Advanced mathematical

methods.
Enrique Alvarez

VI Maii MMXVIII
2
Index

1 Curves and surfaces in R3 . 5

1.1 Curves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Surfaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Tensor calculus in vector spaces 13

2.1 Differential forms. . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 The metric tensor. . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Winding numbers and such. . . . . . . . . . . . . . . . . . . . 27

3 Gauss’ integral 35

4 Surfaces revisited. 37

5 Differential Geometry 41
5.1 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.2 Covariant derivative and curvature. . . . . . . . . . . . . . . . 46
5.3 Differential manifolds . . . . . . . . . . . . . . . . . . . . . . . 47
5.4 The two-sphere, S2 . . . . . . . . . . . . . . . . . . . . . . . 50

6 Moving frames and curvature. 53

7 The Gauss-Codazzi equations 61

8 Distributions 65
8.1 Fourier transform . . . . . . . . . . . . . . . . . . . . . . . . . 71
8.2 Distributions on submanifolds. . . . . . . . . . . . . . . . . . 75

9 Finite groups. 79
9.1 Normal subgroups . . . . . . . . . . . . . . . . . . . . . . . . 83
9.2 Schur’s lemma . . . . . . . . . . . . . . . . . . . . . . . . . . 86
9.3 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
9.4 Partitions and representations of Sn . . . . . . . . . . . . . . 95

3
10 Lie groups. 103
10.1 Matrix groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
10.2 Representations of SU p2q and SOp3q through tensor methods. 109
10.3 Representations of GLpnq through tensor methods. . . . . . . 111
10.4 Representations of U pnq . . . . . . . . . . . . . . . . . . . . . 113
10.5 Representations of Opnq . . . . . . . . . . . . . . . . . . . . . 114

11 The rotation group SOp3q „ SU p2q{Z2 . 119

11.1 The Lie algebra SUp2q . . . . . . . . . . . . . . . . . . . . . . 125
11.2 Highest weight representations of SUp2q . . . . . . . . . . . . 126
11.3 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . . 128
11.4 Spinor representations . . . . . . . . . . . . . . . . . . . . . . 130
11.5 Product representations . . . . . . . . . . . . . . . . . . . . . 133
11.6 Wigner-Eckart . . . . . . . . . . . . . . . . . . . . . . . . . . 135

12 Roots and weights 137

12.1 SU(3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
12.2 Dynkin diagrams . . . . . . . . . . . . . . . . . . . . . . . . . 147
12.3 The exceptional algebra G2 . . . . . . . . . . . . . . . . . . . 149
12.4 Fundamental weights . . . . . . . . . . . . . . . . . . . . . . . 155

13 Representations. 157
13.1 The Weyl group . . . . . . . . . . . . . . . . . . . . . . . . . . 160

14 The unitary groups AN ´1 “ SU pN q 163

15 Orthogonal algebras 167

15.1 Dn “ SOp2nq . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
15.2 SOp2n ` 1q ” Bn . . . . . . . . . . . . . . . . . . . . . . . . . 168
15.3 Clifford algebras . . . . . . . . . . . . . . . . . . . . . . . . . 171

16 Automorphisms 173

4
1

Curves and surfaces in R3.

1.1 Curves.

x “ xpuq
y “ ypuq
z “ zpuq
u1 ď u ď u2 (1.1)

or else
xi “ xi puq i “ 1, 2, 3. (1.2)
One example is the circular helix

x “ a cos u
y “ a sin u
z “ bu
0ďuď8 (1.3)

which for b “ 0 reduces to a circle of radius r “ a in the z “ 0 plane.

The arc length is given by Pythagoras’ theorem
˜ˆ ˙ ˆ ˙2 ˆ ˙2 ¸
2 2 2 2 dx 2 dy dz
ds ” dx ` dy ` dz “ ` ` du2 (1.4)
du du du

Then the arc between two points is given by

d
ż u1 ˆ ˙2 ˆ ˙2 ˆ ˙2
dx dy dz
s01 ” ` ` du (1.5)
u0 du du du

Let us compute it for a circle

a
y“ R 2 ´ x2 (1.6)

5
 c ´ ¯2
dx2 ` ?R2x´x2 dx2 “ ? dx
ş ş ş
s “ dsa “ R2 ´x2
“
ş dt ´1 x
“ ?1´t 2 “ sin R (1.7)

Then
x
sin s “ (1.8)
R
For example, for the helix,
a
s01 “ a2 ` b2 pu1 ´ u0 q (1.9)

The tangent vector is defined as

d~
x
~t ” d~x ” du
” c´
1 d~x
(1.10)
ds ds ¯2 du
du d~
x
du

For the helix,

~t “ ? 1 p´a sin u, a cos u, bq (1.11)
a2 ` b2
The normal to the tangent at a given point,

~t.~n “ 0 (1.12)

and normalized such that

~n2 “ 1 (1.13)
In our example
~n “ ˘ pcos u, sin u, 0q (1.14)
It is clear that
~t2 “ 1 (1.15)
which implies that
~t~t9 “ 0 (1.16)
We can write
d~t
” κ ~n (1.17)
ds
where κ is called the curvature at a given point (there is a sign that must
be fixed by some convention); and the radius of curvature is defined by
1
κ” (1.18)
R
In the example
d~t a
“ 2 p´ cos u, ´ sin u, 0q (1.19)
ds a ` b2

6
so that
a2 ` b2
R“ (1.20)
a
which reduces to a when b “ 0.
It is clear that ´ ¯2
κ2 “ ~t9 (1.21)
We define the binormal as
~b ” ~t ˆ ~n (1.22)
In the example
~b “ ? 1 p´b sin u, b cos u, ´aq (1.23)
a2 ` b2
It is clear that the vectors p~t, ~n, ~bq form a moving trihedron along the curve.
Consider now
~
~b9 ” db “ ~t9 ˆ ~n ` ~t ˆ ~n9 “ ~t ˆ ~n9 (1.24)
ds
It is clear that this vector is orthognonal to both ~t as well as to ~b, so that it
must lie in the direction of ~n
d~b
” ´τ ~n (1.25)
ds
ahere τ is called the torsion of the curve at the point considered. For the
helix
~b “ ? 1 p´b sin u, b cos u, aqq (1.26)
a2 ` b2
and then
d~b 1
“ 2 p´b cos u, ´b sin u, 0q (1.27)
ds a ` b2
and then
b
τ “¯ 2 (1.28)
a ` b2
which vanishes for b “ 0 as it does for any plane curve. Finally, the derivative
of the normal vector has to lie in the plane spanned by p~t, ~bq
d~n
“ C1~t ` C2~b (1.29)
ds
We find that

C1 “ ~t.~n9 “ ´~n.~t9 “ ´κ
9
C2 “ ~b.~n9 “ ´~n.~b “ τ (1.30)

conveying the fact that

d~n
“ ´κ~t ` τ ~b (1.31)
ds

7
This is the last of Frenet-Serret’s formulas.
Frenet-Serret’s formulas also imply that the acceleration is given by
d2~t
ds2
“ κ d~n
ds ` κ~ 9 n “ ´κ2~t ` κτ~b ` κ~
9n
d2 ~ ~ ~
´ κ9 ~t ` τ9~b “ ´ κ2 ` τ 2 ~n ´ κ9 ~t ` τ9~b
n dt db
` ˘
ds2
“ ´κ ds ` τ ds (1.32)

Neglecting the derivatives of the curvature and the torsion, this yields the
familiar centripetal acceleration for plane curves, for which τ “ 0.

1.2 Surfaces.
xi “ xi pu, vq u1 ď u ď u2 v1 ď v ď v2 (1.33)
For example, the circular cone z2 “ x2 ` y2

x “ u sin v
y “ u cos v
z“u (1.34)

It has a singular point at u “ 0. Another example is the cylinder

x “ cos u
y “ sin u
z“v (1.35)

The induced metric on the surface by the euclidean metric in R3 is

a,b“2
ÿ Bxi Bxj a b
ds2 ” δij dx dx ” Edu2 ` 2F dudv ` Gdv 2 (1.36)
a,b“1
Bxa Bxb

It used to be called the first fundamental form on the surface.

For the cone
ds2 “ 2du2 ` u2 dv 2 (1.37)
and for the cylinder
ds2 “ du2 ` dv 2 (1.38)
The tangent plane to the surface at a given point is generated by the two
vectors
~ta ” Ba ~x a “ 1, 2. (1.39)
and normalized in such a way that
~t2a “ 1 (1.40)

Foe the unit sphere

ds21 “ dθ2 ` sin2 θdφ2 (1.41)

8
 ~tθ “ pcos θ cos φ, cos θ sin φ, sin θq
~tφ “ p´ sin φ, sin cos φ, 0q (1.42)

Again for the cone

~tu ” ?1 psin v, cos v, 1q

2
~tv ” pcos v, ´ sin v, 0q (1.43)

For the cylinder

~tu ” p´ sin u, cos u, 0q

~tv ” p0, 0, 1q (1.44)

The normal vector is uniquely defined as the unit vector proportional to

~ ” ~xu ˆ ~xv
N (1.45)
|~xu ˆ ~xv |

For the sphere it reads

~ “ psin θ cos φ, sin θ sin φ, cos θq ” ~x

N (1.46)

For our cone

~ ” ?1 psin v, cos v, ´1q
N (1.47)
2
And for the cylinder
~ ” pcos u, sin u, 0q
N (1.48)
Consider now a curve on the surface; its tangent vector surely lies on the
tangent plane. We can project the derivative of the tangent vector with
respect to the arc (the normal curvature vector) on a tangential and a normal
component.
d~t ~
” kn ` ~kt (1.49)
ds
where ´ ¯
~kn ” κn N
~ ” ~k.N ~ N ~ (1.50)

and the tangent or geodesic curvature vector is ~kt .

Now, the fact that N~ .~t “ 0 implies that

d~t ~ ~
dN ~
d~x dN
κn ” .N “ ´~t. “´ . (1.51)
ds ds ds ds
The second fundamental form is defined as

~ ” edu2 ` 2f dudv ` gdv 2

´d~x.dN (1.52)

9
This means that for the sphere the first and second fundamental forms are
the same.
ds21 “ ds22 (1.53)
For the cone it gives
ds22 “ udv 2 (1.54)
and for the cylinder
ds22 “ du2 (1.55)
For the sphere, the determinant

g2 ” eg ´ f 2 “ sin2 θ ě 0 (1.56)

For the other two surfaces, however,

g2 “ 0 (1.57)

It so happens that

edu2 ` 2f dudv ` gdv 2 e ` 2f λ ` gλ2

κn “ “ (1.58)
Edu2 ` 2F dudv ` Gdv 2 E ` 2F λ ` Gλ2
where
dv
λ” (1.59)
du
This defines a function κn pλq. The extrema of this function ara the
directions of principal curvature, κ1 and κ2 . The condition of an extrema
can be written as

rpE ` F λq ` λpF ` Gλqs pf ` gλq “ rpe ` f λ ` λpf ` gλqs pF ` Gλq (1.60)

For those λ we can write

e ` f λ ` λpf ` gλq f ` gλ e ` fλ
κpλq “ “ “ (1.61)
E ` F λ ` λpF ` Gλq F ` Gλ E ` Fλ
Then

pe ´ κEqdu ` pf ´ κF qdv “ 0
pf ´ κF qdu ` pg ´ κGqdv “ 0 (1.62)

and eliminating κ we get det M “ 0 where

¨ 2
dv ´dudv du2
˛

M ”˝E F G‚ (1.63)
e f g

from which we get the two directions of principal curvature.

It is also easy to prove ([2]) that they are mutually orthogonal.

10
 In terms of those, the mean curvature is defined as
κ1 ` κ2 Eg ´ 2f F ` eG
M” “ (1.64)
2 2 pEG ´ F 2 q
and the gaussian curvature as
eg ´ f 2
K ” κ1 .κ2 “ (1.65)
EG ´ F 2
It is clear that when
g2 ě 0 (1.66)
the normal chapters are all convex; those points are dubbed elliptic points.
When
g2 “ 0 (1.67)
there is one direction with κ “ 0; those are parabolic points. Finally, when
g2 ď 0 (1.68)
some normal chapters are convex and others are concave; those are hyperbolic
points.
For the cone
2u 1
M“ 2u2
“ u
K“0 (1.69)
Clearly something special happens at the apex of the cone, u “ 0, although
the gaussian curvature does not see it.
The three vectors ´ ¯
~
~xu , ~xv , N (1.70)
onstitute a moving frame (that is, a frame at each point of the surface).
Consequently, we can expand
~
~xuu ” Γ111 ~xu ` Γ211 ~xv ` e N
~xuv ” γ 1 ~xu ` Γ2 ~xv ` f N ~
12 12
~xvv ” Γ122 ~xu ` Γ222 ~xv ~
`gN (1.71)
where the Christoffel sumbols are given by
GEu ´2F Fu `F Eu
Γ111 “ 2pEG´F 2 q
GEv ´F Gu
Γ112 “ 2pEG´F 2 q
2GFv ´GGu ´F Gv
Γ122 “ 2pEG´F 2 q
2EFu ´EEv ´F Eu
Γ211 “ 2pEG´F 2 q
EGu ´F Ev
Γ212 “ 2pEG´F 2 q
EGv ´2F Fv `F Gu
Γ222 “ 2pEG´F 2 q
(1.72)

11
 ~ 2 “ 1 we know that
Also, from N
~ u “ p1 ~xu ` p2 ~xv
N
~ v “ q1 ~xu ` q2 ~xv
N (1.73)

Gauss’ theorema egregium states that the Gaussian curvature depends

onlt on E, F, G, and their first derivatives. This shows that it is a bending
invariant, in Struik’s words. This means that those properties are intrinsic
to the surface, and they do not depend on how the surface is imbedded in
euclidean ambient space. The theorem can be proven by demanding that

~xuuv “ ~xuvu
~xvvu “ ~xuvv (1.74)

12
2

Tensor calculus in vector

spaces

Consider a n-dimensional vector space, V with a basis

n
ÿ
@v P V v“ v i ei ” v i ei (2.1)
i“1

where we have introduced the summation convention. This only affects

contravariant coupled with covariant indices. Given a nonsingular n ˆ n
matrix, we can change to a different basis, fa

fa ” Aia ei (2.2)

Then the vector v can be expressed in the new basis

v “ v a fa “ v i ei “ v a Aja ej (2.3)

and owing to the fact the the basis elements are linearly independent,

v j “ v a Aja ÝÑ v a “ Bia v i (2.4)

where the matrix B “ A´1

Aja Bia “ δij (2.5)
For the time being, indices cannot be raised or lowered. Consider now the
dual space, V ˚ .

θPV θpvq P R (2.6)

We can define the dual basis of the basis of V through

E i pej q ” δji (2.7)

13
Please note carefully the position of the indices in the Kronecker delta.
Those are the only deltas that are allowed in this course. Any element
ω P V ˚ can be expanded in the dual basis

ω ” ω i i (2.8)

Under a change of basis in V

ω i Ñ Aia ω a (2.9)

Everybody heard aboud some wild and ferocious animals called tensors.
What are those? Consider bilinear mappings from

V ˆV˚ Ñ R (2.10)

T : pv, θq Ñ T pv, θq P R (2.11)

Owing to linearity, it is enough to know the values on the basis, because

T pv, θq “ v i θj T pei , E j q ” v i θj Ti j (2.12)

The space of those animals is called the tensor product of V ˚ b V , and its
elements are called (1-covariant 1-contravariant) tensors. Under a change of
basis
Ti j Ñ Aai Bbj Ta b (2.13)
The set of all those (1,1) tensors is another vector space, which is calles the
tensor product of V ˚ b V

E l b ei P V ˚ b V (2.14)

Pleasee note carefully that

V bV˚ ‰V˚bV (2.15)

that is
Ti j ‰ T j i (2.16)

Ordinary vectors and ordinary dual vectors are particular instances (0,1)
and (1,0) respectiveli. The generalization to

T i1 ...ip j1 ...jq ei1 b . . . eip b E j1 b E jq P V b . . . ppq . . . b V b V ˚ b . . . pqq .(2.17)

..V ˚

is immediate. The contravariant or covariant character of the indices is

an absolute property. There is no in general a canonical way of raising or
lowering indices. When there is a metric, there is such a canonical way.

14
But before introducing a metric, let us examine some particularly intere-
sting tensors which are defined independently of the metric. As a matter of
notation, let us define the symmetrization operator

1 ÿ
Tpa1 ...ap q ” Taπp1q ...aπqnq (2.18)
p! πPS
p

where the sum extends over all p! elements of the permutation group Sp ; as
well as the antisymmetrization operator

1 ÿ
Tra1 ...ap s ” p´1qPπ Taπp1q ...aπqnq (2.19)
p! πPS
p

where Pπ is the parity of the permutation π.

2.1 Differential forms.

Let us identify tangent vectors ~v P Tx with directional derivatives of func-
tions defined at a given point

~v pf q ” v µ Bµ f (2.20)

A particular basis is given by the vectors

Bµ (2.21)

Given an arbitrary function, its differential is defined as df P Tx˚

df p~v q ” ~v pf q (2.22)

Differential forms are antisymmetric linear maps

ω1 : v P Rn Ñ ωpvq P R (2.23)

A local basis is given by

dxa pBb q “ δba (2.24)

Let us define a p-form A P Λp as a tensor with p cpvariant indix, totally

antisymmetric
Aa1 ...ap ” Ara1 ...ap s (2.25)

Dor examplem ω P Λ2 :

ω2 pv, wq P Rn ˆ Rn Ñ ω2 pv, wq P R (2.26)

15
• Exterior product. The exterior product of two one-forms yields a
two-form
ˆ ˙
ω1 pv1 q α1 pv1 q
pω1 ^ α1 qpv1 , v2 q ” det (2.27)
ω1 pv2 q α1 pv2 q

In the general case, the product of a p-form and a q-form is a (p+q)-

form

ÿ
pωk ^ ωl q pv1 . . . vk`l q ” ˘ωk pvi1 . . . vik qωl pvik`1 . . . vik`l q (2.28)

The basic identity reads

ωp ^ ωq “ p´1qpq ωq ^ ωp (2.29)
Sometimes we shall write
dxµ1 ...µp ” dxµ1 ^ . . . ^ dxµp (2.30)
This means that for every odd degree form
ω2p`1 ^ ω2p`1 “ 0 (2.31)

• Coordinate basis.
In the basis of the tangent space associated to a local chart, pxα q,
ÿ
ωk ” ωι1 ...ιk dxι1 ^ . . . ^ dxιk (2.32)
ι1 ă...ăιk

dxµ ^ dx “ dxµ b dxν ` dxν b dxµ

ν
(2.33)

We shall write in local coordinates

1
α ” αµ1 ...µp dxµ1 ^ . . . ^ dxµp (2.34)
p!
It is exceedingly useful to introduce the Kronecker symbols
λ1 λ
λµ11 ...λ p p
...µp ” p! δrµ1 . . . δµp s (2.35)
It is a good exercise to prove that
ρµ11...ρ p
...µp αρ1 ...µp “ p! αµ1 ...µp

λµ11 ...λ q µ1 ...µq σ1 ...σp

...µq ν1 ...νp`q “ q! λν11...ν
...λq σ1 ...σp
p`q

λµ11 ...λ q ρ1 ...ρp λ1 ...λq

...µq ρ1 ...ρp “ p! µ1 ...µq (2.36)
A general formula for the exterior product is given by
1 1
α^β “ αλ1 ...λp βµ1 ...µq dxλ1 ^ . . . ^ dxλp dxµ1 ^ . . . ^ dxµq (2.37)
p! q!

16
• Exterior differential. The differential of a function is given by a
one-form ÿ
df ” Ba f dxa (2.38)
In the general case, the differential of a p-form is a (p+1)-form
ÿ
dω ” dωι1 ...ιk ^ dxι1 ^ . . . ^ dxιk (2.39)
ι1 ă...ăιk

A general formula can also be given

1
pdαqµ0 µ1 ...µp ” λ0 λ1 ...λp Bλ αλ ...λ (2.40)
pp ` 1q! µ0 µ1 ...µp 0 1 p
The uselfuness of exterior calculus stems essentially from the basic fact
that
d2 “ 0 (2.41)
It is also a fact that the graded Leibnitz rule holds, id est,

d pαp ^ βq q “ dαp ^ βq ` p´1qp αp ^ dβq (2.42)

• Pullback.
φ : x P Mp Ñ y P Nq (2.43)
By i
ω “ ai dy i P ΛpN q Ñ φ˚ ω ” ai pypxqq dxa P ΛpM q (2.44)
Bxa
It is fact of life that
d pφ˚ ωq “ φ˚ dω (2.45)

• Poincaré. Everybody knows that in R3

~ ˆ ~v “ ~0 ùñ ~v “ ∇φ
∇ ~ (2.46)

In fact Poncaré was able to show that in Rn

dω “ 0 ùñ ω “ dα (2.47)

This is not true in general, and the number of independent ω that fail
to satisfy that is called the Bettti number of the manifold. Let us prove
this theorem. Given a p-form,
1
ωp P ΛpRn q ” ωa ...a px1 , . . . xn q dxa1 ^ dxa2 . . . ^ dxap (2.48)
p! 1 p
, define the homotopy operator, K in two steps. First define a Λp`1
form
1
pφ ˚ ωq ” ωa ...a ptx1 , . . . , txn q pxa1 dt ` tdxa1 q^pxa2 dt ` tdxa2 q^. . .^pxap dt ` tdxap q
p! 1 p
(2.49)

17
 Now the operator K is defined in two steps. In those monomials of
φ˚ ω not involving dt
Kω “ 0 (2.50)
On monomials of φ˚ ω involving dt
ˆż 1 ˙
pKωq ” pφ ωq~a ptxqdt dx~a
˚
(2.51)
0

Let us work out an example in n “ 3 dimensions.

ω “ xdx ^ dy ` ez dy ^ dz (2.52)

φ˚ ω “ tx ptdx ` xdtq ^ ptdy ` ydtq ` etz ptdy ` ydtq ^ ptdz ` zdtq “

` ˘
“ ´t2 xy dt ^ dx ` t2 x2 ´ etz tz dt ^ dy ` etz ty dt ^ dz ` no dt terms
(2.53)

ş1 ` ˘
α ” Kω “ ´t2 xy dt ^ dx ` t2 x2 ´ etz tz dt ^ dy ` etz ty dt ^ dz “
0
´ 2 z
¯ z
“ ´ 13 xydx ` x3 ´ e pz´1q`1
z dy ` y e pz´1q`1
z2
dz (2.54)

since ż
ueu du “ eu pu ´ 1q (2.55)

And lo and behold,

ω “ dα (2.56)

• Hodge dual. Let us introduce the so called volume element defined

as
a
ηµ1 ...µn ” |g| 1...n
µ1 ...µn (2.57)

Actually, 1...n
µ1 ...µn is not a tensor. Let us work it out in two dimensions.
Denote the jacobian matrix
Bxb
Jab1 ” (2.58)
Bxa1
and its determinant by J ” detJab1 Also the determinant of the metric
itself does not transform as a true scalar, but rather

g 1 px1 q ” J 2 gpxq (2.59)

Then
? 1 1 1a 1 1 1 1 1
ηab dxa ^dxb “ gab Jaa1 Jbb1 dxa ^dxb “ g Ja1 b1 dxa ^dxb “ ηa1 b1 dxa ^dxb
J
(2.60)

18
This means that
?
g ab (2.61)
is a true tensor. Some properties;
1 µ1 ...µn
η µ1 ...µn “ a 1...n
|g|
µ ...µ
ηλ1 ...λp λp`1 ...λn η λ1 ...λp µp`1 ...µn “ p! λp`1
p`1 n
...λn
∇ρ ηµ1 ...µn “ 0
a
dpvolq ” ηµ1 ...µn dxµ1 ^ . . . ^ dxµn “ |g|dx1 ^ . . . ^ dxn
dxµ1 ^ . . . ^ dxµn “ η µ1 ...µn dpvolq (2.62)

To verify these formulas is excellent gymnastics.

The Hodge operator maps p-forms into (n-p)-forms.

˚ : Λp Ñ Λn´p (2.63)

It is defined by
1
p˚Aqµp`1 ...µn ” ηµ ...µ Aµ1 ...µp (2.64)
p! 1 n

It is clear that in R3

˚dz “ dx ^ dy
˚dy “ dz ^ dx
˚dx “ dy ^ dz (2.65)

Its square depends on the dimension of spacetime as well as on the

degree of the form
˚2 : Λp Ñ Λp (2.66)
First of all
1 1
p˚Aqa1 ...an´p “ ηb ...b a ...a Ab1 ...bp “ p´1qppn´pq ηa1 ...an´p b1 ...bp Ab1 ...bp
p! 1 p 1 n´p p!
(2.67)
and
1 1 b1 ...bn´p d1 ...dp A
p˚2 Aqc1 ...cp “ p! pn´pq! ηc1 ...cp b1 ...bn´p η d1 ...dp “
d ...d
“ p!1 c11...cp Ad1 ...dp “ Ac1 ...cp
p
(2.68)

˚2 “ p´1qppn´pq (2.69)
In four dimensions (actually, in any even dimension)

˚2 “ p´1qp (2.70)

19
 In R4
˚dx ^ dy “ dz ^ dw (2.71)

There are then euclidean self-dual two-forms

ω2 ” dx ^ dy ` dz ^ dw (2.72)

In three-dimensions Hodge squared it is always +1

˚2 “ `1. (2.73)

The exterior codifferential is the adjoint of the exterior differential

pα, δβq ” pdα, βq (2.74)

It is given by
δ ” p´1qp ˚´1 d˚ (2.75)

It is possible to give a simple formula

1 µ1 ...µp
pδαqρ1 ...ρp´1 “ ´  ∇ν αµ1 ...µp (2.76)
p! νρ1 ...ρp´1

The interior product of a p-form and a vector, X, is the (p-1)-form

given by
pipXqωq pv1 . . . vp´1 q ” ωp pX, v1 . . . vp´1 q (2.77)

• Stokes’ theorem We start from the properties of the volume defined

by an elementary cell of R3

– It vanishes if the vectors are linearly dependent.

– It stays the same when we add to a given vector a linear combi-
nation of the other vectors.
– Depends in a linear way on all vectors.

Al these properties are enjoyed by the elementary formula

ÿ
V “ ijk v1i v2j v3k “ η p~v1 , ~v2 , ~v3 q (2.78)

where the volume element is defined by

η ” dx1 ^ dx2 ^ dx3 (2.79)

This leads in a natural way to define volumes through integration. For

example, in R4 ,

20
 1 bcd
– Codimension-1 hypersurfaces dSa ” 3! ηabcd dx Consi-
der for example the hypersurface

S ” tx4 “ T u (2.80)

The normal is the vector

n “ p0, 0, 0, 1q (2.81)

The hyprersurface can be paremeterized by

xi “ ξ i (2.82)

so that
1 a
dSa ” n gpx4 “ T q d3 ξ (2.83)
3!
1 cd
– Codimension-2 hypersurfaces dVab ” 2! ηabcd dx Consi-
der the two-sphere
S2 ãÑ R4 (2.84)

x4 “ T
x2 ` y 2 ` z 2 “ R 2 (2.85)

It can be parameterized by polar coordinates

xi “ xi pθ, φq (2.86)

There are two normal vectors, namely

n1 “ p0, 0, 0, 1q
n2 “ px, y, z, 0q (2.87)

and the volume element is

a
dVab “ n1a n2b gpT, θ, φq dθ ^ dφ (2.88)

1 d
– Codimension-3 hypersurfaces dVabc ” 3! ηabcd dx

For a trivial example, consider

xi “ xi0 (2.89)

which can be pameterized as

x4 “ σ (2.90)

21
there are now three normals

ni ” e i (2.91)

and the volume element reads

b
dVabc “ n1a n2b n3c gpxi0 q dσ (2.92)

Stokes’ theorem in general states that.

ż ż
ω“ dω (2.93)
BV V

The classical theorems of Gauss, Stokes and the divergence are but
particular instamces of this. For example
ż ż
dA1 “ A1 (2.94)
S2 C1 ”BS2

If A1 is a 1-form of R3
A1 ” Ai dxi (2.95)
then
1
dA2 “ pBi Aj ´ Bj Ai q dxi ^ dxj (2.96)
2
It is customary to define the rotational or curl as

protAqi ” ijk Bj Ak (2.97)

The surface integral

ż ż
1
dA2 “ pBi Aj ´ Bj Ai q dxi ^ dxj “ (2.98)
S S 2

It is customary to define

1
ni dS ” ijk dxj ^ dxk (2.99)
2
so that
ÿ ÿ
protAqi ni dS “ pBj Ak ´ Bk Aj q dxj ^ dxk (2.100)
i jk

and we recover Stokes’original theorem

ż ż
~ A
rot ~ ~n dS “ ~ d~x
A (2.101)
S BS

22
Let us now apply it to
ż ż
dω2 “ ω2 (2.102)
V3 BV3

Write
1
ω2 ” ωij dxi ^ dxj (2.103)
2
so that
1 1
dω2 ” Bk ωij dxk ^ dxi ^ dxj “ Bk ωij kij dV (2.104)
2 2
Now we define the dual one-form
1
Ωi dxi ” p˚ω2 q1 ” ijk ωjk (2.105)
2
then
~
dω2 “ Bk Ωk ” divΩ (2.106)
and we recover Gauss’divergence teorem
ż ż
~
div Ω dV “ ~ ~n dS
Ω (2.107)
V BV

Let us work out the integral over a two-sphere

x “ R sin θ cos φ
y “ R sin θ sin φ
z “ R cos θ (2.108)

z is no an independent variable; rather,

a
z ” 1 ´ x2 ´ y 2 (2.109)

sin φ “ ? 2y 2
x `y
a
sin θ “ x2 ` y 2 (2.110)

Bθ cos φ
Bx “ cos θ
Bθ sin φ
By “ cos θ
Bφ y
Bx “ ´ sin2 θ
Bφ x
By “ sin2 θ (2.111)

Exterior normal
~ “ xv x `yv y `zv z
~v .ds R R2 sin θdθdφ “
“ R pv x sin θ cos φ ` v y sin θ sin φ ` v z cos θq sin θdθdφ
(2.112)

23
 Assume, for example,
B
v“z (2.113)
Bz
(such that
~ v “ 1q
∇~ (2.114)
żR 3 ˇR
ż ż ˇ
~ v dpvolq “ r 4
∇~ 2
r dr sin θdθdφ “ 2π ˇˇ “ πR3 (2.115)
S2 0 3 0 3
This equals
ż
3
ˇπ 4π 3
R sin θ cos2 θdφ “ ´2πR3 cos3 θˇ0 “ R (2.116)
3

• Lie derivative. The Lie derivative of a function is defined as the

directional derivative
~v pf q “ £p~v qf (2.117)
The Lie derivative of a one-form is defined in a natural way.

£p~v qdf ” d~v pf q (2.118)

This definition extends to a general case simply by postulating that

Leibnitz’ rule holds true

£p~v qaa dξ a “ p£p~v qaa qdξ a ` αa £p~v qdξ a (2.119)

In the case of vectors we use the dual application

~ “ x£p~v qα, Xy
£p~v qxα, Xy ~ ` xα, £p~v qXy
~ (2.120)

It is a fact that
~ Y
£pXq ~ “ rX,
~ Y~s
~ “ ipXqd
£pXq ~ ` dipXq
~ (2.121)

• Diffeomorfisms An active diffeomorphism

ξ : x P M Ñ y “ ξpxq P M (2.122)

Acting on vectors, given g : y Ñ R, then g ˝ ξ : x Ñ R and v P Tx , we

define a different vector ξ˚ v P Ty through

ξ˚ pvqpgq ” vpg ˝ ξq (2.123)

In a local coordinate basis

pξ˚ vqµ pyq “ v ρ Bρ ξ µ pxq (2.124)

24
 Given a one-form ω P Ty˚ we define another form ξ ˚ ω P Tx through

ξ ˚ ωpvq ” ωpξ˚ vq (2.125)

In a local coordinate basis

pξ ˚ ωqα pxq “ ωµ pyqBα ξ µ pxq (2.126)

If it were a 2-form
pξ ˚ ωq pv, wq “ ω pv, wq (2.127)
that is
pξ ˚ ωqαβ pxq “ ωµν pyqBα ξ µ Bβ ξ ν (2.128)

2.2 The metric tensor.

The metric tensor in Rn is defined through

ds2 “ gab pxqdxa dxb (2.129)

with
g ” det gab ‰ 0 (2.130)
so that there is the inverse matrix

g ac gcb “ δba (2.131)

For example, in polar coordinates

ds2 “ dr2 ` r2 dΩ2 (2.132)

where
dΩ2 ” dθ2 ` sin2 θdφ2 (2.133)
Then, there is a canonical mapping from

V ÑV˚ (2.134)

V ˚ pV q ” gpV, V q (2.135)
This is
Va “ gac V c V a “ g ab Vb (2.136)
Ordinary derivatives of any object more complicated than scalar (id est, a
vector, or any higher rank tensor) are not tensors, not even in Rn . This is
because under the change of coordinates

xa Ñ y α pxa q (2.137)

25
under which
By α b
V α pyq ” V pxq (2.138)
Bxb
derivatives transform as
BV α pyq B 2 y α Bxc b By α BV b pxq Bxc
“ V pxq ` (2.139)
By γ Bxb Bxc By γ Bxb Bxc By γ
Let us lighten the notation a little bit. First of all,
B
Bα ”
By α
B
Bb ” b (2.140)
Bx
Er also introduce
By α
Jbα ” (2.141)
Bxb
Then
` ˘a Bxa
J ´1 β
” (2.142)
By β
is the inverse matrix ` ˘ ` ˘
J J ´1 “ J ´1 J “ 1 (2.143)
The previous equation reads
˘c ˘c
Bγ V α “ J ´1 γ Bc Jbα V b ` Jbα Bc V b J ´1 γ
` `
(2.144)

It is conceptually much simpler if we imagine matrices with rows defined

by the covariant indices and columns by the contravariant indices. The
equation then reads
` ˘
BV pyq “ J ´1 BJV ` J ´1 BV pxqJ (2.145)

Let us now ask the question: is it possible to modify the definition of

derivative in such a way that
` ˘
∇V “ J ´1 ∇V J (2.146)

Let us try the ansatz

∇V “ BV ` ΓV (2.147)
(Γ is a three-index beast). In order for that to be true the transformed
covariant derivative
` ˘
BV pyq ` ΓpyqV pyq “ J ´1 BJV ` J ´1 BV pxqJ ` ΓpxqV pxq (2.148)

ought to be equal to
ˆ ˙
´1
J BV pxq ` ΓpxqV pxq J (2.149)

26
This would be true provided the Γ transform as

Γpyq “ J ´1 BJ ` J ´1 ΓJ (2.150)

When such an object exists, there is an invariant concept of derivative.

This is what mathematicians call a connection. The surprising thing is that
whenever there is a metric, there is such a connection, which is called the
Levi-Civita one, and the coefficients, the Christoffel symbols,
1
Γαβγ ” g αλ p´Bλ gβγ ` Bβ gλγ ` Bγ gλβ q (2.151)
2
Let us check that
" ¯*
1 α λ al ´ ¯ ´ ¯ ´
Γαβγ
“ Ja Jl g b c l c
´ Bλ Jβ Jγ gbc ` Bβ Jλ Jγ glc ` Bγ Jβ Jλ gbl b l
“
2
"
1
“ Jaα Juλ g au ´ Bλ Jβb Jγc gbc ´ Jβb Bλ Jγc gbc ´ Jβb Jγc Bλ gbc `
2
`Bβ Jλl Jγc glc ` Jλl Bβ Jλc glc ` Jλl Jγc Bβ glc `
*
b l b l b l
`Bγ Jβ Jλ gbl ` Jβ Bγ Jλ gbl ` Jβ Jλ Bγ gbl (2.152)

The three terms in the right of the rows yield

Jaα Jβb Jγc Γabc (2.153)

If this were all, this would have been a true tensor. But there is more.
Taking into account that
Bα Jβa “ Bβ Jαa , (2.154)
the terms in the paces 11 and 21 cancel, as do the terms 12 and 32. The
rest (22+31) yield

Jaα Juλ g au Bγ Jβb Jλl gbl “ g aλ Bγ Jβb Jλl gbl (2.155)

QED.
The two basic properties of the Levi-Civita connection are

Γαβγ “ Γαγβ
∇α gβγ “ 0 (2.156)

2.3 Winding numbers and such.

Spheres are defined as
n
ÿ
S” x2a “ L2 (2.157)
a“1

27
The n-dimensional ball is defined as
n
ÿ
Bn : x2a ď L2 (2.158)
a“1

The normal vector is

na “ cBa S “ c xa (2.159)
Let us study the one form in Rn
ÿ
rdr ” xi dxi “ x1 dx1 ` x2 dx2 ` . . . ` xn dxn (2.160)
i

It is clear that

˚rdr “ x1 dx2 ^ . . . ^ dxn ´ x2 dx1 ^ dx3 ^ . . . ^ dxn ` . . . “

“ p´1qi´1 xi dx1 ^ . . . ^ dxxi ^ . . . ^ dxn
ř
(2.161)

It so happens that the measure on the S n´1 sphere is proportional to this

(n-1)-form
dSa “ cna ˚ rdr (2.162)
On the other hand
ÿ
d ˚ rdr “ p´1qi´1 dxi ^ dx1 ^ . . . ^ dx
xi ^ . . . ^ dxn “ nη (2.163)

This shows that actually c “ 1. You can show as an exercise that

L
V pBn q “ nV pSn´1 q
π n{2
V pBn q “ Γpn{2`1q (2.164)

In particular, we recover

V pB3 q “ 34 πL3
V pS2 q “ 4πL2 (2.165)

Let us denote the volume element in euclidean space by

ωn ” dpvolq ” dx1 ^ . . . ^ dxn (2.166)

In ordinary euclidean space, E3

ω3 “ r2 sin θ dr ^ dθ ^ dφ (2.167)

and by σ 1 the volume element on the codimension-one unit sphere

r“L (2.168)

Consider ÿ
rdr “ xi dxi (2.169)

28
It is plain that
ÿ
˚rdr “ p´1qi´1 xi dx1 ^ . . . ^ dx
xi ^ . . . ^ dxn (2.170)

In E2
˚rdr “ xdy ´ ydx “ r2 dθ (2.171)
Again, in E3 , it is easy to work out that

˚rdr “ xdy ^ dz ´ ydx ^ dz ` zdx ^ dy “ r3 sin θ dθ ^ dφ (2.172)

It is a fact that
σ 1 “ ˚rdr (2.173)
(on the sphere Sn´1 ), because

Demonstratio.
d p˚rdrq “ nωn (2.174)
and in particular in E3
1
dp˚rdrq “ ω3 (2.175)
r2

Consider now the projection

π : En z0 ÝÑ Sn´1 (2.176)

~x
πp~xq ” (2.177)
|x|
We know that ` ˘
d π ˚ σ 1 “ π ˚ dσ 1 “ 0 (2.178)
(because there are no n-forms in Sn´1 ). Let us show that
σ
π˚σ1 “ ”τ (2.179)
rn
(σ 1 is the restriction of σ to Sn´1 ).

Demonstratio. First of all,

1 n n n
dτ “ n
dσ ´ n`1 prdrq ^ σ “ n ω ´ n ω “ 0 (2.180)
r r r r
Now, define

p´1qi´1 πi dπ1 ^ . . . ^ dπ
ř
π˚σ1 “ xi ^ . . . ^ dπn “
p´1qi´1 xri r 1
ř
“ 2pn´1q prdx1 ´ x1 drq ^ . . . ^ dπi ^ . . . ^ prdxn ´ xn drq “
x
(2.181)

29
In E2
x y y x 1
π˚σ “ d ´ d “ 2 σ “ dθ (2.182)
r r r r r
In E3 is clear that it is going to give the same as σ without the radial
coordinate. This defines the angular measure, τ .

Given two closed and oriented manifolds, M and N, and a mapping

f: M ÝÑ N (2.183)

then it is a fact that f˚ N is an integral multiple on M plus a boundary ([3]).

This is dubbed the degree of f, deg f . When Σ ãÑ En z0, we can deinne the
projection as above
π : Σ Ñ Sn´1 (2.184)
Then
ż ż ż ż
τ“ π˚σ1 “ σ 1 “ deg π σ 1 “ deg π An´1 (2.185)
Σ Σ πpΣq Sn´1

We can generalize this a little bit. Consider a closed manifold

f π
Mn´1 ÝÑ En z0 ÝÑ Sn´1 (2.186)

The winding number of this hypersurface around the origin is given by

ż
1
w” f˚ τ (2.187)
An´1 M

In general, given
f
Mn ÝÑ Nn (2.188)
and a volume form β in N normalized to 1
ż
β“1 (2.189)
N

we have ż
degf “ f ˚β (2.190)

This is essentially the mathematics behind the WZNW lagrangian.

Let us now study the Hopf invariant. Normalize
ż
σn “ 1 (2.191)
Sn

Consider a map
f
S3 ÝÑ S2 (2.192)

30
Then
dpf ˚ σ2 q “ f ˚ pdσ2 q “ 0 (2.193)
Now, it is known that S3 does not have nontrivial cycles, so that D α1

dα1 “ f ˚ σ2 (2.194)

It so happens that the integral

ż
α1 ^ f ˚ σ2 (2.195)

is an integar dubbed the Hopf invariant. Represent the sphere S3 as pz, wq

|z|2 ` |w|2 “ 1, and the sphere S2 as

zS ” rS eiφS (2.196)

in such a way that

1 1
ds2 ”
` 2
drS ` rS2 dφ2S
˘
2 2 dzS dz¯S “ ` ˘2 (2.197)
p1 ` |zS | q 1 ` rS2

In order to get that we have to rescale

xS
xS Ñ (2.198)
2L
so that
1 2Lx
xS ” 2L L`xn
ds2 Ñ 4L2 ds2 (2.199)

is now dimensionless.
ˆ ˙
4 1
σ2 “ C ` ˘2 rS drS ^ dφS “ ´2Cd ^ dφ (2.200)
1 ` rS2 1 ` rS2

The normalization is
1 ˇˇ8
ż ż8 ˇ
rdr
σ2 “ 8πC “ ´4πC “ 4πC (2.201)
0 p1 ` r2 q2 1 ` rS2 ˇ0

Then the mapping

f z1
pz1 , z2 q ” x1 ` ix2 , x3 ` ix4 ” pr1 eiφ1 , r2 eiφ2 q P S3 ÝÑ zS ”
` ˘
P S2
z2
(2.202)
The condition

|z1 |2 ` |z2 |2 ” x21 ` y12 ` x22 ` y22 “ 1 ðñ r1 ” cos ψ r2 ” sin ψ (2.203)

31
We shall see in due course that he standard parameterization in terms of
Euler angles of SU(2) is
i i i
g pθ, φ, χq ” e 2 χσ3 e 2 θσ1 e 2 φσ3 (2.204)

where the range of the different angles is

0ďθďπ
0 ď φ ď 2π
0 ď χ ď 4π (2.205)

The left-invariant one-forms read

g ´1 dg “ 2i σa ωaL
ω1L “ cos φ dθ ` sin θ sin φdχ
ω2L “ sin φ dθ ´ sin θ cos φ dχ
ω3L “ dφ ` cos θ dχ
(2.206)

It is convenient to define
χ`φ
z1 ” ei 2 cos θ
2
χ´φ
z2 ” e i 2 sin θ
2 (2.207)

The round metric in S3 then reads

ÿ
ds2 ” 2
ωaL “ dθ2 ` dφ2 ` 2 cos θ dφ dχ ` dχ2 (2.208)

The Hopf fibering goes as follows. In the neighborhood z1 ‰ 0

z2
p : S3 Ñ S2 ppz1 , z2 q ” ”z (2.209)
z1
and if z2 ‰ 0
z1 1
p : S3 Ñ S2 ppz1 , z2 q ”
” (2.210)
z2 z
Denoting by H˘ the two hemispheres of the two-sphere S2 ,

ˆ ˙ ´ ¯
? u` i χ`φ θ i χ´φ
H` : pz, u` q P S2 ˆ S1 Ñ 2
, ?zu` ” e cos2
2, e sin
2
θ
2 `
1`|z| 1`|z|2
ˆ ˙ ´ ¯
?|z|u´ 2 , ?|z|u´ i χ`φ θ i χ´φ θ
H´ : pz, u` q P S2 ˆ S1 Ñ 2
” e 2 cos 2 , e 2 sin 2
1`|z| z 1`|z| ´

Now, start with

1
σ” pu1 du2 ^ du3 ` u2 du3 ^ du1 ` u3 du1 ^ du2 q (2.211)
4π

32
in such a way that
ÿ ui dui r
dr ^ σ ” ^σ “ du1 ^ du2 ^ du3 (2.212)
i
r 4π

Using the relationship ÿ

ui dui “ 0 (2.213)

we get
1 du1 ^ du2
σ“ (2.214)
4π u3
Now the sterographic projection of CP1 ÝÑ S2 Ă R3 reads

´1 ` x2 ` y 2
ˆ ˙
2x 2y
z ” z ` iy ÝÑ gpzq ” u1 “ , ,
1 ` x2 ` y 2 1 ` x2 ` y 2 1 ` x2 ` y 2
(2.215)
The form g ˚ σ is given by

i dz ^ ds z z1 zs0 ´ zs0 ds
i pz1 dz0 ´ z0 dz1 q ^ pds z1 q
g˚σ “ ´ 2
“ 2 2
(2.216)
2π p1 ` |z| q 2π p|z0 | ` |z1 | q

Using now real coordinates

|z0 |2 ` |z1 |2 “ 1 “ x21 ` x22 ` x23 ` x24 (2.217)

we get, after some calculation,

1 1
f ˚σ “ pdx1 ^ dx2 ` dx3 ^ dx4 q “ dα (2.218)
π π
where
1
α” px1 dx2 ` x3 dx4 q (2.219)
π
and finally [?]
ż ż żπ żπ ż 2π
2 2
Hpf q “ αdα “ 2 x2 dx2 ^dx3 ^dx4 “ 2 dθ dφ dξ sin4 ξ sin3 φ cos2 θ “ 1
S3 π π 0 0 0
(2.220)

Assuming

zS “ rS eiφS
u` “ eiα (2.221)

33
This implies on H`
θ ? 1
cos 2 ” 1`r2
θ ? r
sin 2 ” 1`r2
χ “ 2α ` φS
φ “ ´φS (2.222)

On H´
1 1
zN ” “ e´iφS (2.223)
zS r

θ´ 1 ? r θ`
sin 2 ”? 2
“ 1`r2
“ sin 2
1`rN
θ´ θ`
cos 2 ” ? rN 2
“ ? 1
1`r2
“ cos 2
1`rN
χ “ 2α ` φS “ 2α ´ φN
φ “ ´φS “ φN (2.224)

On the equator

|z|
u` ” eiα “ u´ ” e´iφS u´ ” eipα`φS q e´iφS (2.225)
z
This is the twist that makes all the difference between the fiber bundle and
the product space.

(2.226)

34
3

Gauss’ integral

35
36
4

Surfaces revisited.

Choose a moving frame in such a way that p~e1 , ~e2 q are a basis for the tangent
plane and ~e3 is the normal to the surface. On the surface itself

d~x “ σ1~e1 ` σ2~e2 (4.1)

where σ1 and σ2 are a couple of 1-forms. For example, in the case of the
two-sphere we can choose
ˆ ? ? ˙
1 B 1 x L2 ´x2 ´y 2 y L2 ´x2 ´y 2 a
2 2
~e1 “ L Bθ “ L
? , ? 2 2 ,´ x ` y “
x2 `y 2 x `y
´ ? ¯
“ L1 ?Lxz ? yz 2 ´ z2
2 ´z 2 , 2
L ´z 2 , ´ L
1 B ? 1
~e2 “ L sin θ Bφ “ L2 ´z 2 p´y, x, 0q

~e3 “ Br B
“ L1 px, y, zq (4.2)

Then
´ ¯
d~x “ dx, dy, ´ xdx`ydy
z “
´ ¯
yz
~e2 dx ` z ?LLy
` ˘
? Lx ~e1 ´ Lx ~e1 ` xz
“ z L2 ´z 2 2 ´z 2 Ly ~
e2 dy (4.3)

This determines the one-forms

σ1 “ ? L pxdx ` ydyq
z L2 ´z 2
σ2 “ ? 1 p´ydx ` xdyq (4.4)
L2 ´z 2

It is also the case that

b“3
ÿ
d~ea “ ωab~eb (4.5)
b“1

37
We find
´ 2
¯
xL2
d~e1 “ 1 ? z dx ` dz, ?L2z´z 2 dy ` 2 yL2 3{2 dz, ?L2z´z 2 dz “
L L2 ´z 2 pL2 ´z 2 q3{2 pL ´z q
1
` ˘
“ LpL2 ´z 2 q3{2
zpL2 ´ z 2 qdx ` xL2 dz, zpL2 ´ z 2 qdy ` yL2 dz, `zpL2 ´ z 2 qdz “
´ ´ ¯ ´ ¯
1 L2 x2 L2 xy L2 y 2 L2 xy
“ z x2 ` y 2 ´ z2
dx ´ L2 ´ z 2 ´
z dy, z z2
dy ´ z dx,
LpL2 ´z 2 q3{2
` ˘ ˘
, ´ L2 ´ z 2 pxdx ` ydyq (4.6)

´ ¯
d~e2 “ ´ ?Ldy 2 ´z 2 ´ yz
pL2 ´z 2 q3{2
dz, ? dx
L2 ´z 2
` xz
pL2 ´z 2 q3{2
dz, 0 “
` ˘
“ pL2 ´z1 2 q3{2 ´pL2 ´ z 2 qdy ´ yzdz, pL2 ´ z 2 qdx ` xzdz, 0 “
` ˘
“ 2 1 2 3{2 xydx ´ x2 dy, y 2 dx ´ xydy, 0 (4.7)
pL ´z q

1
d~e3 “ L pdx, dy, dzq (4.8)

Then
xdy ´ ydx
ω12 ” ~e2 d~e1 “ z (4.9)
LpL2 ´ z 2 q

` ˘ `
ω13 “ ~e3 .d~e1 “ ?1 z pxdx ` ydyq ` pL2 ` z 2 qdz “ ?1 ´z pxdx ` ydyq ` pL2 ´
L2 L2 ´z 2 L2 L2 ´z 2
“ ´ z ?L12 ´z 2 pxdx ` ydyq

(because on the sphere xdx ` ydy ` zdz “ 0)

1
ω23 “ ~e3 .d~e2 “ ? pydx ´ xdyq (4.11)
L L2 ´ z 2
But we have normalized in such a way that

~ea .~eb “ δab (4.12)

so that
d~ea .~eb ` ~ea .d~eb “ 0 (4.13)
that is that the matrix of one-forms

Ω ” ωab (4.14)

is antisymmetric. In fact
z
ω21 “ ~e1 .d~e2 “ pydx ´ xdyq “ ´ω12 (4.15)
LpL2 ´ z2q

38
 ¨ ˛
0 $ ´ω1
Ω “ ´ΩT ” ˝´$ 0 ´ω2 ‚ (4.16)
ω1 ω2 0
We can the write in gory detail
d~e1 “ $~e2 ´ ω1~e3
d~e2 “ ´$~e1 ´ ω2~e3
d~e3 “ ω1~e1 ` ω2~e2
(4.17)
As a consequence of our definitions we have
A“2
ÿ A“2
ÿ
0 “ d2 ~x “ dσA~eA ´ σA d~eA “ 0 “ dσA~eA ´ σA ωAb~eb (4.18)
A“1 A“1

This implies
dσ “ σΩ (4.19)
That is
dσ1 “ $ ^ σ2
dσ2 “ ´$ ^ σ1
dσ3 ” 0 “ σ1 ^ ω1 ` σ2 ^ ω2 (4.20)
We also deduce
0 “ d2 σ “ dσΩ ´ σdΩ “ σΩ2 ´ σdΩ (4.21)
so that
dΩ “ Ω2 (4.22)
¨ ˛ ¨
´$2 ´ ω12
˛
0 $ ´ω1 ´ω1 ω2 ´$ω2
d ˝´$ 0 ´ω2 ‚ “ ˝ ´ω2 ω1 ´$2 ´ ω22 $ω1 ‚ (4.23)
ω1 ω2 0 ´ω2 $ ω1 $ ´ω12 ´ ω22
To be specific
d$ “ ´ω1 ^ ω2
dω1 “ $ ^ ω2
dω2 “ ´$ ^ ω1 (4.24)
Let us recap.
σ1 “ ? L pxdx ` ydyq
z L2 ´z 2
σ2 “ ?L21´z 2 p´ydx ` xdyq
ω1 “ z ?L12 ´z 2 pxdx ` ydyq “ L1 σ1
ω2 “ ´ L?L12 ´z 2 pydx ´ xdyq “ L1 σ2 (4.25)

39
Since there is only one linearly independent 2-form om the two-dimensional
surface Σ, we have
ω1 ^ ω2 “ Kσ1 ^ σ2 (4.26)
where K is our old friend the Gaussian curvature. For the sphere S2 we find
1
K“ (4.27)
L2
It is actually independent on the choice of the basis vectors p~e1 , ~e2 q. Exactly
the same reasoning tells us that

σ1 ^ ω2 ´ σ2 ^ ω1 “ 2Hσ1 ^ σ2 (4.28)

Here H is the mean curvature of the surface Σ. For the sphere S2 it reads
H “ L1 . In order to write the forms pω1 , ω2 q in terms of the forms pσ1 , σ2 q,
we have to be consistent wit the equation

σ1 ^ ω1 ` σ2 ^ ω2 “ 0 (4.29)

The general solution Flanders claims to be

ω1 “ pσ1 ` qσ2
ω2 “ qσ1 ` rσ2 (4.30)

Actually, for the sphere, this is satisfied in a trivial way, namely

1
ω1 “ L σ1
1
ω2 “ L σ2 (4.31)

(That is q “ 0 and p “ r.)

It follows that
p`r
H“ 2
K “ pr ´ q 2 (4.32)

Now the relation

d$ ` ω1 ^ ω2 ùñ d$ ` Kσ1 ^ σ2 “ 0 (4.33)

which determines K in terms of p$, σ1 , σ2 q. But

dσ1 “ $ ^ σ2
dσ2 “ ´$ ^ σ1 (4.34)

determine $ in terms of pσ1 , σ2 q; actually

$ “ aσ1 ` bσ2 (4.35)

Then K is completely determined by pσ1 , σ2 q. Lo and behold, this is Gauss’

theorema egregium.

40
5

Differential Geometry

It is important to be able to pinpoint characteristics that are intrinsic, that

is, independent of the coordinates used in overlaps of open sets in a covering.
The two main ones are

• The contravariant vector interpreted a a directional derivative. Gi-

ven the linear space FpM q of all functions f : M Ñ R, and a local
coordinate system xµ : M Ñ U Ă Rn

V P T : FpM q Ñ R (5.1)

V pf q ” V µ Bµ f (5.2)

It follows that

B 1 B
Vµ µ
“ V µ µ1 (5.3)
Bx Bx

• A covariant vector interpreted as the differential of a function

W P T ˚ : f P FpM q Ñ dFpM q (5.4)

df ” Bµ f dxµ (5.5)

this means that

Bf Bf 1
µ
dxµ “ µ1 dxµ (5.6)
Bx Bx

Starting with those elements, more complicated transformation laws can be

easily derived.

41
5.1 Geodesics
The integral given the distance is extended over the parameterized curve γ

xµ “ xµ pλq (5.7)

and we have denoted by

dxµ
x9 µ ” (5.8)
dλ
We can normalize the tangent vector
x9 µ
uµ ” ? (5.9)
x9 2
The extrema of the action are by definition the geodesics of the manifold.
We get
ż ż "
ρ µ ν µ ν ρ
δS “ ´mc dλ tBρ gµν δx x9 x9 ` 2gµν x9 δ x9 u “ ´mc dλ δx
*
µ ν λ µ µ
Bρ gµν x9 x9 ´ pBλ gµρ ` Bµ gλρ q x9 x9 ´ 2gµρ x
: (5.10)

Expressed in the form of four ordinary differential equations for the four
functions of one variable xµ psq they read

d2 xµ α
µ dx dx
β
` Γαβ “0 (5.11)
ds2 ds ds
Here the Christoffel symbols are given by
1 ρσ
Γλ,µν ” gλρ Γρµν ” gλρ g p´Bσ gµν ` Bµ gνσ ` Bµ gνσ q (5.12)
2
This is true insofar as we are parametrizing the curve using the arc length
(which is anyway possible only when the tangent to the curve is everywhere
timelike or spacelike). Assume now we use another parameter,

λ “ λpsq (5.13)

Then
dxα dxα dλ
uα ” ds “ dλ ds
duα d2 xα dλ 2 α d2 λ
` ˘
ds “ dλ2 ds ` dx
dλ ds2 (5.14)

This means that the geodesic equations now read

d2 xα β
α dx dx
γ dxα
` Γβγ “ f pλq (5.15)
dλ2 dλ dλ dλ

42
where the function f pλq is given by
d2 λ
2
cpλq ” ´ ` ds ˘2 (5.16)
dλ
ds

The Christoffel symbols are in a sense the gauge field associated to

diffeomorphisms. An element V P T behaves as
1
µ1 Bxµ
1
V px q ” V ρ pxq (5.17)
Bxρ
or in matrix motation
V 1 “ J.V (5.18)
It is obvious that
Bα V µ (5.19)
does not transform as a tensor unless the diffeomorphism is a linear one,
because
dV 1 “ dJ.V ` JdV (5.20)
The idea on a connection (” gaugef ield) is to modify the ordinary detivative
into a covariant derivative

DV ” dV ` ΓV (5.21)

To be specific, the gauge fields are defined in such a way that

∇ρ V µ ” Bρ V µ ` Γµρσ V σ (5.22)

does transform as a tensor, that is

1
1 Bxσ Bxµ
∇ρ1 V µ px1 q “ ∇σ V λ (5.23)
Bxρ1 Bxλ
It is easier to vidualize all this in matrix notation

pDV q1 “ dV 1 ` Γ1 V 1 “ dpJ.V q ` Γ1 .J.V “ dJ.V ` J.dV ` Γ1 .J.V “

“ JDV “ J pdV ` ΓV q (5.24)

It is plain that for this to be true, it is enough that

Γ1 J ` dJ “ JΓ (5.25)

that is
Γ1 “ JΓJ ´1 ´ dJ.J ´1 (5.26)
In gory detail
λ1 B 2 xµ Bxσ
1
Bxσ Bxµ δ
1
µ1 Bx
Γρ1 λ1 λ` λ σ ρ1 “ ρ1 Γ (5.27)
Bx Bx Bx Bx Bx Bxδ σλ

43
Because of the inhomogeneous term the Christoffel symbols are NOT ten-
sors. Thay are connections, that is, gauge fields. It is useful exercise to
check at least that the Christoffel symbols are a solution of these equations.
Actually they are the unique solution involving the metric tensor alone.
It is also useful to check that for covariant tensors.

∇µ ων ” Bµ ων ´ Γλµν ωλ (5.28)

Using this formula, it is plain to check that the metric is covariantly constant

∇α gβγ “ Bα gβγ ´ Γλαβ gλγ ´ Γλαγ gλβ “ 0 (5.29)

• Let us begin by computing geodesics on the plane

ds2 “ dr2 ` r2 dθ2 (5.30)

Γrθθ “ 12 g rr p´Br gθθ q “ ´r

Γθrθ “ 12 g θθ pBr gθθ q “ 1
r (5.31)

r: ´ rθ92 “ 0
θ: ` 1 θ9r9 “ 0
r (5.32)

First integral
r9 2 ` r2 θ92 “ 1 (5.33)
It easier to start from
d ˆ ˙
dθ 2
ż
L“ dr r2 `1 (5.34)
dr

Euler-Lagrange ¨ ˛
d ˝ 2 θ1
r b ` ˘ ‚“ 0 (5.35)
dr 2 dθ 2
r dr ` 1

θ1
6 r2 b ` ˘ “C (5.36)
2
r2 dθ
dr ` 1

It is not difficult to check that the equation of a general planar straight

line
r sinpθ ` θ0 q “ r0 sin θ0 cos θ0 (5.37)
is a solution of the first integral.

44
 • Let us compute now the geodesics on the ordinary two-sphere. The
arc distance is ż a
S” dθ2 ` sin2 θ dφ2 (5.38)
Let us describe the curve as
φ “ φpθq (5.39)
Then d
ż ˆ ˙2
dφ
S” dθ 1` sin2 θ (5.40)
dθ
Euler-Lagrange
ˆ ˙
d
b b
B 2 B
1 2
1 ` sin θ pφ q “ 1 ` sin2 θ pφ1 q2 “ 0 (5.41)
dθ Bφ1 Bφ
This means that
sin2 θ φ1
a “C (5.42)
1 ` sin2 θ pφ1 q2
That is
C
φ1 “ ? (5.43)
sin θ sin2 θ ´ C 2
ż
C
φ “ dθ ? (5.44)
sin θ sin2 θ ´ C 2
Let us change variables
dθ
u “ cot θ, du “ ´ dθ (5.45)
sin2 θ
?
1 ´ C2
ż ż
du du ´1 u
φ “ ´C a “´ ? “ cos `φ0 ; a”
1 ´ C 2 p1 ` u2 q a2 ´ u2 a C
(5.46)

cot θ “ a cos pφ ´ φ0 q (5.47)

This equation has got a nice geometric interpretation. Consider a fixed
unit vector defined by pθ0 , φ0 q. The plane orthogonal to it is generated
by the vectors such that
sin θ sin θ0 cos pφ ´ φ0 q ` cos θ cos θ0 “ 0 (5.48)
The interchapter of this plane with unit sphere yields the desired
geodesic.
The geodesic equation can be written as
∇u u “ f u (5.49)
that is, the tangent vector to the curve propagates parallel to itself, in the
sense that the tangent component of the covariant derivative is proportional
to the tangent vector itself.

45
5.2 Covariant derivative and curvature.
In fact the metric connection (Christoffels) is the unique symmetric connec-
tion such that the coveriant derivative of the metric vanishes.
Bα gµν ´ Γλαµ gλν ´ Γλαν gλµ “ 0
Bµ gαν ´ Γλνµ gλα ´ Γλαµ gλν “ 0
Bν gµα ´ Γλαν gλµ ´ Γλαµ gλα “ 0 (5.50)
1-2+3 yields
´2 Γλαν gλµ “ Bα ` Bν ´ Bµ (5.51)
The commutator of two vectors, X, Y P T is defined as the vector
T ˆT ÑT (5.52)
rX, Y sα ” X µ Bµ Y α ´ Y µ Bµ X α “ X µ ∇µ Y α ´ Y µ ∇µ X α (5.53)
Ths covariant derivative in the direction of a vector V is
∇V : T Ñ T (5.54)

´ ¯
α µ αα α µλ
p∇V W q ” V ∇µ W ” V Bµ W ` Γµλ W (5.55)
The curvature of the connection ∇ is defines as the operator
R : T3 ” T ˆ T ˆ T Ñ T (5.56)
Z Ñ RXY Z ” r∇X , ∇Y s Z ´ ∇rX,Y s Z (5.57)
Let us slowly work this out
p∇Y Zqα “ Y λ pBλ Z α ` Γαλσ Z σ q (5.58)
The commmutator of two covariant derivatives reads
´ ¯
α ρ α α δ
p∇X ∇Y Zq ” X Bρ p∇Y Zq ` Γρδ p∇Y Zq “
" ´ ¯*
ρ
` λ α λ α σ
˘ α σ δ σ δ β
“ X Bρ Y Bλ Z ` Y Γλσ Z ` Γρδ Y Bσ Z ` Y Γσβ Z “
"
“ X Bρ Y λ Bλ Z α ` Y λ Bρλ Z α ` Bρ Y λ Γαλσ Z σ ` Y λ Bρ Γαλσ Z σ ` Y λ Γαλσ Bρ Z σ `
ρ

´ ¯*
α σ δ
`Γρδ Y Bσ Z ` Y Γσβ Z σ δ β (5.59)

In the opposite order

"
α
p∇Y ∇X Zq ” Y Bρ X λ Bλ Z α ` X λ Bρλ Z α ` Bρ X λ Γαλσ Z σ ` X λ Bρ Γαλσ Z σ ` X λ Γαλσ Bρ Z σ `
ρ

´ ¯*
α σ δ
`Γρδ X Bσ Z ` X Γσβ Z σ δ β (5.60)

46
On the other hand, the covariant derivative in the direction of the commu-
tator reads
` ˘α ` ˘
∇rX,Y s Z “ X µ ∇µ Y λ ´ Y µ ∇µ X λ pBλ Z α ` Γαλσ Z σ q (5.61)

In the commutator all terms proportional to derivatives of the vector we are

mapping, BZ disappear, what is left out is a linear mapping

pX, Y, Zq P T 3 Ñ pRXY Zqα ” Rα σρλ Z σ X ρ Y λ (5.62)

The tensor Rα σρλ is called the Riemann tensor ans is by construction

antisymmetric in the last two indices

Rα σρλ “ ´Rα σλρ (5.63)

Its value can be easily read out from the preceding formulas

Rα σρλ ” Bρ Γαλσ ´ Bλ Γαρσ ` Γαδρ Γδσλ ´ Γαδλ Γδσρ (5.64)

Lets work out the two-sphere as an example. We shall actually consider a

rugby ball.

5.3 Differential manifolds

Differential manifolds are smoooth objects that locally are similar to Rn , but
globally are different. Instead of giving the general theory we shall content
ourselves here with a detailed study of the simplest non-trivial example. Let
us first consider the simpler case of ordinary spheres embedded in euclidean
space.
The sphere Sn of radius l embedded in Rn`1 is defined thtough the
equations
A“n`1
ÿ
2
XA “ l2 (5.65)
A“1

where a point in Rn`1 is represented by the (n+1) coordinates pX1 , X1 , . . . Xn`1 q.

We are all used to polar coordinates, a generalization of the polar angles pθ, φq
for the two-sphere S2 . We need n angles to define a point in the n sphere.
We shall call these angles, θ1 . . . θn , and to be specific,

Xn`1 “ r cos θn
Xn “ r sin θn cos θn´1
...
X2 “ r sin θn sin θn´1 . . . cos θ1
X1 “ r sin θn sin θn´1 . . . sin θ1 (5.66)

47
(were we to use r itself as the radial coordinate, those would be polar
coordinates in Rn`1 , in them the equation of the sphere is simply

r “ l “ constant (5.67)

here

0 ď θ1 ď 2π
0 ď θj ď π for j ‰ 1 (5.68)

The Xn`1 axis is special in those coordinates; any axis however can be taken
as the Xn`1 axis. The metric induced on S n by the euclidean metric in Rn`1
is

ds2n “ δAB dX A pθqdX B pθq “

“ dθn2 ` sin2 θn dθn´1
2 ` . . . ` sin2 θn sin2 θn´1 . . . sin2 θ2 dθ12 (5.69)

id est, in a recurrent form

ds21 “ dθ12
ds2n “ dθn2 ` sin2 θn ds2n´1 (5.70)

The tangent space is a vector space Tn with the same dimension as the
manifold itself. It can be defined as the set of vectors orthogonal to the
normal vector

nA “ XA (5.71)
In general, given a surface in Rn`1 defined by the equation

f pXA q “ 0 (5.72)

the normal vector is given by the gradient

nA ” BA f (5.73)

To come back to the sphere, the tangent space is defined as those vectors
that obey ÿ
xA tA “ 0 (5.74)
A
Particularizing to the two-dimensional sphere, the tangent space is now the
tangent plane, that is, the set of vector in R3 such that

n1 .sin θ cos φ ` n2 .sin θ sin φ ` n0 cos θ “ 0 (5.75)

In the North or South pole (θ “ 0, πq the tangent plane is just the plane

X0 “ ˘l (5.76)

48
that is, the set of vectors
p0, n1 , n2 q (5.77)
and in the equator (θ “ π2 )

n1 cos φ ` n2 sin φ “ 0 (5.78)

Polar coordinates do not cover the whole sphere (neither do they cover eu-
clidean space). They are not well defined at the two poles. It is interesting
to study other set of coordinates, which are actually close to what carto-
graphers do when drawing maps. The stereographic coordinates are defined
out of one of the poles (either North or South) Northern pole stereographic
projection
2l Xµ
xµS ” Xµ ” (5.79)
X0 ` l ΩS
(µ “ 1 . . . n). Let us choose cartesian coordinates in Rn`1 with origin in
the South pole itself. This meags tgeat the South pole is represented by
X A “ 0, and the norh pole by XA “ pl, 0, . . . , 0q. One can imagine that one
is projecting a point P pXA q P Sn from the South pole into into a point xµS
that one van view as living on the tangent plane at the North pole.
x2S
1´ 4l2
X0 “ l x2S
“ lp2ΩS ´ 1q “ lp2ΩN ` 1q (5.80)
1` 4l2

1
ΩS ” x2S
(5.81)
1` 4l2

x2S l ´ X0
2
“ (5.82)
4l l ` X0
This means that when X0 “ l (the North pole) then

x2S
“0 (5.83)
4l2
and when X0 “ ´l (the South pole) then

XS2 “ 8 (5.84)

The jacobians of the embedding is

xµ
Bµ X 0 “ ´Ω2S
l
xα xµ
Bµ X α “ ΩS δµα ´ Ω2S (5.85)
2l2
The induced metric

ds2 “ δAB Bµ X A Bν X B dxµ dxν “ Ω2S δµν dxµ dxν (5.86)

49
Performing the North pole projection

2l
xµN ” Xµ (5.87)
X0 ´ l

Uniqueness of X0 means that

2ΩN ` 1 “ 2ΩS ´ 1 (5.88)

and uniqueness of X µ

ΩS µ 4l2
xµN “ xS “ ´ 2 xµS (5.89)
ΩN xS

Conversely,
4l2 µ
xµS “ ´ x (5.90)
x2N N
This leads to
1
ΩN “ ´ x2N
(5.91)
1` 4l2

The antipodal map

X A Ø ´X A (5.92)
is equivalent to
xµS Ø xµN (5.93)
and the jacobian is

BxµN 4l2 xµS xνS

ˆ ˙
µ
“ ´ δν ´ 2 2 (5.94)
BxνS x2S xS

Only functions which are invariant under the exchange of North and South
pole stereographic coordinates are well defined on the sphere.

5.4 The two-sphere, S2

Let us work out in detail the two dimensional case. Define dimensionless
corrdinates asd
x x
ξ1 ” ´ z`l η1 “ l´z
y y
ξ2 ” z`l η2 “ l´z (5.95)

It then follows that

1 ´ ξ2 η2 ´ 1
z“L “ l (5.96)
1 ` ξ2 η2 ´ 1

50
and the change N/S now reads
1 1
η1 “ ξ2
ξ
η2 “ ´ ξ12 ξ 2 (5.97)

Then defining the complex variable

z ” ξ 1 ` iξ 2 (5.98)

the change N/S reduces to

1
zÑw” (5.99)
z
Consider now a field of vectors
B B 1 B B
ln pzq “ zn “ ´ 2 ln pzq “ ´w2´n (5.100)
Bz Bz z Bw Bw
If we want the field to be non-singulkar for all values of z and w, then

ně0 & 2´ně0 (5.101)

so that the only vector fields globally defined on the two-sphete S2 are
˘ B ˘ B
a ` bz ` cz 2 “ ´ aw2 ` bw ` c
` `
(5.102)
Bz Bw

(5.103)

The induced metric on the sphere reads

dx2S dx2N
ds2 “ x2S 2
“ x2N 2
(5.104)
p1 ` 4l2
q p1 ` 4l2
q

which is conformally flat. This is the main virtue of these coordinates, and
the reason why cartographers are fond of them, We shall call a frame a basis
on the tangent space to the sphere as a manifold. Let us define a frame
through
δab eaµ ebν “ gµν (5.105)
The frames are given by
1
peS qaµ “ δaµ x2S
(5.106)
1` 4l2

1
peN qaµ “ ´δaµ x2N
(5.107)
1` 4l2

51
It is easy to check that
xµ
S xa
S
δµa ´ 2 xS2 BxνN
Lab pxq peS qbµ ” “ peN qaν (5.108)
1`
x2S BxµS
4l2

where the position dependent rotation is given by

xa xb
Lab ” δba ´ 2 (5.109)
x2
In fact this was the reason for the apparently arbitrary minus sign in front
of the definition of eN , which is unneccessary to reproduce the metric.
There are many reasons to be drawn from this example. First of all,
it is never possible to cover a non trivial manifold with a single coordinate
system. In this case we need at least two, namely North and South stereogra-
phic coordinates. Second, at each coordinate system, there is a frame in the
tangent space, and if we refer all quantities to this frame formal operations
are similar to the flat space ones.

52
6

Moving frames and

curvature.

The analogous to the field strength tensor for gauge theories is then the
Riemann-Christoffel tensor

Rµ ναβ ” Bα Γµνβ ´ Bβ Γµνα ` Γµσα Γσνβ ´ Γµσβ Γσνα (6.1)

The Ricci tensor is defined by contracting indices

Rµν ” Rλ µλν (6.2)

Recall the algebraic Bianchi identity

Rµ αβγ ` Rµ γαβ ` Rµ βγα “ 0 (6.3)

Clever use of this identity allows to prove that

Rαβγδ “ Rγδαβ (6.4)

Let us see it. We start with

Rαλµν ` Rαµνλ ` Rανλµ “ 0

Rλαµν ` Rλναµ ` Rλµνα “ 0 (6.5)

Substracting

2Rαλµν ` Rαµνλ ` Rανλµ ´ Rλναµ ´ Rλµνα “ 0 (6.6)

The same equation with the indices interchanged

pαλq Ñ pµνq (6.7)

2Rµναλ ` Rµαλν ` Rµλνα ´ Rνλµα ´ Rναλµ “ 0 (6.8)

53
conveys the fact that
Rµναλ “ Rαλµν (6.9)
We have then a symmetric tensor RIJ where each index is in the antisym-
metric rαβs (that is, D ” npn´1q
2 values). This yields

n2 pn2 ´ 1q
ˆ ˙
DpD ` 1q n
´ “ (6.10)
2 4 12

(we withdraw n4 because of the algebraic Bianchi identity) independent

` ˘

components. Id est, 20 in n=4 dimensions.

There are also some differential identities, the Bianchi identities

∇α Rµ νβγ ` ∇γ Rµ ναβ ` ∇β Rµ νγα “ 0 (6.11)

Contracting δµβ
∇α Rνγ ´ ∇γ Rνα ` ∇µ Rµ νγα “ 0 (6.12)
Contracting again g να

∇α Rαγ ´ ∇γ R ` ∇µ Rγµ “ 0 (6.13)

We shall derive most of these equations in a short while. Many useful formu-
las of tensor calculus are to be found in Eisenhart’s book, still indispensable.
Also very useful are the Ricci identities that state that

r∇α , ∇β s ωγ “ Rαβγδ ω δ (6.14)

This can actually be taken as the definition of the Riemann tensor, as is

done in many books.
All this means that the geodesic equations can be written as

uµ ∇µ uα “ 0 (6.15)

where the four-velocity of the massive particle is given by

dxα
uα ” (6.16)
ds
In general, the metric

da2 “ gµν pxq dxµ dxν (6.17)

is not flat; to the extent that it differs from the flat metric, it indicates the
presence of a gravitational field. At each point there are tensors (or spinors)
that represent physical observables. For example, the energy momentum
tensor
Tµν pxq (6.18)

54
This tensor live in the tangent space; the set of all tangent spaces of the
manifold is the tangent bundle. A frame is a basis of the tangent vector
space at a given point of the space-time manifold. This four vectors are
represented by
Eaµ Bµ (6.19)
where the index a “ 0, 1, 2, 3 labels the four different vectors. The simplest
possibility is to choose one of them timelike (this is the one labeled E0 ) ,
and the other three spacelike. Furthermore, they can be normalized in such
a way that
gµν Eaµ Ebν “ ηab (6.20)
This is the reason why latin indices are dubbed Lorentz indices, whereas
the ordinary spacetime indices are called Einstein indices. Such a frame is
precisely a LIF (where FREFOS live) and the physical observables measured
in the LIF are simply
Tab ” Tµν Eaµ Ebν (6.21)
The determinant of E considered as a matrix cannot vanish. We can
then define the coframe made out of the dual one-forms
ea pEb q “ δβa (6.22)
When indices are put in place, this is equivalent to computing the inverse
matrix
eaµ Ebµ ” δba
eaµ Eaν “ δµν (6.23)
From the normalization condition
gµν Eaµ Ebν “ ηab
and multiplying both members by the dual form eaσ
ñ eaµ “ gµν η ab Ebν
This means that the dual form is simply the frame with the Einstein indices
lowered with the spacetime metric, and the Lorentz indices raised with the
Lorentz metric. Following most physicists we shall represent both the frame
and the coframe with the same letter, although when neccessary we will
indicate explicitly its nature, as in
~ea ” eµa Bµ
ea ” eaµ dxµ (6.24)
The parallel propagator is defined once frames at different points are
selected by some mechanism
g α β 1 px, x1 q ” eαa pxqeaα1 px1 q (6.25)

55
Then physical quantities at different points are related through
1
Aα pxq ” g α β 1 px, x1 qaα px1 q (6.26)

For n-dimensional spheres in stereographic coordinates

ds2 “ Ω2 δµν dxµ dxν (6.27)

where
1
Ω” x2
(6.28)
1` 4L2
and the frame is defined by
eaµ “ Ωδµa (6.29)
in such a way that
1 µ
eµa “ δ (6.30)
Ω a
The Sn Christoffels read
Ωβ α Ωγ α Ωα
Γαβγ “ δ ` δ ´ δβγ (6.31)
Ω γ Ω β Ω
Under a local Lorentz transformation

Ea1 “ La1 b pxqEb (6.32)

Eaµ pxq is a nonsingular square n ˆ n matrix. The commutators are given by

c
rEa , Eb s “ Cab Ec

It is a fact that
1` a 1` a
dea “ Brµ eaρs dxρ ^ dxµ “ Bµ eν ´ Bν eaµ dxµ ^ dxν “ Bµ eν ´ Bν eaµ eµc eνd ec ^ ed “
˘ ˘
2 2
1` a ν a µ
˘ c d 1`
ed pec qeµ ´ eν ec peνd q ec ^ ed “
µ a a
˘
“ ec peν qed ´ ed peµ qec e ^ e “
2 2
1 1
“ red , ec sµ eaµ ec ^ ed “ ´ Ccd a c
e ^ ed (6.33)
2 2
To be specific, the structure constants read
´ ¯ ´ ¯
c
Cab “ ecµ eλa Bλ eµb ´ eλb Bλ eµa “ ecµ eλa ∇λ eµb ´ eλb ∇λ eµa (6.34)

(The Christoffels cancel when taking the antisymmetric part). In our Sn

example,
c Ωb Ωa
Cab “ 2 δac ´ 2 δbc (6.35)
Ω Ω
Under a local Lorentz transformation the vierbein transforms as
1
ea “ La b pxqeb (6.36)

56
This is not true of the derivatives of the vierbein, dea , owing to the term
in dLa b . We would like to introduce a gauge field (connection) in the LIF,
the so called spin connection, such that the two-form

Dea ” dea ` ω a b ^ eb (6.37)

transforms as
pDea q1 “ La bDeb (6.38)
For this to be true we need
´ ¯ ` ˘ ´ ¯ ´ ¯
a
d La b eb ` ω 1 b ^ Lb c ec “ La b deb ` ω b c ^ ec (6.39)

This is equivalent to
` ˘a
dLa b ^ eb ` ω 1 b ^ Lb c ec “ La b ω b c ^ ec (6.40)

which is kosher provided

` ˘a
dLa c ` ω 1 b Lb c “ La b ω b c (6.41)

Lorentz transformations are such that

Lac Lad “ δdc “ Lca Lda (6.42)

Finally we get the transformation law for the gauge field

` 1 ˘a a b c a c
ω d “ L b ω c Ld ´ dL c Ld (6.43)

At the linear level, where

Lab ” ηab ` ab (6.44)
δω a bµ “ ´Bµ ω a b ` r, ωsa b (6.45)
This should be valid for any field living in the LIF that transforms with a
representation of the Lorentz group. But any field can be so represented.
For example, a vector field, V µ is projected on the LIF by a FREFO as V a ”
eaµ V µ . We want that its Lorentz covariant derivative is also the projection
of Einstein’s covariant derivative, that is

∇L pV a q “ eaµ p∇E V qµ (6.46)

This physical requirement determines the relationship between Lorentz and

Einstein connections to be

ω a bσ “ eaλ Γλµσ eµb ´ eρb Bρ eaσ (6.47)

It is a fact (confer [15]) that the torsion can be defined through the connec-
tion ωba by
1 a b
dea ` ωba ^ eb ” T a ” Tbc e ^ ec
2

57
Demanding that the tangent metric is covariantly constant we learn that
d d
∇a ηbc “ 0 “ ´ωab ηdc ´ ωac ηdb ” ´ωc|ab ´ ωb|ac (6.48)

When the torsion vanishes, and in tensor form

Bρ eaσ ´ Bσ eaρ ` ω a σρ ´ ω a ρσ “ 0 (6.49)

it follows that

ωa|bc ´ ωa|cb “ pBρ eaσ ´ Bσ eaρ q eσb eρc ” ~eb Bc ea ´ ~ec Bb ea “ ea Bb~ec ´ ea Bc~eb “
d~
“ ea . r~eb , ~ec s “ ea .Cbc ed ” Ca|bc (6.50)

where we have used the fact that

~eb Bc ea “ ´ea Bc~eb (6.51)

This means that the torsion-free condition completely determines the anti-
symmetric part of the connection. One often is interested in the case when
the connection lies in the Lie algebra of a simple group. For example, if
ωµ P SOpnqq
ωµ|ab “ ´ωµ|ba (6.52)
For spheres we have
ˆ ˙
1 Ωc Ωb
ωa|bc “ δab ´ 2 δac (6.53)
2 Ω2 Ω
ˆ ˙
Ωb Ωa Ωb Ωµ Ωa Ωµ
2ωµ|ab “ δµa ´ δµb “ δaµ ` δab ´ δbµ ´ δab (6.54)
Ω Ω Ω Ω Ω Ω

We see that this is equivalent to our physical postulate of FREFOs and

FIDOS. The curvature of the connection is defined through
1 a c
dωba ` ωca ^ ωbc ” Ra b ” Rbcd e ^ ed
2
It is asy to check that this a true Lorentz tensor; that is, under a local
Lorentz transformation
Rba Ñ La c Rc d Lb d (6.55)
This leads immediately to Bianchi identities

dT a “ dωba ^ eb ´ ωba ^ deb “ pRba ´ ωca ^ ωbc q ^ eb ´ ωba ^ pT b ´ ωcb ^ ec q “

“ Rba ^ eb ´ ωba ^ T b
dRba “ dωca ^ ωbc ´ ωca ^ dωbc “
pRca ´ ωda ^ ωcd q ^ ωbc ´ ωca ^ pRbc ´ ωdc ^ ωbd q “ Rca ^ ωbc ´ ωca ^ Rbc (6.56)

58
For a Levi-Civita connection the algebraic Bianchi identity in a natural basis
reads
1 a b µνλ
Rba ^ eb “ 0 “ Rbµν eλ dx (6.57)
2
In gory detail

Rα rλµνs “ 0 “ Rα λµν ` Rα µνλ ` Rα νλµ (6.58)

Clever use of this identity allows to prove that

Rαβγδ “ Rγδαβ (6.59)

Let us see it. We start with

Rαλµν ` Rαµνλ ` Rανλµ “ 0

Rλαµν ` Rλναµ ` Rλµνα “ 0 (6.60)

Substracting

2Rαλµν ` Rαµνλ ` Rανλµ ´ Rλναµ ´ Rλµνα “ 0 (6.61)

The same equation with the indices interchanged

pαλq Ñ pµνq (6.62)

2Rµναλ ` Rµαλν ` Rµλνα ´ Rνλµα ´ Rναλµ “ 0 (6.63)

conveys the fact that
Rµναλ “ Rαλµν (6.64)
We have then a symmetric tensor RIJ where each index is in the antisym-
metric rαβs (that is, D ” npn´1q
2 values). This yields

n2 pn2 ´ 1q
ˆ ˙
DpD ` 1q n
´ “ (6.65)
2 4 12

(we withdraw n4 because of the algebraic Bianchi identity) independent

` ˘

components. Id est, 20 in n=4 dimensions. The differential identity in a

natural basis reads

∇rα Rµ βγδs ” ∇α Rµ βγδ ` ∇γ Rµ βδα ` ∇δ Rµ βαγ “ 0 (6.66)

where the overline on an index means that this particular index is absent
from the antisymmetrization. Now

∇rα Rµ βγδs ” Brα Rµ βγδs ` Γµrασ Rσ βγδs ´ Γσrαβ Rµ σγδs ´ Γσrαγ Rµ βsσδs ´ Γσαδ Rµ βγσs “
“ Brα Rµ βγδs ` Γµrασ Rσ βγδs ´ Γσrαβ Rµ σγδs (6.67)

59
Using the relationship between ωbµa and Γα derived above we are done. On
βµ
the other hand
´ ¯ ´ ¯
Bα Rµ βγδ “ Bα eµa ebβ Ra bγδ “ pBα eµa q ebβ Ra bγδ `eµa Bα ebβ Ra bγδ `eµa ebβ Bα Ra bγδ
(6.68)
It is a fact of life that

∇Ea pEb q ” Γcab Ec

a
Tbc “ Γabc ´ Γacb ´ Cbc
a

a
Rb,cd “ Ec Γadb ´ Ed Γacb ` Γedb Γace ´ Γecb Γade ´ Ccd
e a
Γeb (6.69)

It is nice exercise to check that the scalar curvature for a two-dimensional

surface

N
R“ D
“ 2
‰
N ” e eu gv ´ 2gv fu ` gu ` f rgv eu ` 2fu p2fv ´ gu q ´ ev p2fv ` gu qs `
“ ‰
`2f 2 revv ´ 2fuv ` gvv s ` g e2v ` eu p´2fv ` gu q ´ 2e pevv ` 2fuv ` gvv q
` ˘
D ” 2 f 2 ´ eg (6.70)

60
7

The Gauss-Codazzi equations

Consider a codimension one hypersurface given by the embedding

Σn´1 ãÑ Mn (7.1)

xα “ σ α py i q (7.2)
The indiced metric is given by

ds2n´1 “ hij dy i dy j “ gµν Bi σ µ Bj σ ν dy i dy j (7.3)

There are then two metric connections: the n-dimensional one, ∇g and
the (n-1)-dimensional one associated to the induced metric, Dh . From the
definition itself of the induced metric follows

0 “ Dk hij “ Bρ gαβ Dk σ ρ Bi σ α Bj σ b ` gαβ Dk pBi σ α qBj σ β ` gαβ Bi σ α Dk pBj σ β q

(7.4)
Cyclic permutations

Bρ gαβ Dj σ ρ Bk σ α Bi σ β ` gαβ Dj pBk σ α qBi σ β ` gαβ Bk σ α Dj pBi σ β q “ 0

Bρ gαβ Di σ ρ Bj σ α Bk σ β ` gαβ Di pBj σ α qBi σ β ` gαβ Bj σ α Di pBk σ β q “ 0

Adding 1+2-3 yields

1
0 “ gαβ Dj Dk σ α Di σ β ` Dk σ ρ Di σ α Dj σ β pBρ gαβ ` Bβ gρα ´ Bα gβρ q “
2
gαβ Dj Dk σ α Di σ β ` Dk σ ρ Di σ α Dj σ β tα, βρu “
´ ¯
“ gαβ Di σ β Dk Dj σ α ` t αβρ uDj σ β Dk σ ρ (7.5)

This means that

Dk Dj σ α “ ´t αβρ uDj σ β Dk σ ρ ` Kjk nα (7.6)

where the normal component reads

´ ¯
Kjk ” nα Dk Dj σ α ` t αβρ uDj σ β Dk σ ρ (7.7)

61
Taking the Dj
´ ¯
0 “ Dj gαβ Bi σ α nβ “ Dj gαβ Bi σ α nβ ` gαβ Dj Di σ α nβ ` gαβ Di σ α Dj nβ
(7.8)
On the other hand,

Dj gαβ “ Dj σ ν Bν gαβ “ Dj σ ν ptαν; βu ` tβν; αuq (7.9)

so that

Kjk “ nα t αβρ uDj σ β Dk σ ρ ´ gαβ Dj σ α Dk nβ ´ nβ Dk σ ρ ptαρ; βu ` tβρ; αuq Dj σ α “

´gαβ Dj σ α Dk nβ ´ nβ Dk σ ρ tβρ; αuDj σ α “ ´ξiα ∇ρ nα ξjρ (7.10)

This tensor is called the extrinsic curvature, and represents the derivative
of the normal vector, projected on the surface.
Our purpose in life is now to relate the Riemann tensor on the hypersur-
face (computed with the induced metric) with the corresponding Riemann
tensor of the spacetime manifold. Those are the famous Gauss-Codazzi
equations, which we purport now to derive. They were one of the pillars
of Gauss’ theorema egregium, [15] which asserts that If a curved surface is
developed upon any other surface whatever the measure of curvature in each
point remains unchanged.
We start with
´ ¯
0 “ Dj gαβ nα nβ “ Dj σ ρ ptαρ; βu ` tρβ; αuq nα nβ ` gαβ Dj nα nβ ` gαβ nα Dj nβ “
´ ¯
gαβ nβ Dj nα ` t αµν uDj σ pµ nνq “ gαβ nβ ∇µ nα Dj σ µ “ nα ∇µ nα ξjµ (7.11)

On the other hand, the explicit expression for the extrinsic curvature reads

Kij “ ´ξiα ∇ρ nα ξjρ (7.12)

First of all let us derive some properties of the extrinsic curvature. It is

symmetric, Kij “ Kji .

´Kij “ ∇β nα ξiα ξjβ “ ´nα ∇β ξiα ξjβ (7.13)

But ” ı
ξjβ , ξiα “ 0 (7.14)

so that
´Kij “ ´nα ξiα ∇β ξjα “ ∇β nα ξiβ ξlα “ Kji (7.15)
This symmetry implies a very useful formula for the extrinsic curvature,
namely
´Kij “ ∇pβ nαq ξiα ξjβ “ £pnqgαβ ξiα ξjβ (7.16)

62
By the way, in the physics jargon when Kij “ 0 it is said that it is a moment
of time symmetry.
On the other hand, remembering that

ξiα ξβi “ gβα ´ nα nβ (7.17)

we deduce that

´Kij ξµi “ ´ gµα ´ nα nµ ∇ρ nα ξjρ “ ´∇ρ nµ ξjρ

` ˘
(7.18)

(because of [7.11]).
Let us analyze the definition of extrinsic curvature in even more detail.
´ ¯
pDk Dj Di ´ Dj Dk Di q σ α “ ξm h Rhijk “ Dk ´t αβρ uξiβ ξjρ ` Kij nα ´
α mh
´ ¯
´Dj ´t αβρ uξiβ ξkρ ` Kik nα “ Bk t αβρ uξiβ ξjρ ´ t αβρ uDk ξiβ ξjρ ´ t αβρ uξiβ Dk ξjρ `

Dk Kij nα ` Kij Dk nα ` Bj t αβρ uξiβ ξkρ ´ t αβρ uDj ξiβ ξkρ ` t αβρ ξiβ Dj ξkρ ´ Dj Kik ´ Kik Dj nα

and using again the defnition of the extrinsic curvature to eliminate the term
with two derivatives,
´ ¯
h Rrijk rhs “ ´Bk t αβρ uξiβ ξjρ ´ t αβρ uξjρ ´t βµν uξiµ ξkν ` Kik nβ ` Dk Kij nα ` Kij Dk nα `
α mr
ξm
´ ¯
Bj t αβρ uξiβ ξkρ ` t αβρ uξkρ ´t βµν uξiµ ξjν ` Kij nβ ´ Dj Kik nα ´ Kik Dj nα “
´ ¯ ´ ¯
nα pDk Kij ´ Dj Kik q ` Kij Dk nα ` t αβρ unβ ξkρ ´ Kik Dj nα ` t αβρ nβ ξjρ ´
´ ¯
´ξiβ ξjρ ξkσ Bσ t αβρ u ´ Bρ t αβσ u ´ t αλρ ut λβσ ` t αλσ ut λβρ (7.19)

Using again the definition of the extrinsic curvature, as well as the one of
the full Riemann tensor, we get

h pRrijk rhs ` Kij Krk ´ Kik Krj q ´ nα pDk Kij ´ Dj Kik q “ ´ξiβ ξjρ ξkσ Rα βσρ rgs
α mr
ξm

This projects into the famous Gauss-Codazzi equations

Rlijk rhs ` Kil Kjk ´ Kik Klj “ ξlα ξiβ ξjρ ξkσ Rαβρσ rgs (7.20)

as well as

Dj Kik ´ Dk Kij “ ´nα ξiβ ξjρ ξkσ Rαβσρ rgs (7.21)

Please note that not all components of the full Riemann tensor can be recove-
red from the knowledge of the Riemann tensor computed on the hypersurface
plus the extrinsic curvature. As a matter of fact,
pnq
R“ pn´1q
Rij ij ` 2 pnq Ri nin “pn´1q R ` K 2 ´ Kij k ij ` 2 pnq Ri nin (7.22)

63
This means that an explicit computation of pnq Ri nin is needed before the
Einstein-Hilbert term could be written in the 1+(n-1) decomposition. To
do that, consider Ricci’s identity

∇γ ∇β nα ´ ∇β ∇γ nα “ Rρ αβγ nρ (7.23)

Now

nβ p∇γ ∇β nγ ´ ∇β ∇γ nγ q “ nβ g αγ Rρ αβγ nρ ” Rnα nα (7.24)

Besides,
´ ¯´ ¯
∇γ nβ ∇β nγ “ ∇γ nβ nβ nµ ` ξiβ ξ µi nγ nν ` ξjγ ξ jν ∇µ nν “

∇γ nβ ξiβ ξ µi ξjγ ξ jν ∇µ nν “ ´Kij K ij (7.25)

Summarizing,
` ˘ ` ˘
Rnα nα “ nβ ∇γ ∇β nγ ´ nβ ∇β ∇γ nγ “ ∇γ nβ ∇β nγ ´ ∇γ nβ ∇β nγ ´ ∇β nβ ∇γ nγ `
`∇β nβ ∇γ nγ “
` ˘
“ ∇γ nβ ∇β nγ ´ nγ ∇β nβ ` Kij K ij ´ K 2 (7.26)

Then
pnq
R“ pn´1q
R ` Kij K ij ´ K 2 ´ Bα V α (7.27)

64
8

Distributions

Dirac introduced a function such that

δpxq “ 0 x‰0 (8.1)
but ż8
dx δpxq “ 1 (8.2)
´8
Consider the function φpx, q defined in such a way that
φpx, q “ 0 rě
2
´
φpx, q “ e 2 ´r 2 rď (8.3)
It is clear that
1
f p0q “ ‰0 (8.4)
e
nevertheless
ż ż 2 ż1
´  ´ 1
φpx, q d3 x “ e 2 ´r2 d3 x “ 4π3 e 1´r2 dr “ C3 (8.5)
´ ´ 0

It is clear that Dirac’s function cannot be a true function. Laurent

Schwartz gave mathematical respectability to Dirac’s ideas by introducing
the concept of distributions. The main idea is to consider the dual of a
convenient function space.
To begin, let us start with the space of test functions K, of real functions
with continuous derivatives to all orders, and with compact support. It is
not empty (actually, our recent friend, the function φpx, q P K).
It can be shown that given any continuous function f(x) with bounded
support, there is always some φpxq P K arbitrarily close to it.
Define a distribution d P K 1 as a continuous linear functional on K
@φpxq P K xd, φpxqqy P R (8.6)
The two essential properties are

65
 •
xd, a1 φ1 ` a2 φ2 y “ α1 xd, φ1 y ` a2 xd, φ2 y (8.7)

• If the sequence tφn u converges to 0 in K, then the sequence

txd, φn yu (8.8)

converges to zero.

It is plain that any locally summable function f pxq is a particular case of a

distribution, just by defining
ż8
xf, φy ” dx f pxq φpxq (8.9)
´8

those are called regular distributions.

But there are distributions (dubbed singular) which can not be written
in such a way. The most important one is precisely the Dirac delta

xδpxq, φpxqy ” φp0q (8.10)

• It is natural to define the behavior under a translation

xdpx ´ aq, φpxqy “ xdpxq, φpx ` aqy (8.11)

• Under a reflexion

xdp´xq, φpxqy “ xdpxq, φp´xqy (8.12)

• Under a rescaling a regular distribution behaves as

ż ż
dx f px{λqφpxq “ λ dx f pxq φpλxq (8.13)

In general, we generalize this in n-dimensions to

xpλdq pxq, φpxqy ” λn xdpxq, φpλxqy (8.14)

Let us now introduce the space S (Schwartz) of infinitely differentiable func-

tions which, together with their derivatives, go to zero faster than any power
of 1r when r Ñ 8. For example
2
e´r P S (8.15)

It is clear that tempered distributions are a subset of distributions

S1 Ă K1 (8.16)

66
(the bigger the starting space, the smaller its dual). The derivative of a
distribution is defined as

xd1 , φpxqy ” ´xd, φ1 pxqy (8.17)

This result holds for regular distributions just by neglecting surface terms.
Let us work out some examples
• Consider the Heaviside function

θpxq “ 0 x ă 0
θpxq “ 1 x ą 0 (8.18)

Let us compute its derivative as a distribution.

ż8
1 1
xθ pxq, φpxqy ” ´xθpxq, φ pxq ” ´ φ1 pxqdx “ φp0q (8.19)
0

This means that in this sense,

θ1 pxq “ δpxq (8.20)

• Let us find now the derivative of the distribution

xλ` (8.21)

defined for ´1 ă λ ă 0 as

xλ` “ 0 x ď 0
xλ` “ xλ xą0 (8.22)

This is locally summable, which is not the case with the ordinary
derivative
λxλ´1 (8.23)
We have to regularize the integral
ż8
λxλ´1 dx (8.24)
0

According to the definition

´ ¯1 ż8 ż8
λ λ 1
x x` , φpxqy “ ´ x φ pxq dx ” ´ lim xλ φ1 pxq dx (8.25)
0 Ñ0 

Let us now integrate by parts with

du “ dφ ùñ u“φ`C
λ
v“x (8.26)

67
 This leads to
ˆ ˇ8 ż 8 ˙
ˇ
λ´1 ˇ λ´1
´ lim pφ ` Cq λx ˇ ´ λx pφ ` Cq dx (8.27)
Ñ0 


It is plain that in order for this to have a finite limit it is neccessary

that
C “ ´φp0q (8.28)
We are then led to the definition
´ ¯1 ż8
λ
x x` , φpxqy ” pφpxq ´ φp0qq λxλ´1 dx (8.29)
0

• Let us compute the derivative of

log px ` i0q ” lim logpx ` iyq (8.30)

yÑ0

Now
logpx ` i0q “ log |x| ` iπθp´xq (8.31)
We have seen that
θ1 pxq “ δpxq (8.32)
Now
θpxq ` θp´xq “ 1 ùñ θ1 pxq “ ´δpxq (8.33)
as well as
d x d 1 x 1
x2 “ |x|2 ùñ |x| “ ùñ log |x| “ “
dx |x| dx |x| |x| x
(8.34)
Then
d 1
logpx ` i0q “ ´ iπδpxq (8.35)
dx x
d
• Let us explore dx |x| in the sense of distributions
ş ş0 ş8
x|x|1 , f y ” ´ |x|f 1 ““ ´8 xf 1 ´ 0 xf 1 “
ş0
“ xf |0´8 ´ ´8 f ` 0 f ´ xf |8
ş8 ş
0 “ σpxqf (8.36)

There is not delta-component of |x|1 .

• Let is now compute the laplacian of the Newtonian potential.

1
∆ (8.37)
r

68
 Following the general definition
ż ż
1 1 1 1
x ∆ , φ y “ x , ∆φ y “ d3 x ∆φ ” lim d3 x ∆φ “
r r r Ñ0 rě r
ż ˆ ˆ ˙ ˆ ˙ ˙
1 i 1
lim d3 x ∇i ∇ φ ´ ∇i ∇i φ “
Ñ0 rě r r
ż ˆ ˆ ˙ ˆ ˙ ˆ ˙ ˙
3 1 i i1 1
“ lim d x ∇i ∇ φ ´ ∇i ∇ φ ` ∆ φ (8.38)
Ñ0 rě r r r

The volume term vanishes because 1r is harmonic on R3 {0. The surface

term is ˆ ˙
2 1 d 1
´ lim 4π φ ` φ 2 “ ´4πφp0q (8.39)
Ñ0  dn 
Then
1
∆ “ ´4πδ 3 pxq (8.40)
r

• It is illustrative to repeat this calculation for n=2.

ż ż
2
∆ log r φpxq d x ” log r ∆φ d2 x “
rě rě
ż " *
“ ∇ plog r ∇φq ´ ∆ log r φ ` ∇ p∇ log r φq “
rě
ż
1
“0`0` 2πr dr φ “ 2π φp0q (8.41)
0 r

• The derivative of a convergent sequence of differentiable functions also

converges to the derivative of the limit, in the sense of distributions.
For example, it is a fact that any series of the form
8
ÿ
an einx (8.42)
´8

whose coefficients increase no faster that a power of n when |n| Ñ 8

converges in the sense of distributions.
Proof. In fact such a series can always be obtained by sufficient number
of term-by-term derivatives of another series of the type
8
ÿ an
einx (8.43)
´8
pinqk

QED

69
• A sequence of distributions

da , d2 . . . dn (8.44)

is defined to converge to the distribution d if @φpxq P K

lim xdn , φy “ xd, φy (8.45)

nÑ8

It is a fact that every singular distribution is the limit of a sequence

of regular functionals.

• Let us work out in detail an important example Consider the series

8
ÿ 1
sin nx (8.46)
n“1
n

It is quite easy to check that it converges to the function

π´x
f pxq ” 0 ă x ă 2π (8.47)
2
Consider the function over two periods
π x
´ ´ ´ 2π ă x ă 0
2 2
π x
´ 0 ă x ă 2π (8.48)
2 2
which has a 2π discontinuity at the origin. Its derivative is
ż ż0 ż8
1 1 π`x 1 π´x 1
xd , φy ” ´ dx dpxq φ pxq “ dx φ pxq ´ φ “
´8 2 0 2
1 0 1 8
ż ż
π π
“ φp0q ´ φ dx ` φp0q ´ φ dx (8.49)
2 2 ´8 2 2 0

Then
1
d1 “ π δpxq ´ (8.50)
2
Differentiating the whole series, we get a delta at each discontinuity
8
ÿ 1 ÿ
cos nx “ ´ ` π δpx ´ 2πnq (8.51)
2 ´8

Euler’s formula now implies that

8
ÿ 8
ÿ
ix 2ix ´ix ´2ix inx
1`e `e `...`e `e `... ” e “ 2π δpx ´ 2πnq
´8 ´8
(8.52)

70
 • A delta convergent sequence tfi u, is one such that

– 1.-For any M ą 0 and for |a| ď M and |b| ď M the quantities

ˇż b ˇ
ˇ ˇ
ˇ fj pξq dξ ˇ ď C (8.53)
ˇ ˇ
a

where C is independent of a, b, j, but it may depend on M .

– 2.- For any fixed nonvanishing a and b
żb
lim fj pξq dξ “ 0 a ă b ă 0 or 0ăaăb
jÑ8 a
żb
lim fj pξq dξ “ 1 aă0ăb (8.54)
jÑ8 a

There are many examples. One of them is

1 
fj “ (8.55)
π x ` 2
2

as  Ñ 0.

8.1 Fourier transform

• The starting point in order to define the Fourier transform (FT) of
distributions is Parseval’s theorem. For regular distributions it asserts
that ż ż
1
˚
dxf pxqgpxq dx “ dk f˜˚ pkqg̃pkq (8.56)
2π
where we have defined the FT as
ż8
˜
f pkq ” eikx f pxq dx (8.57)
´8

This can be used to define a distribution in some space Z 1 (to be defined

in a moment) for every distribution in K 1 . This is by definition the FT
of the original distribution. Fourier transform establishes a one-to-one
mapping
K ÐÑ Z (8.58)
where Z is defined as follows. It consists of slowly increasing functions,
that is, all entire functions ψpsq (where s ” σ ` iτ ) such that

|s|q |ψpsq| ď Cea|τ | q “ 0, 1, 2 . . . (8.59)

where the constants a and Cq may depend on ψ.

71
• Let us compute the FT of Dirac’s delta.
ż
pδ̃, φ̃q ” 2πpδ, φq “ 2πφp0q “ φ̃pkq ” p1, φ̃q (8.60)

Ergo,
F T rδs “ δ̃ “ 1 (8.61)
Also
ż
p1̃, φ̃q “ 2πp1, φq “ 2π φpxq dx “ 2π φ̃p0q “ 2πpδ, φq (8.62)

F T r1s “ 1̃ “ 2πδ (8.63)

Similar computations lead to

F T rδ p2mq pxqs “ p´1qm k 2m

F T rδ p2m`1q pxqs “ p´1qm`1 ik 2m`1 (8.64)

• Consider now the function

1 sin νx
fν pxq ” (8.65)
π x
for 0 ă ν ă 8). First of all,
ż8
fν pxq dx “ 1 (8.66)
´8

It is easily done by using Cauchy’s theorem to compute

eiz
ż
“ 2πie (8.67)
z ´ i

Furthermore, for 0 ă a ă b the integrals

żb ż bν
1 sin y
fν pxq dx “ (8.68)
a π aν y

go to zero as ν Ñ 8. Moreover, this same integral is bounded uni-

formly @ν. Therefore we are dealing with a delta-convergent sequence.

lim fν pxq “ δpxq (8.69)

νÑ8

Now observe that żν

sin νx eiξx
“ dξ (8.70)
x ν 2π

72
 Then what we have just proved is that
żν
lim eiξx dξ “ 2πδpxq (8.71)
νÑ8 ´ν

In fact it is a theorem that every integrable function f pxq which does

not grow at infinity faster that some power of |x|, has got a Fourier
transform in the sense of distributions.
• Let us recall that the convolution of two ordinary functions (regular
distributions) is defined as
ż ż
xf ˚ g, φpxqy ” f pξqgpx ´ ξqdξdx “ dξdηf pξqgpηqφpξ ` ηq (8.72)

Let us examine the convolution of singular distributions. It is natural

to generalize this last version of convolution to
xt1 ˚ t2 , φy ” xt1 pxqt2 pyq, φpx ` yqy (8.73)
It is plain that
t1 ˚ t2 “ t2 ˚ t1 (8.74)
as well as
t1 ˚ pt2 ˚ t3 q “ pt1 ˚ t2 q ˚ t3 (8.75)
Let us now compute the convolution of Dirac’s delta.
xδ ˚ t, φy “ xδpxqtpyq, φpx ` yqy “ xtpyq, φpyqy “ xt, φy (8.76)
That is, Dirac’s delta acts as a unit with respect to convolutions
δ˚t“t (8.77)
Also,
B pt ˚ sq “ pBtq ˚ s “ t ˚ pBsq (8.78)
Indeed
xB pt ˚ sq , φy “ ´xt ˚ s, Bφy ” ´xtpxq, pxspyq, Bφpx ` yqyqy “ xBt ˚ s, φy
(8.79)
• Consider a linear differential equation with constant coefficients
P pBq y “ Jpxq (8.80)
Define an elementary solution or Green function as
P pBq G “ δ (8.81)
Then we can write solutions of our PDE as
y “J ˚G (8.82)
because
P pBq y “ J P pBq G “ J ˚ δ “ J (8.83)

73
• Consider the ODE
d2
ˆ ˙
` ω2 xptq “ jptq (8.84)
dt2
Let us first show that the function
eiω|t|
Gptq ” (8.85)
2iω
is an elementary solution. Indeed
d
Gptq “ iωGptq.σptq (8.86)
dt
and
d2 ` 2 ˘
Gptq “ ´ω ` iω2δptq Gptq (8.87)
dt2
Incidentalt, the same thing happens with
eiωt
GR ptq “ θptq (8.88)
2iω
as well as
eiωt
GR ptq “ ´θp´tq (8.89)
2iω
• We have seen previously that
1 1
Gpxq “ ´ n´2
pn ą 2q
pn ´ 2qΩn r
1 1
Gpxq “ ´ log pn “ 2q (8.90)
2π r
is an elementary solution of the laplacian. This leads to Poisson’s
formula for the newtonian potential due to a density ρpxq
ż ż
V pxq “ dξρpξqGpx ´ ξq (8.91)

That is,
ż
1 ρpξ1 , ξ2 , ξ3 q
V px, y, zq “ ´ a dξ1 dξ2 dξ3
4π pξ1 ´ xq2 ` pξ2 ´ yq2 ` pξ3 ´ zq2
(8.92)
• We know that every periodic locally summable fuction f pθq can be
written in the form of a Fourier series
ÿ8
f pθq “ cn einθ (8.93)
´8

Taking the Fourier transform term by term we easily get

8
ÿ
f˜pkq “ cn δpk ` nq (8.94)
´8

74
 • Let us derive the marvelous Poisson summation formula. Starting
from ÿ ÿ
einx “ 2π δpx ´ 2πnq (8.95)
we easily get
ÿ ÿ ÿ
f pxq ” einπx{L “ 2π δpπx{L ´ 2πnq “ 2L δpx ´ 2nLq (8.96)

Its Fourier transform

ż
˜
ÿ ÿ 1 ÿ n
f pkq ” dxe´2πixk δpx ´ 2nL “ 0 e´2πikp2nLq “ δpk ´ q
n 2L 2L
(8.97)
Now, the transform of a gaussian
2
gpxq ” e´x (8.98)

is another gaussian
? 2 x2
f˜pkq ” πe´π (8.99)
Let us apply Parseval’s theorem to this couple of functions
ż ż
f pxqgpxqdx “ f˜pk “ g̃p´kqdk (8.100)

We get Pôisson’s formula

ÿ 2 L2 1 ÿ ´ π2 m22
e´4m “ e 4L (8.101)
2L
This has got plentiful physical applications.

8.2 Distributions on submanifolds.

• Consider
δpf pxqq (8.102)
It is clear that in the simplest case
ż ż
δpf pxqqgpxqdx “ dtδptqgpxptqq (8.103)

where
t ” f pxq ñ dt “ f 1 pxqdx (8.104)
In some case, some care must be taken. For example, consider

δpx2 ´ m2 q (8.105)

75
 4

-3 -2 -1 1 2 3

-2

-4

Descriptio 8.1: Exmaple.

Then
ż8 ż ´µ ż8
dt dt
dx “ a ` a (8.106)
´8 8 ´2 t ` µ2 ´µ 2 t ` µ2
Then ż8
gp´µq gpµq
dxδpx2 ´ µ2 q gpxq “ ` (8.107)
´8 2µ 2µ

• Every functional concentrated on a point is a linear combination of

the delta function and its derivatives.
• Consider codimension one hypersurfaces given by

P px1 . . . xn q “ 0 (8.108)

We sould like to define such things as δpP q, etc. Let us assume that

Bµ P |P ‰ 0 (8.109)

Let us first define the Leray form ω as such that

dP ^ ω “ dpvolq (8.110)
BP
Provided Bx 1 ‰ 0, there is always some coordinate system such that
the equation of the surface reads

u1 “ P u2 “ x2 ... un “ xn (8.111)

76
Then
ˆ ˙
n Bx
1 2
dpvolq ” dx ^ dx ^ . . . dx “ det du1 ^ du2 . . . ^ dun “
Bu
1
BP
du1 ^ du2 . . . ^ dun (8.112)
Bx1

Ergo,
1
ω“ BP
dx2 . . . ^ dxn (8.113)
Bx1
In fact it can be shown that ω has an intrinsic meaning. It is only
natural to define ż
xδpP q, φy ” φpxq ω (8.114)
P “0
As an example, let us work out

δpxy ´ cq (8.115)

in two dimensions. Using the coordinates

u1 “ xy ´ c
u2 “ y (8.116)

Then
dy
ω“ (8.117)
y
because
dy
pxdy ` ydxq ^ “ dx ^ dy (8.118)
y
It is a fact ż ˆ ˙
c dy
xδ pxy ´ cq , φpx, yqy “ φ ,y (8.119)
y y
Let us work out another example, namely δpr ´ Rq The Leray form
coincides with the euclidean area element Rn´1 dΩ
ż
xδpr ´ Rq, φy “ Rn´1 φ dΩ (8.120)
r“R

It is to be noted that this vanishes when R “ 0, unless φ diverges in

an appropiate way (in which case pδ n p~xq, φq would diverge). If it were
instead δpr2 ´ c2 q, we could define

u1 “ r2 ´ R2
u2 “ θ1
...
un “ θn´1 (8.121)

77
 This means that

1 1
ω“dΩ “ Rn´1 dΩ (8.122)
2r 2R
and
Rn´2
ż
xδpr2 ´ R2 q, φy “ φ dΩ (8.123)
2 r“R
If we define a Heaviside function
θpP q “ 1 ðñ P pxq ě 0
θpP q “ 0 ðñ P pxq ă 0 (8.124)
Then it can be shown that
θ1 pP q “ δpP q (8.125)

• Let us now introduce a family of forms that depend both on φ P K

and on P.
ω0 pφq ” φ ω
dω0 ” dP ^ ω1 pφq
...
dωk´1 pφq ” dP ^ ωk pφq
... (8.126)
Then we define the derivatives of the delta function as
ż
pkq k
xδ pP q, φy ” p´1q ωk pφq (8.127)
P “0

For example, let us compute δ pkq pr ´ Rq. Using the same coordinates
as before, we recover
ω “ rn´1 dΩ (8.128)
Then
ω0 “ φ rn´1 dΩ (8.129)
and ` ˘
B φ rn´1
ω1 pφq “ dΩ (8.130)
Br
In fact
` ˘
B k φ rn´1
ωk pφq “ dΩ (8.131)
Brk
Then
` ˘
p´1qk B k φ rn´1
ż
pkq
xδ pr ´ Rq, φy “ n´1 dΩ (8.132)
R r“R Brk

78
9

Finite groups.

A group G is a set with a product

GˆGÑG (9.1)

such that
1.- g1 , g2 P G Ñ g1 g2 P G
2.- The composition law is associative: g1 pg2 g3 q “ pg1 g2 q g3 .
3.- There is a unit e P G, such that eg “ ge “ g @g P G.
4.- Every element has got an inverse g ´1 g “ gg ´1 “ e

• A group is finite if the set has a finite number of elements. This is called
the order of the group, |G|. Cyclic groups are particular instances such
that
@g P G, g n “ 1 (9.2)
for some integer n. For example, Z3 such that a3 “ b3 “ e, has got
the multiplication table

e a b
e e a b
(9.3)
a a b e
b b e a

An abelian group obeys

gh “ hg @g, h P G (9.4)

• A representation is a mapping

g P G Ñ Dpgq (9.5)

where Dpgq is a linear operator acting in some linear space V and such
that

79
 – Dpeq “ 1
– Dpg1 qDpg2 q “ Dpg1 g2 q

The dimension of the representation is the dimension of the linear

space V . For example a three dimensional representation of the cyclic
group Z3 in R3 is
¨ ˛
1 0 0
e Ñ Dpeq ” ˝0 1 0‚
0 0 1
¨ ˛
0 0 1
a Ñ Dpaq ” ˝1 0 0‚
0 1 0
¨ ˛
0 1 0
b Ñ Dpbq ” ˝0 0 1‚
1 0 0
(9.6)

• This is in fact the adjoint representation. We associate the elements

of the group with a basis in VN , a vector space of dimension equal to
the order of the group. For example,
¨ ˛
1
e Ø epeq ” e1 ” 0‚
˝
0
¨ ˛
0
a Ø epaq ” e2 ” 1‚
˝
0
¨ ˛
0
b Ø epbq ” e3 ” ˝0‚ (9.7)
1

Then we define
Dad pg1 qepg2 q ” epg1 g2 q (9.8)
With the natural definition

Dij pgq ” eTi Dpgqej (9.9)

It follows
3
ÿ
Dij pghq ” eTi Dpghqej “ eTi DpgqDphqej “ eTi Dpgq ek eTk Dphqej “
k“1
ÿ
“ Dik pgqDkj phq (9.10)
k

80
 If we change the basis of the linear space on which the representation
acts ÿ
ei Ñ ẽi ” Sij ej (9.11)
j

Then

ẽi Ñ D̃ij ẽj

ei Ñ Dij ej
` ´1 ˘l ˘k
ẽ Ñ Dil S ´1 q j ẽk
`
S i l
ẽ Ñ SDS ´1 ẽ (9.12)

That is
D̃ “ SDS ´1 (9.13)
It is said that D and D̃ are equivalent representations.
• It is fact that all representations of finite groups are equivalent to
unitary representations, that is, one such that

DD` “ D` D “ 1 (9.14)

This is easy to show, by considering the positive semidefinite matrix

ÿ
S” D` pgqDpgq “ U ´1 ΛU (9.15)
gPG

where
Λ “ diag pλ1 . . . λn q (9.16)
and all eigenvalues λi ě 0. Actually all λi ą 0 because if it were one
zero eigenvalue, then there must be a vector such that
ÿ
Sv “ 0 “ v ` Sv “ ||Dpgqv||2 (9.17)
gPG

which is impossible, because in particular Dpeq “ 1. This means that

there is a matrix
1
S 1{2 ” U ´1 Λ 2 U (9.18)
and defining
D̃pgq ” S 1{2 DpgqS ´1{2 (9.19)
we are done, because
« ˜ ¸ ff
“ ‰ ÿ
D̃` pgqD̃pgq “ S ´1{2 D` pgqSDpgq S ´1{2 “ S ´1{2 D` pgq D` phqDphq Dpgq S ´1{2 “
hPG
« ff « ff
ÿ ÿ
“ S ´1{2 D` phgqDphgq S ´1{2 “ S ´1{2 D` pkqDpkq S ´1{2 “ S ´1{2 SS ´1{2 “ 1(9.20)
hPG kPG

81
• A representation is reducible if it has an invariant subspace. If P 2 “ P
is the projector on this subspace, the condition is

@g P G P DpgqP “ DpgqP (9.21)

In the case of the regular representation of Z3 , there is an invariant

subspace with projector
¨ ˛
1 1 1
1˝
P “ 1 1 1‚ (9.22)
3
1 1 1

Actually what happens is that

@g P Z3 DpgqP “ P (9.23)

The restriction of the representation to the subspace is itself a repre-

sentation (in this case, the trivial representation). When this is not the
case, the representation is irreducible. A representation is completely
reducible if it can be written in block diagonal form as the direct sum
of irreducible subrepresentations

Dpgq “ D1 pgq ‘ D2 pgq ‘ . . . ‘ Dn pgq (9.24)

Again, for finite groups, any representation is completely reducible. In

our favorite Z3 example, defining
¨ ˛
1 1 1
1
S “ ˝1 α2 α ‚ (9.25)
3
1 α α2

where
2πi
α“e 3 (9.26)
Then
¨ ˛
1 0 0
D̃peq “ 0
˝ 1 0‚
0 0 1
¨ ˛
1 0 0
D̃paq “ 0
˝ α 0‚
0 0 α2
¨ ˛
1 0 0
D̃pbq “ 0
˝ α2 0 ‚
0 0 α
(9.27)

82
 In fact, every representation of a finite group is completely reduci-
ble. Let us work with the unitary form of the representation. It it is
reducible, it means that there is a projector P such that

@g P DpgqP “ DpgqP (9.28)

Taking the adjoint

@g P Dpgq` P “ P Dpgq` (9.29)

But D` pgq “ D´1 pgq “ Dpg ´1 q and g ´1 runs over G as well as g does
(because for every g, there is a unique g ´1 . To summarize, we claim
that
@g P DpgqP “ P Dpgq (9.30)
It follows that

@g p1´P q Dpgqp1´P q “ D´P D´DP `P DP “ Dpgqp1´P q (9.31)

and 1 ´ P also projects into an invariant subspace.

9.1 Normal subgroups

• Given a subgroup H, we define a (left) coset

gH (9.32)

as the set of all elements

gh @g P G @h P H (9.33)

It is plain that every element in G must be in one (and only one) coset,
because we first pick g1 R H and construct the |H| elements g1 H those
are all different. Then we pick some g2 R H, g2 R g1 H, and so on. This
proves the theorem of Lagrange.

|gG| ˆ |H| “ |G| (9.34)

There is an equivalence relationship when g1 , g2 P g1 H

g1 „ g2 ô g1 “ g2 h, hPH (9.35)

Then the quotient

G{ „ (9.36)
is the set of those left cosets.
It is important to distinguish that from another equivalence relatioship

g1 „ g2 ô Dh P G, hg1 “ g2 h (9.37)

83
 The set of those (conjugacy) classes is

G{ „ (9.38)

Those subgroups such that

@g P G gH “ Hg (9.39)

are dubbed normal. In this case, the coset space is also a group,
because

pg1 Hq pg2 Hq “ g1 Hg1´1 g1 g2 H “ Hg1 g2 H “ g1 g2 g2´1 g1´1 Hg1 g2 H “ g1 g2 HH “ g1 g2 H

(9.40)
G{H is called the factor group of G by H.

• The center of a group, Z is the set of all elements that commute with
all elements of the group,

z P Z Ø zg “ gz @g P G (9.41)

The center is an abelian invariant subgroup.

• Consider the permutation group S3 . Permutation groups are very

important for a variety of reasons. One of them is Cayley’s theorem:
Eevery finite group |G| “ n is isomorphic so a subgroup of Sn . The
elements of S3 are

e
a1 ” p123q
a2 ” p321q
a3 ” p12q
a4 ” p23q
a5 ” p31q
(9.42)

The multiplicaction table is given by

e e a1=(123) a2=(132) a3=(12) a4=(23) a5=(13)

a1=(123) a1 a2 e a5 a3 a4
a2=(132) a2 e a1 a4 a5 a3
(9.43)
a3=(12) a3 a4 a5 e a1 a2
a4=(23) a4 a5 a3 a2 e a1
a5=(13) a5 a3 a4 a1 a2 e

84
 First of all, let us notice that the subset te, a1, a2u is a subgroup, the
alternating group A3 , consisting on all even premutations, and which
in this case happens to be isomorphic to Z3 . Moreover

a3 A3 “ ta3 , a4 , a5 u
a4 A3 “ ta4 , a5 , a3 u
a5 A3 “ ta5 , a3 , a4 u (9.44)

• Coming back to our S3 example, A3 „ Z3 is a normal subgroup,

because

A3 a3 “ ta3 , a5 , a4 u
A3 a4 “ ta4 , a3 , a5 u
A3 a5 “ ta5 , a4 , a3 u (9.45)

Note however that there is another subgroup, H ” te, a4 u (remember

that a24 “ e) which is not normal.

a5 H “ ta5 , a2 u ‰ Ha5 “ ta5 , a1 u (9.46)

• The conjugacy classes are sets such that if they contain an element s,
they also contain all its conjugates

S ” tg ´1 sg @g P Gu (9.47)

It is a plain that for such a set

g ´1 Sg “ S (9.48)

e is always a conjugacy class. In S3 , taking into account that

a´1
1 “ a2
a23 “ a24 “ a25 “ e (9.49)

the conjugacy classes are: frist the two three-cycles

ta1 , a2 u (9.50)
and then the three two-cycles

ta3 , a4 , a5 u (9.51)

We note here a general trend in the symmetric group: conjugate per-

mutations have the same cycle structure; in particular the permuta-
tions in the same class are either all even or else all odd.

85
 • For fixed g P G, the mapping

h P G Ñ ghg ´1 P G (9.52)

is an inner automorphism. This is plain, because

` ˘` ˘
gg1 g ´1 gg2 g ´1 “ gg1 g2 g ´1 (9.53)

Besides, it is 1-1, because if

gg1 g ´1 “ gg2 g ´1 ñ g1 “ g2 (9.54)

Outer automorphisms are all those automorphisms that are not inner.

9.2 Schur’s lemma

• If there are two inequivalent irreducible representations D1 and D2 of
a group G, such that there is a matrix A that obeys

D1 pgqA “ AD2 pgq @g P G (9.55)

then it follows that A “ 0. In fact, assume there is a vector such that

Av “ 0. Then there is a projector P onto the subspace that annihilates
A on the right. This subspace is invariant under D2 , because

AD2 P “ D1 AP “ 0@g P G (9.56)

But D2 is irreducible, so that P “ 1 and A “ 0. If A annihilates one

state, it must annihilate them all.
If no vector annihilates A on either side, then it must be an invertible
square matrix, Then
D1 “ AD2 A´1 (9.57)
and the two representations are equivalent.
Another proof is as follows. Define

Dpei q ” ej Dji (9.58)

1 2
Dij Aja “ Aib Dba (9.59)
then
1
ei Dij Aja “ D1 pej Aja q “ ei Aib Dba
2
(9.60)
Denoting
Eb ” ei Aib (9.61)
this shows that
D1 pEa q “ Eb Dba
2
(9.62)
which is not possible if D1 is irreducible.

86
• On the other hand, it is a fact that if there is a finite dimensional
irreducible representation D such that

DpgqA “ ADpgq @g P G (9.63)

then A “ 1. This is obvious, because any finite dimensional matrix,

A, has at least one eigenvalue Then

Dpgq pA ´ λ1q “ pA ´ λq D (9.64)

and the matrix

A ´ λ1 (9.65)
has a null eigenvector. Then the former argument shows that

A ´ λ1 “ 0 (9.66)

One consequence of Schur’s lemma is that once the form of D is fixed,

there is no further freedom to make nontrivial similarity transforma-
tions on the states.

9.3 Characters

Given an arbitrary matrix, let us say, X, consider the matrix

ÿ
A” Dpgq X Dpg ´1 q (9.67)
gPG

where Dpgq is an matrix irrep of G with dimension dR . It is fact of

life that
rDphq, As “ 0 @h P G (9.68)

Indeed
ř ř ř
DphqA ” Dphq gPG Dpgq X Dpg ´1 q “ gPG DphgqXDpg ´1 q “ gPG DphgqXDpg ´1 h´1 qDphq “
ř
“ gPG Dpgq X Dpg ´1 qDphq ” ADphq (9.6

Schur’s now implies that

A “ λ1 (9.70)
Let us now choose as starting point the particular matrix

X ” pElm qij ” δil δjm (9.71)

Then ÿ
Dil pgqDmj pg ´1 q “ λlm δij (9.72)
gPG

87
Taking the trace δ ij we learn that

|G|δlm “ λlm dR (9.73)

This proves the orthogonality relation

ÿ |G|
Dil pgqDmj pg ´1 q “ δlm δij (9.74)
gPG
dR

Let us now repeat the same procedure using two different representa-
tions, id est, ÿ
B” D2 pgqXD1 pg ´1 q (9.75)
gPG

It is plain that
D2 phqB “ BD1 phq (9.76)

Schur’s tell us that

B“0 (9.77)

and using
X ” Elm (9.78)

we learn that ÿ
Dil2 pgq Dmj
1
pg ´1 q “ 0 (9.79)
gPG

We can characterize both orthonomality relations in the following way.

Consider the set of all irreps
µ
Dij pgq (9.80)

This can be considered as a |G|-dimensional vector for every value of

pµ, i, jq. These vectors are orhogonal in the sense that

ÿ |G| µν
Dilµ pgq Dmj
ν
pg ´1 q “ δ δlm δij (9.81)
gPG
dR

For each irrep, µ there are d2R mutually orthogonal vectors in K|G| .
This is possible only provided that
ÿ
d2µ ď |G| (9.82)
µ

88
• The character of a given irrep is just the trace
ÿ
χµ pgq ” T r Dµ pgq ” Dij pgq (9.83)
i

The character is a class function because

tr D “ tr hDh´1 (9.84)

From our master relationship we learn that

ÿ
χµ pgqχν pg ´1 q “ |G| δ µν (9.85)
gPG

Assume the classes of G are K1 . . . KC ; that is that there are C classes

with number of elements
C
ÿ
dKi “ |G| (9.86)
i“1

Then restricting to unitaries

Dpgq` Dpgq “ 1 (9.87)

implies
χpgq “ χpg ´1 q (9.88)

ÿ
χµi χνi dKi “ dG δ µν (9.89)
i
This means that the number of irreps must be smaller or equal to the
number of classes.
We can use the orthogonality relations to decompose the adjoint re-
presentation.
First of all, assume a reducible representation

D “ D1 ‘ D2 ‘ . . . ‘ Dk (9.90)

so that ÿ
χ“ χj (9.91)
The number of times the irrep piq appears in this decomposition is
equal to
xχ|χi y (9.92)

Remember that
Dpsqet ” est (9.93)

89
Now if s ‰ 1 then st ‰ t, so that the diagonal terms in the matrix
Dpsq just vanish. Then
χps ‰ eq “ 0 (9.94)
and
χpeq “ dG (9.95)
Then
1 ÿ ad ´1
xχad |χi y ” χ pt qχi ptq “ di (9.96)
dG tPG

Ergo ÿ
d2i “ dG (9.97)
i

Assume now a central function (f pgq “ gphgh´1 q@h P G). Define a

matrix in V aasociated to an irrep, DR
ÿ
DfR ” f ptqDR ptq (9.98)
tPG

It is plain that
rDfR , Dphqs “ 0, @h P G (9.99)
Then by Schur’s lemma
DfR “ λ1 (9.100)
We can compute λ ÿ
dR λ “ f ptqχptq (9.101)
tPG

We now claim that the characters χ1 . . . χh yield an orthonormal basis

of H, the space of central functions on the group.
We need to prove that any element of H orthogonal to all the characters
is zero.
Assume
xf |χµ y “ 0 @µ (9.102)

This shows that

λ“0 (9.103)
for all irreps. Then
Dfµ “ 0 (9.104)
for all representations direct sum of irreps. Let us work this out for
the regular representation.

0 “ Dfad e1 ”
ř ř
tPG f ptqDptqe1 “ tPG f ptqet (9.105)

90
 It follows that
f ptq “ 0 @t P G (9.106)

QED.
It follows that the number of irreps is equal to the number of classes.
In conclusion, ÿ
d2R “ |G| (9.107)
R
Let us check this in the abelian group ZN ” z0 ” e, z1 . . . zN ´1
zi zj “ zi`jpmod Nq (9.108)
The irreps are given by
2πk
i
Dn pak q “ e N (9.109)
The orthogonality relationship means now that
N ´1
1 ÿ ´ 2πn1 k i 2πn2 k i
e N e N “ δn1 n2 (9.110)
N k“0
It is actually vary easy to prove that all irreps of an abelian group are
one-dimensional. Every element is a conjugacy class by itself. Then
the number of irreps is equal to the order of the group. Each of them
is got to be one-dimensional
• Let us repeat the former theorem in a different language. Given any
class function, F pgq, we can expand it as
ÿ
F pgq “ Fajk Da pgqjk (9.111)
ajk

We can actually write

1 ÿ 1 ÿÿ
F pgq “ F ph´1 ghq “ Fajk Da ph´1 ghqjk “
|G| hPG |G| hPG ajk
1 ÿÿ ÿ 1
“ Fajk Da ph´1 qjj1 Da pgqj1 j2 Da phqj2 k “ Fajk Da pgqj1 j2 δj1 j2 δjk “
|G| hPG ajk da
ÿ 1 ÿ 1
a
“ da Fajj Dkk pgq “ f a χa pgq (9.112)
ajk
da aj
da

This means that the number of irreps is actually equal to the number
of conjugacy classes. If we label conjugacy classes by α , |α| being the
number of elements of the class α, then defining the square matrix
d
|α|
Vαa ” χD pgα q (9.113)
|G| a

91
the orthogonality relation
ÿ
χ˚Da pgq χDb pgq “ |G| δab (9.114)
gPG

means that
VV` “1 6 V `V “ 1 (9.115)
To be specific
ÿ |G|
χ˚Da pgα qχDa pgβ q “ δαβ (9.116)
a |α|
Given any rep, it will containg all irreps Da some number of times,
mD
a . This can be easily computed using
ÿ
χDa pgq˚ χD pgq “ |G|mD
a (9.117)
gPG

For example, the characters of the adjoint are

χpeq “ |G|
χpg ‰ eq “ 0 (9.118)

Then
mD
a “ χa peq “ |Da | (9.119)
Each irrep appears in the adjoint a number of times equal to its
dimension.
Consider again the case of S3 . In this case |G| “ 6. We know the one
dimensional irrep
Dpgq “ 1 (9.120)

It is such that
χ0 pgq “ 1 (9.121)
Now ÿ
1` n2µ “ 6 (9.122)
µ‰0

This means that

nµ “ 1, 2 (9.123)
Let us try and determine the characters using orthogonality

µ {e} ta1 , a2 u ta3 , a4 , a5 u

0 1 1 1
(9.124)
1 1 1 -1
2 2 -1 0

92
Let us prove that given an arbitrary ( in general reducible) represen-
tation, D, the operator

da ÿ
Pa ” χ̄D pgq Dpgq (9.125)
dG gPG a

is a projector onto the subspace that transforms under the rep Da .

In fact taking the trace of the first orthogonality relation, we learn

that
da ÿ ˚ b
χ pgq Dlm pgq “ δ ab dlm (9.126)
dG gPG Da

Let us see how this works in the three dimensional rep of S3

¨ ˛
1 0 0
e Ñ Dpeq “ 0 1
˝ 0‚
0 0 1
¨ ˛
0 0 1
a1 Ñ Dpa1 q “ ˝1 0 0‚
0 1 0
¨ ˛
0 1 0
a2 Ñ Dpa2 q “ ˝0 0 1‚
1 0 0
¨ ˛
0 1 0
a3 Ñ Dpa3 q “ ˝1 0 0‚
0 0 1
¨ ˛
1 0 0
a4 Ñ Dpa4 q “ ˝0 0 1‚
0 1 0
¨ ˛
0 0 1
a5 Ñ Dpa5 q “ 0
˝ 1 0‚ (9.127)
1 0 0

With this notation it is plain that

ÿ ÿ
Dpgq|jy “ |kyxk|Dpgq|jy ” |kyDkj pgq (9.128)
k k

93
 Let us now emply this three-dimensional rep to determine the projec-
tion operators
¨ ˛
˜ j“5
¸ 1 1 1
1 ÿ 1˝
P0 “ D3 peq ` D3 paj q “ 1 1 1‚
6 3
j“1 1 1 1
˜ j“2 j“5
¸
1 ÿ ÿ
P1 “ D3 peq ` D3 paj q ´ D3 aj “ 0
6 j“1 j“3
¨ ˛
˜ j“2
¸ 2 ´1 ´1
1 ÿ 1˝
P2 “ 2D3 peq ´ D3 paj q “ ´1 2 ´1‚(9.129)
6 3
j“1 ´1 ´1 2

This makes vary explicit that

D3 “ D0 ‘ D2 (9.130)

• Let us work out the regular representation of S3

¨ ˛
1 0 0 0 0 0
˚0 1 0 0 0 0‹
˚ ‹
˚0 0 1 0 0 0‹
Dpeq “ ˚˚0 0 0 1 0
‹ (9.131)
˚ 0‹‹
˝0 0 0 0 1 0‚
0 0 0 0 0 1
¨ ˛
0 0 1 0 0 0
˚1 0 0 0 0 0‹
˚ ‹
˚0 1 0 0 0 0‹
Dpa1 q “ ˚
˚0
‹ (9.132)
˚ 0 0 0 1 0‹‹
˝0 0 0 0 0 1‚
0 0 0 1 0 0
¨ ˛
0 1 0 0 0 0
˚0 0 1 0 0 0‹
˚ ‹
˚1 0 0 0 0 0‹
Dpa2 q “ ˚
˚0
‹ (9.133)
˚ 0 0 0 0 1‹‹
˝0 0 0 1 0 0‚
0 0 0 0 1 0
¨ ˛
0 0 0 1 0 0
˚0 0 0 0 1 0‹
˚ ‹
˚0 0 0 0 0 1‹
Dpa3 q “ ˚
˚1
‹ (9.134)
˚ 0 0 0 0 0‹‹
˝0 1 0 0 0 0‚
0 0 1 0 0 0

94
 ¨ ˛
0 0 0 0 1 0
˚0 0 0 0 0 1‹
˚ ‹
˚0 0 0 1 0 0‹
Dpa4 q “ ˚
˚0
‹ (9.135)
˚ 0 1 0 0 0‹‹
˝1 0 0 0 0 0‚
0 1 0 0 0 0
¨ ˛
0 0 0 0 0 1
˚0 0 0 1 0 0‹
˚ ‹
˚0 0 0 0 1 0‹
Dpa5 q “ ˚
˚0
‹ (9.136)
˚ 1 0 0 0 0‹‹
˝0 0 1 0 0 0‚
1 0 0 0 0 0

It is the case that

¨ ˛
1 1 1 1 1 1
˚1 1 1 1 1 1‹
˚ ‹
1˚ 1 1 1 1 1 1‹
P0 “ ˚ ‹ (9.137)
6 ˚1 1 1
˚ 1 1 1‹‹
˝1 1 1 1 1 1‚
1 1 1 1 1 1
¨ ˛
1 1 1 ´1 ´1 ´1
˚1 1 1 ´1 ´1 ´1‹
˚ ‹
1˚ 1 1 1 ´1 ´1 ´1‹
P1 “ ˚ ‹ (9.138)
6˚˚´1 ´1 ´1 1 1 1‹‹
˝´1 ´1 ´1 1 1 1‚
´1 ´1 ´1 1 1 1
¨ ˛
2 ´1 ´1 0 0 0
˚´1 2 ´1 0 0 0‹
˚ ‹
1˚ ´1 ´1 2 0 0 0‹
P2 “ ˚ ‹ (9.139)
3˚˚0 0 0 2 ´1 ´1‹ ‹
˝0 0 0 ´1 2 ´1‚
0 0 0 ´1 ´1 2

9.4 Partitions and representations of Sn

• Let us recall that a cycle is a cyclic permutation of a subset. An
arbitray permutation has got kj j-cycles, where

j“n
ÿ
jkj “ n (9.140)
j“1

95
• Let us quickly revew a few general properties of cycles.
Every cycle can be written as a product of transpositions, allowing for
an index to appear several times:

pi1 i2 . . . in q “ pi1 i2 qpi2 i3 q . . . pin´1 in q (9.141)

The number of such transpositions is even for even permutations, and

odd otherwise. The canonical way of writing a permutation is as a
product of cycles without any common element. There are then n1
one-cycles (usually not written down), n2 two cycles, n3 three-cycles,
ad so on, in such a way that

n “ k1 ` 2k2 ` 3k3 ` . . . ` kn n (9.142)

We say that the set of numbers (n1 . . . nnn constitute a partition of

the number n.)
The cycle structure is invariant under conjugation. We claim that
ˆ ˙ ˆ ˙
12 . . . n s1 s2 . . . sn
pa1 . . . ap q “ pb1 . . . bp q (9.143)
s1 s2 . . . sn 12 . . . n

The reason is that for the numbers not involved in the cycle (let us
say, 3) the cycle is irrelevant in the sense that
ˆ ˙ˆ ˙
3 s3
“ ps3 q (9.144)
s3 3
so that s3 remains invariant.
For example
p123qp12qp132q “ p23q (9.145)
Also
p12qp123qp12q “ p132q (9.146)
How many elements are there in each conjugacy class? There are n!
permutations to begin with. But order is immaterial between cycles
of the same length; so we must divide by kj |. Also cyclic order does
not matter within a cycle; this yields a factor j kj . Altogether we have
n!
Nj “ ś kj k !
(9.147)
jj j

It is useful to represent conjugacy classes by Young frames; columns

of boxes of length j, top justified and arranged in decreasing j from
3!
left to right. For example, in S3 , the identity 13 (with 3! “ 1 element)
is
(9.148)

96
 3!
The class (2,1) (with 2 “ 3 elements)

(9.149)

3!
and the class 3 (with 3 “ 2 elements)

(9.150)

Altogether we recover the 6 elements of the group S3 .

It is a fact that each tableau yields an dirrep of Sn with dimension

n!
dR “ (9.151)
H
where H is the hooks factor. To be specific,

3!
d “ “1 (9.152)
3.2

3!
d “ “2 (9.153)
3

3!
d “ “1 (9.154)
3.2

• The inequivalent irreducible representations of Sn may be labelled by

the partitions of the integer n. An unlabelled Young diagram or Young
frame corresponds to a partition of the integer n, consisting of n boxes
arranged in r rows
ÿr
n“ λj (9.155)
i“1

λ1 ě λ2 ě . . . ě λr (9.156)

The usual notation is

32 1 “ t331u “
(
(9.157)

97
• A Young tableau, or labelled Young diagram is an assignment of the
numbers 1, 2 . . . n to the boxes of a Young frame. The tableau is stan-
dard if the numbers are increasing both along rows from left to right
and along columns from top to bottom.
The Young operator corrresponding to a given tableau is obtained by
first symmetrizing rows (let us call p the horizontal permutations) and
then antisymmetrizing columns (denote by q vertical permutations)
˜ ¸˜ ¸
ź ź
P “C δπ π π (9.158)
q p

It is possible to check that this is a projector, and even to compute

the constant C. We shall do it in some examples. We shall define
a mapping from a given tableau to a state in the adjoint (that is, an
element of Sn ) by defining a em lexicografic ordering: from left to right
and then top down, like reading a page in usual latin conventions. For
example
6 3 4
5 1 2
7 8
9 ÝÑ p634512789q (9.159)

• Let us now work in gory detail the case of S3 First of all, consider the
frame
(9.160)

There is only one standard tableau,

1
2
3 ÝÑ p123q (9.161)
and six others (all possible permutations). The Young operator maps
ˆ ˙
p123q ÝÑ YS ” C 1`p12q`p13q`p23q`p123q`p132q (9.162)

The projector is
1
PS ” Y0 (9.163)
6
This is a one-dimensional subspace corresponding to the trivial repre-
sentation
π ÝÑ 1 (9.164)

Consider now the frame

(9.165)

98
Again, there is only one standard tableau

1 2 3 ÝÑ p123q (9.166)

The Young operator maps

ˆ ˙
p123q ÝÑ YA ” C 1´p12q´p13q´p23q`p123q`p132q (9.167)

Again, this is a one-dimensional subspace. It correspomds to the

representation
1
PA ” YA (9.168)
6
π ÝÑ p´1qπ (9.169)
Let us now turn to the hook.

(9.170)

There are two standard tableaux. Let us write them with their opera-
tors.

1 2 ˆ ˙ˆ ˙ " *
3 ÝÑ p123q ÝÑ Y1 ” C 1 ´ p13q 1 ` p12q “ C 1 ´ p13q ` p12q ´ p123q

1 3 ˆ ˙ ˆ ˙
2 ÝÑ p132q ÝÑ Y2 ” C p1 ´ p12qq p1 ` p13qq “ C 1 ` p13q ´ p12q ´ p132q

Let us compute

Y12 “ p1 ´ p13q ` p12q ´ p123qq r1 ´ p13q ` p12q ´ p123qs “ r1 ´ p13q ` p12q ´ p123qs `
` r´p13q ` 1 ´ p123q ` p12qs ` rp12q ´ p132q ` 1 ´ p23qqs ` r´p13q ` p32q ´ p13q ` p132qs “
“ 3Y1 (9.171)

Also
Y22 “ 3Y2 (9.172)
This means that
1 1
Y1 , Y2 (9.173)
3 3
are true projectors.

Besides

99
 " *" *
P1 .P2 “ 1 ´ p13q ` p12q ´ p123q 1 ` p13q ´ p12q ´ p132q “ r1 ´ p13q ` p12q ´ p123q

` r`p13q ´ 1 ` p132q ´ p23qs ` r´p12q ` p123q ´ 1 ` p13qs ` r´p132q ` p23q ´ p13q ` 1s “

P1 .PS “ P1 .PA “ 0
P2 .PS “ P2 .PA “ 0 (9

Threre is closure, in the sense that

PS ` PA ` P1 ` P2 “ 1 (9.175)

There are also four nonstandard ones

2 3 ˆ ˙
1 ÝÑ p231q ÝÑ P3 ” r1 ´ p12qs r1 ` p23qs “ C 1 ` p23q ´ p12q ´ p123q

2 1 ˆ ˙
3 ÝÑ p213q ÝÑ P4 ” r1 ´ p23qs r1 ` p12qs “ C 1 ` p12q ´ p23q ´ p132q

3 1 ˆ ˙
2 ÝÑ p312q ÝÑ P5 ” r1 ´ p23qs r1 ` p13qs “ C 1 ` p13q ´ p23q ´ p123q

3 2 ˆ ˙
1 ÝÑ p321q ÝÑ P6 ” r1 ´ p13qs r1 ` p23qs “ C 1 ´ p13q ` p23q ´ p132q

Now by direct inspection we find that

P1 ` P2 “ P3 ` P4
P6 ` P5 “ P1 ` P2
(9.176)

In the regular representation the Young operators read

¨ ˛
1 0 ´1 1 0 ´1
˚´1
˚ 1 0 ´1 1 0‹‹
˚0 ´1 1 0 ´1 1 ‹
Y1 ” C ˚
˚1
‹ (9.177)
˚ ´1 0 1 ´1 0 ‹ ‹
˝0 1 ´1 0 1 ´1‚
´1 0 1 ´1 0 1

The structure of this matrix is

ˆ ˙
A A
Y1 “ C (9.178)
B B

100
 with
¨ ˛ ¨ ˛
1 0 ´1 1 ´1 0
A ” ˝´1 1 0 ‚ B”˝ 0 1 ´1‚ (9.179)
0 ´1 1 ´1 0 1

Eigenvectors read
¨ ˛ ¨ ˛ ¨ ˛¨ ˛
1 ´1 0 0
˚´1‹ ˚0‹ ˚ 0 ‹˚ 0 ‹
˚ ‹ ˚ ‹ ˚ ‹˚ ‹
˚0‹ ˚1‹ ˚ 0 ‹˚ 0 ‹
˚ ‹
˚0‹
˚ ‹
˚0‹
˚ ‹˚ ‹
˚ 1 ‹ ˚´1‹ (9.180)
˚ ‹ ˚ ‹ ˚ ‹˚ ‹
˝0‚ ˝´1‚ ˝ 0 ‚˝ 1 ‚
0 1 ´1 0
¨ ˛
1 ´1 0 ´1 0 1
˚0 1 ´1 1 ´1 0 ‹
˚ ‹
˚´1 0 1 0 1 ´1‹
Y2 ” C ˚
˚ ‹ (9.181)
˚´1 1 0 1 0 ´1‹‹
˝ 0 ´1 1 ´1 1 0‚
1 0 ´1 0 ´1 1
The structure is ˆ ˙
B D
Y2 “ C (9.182)
´B ´D
with B as above and
¨ ˛
´1 0 1
D ” ˝ 1 ´1 0 ‚ (9.183)
0 1 ´1

Different tableaux corresponding to the same frame yield equivalent,

although not identical representations.

• The dimension of a representation corresponding to a Young frame λ

is computed by dividing n! by the factorial of the hook length of each
box in the first column of λ and multiply by the difference of each pair
of such hook lengths. For example,

4!
dim “ p4 ´ 1q “ 3 (9.184)
4!1!
pp ` 4q! pp ` 4qpp ` 1q
dim tp`2, 2u ” ... “ pp`1q “
pp ` 3q!2! 2

(9.185)

101
102
10

Lie groups.

Lie groups are particular instances of continuous groups, where each element
g P G depends on a finite number of real continuous pareters

gpaq ” gpa1 . . . an q (10.1)

in such a way that

gpaqgpbq “ gpcq (10.2)
where the functions
ci “ fi paj , bk q (10.3)
are sufficiently regular. The canonical example is the thee-dimensional ro-
tation group, SOp3q, where the parameters are the three Euler angles. In is
often convenient to choose the parameters in such a way that

gp0q “ e (10.4)

Lie groups are n-dimensional manifolds (symmetric spaces). As is the case

for all manifolds, Lie groups can be compact or non compact. (Remember
that in Rn this means that a set is both closed and bounded). Compact
groups (again SO(3) is the simplest nontrivial example) are much simpler,
and in many ways analogous to finite groups.
The Lie magic is that many of the characteristics of a Lie group are
determined by the properties of the neighborhood of the origin (which can
be chose arbitrarily in the group manifold). The tangent space at the origin
is dubbed the Lie algebra, G. The relationship between a Lie group and its
Lie algebra is the exponential mapping
ři“n
g “ ei i“1 ai Ti
(10.5)

Indeed we can define the Lie algebra by analiticity

lim gpαi q “ 1 ` αi Ti ` . . . (10.6)

αÑ0

103
On the other hand, it is plain that
gppt ` sqαi q “ gptαi qgpsαi q (10.7)
ˇ
dˇ
Taking the derivative with respect to dt t“0
we get
d
gpsαi q “ αi Ti gpsαi q (10.8)
ds
whose solution is the matrix exponential. Please note that
eA eB ‰ eA`B (10.9)
(BCH)
Let us work out the composition law
i j i
´ eiα Ti eiβ Tj¯” ´e
iγ pα,βqTi “
¯
k
α α
1 ` iαi Ti ´
l
T T ` . . . 1 ` iβ iT ´ βk βl T T ` . . . “
2 k l i 2 k l
i j
`1 i j 1 i j i j
˘
“ 1 ` iα Ti ` iβ Tj ´ 2 α α ` 2 β β ` α β Ti Tj ` . . .
` i ˘ ` ˘` ˘ ` ˘
“ 1 ` i α ` β i Ti ` 12 ai ` β i αj ` β j Ti Tj ` 12 αi β j ´ αj β i Ti Tj ` . . . “
` ˘ ` ˘` ˘
“ 1 ` i αi ` β i Ti ` 12 ai ` β i αj ` β j Ti Tj ` 12 αi β j rTi Tj s ` . . . (10.10)
The elements Ti are a basis for the Lie algebra, which is nothing else than
a vector space with an internal composition law, the commutator.
k“n
ÿ k“n
ÿ
k
rTi , Tj s “ Cij Tk “ i fijk Tk (10.11)
k“1 k“1

In this way
γ i pα, βq “ αi ` β i ` fkl
i k l
α β ` ... (10.12)
The constants (ork
Cij fijk )
are denoted the structure constants of the algebra.
A consequence of this is that the generators are traceless
T r Ti “ 0 (10.13)
Jacobi’s identity reads
rTi , rTj , Tk s`rTk , rTi , Tj s`rTj , rTk , Ti s “ 0 ðñ Cilm Cjk
l m l
`Ckl m l
Cij `Cjl Cki “ 0
(10.14)
Let us now define the adjoint representation as
´ ¯i
Tkad ” ifkji
(10.15)
j

Let us check that this constitutes a representation. The first member is

equal to
i Tj ´ Ti Tj “ fi fj ´ fi fj “ fj fi ` fj fi “
“ ad ad ‰
Tk , Tl “ Tkj lm lj km lj km kj lm lm jk mk jl
j i j ` ad ˘i
“ ´fkl fjm “ ifkl Tj m (10.16)

104
Let us define a matrix in the algebra, G, the Killing metric as
q p
gkl ” tr Tkad Tlad ” ´fkp flq (10.17)

It can be shown that for compact groups the Killing matrix is definite
positive.
An intrinsic definition [13] is as follows. Consider an endomorphism of
L:
AdpXq : Y P L Ñ rX, Y s P L (10.18)
and the Killing form as
j
κpX, Y q ” Tr pAdpXq, AdpY qq “ X i Y l Cij
k
Clk (10.19)

If we change basis in the Lie algebra

Ti ” Mia T̃a (10.20)

the new structure constants are

w
` ˘i ` ˘j k w
C̃uv “ M ´1 u M ´1 v Cij Mk (10.21)

and the new Killing form

` ˘i ` ˘j
κ̃ab ” M ´1 a M ´1 b κij (10.22)

Then we can define

fijk ” fijl glk “ fijl flba fka

b
(10.23)

We known of course, that

k
fpijq “0 (10.24)
Let us compute

fijk ` fikj “ fijl flba fka

b ` fl fafb
ik lb ja (10.25)

But we can write

´ ¯ ´ ¯ ´ ¯
flba fijl fka
b
“ ´ fbla fijl fka
b
“ fila fjb
l
` fjla fbil fka
b
(10.26)

as well as
´ ¯ ´ ¯ ´ ¯
l a b
fik flb fja “ fbla fki
l b
fja a l
“ ´ fki fib ` fila fbk
l b
fja (10.27)

The structure constants thus are completely antisymmetric.

If is customary to define the quadratic Dynkin index as
p2q
tr Ta Tb ” IR δab (10.28)

105
106
We define an invariant subalgebra, A as a set of generators that maps into
itself under commutation with any element of the algebra G.That is, it is a
subalgebra which is also an ideal in the algebraic sense. There are always
two trivial invariant subalgebras, namely,e and G itself.
A simple Lie algebra is such that it does not have any nontrivial invariant
subalgebra. This the only type of Lie alegras we are going to study in this
course.
In this case, the adjoint representation is irreducible. (Compare with
finite groups). Assume there is an invariant subspace, generated by TA .
Call the other generators Tα . This means that

rA, Gs Ă A ÝÑ fiAα “ 0 (10.29)

Then by antisymmetry all structure constants with two indices A (in A) or

with two indices α vanish; the only possibility is to have three A or three α;
so that the algebra is not simple to begin with.
A semisimple algebra is such that there is no any abelian invariant su-
balgebra. They consist of direct products of simple algebras. The necessary
and sufficient condition for an algebra to be semisimple is that the Killing
form is non-degenerate, that is,

det gij ‰ 0 (10.30)

Let us prove the first part. Assume there is an invariant abelian subalgebra,
B generated by Tα ( the full set of generators will be denoted by i “ pα, Aq).
This means that
rB, Gs Ă B ÝÑ fiαA “ 0 (10.31)
Then the row of the Killing metric corresponding to the subalgebra, that is,

giα “ fijk fαkj “ fiβk fαkβ “ 0 (10.32)

This means that a whole row of the Killing matrix vanishes, and so does its
determinant.
Semisimple Lie algebras are direct products of simple Lie algebras, such
as
G “ G 1 ˆ G2 ˆ . . . (10.33)
where all Gi are simple.

10.1 Matrix groups

Most important are matrix groups.

• The group of n-dimensional nonsingular matrices in the field of com-

plex (real) numbers is denoted as GLn pCq (GLn pRq).

107
• The group of n-dimensional unimodular (unit determinant) matrices
in the field of complex (real) numbers is denoted as SLn pCq (SLn pRq).
• The group of n-dimensional unitary (complex matrices in the field of
complex (real) numbers is denoted as SUn pCq (unitary group).
g ` g “ gg ` “ 1 (10.34)
The unitary Lie algebra SU pnq is such that
`
e´iaT eiaT “ 1 “ 1 ` ia T ´ T ` ` Opa2 q
` ˘
(10.35)
That is, elements of the Lie algebra are hermitian matrices. How many
are those? The condition is
Tij “ Tji˚ (10.36)
2
The n diagonal elements are real; and the n 2´n complex elements
below diagonal are deternined by those above; altogether we have
(deleting the trace)
n2 ´ n
n`2 ´ 1 “ n2 ´ 1 (10.37)
2
real parameters.
• The group of n-dimensional real orthogonal matrices is denoted as
SOpnq (orthogonal group).
g T g “ gg T “ 1 (10.38)
The Lie algebra SOpnq is given by
T
eiaT eiaT “ 1 ` iapT T ` T q ` Opa2 q (10.39)
antisymmetric matrices. The number of parameters is then
npn ´ 1q
(10.40)
2
• The group of matrices the leave invariant the diagonal quadratic form
with p values of +1 and q values of -1
¨ ˛
1 0 ... 0 0
˚0 1 . . . 0 0‹
˚ ‹
Ip,q ” ˚
˚ ... ‹
‹ (10.41)
˝0 0 . . . ´1 0 ‚
0 0 . . . 0 ´1
is denoted SOpp, qq and they are non-compact as soon as either p or q
are non-vanishing. The Lorentz group SOp1, 3q belongs to this class.

108
 • The group of matrices that leave invariant the quadratic form
ˆ ˙
0 I
J” (10.42)
´I 0

where I is the n ˆ n unit matrix is the symplectic group, Spp2nq.

• The group of matruces that leave invariant the quadratic form

ds2 “ dz1 dz̄1 ` . . . dzn dz̄n (10.43)

is the unitary group U pnq „ SU pnq ˆ U p1q,

10.2 Representations of SU p2q and SOp3q through

tensor methods.
First of all, it is a fact that

M P SU p2q ùñ M
Ď P SU p2q (10.44)

Consider ua that transforms with M P SU p2q and ua that does it with

respect to M
Ď. It so happens that

Ďa M w δ v “ M M ` a “ δ a
δba Ñ M
` ˘
v b w b b (10.45)

also

ab ÝÑ ab Mva Mwb “ Mv1 Mm2 ´ M2 M1 “

v w
ˆ 1 2 2 1 1 2 2 1
˙
M1 M1 ´ M1 M1 “ 0 M1 M1 ´ M1 M2 “ det M
“ “
´det M M21 M22 ´ M22 M21 “ 0
“ det M vw “ vw (10.46)

as well as
ab ÝÑ ab (10.47)
From an upper and a lower index we can always form a simpler representa-
tion with two indices less

Tγαβ “ tα δγβ ` hβ δγα (10.48)

where

T α ” Tγαγ “ 2tα1 ` tα2

T .β ” Tγγβ “ tα1 ` 2tα2 (10.49)

Then
` α
1
˘
tα1 “ 2T ´ T .β
3
` .β
1
˘
tα2 “ 2T ´ T α
3 (10.50)

109
We can then eliminate either contravariant or covariant indices; the only
thing that matters is the difference, which we will write downstairs. Even
then, from say
Tαβ (10.51)

we can form
αβ Tαβ (10.52)

unless it is totally symmetric. Those we cannot reduce further.

We know that the combinations with repetition of n objects taken m at
a time is the number of ways of combining n ´ 1 bars and m crosses; such
that the number of crosses to the left of the first bar stands for the number
of times times the first object appears; the number of crosses between the
first and the second bar stands for the`number ˘ of times the second object
n`m´1
appears and so on. This is pn`m´1q!
m!pn´1q! “ m .
In our case n “ 2 and the irreps are generated˘ by symmetric covariant
tensors with m indices, of which there are m`1
`
m “ m ` 1.
In the case of SOp3q all irreps are real, so we need to consider covariant
indices only. In spite of that

δij ÝÑ δij Rki Rlj “ δkl (10.53)

as well as
ijk ÝÑ ijk Rli Rm
j k
Rm “ det R lmn “ lmn (10.54)

So that irreps are generated by symmetric traceless covariant tensors with

j indices. In our case n “ 3, so this yields
ˆ ˙
j`2 pj ` 2qpj ` 1q
“ (10.55)
j 2

We have to withdraw all traces, of which there are

ˆ ˙
j jpj ´ 1q
“ (10.56)
j´2 2

The difference is just the dimension of the representation,

d “ 2j ` 1 (10.57)

It is worth remarking that only when

m “ 2j (10.58)

(that is, m is an even integer) does the SU p2q irrep be also an irrep of SOp3q.

110
10.3 Representations of GLpnq through tensor me-
thods.
Weyl’s treatment of finite dimensional group representations rests on a
simple fact. Consider any tensor that under L P GLpnq transforms as

Tµ1 1 µ2 ...µn “ Lλµ11 . . . Lλµnn Tλ1 ...λn (10.59)

Assume now that the tensor T is invariant under some permutation π P Sn

Tλπp1q ...λπpnq “ Tλ1 ...λn (10.60)

Then

Tµ1 1 µ2 ...µn “ Lλµ11 . . . Lλµnn Tλ1 ...λn “ Lλµ11 . . . Lλµnn Tλπp1q ...λπpnq “ Lλµ1πp1q . . . Lλµnπpnq Tλ1 ...λn
(10.61)
It follows that
Tµ1 πp1q µπp2q ...µπpnq “ Tµ1 1 µ2 ...µn (10.62)
That is, the subspace of tensors invariant under any permutation symm-
metry is invariant under GLpnq transformations. Let us perform now some
elementary checks.
• n “ 2. There are only two symmetry classes: antisymmetric

T (10.63)
µ
ν
and symmetric
T (10.64)
µ ν
The Young projectors are
1 ˘ pµνq
(10.65)
2
• n “ 3. There are now three classes.

T
µ ν λ
T
µ
ν
λ
T (10.66)
µ ν
λ
Let us work out this third case in detail. The Young operator is given
by

Y ” P Q ” p1 ´ pµλqq p1 ` pµνqq “ 1 ` pµνq ´ pµλq ´ pµλνq

(10.67)

111
 pY T qµνλ “ Tµνλ ` Tνµλ ´ Tλνµ ´ Tνλµ (10.68)
It is clear that
Y 2 “ 3Y (10.69)
so that
1
P ” Y (10.70)
3
is a projector. Any tensor in this class is such that

2Tµνλ “ Tνµλ ´ Tλνµ ´ Tνλµ (10.71)

But then it is also a fact that

2Tµ1 ν 1 λ1 “ Lµµ1 Lνν 1 Lλλ1 pTνµλ ´ Tλνµ ´ Tνλµ q “ Tν 1 µ1 λ1 ´ Tλ1 ν 1 µ1 ´ Tν 1 λ1 µ1

(10.72)
It is a fact that in general these tensors form a basis for an irreducible
representation of GLpnq, without any further ado. Let us compute some
dimensions of those representations. The dimension of the space

α β ... δ pr slotsq (10.73)

is the numer of combinations with repetition of n objects taken in packs of

r. This can be computed as imagining (n-1) vertical lines and r crosses. The
number is ˆ ˙
pn ` r ´ 1q! n`r´1
CRrn “ “ (10.74)
r!pn ´ 1q! r
For example in the case n “ 3, r “ 3 this formula yields

D “ 10 (10.75)

To be specific, the components are

T111 T112 T113 T122 T123 T133 T222 T223 T233 T333 (10.76)

In order to count dimension for lower representations, it is useful to consider

outer products. For example

b “ ‘ (10.77)

This just expresses the trivial identity

npn ` 1q npn ´ 1q
n2 “ ` (10.78)
2 2
One can also work out slightly more complicated examples; for example

b “ ‘ (10.79)

112
The dimension of the representetion easily follows:

n2 pn ` 1q
ˆ ˙
n`2
“ `D (10.80)
2 3

and we recover that

npn2 ´ 1q
D “ (10.81)
3

10.4 Representations of U pnq

First of all, U pnq „ SU pnq ˆ U p1q. It is plain that all irreps of GL(n) are
also reps of any subgroup; but not necessarily irreps. As a trivial example,
the tensor
Trabs ” T ab (10.82)

transforms as a one-dimensional r12 s ” irrep of GLp2q:

Trijs Ñ gai gbj Trijs “ T ab det g (10.83)

It is also the case that the diagram r2n s has only one standard tableau,
1 1
for example 2 2 . There is only one basis element. Actually the Young
projector reads in this case

pP T qα1 α2 α3 α4 „ Tα1 α2 α3 α4 ` pTα2 α1 α3 α4 ` Tα1 α2 α4 α3 q ´ pTα3 α2 α1 α4 ` Tα2 α3 α1 α4 ` Tα3 α2 α4 α1 q ´

´ pTα1 α4 α3 α2 ` Tα4 α1 α3 α2 ` Tα1 α4 α2 α3 ´ Tα3 α4 α1 α2 ´ Tα4 α3 α1 α2 ´ Tα3 α4 α2 α1 q (10.84)

This result can be written as

pP T qα1 α2 α3 α4 “ T α1 α3 α2 α4 (10.85)

Under the action of GL(n)

" ´ ¯ ´
pP T qα1 α2 α3 α4 Ñ gαβ11 gαβ22 gαβ33 gαβ44 ` gαβ12 gαβ21 gαβ33 gαβ44 ` gαβ11 gαβ22 gαβ34 gαβ43 ´ gαβ13 gαβ22 gαβ31 gαβ44 ` gαβ12 gαβ23 gαβ31 gαβ44 `
¯ ´
`gαβ13 gαβ22 gαβ34 gαβ41 ´ gαβ11 gαβ24 gαβ33 gαβ42 ` gαβ14 gαβ21 gαβ33 gαβ42 ` gαβ11 gαβ24 gαβ32 gαβ43 ´ gαβ13 gαβ24 gαβ31 gαβ42 `
¯*
β1 β2 β3 β4 β1 β2 β3 β4
´gα4 gα3 gα1 gα2 ´ gα3 gα4 gα2 gα1 Tβ1 β2 β3 β4 (10.86)

This condenses into

pP T qα1 α2 α3 α4 Ñ pdet gq2 T α1 α3 α2 α4 (10.87)

113
This result is actually quite generic. If we have a rep of GL(n), say rλ1 . . . λn s,
and we add to it a column of n boxes to it, the only set of indices in standard
order that can be inserted in the additional column of rλ1 ` 1 . . . λn ` 1s is
p1, 2, . . . , nq.
Thus the number of standard tableaux of the representation rλ1 ` 1 . . . λn ` 1s
is the same as this number for rλ1 . . . λn s; the only change is a new factor
of det g. This means that thse two patters are equivalent for unimodular
groups.
Then for unimodular groups we need to consider only patters with fewer
than n rows.
rλ1 . . . λn s “ rλ1 ´ λn . . . λn´1 ´ λn s (10.88)
There is a second equivalence (related to Hodge duality), namely
“ n´1 ‰
1 “ r1s (10.89)

This can be easily generalized to

“ n´p ‰
1 “ r1p s (10.90)

The general theorem is that for unimodular transformations

rλ1 , λ2 , . . . , λn s “ rλ1 ´ λn , λ1 ´ λn´1 , . . . , λ1 ´ λ2 s (10.91)

Which is equivalent to

rµ1 , µ2 , . . . , µn´1 s “ rλ1 ” λ, λ2 ” λ ´ µn´1 , λ3 ” λ ´ µn´2 , . . . , λn´1 ” λ ´ µ2 , λn ” λ ´ µ1 s

(10.92)

10.5 Representations of Opnq

This is the only case in which the Kronecker delta with two covariant or else
two contravariant indices makes sense, because

g.g T “ 1 ðñ δ ab gai gbj “ δ ij (10.93)

Contractions commute with group transformations

δ ab gai gbj . . . Tij... “ δ ij . . . Tij... (10.94)

Traceless tensors are transformed into traceless tensors. There is a complete
decomposition of an arbitary tensor into a traceless piece plus other terms
containing Kronecker deltas. For example
ˆ ˙
1 1 1
Tij “ Tij ´ T δij ` T δij ” Tij0 ` T δij (10.95)
n n n

114
where
T ” δ ij Tij (10.96)

Another example

0
Tijk ” Tijk ` δij Ak ` δik Bj ` δjk Ci (10.97)

Let us denote the three possible traces

Ti23 ” δ jk Tijk
Tj13 ” δ ik Tijk
Tk12 ” δ ij Tijk (10.98)

This means that

Ti23 “ Ai ` Bi ` n Ci
Tj13 “ Aj ` nBj ` Bj
Tk12 “ n Ak ` Bk ` Ck (10.99)

then

1 ` 23
T ` T 13 ´ p1 ` nqT 13
˘
A“´
n2
`n´2
1 ` 23
T ´ p1 ` nqT 13 ` T 13
˘
B“´ 2
n `n´2
1
´p1 ` nqT 23 ` T 13 ` T 13
` ˘
C“´ 2 (10.100)
n `n´2

It is plain that a permutation of the indices maps a traceless tensor into

another traceless tensor. We can then apply Young operators to a traceless
tensor to obtain traceless tensors of a given symmetry type.
In fact, there is a theorem that states that the traceless tensors corre-
sponding to Young diagrams in which the sum of the lengths of the first two
columns exceeds n must vanish.

Let us work out the hook in n=2 dimensions (its traceless part
should vanish in agreement with the preceding theorem). The action of the
Young projector is proportional to
´ ¯
P Hook T ” tijk ` tjik ´ tkji ´ tjki (10.101)
ijk

115
Let us compute components in gory detail
H
T111 “0
H
T112 “ t112 ` t112 ´ t211 ´ t121
H
T121 “ t121 ` t211 ´ t121 ´ t211 “ 0
H
T122 “ t122 ` t212 ´ t221 ´ t221
H
T222 “0
H
T211 “ t211 ` t121 ´ t112 ´ t112
H
T212 “ t212 ` t122 ´ t212 ´ t122 “ 0
H
T221 “ t221 ` t221 ´ t122 ´ t212 (10.102)

Of the four non-vanishing components only two are independent because

H H
T211 “ ´T112
H H
T221 “ ´T122 (10.103)

Imposing now tracelessness,

H H H
T112 ` T222 “ T112 “0
H H H
T122 ` T111 “ T122 “0 (10.104)

there is nothing left QED.

Let us define associate diagrams. Assume the length of the first column
in T, say a ă n ď 2. Then the legth of the first colum of T 1 is a1 ” n ´ a,
and all othe columns of T and T 1 have the same length.
Fot example, for n=3

T ” Ñ T1 ” (10.105)

In n=4

T ” Ñ T1 ” (10.106)
In general the pattern T will contain a given number of indices , r “ 1, 2, . . .
and
µ 1 ` µ2 ` . . . ` µν ” r (10.107)
indices (where as usual µ1 ě µ2 ě . . . ě µν ). When n is an even number,
then
n
ν” (10.108)
2

116
 n
(in this case diagrams with 2 rows are self-conjugate) and if n is an odd
number, then
n´1
ν” (10.109)
2
In SO(n) the reps corresponding to associate diagrams T and T 1 are
equivalent.For SO(3), ν “ 1, and irreps are desibed by the diagram

... (10.110)

(symmetric traceless tensors).

We shall denore by so(n) the Lie algebra of SO(n).

117
118
11

The rotation group

SOp3q „ SU p2q{Z2.

Assume we are interested in the matrix that relates two different orthonor-
mal frames
~ea “ Ra b~eb (11.1)
This a matrix R P SOp3q. This means that
RRT “ RT R “ 1 (11.2)
Put it into another form, this is the condition that
x2 ` y 2 ` z 2 (11.3)
remains invariant under such a linear transformation.
Any rotation is always a rotation around an axis, which is the locus of
the fixed points of the rotation. Let us characterize the axis by a unit vector,
n̂.
Given any vector, ~v P R3 , it is plain that the component of it in the
direction of the axis, ~vk ” p~v .n̂q n̂ will be unaffected, whereas the orthogonal
component ~vK ” ~v ´ ~vk will become a combination of ~vK and n̂ ˆ ~v .
~vk1 “ ~vk
1 “ α~
~vK vK ` β n̂ ˆ ~vK (11.4)
The conservation of the norm implies that
α2 ` β 2 “ 1 (11.5)
Altogether
~v Ñ α ~v ` p1 ´ α q p~v .n̂q n̂ ` β n̂ ˆ ~vK (11.6)
and the rotation matrix is
Rn̂ “ α δij ` p1 ´ αq ni nj ` β ikj nk (11.7)

119
It is easy to check that this matrix is orthogonal,
ÿ
Rij Rjk “ δik (11.8)
j

Choosing

α ” cos α
β ” sin α (11.9)
¨
n21 ` pn22 ` n23 q cos α
˛
p1 ´ cos αqn1 n2 ´ n3 sin α p1 ´ cos αqn1 n3 ` n2 sin α
R “ ˝ n3 sin α ` n1 n2 p1 ´ cos αq n22 ` pn21 ` n23 q cos α p1 ´ cos αqn2 n3 ´ n1 sin α‚
´n2 sin α ` p1 ´ cos αqn3 n1 p1 ´ cos αqn3 n2 ` n1 sin α n23 ` pn22 ` n21 q cos α
(11.10)
All this yields, for n̂ ” p0, 0, 1q
¨ ˛
cos α sin α 0
R “ ˝´ sin α cos α 0‚ (11.11)
0 0 1

which when α “ π2 transforms the positive OX axis, p1, 0, 0q into the negative
OY axis, p0, ´1, 0q. The opposite sign corresponds
¨ ˛ to α Ø ´α.
0
For arbritrary n̂ transforms the vector ˝0‚ into
1
¨ ˛ ¨ ˛
0 p1 ´ cos αqn1 n3 ` n2 sin α
Rn ˝0‚ “ ˝p1 ´ cos αqn2 n3 ´ n1 sin α‚ (11.12)
1 2 2 2
n3 ` pn2 ` n1 q cos α

This corresponds to the polar direction

cos Θ “ n23 ` p1 ´ n23 q cos α

n2 n3 p1´cos αq´n1 sin α
tan Φ “ n1 n3 p1´cos αq`n2 sin α (11.13)
¨ ˛
sin θ cos φ
This depends on three parameters, as it should: two from n̂ ” ˝ sin θ sin φ ‚
cos θ
and another one from α.

cos Θ “ cos2 θ ` p1 ´ cos2 θq cos α

sin φ cos θp1´cos αq´cos φ sin α
tan Φ “ cos φ cos θp1´cos αq`sin φ sin α (11.14)

We can ask, for example, what is the rotation that transforms a given
univ vector, say n̂1 into another one, say, n̂2 . Let us denote

nˆ1 nˆ2 ” cos θ (11.15)

120
It is plain that the axis of rotation will be
n̂1 ˆ n̂2
n̂ ” (11.16)
sin θ
We need
n̂2 ´ cos θ n̂1
n̂2 “ cos α n̂1 ` sin α (11.17)
sin θ
This clearly needs α “ θ.
The groups SOp3q and SU p2q{Z2 are intimately related. Indeed any
unitary matrix can be parameterized as
ˆ ˙
cos α eiβ sin α eiγ
u“ (11.18)
´sin α e´iγ cosα e´iβ

It is clear that the range of the angles is

0 ď β ď 2π
0ďαďπ
0 ď γ ď 2π (11.19)

Consider an arbitrary hermitian matrix

ˆ ˙
1 ` z x ´ iy
M” (11.20)
x ` iy 1 ´ z

Its determinant is
det M “ 1 ´ r2 (11.21)
It is plain that the transformation

M Ñ uM u` (11.22)

leaves this determinant unchanged. Then there is a map

u P SU p2q Ñ R P SOp3q (11.23)

It is plain that both ˘u yield the same rotation; this is the reason for a
factor Z2 . To be specific, when β “ γ “ 0
ˆ ˙
` 1 ` z cos 2α ` x sin 2α ´i y ` x cos 2α ´ z sin 2α
uM u “
i y ` x cos 2α ´ z sin 2α 1 ´ z cos 2α ´ x sin 2α
(11.24)
which means that
¨ 1˛ ¨ ˛¨ ˛
x cos 2α 0 ´sin 2α x
˝y 1 ‚ “ ˝ 0 1 0 ‚˝y ‚ (11.25)
z 1 ´sin 2α 0 cos 2α z

121
It represents a rotation of angle 2α around the y axis, R2 p´2αq. This
rotation is negative, because when 2α “ π2 this yields
¨ ˛ ¨ ˛
1 0
˝0‚ Ñ ˝0‚
0 1
¨ ˛ ¨ ˛
0 ´1
˝0‚ Ñ ˝ 0 ‚ (11.26)
1 0
Also, when, α “ 0,
¨ 1˛ ¨ ˛¨ ˛
x cos 2β sin 2β 0 x
˝y 1 ‚ “ ˝´sin 2β cos 2β 0‚˝y ‚ (11.27)
z1 0 0 1 z
namely, R3 p´2βq. It is curious that when
π
α“
2
β“0 (11.28)

we recover again a rotation R3 p´2γq.

In the general case,
ˆ ˙
` 1 ` z 1 x1 ´ iy 1
uM u “ (11.29)
x1 ` iy 1 1 ´ z 1

´ ¯
1 ` z 1 ” 1 ` z cos 2α ` eipβ´γq px ´ iyq ` eipγ´βq px ` iyq sin 2α
x1 ´ iy 1 ” e2iβ px ´ iyqcos2 α ´ e2iγ px ` iyq sin2 α ´ eipβ`γq z sin 2α
x1 ` iy 1 ” e´2iβ px ` iyq cos2 α ´ e´2iγ px ´ iyqsin2 α ´ e´ipβ`γq z sin 2α
´ ¯
1 ´ z 1 ” 1 ´ z cos 2α ´ eipβ´γq px ´ iyq ` eipγ´βq px ` iyq sin 2α(11.30)

That is
¨ ` ˘
cos2 α cos 2β ´ sin2 α cos 2γ ´ cos2 α sin 2β ` sin2 α sin 2β ´ sin 2α cos pβ ` γq
˛

R “ ˝´ cos2 α sin β ` sin2 α sin 2γ cos2 α cos 2β ` sin2 α cos 2γ sin 2α sinpβ ` γq ‚
sin 2α cos pβ ´ γq sin 2α sin pβ ´ γq cos 2α
(11.31)
This means that in order to go from the unit vector along the third axis, ~e3
to an arbitrary unit vector corresponding to the polar angles pθ, φq all we
have to do is identify

´ sin 2α cos pβ ` γq ” sin θ cos φ

sin 2α sinpβ ` γq ” sin θ sin φ
cos 2α ” cos θ (11.32)

122
which can be achieved by letting
θ
α“ 2
β`γ “π´φ (11.33)
in SU p2q language
cos 2θ ´ sin 2θ e´iφ
ˆ ˙
u“ (11.34)
sin 2θ eiφ cos 2θ
Staring again at this formula, we learn that when precisely
β“0
π
γ“ (11.35)
2
we recover a rotation around the first axis, R1 p2αq
x1 “ x
y 1 “ y cos 2α ` z sin 2α
z 1 “ ´y sin 2α ` z cos 2α (11.36)
Euler showed that every rotation R P SOp3q can be written in the form
R “ R3 pψq R1 pθq R3 pφq (11.37)
The range of the Euler angles is
0 ď φ ď 2π
0ďθďπ
0 ď ψ ď 2π (11.38)
In our SU p2q language this is
˜ ψ ¸ˆ ˙ ˜ iφ ¸ ˜ φ`ψ ψ´φ
¸
ei 2 0 cos 2θ i sin 2θ e 2 0 ei 2 cos 2θ i ei 2 sin θ
2
u“ ´i ψ θ π φ “ φ´ψ φ`ψ
0 e 2 i sin 2 cos 2 0 e´i 2 i ei 2 sin 2θ e´i 2 cos θ
2

It is plain that this covers the whole group manifold, provided

0 ď φ ` ψ ď 4π
0 ď φ ´ ψ ď 4π
0ďθďπ (11.39)
Indeed
ψ`φ“β
π
ψ´φ“γ´
2
θ
α“ (11.40)
2

123
Descriptio 11.1: Euler angles

124
The relationship with Gel’fand’s notation is
ψ Ñ ´φ1
φ Ñ ´φ2
θ Ñ ´θ (11.41)
In SOp3q language this is
¨ ˛¨ ˛¨ ˛
cos ψ sin ψ 0 1 0 0 cos φ sin φ 0
R “ ˝´sin ψ cos ψ 0‚˝0 cos θ sin θ ‚˝´sin φ cos φ 0‚ “
0 0 1 0 ´sin θ cos θ 0 0 1
¨ ˛
cos ψ cos φ ´ cos θ sin φ sin ψ sin φ cos ψ ` cos θ sin ψ cosφ sin θ sin ψ
˝´cos φ sin ψ ´ cos θ cos ψ sin φ ´sin ψ sin φ ` cos θ cosψ cos φ sin θ cos ψ ‚
sin θ sin φ ´sin θ cos φ cos θ
Please note that this matriz transforms the unit vector along the third axis
to the vector ¨ ˛ ¨ ˛
0 sin θ sin ψ
˝0‚ Ñ ˝sin θ cos ψ ‚ (11.42)
1 cos θ
corresponding to the direction n “ pθ, π2 ´ ψq.

11.1 The Lie algebra SUp2q

Start with
T P SUp2q ðñ T “ T ` & tr T “ 0 (11.43)
The most general solotion is
ˆ ˙
a b
T “ (11.44)
c d
with
a “ ā
b “ c̄
d “ ´a (11.45)
That is
ˆ ˙
z x ´ iy
T “ ” xσ1 ` yσ2 ` zσ3 ” ~x~σ (11.46)
x ` iy ´z
The Puli matrices generate the simplest Clifford algebra.
tσi , σj u “ 2δij (11.47)
rσi , σj s “ 2iijk σk (11.48)
so that
σi σj “ δij ` iijk σk (11.49)

125
11.2 Highest weight representations of SUp2q
Let us review the representations of the SU(2) algebra that you are already
familiar with from quantum mechanics. We shall do it in a way which
generalizes to arbitrary groups. The algebra reads

rJi , Jj s “ iijk Jk (11.50)

Define
J1 ˘ J2
J˘ ” ? (11.51)
2
We are looking for finite dimensional unitary representations. Let us call
j the highest value of J3 .
J3 |j, αy “ j|j, αy (11.52)
First of all, just because it is a highest weight state, we can easoily determine
the value of the casimir
J 2 ” J12 ` J22 ` J32 (11.53)

J´ J` |j, jy “ 0 “ J ´ J32 ´ J3 |j, jy

` 2 ˘
(11.54)
then
J 2 |j, jy “ jpj ` 1q|j, jy (11.55)
Were there more than one highest weight state, we normalize as

xjα|jβy “ δαβ (11.56)

If we define
J1 ˘ iJ2
J˘ ” ? (11.57)
2
Then

rJ3 , J ˘ s “ ˘J ˘
rJ ` , J ´ s “ J3 (11.58)

so that
J3 J ˘ |my “ J ˘ m|my ˘ J ˘ |my “ pm ˘ 1qJ ˘ |my (11.59)
We have assumed from the beginning that there is no state with J3 “ m ` 1;
then it must be the case that @α

J ` |j, αy “ 0 (11.60)

as well as
J ´ |jαy “ Nj pαq|j ´ 1, αy (11.61)

126
Let us compute

Nj pβqNj pαqxj ´ 1, β|j ´ 1, αy “ xj, α|J ` J ´ |j, αy “ xj, α| rJ ` J ´ s |j, αy “

“ xj, α|J3 “ |j, αy “ jδαβ (11.62)

Then we learn that a

Nj pαq “ Nj “ j (11.63)
On the other hand
1 ` ´ a
J ` |j ´ 1, αy “ |J J |j, αy “ j|j, αy (11.64)
Nj

In general

J ´ |j ´ k, αy “ Nj´k |j ´ k ´ 1, αy
J ` |j ´ k ´ 1, αy “ N
sj´k |j ´ k, αy (11.65)

Actually,

Nj´k “ xj ´ k ´ 1|J´ |j ´ ky
Nj´k “ xj ´ k|J ` |j ´ k ´ 1y “ Nj´k
s ˚ (11.66)

We choose phases in such a way that

Nj´k “ N
sj´k (11.67)

|Nj´k |2 “ xj ´ k, α|J ` J ´ |j ´ k, αy “ xj ´ k, α| rJ ` J ´ s |j ´ k, αy ` xj ´ k, α|J ´ J ` |j ´ k, αy “

“ |Nj´k´1 |2 ` j ´ k (11.68)

Then we have a series of the type

kpk ` 1q
ak “ ak´1 `j ´k “ ak´2 `j ´k `j ´k `j ´pk ´1q “ . . . “ a0 ´kj ´
2
(11.69)
that is
2 kpk ` 1q k`1
Nj´k “ pk ` 1qj ´ “ p2j ´ kq (11.70)
2 2
in other words, c
pj ` mqpj ´ m ` 1q
Nm “ (11.71)
2
We are looking for finite dimensional representations. This means that
necessarily we must real some m ” j ´ l such that

J ´ |j ´ l, αy “ 0 (11.72)

127
This is only possible if there is a certain value of k “ l such that
c
p2j ´ lqpl ` 1q
0 “ Nj´l “ (11.73)
2
which means
l “ 2j (11.74)
We learn that
l
j“ (11.75)
2
where l P N, just because it counts the number of times we have applied the
operator J´ . Besides, from now on we can drop the index α.
We can summarize, in the usual notation

xj, m1 |J3 |j, my “ m δm1 ,m

b
xj, m1 |J ` |j, my “ pj`m`1qpj´m
2 δm1 ,m`1
b
xj, m1 |J ´ |j, my “ pj`mqpj´m`1q
2 δm1 ,m´1
(11.76)

11.3 Spherical Harmonics

Let us assume there is an action of G in M , that is

GˆM ÑM (11.77)

pg, xq Ñ g.x (11.78)

Then there is a representation of the group in the space of functions on M ,
FpMq ` ˘
f P FpMq Ñ pTg f q pxq ” f g ´1 x P FpMq (11.79)
It is indeed a representation, because
` ˘ ` ˘ ´ ¯
Tg pTh f q pxq “ Tg f h´1 x “ f h´1 g ´1 x “ f pghq´1 x “ pTgh f q pxqq
(11.80)
Consider now the two-sphere, M ” S2 . Let us consider an infinitesimal
(negative) rotation around the axis OZ. It must be so that

Bf
pTg f q pθ, φq ” f pθ, φ ´ αq “ f pθ, φq ´ α ` ... (11.81)
Bφ

Then
B
A3 ” ´ (11.82)
Bφ

128
Now consider the (again, negative) rotation around the axis OX.

x1 “ x
y 1 “ y cos α ` z sin α
z 1 “ ´y sin α ` z cos α (11.83)

It follows that
dφ
dx ˇ
ˇ dθ
dα α“0 “ 0 “ cos θ dα cos φ ´ sin φ dα sin θ
ˇ
dy ˇ dθ dφ
dα ˇ “ ´z “ ´ cos θ “ cos θ dα sin φ ` sin θ cos φ dα
α“0
ˇ
dz ˇ dθ
dα α“0 “ y “ sin θ cos φ “ ´ sin θ dα (11.84)

which yields immediately

dθ
dα “ ´ sin φ
dφ cos θ
dα “ sin θ cos φ (11.85)

so that
B B
A1 “ sin φ ` cot θ cos φ (11.86)
Bθ Bφ
In an analogous way we get
B B
A2 “ ´ cos φ ` cot θ sin φ (11.87)
Bθ Bφ
The hermiyian generators are

Hi ” iAi (11.88)

´ ¯
H` ” H1 ` iH2 ” iA1 ´ A2 “ eiφ B
Bθ
B
` i cot θ Bφ
´ ¯
B B
H´ ” H1 ´ iH2 ” iA1 ` A2 “ e´iφ ´ Bθ ` i cot θ Bφ
B
H3 “ iA3 “ ´i Bφ (11.89)

Let us denot the p2l ` 1q eigenfunctions corresponding to weight l by

Ylm pθ, φq m “ ´l, . . . , l (11.90)

First of all, we want that

B
H3 Ylm “ ´i Ylm “ mYlm (11.91)
Bφ
Then
1
Ylm pθ, φq “ ? eimφ Flm pθq (11.92)
2π

129
is normalized by ż
|Ylm pθ, φq| sin θ dθ dφ “ 1 (11.93)

provided żπ
sin θ dθ |Flm pθq|2 “ 1 (11.94)
0
Let us now impose that

H 2 Ylm “ lpl ` 1qYlm (11.95)

B2
ˆ ˙
2 1 B B 1
´H “ sin θ ` (11.96)
sin θ Bθ Bθ sin2 θ Bφ2
In terms of µ ” cos θ, and defining Plm pµq ” Flm pcos θq, the ODE reads
m m2
ˆ ˙ ˆ ˙
d 2 dPl pµq
p1 ´ µ q ` lpl ` 1q ´ Plm pµq “ 0 (11.97)
dµ dµ 1 ´ µ2

which defines the (normalized) associated Legendre functions

d c
pl ` mq! 2l ` 1 1 ` ˘ m l´m 2
2 ´2 d
Plm pµq ” 1 ´ µ pµ ´ 1ql (11.98)
pl ´ mq! 2 2l l! dµl´m

The functions Pl pµq ” Pl0 pµq happen to be polynomials; the Legendre

polynomial of order l.
c
2l ` 1 1 dl ` 2 ˘l
Pl pµq ” µ ´ 1 (11.99)
2 2l l! dµl

11.4 Spinor representations

No all representations of SU p2q are also representations of SOp3q, only those
with l P N qualify for that. The rest, that is, the ones such that l P 2N`1
2 are
the famous spinor representations, sometimes called somewhat confusingly,
bivalued representations of SOp3q.
First of all, for the s “ 12 representation

1 a α β
vα Ñ pσ qβ v (11.100)
2
In particular ˆ ˙ ˆ ˙
1 1 1 1
H3 ” ´ σ3 “´ (11.101)
2 0 2 0
ˆ ˙ ˆ ˙
1 0 1 0
H3 ” ´ σ 3 “ (11.102)
2 1 2 1

130
Let us denote ˆ ˙ ˆ ˙
1 0
eα : e1 ” ; e2 ” (11.103)
0 1
It is easy to find the space of functions for such representations. It is the
space of symmetric spinors with n “ 2s indices.

A “ apα1 ...αn q eα1 b . . . b eαn ai “ 1, 2. (11.104)

The SU p2q action is given by

1 1 α1 α12s
apα1 ...α2s q eα11 b . . . b eα1n ” apα1 ...α2s q τα11 b . . . b τα2s eα1 b . . . b eαn (11.105)

It is a fact that
ˆ 1 ˙ ˆ 1 ˙
´2 0 ´2 0
H3 apa1 ...an q eα1 b . . . b eαn ” apa1 ...an q 1 b ... b 1 apa1 ...an q “
0 2 0 2
p2 ´p1
“ 2 apa1 ...an q eα1 b . . . b eαn (11.106)

where p1 counts the number of times the value 1 appears amongst the set
of indices, and p2 likewise for the value 2.
We need

p1 “ l ´ m
p2 “ l ` m (11.107)

in order that

H3 apa1 ...an q eα1 b . . . b eαn “ m apa1 ...an q eα1 b . . . b eαn (11.108)

• Let us compute J~2 for s “ 12 .

~2 ~σal ~σlj 1
J ψa ” ψj “ 3ψa “ sps ` 1qψa (11.109)
2 2 4

• Let us repeat now the computation for s “ 1

σa
~ ~σb
´ ¯
~
Jψ ” 2pi ψa|jq ` 2pj ψi|b
pijq
a σb
~
σpi ~ σju
´ ¯ ´ ¯ ´ ¯ ´ u ¯
σa
~
J~2 ψ ” 2
~
Jψ ` 2pj Jψ ~ “ 2i ~σ2a ψuj ` 2 ψau `
pijq a|jq i|b
σjb σbu
´ u ¯
~
σi
` 2 2 ψub ` 2 ψiu “ 2ψij “ sps ` 1qψij (11.110)

We have used repeteadly

~σij ~σkl “ 2δil δkj ´ δij δkl (11.111)

131
• Diagonalizing σ3
pσ3 qji ζjẑ˘ ” ˘ζiz̄˘ (11.112)

An eigenstate pf J3 with eigenvalue jẑ will be given by

´ ¯s`jẑ ´ ¯s´jẑ
ψ sjẑ “ ζ ẑ` ζ ẑ´ (11.113)

We can simplify the notation as shown because all indices are totally
symmetrized anyway.
Denote
ζi ” α` ζiẑ` ` α´ ζiẑ´ (11.114)

Then ÿ s`jẑ s´jẑ

ζ i1 . . . ζ i2s ψi1 ...i2s “ α` αl xsjẑ |ψy (11.115)
jẑ

For example, when s “ 1{2

B ˇ F B ˇ F
`i ´i 1 0 1 1 ˇˇ 0 1 1 1 ˇˇ
α` ζ ψi ` α´ ζ ψi “ α` α´ ψ ` α` α´ ´ ˇψ
2 2ˇ 2 2
(11.116)

• For a general direction, n̂ ” pθ, φq

ˆ n̂,` ˙ ˆ ˙ ˜ ẑ,` ¸
ζ c ´s˚ ζ
n̂,´ “ (11.117)
ζ s c ζ k̂,´

with

θ
c ” cos 2
θ
s ” eiφ sin 2 (11.118)

Then
` ˘s`jn̂ ` n̂,´ ˘s´jn̂ ` ẑ,` ˘s`jn̂ ` ˚ ẑ,` ˘s´jn̂
ψ s,jn̂ “ ζ n̂,` ζ “ cζ ´ s ζ ẑ,´ s ζ ` c ζ ẑ,´ “
ř `s`jn̂ ˘ `s´jn̂ ˘ ` ẑ,` ˘s`jn̂ ´p ` ˘p ` ˘ s´j ´q ` ˘ q
“ pq p q cζ ´s ζ ẑ,´ s˚ ζ ẑ,` n̂
c ζ ẑ,´ ”
” jẑ Rjsn̂,ẑ pθ, φq ψ s,jẑ
ř
(11.119)

where
ˆ ˙ˆ ˙
ÿ s ` jn̂ s ´ jn̂ m`
Rjsn̂,ẑ pθ, φq ” c p´sqs`jn̂ ´m` cs´jn̂ ´m´ ps˚ qm´
m˘ ; m` `m´ “s`jẑ
m` m´
(11.120)

132
 • Let us revisit spherical harmonics from the spinor viewpoint. Let us
define as usual ˆ ˙
j j z x ´ iy
xi ” ~x ~σi “ (11.121)
x ` iy ´z
AH4 also define
ˆ ˙
´px ´ iyq z
xij ” ik xkj “ (11.122)
z x ` iy

in order to build irreps

ζ i1 ζ j1 . . . ζ is ζ js xi1 j1 . . . xis js ” pζζxqs (11.123)

Let us choose x as the unit vector with

x ` iy “ sin θeiφ
z ” cos θ
ζi “ pα` , α´ q
ζi ” ij ζj “ p´α´ , α` q (11.124)

Now it is a fact that

ˆ ˙
` ˘ ´s˚ c ` ˘ 2 2
´α´ α` ´α´ α´ “ ´s˚ α´ ´ 2cα` α´ ` sα`
c s
(11.125)

11.5 Product representations

It is possible to construct the tensor product of two irreps.

Dpgq ” D1 b D2 (11.126)

The basis of the product space is just the tensor product of the two basis

e1 b e2 (11.127)

This is trivally a representation. Its action on the natural basis is given by

Dg pe1 b e2 q “ eiαA pe1 b e2 q “ e1 b e2 ` pαAe1 q b e2 ` e1 b pαAe2 q ` . . .

(11.128)
This should be familiar from the addition of angular momentum in
quantum mechanics. It is clear that the generators of

D ” p1 ` T q b p1 ` T q (11.129)

are
p1 b T q ‘ pT b 1q (11.130)

133
In SU(2), since we work in a basis where J3 is diagonal, the values of J3 just
add.

J3 p|j1 m1 y b |j2 m2 yq “ pm1 ` m2 q3 |j1 m1 y b |j2 m2 y (11.131)

Consider, for example, the product of the three-dimensional irrep with the
two-dimensional one, 1 b 1{2. We shall analyze this tensor product by the
familiar highest weight technique.
The highest weight state is unique

|3{2, 3{2y ” |1{2, 1{2y b |1, 1y (11.132)

Now, remembering that

c
pj ` mqpj ´ m ` 1q
J ´ |j, my “ |j, m ´ 1y (11.133)
2
we get
c c
´ 3 1
J |3{2, 3{2y “ |3{2, 1{2y “ |1{2, ´1{2y b |1, 1y ` |1{2, 1{2y b |1, 0y
2 2
c c (11.134)
1 2
|3{2, 1{2y “ |1{2, ´1{2y b |1, 1y ` |1{2, 1{2y b |1, 0y (11.135)
3 3
There is an state orthogonal
c c
2 1
|ψy “ |1{2, ´1{2y b |1, 1y ´ |1{2, 1{2y b |1, 0y (11.136)
3 3
This will be later taken as the highest weight of another chain.

? b b b
J ´ |3{2, 1{2y “ 2|3{2, ´1{2y “ 13 |1{2, ´1{2y b |1, 0y ` 23 12 |1{2, ´1{2y b |1, 0y `
b b b
2 4 1
3 |1{2, 1{2y b |1, ´1y “ 3 |1{2, ´1{2y b |1, 0y ` 3 |1{2, 1{2y b |1, ´1y (11.137)

then
c c
2 1
|3{2, ´1{2y “ |1{2, ´1{2y b |1, 0y ` |1{2, 1{2y b |1, ´1y “
3 3
(11.138)

Here also there is another state orthogonal

c c
2 1
|χy ” |1{2, ´1{2y b |1, 0y ´ |1{2, 1{2y b |1, ´1y (11.139)
3 3

134
 Let us apply the operator J ´ once more
c c
2 1
J ´ |3{2, ´1{2y “ |3{2, ´3{2y “ |1{2, ´1{2yb|1, ´1y` |1{2, ´1{2yb|1, ´1y “ |1{2, ´1{2yb|1, ´1y
3 3
(11.140)
Let us now check that the state |ψ is a good candidate for a highest weight
state. For this to be true it is necessary that

J ` |ψy ” 0 (11.141)

then
b b b b
2 1 1 1
J ´ |ψy “ 3 |1{2, ´1{2y b |1, 0y ´ 3 2 |1{2, ´1{2y b 1, 0y ´ 3 |1{2, 1{2y b |1, ´1y “
b b
2 1
“ 3 |1{2, ´1{2y b 1, 0y ´ 3 |1{2, 1{2y b |1, ´1y (11.142)

that is, we identify the two orthogonal states we have obtained as

ψy ” |1{2, 1{2y
|χy ” |1{2, ´1{2y (11.143)

That is,
1 b 1{2 “ 3{2 ‘ 1{2 (11.144)

11.6 Wigner-Eckart
A tensor operator Ols l “ ´s . . . ` s) transforming under the spin-s repre-
sentation of SU(2) is a set of 2s+1 operators such that

rJa , Ols s “ Om
s
pJas qml (11.145)
In the standard basis
pJ3s qll1 “ lδll1 (11.146)
(´s ď l, l1 ď s); so that
rJ3 , Ols s “ lOls (11.147)
A trivial example is a particle in an spherically symmetric potential. Then

Ja “ La ” abc xb pc (11.148)

and ´ ¯
rJa , xb s “ ´iacb xc “ xc Jaadj (11.149)
cb
To go the canonical basis, first realize that

x0 “ x3 (11.150)

135
ans then
x1 ˘ ix2
rJ ˘ , x0 s ” x˘1 “ ¯ ? (11.151)
2
Twnsor operators have got the interesting property that
` ˘
Ja Ols |jmαy “ Ols1 |jmαy pJas ql1 l ` Ols |jm1 αy Jaj m1 m (11.152)
this is the transformation of the tensor product
sbj (11.153)
Note inprticular that
J3 Ols |jmαy “ pl ` mqOls |jmαy (11.154)
The Wigner-Eckart theorem states that
xJ, m1 , β|Ols |j, m, αy “ δm1 ,l`m xJ, l ` m|s, j, l, myxJ, β|Os |j, αy (11.155)
Let us work out an example in detail.Let us assume known the matrix
element
1 1 1 1
x , , α|x3 | , , βy ” A (11.156)
2 2 2 2
and we would like to compute x 12 , 21 , α|x1 | 21 , 12 , βy. First,
1
x1 ” ? p´x`1 ` x´1 q (11.157)
2
Starting with the highest weight state
ˇ F ˇ F
ˇ3 3 ˇ1 1
ˇ ,
ˇ 2 2 ” x`1 ˇ 2 , 2 (11.158)
ˇ

we get ˇ F c ˇ F c ˇ F
ˇ3 1 2 ´
ˇ3 3 2 ´ ˇ1 1
ˇ ,
ˇ 2 2 “ 3 J ˇ 2 , 2 “ 3 J x`1 ˇ 2 , 2 (11.159)
ˇ ˇ

But using
J ´ x`1 “ x0 ` x`1 J ´ (11.160)
ˇ F c ˇ F c ˇ F
ˇ3 1 2 ˇ1 1 1 ˇ1 1
ˇ ,
ˇ 2 2 “ 3 x0 ˇ 2 , 2 ` 3 x`1 ˇ 2 , ´ 2 (11.161)
ˇ ˇ

Finally
@1 1 3 1
D b2 @1 1 1 1
D b1 @1 1 1 1
D
0“ , | ,
2 2 2 2 “ ,
3 2 2 |x 0 | ,
2 2 ` 3 2 , 2 |x`1 | 2 , ´ 2 (11.162)

This implies that B ˇ ˇ F

1 1 ˇˇ ˇ1 1 ?
, ˇ x`1 ˇ , ´
ˇ “ ´ 2A (11.163)
2 2 2 2
and finally B ˇ ˇ F
1 1 ˇˇ ˇˇ 1 1
, x1 , ´ “A (11.164)
2 2ˇ ˇ2 2

136
12

Roots and weights

A Cartan subalgebra is a maximal abelian subalgebra; that is, a set of

commuting generators Hi “ Hi` as large as possible

rHi , Hj s “ 0 (12.1)

The dimension of the Cartan subalgebra is called the rank of the group.
In the case of SU p2q, the rank is one and the only H is precisely J3 . The
normalization in a given irrep is defined (Georgi) as

tr pHi Hj q ” kD δij (12.2)

Humphreys defines a symmetric bilinear form as

βpX, Y q ” tr pDpXq, DpY qq (12.3)

which then uses to define the dual basis of the Lie algebra L. .
In a more intrinsic way, AdL H is simultaneously diagonalizable. That is,
L is the direct sum of the subspaces

Lα ” t X P L rH, Xs “ αpHq X @H P Hu (12.4)

where α P H ˚ . It is plain that L0 is simply CL H, the centralizer of H.

The set of nonzero roots α P H ˚ is denoted by Φ. This yields the Cartan
decomposition of the Lie algebra

L “ CL H ‘ YαPΦ Lα (12.5)

It can be proved that the restriction of κ to H is nondegenerate. This allows

to identificate H with H ˚ :

φ P H˚ Ñ Tφ P H (12.6)

such that
φpHq “ κpTφ , Hq (12.7)

137
 The states of a given irrep will read

Hi |µy “ µi |µy (12.8)

The eigenvalues are dubbed weights. They obey

µi “ µ̄i (12.9)

because they are eigenvalues of a hermitian operator. The vector µi P Rm

is called a weight vector.
Let us remind ourselves of the adjoint representation. In order to define
it, consider a linear space with an state associated to every generator

Xa ÐÑ |Xa y (12.10)

with the scalar product defined as

1
xXa |Xb y ” tr Xa` Xb (12.11)
λ
in such a way that
xHi |Hj y “ δij (12.12)
It is plain that
ř ř ` adj ˘
Xa |Xb y “ c |Xc yxXc |Xa |Xb y “ c Xc Da cb ” ´ifacb |Xc “
“ ifabc |Xc y “ |ifabc Xc y “ | rXa , Xb sy (12.13)

It is plain that for the states corresponding to the Crtan generators the
weight vanishes
Hi |Hj y “ | rHi , Hj sy “ 0 (12.14)
The other states have non zero weight vectors

Hi |Eα y “ αi |Eα y (12.15)

This equivalent to
rHi , Eα s “ αi Eα (12.16)
This generators cannot be hermitian, because
“ ‰
Hi , Eα` “ ´αi Eα` (12.17)

which means that

E´α “ Eα` (12.18)
This is the generalization of the well-known elements J˘ in the SU p2q case.
Is it possible to normalize in such a way that
1
xEα |Eβ y “ tr Eα` Eβ “ δαβ (12.19)
λ

138
It is fact of life that the E˘α are lowering and raising operators for the
weights. Starting from
Hi |µy “ µi |µy (12.20)
we get

Hi E˘α |µy “ prHi , E˘α s ` E˘α Hi q |µy “ p˘αi ` µi q E˘α |µy (12.21)

In particular, the state

Eα |E´α y (12.22)
has zero weight, so that it must be a linear combination of Cartan generators.

Eα |E´α y “ βi |Hi y “ |β.Hy “ | rEα , E´α sy (12.23)

The constants βi are given by

1 1 1
βi “ xHi |Eα |E´α y “ tr pHi rEα , E´α sq “ tr pE´α rHi , Eα sq “ αi tr E´α Eα “ αi
λ λ λ
(12.24)
We conclude that
rEα , E´α s “ α.H (12.25)
It so happens that for any non-zero pair of root vectors, ˘α, there is an
SU p2q subalgebra, with generators
1
J˘ ” |α| E˘α
J3 ” α.H
|α|2
(12.26)

Indeed,
” ı
1 1 α.H
|α| Eα , |α| E´α “ α2
” ı
α.H
|α|2
, E˘α “ E˘α (12.27)

From that we can prove, for example, that root vectors correspond
to unique generators.

Demonstratio. Let us assume that there are two, Eα and Eα1 and we shall
get a contradiction. Choose adequate linear combinations in such a way
that
1 ` ˘ 1 ` ˘
xEα |Eα1 y ” tr Eα` Eα1 “ tr E´α Eα1 “ 0 (12.28)
λ λ
We now act with the J´ . This has zero weight vector, so that it is in the
Cartan subalgebra. But
1 ` “ ‰˘ 1 ` “ ‰˘ αi ` ˘
xHi |J´ |Eα1 y “ tr Hi J´ , Eα1 “ ´ tr J´ Hi , Eα1 “ ´ tr J´ Eα1 “ 0
λ λ λ
(12.29)

139
It follows that
J´ |Eα1 y “ 0 (12.30)
But this is not possible, because

J3 |Eα1 y “ |Eα1 y (12.31)

and the lowest state in a spin 1 representation vcannot hace J3 eigenvalue

`1.

More is true: If α is a root, then no non-zero multiple of α (except

´α) is also a root.

Demonstratio. It is not difficult to stablish a contradiction between the

SU(2) associated to 2α and the SU(2) associated to α.

Assume now we have a rep D with weights µi . Consider the action of

the SU(2) associated to some root α

α.H α.µ
J3 |µy ” 2
|µy “ 2 |µy (12.32)
|α| α

But we know that the J3 allowed values are either integers or half-integers.
Ergo
2α.µ
PZ (12.33)
α2
Now the state |µy can always be written as a linear combination of states
transforming according to definite reps of SU p2q. Assume the highest spin
state appearing in this linear combination is j. There must necessarily exist
an integer p such that
p
J` |µy ‰ 0 (12.34)
but
p`1
J` |µy “ 0 (12.35)
Then
α.pµ ` pαq α.µ
“ 2 `p“j (12.36)
α2 α
Likewise, there must be another integer, q such that
q
J´ |µy ‰ 0 (12.37)

but
q`1
J´ |µy “ 0 (12.38)
Then
αpµ ´ qαq α.µ
2
“ 2 ´ q “ ´j (12.39)
α α

140
It follows that
a.µ p´q
2
“´ (12.40)
α 2
One can also consider the α string through α itself. It is clear that in
this case
q´p“2 (12.41)
But we know that p “ 0, because 2α is not a root. Then q=2. This just
reexpresses the fact that 0 and ´α are also roots.
In [13] ia defined the α-string through β as the set of roots

β ´ qα, β ´ pq ´ 1qα, . . . , β, . . . , β ` pα (12.42)

and it is a fact that

βpHα q “ ´ pp ´ qq (12.43)
There is a formal identification of H ˚ with H:

α P H˚ ðñ Hα P H such that αpHq “ κ pHα , Hq @H P H (12.44)

Let us, against the famous Coleman’s advice, belabor this point.
Given a basis Hi P H and the dual basis αi in H˚

αi pHj q ” δji (12.45)

hi Hi and αpHq “ hi so that

ř
Then any H P H, H “

hi “ κkl pHαi qk hl (12.46)

and
κkl pHαi qk “ δli ñ pHαi qk “ κki (12.47)
and the formal identification is explicitly given by
ÿ ÿ
β “ βi αi ñ Hβ “ βi κki Hk ” β k Hk (12.48)

Also, a scalar product in the root space is defined through

ÿ
pα, βq ” κ pHα , Hβ q ” αi βi (12.49)
i

One immediate consequence is as follows. Defining the SU(2) algebra

with Eα
α.β 1
2
“ ´ pp ´ qq (12.50)
α 2
Defining the SU(2) algebra with Eβ yields

β.α 1
2
“ ´ pp1 ´ q 1 q (12.51)
β 2

141
The angle between both roots then is determined by
pα.βq2 pp ´ qqpp1 ´ q 1 q
cos2 θ ” “ (12.52)
α2 β 2 4
There are only four possibilities

pp ´ qqpp1 ´ q 1 q θ
0 900 “ π2
1 600 “ π3 ; 2π
3 (12.53)
2 450 “ π4 ; 3π
4
3 300 “ π6 ; 5π
6

12.1 SU(3)
Let us define the Gell-Mann matrices
¨ ˛
0 1 0
λ1 ” ˝1 0 0‚ (12.54)
0 0 0
¨ ˛
0 ´i 0
λ2 ” ˝ i 0 0 ‚ (12.55)
0 0 0
¨ ˛
1 0 0
λ3 ” ˝0 ´1 0‚ (12.56)
0 0 0
¨ ˛
0 0 1
λ4 ” ˝0 0 0‚ (12.57)
1 0 0
¨ ˛
0 0 ´i
λ5 ” ˝0 0 0 ‚ (12.58)
i 0 0
¨ ˛
0 0 0
λ6 ” ˝0 0 1‚ (12.59)
0 1 0
¨ ˛
0 0 0
λ7 ” ˝0 0 ´i‚ (12.60)
0 i 0
¨ ˛
1 0 0
1 ˝
λ8 ” ? 0 1 0‚ (12.61)
3 0 0 ´2

142
We define the SU(3) generators in such a way that
1
Ta ” λa (12.62)
2
and
1
tr Ta Tb “ δab (12.63)
2
It is clear that
tT1 , T2 , T3 u (12.64)
generate an SU(2) subgroup.
Let us choose the Cartan subalgebra to be

tH1 ” T3 ; H2 ” T8 u (12.65)

The weights in the fundamental representation are

´ ? ¯
e1 Ñ 21 , 63
´ ? ¯
e2 Ñ ´ 21 , 63
´ ? ¯
e3 Ñ 0, ´ 33 (12.66)

Weights for the vertices of an equilateral triangle of side 1 in the pH1 , H2 q

plane
The roots are differences of weights. This often the best way to compute
them.

e1 ´ e2 “ p1, 0q
?
3
e1 ´ e3 “ p 12 , 2 q
?
e2 ´ e3 “ p´ 12 , 23 q (12.67)

It is a fact that
T1?
˘iT2
E˘1,0 ” 2
T4?
E˘ 1 ,˘ ?3 ” ˘iT 2
5
2 2
T6?
˘iT7
E¯ 1 ,˘ ?3 ” 2
(12.68)
2 2

Roots form a regular hexagon in the pH1 , H2 q plane.

In an arbitrary Lie algebra (and in some basis) we will say that a given
weight µ is positive if its first non-zero component is positive, and negative
if its first non-zero component is negative. This property defines an ordering,
to wit
µ ą ν ðñ µ ´ ν ą 0 (12.69)

143
Descriptio 12.1: Roots and weights of SU p3q.

144
The highest weight in a representation is then defined in an obvious way.
In the adjoint representation, positive roots will correspond to raising
operators, and negative roots to lowering operators. The highest weight
state must be annihilated by all positive roots.
In the particular case of SU(3), positive roots are on the right half of
the cartesian pH1 , H2 q plane, and negative roots are on the left hand side of
it.
Again, in a general setting, we define simple roots as positive roots
that cannot be written as sums of other positive roots. Let us call ∆ the
set of all simple roots. It is fact of life that from the geometry of the simple
roots, it is possible to reconstruct the whole Lie algebra. Let us see how.

• If α and β are different simple roots, then α ´ β is not a root. Proof.

This is so because otherwise either

α “ β ` pα ´ βq (12.70)

or else
β “ α ` pβ ´ αq (12.71)
(depending on whether α ´ β ą 0 or β ´ α ą 0).

• This implies that

E´α |Eβ y “ E´β |Eα y “ 0 (12.72)
Then using the master formula
α.β p´q
2
“´ (12.73)
α 2
we learn that q “ 0. Also,

β.α p1 ´ q 1
“ ´ (12.74)
β2 2

implies that q 1 “ 0. This means that we know the relative length of

the roots, as well as the angle between them.

β2 p
2
“ 1 (12.75)
α p
? 1
pp
cos θα,β “ ´ (12.76)
2
• A trivial consequence is that
π
ď θα,β ď π (12.77)
2
(remember that simple roots are positive).

145
• Then all simple roots are linearly independent.

Demonstratio. Assume ÿ
Cα α “ 0 (12.78)
α

which can be rewritten as

µ ` “ µ´ (12.79)

with ÿ
µ` “ α (12.80)
Cα ą0
ÿ
µ´ “ α (12.81)
Cα ă0

But this cannot be, because

pµ` ´ µ´ q2 “ µ2` ` µ2´ ´ 2µ` .µ´ ě µ2` ` µ2´ ě 0 (12.82)

and µ` µ´ ď 0.

• Any positive root can be written as a linear combination of simple

roots with non-negative integer coefficients
ÿ
φ“ Kα α (12.83)
α

• There are exactly l (rank) simple roots.

Demonstratio. It this were not true there would be some vector ξ

orthogonal to all simple roots (and therefore orthogonal to all roots),

@φ P Φ, rξ.H, Eφ s “ 0 (12.84)

This would mean that the algebra is not simple.

• When we write ÿ
β“ Kα α (12.85)
αP∆

Call the height of a root the number

ÿ
ht β ” Kα (12.86)

If all Kα ě 0 we say that β is positive.

146
• Let us spell in detail how to build all roots out of the simple roots in
the simple case of SU(3); this then is easily generalized by induction.
The simple roots are
?
3
α1 “ p 12 , 2 q
?
α2 “ p 21 , ´ 23 q (12.87)

with

α12 “ α22 “ 1
α1 .α2 “ ´ 12
2α1 .α2 2α2 .α1
α21
“ α22
“ ´1 (12.88)

Thus p=1 for both α1 acting on |α2 y as well as for α2 acting on |α1 y.
Then
α1 ` α2 (12.89)

is a root, but neither α1 ` 2α2 nor α2 ` 2α1 are roots.

12.2 Dynkin diagrams

Remember that we found some time ago that

α.µ
α2
`p“j
α.µ
α2
´ q “ ´j (12.90)

It could be the case that |µy has lower spin components; but j is the
highest one. The value of j is determined by

p ` q “ 2j (12.91)

In case |µy is a root |µy “ |βy in the adjoint representation, the situa-
tion is simpler, because we know that each root appears only once in
the adjoint, and we conclude that

α.β
|βy “ |j; y (12.92)
α2
which is completely determined up to a phase. Let us check this in
the case of SUp3q. The root diagram is built out of

tα1 , α2 , α1 ` α2 , ´α1 , ´α2 , ´α1 ´ α2 u (12.93)

147
 To this one has to add the two null roots to get the eight fimensions
of the algebra. Besides, we know that

α12 “ α22 “ 1
α1 .α2 “ ´ 12
α1 .α2 α1 .α2
α21
“ α22
“ ´ 21 (12.94)

Also, we know how Hi commutes with everything, so that the only

thing missing is the
rHi , Eα s (12.95)

Consider the operator Eα1 `α2 . We know that p “ 1 and q “ 0, so that

p ` q “ 1 “ 2j (12.96)

We have
1
Jpα1 q` |Eα2 y ” |α1 | Eα1 |Eα2 y “ Eα1 |Eα2 y “ | rEa1 , Eα2 sy (12.97)

Under the SUp2qα1 ˇ F

ˇ1 1
|Eα2 y “ ˇ , ´
ˇ (12.98)
2 2
because
α1 .α2 1
J3α1 |Eα2 y “ 2 |Eα2 y “ ´ |Eα2 y (12.99)
α1 2
But we know that ˇ F ˇ F
α1
ˇ1 1 1 ˇˇ 1 1
J` ˇ ,´
ˇ 2 2 “ ?2 ˇ 2 , 2 (12.100)

so that we learn that

ˇ F
1 ˇˇ 1 1 1
? ˇ , “ ? η |Eα1 `α2 y (12.101)
2 2 2 2
where η is a phase, which we can choose equal to 1, as our convention. It
follows that ?
|Eα1 `α2 y “ 2 |rEα1 , Eα2 sy (12.102)
so that ?
Eα1 `α2 “ 2 rEα1 , Eα2 s (12.103)
The Jacobi identity applied to rE´α2 , rEα1 , Eα2 ss ` . . . now determines both
1
rE´α1 , Eα1 `α2 s “ ? Eα2 (12.104)
2
(which is part of the SUp2qα1 algebra, so that it was already known) as well
as

148
 1
rE´α2 , Eα1 `α2 s “ ´ ? Eα1 (12.105)
2
The phase p´1q is fully determined now.
The Dinkin diagram associetes simple roots with open circles. Pairs of
circles are connected by lines, depending on the angle between both roots:
π
• No line if the angle is 2 “ 900 ùñ |α.β| “ 0
2π
• One line if the angle is 3 “ 1200 ùñ |α.β| “ ´ 12 .|α||β|
?
3π 2
• Two lines if the angle is 4 “ 1350 ùñ |α.β| “ ´ 2 .|α||β|
?
• Three lines if the angle is 5π
6 “ 1500 ùñ |α.β| “ ´ 23 .|α||β|

In the figure we have indicated the real compact forms of the complex
Lie algebra. There are also non-compact real forms of the same complex
algebras, for example, a non-compact form of SU pnq is SLpnq.
The dynkin diagrams evidences some isomorphisms between lower rank
algebras.

SOp3q „ SU p2q „ spp2q

SOp4q „ SU p2q ˆ SU p2q
SOp6q „ SU p4q
SOp5q „ spp4q (12.106)

12.3 The exceptional algebra G2

This algebra has got two simple roots

α1 ” p0, 1q
´? ¯
α2 ” 23 , ´ 32 (12.107)

It follows that

α12 “ 1
α22 “ 3
α1 .α2 “ ´ 32
2α1 .α2
α21
“ ´3
2α1 .α2
α22
“ ´1
?
α1 .α2 3
|α1 |.|α2 | ” cos θ12 “ ´ 2
θ12 “ 1500 (12.108)

149
Descriptio 12.2: SU(3) simple roots.

150
Descriptio 12.3: Allowed angles between roots.

151
Descriptio 12.4: The Classification of simple groups.

152
The Dynkin diagram is simply two circles united by a triple line.
The α1 string through α2 has got p “ 3. The α2 string going through
α1 instead has p “ 1. This means that

φ2 ” α1 ` α2
φ3 ” 2α1 ` α2
φ4 ” 3α1 ` α2 (12.109)

are all roots.

• We known that the φ3 state is unique because α1 ` 2α2 is not a root.

• In order to check whether there is another state at level 4, we have to

check whether 2α1 ` 2α2 is a root (could it be reached by acting on
φ3 with a simple root (α2 )?)

2α2 p2α1 ` α2 q
“ ´2 ` 2 “ 0 “ ´pp ´ qq (12.110)
α22

But we already know that q “ 0 because 2α1 is not a root, so that

p “ 0 and 2α1 ` 2α2 is not a root. Another argument is that it is twice
a root, namely α1 ` α2 , and no multiple of a root can ever be a root.

• We know that 4α1 ` α2 is not a root. The remaining possibility at

level 5 is 3α1 ` 2α2 .

2α2 p3α1 ` α2 q
“ ´3 ` 2 “ ´1 (12.111)
α22

But we know that q “ 0 which means that p “ 1, so that 3α1 ` 2α2 is

a root.

• Also, 3α1 ` 3α2 is not a root, so that at level 6 we only need to check
4α1 ` 2α2
2α1 p3α1 ` 2α2 q
“6´6“0 (12.112)
α12
We know that q “ 0, so that we are done.

We have uncovered the 12+2 roots of G2 .

In general, in order to keep track of the integers pi and qi cooresponding
ř of a simple root αi on a state |φy, Aassume that the positive
to the action
root φ “ ki αi (with ki ą 0); then

2φ.αi ÿ 2αj .αi ÿ

qi ´ p i “ “ kj ” kj Aji (12.113)
αi2 j
αi2 j

153
where the Cartan matrix is defined as

2αj .αi
Aji ” (12.114)
αi2

Its diagnonal entries are all equal to 2. For SU(3) the Cartan matrix is
ˆ ˙
2 ´1
A“ (12.115)
´1 2

And for G2 ˆ ˙
2 ´1
A“ (12.116)
´3 2
Now when we go from φ to φ ` αl by the action of the raising operator Eαl ,
this changes kl to kl ` 1 so that

qi ´ p i ÝÑ qi ´ pi ` Ali (12.117)

It is now easy to work this out in gory detail in the SU(3) case. The Cartan
matrix gives the qi ´ pi , and we know the value of qi , namely qi “ 2 for the
root αi itself (because it is the J` of an SU(2)), whereas qi “ 0 for any other
root (because αi ´ αj is not a root). Let us work out A2 “ SUp3q again in
detail. Cartan’s matrix is
ˆ ˙
2 ´1
A“ (12.118)
´1 2

• Consider the α2 string through α1 (in this case q=0 because α1 ´ α2

is not a root).
A12 “ A21 “ ´1 “ q ´ p (12.119)
Then α1 ` α2 is a root; but neither α1 ` 2α2 nor α2 ` 2α1 are. We
have then three roots (plus the negatives) plus two H; these exhaust
the 8 dimensions of the algebra.

In the case of G2 , we start with

root q´p
α1 r2, ´1s (12.120)
α2 r´3, 2s

This means, for the α2 -string through α1 (q=0, because α1 ´ α2 is nort a

root), that p “ 1, so that α1 ` α2 is indeed a root, but α1 ` 2α2 is not a
root. α1 is then the highest weight of a doublet of SU p2qα2 . On the other
hand α2 is a triplet under SU p2qα1 : α2 ` 2α1 and α2 ` 3α1 are also roots

154
(nut this is not the case with α2 ` 4α1 ). Consider the α2 -string through
β ” a2 ` 3α1 . It so happens that
ˆ ˆ ˙˙
2β.α2 2 ´3
2 “ 3`3 “ ´1 “ q ´ p (12.121)
α2 3 2

Giving the fact that we know that q=0 (because 3α1 is not a root), this
means that γ ” 2α2 ` 3α1 is also a root. pβ, γq form a doublet under
SU p2qα2 .

12.4 Fundamental weights

The highest weight of a rep is such that

µ`φ (12.122)

is not a weight for any positive root φ. This is equivalkent to

2αi .µ
Eαi |µy “ 0 ô “ li ě 0 (12.123)
αi2
The integers li are the Dynkin coefficients. It is useful to introduce the
fundamental weights which are m vectors such that
2αi .µj
“ δij (12.124)
αi2
The highest weight can the be written as
ÿ
µ“ li µi (12.125)

For example, for A2 ” SUp3q, where the simple roots are

?
3
α1 “ p 21 , 2 q
?
α2 “ p 12 , ´ 23 q (12.126)

they read

?
3
µ1 “ p 12 , 6 q
?
µ2 “ p 12 , ´ 63 q (12.127)

The defining representation generated by Gell-Mann’s matrices has got µ1

as its highest weight. Its Dynkin indices are then p1, 0q. Start with

H1 |µ1 y “ 12 |µ1 y
?
3
H2 |µ1 y “ 6 |µ1 y (12.128)

155
It is clear that
E´α2 |µ1 y “ 0 (12.129)
because it is a highest weight state, and by definition, µ1 .α2 “ 0.
2µ1 .α1
“1 (12.130)
α12

This tells us that µ1 ´ α1 is a weight, but µ1 ´ 2α2 is not.

H1 E´α1 |µ1 y “ E´α1 12 |µ1 y ´ pα1 q1 E´α1 |µ1 y “ 0

? ´? ? ¯
H2 E´α1 |µ1 y “ E´α1 63 |µ1 y ´ pα1 q2 E´α1 |µ1 y “ 63 ´ 23 |µ1 y “
“ ´ ?13 |µ1 y (12.131)

This is then the weight

1
µ1 ´ α1 “ p0, ´ ? q (12.132)
3
Now
pµ1 ´ α1 q.α2
2 “1 (12.133)
α22
This tells us that µ1 ´ α1 ´ α2 must be a weight. We can represent this
procedure as follos

1 0 µ1
´1 1 µ1 ´ α 1
0 ´1 µ1 ´ α 1 ´ α 2
(12.134)

The rationale is as follows. The Cartan matrix is

ˆ ˙
2 ´1
(12.135)
´1 0

We start with the highest weight which is the top of an α1 -doublet. We

substract the first row of the Cartan matrix, and get to ´1 1 , which
must be the top of an α2 doublet. We then substract the second row of the
Cartan matric and end up into 0 ´ 1 and we are done

156
13

Representations.

Let us rewrite again the SU(3) Cartan matrix in another way (corresponding
to level 0 and level one roots). The boxed numbers represent tha value of
q´p

k“1 2 ´1 ´1 2
k“0 0 0 (13.1)

Then we start, knowing that the q-values are

q“ 2 0 0 2 (13.2)

because α1 ´ α2 is not a root, and each root is in a j “ 1 of its own SU p2q.

From that, we can go up one step in level

k“2 1 1 α1 ` α2
k“1 2 ´1 ´1 2 α1 |α2
k“0 0 0 (13.3)

We know that q “ 1 in both case, so that this is telling us that p “ 0 and

we are done with the positive roots.
To construct the µ2 irrep (Dynkin indices p0, 1q) we proceed in a similar
way, and get

0 1 µ2
1 ´1 µ2 ´ α2
´1 0 µ2 ´ α2 ´ α1 (13.4)

All states in a given irrep can easily be built out of the highest weight
state as
E´αan . . . E´αa1 |µy (13.5)

157
Descriptio 13.1: Roots of G2 .

158
Descriptio 13.2: Weights of the 3 and 3̄ of SU p3q.

159
where
αai P ∆ (13.6)
A scalar product exists in this linear space which is such that given two
~ , β~ Ă ∆
subsets of ∆, α
ˇ ˇ
xµ ˇEα~ E´β~ ˇ µy „ δα~ ,β~ (13.7)
ˇ ˇ

The explicit computation of an orthonormal basis can become easily painful

for large irreps.

13.1 The Weyl group

This is the set of all Weyl reflections. They stem for the fact that the SU(2)
irreps are symmetrical under

J3 Ñ ´J3 (13.8)

Remember that
2α.µ
q´p“ (13.9)
α2
so that
ˇ F
α.µ α.µ α.µ ˇˇ 2α.µ
J3 |µy “ |µy ÝÑ |µ ´ pq ´ pqαy “ ´ |µ ´ pq ´ pqαy ” ´ µ ´ α
α2 α2 α2 ˇ α2

In slightly more formal terms we are multiplying the weigh by the idempo-
tent
2αi αj
pI α qji ” δij ´ (13.10)
α2
It is easy to show that
Iα2 “ 1 (13.11)
I α .α “ ´α (13.12)
In general, we can decompose any vector with respect to the direction of α

v “ vK ` vk (13.13)

then
I α .v “ vK ´ vk (13.14)
In the particular case of the 3̄ ” p0, 1q of SU(3), all weights are just the
negative of the weights of the 3 ” p1, 0q. This means that the two irreps are
related by complex conjugation.

rDa , Db s “ ifabc Dc ùñ rDa , Db s˚ “ ´ifabc Dc˚ ùñ r´Da˚ , ´Db˚ s˚ “ ifabc p´Dc˚ q

(13.15)

160
This irrep is usually dubbed D̄. The irrep D is said to be real if it is
equivalent to its complex conjugate. Otherwise, it is said to be complex.
Given the fact that Hi` “ Hi if µ is a weight in D, then ´µ is a weight
in D̄. Then the lowest weight of p1, oq is minus the highest weight of p0, 1q
and the other way around.
The highest weight of pn, mq is nµ1 ` mµ2 , and the lowest weight of
pn, mq is ´nµ2 ´ mµ1 , so that that highest weight of pm, nq is nµ2 ` mµ1 .
The irreps pn, mq ans pm, nq are complex conjugates.

Let us work out the p2, 0q irrep of SU(3). Remember the SU p3q Cartan
matrix ˆ ˙
2 ´1
(13.16)
´1 2
Then the string of weights looks as follows

2 0 2µ1
0 1 2µ1 ´ α1
´1 2 2µ1 ´ 2α1 1 ´1 2µ1 ´ α1 ´ α2

(13.17)

Let us now look at the Weyl reflections

2α.µ1
J3α1 |2µ1 y “ |2µ1 y “ |2µ1 y (13.18)
α2
Let us begin with the Weyl reflections of µ.

I α1 pµ ” 2µ1 q “ 2µ1 ´ 2α1

I α2 pµ ” 2µ1 q “ µ
I α2 p2µ1 ´ 2α1 q “ 2µ1 ´ 2α1 ´ 2α2
I α1 p2µ1 ´ 2α1 ´ 2α2 q “ 2µ1 ´ 2α1 ´ 2α2 (13.19)

Let is now examine the Weyl reflections of µ ´ α1 ” 2µ1 ´ α1 . First of all,

I α1 leaves this weight invariant, because it is orthoginal to α1 . Otherwise

I α2 p2µ1 ´ α1 q “ 2µ1 ´ α1 ´ α2
I α1 p2µ1 ´ α1 ´ α2 q “ 2µ1 ´ 2α1 ´ α2
I α2 p2µ1 ´ 2α1 ´ α2 q “ 2µ1 ´ 2α1 ´ α2 (13.20)

Altogether, this is a six-dimensional irrep

p2, 0q “ 6 (13.21)

161
Consider now the irrep p1, 1q. It so happens that
µ1 ` µ1 “ α 1 ` α 2 (13.22)
which is the highest weight of the adjoint of SU(3), already studied. We
know thet the zero weight is doubly degenerate. Let us check now that the
two ways of getting zero weight ara actually linearly independent.
|01 y ” E´α1 E´α2 |µ1 ` µ2 y
|02 y ” E´α2 E´α1 |µ1 ` µ2 y (13.23)
Our task is to show that
x01 |02 y2 ‰ x01 |01 yx02 |02 y (13.24)
Demonstratio. This is easy, because
x01 |01 y ” xµ |Eα2 Eα1 E´α1 E´α2 | µy “ xµ |Eα2 pE´α1 Eα1 ` α1 .Hq E´α2 | µy “
ˇ ` ˘ˇ
“ xµ ˇEα αi E´α Hi ´ αi E´α ˇ µy “ µ.α1 ´ α1 .α2 “ 1 ` 1 “ 1 (13.25)
2 1 2 2 2 2 2

x02 |02 y ” xµ |Eα1 Eα2 E´α2 E´α1 | µy “ xµ |Eα1 pE´α2 Eα2 ` α2 .Hq E´α1 | µy “
ˇ ` ˘ˇ
“ xµ ˇEα αi E´α Hi ´ αi E´α ˇ µy “ µ.α2 ´ α2 .α1 “ 1 ` 1 “ 1 (13.26)
1 2 1 1 1 2 2

x01 |02 y ” xµ |Eα2 Eα1 E´α2 E´α1 | µy “ xµ |Eα2 E´α2 Eα1 E´α1 | µy “
“ xµ |α2 .Hα1 .H| µy “ pα2 .µq pα1 .µq “ 21 . 12 “ 1
4 (13.27)

The p3, 0q, with highest weight µ “ 3µ1 . It follows that the string of weights
reads
3 0 3µ1
1 1 3µ1 ´ α1
´1 2 3µ1 ´ α1 ´ α2 2 ´1 3µ1 ´ 2α1
0 0 3µ1 ´ 2α1 ´ α2 ´3 3 3µ1 ´ 3α1
1 ´2 3µ1 ´ 2α1 ´ 2α2 ´2 1 3µ1 ´ 3α1 ´ α2
´1 1 3µ1 ´ 3α1 ´ 2α2
0 ´3 3µ1 ´ 2α1 ´ 3α2 (13.28)
All states are obviously unique except the 0 0 . But this is also unique
because as you can undoubtly prove
E´α1 E´α2 E´α1 |3µ1 y „ E´α2 E´α1 E´α1 |3µ1 y (13.29)
So that p3, 0q “ 10. Its complex conjugate 1Ď0 “ p0, 3q

162
14

The unitary groups

AN ´1 “ SU pN q

Our normalization will be

1
tr pTa Tb q “ δab (14.1)
2
The generators of the Cartan subalgebra in the fundamental irrep N are
given by (A “ 1, . . . , N ´ 1; i, j, k “ 1 . . . N )
˜ ¸
A
` A˘ 1 ÿ
H ij “ a δik δjk ´ aδi,A`1 δj,A`1 (14.2)
2ApA ` 1q k“1

The N weights (as many as the dimension of the fundamental) are pN ´

1q-dimensional vectors, which are the eigenvalues of the H in the Cartan
subalgebra ˜ ¸
b“A
a 1 ÿ
pµ qA ” a δab ´ Aδa,A`1 (14.3)
2ApA ` 1q b“1
For example
´ ¯
?1 , ?1 , ?1 , ..., ? 1
µ1 “ 4 12 24 2N pN ´1q (14.4)
´ 1 1 1 1
¯
µ2 “ ´ ?4 , ?12 , ?24 . . . , ?2N pN ´1q (14.5)
´ 2 1 1
¯
µ3 “ 0, ´ ?12 , ?24 , . . . , ?2N pN ´1q (14.6)
´ ¯
4 0, 0, ´2 ?1 , . . . , ? 1
µ “ 24 2N pN ´1q (14.7)

... (14.8)
´ 1
¯
µN “ 0, 0, 0, . . . , 0, ? (14.9)
2N pN ´1q

163
We can compute the weight length
Nÿ
´1 N ´1 ˆ ˙
1 1 1 1 ÿ 1 1
µ .µ “ “ ´ (14.10)
A“1
2ApA ` 1q 2 A“1 A A ` 1

Let us dub
Nÿ
´1 Nÿ
´1
1 1 1
f pN q ” “1` ´ (14.11)
A“1
A A“1
A`1 N
Then
N ´1
µ1 .µ1 “ (14.12)
2N
and in fact this results holds for all other weights.
For a ă b, for example,
` ˘2
µ1 .µ2 “ µ1 ´ 21 “ ´ 2N 1
(14.13)

Again, this results turns out to be generic. We can then write

1 1
µa .µb “ ´ ` δab (14.14)
2N 2
We shall adopt here a backwards convention: a positive weight is one
such that the last non-zero component is positive. Then

µ1 ą µ2 ą . . . ą µN (14.15)

The roots are differences of weights

µa ´ µb pa ‰ bq (14.16)

Positive roots are

µa ´ µb pa ă bq (14.17)
The simple roots are
αA ” µA ´ µA`1 (14.18)
It so happens that
1
` 1 ˘ ` 1 ˘ ` 1 ˘
αa .αb “ ´ 2N ` 21 δab ´ ´ 2N ` 12 δa,b`1 ´ ´ 2N ` 12 δa`1,b ` ´ 2N ` 12 δa`1,b`1 “
“ 12 δab ´ 12 δa,b`1 ´ 12 δa`1,b ` 12 δa`1,b`1 “ δab ´ 12 δa,b`1 ´ 12 δa`1,b (14.19)

This explains the shape of the Dynkin diagram, the simplest of them all. It
is then plain that the fundamental weights are given by
a“A
ÿ
MA ” µa (14.20)
a“1

164
Indeed

2αB M A “ 2 a“A
ř a
` B ˘
B`1 “
a“1 µ µ ´ µ
“ a“A
ř
a“1 pδaB ´ δa,B`1 q “ δAB (14.21)

Oeing to the fact that the Cartan generators are traceless,

a“N
ÿ
µa “ 0 (14.22)
a“1

Then
a“N
ÿ´1
µN “ ´ µa “ ´µN ´1 (14.23)
a“1

Then
p1, 0 . . . 0q “ p0, . . . , 1q (14.24)
and so on.

165
166
15

Orthogonal algebras

The Dynkin diagrams of SOp2nq and SOp2n ` 1q are different, and this re-
flects some important differences between the two sets of orthogonoal groups.
Let us first examine the structure of both algebras.

15.1 Dn “ SOp2nq
The Lie algebra consists on imaginary antisymmetric matrices of dimesion
2n- There are np2n ´ 1q of those. The Cartan generators in the fundamental
representation can be chosen as
a
Hjk ” ´i pδj,2m´1 δk,2m ´ δk,2m´1 δj,2m q (15.1)

(a “ 1 . . . n, the rank of Dn ) For example, for D2 in block form

ˆ ˙
1 σ2 0
H “
0 0
ˆ ˙
2 0 0
H “ (15.2)
0 σ2
The corresponding eigenvectors are

˘ek ” δj,2k´1 ˘ iδj,2k (15.3)

For example, ¨ ˛
1
˚˘i‹
˘e1 ” ˚
˝0‚
‹ (15.4)
0
¨ ˛
0
2
˚0‹
˘e “ ˚˝1‚
‹ (15.5)
˘i

167
In general
Ha p˘eb q “ δab p˘eb q (15.6)
This means that the weight vectors are given by

pa qb ” δba (15.7)

Roots are given by

˘a ˘ b pa ‰ bq (15.8)
There are npn ´ 1q of those (“ np2n ´ 1q ´ n). The positive roots are given
by
a ˘ b pa ă bq (15.9)
Finally, the simple roots are given by

a ´ a`1 a “ 1...n ´ 1
n´1 ` n (15.10)

It is plain that
` ˘2
a ´ a`1 “ 2
` a ˘` ˘
 ´ a`1 a`1 ´ a`2 “ ´1
cos θ “ ´ 14 (15.11)

On the other hand the two last simple roots are orthogonal
` n´1 ˘` ˘
 ` n n´1 ´ n “ 0 (15.12)

15.2 SOp2n ` 1q ” Bn
This algebra has an extra one-dimensional subspace associated with a zero
weight. The dimension of the algebra is np2n ` 1q. The Cartan subalgebra
is the same, with one extra row and column. For example
ˆ ˙
σ2 0
H“ (15.13)
0 0

There are extra roots connecting the extra dimensional subspace with the
others:

˘a ˘ b
˘a (15.14)

168
Altogether, we have 2n extra roots, which is the difference between the
dimensions of Bn and Dn . The positive roots are just

a ˘ b pa ă bq
a (15.15)

The simple roots

a ´ a`1 pa “ 1 . . . n ´ 1q
n (15.16)

What happens is that n´1 ` n is not simple anymore, because it is pn´1 ´

n q ` 2n . This changes the angle between the two last roots

pn´1 ´ n q.n “ ´1
cos θ “ ´ 12 (15.17)

The fundamental weights are

i“a
ÿ
Ma ” i a “ 1...n ´ 1 (15.18)
i“1

i“n
1ÿ i
Mn ”  (15.19)
2 i“1
Indeed
ři“a ` ˘
2M a αb “ i b
i“1   ´ 
b`1 “ δ
ab
1 ři“n i a
n a
` a`1
˘
2M α “ 2 j“1   ´  “0
n n
ři“n i n
2M α “ j“1  . “ 1 (15.20)

Weyl reflexions of M N on all roots a yields the set of weights

a 1ÿ c 1ÿ c 1` 1
I  M N “ p1 ´ 2a b a q . p ´ 2a q Ñ ˘ ˘ 2 ˘ . . . ˘ n
˘
 “
2 c 2 c 2
(15.21)
n
This is a 2 dimensional representation, the spinor representation. We shall
work in a rep space which the n-th tensor product of the two-dimensional
space S of the spin 1{2 irrep.

S b . . .pn b S (15.22)

Then the Cartan subalgebra is given by one 12 σ3 in the j-th position

1 1
Hj ” 1 b . . . b σ3 b . . . b 1 ” σ3j (15.23)
2 2

169
Then it can be shown that

E˘1 “ 12 σ˘
1

E˘2 “ 12 σ31 σ˘
2

...
1 1 j´1 j
E˘j “ 2 σ3 . . . σ3 σ˘ (15.24)

To summarize,

M2j´1,2n`1 “ 12 σ31 . . . σ3j´1 σ1j

M2j,2n`1 “ 12 σ31 . . . σ3j´1 σ2j (15.25)

and then all other generators are determined by the algebra

Mab ” ´i rMa,2n´1 , Mb,2n´1 s (15.26)

In the case of Dn`1 ” SOp2n ` 2q the roots are

αj “ j ´ j`1
αn`1 ” n ` n`1 (15.27)

There are two special representations corresponding to the last two

fundamental weights. Let us call them Dn and Dn`1

1
` ˘
µn ” 2 1 ` . . . ` n ´ n`1
` ˘
µn`1 ” 12 1 ` . . . ` n ` n`1 (15.28)

Under the SOp2n ` 1q subgroup generated by

Mjk j, k ď 2n ` 1 (15.29)

both representations transform like the spinor representation. It can be

shown that in Dn the extra generator in the Cartan subalgebra reads

1
Hn`1 ” M2n`1,2n`2 “ ´ σ31 . . . σ3n (15.30)
2

and in the other representation Dn`1

1
Hn`1 ” M2n`1,2n`2 “ σ31 . . . σ3n (15.31)
2

170
15.3 Clifford algebras
The simplest definition of a Clifford algebra is through the relations

tγi , γj u “ 2δij i, j “ 1, . . . , N (15.32)

Given a representation of the Clifford algebra, there is an associated repre-

sentation of SOpN q given by

i
Mij ” ´ rγi , γj s (15.33)
4

The gamma matrices themselves transform with the fundamental D1 ” N

of SOpN q

rMjk , γl s “ i pδjl γk ´ δkl γj q (15.34)

For Bn ” SOp2n ` 1q there is an explicit representation of the Clifford
algenbra that yields precisely the spinor representation of Bn .

γ1 ” σ21 σ32 . . . σ3n

γ2 ” ´σ11 σ32 . . . σ3n
γ3 ” σ22 σ33 . . . σ3n
γ4 ” ´σ12 σ33 . . . σ3n
...
γ2n´1 ” σ2n
σ2n ” ´σ1n
γ2n`1 ” σ31 σ32 . . . σ3n (15.35)

It is fact of life that

γ1 γ2 . . . γ2n`1 “ in (15.36)
We do jot have enough elements to construct a representation of SOp2n`2q;
but we can construct the SOp2nq algebra just by leaving out γ2n`1 . This is
a reducible representation; there is a nontrivial matrix that commutes with
all the generator, namely γ2n`1 itself. There are two projectors. One onto
Dn´1
1
p1 ´ γ2n`1 q (15.37)
2
and another onto Dn

1
p1 ` γ2n`1 q (15.38)
2

171
There is a natural SU pN q subgroup of SOp2N q. In fact from a Clifford
algebra one can construcy the operators
1
aj ”2 pγ2j´1 ´ iγ2j q
aj ” 12 pγ2j´1 ` iγ2j q
`
(15.39)

They obey
! )
taj , ak u “ a` j , a`
k “0
(
aj , a`
k “ δjk (15.40)

Then out of the matrix elements in the N of SU pN q

ÿ
Ta ” a`
i pTa qij aj (15.41)
ij

In order to show that this is in fact a subalgebra of SOp2N q, let us write

1
( 1“ ` ‰ 1 i 1
a` `
i aj “ 2 ai , aj ` 2 ai , aj “ 2 δij ` 2 M2i´1,2j´1 ` 2 M2i´1,2j ´
´ 12 M2i,2j´1 ` 2i M2i,2j (15.42)

The Fock states generate the representation DN for N even, and DN ´1 for
Nodd. There is an SOp2N q generator which commutes with the SU pN q
subgroup, namely
N N
ÿ ÿ N
S” M2j´1,2j “ a`
i ai ´ (15.43)
j“1 i“1
2

This geberates a U p1q algebra. Actually

N
1ÿ i
S“ σ (15.44)
2 i“1 3

in such a way that

N
S |0y “ ´ |0y (15.45)
2

172
16

Automorphisms

We shall dub inner such automorphisms that are equivalent to a conjugation

Ta Ñ RTa R´1 (16.1)

where
aT
R ” eiθ a
(16.2)

All other automorphisms are called outer. Complex conjugation acts as

Ta Ñ ´Ta˚ (16.3)

This means that an algebra can hace complex representations only it it

enjoys nontrivial automorphisms. Sometimes this is trivial, like in the SU p4q
exchanging
Eα1 Ø Eα3 (16.4)

which exchanges the representation D1 with the D3 ” D1 which are non-

equivalent. In fact all complex conjugations automorphisms can be obtained
from reflexion symmetries of the Dynkin diagram. The opposite is not true:
not all reflecion symmetries correspond to complex conjugation. The cano-
nical example is SOp8q. Nontrivial automorphisms allow to classify all real
forms of complex Lie algebras. Let us see how thos work for the complex al-
gebra A1 . In order to do that it is better to forget about physicist’s notation
and wrote
g ” eα.T (16.5)

This algebra is generated by

173
 ˆ ˙
1 0
T3 ”
0 ´1
ˆ ˙
0 1
T1 ”
0 0
ˆ ˙
0 0
T2 ”
1 0
ˆ ˙
i 0
I3 ”
0 ´i
ˆ ˙
0 i
I1 ”
0 0
ˆ ˙
0 0
I2 ”
i 0
(16.6)

Restriction to real matrices leaves the algebra of SLp2, Rq. The algebra is

rT1 ” T` , T2 ” T´ s “ T3
rT3 , T` s “ 2T`
rT3 , T´ s “ 2T´ (16.7)

This is exactly what we have been advocating for SU p2q. But were we to
stick to real generators the algebra would really have been

Ji ” iHi (16.8)

rJi , Jj s “ iijk Jk Ñ rHi , Hj s “ ijk Hk (16.9)

Defining
H˘ ” H1 ˘ H2 (16.10)

rH3 , H˘ s “ ¯H˘ sH` , H´ s “ ´2H3 (16.11)

This algebra is almost the same as SL(2,R). They differ only in

T3 Ñ ´H3 (16.12)

This can be interpresed as dua to the existence of an involutive automosphi-

sm in SL(2,R)

φ pT˘ q “ ´T˘
φpT3 q “ T3 (16.13)

174
Now Weyl’s unitary trick intruct to consider the algebra

T˘ Ñ iH˘ (16.14)

and this is the real compact form of the complex Lie algebra. In this case,
SUp2q Ă SLp2, Cq. A complex Lie algebra includes menay real forms in
general, although only one of them is compact.

175
176
Acknowledgments

We have been partially supported by the European Union FP7 ITN INVI-
SIBLES (Marie Curie Actions, PITN- GA-2011- 289442)and (HPRN-CT-
200-00148) as well as by FPA2012-31880 (Spain), FPA2014-54154-P, COST
action MP1405 (Quantum Structure of Spacetime) and S2009ESP-1473 (CA
Madrid). This work is supported by the Spanish Research Agency (Agencia
Estatal de Investigaciń) through the grant IFT Centro de Excelencia Severo
Ochoa SEV-2016-0597.

177
178
Conspectus librorum

[1] L. Schwartz, Methodes mathematiques pour les sciences physiques

(Hermann)
[2] Dirk Struik, Lectures in classical differential geometry (Dover,1961)
[3] H. Flanders, Differential forms (Academic press,1963)
[4] Dubrovin, Novikok y Fomenko,
Geometria contemporánea (MIR, Moscu, 1979)
[5] M.J. Lighthill, An introduction to Fourier analysis and generalized
functions (Cambridge,1958)
[6] IM Gelfand and GE Shilov, Generalized Functions, Vol 1 (Academic
Press)
[7] P. Jacobson, Lie algebras (Dover)
[8] Hermann Weyl, The classical groups (Princeton)
[9] H. Borner Representations of groups (North Holland 1970).
[10] A. O. Barut and R. Raczka, “Theory Of Group Representations And
Applications,” Singapore, Singapore: World Scientific ( 1986) 717p
[11] M. Hamermesh, Group theory (Addison Wesley, 1964)
[12] JP. Serre, Representations lineaires des groupes finis (Hermann,Paris,
1998)
[13] James E. Humphreys, Introduction to Lie algebras and representation
theory (Springer-Verlag,1980)
[14] H. Georgi, “Lie Algebras In Particle Physics. From Isospin To Unified
Theories,” Front. Phys. 54, 1 (1982).
[15] S. Sternberg, Curvature in mathematics and physics (Dover,2012)
Group theory and physics (Cambridge, 1995)
[16] Pierre Ramond, Group Theory (Cambridge,2010)

179