {I1 · · · Ip } must be
the same as {J1 · · · Jp } because both sets are totally antisymmetrized
with the indices {ν1 · · · νn−p }.
Since both sets are ordered, it follows that we can replace J by
I,
p
|g| X
?ω ∧ ω = ν1 ···νn−p I1 ···Ip g I1 I1 · · · g Ip Ip ωI1 ···Ip ωI1 ···Ip ×
(n − p)!
I
p dxν1 ∧ · · · ∧ dxνn−p ∧ dxI1 ∧ · · · ∧ dxIp

|g| X
= ν1 ···νn−p I1 ···Ip ω I1 ···Ip ωI1 ···Ip ×
(n − p)!
I
dxν1 ∧ · · · ∧ dxνn−p ∧ dxI1 ∧ · · · ∧ dxIp . (17.16)
In each term of this sum, the indices {ν1 · · · νn−p } are completely
determined, so we can replace them by the corresponding ordered
set K = K1 < · · · < Kn−p , which is completely determined by the
set I , so that
p X
?ω ∧ ω = |g| K1 ···Kn−p I1 ···Ip ω I1 ···Ip ωI1 ···Ip ×
I
dxK1 ∧ · · · ∧ dxKn−p ∧ dxI1 ∧ · · · ∧ dxI(17.17)
p
.

The indices on this are a permutation of {1, · · · , n} , so is ±1.

But this sign is the same as that for the permutation to bring the
basis to the order dx1 ∧ · · · ∧ dxn , so the overall sign to get both to
69
the standard order is positive. Thus we get

p X I ···I
?ω ∧ ω = |g| ω 1 p ωI1 ···Ip 1···n dx1 ∧ · · · ∧ dxn
I
p 1
= |g| ω µ1 ···µp ωµ1 ···µp dx1 ∧ · · · ∧ dxn
p!
1 µ1 ···µp
= ω ωµ1 ···µp (vol) (17.18)
p!
If we are in a basis where the metric is not diagonal, it is still
symmetric. So we can diagonalize it locally by going to an appro-
priate basis, or set of coordinates, at each point. In this basis, the
components of ω may be ωµ0 1 ···µp , so we can write

1 µ01 ···µ0p
?ω ∧ ω = ω ωµ01 ···µ0p (vol0 ) (17.19)
p!
But both factors are invariant under a change of basis. So we can
now change back to our earlier basis, and find Eq. (17.18) even when
the metric is not diagonal. Note that the metric may not be diago-
nalizable globally or even in an extended region.

Chapter 18
Maxwell equations
We will now consider a particular example in physics where differen-

tial forms are useful. The Maxwell equations of electrodynamics are,
with c = 1 ,
∇·E =ρ (18.1)
∂E
∇×B− =j (18.2)
∂t
∇·B =0 (18.3)
∂B
∇×E+ = 0. (18.4)
∂t
The electric and magnetic fields are all vectors in three dimensions,
but these equations are Lorentz-invariant. We will write these equa-
tions in terms of differential forms.
Consider R4 with Minkowski metric gµν = diag(−1, 1, 1, 1) . For
the magnetic field define a 2-form
B = Bx dy ∧ dz + By dz ∧ dx + Bz dx ∧ dy . (18.5)
For the electric field define a 1-form
E = Ex dx + Ey dy + Ez dz . (18.6)

Combine these two into a 2-form F = B + E ∧ dt . Let us calculate

dF = d(B + E ∧ dt) = dB + dE ∧ dt . As usual, We will write 1, 2, 3
70
71
for the component labels x, y, z .

dB = d(B1 dy ∧ dz + B2 dz ∧ dx + B3 dx ∧ dy)
= ∂t B1 dt ∧ dy ∧ dz + ∂1 B1 dx ∧ dy ∧ dz
+∂t B2 dt ∧ dz ∧ dx + ∂2 B2 dy ∧ dz ∧ dx
+∂t B3 dt ∧ dx ∧ dy + ∂3 B3 dz ∧ dx ∧ dy . (18.7)
And
d(E ∧ dt) = d(E1 dx ∧ dt + E2 dy ∧ dt + E3 dz ∧ dt)
= ∂2 E1 dy ∧ dx ∧ dt + ∂3 E1 dz ∧ dx ∧ dt
+∂1 E2 dx ∧ dy ∧ dt + ∂3 E2 dz ∧ dy ∧ dt
+∂1 E3 dx ∧ dz ∧ dt + ∂2 E3 dy ∧ dz ∧ dt . (18.8)
Thus, remembering that the wedge product changes sign under each
exchange, we can combine these two to get
dF = (∂t B1 + ∂2 E3 − ∂3 E2 ) dt ∧ dy ∧ dz
+ (∂t B2 + ∂1 E3 − ∂3 E1 ) dt ∧ dz ∧ dx
+ (∂t B3 + +∂1 E2 − ∂2 E1 ) dt ∧ dx ∧ dy
+ (∂1 B1 + ∂2 B2 + ∂3 B3 ) dx ∧ dy ∧ dz
= (∂t B1 + (∇ × E)1 ) dt ∧ dy ∧ dz
+ (∂t B2 + (∇ × E)2 ) dt ∧ dz ∧ dx
+ (∂t B3 + (∇ × E)3 ) dt ∧ dx ∧ dy
+ (∇ · B) dx ∧ dy ∧ dz . (18.9)
Thus two of Maxwell’s equations are equivalent to dF = 0 .
For the other two equations we need ?F . Using the formula
(17.11) for dual basis forms, it is easy to calculate that
?(dx ∧ dy) = dt ∧ dz , ?(dy ∧ dz) = dt ∧ dx , ?(dz ∧ dx) = dt ∧ dy ,
?(dx ∧ dt) = dy ∧ dz , ?(dy ∧ dt) = dz ∧ dx , ?(dz ∧ dt) = dx ∧ dy .
(18.10)
We use these to calculate

?F = ?(B + E ∧ dt)
= B1 dt ∧ dx + B2 dt ∧ dy + B3 dt ∧ dz
+E1 dy ∧ dz + E2 dz ∧ dx + E3 dx ∧ dy . (18.11)
72 Chapter 18. Maxwell equations
Then in the same way as for the previous calculation, we find
d?F = (∇ · E) dx ∧ dy ∧ dz
+ (∂t E1 − (∇ × B)1 ) dt ∧ dy ∧ dz
+ (∂t E2 − (∇ × B)2 ) dt ∧ dz ∧ dx
+ (∂t E3 + (∇ × B)3 ) dt ∧ dx ∧ dy . (18.12)
We need to relate this to the charge-current.

Define the current four-vector as
j µ ∂µ = ρ∂t + j 1 ∂1 + j 2 ∂2 + j 3 ∂3 . (18.13)
Then there is a corresponding one-form jµ dxµ with jµ = gµν j ν . So

in terms of components,
jµ dxµ = −ρdt + j1 dx1 + j2 dx2 + j3 dx3 . (18.14)
Then using Eq. (17.11) it is easy to calculate that
?j = −ρ dx ∧ dy ∧ dz + j1 dt ∧ dy ∧ dz
+j2 dt ∧ dz ∧ dx + j3 dt ∧ dx ∧ dy . (18.15)
Comparing this equation with Eq. (18.12) we find that the other two
Maxwell equations can be written as
d?F = −?j . (18.16)
Finally, using Eq. (17.18), we see that the action of electromagnetism

can be written as
Z
1
− F ∧ ?F (18.17)
2
This expression holds in both flat and curved spacetimes. For the
latter, with local coordinates (t, x, y, z) we find
√
F ∧ ?F = (B 2 − E 2 ) −g dt ∧ dx ∧ dy ∧ dz . (18.18)

Chapter 19
Stokes’ theorem
We will next discuss a very beautiful result called Stokes’ formula.

This is actually a theorem, but we will not prove it, only state the
result and discuss its applications. So for us it is only a formula, but
still deep and beautiful.
• A submanifold S is a subset of points in M such that any point
in S has an open neighbourhoood in M for which there is some chart
where (n − m) coordinates vanish. S is then m-dimensional. 2
• Suppose U is a region of an oriented manifold M . The bound
ary ∂U of U is a submanifold of dimension n − 1 which divides M
in such a way that any curve joining a point in U with a point in U c
must contain a point in ∂U .
Now suppose U has an oriented smooth boundary ∂U . Then ∂U
is automatically an oriented manifold, by considering the restrictions
of the charts on U to ∂U .
• Consider a smooth (n − 1) form in M . Stokes’ formula says
that
Z Z
dω = ω . (19.1)
U ∂U
If M is a compact manifold with boundary ∂M , this formula can

be applied to all of M . If ω vanishes outside some compact region
we can again set U = M . Also, U can be a submanifold in another
manifold, like a 2-surface in a 3-manifold. 2

Example: Let U = [0, 1] . Then a function f : M → R is a

0-form, and df = f 0 (x)dx is a 1-form. Take the orientation of M to
be from 0 to 1. Then ∂M consists of the points x = 0 and x = 1 ,
73
74 Chapter 19. Stokes’ theorem
and Stokes’ formula says that

Z Z
df = f
M ∂M
Z1
i.e. f 0 (x)dx = f (1) − f (0) . (19.2)
0
Example: Consider a 2-d disk D in R2 , with boundary ∂D .

Take a 1-form A . Then Stokes’ formula says
Z Z
A = dA . (19.3)
∂D D
Let us seee this equation in a chart. We can write
A = Ai dxi
dA = ∂i Aj dxi ∧ dxj (19.4)

d d
A evaluated on ∂D can be written as A where is tangent
dt dt
dxi

d
to ∂D . So we can write A = Ai dt , and
dt dt
Z Z
i
Ai dx = ∂i Aj dxi ∧ dxj
∂D D
Z
= (∂1 A2 − ∂2 A1 ) dx1 ∧ dx2
D
Z
= (∂1 A2 − ∂2 A1 ) d2 x . (19.5)
ϕ(D)
Similarly for higher forms on higher dimensional manifolds.

• Gauss’ divergence theorem is a special case of Stokes’ the-
orem. Before getting to Gauss’ theorem, we need to make a new
definition. Consider an n-form ω 6= 0 on an n-dimensional manifold.
We can write this in a chart as

ω = f dx1 ∧ · · · ∧ dxn
1
= f µ1 ···µn dxµ1 ∧ · · · ∧ dxµn . (19.6)
n!
75
Given a vector field v , its contraction with ω is

1
ιv ω = ω(v, · · · ) = ωµ µ ···µ v µ1 dxµ2 ∧ · · · ∧ dxµn
(n − 1)! 1 2 n
= f v 1 dx2 ∧ · · · ∧ dxn − f v 2 dx1 ∧ dx3 ∧ · · · ∧ dxn + · · ·
(19.7)
Then we can calculate
d(ιv ω) = dω(v, · · · ) = ∂1 (f v 1 ) dx1 ∧ dx2 ∧ · · · ∧ dxn
+∂2 (f v 2 ) dx1 ∧ dx2 ∧ · · · ∧ dxn
+ · · · + ∂n (f v n ) dx1 ∧ dx2 ∧ · · · ∧ dxn
= ∂µ (f v µ ) dx1 ∧ dx2 ∧ · · · ∧ dxn
1
= ∂µ (f v µ ) ω . (19.8)
f
In particular, if ω is the volume form, we can write
p
|g|
ω = µ1 ···µn dxµ1 ∧ · · · ∧ dxµn ,
n!
1 p
d(ιv (vol)) = p ∂µ (v µ |g|)(vol) . (19.9)
|g|
• This is called the divergence of the vector field v .
There is another expression for the divergence. Remember that
given a vector field v , we can define a one-form, also called v , with
components defined with the help of the metric,
0
vµ = gµµ0 v µ (19.10)
Consider ?v , which has components
p 0
(?v)µ1 ···µn−1 = |g|µ1 ···µn−1 µ g µµ vµ0
p
= |g|µ1 ···µn−1 µ v µ . (19.11)
p
|g|
⇒ ?v = µ ···µ µ v µ dxµ1 ∧ · · · ∧ dxµn−1
(n − 1)! 1 n−1
p !
|g|
d?v = ∂µn µ ···µ µ v µ dxµn ∧ dxµ1 ∧ · · · ∧ dxµn−1
(n − 1)! 1 n−1

(−1)n−1 p
= µ1 ···µn−1 µ ∂µn |g|v µ dxµ1 ∧ · · · ∧ dxµn
(n − 1)!
(19.12)
76 Chapter 19. Stokes’ theorem
Both µ and µn must be different from (µ1 , · · · , µn−1 ) , so µ = µn .

Thus in each term of the sum, the choice of (µ1 , · · · , µn−1 ) auto-
matically selects µn (= µ) , so a sum over (µ1 , · · · , µn ) overcounts n
times. So we can write
(−1)n−1 p
d?v = µ1 ···µn ∂µ |g|v µ dxµ1 ∧ · · · ∧ dxµn
(n − 1)!
1 p
= (−1)n−1 p ∂µ ( |g|v µ )(vol) . (19.13)
|g|
Since this is an n-form in n dimensions, we can calculate from

here that
(−1)n+s−1 p
?d?v = p ∂µ ( |g|v µ ) , (19.14)
|g|
where as before s is the signature of the manifold, i.e. the number

of negative entries in the metric in a locally diagonal form.
Let us now go back to Stokes’ formula. Take a region U of M
which is covered by a single chart and has an orientable boundary
∂U as before. Then we find
Z Z
1 p
p ∂µ ( |g|v µ )(vol) = d(ιv (vol))
|g|
U U
Z
= ιv (vol) . (19.15)
∂U

d
Now suppose b is a 1-form normal to ∂U , i.e. b = 0 for any
dt
d
vector tangent to ∂U , and α is an (n − 1)-form such that b ∧ α =
dt
(vol) . Since all n-forms are proportional, α always exists. For the
same reason, if b 6= 0 on ∂U , it is unique up to a factor. And b 6= 0 on
∂U because ∂U is defined as the submanifold where one coordinate is
d
constant, usually set to zero, so that one component of vanishes
dt
at any point on ∂U , and therefore the corresponding component of
b can be chosen to be non-zero.

So b is unique up to a rescaling b → b0 = f b for some nonvanishing

−1
function f . But we can always scale α → α0 = f α so that b0 ∧ α0 =
b ∧ α . Further, if we restrict α to ∂U , i.e. if α acts only on tangent
77
vectors to ∂U , we find that α is an (n − 1)-form on an (n − 1)-

dimensional manifold, so it is unique up to scaling. Therefore, α is
unique once b is given. Finally, for any vector v ,

ιv (vol) = ιv (b ∧ α) (19.16)

∂U ∂U
is an (n − 1)-form on ∂U which acts only on vectors tangent to ∂U .

Then

ιv (b ∧ α) = b(v)α (19.17)

∂U ∂U
because all terms of the form b ∧ ιv α gives zero for any choice of
(n − 1) vectors on ∂U .
Then we have
Z Z
1 p µ
p ∂µ ( |g|v )(vol) = b(v)α
|g|
U ∂U
Z
= (nµ v µ ) α . (19.18)
∂U
Usually b is taken to have norm 1. Then α is the volume form on

∂U , and we can write
Z Z
1 p q
p ∂µ ( |g|v µ )(vol) = (nµ v µ ) |g(∂U ) |dn−1 x . (19.19)
|g|
U ∂U

Chapter 20
Lie groups
We start a brief discussion on Lie groups, mainly with an eye to

their structure as manifolds and also application to the theory of
fiber bundles.
• A Lie group is a group which is also an analytic manifold. 2
We did not define a Lie group in this way in Chap. 2 , but said that
a Lie group was a manifold in which the group product is analytic in
the group parameters, or alternatively the group product and group
inverse are both C ∞ .
The definition above comes from a theorem that given a con-
tinuous group G in which the group product and group inverse are
C ∞ functions of the group parameters, it is always possible to find
a set of coordinate charts covering G such that the overlap functions
are real analytic, i.e. are C ∞ and their Taylor series at any point
converge to their respective values.
• A Lie subgroup of G is a subset H of G which is a subgroup of
G , a submanifold of G , and is a topological group, i.e., a topological
space in which the group product and group inverse are continuous
maps. 2
• Sometimes this expressed in terms of another definition. P is
an immersed submanifold of M if the inclusion map j : P ,→ M
is smooth and at each point p ∈ P its differential djp is one to one,
with djp being defined by djp : Tp P → Tj(p) M such that djp (v)(g) =
v(g · jp ) . 2
We have mentioned some specific examples of Lie groups earlier.

Let us mention some more examples.

Example: Rn is a Lie group under addition. So is Cn .
78
79
Example: Rn \{0} is a Lie group under multiplication. So is

Cn \{0} .
Example: The direct product of two Lie groups is itself a Lie
group, with multiplication (g1 , h1 )(g2 , h2 ) = (g1 g2 , h1 h2 ) . 2
Example: The set of all n × n real invertible matrices forms
a group under matrix multiplication, called the General Linear
group GL(n, R) . This is also the space of all invertible linear maps
of Rn to itself. We can similarly define GL(n, C) . 2
The next few examples are Lie subgroups of GL(n, R) .
Example: The Special Linear group SL(n, R) is the subset
of GL(n, R) for which all the matrices have determinant +1 , i.e.,
SL(n, R) = {A ∈ GL(n, R)| det A = 1} . One can define SL(n, C) in
a similar manner. 2
Example: The Orthogonal group O(n) = {R ∈ GL(n, R) |
R T R = I} . 2
Example: The Unitary group U (n) = {U ∈ GL(n, C) | U U = †
I} . 2
Example: The Symplectic group Sp(n) , defined as the sub-
group of U (2n) given by AT JA = J , where
−1n×n

0
J=
1n×n 0
2
Example: O(p, q) = {R ∈ GL(p+q, R) | RT ηp,q R = ηp,q } , where
1p×p

0
ηp,q =
0 −1q×q
2
Example: U (p, q) = {U ∈ GL(p + q, C) | U † ηp,q U = ηp,q } .
Example The Special Orthogonal group SO(n) is the sub-
group of O(n) for which determinant is +1. Similarly, the Special
unitary group SU (n) is the subgroup of U (n) with determinant
+1. Similarly for SO(p.q) and SU (p, q) .
The group U (1) is the group of phases U (1) = {eiφ |φ ∈ R} . As a
manifold, this is isomorphic to a circle S 1 .

The group SU (2) is isomorphic as a manifold to a three-sphere

S 3 . These are the only two spheres (other than the point S 0 ) which
admit a Lie group structure.
80 Chapter 20. Lie groups
An important property of a Lie group is that the tangent space

of any point is isomorphic to the tangent space at the identity by
an appropriate group operation. Of course, the tangent space at any
pooint of a manifold is isomorphic to the tangent space at any other
point. For Lie groups, the isomorphism between the tangent spaces
is induced by group operations, so is in some sense natural.
For any Lie group G , we can define diffeomorphisms of G labelled
by elements g ∈ G , called
• Left translation lg : G → G g 0 7→ gg 0 ; 2
• Right translation rg : G → G g 0 7→ g 0 g . 2
These can be defined for any group, but are diffeomorphisms for
Lie groups. We see that
lg−1 lg (g 0 ) = lg−1 (gg 0 ) = g −1 gg 0 = g 0 ⇒ (lg )−1 = lg−1

rg−1 rg (g 0 ) = rg−1 (g 0 g) = g 0 gg −1 = g 0 ⇒ (rg )−1 = rg−1 . (20.1)
It is easy to check that
lg1 lg2 = lg1 g2 rg1 rg2 = rg2 g1 (20.2)
Further, lg−1 (g) = e and rg−1 (g) = e , so any element of G can be

moved to the identity by a diffeomorphism. The tangent space at
the identity forms a Lie algebra, as we shall see. The left and right
translations lead to diffeomorphisms which relate the tangent space
at any point to this Lie algebra, as we shall see now.

Chapter 21
Tangent space at the identity
A point on the Lie group is a group element. So a vector field on

the Lie group selects a vector at each g ∈ G . Since left and right
translations are diffeomorphisms, we can consider the pushforwards
due to them.
• A left-invariant vector field X is invariant under left ttrans-
lations, i.e.,
X = lg∗ (X) ∀g ∈ G . (21.1)
In other words, the vector (field) at g 0 is pushed forward by lg to the

same vector (field) at lg (g 0 ):
lg∗ (Xg0 ) = Xgg0 ∀g, g 0 ∈ G . (21.2)
• Similarly, a right-invariant vector field X is defined by
X = rg∗ (X) ∀g ∈ G ,
i.e. rg∗ (Xg0 ) = Xg0 g ∀g, g 0 ∈ G . (21.3)
A left or right invarian vector field has the important property

that it is completely determined by its value at the identity element
e of the Lie group, since
lg∗ (Xe ) = Xg ∀g ∈ G , (21.4)
and similarly for right-invariant vector fields.

Write the set of all left-invariant vector fields on G as L(G) . Since

the push-forward is linear, we get
lg∗ (aX + Y ) = alg∗ X + lg∗ Y , (21.5)
81
82 Chapter 21. Tangent space at the identity
so that if both X and Y are left-invariant,
lg∗ (aX + Y ) = aX + Y , (21.6)
so the set of left-invariant vector fields form a real vector space.

We also know that push-forwards leave the Lie algebra invariant,
i.e., for lg∗ ,
[lg∗ X , lg∗ Y ] = lg∗ [X , Y ] . (21.7)
Thus if X, Y ∈ L(G) ,
lg∗ [X , Y ] = [lg∗ X , lg∗ Y ] = [X , Y ] , (21.8)
so [X , Y ] ∈ L(G) . Thus the set of all left-invariant vector fields on

G forms a Lie algebra.
• This L(G) is called the Lie algebra of G . 2
The dimension of this Lie algebra is the same as that of G because
of the
Theorem: L(G) as a real vector space is isomorphic to the tan-
gent space Te G to G at the identity of G .
Proof: We will show that left translation leads to an isomor-
phism.
For X ∈ Te G , define the vector field LX on G by
LX g ≡ LX

g := lg∗ X ∀g ∈ G (21.9)
Then for all g, g 0 ∈ G ,
lg0 ∗(LX X
g ) = lg 0 ∗ (lg∗ X) = lg 0 g∗ X = Lg 0 g . (21.10)
Note that for two diffeomorphisms ϕ1 , ϕ2 , we can write
(ϕ1∗ (ϕ2∗ v))(f ) = (ϕ2∗ v)(f ◦ ϕ1 )

= v(f ◦ ϕ1 ◦ ϕ2 )
= ((ϕ1 ◦ ϕ2 )∗ v)(f )
⇒ ϕ1∗ (ϕ2∗ v) = (ϕ1 ◦ ϕ2 )∗ v (21.11)

Since left translation is a diffeomorphism,
lg0 ∗ (lg∗ X) = (lg0 ◦ lg )∗ X = (lg0 g∗ )X (21.12)

83
So it follows that LX is a left-invariant vector field, and we have a

map Te G → L(G) . Since the pushforward is a linear map, so is the
map X → LX . We need to prove that this map is 1-1 and onto.
If LX = LY , we have
LX Y
g = Lg ∀g ∈ G , (21.13)
so
lg−1 ∗ LX Y
g = lg −1 ∗ Lg ⇒ X=Y (∈ Te G) . (21.14)
So the map X → LX is 1-1.

Now given LX , define Xe ∈ Te G by
Xe = lg−1 ∗ LX
g for any g ∈ G . (21.15)
We can also write
Xe = LX
e . (21.16)
Then
lg∗ Xe = lg∗ lg−1 ∗ LX X

g = Lg . (21.17)
So the map X 7→ LX is onto.

Then we can define a Lie bracket on Te G by
[u , v] = [Lu , Lv ]e .

(21.18)
The Lie algebra of vectors in Te G based on this bracket is thus the

Lie algebra of the group G . It follows that
dim L(G) = dim Te G = dim G . (21.19)
Note that since commutators are defined for vector fields and not
vectors, the Lie bracket on Te G has to be defined using the com-
mutator of left-invariant vector fields on G and the isomorphism
Te G ↔ L(G) .
• If for an n-dimensional Lie group G , {t1 , · · · , tn } is a set of basis
vectors on Te G ' L(G) , the Lie bracket of any pair of these vectors

must be a linear combination of them, so

X
k
[ti , tj ] = Cij tk (21.20)
k
84 Chapter 21. Tangent space at the identity
for some set of real numbers Cijk . These numbers are known as the
structure constants of the Lie group or algebra. 2

Since L(G) is a Lie algebra, with the Lie bracket as the product,
the Lie bracket is antisymmetric,
[ti , tj ] = [tj , ti ]
X X
k k
⇒ Cij tk = Cji tk
k k
k k
⇒ Cij = Cji , (21.21)
and the structure constants satisfy the Jacobi identity
[ti , [tj , tk ]] + [tj , [tk , ti ]] + [tk , [ti , tj ]] = 0

l m l
⇒ Cij Ckl + Cjk l
Cilm + Cki m
Cjl = 0. (21.22)
A similar construction can be done using a set of right-invariant

vector fields defined by
RgX := rg∗ X for X ∈ Te G (21.23)
and its ‘inverse’ Xe = rg−1 ∗ RgX .

Chapter 22
One parameter subgroups
There is another characterization of Te G for a Lie group G as the set

of its one parameter subgroups, which we will now define. This is
also called the “infinitesimal” description of a Lie group, and what
Lie called an infinitesimal group.
• A one parameter subgroup of a Lie group G is a smooth
homomorphism from the additive group of real numbers to G , γ :
(R, +) → G . Then γ : R → G is a curve such that γ(s + t) =
γ(s)γ(t) , γ(0) = e , and γ(−t) = γ(t)−1 . 2
Also, since this is a homomorphism, the one parameter subgroup
is Abelian.
Example: For G = (R\{0}, ×) the multiplicative group of non-
zero real numbers, γ(t) = et is a 1-p subgroup.
Example: G = U (1) , γ(t) = eit.
cos t sin t
Example: G = SU (2) , γ(t) = .
−
sin t cos t 
cos t sin t 0
Example: G = GL(3, R) , γ(t) =  − sin t cos t 0  .
0 0 et
The relation between 1-p subgroups and and Te G is given by the
Theorem: The map γ 7→ γ̇(0) = γ̇ e defines a 1-1 correspon-
dence between 1-p subgroups of G, and Te G .
Proof: For any X ∈ Te G define LX = lg∗ X as the corresponding
left-invariant vector field. We need to find a smooth homomorophism
from R to G using LX . This homomorphism is provided by the flow

or integral curve of LX , but let us work this out in more detail.
85
86 Chapter 22. One parameter subgroups
Denote the integral curve of LX by γ X (t) , i.e.,
γ X (0) = LX
e =X
and γ̇ X (t) = LX
γ(t) = lγ(t)∗ X (22.1)
Since LX is left-invariant, lg0 ∗ LX X

g = Lg 0 g . Consider the equation

d X d
γ(t) = Lγ(t) = lγ(t)∗ X ≡ γ∗ . (22.2)
dt dt t
Given some τ , replace γ(t) by γ(τ + t) to get
γ(τ + t) = lγ(τ +t)∗ X . (22.3)
Remember that γ(t) is an element of the group for each t . Now

replace γ(t) in Eq. (22.2) by γ(τ )γ(t) to get

d
= LX

γ(τ )γ(t) ∗ γ(τ )γ(t) (22.4)
dt
We see that γ(t + τ ) and γ(τ )γ(t) are both integral curves of LX , i.e.
both satisfy the equation of the integral curve of LX , and at t = 0
both curves are at the point γ(τ ) . Thus by uniqueness these two are
the same curve,
γ(τ + t) = γ(τ )γ(t) , (22.5)
and t 7→ γ(t) is the homomorphism R → G that we are looking for.

Thus for each X ∈ Te G we find a 1-p subgroup γ(t) given by the
integral curve of LX ,
γ̇(t) = LX
γ(t) = lγ(t)∗ X , (22.6)
where, as mentioned earlier, (γ̇(0) = X) . 2

In a compact connected Lie group G , every element lies on some
1-p subgroup. This is not true in a non-compact G , i.e. there are
elements in G which do not lie on a 1-p subgroup. However, an
Abelian non-compact group will always have a 1-p subgroup, so this
remark applies only to non-Abelian non-compact groups.

For matrix groups, every 1-p subgroup is of the form

n o
γ(t) = etM M fixed, t ∈ R . (22.7)

87
Let us see why. Suppose {γ(t)} is a 1-p subgroup of the matrix

group. Then γ(t) is a matrix for each t , and
γ(s)γ(t) = γ(s + t) . (22.8)
Differentiate with respect to s and set s = 0 . Then
γ̇(0)γ(t) = γ̇(t) . (22.9)
Write γ̇(0) = M . Since G is a matrix group, M is a matrix. Then

the unique solution for γ is
γ(t) = etM . (22.10)
The properties of M are determined by the properties of the group

and vice versa, not every matrix M will generate any group.
The allowed matrices {M } for a given group G are the {γ̇(0)} for
all the 1-p subgroups γ(t) , so these are in fact the tangent vectors
at the identity. The allowed matrices {M } for a given matrix group
G thus form a Lie algebra with the Lie bracket being given by the
matrix commutator. This Lie algebra is isomorphic to the Lie algebra
of the group G . (We will not a give a proof of this here.)
We can find the Lie algebra of a matrix group by considering
elements of the form γ(t) = etM for small t , i.e.,
γ(t) = I + tM (22.11)
for small t . Conversely, once we are given, or have found, a Lie

algebra with basis {ti } , we can exponentiate the Lie algebra to find
the set of 1-p subgroups
γ(a) = exp ai ti

(22.12)
• This is the infinitesimal group, for compact connected groups

this is identical to the Lie group itself. So in such cases, the entire
group can be generated by exponentiating the Lie algebra. Non-
compact groups cannot be written as the exponential of the Lie al-
gebra in general. 2
Example: Consider SO(N ) , the group of N ×N real orthogonal

matrices R with RT R = I , det R = 1 . Write R = I + A , then AT =

−A , i.e. the Lie algebra is spanned by N × N real antisymmetric
matrices. Let us construct a basis for this algebra.
88 Chapter 22. One parameter subgroups
An N × N antisymmetric matrix has N (N − 1)/2 independent

elements. So we define N (N − 1)/2 independent antisymmetric ma-
trices, labelled by µ, ν = 1, · · · , N ,
Mµν = −Mνµ µ , ν are not matrix indices

(Mµν )ρσ = (Mµν )σρ , ρ , σ are matrix indices . (22.13)
A convenient choice for the basis is given by
(Mµν )ρσ = δµρ δνσ − δµσ δνρ . (22.14)
Then the commutators are calculated to be
[Mµν , Mαβ ] = δνα Mµβ − δµα Mνβ + δµβ Mνα − δνβ Mµα . (22.15)
This defines the Lie algebra.

Example: For SU (N ) , the group of N × N unitary matri-
ces U with U † U = I , det U = 1 , the 1-p subgroups are given by
γ(t) = etM with M † + M = 0 in the same way as above, and
det(I + tM ) = 1 ⇒ Tr M = 0 . So the SU (N ) Lie algebra consists
of traceless antihermitian matrices. Often the basis is multiplied by
i to write γ(a) = exp(iaj tj ) , where tj are now Hermitian matrices,
with
[ti , tj ] = ifabc tc . (22.16)

Chapter 23
Fiber bundles
S
Consider a manifold M with the tangent bundle T M = TP M .
P ∈M
Let us look at this more closely. T M can be thought of as the original
manifold M with a tangent space stuck at each point P ∈ M . Thus
there is a projection map π : T M → M , TP M 7→ P , which
associates the point P ∈ M with TP M .
Then we can say that T M consists of pooints P ∈ M and vectors
v ∈ TP M as an ordered pair (P, vP ) . Then in the neighbourhood of
any point P , we can think of T M as a product manifold, i.e. as
the set of ordered pairs (P, vP ) .
This is generalized to the definition of a fiber bundle. Locally
a fiber bundle is a product manifold E = B × F with the following
properties.
• B is a manifold called the base manifold, and F is another
manifold called the typical fiber or the standard fiber.
• There is a projection map π : E → B , and if P ∈ B , the pre-
image π −1 (P ) is homeomorphic, i.e. bicontinuously isomorphic, to
the standard fiber. 2
E is called the total space, but usually it is also called the bun-
dle, even though the bundle is actually the triple (E, π, B) .
• E is locally a product space. We express this in the following
way. Given an open set Ui of B , the pre-image π −1 (Ui ) is homeomor-
phic to Ui ×F , or in other words there is a bicontinuous isomorphism
ϕi : π −1 (Ui ) → Ui × F . The set {Ui , ϕi } is called a local trivializa

tion of the bundle. 2

• If E can be written globally as a product space, i.e. E = B × F ,
it is called a trivial bundle. 2
89
90 Chapter 23. Fiber bundles
• This description includes a homeomorphism π −1 (P ) → F for

each P ∈ Ui . Let us denote this map by hi (P ) . Then in some overlap
Ui ∩ Uj the fiber on P , π −1 (P ) , has homeomorphisms hi (P ) and
hj (P ) onto F . It follows that hj (P ) · hi (P )−1 is a homeomorphism
F → F . These are called transition functions. The transition
functions F → F form a group, called the structure group of F . 2
Let us consider an example. Suppose B = S 1 . Then the tan-
gent bundle E = T S 1 has F = R and π(P, v) 7→ P , where
P ∈ S 1 , v ∈ T S 1 . Consider a covering of S 1 by open sets Ui , and
let the coordinates of Ui ⊂ S 1 be denoted by λi . Then any vector at
d
TP S 1 can be written as v = ai (no sum) for P ∈ Ui .
dλi
So we can define a homeomorphism hi (P ) : TP S 1 → R, v 7→
ai (fixed i). If P ∈ Ui ∩ Uj there are two such homeomorphisms
T S 1 → R , and since λi and λj are independent, ai and aj are also
independent.
Then hi (P ) · hj (P )−1 : F → F (or R → R) maps aj to ai .
The homeomorphism, which in this case relates the component of
the vector in two coordinate systems, is simply multiplication by the
ai
number rij = ∈ R\{0} . So the structure group is R\{0} with
aj
multiplication.
For an n-dimensional manifold M , the structure group of T M
is GL(n, R) .
• A fiber bundle where the standard fiber is a vector space is called
a vector bundle . 2
A cylinder can be made by glueing two opposite edges of a flat
strip of paper. This is then a Cartesian product of acircle S 1 with
a line segment I . So B = S 1 , F = I and this is a trivial bundle,
i.e. globally a product space. On the other hand, a Möbius strip is
obtained by twisting the strip and then glueing. Locally for some
open set U ( S 1 we can still write a segment of the Möbius strip as
U × I , but the total space is no longer a product space. As a bundle,
the Möbius strip is non-trivial.
• Given two bundles (E1 , π1 , B1 ) and (E2 , π2 , B2 ) , the relevant
or useful maps between these are those which preserve the bundle
structure locally, i.e. those which map fibers into fibers. They are

called bundle morphisms. 2

A bundle morphism is a pair of maps (F, f ) , F : E1 → E2 , f :
91
B1 → B2 , such that π2 ◦ F = f ◦ π1 . (This is of course better

understood in terms of a commutative diagram.)
Not all systems of coordinates are appropriate for a bundle. But it
is possible to define a set of fiber coordinates in the following way.
Given a differentiable fiber bundle with n-dimensional base manifold
B and p-dimensional fiber F , the coordinates of the bundle are given
by bundle morphisms onto open sets of Rn × Rp . 2
• Given a manifold M the tangent space TP M , consider AP =
(e1 , · · · , en ) , a set of n linearly independent vectors at P . AP is a
basis in TP M . The typical fiber in the frame bundle is the set of
all bases, F = {AP } . 2
Given a particular basis AP = (e1 , · · · , en ) , any basis AP may
be expressed as
ei = aji ej . (23.1)
The numbers aji can be thought of as the components of a matrix,

which must be invertible so that we can recover the original basis
from the new one. Thus, starting from any one basis, any other basis
can be reached by an n×n invertible matrix, and any n×n invertible
matrix produces a new basis. So there is a bijection between the set
of all frames in TP M and GL(n , R) .
Clearly the structure group of the typical fiber of the frame bun-
dle is also GL(n , R) .
• A fiber bundle in which the typical fiber F is identical (or home-
omorphic) to the structure group G, and G acts on F by left trans-
lation is called a principal fiber bundle. 2
Example: 1. Typical fiber = S 1 , structure group U (1) .
2. Typical fiber = S 3 , structure group SU (2) .
3. Bundle of frames, for which the typical fiber is GL(n , R) , as
is the structure group.
• A section of a fiber bundle (E, π, B) is a mapping s : B →
E , p 7→ s(p) , where p ∈ B , s(p) ∈ π −1 (p) . So we can also say π ◦ s
= identity. 2
Example: A vector field is a section of the tangent bundle, v :
P 7→ vP .

Example: A function on M is a section of the bundle which

locally looks like M × R (or M × C if we are talking about complex
functions).
• Starting from the tangent bundle we can define the cotangent

bundle, in which the typical fiber is the dual space of the tangent
space. This is written as T ∗ M . As we have seen before, a section of
T ∗ M is a 1-form field on M . 2
π
• Remember that a vector bundle F → E → B is a bundle in
which the typical fiber F is a vector space. 2
• A vector bundle (E, π e, B, F, G) with typical fiber F and struc-
ture group G is said to be associated to the principal bundle
(P, π, B, G) by the representation {D(g)} of G on F if its transi-
tion functions are the images under D of the transition functions of
P. 2
That is, suppose we have a covering {Ui } of B , and local trivial-
ization of P with respect to this covering is Φi : π −1 (Ui ) → Ui × G ,
which is essentially the same as writing Φi,x : π −1 (x) → G , x ∈ Ui .
Then the transition functions of P are of the form
gij = Φi ◦ Φj −1 : Ui ∩ Uj → G . (23.2)
The transition functions of E corresponding to the same covering of

B are given by φi : π e−1 (Ui ) → Ui × F with φi ◦ φj −1 = D(gij ) .
That is, if vi and vj are images of the same vector vx ∈ Fx under
overlapping trivializations φi and φj , we must have
vi = D (gij (x)) vj . (23.3)
A more physical way of saying this is that if two observers look at

the same vector at the same point, their observations are relatted by
a group transformation (p, v) ' (p, D(gij v) .
• These relations are what are called gauge transformations in
physics, and G is called the gauge group. Usually G is a Lie group
for reasons of continuity. 2
Fields appearing in various physical theories are sections of vector
bundles, which in some trivialization look like Uα × V where Uα is
some open neighborhood of the point we are interested in, and V is
a vector space. V carries a representation of some group G , usually
a Lie group, which characterizes the theory.
To discuss this a little more concretely, let us consider an associ-

ated vector bundle (E, π e, B, F, G) of a principal bundle (P, π, B, G).

Then the transition functions are in some representation of the group
G . Because the fiber carries a representation {D(g)} of G , there are
93
always linear transformations Tx : Ex → Ex which are members of

the representation {D(g)} . Let us write the space of all sections of
this bundle as Γ(E) . An element of Γ(E) is a map from the base
spacce to the bundle. Such a map assigns an element of V to each
point of the base space.
• We say that a linear map T : Γ(E) → Γ(E) is a gauge trans
formation if at each point x of the base space, Tx ∈ {D(g)} for
some g , i.e. if
Tx : (x, v)α 7→ (x, D(g)v)α , (23.4)
for some g ∈ G and for (x, v)α ∈ Uα × F . In other words, a gauge

transformation is a representation-valued linear transformation of
the sections at each point of the base space. The right hand side is
often written as (x, gv)α . 2
This definition is independent of the choice of Uα . To see this,
consider a point x ∈ Uα ∩ Uβ . Then
(x, v)α = (x, gβα v)β . (23.5)
In the other notation we have been using, vα and vβ are images of the
same vector vx ∈ Vx , and vβ = D(gβα )vα . A gauge transformation
T acts as
Tx : (x, v)α 7→ (x, gv)α . (23.6)
But we also have
(x, gv)α = (x, gβα gv)β (23.7)
using Eq. (23.5) . So it is also true that
Tx : (x, gβα v)β 7→ (x, gβα gv)β . (23.8)
Since F carries a representation of G , we can think of gv as a

change of variables, i.e. define v 0 = gβα v . Then Eq. (23.8) can be
written also as
Tx : (x, v 0 )β 7→ (x, g 0 v 0 )β , (23.9)

where now g 0 = gβα ggβα −1 . So T is a gauge transformation in Uβ as

well. The definition of a gauge transformation is independent of the
choice of Uα , but T itself is not. The set of all gauge transformations

G is a group, with
(gh)(x) = g(x)h(x) , (g −1 )(x) = (g(x)) −1 . (23.10)
• The groups G and G arre both called the gauge group by

different people. 2

Chapter 24
Connections
There is no canonical way to differentiate sections of a fiber bundle.

That is to say, no unique derivative arises from the definition of a
bundle. Let us see why. The usual derivative of a function on R is
of the form
f (x + ) − f (x)
f 0 (x) = lim . (24.1)
→0
But a section of a bundle assigns an element of the fiber at any point
P ∈ M to the base point P . Each fiber is isomorphic to the standard
fiber but the isomorphism is not canonical or unique. So there is no
unique way of adding or subtracting points on different fibers. Thus
there are many ways of differentiating sections of fiber bundles.
• Each way of taking a derivative, i.e. of comparing, is called a
connection. Let us consider a bundle π : E → B , where Γ(E) is the
space of all sections. Then a connection D on B assigns to every
vector field v on B a map Dv : Γ(E) → Γ(E) satisfying
Dv (s + αt) = Dv s + α Dv t
Dv (f s) = v(f )s + f Dv s
Dv+f w (s) = Dv s + f Dw s , (24.2)
where s, t are sections of the bundle, s, t ∈ Γ(E) , v, w are vector fields

on B , f ∈ C ∞ (B) and α is a real number (or complex, depending
on what the manifold is). 2

Note that this is a connection, not some unique connection.

In other words, we need to choose a connection before we can talk
about it. In what follows, whenever we refer to a connection D , we
95
96 Chapter 24. Connections
mean that we have chose a connection D and that is what we are

discussing.
• We call Dv s the covariant derivative of s . 2
To be specific, let us consider the bundle to be a vector bundle
on a manifold M , and try to understand the meaning of D by going
to a chart. Consider coordinates xµ in an open set U ⊂ M , with ∂µ
the coordinate basis vector fields. Write Dµ = D∂µ . Also choose a
basis for sections, which is like a basis for the fiber (vector space) at
each point of M , i.e. like a set of basis vector fields, but the vectors
are not along M , but along the fiber at M .
Call this basis {ei } , then {ei (x)} is a basis for the fiber at P ∈ M ,
with {x} being the set of coordinates at P . Any element of V ' Fx
can be written uniquely as a linear combination of ei (x) . But then
Dµ ej can be expressed uniquely as a linear combination of the ei ,
Dµ ej = Aµj i ei . (24.3)
• These Aiµj are called components of the vector potential or
the connection one-form. 2
Given a section s = si ei with si ∈ C ∞ (M) , we can write
Dv s = Dvµ ∂µ s = v µ Dµ s . (24.4)
Also,
Dµ s = Dµ (si ei ) = ∂µ si ei + si Dµ ei

= ∂µ si ei + si Aµij ej

= ∂µ si + Aµj i sj ei , (24.5)
so writing Dµ s = (Dµ s)i ei , we can say

(Dµ s)i = ∂µ si + Aµj i sj . (24.6)
We have considered connections on an associated vector bundle,
which may be a principal fiber bundle such as a frame bundle. So
we should be able to talk about gauge transformations.
Remember that a gauge transformation is a linear map T : E →
E , (x, v) 7→ (x, gv) for some g ∈ G and for all v ∈ V ' Fx . Let us

apply this idea to the section s . We claim that given a connection

D , there is a connection D0 on E such that
D0 (gφ) = gDv φ , (24.7)
97
where v is a vector field on M and φ ∈ Γ(E) (i.e. φ = s).

Let us first check if the definition makes sense. Since g(x) ∈ G
for all x ∈ M , we know that g −1 (x) exists for all x. So
Dv0 (φ) = Dv0 gg −1 φ

= gDv g −1 φ ,

(24.8)
and thus D0 is defined on all φ for which Dv φ is defined, i.e. D0 exists
because D does. We have of course assumed that g(x) is differentiable
as a function of x.
Let us now check that D0 is a connection according to our defi-
nitions. D0 is linear since
Dv0 (φ1 + αφ2 ) = gDv g −1 (φ1 + αφ2 )

= gDv g −1 φ1 + αgDv g −1 φ2

= Dv0 φ1 + αDv0 φ2 . (24.9)

And it satisfies Leibniz rule because
Dv0 (f φ) = gDv g −1 f φ

= gDv f g −1 φ

= gv(f )g −1 φ + gf Dv g −1 φ

= v(f )φ + f gDv g −1 φ

= v(f )φ + f Dv0 (φ) . (24.10)

Similarly,
0
φ = gDv+αw g −1 φ

Dv+αw
= g Dv g −1 φ + αDw g −1 φ

= gDv (g −1 φ) + αgDw g −1 φ

= Dv0 φ + αDw
0
φ. (24.11)
So D0 is a connection, i.e. there is a connection D0 satisfying
Dv0 (gφ) = g (Dv φ) .
Since φ is a section, i.e. φ ∈ Γ(E), so is g −1 φ, and thus
Dv g −1 φ ∈ Γ(E) and also gDv g −1 φ ∈ Γ(E) . Therefore, Dv0 maps
sections to sections, Dv0 : Γ(E) → Γ(E) . This completes the defini-

tion of the gauge transformation of the connection. We can now

write
Dµ0 φ = ∂µ φi + A0µij φj ei .

(24.12)
98 Chapter 24. Connections
Using the dual space, let us write
Aµ = Aµij ei ⊗ θj ,
A0µ = A0µij ei ⊗ θj , (24.13)
where θi is the dual basis to {ei } . The gauge transformation is

then given by
Dv0 φ = gDv g −1 φ

Dµ0 φ = gDµ g −1 φ

⇒
h i j i
∂µ φi + A0µij ei = g ∂µ g −1 φ + Aµij g −1 φ ei ,

⇒
(24.14)
where, as always, the g’s are in some appropriate representation of

G. Then we can write the right hand side as
h i i i
∂µ φi + g∂µ g −1 j φj + gAµ g −1 j φj ei
h i i
= ∂µ φi + g∂µ g −1 + gAµ g −1 j φj ei . (24.15)
From this we can read off
A0µ = gAµ g −1 + g∂µ g −1 . (24.16)
• A connection which transforms like this is also called a G-

connection. 2
Example: Consider G = U (1) . Suppose E is a trivial complex
line bundle over M , i.e. E = M×C , so that the fiber over any point
p ∈ M is C . A connection D on E may be written as Dµ = ∂µ + Aµ .
We can make E into a U (1) bundle by thinking of the fiber C as
the fundamental representation space of U (1) . Then sections are
complex functions, and a gauge transformation is multiplication by
a phase. 2

Chapter 25
Curvature
We start with a connection D , two vector fields v, w on B , and a

section s , all on some associated vector bundle of some principal
G-bundle E . Then Dv , Dw are both maps Γ(E) → Γ(E) .
We will define the curvature of this connection D as a rule F
which, given two vector fields v , w , produces a linear map F (v, w) :
Γ(E) → Γ(E) by
F (v, w)s = Dv Dw s − Dw Dv s − D[v,w] s . (25.1)
Remember that
D v s = D v µ ∂µ s = v µ D µ s

= v µ ∂µ si + Aµij sj ei
= v(si )ei + v µ Aµij sj ei . (25.2)
Dµ s is again a section. So we can act with D on it and write

h i
Dv Dw s = Dv w(sj )ej + wν Aν j k sk ej

= v w sj ej + v wν Aν j k sk ej

+ w sj + wν Aν j k sk v µ Aµij sj ei .

(25.3)

Since the connection components Aν j k are functions, we can write

v wν Aν j k sk = v(wν )Aν j k sk +wν v(Aν j k )sk +wν Aν j k v(sk ) . (25.4)
99
100 Chapter 25. Curvature
Inserting this into the previous equation and writing Dw Dv s simi-

larly, we find
Dv Dw s − Dw Dv s = [v , w](si )ei + [v , w]µ Aµij sj ei

+ wν v Aν ij − v µ w Aµij sj ei

+ v µ wν Aµij Aν j k − Aν ij Aµj k sk ei . (25.5)
Also,
D[v ,w] s = [v , w](si )ei + [v , w]µ Aµij sj ei , (25.6)
so that

F (v, w)s = v µ wν ∂µ Aν ij − ∂ν Aµij + Aµik Aν kj − Aν ik Aµkj sj ei .
(25.7)
Thus we can define Fµν by
F (∂µ , ∂ν )s = Fµν s = (Fµν s)i ei = (Fµν )ij sj ei , (25.8)
so that
(Fµν )ij = ∂µ Aν ij − ∂ν Aµij + Aµik Aν kj − Aν ik Aµkj . (25.9)
Note that since coordinate basis vector fields commute, [∂µ , ∂ν ] = 0 ,
Fµν s = F (∂µ , ∂ν )s = Dµ Dν s − Dν Dµ s = [Dµ , Dν ]s . (25.10)
• It is not very difficult to work out that the curvature acts linearly
on the module of sections,
F (u, v)(s1 + f s2 ) = F (u, v)s1 + f F (u, v)s2 , (25.11)
where f ∈ C ∞ (B) . Also,
F (u, v + f w) s = F (u, v)s + f F (u, w)s . (25.12)
• For coordinate basis vector fields [∂µ , ∂ν ] = 0 , so
Fµν s = F (∂µ , ∂ν )s = Dµ Dν s − Dν Dµ s = [Dµ , Dν ]s . (25.13)

Since F (∂µ , ∂ν )s is a section, so is
Dλ (Fµν s) = Dλ [Dµ , Dν ]s . (25.14)

101
Similarly, since Dλ s is a section, so is
Fµν Dλ s = [Dµ , Dν ]Dλ s . (25.15)
Thus
Dλ (Fµν s) − Fµν Dλ s = [Dλ , [Dµ , Dν ]]s . (25.16)
Considering C ∞ sections, and noting that maps are associative under
map composition, we find that
[Dλ , [Dµ , Dν ]]s + cyclic = 0 . (25.17)
On the other hand,

Fµν s = (Fµν s)i ei = Fµν ij sj ei , (25.18)
where Fµν ij and si are in C ∞ (B) . So we can write

Dλ (Fµν s) = ∂λ Fµν ij sj + Fµν ij sj Dλ ei

= ∂λ Fµν ij sj ei + Fµν ij ∂λ sj ei + Fµν ij sj Aλki ek

= ∂λ Fµν ij + Fµν kj Aλik sj ei + Fµν ij ∂λ sj ei

− Fµν ik Aλkj sj ei + Fµν ij Aλj k sk ei

= (Dλ Fµν )ij sj ei + Fµν ij (Dλ s)j ei , (25.19)
where we have defined (Dλ Fµν ) by (Dλ Fµν )ij in this. Then this is a
Leibniz rule,
Dλ (Fµν s) = (Dλ Fµν ) s + Fµν (Dλ s) . (25.20)
Then we can write
Dλ (Fµν s) − Fµν (Dλ s) + cyclic = 0

⇒ (Dλ Fµν ) s + cyclic = 0 ∀s
⇒ Dλ Fµν + cyclic = 0 . (25.21)
This is known as the Bianchi identity . 2

0
Given D and g such that g(x) ∈ G , we have D given by
Dv0 φ = gDv g −1 φ .

(25.22)
102 Chapter 25. Curvature
Then
Du0 Dv0 φ = Du0 gDv g −1 φ = gDu Dv g −1 φ ,

(25.23)
and thus

F 0 (u, v) φ ≡ Du0 Dv0 − Dv0 Du0 − D[u
0
,v] φ
= gDu Dv g −1 φ − Dv Du g −1 φ − gD[u ,v] g −1 φ

= gF (u, v)g −1 φ
0
⇒ Fµν = g ◦ Fµν ◦ g −1 . (25.24)
As before, g is in some representation of G , and D (and thus F ) acts

on the same representation. This is the meaning of the statement
that the curvature is gauge covariant . 2

(Lahiri) Lecture Notes On Differential Geometry PDF

Uploaded by

Document Information

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

(Lahiri) Lecture Notes On Differential Geometry PDF

Uploaded by

Copyright:

Available Formats

c Amitabha Lahiri: Lecture Notes on Differential Geometry for Physicists 2011

We start by defining a topological space.

ii) the intersection of a finite number of open sets is open; if

the union of any number of open sets is open, if Ui ∈ O, then

via the sets {(x1 , · · · , xn ) ∈ Rn | ai < xi < bi }. This is called the

• The discrete topology on a set S is defined by O = {A | A ⊂ S},

ii) the union of a finite number of open sets is open; if A1 , A2 ∈ C ,

iii) the intersection of any number of open sets is open, if Ai ∈ C ,

its inverse image (or pre-image, what it is an image of) f −1 (U ) ⊂ S1

the same as the usual  − δ definition of continuity, which says that

times differentiable. Smooth means k is large enough for the purpose

(x1 )2 + · · · + (xn+1 )2 = 1 (2.1)

It is not possible to cover the sphere by a single chart, but it is

the dimension of the group, n = dim G. (This is a different defi-

Then any element of the group can be written as g(a) where

time reflection) Lorentz transformations, is a 6-dimensional con-

Vectors on a manifold are to be thought in terms tangents to the

and the transition functions (ϕα ◦ ϕβ −1 ) are differentiable.

In other words, f is differentiable at P if the coordinates y i = f i (xµ )

both differentiable, we say that f is a diffeomorphism and that M

ii) Derivation approach: A vector is defined as a derivation of func-

iii) Curves approach: A vector tangent to a manifold is tangent to

Consider a smooth function f : M → R. Given a curve γ :

write using chain rule

This notation makes sense when we remember Eq. (3.9). Using it we

Note that ϕ ◦ γ : I → Rn , t 7→ (x1 (γ(t)), · · · , xn (γ(t))) are the coor-

The first factor is exactly as shown in Eq. (4.7), so we can write

i.e., we can write

to be thought of as tangents to the coordinate curves inϕ. These

vP (f + αg) = vP (f ) + αvP (g) (4.14)

• The dual space TP∗ M of TP M is the space of linear mappings

ω(uP + avP ) = ω(uP ) + aω(vP ) , (5.1)

a1 ω1 + a2 ω2 : vP 7→ a1 ω1 (vP ) + a2 ω2 (vP ) . (5.2)

• The elements of TP∗ M are called dual vectors, covectors,

Vector Dual vector

• Given a function on a manifold f : M → R , every vector at

covector df , given by df (vP ) = vP (f ) called the differential or

The differentials dxi are covectors, as we already know. So we

Also, as we have just seen, any covector ω ∈ TP∗ M can be written

so in particular for ω = df , we get

This justifies the name gradient.

ω a 7→ ω 0a = Aab ω b , ea 7→ e0a = (A−1 )ba eb , (5.11)

with A a non-singular matrix, so that ω 0a (e0b ) = δba .

λ = λa ω a = λ0a ω 0a = λ0a Aab ω a , (5.12)

from which it follows that λ0a = (A−1 )ba λb .

Similarly, if v is a vector, we can write

v = v a ea = v 0a e0a = v 0a (A−1 )ba eb , (5.13)

and it follows that v a = Aab v b .

• Consider the (disjoint) union of tangent spaces at all points,

This is called the tangent bundle of M. 2

equations guarantees, at least for small t (i.e. locally), the existence

We could also choose γ(t0 ) = P.

• This φ defines a map φt : U → M at each t by φt (Q) = φ(t, Q) =

the curve γQ (t). This φt is called the local flow of v. 2

ii) φs ◦ φt = φs+t for all s, t, s + t ∈ U ;

iii) each flow is a diffeomorphism with φt −1 = φ−t .

v(f + αg) = v(f ) + αv(g) (6.8)

(u + αv)(f ) = u(f ) + αv(f ), u, v ∈ V (M), α ∈ R.

It is possible to replace α by some function on C ∞ (M). If u, v

(u + αv)P (f ) = uP (f ) + α(P )vP (f ) ∀f ∈ C ∞ (M), P ∈ M.

This looks like a vector space but actually it is what is called a

multiplicative identity elements, 0 + x = x + 0 = x , 1x = x1 =

• A module becomes a vector space when this ring is a commu

the same as the usual − δ definition of continuity, which says that

The commutator is useful