November 12, 2011

Group Theory for Physicists

Kasper Peeters

These notes are incomplete, unfinished and most likely contain er-
rors. They are solely meant as a guide to complement your notes
taken during the lectures and cover a lot of the material discussed
there, but certainly not everything.

Lecture notes for the “Group Theory” module of the M.Sc. in Particles, Strings and
Cosmology at Durham University.
Copyright
c 2011 Kasper Peeters

Department of Mathematical Sciences

University of Durham
South Road
Durham DH1 3LE
United Kingdom

kasper.peeters@durham.ac.uk
1 Introduction 5
1.1 Why group theory? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 A quick tour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Definitions and basic concepts . . . . . . . . . . . . . . . . . . 5
1.2.2 Observing symmetries in quantum mechanics . . . . . . . . . 8
1.2.3 Abstract groups/algebras versus representations . . . . . . . 8
1.3 Literature and software . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Lie algebra classification 11

2.1 Main ingredients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 The adjoint representation . . . . . . . . . . . . . . . . . . . . . 11
2.1.2 Representation theory of sl (2) . . . . . . . . . . . . . . . . . . 11
2.1.3 Preview: extending sl (2) . . . . . . . . . . . . . . . . . . . . . . 13
2.1.4 Real subalgebras . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.5 Simple, semi-simple, ideals . . . . . . . . . . . . . . . . . . . . 14
2.2 Cartan classification and Dynkin diagrams . . . . . . . . . . . . . . . 14
2.2.1 Root systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Rank-two examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 Representation theory 19
3.1 Weight space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.1 Weight vectors versus roots . . . . . . . . . . . . . . . . . . . . 19
3.1.2 Building weight space . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Example: the eightfold way . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Assorted topics 21
4.1 Young tableaux for sl (n) . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Bianchi identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3 Branching rules and grand unified theories . . . . . . . . . . . . . . . 24

3
4
 Introduction
1
1.1. Why group theory?
Group theory is, in short, the mathematics of symmetries. You already know that
symmetries can be very important in understanding or simplifying physics prob-
lems. When you study classical mechanics, you learn that symmetries of a system
are intimately related to the existence of conserved charges. Their existence often
makes solving for the dynamics a lot simpler. Even if a symmetry is not present
exactly (for instance, when a system is almost-but-not-quite spherically symmet-
ric), we can often describe the system as a small perturbation of a system that does
exhibit symmetry. A surprisingly large number of physics problems is built around
that idea; in fact, practically all systems for which we can solve the dynamics exactly
exhibit some sort of symmetry that allow us to reduce the often horrible second-
order equations of motion to much simpler first-order conservation equations. Dynamics (governed by 2nd order
Symmetries have three properties which are key in understanding their math- equations) simplifies tremendously
in the presence of conserved
ematical structure: they are associative, there is an identity element, and for ev- charges (involving only 1st order
ery symmetry transformation there is an inverse transformation that cancels it and derivatives, typically).
brings us back to the original system. These three properties are what defines a
“group”, and the theory of these mathematical structures is “group theory”. The
goal of this module is then, simply put, to show you which types of symmetries there
are (the “classification” of groups) and how they can be made to work in concrete physical
systems (how their “representation” on physical systems works).

1.2. A quick tour

1.2.1 Definitions and basic concepts
Translating the basic properties of a symmetry transformation, mentioned above,
into a mathematical language leads us to the definition of a group G: Properties of a group

1. For two elements x, y ∈ G the composition x · y ∈ G, too (a property called

“closure”).
2. There exists an identity element, denoted e, such that e · x = x · e = x for every
x ∈ G.
3. For every x ∈ G there exists an inverse x −1 such that x · x −1 = x −1 · x = e.
4. The composition is associative, that is, ( x · y) · z = x · (y · z) for any x, y, z ∈ G.
This sounds like a bunch of trivialities, but there certainly are counter-examples. For
instance, the set R of real numbers with multiplication as the composition operation

5
 1.2 A quick tour

does not form a group: the element 0 does not have an inverse.
There are two basic classifications of groups which we can already discuss. First,
note that generically, a group does not need to be commutative, that is, x · y 6= y · x
in general (matrix multiplication for instance, typically does not satisfy this, except
for special matrices). If commutativity does hold for all elements, then we have a
so-called Abelian group.
The other classification is based on whether we have a discrete number of ele-
ments in the group, or a continuum of them. Discrete groups, for instance, include
the permutation group Sn of n elements. Continuous groups include the rotation
group SO(3) of points in three dimensions. Groups which depend continuously on
one or more parameters are also called Lie groups. In this module, we will mostly be
interested in the latter. The classification of discrete groups follows quite a different
set of steps than the classification of Lie groups, for reasons that will become clear
shortly.
The most well known Lie groups are what are known as the classical matrix
groups. These can, as the name suggests, be defined in terms of generators which
are given by matrices, with some additional constraints which occur often in classi-
List of classical groups cal or quantum mechanical properties. The full list is:

GL( N, R)
The general linear group in N dimensions.
Given by all invertible N × N matrices with elements in R. A similar story
holds for matrices with elements in C.

SL( N, R)
The special linear group.
Those matrices of GL(n, R) which have det A = 1.

O( N )
Orthogonal group,
All matrices satisfying AA T = 1.

SO( N )
Special orthogonal group, again adding det A = 1 as a condition to the or-
thogonal group.

U(N)
Unitary group, for which A† A = 1 where A† := ( A∗ ) T .

SU ( N )
Special unitary group, those unitary matrices which also satisfy det A = 1.

Sp(2N )
All matrices that satisfy ΩA + A T Ω = 0, with

1N
 
0
Ω= .
−1 N 0

There are various other matrix Lie groups not covered by this list, but those have
more complicated matrix conditions not easily guessed from some physics problem.
We will encounter them when we turn to the classification problem.
In order to illustrate some more of the concepts of Lie groups, and to connect
them to what are known as Lie algebras, let us focus on the rotation group in 3
dimensions (or SO(3) in the language introduced above) and discuss a few of its

6
 1.2 A quick tour

elements. Rotations around the z-axis are obtained by acting on points with the
one-parameter matrix
 
cos φ − sin φ 0
Rz (φ) =  sin φ cos φ 0 ≈ 1 − iδφTz (1.1)
0 0 1
where  
0 −i 0
Tz =  i 0 0 . (1.2)
0 0 0
In fact, the relation between Rz (φ) and Tz can be given to all orders, not just the first
order, and reads
e−iφTz = Rz (φ) . (1.3)
The matrix Tz is called an element of the Lie algebra corresponding to the Lie group
of which Rz (φ) is an element. Unsurprisingly, the group formed from just Rz (φ) is
called a subgroup of SO(3), in fact, these matrices form SO(2).
In order to compute the product of Lie group elements from their definition in
terms of Lie algebra generators, we in fact only need to know the commutators
of those generators. This is because the Baker-Campbell-Hausdorff expressions,
which give the product of matrix exponentials, only involve commutators on the
right hand side. For example, if [ X, Y ] = 1 · const., we have
1
e X eY = e X +Y + 2 [ X,Y ] . (1.4)
The general case is more complicated, but never requires knowledge of the prod-
uct of two Lie algebra matrices, only knowledge of the commutator. Therefore, it
makes sense to consider the abstract mathematical structure of a Lie algebra, which
is formed by elements Xa together with a Lie bracket,
[ Xa , Xb ] = i f abc Xc . (1.5)
The f abc are called structure constants. For matrix groups these Xa are of course also
matrices, and the Lie bracket is a commutator. However, the important thing is that
we never really need to know the product of matrices Xa , only their commutator, so
it makes more sense to view the Lie bracket as a single operation, rather than one
built from a product and a difference. The Lie product should furthermore satisfy
the condition that it is linear in its arguments, that it is anti-symmetric, and that it
satisfies the Jacobi identity,
[ Xa , [ Xb , Xc ]] + [ Xc , [ Xa , Xb ]] + [ Xb , [ Xc , Xa ]] = 0 . (1.6)
Again, these are automatic for matrix algebras, but can be viewed as a more general
definition of a mathematical structure.
We will here only consider Lie algebras with a finite number of elements. There
also exist Lie algebras with an infinite number of elements, of which the most com-
monly encountered ones in physics are the Kac-Moody algebras.. They occur in the Kac-Moody algebras
study of two-dimensional conformal field theories.
One finally also encounters non-matrix Lie algebra, that is, algebras which are
not commonly expressed or defined in terms of matrix generators. An example is
the Heisenberg algebra,
[ X, Y ] = [ X, Z ] = 0 , [Y, Z ] = X . (1.7)
The standard representation is in terms of differential operators acting on functions,
d
Y=
, Z = t, X = 1. (1.8)
dt
We will not have much to say about such representations either.

7
 1.2 A quick tour

1.2.2 Observing symmetries in quantum mechanics

We will see later that there are also strong consequences of symmetry in quantum
field theory, but let us stick to quantum mechanics for the time being. In quantum
mechanics, the fact that a system is invariant under a particular symmetry transfor-
mation means that the Hamiltonian remains unchanged if we perform that transfor-
mation. In a formula, we have
[ Ĥ, Â] = 0 , (1.9)
where Ĥ is the Hamiltonian and  is one of the generators of the symmetry group.
Let us try to see what this implies for the time evolution of states. States are labelled
by eigenvalues of the conserved charges. If we consider an SO(3) system as exam-
ple, this means that we label states by two eigenvalues, of Ĵ 2 and Ĵz for instance. Call
these l and m. Now in general, the time evolution of a state is given by

|ψ(t)i = ei Ĥt |ψ(0)i , (1.10)

so what appears in expressions for transition amplitudes are matrix elements of the
Hamiltonian. The symmetry condition (1.9) puts a strong constraint on these matrix
elements, as can be seen from the following expression
0 = hl 0 , m0 |[ L̂z , Ĥ ]|l, mi = (m0 − m)hl 0 , m0 | H |l, mi . (1.11)
The matrix element can only be non-zero when m = m0 .
This is a so-called selection
rule: there can be no transition from a state with m 6= m0 in systems which are
rotationally invariant.1
Another way in which the consequences of symmetry can be measured is by not-
ing that if a state transforms as a vector under some symmetry group, it necessarily
comes in a multiplet together with other states carrying the same energy. If you
now perturb the system slightly, in such a way that you break the symmetry, you
(end of lecture 1)2 can observe level splitting.

1.2.3 Abstract groups/algebras versus representations

One of the main reason for being a bit pedantic about defining groups and algebras
in an abstract sense is that there is almost always more than one way to realise
the same given group or algebra in a physical system. This leads us to define a
representation of a group, of dimension p, as a map from the group elements to a set
of p × p matrices D ( X ) acting on a p-dimensional vector space, which preserve the
group composition law,
D ( X1 · X2 ) = D ( X1 ) D ( X2 ) , for all X1 , X2 ∈ G . (1.12)
The product on the right denotes ordinary matrix multiplication. For a representa-
tion of an algebra, we would require
D ([ x1 , x2 ]) = [ D ( x1 ), D ( x2 )] , for all x1 , x2 ∈ g . (1.13)
Let us have a look at some examples of different representations. Consider the
Lie algebra sl (2, C), which is formed by all trace-less 2 × 2 complex matrices. Since
we consider the algebra over C, these matrices can be built using 3 generators; all
other matrices are then obtained by taking suitable complex linear combinations. A
useful basis is3
     
(f) 0 i (f) 0 −1 (f) i 0
T1 = , T2 = , T3 = . (1.14)
i 0 1 0 0 −i

1A refinement of this argument goes under the name of the Wigner-Eckart theorem, to be discussed
later.
3 These anti-hermitean matrices are related to the perhaps more familiar Pauli matrices by multipli-

cation with plus or minus i.

8
 1.3 Literature and software

These matrices satisfy the algebra

(f) (f) (f)
[ Ti , Tj ] = 2eijk Tk . (1.15)

They form the so-called fundamental or defining representation of the algebra sl (2).
(f)
However, the Ti are certainly not the only matrices that satisfy these commutation
relations. It is straightforward to verify that the same commutators are obtained
using the following 3 × 3 matrices,
     
0 0 0 0 0 2 0 −2 0
( a) ( a ) ( a )
T1 = 0 0 −2 , T2 =  0 0 0 , T3 = 2 0 0 . (1.16)
0 2 0 −2 0 0 0 0 0

This representation is called the adjoint representation, for reasons that will become
clear in the next chapter. In any case, the existence of these matrices shows that the
same algebra can have different representations.
Fundamental representation, conjugate fundamental representation, trivial rep-
resentation.
Tensor products, show first by writing out, then do in index notation following
Mukund. Discuss reducibility -¿ block diagonal. Discuss complete reducibility and
Maschke’s theorem.
2⊗2 = 3⊕1. (1.17)
Some Lie algebras do not admit representations which are finite-dimensional
(i.e. the generators cannot be written down as matrices with a finite number of
rows/columns). Such Lie algebras are called non-matrix Lie algebras, and their Non-matrix Lie algebras
existence shows that it is important to consider Lie algebras as mathematical struc-
tures by themselves, not necessarily defined in terms of matrices. Do not confuse
these with the infinite-dimensional algebras we mentioned earlier.

1.3. Literature and software

There are plenty of unreadable books on group theory that focus on all the gory
mathematical details without ever discussing a physics example. A few books not
in this category are

• H. Georgi, “Lie algebras in particle physics: from isospin to unified theories”,

Front. Phys. 54 (1982) 1–255.
A nice and compact book written for physicists.

• J. Fuchs and C. Schweigert, “Symmetries, Lie algebras and representations: A

graduate course for physicists”, Cambridge University Press, 1997.
More thorough and mathematical book, with lots of examples and exercises,
written from the perspective of mathematical physics.

• R. Cahn, “Semisimple Lie algebras and their representations”, Benjamin/Cummings,

1985.
Another very compact book, perhaps a bit too compact. Now also available
for free online,

http://phyweb.lbl.gov/~rncahn/www/liealgebras/book.html

A few others which I personally do not like too much, but some people swear by
these:

9
1.3 Literature and software

• H. Jones, “Groups, representations and physics”, Hilger, 1990.

More physics than some of the other books on this list. Has a nice discussion
of Young tableaux and of the Lorentz group.
• H. Samelson, “Notes on Lie algebras”, 1989.
More mathematically oriented book, now out of print, but available in online
from

http://www.math.cornell.edu/~hatcher/Other/Samelson-LieAlg.pdf

• W. Fulton and J. Harris, “Representation theory, a first course”, Springer, 1991.

Much more mathematically oriented. Starts (as the title suggest) right from
representation theory.

When we get a bit further on in the module, you might be interested to play
with Lie algebras yourself. Rather than doing it by hand, which can be cumbersome,
there is some good software out there which can help you build a better understand-
ing:
• LiE, a computer algebra package for Lie group computations, M.A.A. van
Leeuwen, A.M. Cohen and B. Lisser.
Can be used to construct weight systems, to find branching rules, and many
many other things which we do in this course. Available under the LGPL li-
cense from

http://www-math.univ-poitiers.fr/~maavl/LiE/

On Debian/Ubuntu Linux systems, install the lie package. Examples using

LiE are included at various points in these notes, in the form of small bits of
code in the margin labelled by “LiE”.
• Cadabra, a field-theory motivated approach to computer algebra, K. Peeters.
Contains various algorithms to deal with Young tableaux. Available from

http://cadabra.phi-sci.com

On Debian/Ubuntu Linux systems, install the cadabra package. Examples

using LiE are included at various points in these notes, in the form of small
bits of code in the margin labelled by “Cadabra”.

10
 Lie algebra classification
2
We now turn to the key question, namely to find a procedure that provides for us a
list of all possible Lie algebras. The key ingredient in this classification is to choose
a convenient basis of the generators, and then use this basis to show that all Lie
algebras can be reduced to combinations of sl (2) algebras, connected to each other
by specific non-zero commutation relations between generators of these sl (2)s. All
that is needed to work this out is knowledge about the representation theory of
sl (2). To give an idea of how this procedure works, we will first discuss sl (3) in this
language. We will then turn to the Cartan classification using Dynkin diagrams.

2.1. Main ingredients

2.1.1 The adjoint representation
We have seen several examples of representations already, but there is one represen-
tation that needs special attention because it is crucial in the present chapter. This
representation is called the adjoint representation, and is obtained by thinking of the
generators themselves as vectors. There is a natural action of the algebra elements
on themselves, given by the action of the commutator.

ad x (y) = [ x, y] , also written as ad x = [ x, ·] . (2.1)

This preserves the algebra structure, as is readily shown by making use of the Jacobi
identity,
[ad x, ad y]w = [ x, [y, w]] − [y, [ x, w]]
= [ x, [y, w] + [y, [w, z]]
(2.2)
= −[w, [ x, y]]
= [[ x, y], w] = ad[ x, y](w) .
The adjoint representation clearly always exists, and has a dimension equal to the
number of generators of the algebra. If required you can of course write it in more
usual matrix-vector multiplication form (like we did earlier for the two-tensor prod-
uct representation).

2.1.2 Representation theory of sl (2)

The second crucial ingredient for this chapter is the representation theory of the
simplest of Lie algebras, namely sl (2). Let us first rewrite the sl (2) algebra in a basis

11
 2.1 Main ingredients

(f)
which is more useful for this chapter. Instead of using the Ti matrices, we will use
the ‘cleaner’ ones
     
1 0 0 1 0 0
H= , E+ = , E− = . (2.3)
0 −1 0 0 1 0

(f)
These are simple complex linear combinations of the Ti , so they generate the same
complex algebra. Their commutation relations read

[ H, E± ] = ±2E± , [ E+ , E− ] = H . (2.4)

It is these commutation relations that we want to use to study sl (2).

So let us now think of the H, E± as abstract generators and study the possible
ways in which we can realise them. The starting point will be that we will choose a
basis of the representation space in which H 1 is diagonal, i.e. in which

Hvλ = λ vλ . (2.5)

The E± operators are ‘ladder operators’, in the sense that they raise or lower this
eigenvalue,
H ( E± vλ ) = E± Hvλ ± 2E± vλ = (λ ± 2) E± vλ . (2.6)
Therefore, if you start with some vλ , you can construct all the vectors in that repre-
sentation by laddering. Their H eigenvalues go up or down by two every time you
act.
We will assume now that the representation is finite-dimensional (this can be
shown to be true by virtue of the fact that the generators are hermitian, but we will
not waste time with that). Since the dimension is finite, there has to be a vector
Representations of sl (2) are which has the highest H 1 eigenvalue, which we will call Λ.
labelled by Λ, the highest weight
(largest eigenvalue of H). vΛ−2n := ( E− )n vΛ . (2.7)

After a finite number of steps, the string has to break,

E− vΛ−2N = 0 . (2.8)

To get more information, we have to look at how the raising operator acts. We get

E+ vΛ−2 = E+ E− vΛ = [ E+ , E− ]vΛ = HvΛ = Λ vΛ . (2.9)

We get back a scalar multiple of vΛ (important: you might have guessed, based
on (2.7), that this number is one, but it is not). For a generic starting point, the scalar
multiple is also easily computed; defining

E+ vΛ−2n = rn vλ−2n+2 , (2.10)

we get

E+ vΛ−2n = E+ E− vΛ−2n+2 = ( E− E+ + H )vΛ−2n+2 = (rn−1 + Λ − 2n + 2) vΛ−2n+2 .

(2.11)
We thus have the recursion relation

rn = rn−1 + Λ − 2n + 2 , (2.12)

with r0 = 0 because the chain breaks at the highest eigenvalue. This recursion
relation has the solution
r n = n ( Λ − n + 1) . (2.13)

12
 2.1 Main ingredients

Just like r0 = 0, we also need the chain to break off at the lower end,

0 = E+ E− vΛ−2N = ( E− E+ + H )vΛ−2N = (r N + Λ − 2N ) vΛ−2N . (2.14)

This leads to
N 2 + (1 − Λ ) N − Λ = 0 , (2.15)
with solutions N = −1 and N = Λ. The dimension of the representation is thus The eigenvalues of the diagonal
generator of sl (2) are integers,
dim = N + 1 = Λ + 1 , (2.16) when the basis of generators (2.4)
is used.

and most important: the value Λ has to be integer. The main result is thus that, in
this basis, the eigenvalues of H are integers.

2.1.3 Preview: extending sl (2)

To get an idea of the logic behind the classification of Lie algebras, we will now
first look at a concrete example, namely the extension of sl (2) by at least one ad-
ditional generator that commutes with H. We will see here how one is forced into
introducing more and more ladder operators, but in such a way that the various
commutators between the ladder operators are severely restricted. In fact, there are
only three non-trivial ways to do this. In the next section we will then turn to a
generalisation of this argument that classifies all finite-dimensional algebras.
Let us thus start from sl (2), and try to see how to systematically extend this alge-
bra to something larger. We start by adding a single generator H 2 which commutes
with H 1 (these mutually commuting operators form what is known as the Cartan
subalgebra or CSA). If we make this new generator commute not just with H 1 but
also with the ladder operators E± 1 , we end up with a rather trivial system. So we Extending by only a diagonal
will want to add at least one other generator, which does not commute with H 2 . generator H 2 is trivial, but if we
want to add more, we need at least
However, given that H 2 is hermitean, we will always need two extra generators, 2 and E2 , because of the
a pair E+ −
and these generators together with H 2 then necesssarily form an sl (2). hermiticity of H 2 .
If we do not make these generators of the two sl (2)’s talk to each other, we end
up with just a direct sum of two independent algebras, which is again rather trivial.
Therefore, the first non-trivial result comes in when we make the generators of the
two algebras have non-zero commutators with one another. For commutators with
the CSA,
j j
[ H i , E± ] = ± A ji E± . (2.17)

We must have Aii = 2 to get the diagonal relations, but A12 and A21 are still arbitrary
numbers. We cannot set them to zero because that would decouple the algebras
completely.
In order to figure out what these numbers can be, we have to look at the result
for the commutator of the step operators,
1 2
[ E± , E± ]. (2.18)

These operators are in fact again step operators, as you can show that

[ H 1 , [ E±
1 2
, E± 1
]] = λ[ E± 2
, E± ] (2.19)

for some value of λ. The essential ingredient here is the Jacobi identity (check this!).
Another result that follows from repeated use of the Jacobi identity is that of the
four commutators
1 2 1 2
[ E± , E± ] and [ E∓ , E± ], (2.20)
we have to have at least two non-zero ones.

13
 2.2 Cartan classification and Dynkin diagrams

From the representation theory of sl (2) we know that A ji are integers. If we keep
only a minimal choice,
1 2 1 2
[ E± , E± ] 6= 0 , [ E∓ , E± ] = 0, (2.21)

and all higher order commutators zero, one can show (Jacobi identities again) that
this corresponds to setting

A2 : A12 = −1 , A21 = −1 . (2.22)

As indicated, this algebra is called ‘A2 ’, but is also known as sl (3). There turn out to
be two other options, given by

B2 : A12 = −2 , A21 = −1 ,
(2.23)
G2 : A12 = −3 , A21 = −1 .

It is at this stage not clear that these are in fact the only choices, but we will see this
in the following sections.

I See also: This down-to-earth approach can also be found in chapter 3 of [2].

2.1.4 Real subalgebras

We have so far discussed complex algebras. That is to say, we were allowed to take
complex linear combinations of the generators of the algebra. For the example of
sl (2), the three generators H, E+ and E1 given in matrix representation in (2.3) can
thus be multiplied with any complex number, to produce any arbitrary matrix with
complex elements, subject only to the condition that the trace vanishes. This algebra
is more appropriately called sl (2, C). For this algebra you could have started with
some arbitrary other basis obtained as a complex linear combination of the basis
generators which we chose; this clearly does not make any difference.
However, it is of course possible to look at restrictions of this algebra which are
obtained by restricting the coefficients to real values. Once you do that, you have to
be careful about which generators you take. If you stick to the generators in (2.3),
an arbitrary linear combination with real coefficients give you all real traceless ma-
The algebras sl (2, R) and su(2) are trices. This algebra is called sl (2, R).
the two real subalgebras (or ‘real You can also choose to first go to a basis formed by the matrices (1.14), and then
forms’) of sl (2, C). As real
algebras, they are inequivalent.
take real combinations of those to form an arbitrary element in the real algebra. This
gives what is known as su(2). This is a different algebra, with structure constants
which differ from sl (2, R) only because we restrict ourselves to real linear combina-
tions.

2.1.5 Simple, semi-simple, ideals

2.2. Cartan classification and Dynkin diagrams

The general structure of a (simple) Lie algebra is given by a number r of mutually
commuting generators H i , together with other operators which are eigenvectors of
The rank of an algebra is the H i in the adjoint representation. The number r is called the rank of the algebra. We
number of CSA generators. will start by slightly relaxing normalisation conditions on the generators, so that we
have, apart from the trivial brackets between the Cartan subalgebra generators H i ,
Eigenvalues of the CSA generators also the basic defining relations
are called weights, or roots when
they concern the adjoint [ Hi , Eα ] = αi Eα . (2.24)
representation.

14
 2.2 Cartan classification and Dynkin diagrams

These αi are called the weights of the adjoint representation, and also called roots.
Because the Eα are not hermitian,

[ Hi , Eα† ] = −αi Eα† , (2.25)

we see that if α is a root, so is −α. The corresponding two operators are raising and Roots always come in pairs, with
lowering operators. A key ingredient which we have seen in the example is that the opposite sign.
commutator of two ladder operators is itself a ladder operator,

[ Hi , [ Eα , Eβ ]] = (α + β)i [ Eα , Eβ ] . (2.26)

This follows from a Jacobi identity. Making this consistent with (2.24) requires that
such a commutator is a multiple of the generator Eα+ β ,

[ Eα , Eβ ] = cα,β Eα+ β . (2.27)

Similarly, when α = − β, we see from (2.26) that the commutator has to be a linear
multiple of the CSA generators,

[ Eα , E−α ] = di H i . (2.28)

Finally, when (α + β)i is not an eigenvalue of Hi (in more technical terms: when
α + β is not a root), then the commutator vanishes,

[ Eα , Eβ ] = 0 if α + β 6∈ Φ. (2.29)

A basis of this type is called a Cartan-Weyl basis.

We have also seen that not all Cartan-Weyl bases are equally nice. In particular,
the Cartan-Killing form
( X, Y ) := Tr(ad X ad Y ) (2.30)
is not always diagonal on the Cartan subalgebra. However, you can always choose
your generators in this subalgebra so that this ‘metric’ becomes

Tr( Hi Hj ) = λδij . (2.31)

Moreover, we can always normalise the ladder operators in the same way,

Tr( E− β Eα ) = δαβ (2.32)

We will use this choice as it makes many formulas much simpler and many inter-
mediate steps redundant. With this choice, you can show that the di in (2.33) are
actually equal to the components of the root, i.e.

[ Eα , E−α ] = αi H i . (2.33)

There are many other bases used in the literature, but dealing with a non-trivial
Killing form is rather a lot of extra work.
For every root α, there is an sl (2) algebra (we have seen this explicitly for the
three roots of sl (3)). In order to get this subalgebra in the canonical form you have
to rescale the generators by factors involving the root and the norms of the root, To every root (not only the simple
ones!) is associated an sl (2).
Ẽ± = |α|−1 E±α , H̃ = 2|α|−2 α · H . (2.34)

The basis obtained in this way is called a Chevalley basis. Because of the norms, it is
here obviously quite useful that we are in a basis where the Killing form is diagonal.
We now need to bring a little bit more structure into the system of roots before
we can continue. In general, given a set of roots Φ, it is always possible to choose a

15
 2.2 Cartan classification and Dynkin diagrams

subset Φ+ of them, such that two conditions hold: if α ∈ Φ+ then −α 6∈ Φ+ , and if

α, β ∈ Φ+ , then α + β ∈ Φ+ . Essentially, this just amounts to picking a basis in root
space (and there are often many ways to do this). The roots we choose in this way
are called positive roots. Those elements α ∈ Φ+ which cannot be written as the sum
of positive roots are called simple roots. These simple roots satisfy the important
The difference of two simple roots property that the difference of two simple roots is not a root (show this!).
is not a root. The key idea will now be the following. We will start from some generator cor-
responding to a particular root β. We can then ladder with the Ẽ± (as constructed
above) for a particular root, just like we did when we discussed sl (2) representation
theory. We found there that the eigenvalues of the H̃ operator have to be integers.
Translated to our problem here, and taking care of the normalisation (2.34) that was
required to obtain a canonically normalised sl (2) algebra, this means that

hα, βi
2 ∈ Z. (2.35)
hα, αi
However, since β is a root, you can also apply this with α and β exchanged. We can
get a better understanding for what this means by observing that the inner product
between two vectors is expressible in terms of the cosine of the angle between them,

hα, βi2 mm0

cos2 θαβ = ∈ . (2.36)
hα, αih β, βi 4

The cosine square cannot be large than one. Moreover, we have to exclude the case
where the cosine equals one, because that implies that the two roots are equal or
opposite, which cannot happen because it would mean that one root occurs multiple
times. Thus, there are only four possibilities,

mm’ θ
0 π
2 (90◦ )
1 2π
3 (120◦ ) (2.37)
2 3π
4 (135◦ )
3 5π
6 (150◦ )

A string of roots can thus never contain more than four roots.
The key quantities to classify a Lie algebra are thus these various integers (2.35).
We have actually already given these a name before: when we extended sl (2) to
sl (3), we implicitly used generators in a nice sl (2) normalisation, and then called
the eigenvalues of the H’s the Cartan matrix elements. We thus now have a more
formal definition of the Cartan matrix in terms of simple roots,1

2h α i , α j i
Aij := . (2.38)
hα j , α j i

These elements are actually constrained further. First of all, the inner product be-
tween two different simple roots is always negative (see also below). Therefore, the
non-diagonal elements of the Cartan matrix are negative. In addition, the Schwarz
inequality tells us that
hαi , α j i ≤ hαi , αi ihα j , α j i (2.39)
so that (again, the maximal case is trivial)

Aij A ji < 4 . (2.40)

1 Pay attention to the order of the indices; the Cartan matrix is in general not symmetric.

16
 2.2 Cartan classification and Dynkin diagrams

The only choices for off-diagonal elements and their transpose partners are thus

A12 = −1 and A21 = −1 ,

A12 = −2 and A21 = −1 , (2.41)
A12 = −3 and A21 = −1 .

Finally, since the matrix is built from αi which should be linearly independent, we
know that det A 6= 0.
The determinant condition is a bit awkward to implement, and in fact eliminates
many possibilities which are still left open by the conditions above (some examples
will be seen in the tutorial problems). Working out which combinations are allowed
is not particularly enlightening, so we will just jump straight to the result. This is
most easily expressed in terms of graphical rules, which lead to the so-called Dynkin
diagrams.
[KP: diagrams and rules to be added; see any decent book on Lie algebras]

2.2.1 Root systems

The logic that led us to conclude that the Cartan matrix elements are integers can
in fact be played with an arbitrary root,µ, not just the simple ones. A key ingredi-
ent is that the lowest and highest weight of an sl (2) representation are −Λ and Λ
respectively. For a generic weight µ in a string of roots, this leads to2

hµ, βi
2 = m− p, (2.42)
hα, αi
where m is the number of steps you can act still with a E− lowering operator and p
the number of steps that you can make with an E+ raising operator before the root
string ends. This is not particularly useful (for instance, a ‘zero’ just tells you that
you are in the middle of the string, not how long it is on either end). However, if
you know that m or p is zero, then this expression determines the length of the root
chain. This is true for instance when you start with µ being a simple root, as the
difference of two simple roots is not a root, and so m = 0 in that case. As a side
result, this also immediately shows that inner products between simple roots are
always negative. The only multiple of a root α which
[KP: laddering procedure to be added; discussed in detail on the board] is also a root is −α.
Once all roots (the simple as well as the non-simple ones) have been obtained,
we can use (2.34) to construct all generators which explicitly generate all the sl (2)
subalgebras for us.
This laddering procedure which we have just described in terms of roots can also
be described directly at the level of the algebra, and is known as the Chevalley-Serre
relations. They read
j j
[ Hi , H j ] = 0 , [ H i , E±
i
] = ± A ji E± , i
[ E+ , E− ] = δij H i ,
(2.43)
i 1− A ji j
(ad E± ) E± = 0.

Here the first line shows us the commutation relations between the generators as-
sociated to the simple roots, while the second line determines which commutators
of step operators are kept non-zero. This mimicks what we have seen in the sl (3)
example earlier.

2 This formula in fact has even wider range of applicability, because we can use it not just for the

eigenvalues in the adjoint representation (the roots), but als also for more general representations. We
will return to this issue in the next chapter.

17
 2.3 Rank-two examples

2.3. Rank-two examples

In order to see how the Cartan classification does its job, let us look in some detail at
the three rank-two algebras that we introduced earlier: A2 , B2 and G2 . Their Cartan
matrices are given by
     
2 −1 2 −2 2 −3
A2 : , B2 : , G2 : . (2.44)
1 2 −1 2 −1 2

The first thing to do is to construct all the step operators, or equivalently, all the
roots. For this we use the logic that led to (2.42). We start from a simple root, and
determine whether we can add one of the other simple roots by computing (2.42). A
general root which we encounter in this way is a linear combination of simple roots,

β= ∑ k i αi . (2.45)
i

The sum of all k i ’s is called the level of a root. Looking at (2.42), we see that what we
need to do is to keep track of the numbers

∑ k j A ji (2.46)
j

at every stage.
For B2 we find ...
For G2 the story is similar, but now the root chains are longer because of the
LiE: presence of the ‘−3’ in the Cartan matrix. We get
> positive roots(G2)
[[1,0], [0,1] ,[1,1]  
,[2,1], [3,1] ,[3,2] 2 −3
] −1 2
[2, −3] [−1, 2]
[1, −1] α1 + α2 m1 = m2 = 1 → p2 = 2 (2.47)
[0, 1] α1 + 2α2
[−1, 3] α1 + 3α2
[1, 0] 2α1 + 3α2

In the third line, going from α1 + 2α2 , you cannot go to 2α1 + 2α2 because m for this
step is zero: there is no root 2α2 , so α1 + 2α2 is actually at the bottom of an α1 string,
not at the first step.

18
 Representation theory
3
The machinery of generalised ladder operators and their eigenvalues, the roots, has
a much wider applicability than just the construction of algebras. It also allows us
to discuss generic representations of those algebras. The logic that allowed us to de-
termine the lengths of chains of roots does not only apply when starting with a root,
but, as we have already briefly mentioned, also when we start with a more general
set of eigenvalues of the Cartan subalgebra generators. In the present chapter we
will see how this works in detail and look at a few physics applications.

3.1. Weight space

3.1.1 Weight vectors versus roots
Consider an arbitrary representation of an algebra. If we diagonalise the Cartan
subalgebra generators H i , we can characterise the representation by looking at the
eigenvalues of these operators when acting on the vectors that span the representa-
tion space; compactly1
H |µi = µ|µi . (3.1)
The formula for the length of a root chain through a weight is exactly the same as
for the special case when the weight is also a root,

hµ, βi
2 = m− p, (3.2)
hα, αi
This formula becomes useful when we know that adding a root to µ leads us to a
weight which does not exist in the spectrum. This is exactly the case when µ is a
highest weight, and in this case p = 0.

3.1.2 Building weight space

Following the logic of the previous subsection, we can construct the entire weight
space given the highest weight, by laddering ‘downwards’ with the roots.2 This is

1 This expression can again be a bit confusing. We have r generators H, so H in this formula is an

r-dimensional vector of operators (matrices) and µ is an r-dimensional vector of eigenvalues.

2 If you would consider lowest-weight representations, then you could ladder upwards from the

starting point and the procedure would mimick the construction of the space of roots more closely; the
reason for looking at highest-weight representations is purely historical.

19
3.3 Example: the eightfold way

best made explicit with another example. Consider the representation with highest
weight [1, 1]. Its weight space follows as
 
2 −1
−1 2
[1, 1] α1 + α2
[−1, 2] [2, −1] α2 , α2 (3.3)
[0, 0] 0
[−2, 1] [1, −2] −α1 , − α2
[−1, −1] − α1 − α2

This gives all weights in the representation with highest weight α1 + α2 .

What the diagram above does not tell you is the multiplicity of each weight. You
might have thought that since you can reach the [0, 0] weight in the middle of the
diagram in two different ways, this leads to a multiplicity two. That is indeed the
correct answer, but the wrong logic: using the same logic you would conclude that
the lowest-weight state has multiplicity four, which is incorrect.
What you have to do in order to figure out the multiplicity is to look explicitly
at the states corresponding to these weights, and determine whether the different
ways of obtaining a state with the given weight lead to linearly independent states.
As an example, it is not hard to show that the two states of weight [0, 0],

E−α1 E−α2 |α1 + α2 i and E−α2 E−α1 |α1 + α2 i (3.4)

are linearly independent (this is because the two lowering operators do not com-
mute). However, it is less clear a priori that the four different ways of obtaining
states with weight [−1, −1] are all linearly dependent. This requires repeated swap-
ping around of lowering operators using commutation relations. Unfortunately this
process is often laborious. There is a clever technique due to ... and a similarly clever
formula do to Freudenthal; we may discuss these later if time permits.

I See also: Georgi is one of the few books discussing the multiplicity of weights in
some detail; see around page 135. We will discuss the multiplicity issue again when
we discuss Casimir operators later.

3.2. Tensor products

3.3. Example: the eightfold way
I See also: The ideas behind the eightfold way are described in some detail in [4]
from page 143 onwards, or in [2] section 3.6.

20
 Assorted topics
4
4.1. Young tableaux for sl (n)
Young tableaux are graphical notations for highest-weight representations of sl (n)
(there exist modified tableau notations for other groups as well, see [7] for details). It
is based on an idea we have seen earlier, namely that you can build representations
using tensor products of only fundamental representations, suitably symmetrised or
anti-symmetrised. This is a special property which only holds true for some algebras
(and hence the Young tableau method is not as often seen for non-sl (n) cases).
The sl (3) example we have discussed reads, in Young tableau notation

⊗ = ⊕
(4.1)
φi φj = φ(i φj) + φ[i φj] .

As indicated, each box on the left-hand side corresponds to a fundamental represen-

tation of dimension 3. The horizontal row of two boxes is the symmetrised tensor
product, while the vertical row of boxes is the anti-symmetrised tensor product.
Young tableaux generalise this idea to more complicated tensor products, in which
the irreducible representations on the right hand side do not in general correspond
to fully symmetrised or fully anti-symmetrised products.
To be more precise, a Young tableau with n boxes can be viewed as the tensor
product of n fundamental representations, with each box being related to a fun-
damental index, with symmetrisation applied to every row and then anti-symme-
trisation applied to each row (the result is then no longer symmetric in the indices in
each row). An example of this procedure:

1 1
i j
: φi φj φk → φi φj φk + φj φi φk
k 2 2
(4.2)
1 1 1 1
→ φi φj φk + φj φi φk − φk φj φi − φj φk φi .
3 3 3 3
The normalisation factor is chosen such that applying the combined symmetrisation
and anti-symmetrisation operation twice gives the same result (in technical lingo:
the Young projector is idempotent). Obviously, since you cannot anti-symmetrise
more than n indices if they only take on n different values, the number of rows in a
Young tableau for sl (n) is at most n.
There is a rather simple way to compute the dimension of a representation from
its associated Young tableau. This is easiest to explain at the level of a concrete Tableau dimension formula.

21
 4.1 Young tableaux for sl (n)

example. Consider in An = sl (n + 1) the representation

. (4.3)

Its dimension is obtained using the ratio of two expressions. For n = 3 we have

4 5 6 7

3 4

2
7! × 4
dY = = = 140 . (4.4)
6 4 2 1
6!/5
3 1

LiE: In the numerator, we start with n + 1 and then fill boxes by following the rule that
> dim([2,1,1],A3) going to the right increases the number by one, and going down decreases by one.
140 In the denominator, we fill each box with the sum of the number of boxes to the
right and the number of boxes below it, plus one (this is known as the hook formula).
To evaluate the diagrams, simply multiply the numbers in all the boxes. The result
gives the number of independent components that you need in order to describe a
tensor with the symmetries of the corresponding Young tableau.
A Young tableaux with bm columns with m boxes can be labelled as (b1 b2 b3 . . .).
In this notation, there is a direct relation between tableaux and the Dynkin labels of
the corresponding highest weight representation. Here are some examples for sl (3):

[1, 0] 3
[0, 1] 3̄
[1, 1] 8
(4.5)
[2, 0] 6
[0, 2] 6̄
[2, 1] 15

The last column indicates the dimension as computed using the dimension formula.
Another useful aspect of Young tableaux is that they can be used to decompose
a tensor product of representations into irreducible ones. To do this, draw the two
diagrams next to one another, and place in each box of the second diagram a symbol
a in the first row, a b in the second row and so on. Now add the boxes of the second
diagram to the first diagram, one by one, using the following rules at every step,
1. Each diagram thus constructed must be a Young tableau.
2. For An (i.e. sl (n + 1)), no diagram can have more than n + 1 rows (you would
anti-symmetrise in more than n + 1 indices which take on only n + 1 different
values, hence you would get zero).
3. Boxes with the same label must not appear in the same column.
4. At any given box position, define n a to be the number of a’s above and to the
right of it (similarly for nb and so on). Then we must have n a ≥ nb ≥ nc and
so on.
5. Two tableaux of the same shape are counted as different only when the la-
belling is different.
You can understand some of these rules simply in terms of symmetries: the third
rule, for instance, says that boxes which are symmetrised in the original diagram

22
 4.2 Bianchi identities

cannot appear in anti-symmetrised locations in the resulting diagram. These rules

go under the name of the Littlewood-Richardson algorithm. Let us see this in action
using examples. The tensor product of two fundamentals of A2 decomposes as

3⊗3 = ⊗ a
(4.6)
= a ⊕ = 6⊕3.
a

The product of a fundamental times a conjugate fundamental gives instead

3̄ ⊗ 3 = ⊗ a

a ⊕
(4.7)
= = 8⊕1.
a

You could also compute this by exchanging the two factors. At the level of the Young
tableau manipulations this leads to different intermediate steps, but the same end
result: Cadabra:
3 ⊗ 3̄ = ⊗ a \tab{#}::FilledTableau(dimension=3).
b \tab{1}\tab{a}{b};
@lr tensor!(%);
→ a ⊕ 1 a 1
+ ;
a (4.8) b a
 b
→ a ⊕ a b ⊕ ⊕ b = 8⊕1.
b a a
b
The second diagram violates rule 4, while the fourth diagram violates rule 3.

I See also: Young tableaux are discussed in e.g. [4] section 8.4.

4.2. Bianchi identities

The more complicated symmetries are multi-term symmetries, such as the Ricci cyclic
identity or the Bianchi identity, which relate a sum of terms with different index dis-
tributions. These symmetries are all manifestations of the so-called Garnir symme-
tries of Young tableaux [8]. These state that the sum over all anti-symmetrisation of
boxes in a Garnir hook is identically zero. Examples of such Garnir hooks are given
below,

which represent the Ricci cyclic identity, the Bianchi identity on a five-form and
a more general Garnir symmetry, respectively. Applying a Garnir symmetry on a
tensor produces a sum of tensors, which means that one can no longer restrict to the
canonicalisation of tensor monomials.
There are, however, a few subtleties. Firstly, it would be prohibitively expen-
sive to write down all terms in a Young-projected tensor. Instead, it is much more
efficient to reduce the Young-projected forms by making use of the mono-term sym-
metries, which are easy to deal with using the methods of [9–11]. One thus obtains,

23
4.3 Branching rules and grand unified theories

for e.g. the Riemann tensor,

1

R abcd → 2 R abcd − R adbc + R acbd , (4.9)
3

instead of the (2!)4 = 16 terms which are produced by the Young projector. The
expression on the right-hand side manifestly satisfies the cyclic Ricci identity, even if
one only knows about the mono-term symmetries of the Riemann tensor. Using the
projector (4.9) it is easy to show e.g. that 2 R abcd R acbd = R abcd R abcd . The monomial
on the left-hand side maps to

R abcd R acbd → 13 R abcd R acbd + R abcd R abcd ,

(4.10)

and similarly R abcd R abcd maps to twice this expression, thereby proving the identity
in a way which easily extends to much more complicated cases.

4.3. Branching rules and grand unified theories

One thing which the Dynkin classification of Lie algebras definitely makes more
transparent is the issue of subalgebras. These are subsets of the generators of an
algebra which form a closed set under the commutation relations. In physics, such
subalgebras play an important role in the discussion of grand unified theories or
GUTs, which are attempts to describe the strong, weak and electromagnetic force in
one framework. This is a very rich topic of which we will here only discuss a few
aspects to give you a flavour and a starting point to extend to more complicated
situations.
As you will see in other modules in this MSc., the strong and weak nuclear force
are related to su(3) and su(2) gauge symmetries respectively, while the electromag-
netic force is described by a u(1) gauge theory. It is a bit ugly to have these as three
different groups, and many physicists have attempted to instead obtain these forces
from one gauge theory which uses a larger gauge symmetry group.
In general, it may be possible to find the same subalgebras inside an algebra in
multiple different ways. Consider for instance the sl (2) subalgebras inside sl (3).
For sl (3) we can write a matrix representation analogous to (2.3),
   
1 0 0 0 0 0
H 1 = 0 −1 0 , H 2 = 0 1 0 ,
0 0 0 0 0 1
   
0 1 0 0 0 0
1 1
E+ = 0 0 0 , E− = 1 0 0 ,
0 0 0 0 0 0
    (4.11)
0 0 0 0 0 0
2 2
E+ = 0 0 1 , E− = 0 0 0 ,
0 0 0 0 1 0
   
0 0 1 0 0 0
θ θ
E+ = 0 0 0 , E− = 0 0 0 .
0 0 0 1 0 0

One way to isolate an sl (2) subalgebra in here is to simply use only H 1 together
with the E±1 generators. This is what is called a regular embedding, as it simply uses

a subset of the CSA generators togeter with the generators corresponding to a subset
of the roots. An alternative embedding, however, is to use the 3-dimensional adjoint

24
 4.3 Branching rules and grand unified theories

representation of sl (2) which we wrote down in (1.16). Those T (a) generators are
expressible as linear combinations of the six raising and lowering operators written
above, but this embedding cannot be viewed as restricting the CSA and the root
system. Such embeddings are called non-regular. From the way we just described the
difference, it should be obvious that regular embeddings are much easier to describe
in the language of roots and weights. We will focus on these for the remainder of
this section.
In the standard model, we have various fermion multiplets which transform un-
der different representations of the su(3) × su(2) × u(1) symmetry. We will not dis-
cuss that in detail, but here just take for granted that before electroweak symmetry
breaking, these representations are

left-handed quark doublets (u, d), (c, s), (t, b) : (3, 2, 16 ) L ,

right-handed up quarks u, c, t : (3, 1, 23 ) R ,
right-handed down quarks d, s, b : (3, 1, − 31 ) R ,
(4.12)
left-handed lepton doublet (e− , νe ) : (1, 2, − 21 ) L ,
right-handed electrons e− : (1, 1, −1) R ,
right-handed neutrinos νe : (1, 1, 0) R .

There are of course three families of each of these; we have only written down one
of them.
We can obviously embed su(3) × su(2) in su(5), using a matrix embedding
 
U3 0
U= . (4.13)
0 U2

Here tr U3 = tr U2 = 0. There is then a one-parameter family of traceless matrices

left, of the form  
a 0 0 0 0
0 a 0 0 0
 
Q= 0 0 a 0 0 , (4.14)

0 0 0 b 0
0 0 0 0 b
where 3a + 2b = 0. Such matrices generate the remaining u(1). The map from the
extended Dynkin diagram to the Dynkin diagram for the subgroups is

...graphic (4.15)

Restricting the 5,

[ 1 0 0 0 ] [ 1 0 ] [0]
[ −1 1 0 0 ] [ −1 1 ] [0]
[ 0 −1 1 0 ] → [ 0 −1 ] [0] . (4.16)
[ 0 0 −1 1 ] [ 0 0 ] [1]
[ 0 0 0 −1 ] [ 0 0 ] [−1]

This means that the weight zn = [0 0 1 0]n is

With the choice q = 1 we get

5̄ → (3̄, 1, 13 ) + (1, 2, − 21 ) . (4.17)

These correspond precisely to the (conjugate of the) down quarks and the left-handed
leptons (electron and neutrino). An extended version of this analysis shows that
in fact 5̄ + 10 contains one entire family of the standard model. For more details
see [12].

25
4.3 Branching rules and grand unified theories

26
Bibliography

[1] H. Georgi, “Lie algebras in particle physics: from isospin to unified theories”,
Front. Phys. 54 (1982) 1–255.
[2] J. Fuchs and C. Schweigert, “Symmetries, Lie algebras and representations: A
graduate course for physicists”, Cambridge University Press, 1997.
[3] R. Cahn, “Semisimple Lie algebras and their representations”,
Benjamin/Cummings, 1985.

[4] H. Jones, “Groups, representations and physics”, Hilger, 1990.

[5] H. Samelson, “Notes on Lie algebras”, 1989.
[6] W. Fulton and J. Harris, “Representation theory, a first course”, Springer, 1991.

[7] P. Cvitanovič, “Group theory”, 2008.

[8] B. E. Sagan, “The symmetric group; representations, combinatorial algorithms
and symmetric functions; 2nd edition”, Springer, 2001.
[9] A. Dresse, “Polynomial poisson structures and dummy variables in computer
algebra”, PhD thesis, Université Libre de Bruxelles, 1993.

[10] R. Portugal, “An algorithm to simplify tensor expressions”, Comp. Phys.

Commun. 115 (1998) 215–230, gr-qc/9803023.
[11] R. Portugal, “Algorithmic simplification of tensor expressions”, J. Phys. A32
(1999) 7779–7789.

[12] A. N. Schellekens, “Beyond the standard model”. Lecture notes for the ’95
Dutch AIO/OIO school, Jan 1995.

27