Budd - Lie Algebras in Particle Physics
Budd - Lie Algebras in Particle Physics
Budd - Lie Algebras in Particle Physics
Timothy Budd
Version: June 13, 2022
Contents
1 Introduction to groups and representations 2
1
These lecture notes are an extended version of notes by F. Saueressig and are loosely based on the book Lie
Algebras in Particle Physics by H. Georgi.
1 Introduction to groups and representations
Group theory is the mathematical theory underlying the notion of symmetry. Understanding the
symmetries of a system is of great importance in physics. In classical mechanics, Noether’s theo-
rem relates symmetries to conserved quantities of the system. Invariance under time-translations
implies the conservation of energy. Invariance with respect to spatial translations leads to the
conservation of momentum, and rotational invariance to the conservation of angular momentum.
These conserved quantities play a crucial role in solving the equations of motion of the system, since
they can be used to simplify the dynamics. In quantum mechanics, symmetries allow one to classify
eigenfunctions of, say, the Hamiltonian. The rotational symmetry of the hydrogen atom implies
that its eigenstates can be grouped by their total angular momentum and angular momentum in
the z-direction, which completely fixes the angular part of the corresponding wave-functions.
In general a symmetry is a transformation of an object that preserves some properties of the object.
The two-dimensional rotational symmetries SO(2) correspond to the linear transformations of the
Euclidean plane that preserve lengths. A permutation of n particles preserves local particle number.
Since a symmetry is a transformation, the composition of two symmetries of an object is always
another symmetry. The concept of a group is introduced exactly to capture the way in which such
symmetries compose.
Definition 1.1 (Group). A group (G, · ) is a set G together with a binary operation · that
sends f, g to an element f · g. It is required to satisfy
(i) closure: if f, g ∈ G then also f · g ∈ G;
(ii) associativity: if f, g, h ∈ G then f · (g · h) = (f · g) · h;
(iii) identity: there exists an element e ∈ G such that e · f = f · e = f for all f ∈ G;
(iv) inverse: for every f ∈ G there is an element f −1 ∈ G such that f · f −1 = f −1 · f = e.
Let us start by looking at a few abstract examples (check that these indeed satisfy the group
axioms!):
Example 1.4. The permutation group (SX , ◦) of a set X: SX is the set of bijections {ψ :
X → X} and ◦ is given by composition, i.e.
Example 1.5. The general linear group G = GL(N, C) consisting of all complex N × N
matrices with non-zero determinant with product given by matrix multiplication.
Based on these examples several remarks are in order. First of all, the binary operation of a group
is not necessarily implemented as a multiplication, as should be clear from examples 2 and 3. One
is free to choose a notation different from f = g · h to reflect the product structure, e.g. f = g ◦ h,
f = g + h, f = gh. Often we will write G instead of (G, ·) to indicate a group if the product is
clear from the context.
Notice that there is no requirement that g · h = h · g in the definition of a group, and indeed
2
examples 1.4 and 1.5 contain elements g and h such that g · h 6= h · g.
Definition 1.7 (Order of a group). The order |G| of a group G is the number of elements
in its set. A group of finite order is called a finite group, and a group of infinite order is an
infinite group.
For example, Sn is a finite group of order n!, while GL(N, C) is an infinite group.
Although one may use different operations to define a group (multiplication, addition, composition,
. . . ) this does not mean that the resulting abstract groups are necessarily distinct. In fact, we
will identify any two groups that are related through a relabeling of their elements, i.e. any two
groups that are isomorphic in the following sense.
Definition 1.8 (Isomorphic groups). Two groups G and G0 are called isomorphic, G ∼ = G0 ,
if there exists a bijective map φ : G → G0 that preserves the group structure, i.e.
Example 1.9 (Cyclic group). For positive integer n the three groups defined below are all
isomorphic (see exercises). Any one of them may be taken as the definition of the cyclic group
of order n.
• the set Zn := {0, 1, 2, . . . , n − 1} equipped with addition modulo n;
• the set Cn ⊂ Sn of cyclic permutations on {1, 2, . . . , n} equipped with composition;
• the nth roots of unity Un := {z ∈ C : z n = 1} equipped with multiplication.
Abstract group theory is a vast topic of which we will only cover a few aspects that are most
relevant for physics. Central in this direction are representations.
3
invertible linear map S : V → V 0 such that
Having introduced abstract groups and their representations, we disentangle two aspects of the
symmetry of a physical system: the abstract group captures how symmetry transformations com-
pose, while the representation describes how the symmetry transformations act on the system.2
This separation is very useful in practice, thanks to the fact that there are far fewer abstract
groups than conceivable physical systems with symmetries. Understanding the properties of some
abstract group teaches us something about all possible systems that share that symmetry group.
As we will see later in the setting of Lie groups, one can to a certain extent classify all possible
abstract groups with certain properties. Furthermore, for a given abstract group one can then try
to classify all its (inequivalent) representations, i.e. all the ways in which the group can be realized
as a symmetry group in the system.
For instance, the isospin symmetry of pions in subatomic physics and the rotational symmetry of
the hydrogen atom share (almost) the same abstract group. This means that their representations
are classified by the same set of quantum numbers (the total (iso)spin and (iso)spin in the z-
direction).
First we will focus on representations of finite groups, while the remainder of the course deals
with Lie groups, i.e. infinite groups with the structure of a finite-dimensional manifold (like
U (1), SO(3), SU(2), . . .).
· e g1 g2 ...
e e g1 g2 ...
g1 g1 g1 · g1 g1 · g2 ... (1.4)
g2 g2 g2 · g1 g2 · g2 ...
.. .. .. .. ..
. . . . .
Note that the closure and inverse axioms of G (axioms (i) and (iv) in Definition 1.1) precisely state
that each column and each row contains all elements of the group exactly once. Associativity on
the other hand is not so easily visualized in the table.
We can easily deduce using the Cayley table that there exists a unique abstract group of order 3.
Indeed, if G = {e, a, b} with e the identity, then we know that the Cayley table is of the form
· e a b
e e a b
a a ? ?
b b ? ?
There is only one way to fill in the question marks such that each column and each row contains
2
By definition the representations we are considering act as linear transformations on a vector space. So we only
cover physical systems with a linear structure in which the symmetries are linear transformations. This is not much
of a restriction, since all quantum mechanical systems are of this form.
4
the three elements e, a, b (à la sudoku!):
· e a b
e e a b
(1.5)
a a b e
b b e a
It is easily checked that this multiplication is associative and thus defines a group. In fact, we have
already encountered this group as a special case of the cyclic group of Example 1.9 when n = 3,
i.e. G is isomorphic to the group Z3 of addition modulo 3 (with isomorphism given by e → 0,
a → 1, b → 2 for instance).
Given the abstract group, one can construct representations. Let us start with one-dimensional
representations and therefore specify the vector space on which the representation acts to be
V = C. Just like any group, Z3 has a trivial representation
Dtriv (e) = 1 , Dtriv (a) = 1 , Dtriv (b) = 1 . (1.6)
A non-trivial one-dimensional representation is given by rotations of the complex plane by 2π/3,
D1 (e) = 1 , D1 (a) = e2πi/3 , D1 (b) = e4πi/3 . (1.7)
A straightforward computation verifies that these representations indeed obey the multiplication
laws indicated in the group multiplication table.
How about higher-dimensional representations? One such representation that can be constructed
naturally for any finite group is the regular representation, which we introduce now.
The dimension of the regular representation thus equals the order of the group, dim(Dreg ) = |G|.
Note that in the basis |g1 i, . . . , |gn i these vectors correspond to the column vectors
1 0
0 ..
|g1 i = . , ··· , |gn i = . ,
.
. 0
0 1
so if the group multiplication table is known, it is easy to construct explicit matrices for the
regular representation. Indeed D(h) corresponds to the matrix that has |h · g1 i, . . . , |h · gn i as its
columns.
In the example of Z3 in which we order the basis elements as |ei, |ai, |bi a quick look at the Cayley
table (1.5) leads to
1 0 0 0 0 1 0 1 0
Dreg (e) = 0
1 0 , Dreg (a) = 1 0 0 , Dreg (b) = 0 0 1 . (1.11)
0 0 1 0 1 0 1 0 0
5
You may check explicitly by matrix multiplication that these matrices indeed constitute a repre-
sentation Dreg : Z3 → GL(C3 ).
Besides constructing the special regular representation, there are other ways to produce higher-
dimensional representations. Given two representations D1 and D2 of a group one may naturally
construct two new representations of the group with larger dimension: their direct sum D1 ⊕ D2
and their tensor product D1 ⊗ D2 . To understand their definition, we need to recall the direct sum
and tensor product of vector spaces.
Definition 1.13 (Direct sum and tensor product of vector spaces). If V is a vector
space with basis |v1 i, . . . , |vn i and W is a vector space with basis |w1 i, . . . , |wm i, then
• the direct sum V ⊕ W is the vector space with basis
As you can check (and the notation already suggests), we have that the dimensions of these vector
spaces are related by
The representations D1 ⊕ D2 and their tensor product D1 ⊗ D2 act precisely on these vector
spaces.
In the exercises you will check explicitly that these abstract definitions satisfy the axioms of a
representation. The direct sum representation has a less abstract interpretation when examined at
the level of the matrices. Since (D1 ⊕ D2 )(g) sends vectors in V ⊂ V ⊕ W to vectors in V ⊂ V ⊕ W
and similarly for W ⊂ V ⊕ W , the corresponding matrix has the block-diagonal form
D1 (g) 0
(D1 ⊕ D2 )(g) = , (1.15)
0 D2 (g)
6
Definition 1.15 (Invariant subspace). An invariant subspace W of a representation D :
G → GL(V ) is a linear subspace W ⊂ V such that
Trivially W = V and W = {0} are invariant subspaces, so we are usually more interested in
proper invariant subspaces W , meaning that W ⊂ V is invariant and W 6= {0} and W 6= V .
Equivalently this is saying that there exists a change of basis S : V → V such that SD(g)S −1 is
block-diagonal for all g ∈ G and such that the blocks are matrices of irreducible representations of
G.
Let us return to the example of the regular representation Dreg of Z3 given in (1.11) and convince
ourselves that it is reducible and completely reducible. To see that it is reducible, we should identify
a proper invariant subspace. An example is given by the 1-dimensional subspace W ⊂ C3 spanned
by the vector (1, 1, 1)T . This vector is mapped to itself by all three matrices of the representation,
so W is indeed invariant. In order to see that Dreg is completely reducible we observe that the
three matrices can be simultaneously diagonalized. Setting ω ≡ e2πi/3 the appropriate similarity
transform is given by
1 1 1 1 1 1
1 1
S = √ 1 ω 2 ω , S −1 = √ 1 ω ω 2 , (1.17)
3 1 ω ω2 3 1 ω2 ω
7
Definition 1.19 (Unitary representation). A representation D : G → GL(CN ) is unitary
if D(g)† D(g) = 1 for all g ∈ G.
In the exercises you will prove the following result, which shows that we may safely restrict ourselves
this special family of representations.
is invariant as well. We thus need to show that for every g ∈ G and v ∈ W ⊥ that D0 (g)v ∈ W ⊥ as
well, or that hD0 (g)v|wi = 0 for all w ∈ W . This we may checked using the unitarity of D0 (g):
unitary
hD0 (g)v|wi = hv|D0 (g)† wi = hv|D0 (g)−1 wi = hv|D0 (g −1 )wi = 0,
where in the last equality we used that w0 = D0 (g −1 )w ∈ W because W is invariant and hv|w0 i = 0
because v ∈ W ⊥ .
Let us now choose a basis w1 , . . . , wn of W and a basis v1 , . . . , vN −n of W ⊥ , which combined
form a new basis of CN . In this new basis D0 (g) is block-diagonal, because hvi |D0 (g)wj i = 0 and
hwj |D0 (g)vi i = 0 for all g ∈ G. We conclude that D0 , and therefore also D, is equivalent to a direct
sum of two representations D1 and D2 that both have smaller dimension.
We may iterate this procedure for D1 and D2 separately, until we are left with only irreducible
representations. Since the dimensions of the representations decrease at every step, this stops
quickly.
It is good to keep in mind that Lemma 1.20 and Theorem 1.18 may fail for infinite groups. For
instance, the group (Z, +) of Example 1.3 has the two-dimensional representation D given by
1 x
D(x) = . (1.20)
0 1
It is reducible, because the subspace spanned by (1, 0)T is invariant. However, it is not completely
reducible. If it were, D would be equivalent to the direct-sum of two one-dimensional represen-
tations or, equivalently, the matrices D(x) would be simultaneously diagonalizable for all x. But
none of the D(x) with x 6= 0 are diagonalizable (let alone simultaneously).
Theorem 1.21 (Schur’s lemma - Part I). Let D1 : G → GL(V1 ) and D2 : G → GL(V2 )
be irreducible representations of a group G, and A : V2 → V1 a linear map. If A intertwines
8
D1 and D2 , meaning that D1 (g)A = AD2 (g) for all g ∈ G, then A = 0 or D1 and D2 are
equivalent.
which shows that ker(A) is a (not necessarily proper) invariant subspace of D2 . Since D2 is
irreducible, ker(A) = {0} or ker(A) = V2 , meaning that A is injective or A = 0.
A is surjective or A = 0: Let us consider the image im(A) = {Av : v ∈ V2 } ⊂ V1 . For every
g ∈ G we have
The results is useful if you want to determine whether two different-looking irreducible representa-
tions are equivalent: if you can find any non-zero linear map A that intertwines them, then they are
equivalent. The second Schur’s lemma specializes this result in the case that both representations
act on the same linear space V1 = V2 .
Proof. If A commutes with all elements of D, then for any x ∈ C the same is true for A − x1.
According to Theorem 1.21, for every x ∈ C we have either that A − x1 = 0 or that A − x1 is
invertible. The latter condition is equivalent to det(A − x1) 6= 0. Since det(A − x1) is a non-
constant polynomial (namely the characteristic polynomial of the matrix A), like any non-constant
polynomial it must have a root at some λ ∈ C. Hence for x = λ the first condition must be satisfied,
namely that A = λ1. This is precisely what we intended to prove.
In other words, if you can find any linear map A that is not proportional to the identity and that
commutes with a representation D, then D must be reducible.
Let us summarize the restrictions stated in Shur’s lemmas for a linear map A : V → V that
commutes with all elements of a completely reducible representation D : G → GL(V ), so this
applies to any representation of a finite group G. Let us consider the matrix A in the basis in which
D is block-diagonal. More precisely, let D1 , D2 , D3 be the inequivalent irreducible representations
of G, then we may assume that D(g) takes the form
Di1 (g) 0 0 ···
0 Di2 (g) 0 · · ·
D(g) = 0 , (1.25)
0 Di3 (g) · · ·
.. .. .. ..
. . . .
9
where the indices i1 , i2 , . . . ∈ {1, 2, . . .} may repeat, since an irrep may in general be included more
than once in the decomposition. Let us examine the matrix A, whose blocks we accordingly label
as
A11 A12 A13 · · ·
A21 A22 A31 · · ·
A = A .
31 A32 A33 · · ·
.. .. .. ..
. . . .
meaning that Ak` intertwines Dik and Di` . If ik 6= i` then Dik and Di` are inequivalent and the
lemma implies that Ak` = 0. If ik = i` then the second lemma implies that there is a complex
number ak` such that Ak` = ak` 1. In other words, there is only very littly freedom left in A, even
though the dimension of the vector space involved can be really large.
For example, if D = D1 ⊕ D2 ⊕ D3 decomposes into the three inequivalent irreducible representa-
tions, then A is diagonal and determined by three complex numbers a11 , a22 , a33 ,
a11 1
D1 (g) 0 0 0 0
D(g) = 0 D2 (g) 0 =⇒ A= 0 a22 1 0 . (1.27)
0 0 D3 (g) 0 0 a33 1
a11 1 a12 1
D1 (g) 0 0 0 0 0
0 D1 (g) 0 0 a21 1 a22 1 0 0
D(g) = =⇒ A= . (1.28)
a33 1 a34 1
0 0 D2 (g) 0 0 0
0 0 0 D2 (g) 0 0 a43 1 a44 1
This situation naturally arises in quantum mechanical systems, where the vector space V is the
Hilbert space of states of the system. If the system possesses a symmetry, then the symmetry group
G acts on the Hilbert space V via a unitary representation D : G → GL(V ). Observables, like the
Hamiltonian, are encoded in hermitian operators Ô : V → V that commute with the action of the
symmetry group, i.e. [D(g), Ô] = 0. In other words, Ô is precisely of the form of the operator A
described above. So if we choose a basis for the Hilbert space such that the representation D is
block-diagonal as in (1.25), then the conclusions of Shur’s lemmas apply to the matrix elements of
Ô.
In particular, in the presence of symmetries time evolution of a quantum mechanical system, en-
coded in the Hamiltonian Ĥ, is heavily constrained. Most of the matrix elements of Ĥ are required
to vanish, as can be seen from the examples (1.27) and (1.28). If an irreducible representation oc-
curs only once in the decomposition of the Hilbert space like in (1.27), then any state transforming
under that irrep will necessarily be an eigenstate of Ĥ. Another consequence is that the energy
spectrum will necessarily come with multiplicities depending on the dimensions of the irreducible
representations.
10
2 Lie groups and Lie algebras
Informally, if the elements of a group G depend smoothly, g = g(α), on a set of real parameters αa ,
a = 1, . . . , n, then G is called a Lie group of dimensions n. A prototypical example is the rotation
group
Definition 2.1 (Lie group). A Lie group of dimension n is a group (G, ·) whose set G
possesses the additional structure of a real n-dimensional smooth manifold, such that the
group structures, composition (g, h) → g · h and inversion g → g −1 , are smooth maps.
Recall that, informally speaking, a set G is an n-dimensional smooth manifold if the neighbourhoud
of any element can be smoothly parametrized by n parameters (α1 , . . . , αn ) ∈ Rn . In terms of such
a parametrization the composition and inversion correspond locally to maps Rn × Rn → Rn and
Rn → Rn respectively, which we require to be smooth in the usual sense.
In practice, we will exclusively work with a special class of Lie groups, the matrix Lie groups.
Definition 2.2 (Matrix Lie group). A matrix Lie group is a Lie group whose elements are
(complex or real) matrices and the group operation is matrix multiplication.
This is not really a restriction since almost all Lie groups that are of relevance in physics are
(isomorphic to) matrix Lie groups3 . Note that the dimension of the matrices is in general unrelated
to the dimension of the Lie group (for instance, the 2×2 matrices of SO(2) comprise a 1-dimensional
Lie group).
11
Definition 2.3 (Generators of a Lie group). The n generators of an n-dimensional Lie
group are defined as
∂
ξa = g(α)α=0 , a = 1, . . . , n. (2.2)
∂αa
Of course the generators depend on the chosen parametrization, but the generators of one parametriza-
tion are necessarily linear combinations of those of another parametrization.4 Put differently,
the vector space spanned by the generators is independent of the parametrization. Since the
parametrization is smooth, we may approximate elements in the close vicinity of the identity via
the Taylor expansion
g(α) = 1 + αa ξa + O(α2 ), (2.3)
where :=
Pn the Einstein summation convention to sum over repeated indices is understood: xa ya
a=1 xa ya .
Applying (2.2) to (2.1) shows that SO(2) possesses a single generator given by
0 −1
ξ= . (2.4)
1 0
There is a natural way to recover group elements from linear combinations of the generators.
Notice that if the vector αa is not particularly small, then the matrix 1 + αa ξa does not give an
element of the group, because we have neglected the contributions from the higher order terms in
(2.3). However, we can get a better approximation of a group element by choosing a large integer
k and considering the matrix 1 + αka ξa , which we can then compose with itself k times. In the
limit of k → ∞ this necessarily results in a group element. This motivates the introduction of the
exponential map.
Definition 2.4 (The exponential map). For any linear combination αa ξa of generators,
the exponential eαa ξa is the group element given by
k ∞
ξ 1
1 + αa a
X
e αa ξa
= lim = (αa ξa )m . (2.5)
k→∞ k m!
m=0
We may check that the exponential map of multiples of (2.4) reproduces the parametrization (2.1).
To see this, note that for any integer n
2n 2n+1
0 −1 n 1 0 0 −1 n 0 −1
= (−1) , = (−1) .
1 0 0 1 1 0 1 0
4
Note that in the physics literature one often uses a convention with an additional imaginary unit in the definition
of the generator, i.e. g() = 1 + i a ξ˜a + . . . and ξ˜a ≡ −iXa . The reason is that most Lie groups encountered in
physics are unitary, in which case the generators ξ˜a are Hermitian while the ξa are antihermitian (see section 2.4
below). We will stick to the mathematical convention without the i, which is more convenient when considering the
algebraic structure.
12
Then we can compute eαX via the series expansion (2.5),
∞ ∞ ∞
(−1)n 2n 1 0 (−1)n 2n+1 0 −1
X
αξ
X 1 n
X
e = (αξ) = α + α
n! (2n)! 0 1 (2n + 1)! 1 0
n=0 n=0 n=0
1 0 0 −1 cos α − sin α
= cos α + sin α = . (2.6)
0 1 1 0 sin α cos α
Note that it is a bit of a coincidence that the parametrization eαξ reproduces our original parametriza-
tion g(α), as we could have started with a different parametrization g(α). In any case we see that
the exponential map of SO(2) covers the whole group. This turns out to be true for almost all the
Lie groups that we will consider in this course.
Definition 2.5 (compactness). A matrix Lie group is compact if the matrix entries of all
group elements are bounded, i.e. smaller in absolute value than some fixed C > 0.
Definition 2.6 (connectedness). A Lie group is connected if every element can be reached
from the identity by a continuous path.
Note that SO(2) is connected and compact. The group O(2) of 2 × 2 orthogonal real matrices is
compact but not connected: these matrices can have determinant ±1, but there is no way to reach
a matrix with determinant −1 via a continuous path from the identity (which has determinant
1).
Fact 2.7. If G is a connected and compact matrix Lie group, then any element g ∈ G can be
obtained as the exponential of a linear combination of generators, g = eαa ξa .
If all generators mutually commute, i.e. [ξa , ξb ] = 0 for all a, b = 1, . . . n, then it is easily seen from
the definition of the exponential map that A = αa ξa and B = βa ξa satisfy
In other words the Lie group is necessarily Abelian and the group multiplication is equivalent to
addition of the corresponding generators.
For general non-Abelian Lie groups however we know that there are A = αa ξa and B = βa ξa that
fail this relation, i.e.
eA eB 6= e(A+B) 6= eA eB . (2.8)
Note that we do have
eλA eµA = e(λ+µ)A for λ, µ ∈ R. (2.9)
In particular, setting λ = 1 and µ = −1 we find for the group inverse of eA ,
Another important property of the exponential map is that it relates the determinant in the group
to the trace of the generators,5
det eA = etr A . (2.11)
5
If you would like to see this proved, recall that every (complex) matrix A can be put in Jordan normal form J
−1
by an invertible matrix S, A = SJS −1 . Then eA = eSJS = SeJ S −1 , so that det eA = det eJ while also tr A = tr J.
It remains to show that det eJ = etr J . Since J is upper triangular, by the definition of the exponential the same is
true for eJ , which has its diagonal entries equal to the exponential of the diagonal entries of J. So det eJ is equal to
the product of the diagonal entries of eJ , which is etr J .
13
2.2 Lie algebras
As we will see below, the generators ξa span a vector space g = {αa ξa : αa ∈ R} with a particular
structure, known as a Lie algebra.
Definition 2.8 (Lie algebra). A Lie algebra g is a vector space together with a “Lie bracket”
[·, ·] that satisfies
(i) closure: if X, Y ∈ g then [X, Y ] ∈ g;
(ii) linearity: [aX + bY, Z] = a[X, Z] + b[Y, Z] for X, Y, Z ∈ g and a, b ∈ C;
(iii) antisymmetry: [X, Y ] = −[Y, X] for X, Y ∈ g;
(iv) Jacobi identity: [X, [Y, Z]] + [Z, [X, Y ]] + [Y, [Z, X]] = 0 for X, Y, Z ∈ g.
When working with matrices, like in our case, we can take the Lie bracket to be defined by the
commutator of the matrix multiplication
[X, Y ] := XY − Y X. (2.12)
It is then easy to see that this bracket satisfies linearity and antisymmetry. In the exercises you will
verify that it also satisfies the Jacobi identity by expanding the commutators. Properties (ii)-(iv)
thus follow directly from working with the generators of a matrix Lie group. We will now see that
the closure (i) of the generators Xa follows from the closure of the group multiplication.
Let A, B ∈ g, i.e. A and B are linear combinations of the generators ξa , and let ∈ R, then
typically
eA eB 6= eB eA , or equivalently e−A e−B eA eB 6= 1. (2.13)
However, by the closure of the group e−A e−B eA eB must be some group element. Let us deter-
mine this group element for small . Expanding to second order in ,
∂
g(t)t=0 = [A, B]
∂t
is a generator. This shows that the closure property (i) holds.
We conclude that to any Lie group G we can associate a Lie algebra g spanned by the generators
of G. In mathematical terms the linear space spanned by the generators is the tangent space of
the manifold G at the identity e ∈ G, so we find that the tangent space of a Lie group G naturally
possesses a Lie algebra structure g. By convention, we will use Fraktur font to denote Lie algebras:
the Lie algebras of the Lie groups SO(N ), SU(N ), . . . , are denoted by so(N ), su(N ), . . .
Since the generators ξa form a basis of the Lie algebra, the commutator of two generators [ξa , ξb ]
is a real linear combination of generators again. The real coefficients are called the structure
constants.
14
Definition 2.9 (Structure constants). Given a Lie group G with generators ξa , a =
1, . . . , n. The structure constants fabc of the Lie algebra g spanned by ξa are defined through
the relation
[ξa , ξb ] = fabc ξc . (2.15)
[ ξa , [ ξb , ξc ] ] + [ ξb , [ ξc , ξa ] ] + [ ξc , [ ξa , ξb ] ] = 0 (2.17)
4) The structure constants are sufficient to compute the product of the group elements exactly.
This follows from the Baker-Campbell-Hausdorff formula for exponentials of matrices
1 1
eX eY = eX+Y + 2 [X,Y ]+ 12 ([X,[X,Y ]]+[Y,[Y,X]])+more multi-commutators . (2.19)
Representations of a Lie group and a Lie algebra are closely related. Indeed, if D is an N -
dimensional representation of G we can consider the generators of G in the representation D given
by
∂
D(eαa ξa )α=0 ,
Xa = a = 1, . . . , n, (2.20)
∂αa
or, equivalently,
15
because for any αa ∈ Rn we have
(2.5)
αa k (2.21) αa k (2.5) αa Xa
D(eαa ξa ) = lim D 1+ ξa = lim 1 + Xa = e .
k→∞ k k→∞ k
We claim that then indeed D([A, B]) = [D(A), D(B)]. On the one hand we have for ∈ R
that
(2.22)
e−D(A) e−D(B) eD(A) eD(B) = D(e−A )D(e−B )D(eA )D(eB )
= D(e−A e−B eA eB )
(2.14)
1 + 2 [A, B] + O(3 )
= D
(2.21)
= 1 + 2 D([A, B]) + O(3 ).
On the other hand, by applying (2.14) directly we also have
determines a representation of g. And one can find a corresponding representation of the Lie group
by considering the exponential (2.22).6 The possibility to go back an forth between Lie groups and
Lie algebras via the exponential map is very powerful, because Lie algebras are often much easier
to work with.
We finish this chapter with a discussion of several explicit families of Lie groups and their Lie
algebras. Two families of Lie groups, the special unitary groups SU(N ) and special orthogonal
groups SO(N ), play an important role in physics. We also include the definition of the less
familiar family of compact symplectic groups Sp(N ), because together SU(N ), SO(N ) and Sp(N )
form the only infinite families of the (yet to be introduced) simple compact Lie groups, as we will
see in the last chapter of these lecture notes.
Its importance as a symmetry group stems from the fact that it leaves invariant the N -dimensional
scalar product familiar from quantum mechanics,
h v | w i ≡ va∗ wa , v, w ∈ CN , (2.26)
6
Warning: This does not always work as we will see later. Often a group element g can be written as g = eA
in multiple ways (e.g. e2πξ = 1 = e0ξ in our example of SO(2)), so D(eA ) := eD(A) is only well-defined if eD(A) is
independent of the choice of A.
16
where we used the summation convention. Indeed, we observe that for any U ∈ SU(N ),
U † U =1
hU v|U wi = (U v)∗a (U w)a = Uab
∗ ∗
vb Uac wc = vb∗ (U † )ba Uac wc = vb∗ δbc wc = hv|wi. (2.27)
Note that we have not used that det U = 1. The most general linear transformations preserving
the inner product thus form the larger unitary group
n o
U(N ) = U ∈ CN ×N : U † U = 1 . (2.28)
Since this phase commutes with all elements of the group it does not play a role in understanding
the algebraic structure of unitary matrices.7 For this reason we concentrate on SU(N ) rather than
U(N ).
In order to determine the corresponding Lie algebra we need to translate the conditions (2.25) for
the Lie group elements into conditions on the generators. Writing an arbitrary group element in
†
terms of the exponential map, U (α) ≡ eαa ξa , one has U † = eαa ξa and U −1 = e−αa ξa . The condition
U † = U −1 then implies that the generators are antihermitian ξa† = −ξa . In addition, det U = 1
imposes that the generators must be traceless, tr ξa =0. This property follows from
!
log det U = 0 = tr log eαa ξa = αa tr ξa . (2.30)
Since this must hold for any value of the αa the generators must be traceless. Thus the Lie algebra
su(N ) consists of all antihermitian, trace-free matrices,
n o
su(N ) = X ∈ CN ×N : X † = −X, tr X = 0 . (2.31)
The number of generators can be deduced from the following counting argument: a N × N matrix
with complex entries has 2N 2 parameters. The condition that the matrix is antihermitian imposes
N 2 constraints. Requiring that the matrices are trace-free removes an additional generator (the
one corresponding to global phase rotations). Hence
h v , w i ≡ v a wa , (2.34)
Again SO(N ) is not the largest group of matrices preserving the inner product,
because we have
N ×N : Λ Λ = 1 is that of all or-
T
not used that det Λ = 1. The largest group O(N ) = Λ ∈ R
thogonal N × N matrices including reflections. Since any orthogonal matrix has determinant ±1,
7
We will see later that the absence of this phase in SU (N ) entails that the Lie algebra su(N ) is simple (in a
precise sense), while u(N ) is not.
17
O(N ) really consists of two disjoint components, one for each sign of the determinant. The com-
ponent containing the identity matrix is precisely SO(N ). We will focus on the latter because
it is connected (and compact), and therefore any element can be obtained via the exponential
map.
T
For any Λ ∈ SO(N ) we may write Λ = eαa ξa with ξa real N × N matrices. Then ΛT = eαa ξa
and Λ−1 = e−αa ξa . Thus ΛT = Λ−1 implies that the generators of the Lie algebra so(N ) are
antisymmetric martrices ξaT = −ξa . The restriction on the determinant implies that the generators
are traceless, but this already follows from antisymmetry. Hence we find that
so(N ) = X ∈ RN ×N : X T = −X .
(2.36)
The number of generators follows from counting the independent entries in an antisymmetric N ×N
matrix and is given by
1
dim so(N ) = N (N − 1) . (2.37)
2
1N
2N
X 0
(v, w) ≡ va Jab wb , J= . (2.38)
−1N 0
a=1
Hence it satisfies n o
Sp(N ) = U ∈ C2N ×2N : U † U = 1, U T JU = J . (2.39)
Together with the fact that the generators have to be antihermitian for U to be unitary, this leads
us to identify the Lie algebra as
n o
sp(N ) = X ∈ C2N ×2N : X † = −X, X T = JXJ . (2.42)
With some work one may deduce from these conditions that sp(N ) has dimensions N (2N +1).
18
3 Irreducible representations of su(2)
The exponential map provides a connection between Lie groups and the underlying algebra. As
a consequence we can construct representations of the Lie group from representations of the cor-
responding Lie algebra. The goal of this section is to construct all finite-dimensional irreducible
representations of the Lie algebra su(2).
A basis of traceless Hermitian matrices, well-known from quantum mechanics, is given by the Pauli
matrices
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = , (3.2)
1 0 i 0 0 −1
satisfying
σa† = σa , tr σa = 0, σa σb = δab 1 + iabc σc . (3.3)
Here abc is the completely anti-symmetric -tensor with 123 = 1. To obtain a basis of antihermitian
traceless matrices, we then only have to multiply the Pauli matrices by i. In particular, if we
set
ξa = − 2i σa (3.4)
then
[ ξa , ξb ] = abc ξc , (a, b, c = 1, 2, 3). (3.5)
It follows that (with this choice of basis) the structure constants of su(2) are fabc = abc .
Fact 3.1. Just like in the case of finite groups, every representation of a compact Lie group
is equivalent to a unitary representation.
We have seen in Section 2.4 that the generators in a unitary representation are antihermitian.
Since SU(2) is compact, it is thus sufficient to look for antihermitian matrices X1 , X2 , X3 satisfying
(3.6).
Recall that a representation D : G → GL(V ) is irreducible if it has no proper invariant subspace
W ⊂ V , i.e. a subspace such that D(g)w ∈ W for all w ∈ W and g ∈ G. An invariant subspace
W can be equivalently characterized in terms of the Lie algebra g.
8
Although there is no confusion possible in the case of su(2), where everyone agrees that the fundamental rep-
resentation is the two-dimensional defining representation, there are some inconsistencies in the literature in the
terminology “fundamental representation” for other Lie groups. For this reason we will stick with using defining
representation.
19
Definition 3.2 (Invariant subspace of a Lie algebra representation). A linear subspace
W ⊂ CN is an invariant subspace of a Lie algebra representation D : g → CN ×N if Xw ∈ W
for all X ∈ g and w ∈ W .
Of course, it is sufficient to check that Xa w ∈ W for all w ∈ W and the n generators Xa of the
representation. An irreducible representation of su(2) therefore corresponds to a set of N × N
antihermitian matrices satisfying (3.6) that do not leave any proper subspace W ⊂ CN invari-
ant.
Such representations can be constructed systematically via the highest-weight construction. The
first step in the construction diagonalizes as many generators as possible. Recall the following
fundamental fact in linear algebra.
Fact 3.3. If A1 , . . . , Ak are hermitian, antihermitian or unitary matrices, then they are simul-
taneously diagonalizable if and only if they all commute, i.e. [Ai , Aj ] = 0 for all i, j = 1, . . . , k.
For su(2) this implies that only one generator may be taken diagonal, which without loss of
generality we take to be X3 . Since X3 is antihermitian,
J3 ≡ iX3 (3.7)
is a real diagonal N × N matrix. The remaining generators X1 and X2 are recast into raising and
lowering operators
1 i
J± ≡ √ (J1 ± i J2 ) = √ (X1 ± i X2 ) , Ja ≡ iXa . (3.8)
2 2
Since Xa† = −Xa , one has (J− ) = (J+ )† . Furthermore, the commutator (3.6) implies
[ J3 , J± ] = ±J± , [ J+ , J− ] = J3 . (3.9)
The key property of the raising and lowering operators is that they raise and lower the eigenvalues
of J3 by one unit. To be precise, suppose |mi is an eigenvector of J3 with eigenvalue m, which is
necessarily real because J3 is Hermitian. Then J± |mi is an eigenvector of J3 with eigenvalue m ± 1
(unless J± |mi = 0), because
J− |ji = Nj |j − 1i , (3.13)
20
Employing the definition (3.13)
We can continue this process and define |j −ki to be the properly normalized state obtained after k
applications of J− on |ji, assuming it does not vanish. In particular we introduce the normalization
coefficients Nj−k > 0 via
J− |j − ki ≡ Nj−k |j − k − 1i . (3.16)
We claim that the states |ji, |j − 1i, . . . span an invariant subspace W of the representation. It is
clear from the construction that W is invariant under J3 and J− , so it remains to verify that for
J+ |j −ki ∈ W for all k. This can be checked by induction: we have seen that always J+ |j −1i ∈ W .
Now suppose J+ |j − ki ∈ W for some k ≥ 1. Then
1 1
J+ |j − k − 1i = J+ J− |j − ki = (J− J+ |j − ki + J3 |j − ki) , (3.17)
Nj−k Nj−k
Noticing that the recursion relation (3.18) is a telescopic sum, the solution for the normalization
coefficients is obtained as
k
X
|Nj−k |2 = |Nj |2 + (j − l) = 12 (k + 1)(2j − k) . (3.20)
l=1
In order for the number of states transforming within the representation to be finite, there must
also be a lowest-weight state with J3 eigenvalue j − l, l ∈ N. By definition this state is annihilated
by J− ,
J− |j − li = 0, (3.21)
21
since there is no state with J3 -eigenvalue j − l − 1. This implies that the norm Nj−l introduced in
(3.16) and computed in (3.20) has to vanish
1 p !
Nj−l = √ (2j − l)(l + 1) = 0 . (3.22)
2
This happens if and only if j = 0, 1/2, 1, 3/2, . . . and l = 2j. In other words, one obtains a finite-
dimensional representation if j is either integer or half-integer. The finite-dimensional irreducible
representations of su(2) can then be labeled by their j-value, and are called the spin-j represen-
tations. The spin-j representation is 2j + 1-dimensional and the J3 -eigenvalues of the states cover
the range −j ≤ m ≤ j with unit steps.
We have proved that any N -dimensional irreducible representation of su(2) is equivalent to the
spin-j representation with 2j + 1 = N , meaning that we can find a basis |mi ≡ |j, mi, m =
−j, −j + 1, . . . , j such that
J3 |j, mi = m |j, mi, J− |j, mi = Nm |j, m − 1i, J+ |j, mi = Nm+1 |j, m + 1i, (3.23)
1 p
Nm = √ (j + m)(j − m + 1). (3.24)
2
In particular we can read off explicit matrices for the generators in the spin-j representation
and
[J+ ]m0 m = hj, m0 | J+ |j, mi = Nm+1 hj, m0 |j, m + 1i
1 p (3.26)
=√ (j + m + 1)(j − m) δm0 ,m+1 .
2
The matrix representation of the lowering operator is obtained via J− = (J+ )† . It remains to
check that this spin-j representation is indeed an irreducible representation. It is easy to verify
that the matrices J3 , J+ , J− satisfy the commutation relations (3.9). Solving (3.7) and (3.8) then
gives explicit matrices Xa satisfying the commutation relations [Xa , Xb ] = abc Xc . To see that the
representation is irreducible let’s consider an arbitrary (non-vanishing) vector
j
X
|ψi = am |j, mi. (3.27)
m=−j
This means that if |ψi is in some invariant subspace then all basis vectors are in this subspace,
so the spin-j representation has no proper invariant subspaces! This completes the proof of the
following theorem.
22
We close this section by checking that for j = 1/2 we reproduce the two-dimensional fundamental
representation of su(2). Using the basis |1/2, 1/2i, |1/2, −1/2i we evaluate (3.25) to
1 1 0 1
J3 = = σ3 = iX3 . (3.30)
2 0 −1 2
23
4 The adjoint representation
4.1 Definition
Let G be a matrix Lie group defined in terms of N -dimensional matrices and g the corresponding
Lie algebra. If the dimension of G, i.e. the number of generators of the Lie algebra g, is n, then
G possesses a natural n-dimensional representation with underlying vector space given by the Lie
algebra g itself.
This definition makes sense because the elements of G and g are both N × N matrices. To see
that gXg −1 ∈ g, notice that
t 7→ g etX g −1 (4.2)
determines a smooth path in G that visits the identity at t = 0. Hence,
Let us look at the corresponding representation Dad : g → gl(g) of the Lie algebra, where gl(g) is
the space of linear mappings from g to itself9 . According to (2.21) we should consider a generator
Y ∈ g and expand the action of Dad (1 + Y ) to first order in ,
Dad (1 + Y )X = (1 + Y )X(1 − Y + . . .)
= X + [Y, X] + . . .
= (1 + [Y, · ] + . . .)X
= (1 + Dad (Y ) + . . .)X.
Hence Dad (Y ) acts on the Lie algebra g by taking the Lie bracket [Y, · ]. Note that the Lie algebra
plays a double role here: to each element Y ∈ g we associate a linear transformation Dad (Y ) of g
given by Dad (Y )X = [Y, X]. Since the notation Dad (Y )X can be a bit awkward, it is convenient
to introduce the ket notation |Xi for the elements X ∈ g on which the adjoint representation acts.
Then Dad (Y )(X) = [Y, X] can be more succinctly summarized as
To obtain explicit matrices [Ta ]bc for this representation, we choose a basis of generators ξa span-
ning the Lie algebra g and consider an inner product10 such that hξa |ξb i = δab . Recall that the
commutator of the generators is given in terms of the structure constants by
[ ξa , ξb ] = fabc ξc . (4.5)
To find the matrix element [Ta ]bc corresponding to the action of Ta = Dad (ξa ) on g we use the
inner product to find
(4.4) (4.5)
[Ta ]bc = hξb | ξa |ξc i = hξb |[ξa , ξc ]i = facd hξb |ξd i = facb . (4.6)
24
Definition 4.2 (Adjoint representation of a Lie algebra). The adjoint representation
Dad : g → gl(g) of a Lie algebra g is given by Y |Xi = |[Y, X]i. Once a basis of g has been
chosen with structure constants fabc , the generators of g in the adjoint representation are
given by the matrices
[Ta ]bc ≡ facb , a, b, c = 1, . . . , n. (4.7)
The generators Ta satisfy the commutator relation [Ta , Tb ] = fabc Tc . This can be verified by the
Jacobi identity (2.18) (see exercises).
The dimension of the adjoint representation equals the dimension of the Lie group. This is similar
to the regular representation encountered in the context of finite groups (definition 1.12), where
the elements of the group served as a basis of the linear space on which the representation acts.
Now it is not the group elements but the generators that span the linear space.
provides a scalar product on the space of generators of the adjoint representation. (Note that
the trace tr is with respect to the matrix-indices of the generators [Ta ]bc .)
Fact 4.4. As a side-remark let us mentions that if g is a simplea Lie algebra and one is not
worried over the precise normalization, one may equally compute the Cartan-Killing metric
by taking the trace γab = c tr[Xa Xb ] in any other irreducible representation than the adjoint
one, because they are all equal up to a positive real factor c.
a
A Lie algebra is simple if its adjoint representation is irreducible. A more useful definition of a simple Lie
algebra will be given in Chapter 7. For now it is sufficient to know that the Lie algebras su(N ) for N ≥ 2 and
so(N ) for N ≥ 3 are all simple.
A natural question concerns the freedom in writing down the generators Ta . Since the generators
ξa constitute a basis of a linear space, there is the freedom of transforming to a new basis Xa0 by
performing an invertible, linear transformation L:
25
This, in turn, implies that the Cartan-Killing metric transforms as
0
= tr Ta0 Tb0 = Lca Ldb tr L−1 Tc L L−1 Td L = Lca Ldb γcd ,
γab = tr [Ta Tb ] 7→ γab (4.13)
where we used the cyclicity of the trace.
Since the Cartan-Killing metric is symmetric in the generator indices a, b, one may choose the
basis transformation L such that that after the transformation γab is of the diagonal form
−1n−
0 0
γab = tr [Ta Tb ] = 0 +1n+ 0 (4.14)
0 0 0 1n0
for some integers n− , n+ and n0 that add up to n. This fact about symmetric matrices may
be familiar to you from general relativity (where the spacetime metric can be brought by a local
basis transformation to Minkowski form, i.e. n− = 1, n+ = 3) and is known under the name of
Sylvester’s law of inertia.
The block containing the negative eigenvalues is generated by the n− “compact” (rotation-type)
generators, the positive eigenvalues belong to the n+ “non-compact” (boost-type) generators and
the zero-entries originate from nilpotent generators where (Xa )p = 0 for some p > 1. Thus the
eigenvalues of the Cartan-Killing metric provide non-trivial information on the compactness of the
Lie algebra.
Definition 4.5 (Compact Lie algebra). A Lie algebra g is compact if the Cartan-Killing
metric is negative definite, meaning that it admits a basis such that γab = −δab .
Remark 4.1. It is important to remark here that compactness of the Lie algebras is a slightly
stronger condition than compactness of the corresponding Lie group (definition 2.5). For example
U (1) = {u ∈ C : |u| = 1} is a compact Lie group, because its (single) matrix element has bounded
absolute value. The corresponding Lie algebra u(1) = iR is not compact, because it has a single
basis generator ξ1 = i with structure constant f111 = 0, implying that the Cartan-Killing metric
on u(1) vanishes. However, it can be shown that compactness of the Lie algebra always implies
compactness of the Lie group. Moreover, for simple Lie groups of dimension larger than one the
two notions of compactness are equivalent.
A consequence of working in the basis where γab = −δab is that the structure constants of the
compact Lie algebra g become completely antisymmetric11
fabc = fbca = fcab = −fbac = −facb = −fcba . (4.15)
The proof of this property is rather instructive. It expresses the structure constants in terms of
the Cartan-Killing metric via
(4.5) (4.8) γdc =−δdc
tr ( [ Ta , Tb ] Tc ) = fabd tr ( Td Tc ) = fabd γdc = −fabc . (4.16)
The cyclicity of the trace then shows that
fabc = − tr ( [ Ta , Tb ] Tc )
= − tr ( Ta Tb Tc − Tb Ta Tc )
= − tr ( Tb Tc Ta − Tc Tb Ta ) (4.17)
= − tr ( [ Tb , Tc ] Ta )
= fbca .
Together with the anti-symmetry of the structure constants in the first two indices, this establishes
the relation (4.15).
11
This is the basis which is typically adopted when studying non-abelian gauge groups in quantum field theory.
26
4.3 Casimir operator
Suppose g is compact with generators Xa in any irreducible representation. Then we can introduce
an important quadratic operator.
Definition 4.6 (Casimir operator). The (quadratic) Casimir operator of a compact Lie
algebra in this representation is given by
C := γ ab Xa Xb , (4.18)
Note that generally C ∈/ g since it is not built from commutators of the generators, but it makes
perfect sense as a matrix acting on the same vectors as Xa .
Theorem 4.7. The Casimir operator is independent of the choice of basis Xa and commutes
with all generators, i.e. [C, Xa ] = 0.
Proof. To see that it is independent of the basis let us look at a transformation Xa 7→ Xa0 = Lba Xb
as before. Then
(4.18)
C 7→ C 0 = (γ 0 )ab Xa0 Xb0 = (γ 0 )ab Lca Xc Ldb Xd (4.19)
(4.13)
= L(Lt γL)−1 Lt cd Xc Xd = γ cd Xc Xd = C.
(4.20)
Without loss of generality we may assume thus assume that we have chosen generators such that
γab = −δab . Then we find for any c,
(4.18)
[C, Xb ] = [Xa Xa , Xb ] = Xa [Xa , Xb ] + [Xa , Xb ]Xa (4.21)
(4.5) (4.15)
= Xa fabc Xc + fabc Xc Xa = 0. (4.22)
Note that the Casimir operator does not just commute with all generators, but also with the group
elements of G in the corresponding representation D. Indeed, using the exponential map (Definition
2.4) any element g ∈ G can be written as D(g) = exp(αa Xa ) and therefore satisfies
Thm. 4.7
[C, D(g)] = [C, exp(αa Xa )] = 0. (4.23)
This puts us precisely in the setting of Schur’s Lemma, Theorem 1.22, which allows us to conclude
that if D is irreducible the Casimir operator is proportional to the identity,
C = CD 1, (4.24)
where CD is a positive real number that only depends on the representation D. It can therefore
be used to label or distinguish irreducible representations.
As an example let us look at su(2) with the standard basis
In the exercises you will determine that the corresponding Cartan-Killing metric is
27
Given what we know about the total angular momentum (or spin) in quantum mechanics it should
not come as a surprise that it commutes the generators Ja . To compute the value Cj in the spin-j
representation it is sufficient to focus on a single state, say the highest-weight state |ji = |j, m = ji.
Using that
(3.8) 1 1
J +J − + J −J + = (J1 + iJ2 )(J1 − iJ2 ) + (J1 − iJ2 )(J1 + iJ2 ) = J12 + J22 , (4.28)
2 2
we find
(4.27) 1 2 1
C|ji = (J3 + J + J − + J − J + )|ji = (J32 + [J + , J − ] + 2J − J + )|ji (4.29)
2 2
(3.9) 1 (3.12) 1
= (J32 + J3 + 2J − J + )|ji = (J 2 + J3 )|ji, (4.30)
2 2 3
and therefore Cj = 12 j(j + 1). We see that the Casimir operator indeed distinguishes between the
different irreducible representations of su(2).
For larger Lie algebras, like su(N ) with N ≥ 3, this is no longer the case since irreducible represen-
tations are typically labeled by more than one parameter. Luckily in general the quadratic Casimir
operator C is not the only operator that commutes with all group elements. Further Casimir in-
variants that are polynomials in Xa of order higher than two can then be constructed. It turns
out that all Casimir invariants together do distinguish the irreducible representations.
28
5 Root systems, simple roots and the Cartan matrix
5.1 Cartan subalgebra
Central to the classification of Lie algebras and their representations is the selection of a maximal
set of commuting generators. Mathematically this is described by the Cartan subalgebra of a Lie
algebra. First we need to know what a subalgebra is.
[h, h] ⊂ h. (5.1)
Remark 5.1. Here you may rightfully object that the Lie algebras g we have encountered so far do
not contain any Hermitian generators (except the trivial 0 ∈ g). Indeed, in the case of the special
unitary and orthogonal groups, the generators ξa were antihermitian ξa† = −ξa . In order to find
Hermitian generators we have to allow taking complex linear combinations βa ξa , βa ∈ C, instead
of just real linear combinations αa ξa ∈ g, αa ∈ R. In mathematical terms such complex linear
combinations take value in the complexified Lie algebra gC = g + ig. Note that we have already
implicitly used such generators in Section 3.2 when introducing the lowering and raising operators
−i
J± = √ 2
(X1 ± iX2 ) which are elements of D(gC ) but not of D(g) (they are not antihermitian).
When dealing with representation theory and classification of Lie algebras it is typically a lot easier
to work with gC . We will do so implicitly in these lectures even when we write just g!
Fact 5.3. Although a Lie algebra g can have many Cartan subalgebras, they are all equivalent
in the sense that they can be related by a basis transformation of g. Therefore we often talk
about the Cartan subalgebra of g.
Definition 5.4 (Rank). The rank of g is the dimension of its Cartan subalgebra.
If the rank of g is r, the Cartan generators H1 , . . . , Hr form a maximal set of generators that can
be simultaneously diagonalized in any representation D : g → gl(CN ). Hence, we can find a basis
{|µ, xi} of CN such that
D(Hi )|µ, xi = µi |µ, xi, (5.2)
where x is some additional label to specify the state (in case of multiplicities).
Example 5.6 (su(2)). The rank of su(2) is 1 since no two generators commute. The Cartan
generator (spanning the Cartan subalgebra of su(2)) is typically chosen to be J3 = iX3 . The
weights of the spin-j representation of su(2) are the eigenvalues of J3 , which are j, j−1, . . . , −j.
29
5.2 Roots
For the remainder of this chapter we focus on the adjoint representation Dad of a compact Lie
algebra g on itself, given by Dad (X)|Y i ≡ X|Y i = |[X, Y ]i for X, Y ∈ g. Suppose g has rank r.
The vectors |Hi i have weight (0, . . . , 0) ∈ Rr , since for all i, j = 1, . . . , r we have
(4.4) Cartan
Hi |Hj i = |[Hi , Hj ]i = 0. (5.3)
Conversely, by the maximality criterion (Definition 5.2), any simultaneous eigenvector |Xi of the
Cartan generators with zero weight is a Cartan generator.
30
The eigenvectors |Hi i and |Eα i can then be normalized such that
Example 5.8 (Roots of su(2)). Since su(2) is three-dimensional and of rank 1 it must have
two roots. These are precisely ±1, corresponding to the generators E±1 = J± .
Definition 5.9 (Root system). A root system Φ or rank r is a finite subset of non-zero
vectors in Rr satisfying
(i) Φ spans Rr .
(ii) The only scalar multiples of α ∈ Φ that also belong to Φ are ±α.
α·β
(iii) For any α, β ∈ Φ the ratio α·α is an integer or a half-integer.
α·β
(iv) For every α, β ∈ Φ the Weyl reflection β − 2 α·α α of β in the hyperplane perpendicular
to α is also in Φ.
We will now show that
Theorem 5.10. The set of roots of a compact Lie algebra g forms a root system.
We will verify the properties one by one. Property (i) follows from the compactness of g:
Proof of property (i). Suppose the roots do not span Rr . Then there exists a nonzero vector β ∈ Rr
such that β · α = 0 for all α ∈ Φ. But that means that all vectors |Eα i are annihilated by βi Hi ,
βi Hi |Eα i = β · α|Eα i = 0, and the same is true by definition for the states |Hj i. Since those
states together form a basis of g, it follows that [βi Hi , g] = 0. Since βi Hi is represented by the
zero matrix in the adjoint representation, we have hβi Hi |βi Hi i = 0. But this is in contradiction
with the compactness of g, which requires that the Cartan-Killing metric is negative definite.
The next steps rely on the fact that to each pair ±α of roots one may associate an su(2)-
subalgebra.
Proof. If α is a root, then [Eα , E−α ] ∈ h has to be in the Cartan subalgebra h. This follows from
(5.7) with α0 = −α,
Hi |[Eα , E−α ]i = (αi − αi )|[Eα , E−α ]i = 0. (5.13)
On the other hand using (5.10),
31
= tr(Hi Eα E−α ) − tr(Hi E−α Eα )
= tr(E−α [Hi , Eα ]) (5.14)
= αi tr(E−α Eα )
= αi hEα |Eα i = αi .
Since the Hi form an orthonormal basis of the Cartan subalgebra we conclude that
This verifies the first relation of (5.12), while the second follows from (5.6).
Let us use this structure to verify that the root α specifies the generator Eα uniquely. Suppose
that there exists another eigenvector |Eα0 i orthogonal to |Eα i with the same root α . Then the
vector E−α |Eα0 i has zero weight and is thus a Cartan generator, which can be identified following
a computation similar to (5.14),
Thus the projection on the Cartan subalgebra vanishes. As a consequence |Eα0 i must be a lowest
weight state, E− |Eα0 i = 0. But
Thus the state |Eα0 i has an E3 -eigenvalue m = +1. This is in contradiction with the fact shown
in Chapter 4 that the lowest weight states of an su(2)-representation must have a negative or
zero E3 -eigenvalue (namely −j in the spin-j representation). Thus we conclude that one cannot
have two generators Eα and Eα0 corresponding to the same root. Similar arguments lead to the
following.
Proof of property (iii). Let us consider a root β different from α. From (5.11) is follows that
α·β
E3 |Eβ i = |α|−2 αi Hi |Eβ i = |Eβ i. (5.18)
α·α
But from the su(2) representation theory the eigenvalue (α · β)/(α · α) has to be half-integer.
Proof of property (ii). Notice that it is sufficient to show that if α is a root that tα for t > 1 cannot
be a root. To this end, suppose to the contrary that α0 = tα is a root. Applying property (iii) to
α and tα we find that both t and 1/t must be half-integers, so the only option is t = 2.
Let us consider the su(2)-subalgebra (5.11) associated to root α. The vector |Eα0 i has E3 -eigenvalue
2 and therefore it has a component in a spin-j representation for some integer j ≥ 2. Acting with
the lowering operator E− on this component we construct a vector with root α. But this one
is not a multiple of |Eα i, since |Eα i lives in the spin-1 representation (because E+ |Eα i = 0 and
E3 |Eα i = |Eα i). We have shown that this is impossible, finishing our proof by contradiction.
Proof of property (iv). Suppose α and β are two distinct roots. The state |Eβ i has E3 -eigenvalue
m = (α · β)/(α · α) with respect to the su(2)-representation (5.11) with m half-integer. Hence |Eβ i
can be decomposed into components transforming in various spin-j representations for j ≥ |m|. If
m ≥ 0 then acting 2m times with the lowering operator produces a non-vanishing state E− 2m |E i
β
satisfying
2m 2m 2m 2m
Hi E− |Eβ i = −2mαi E− |Eβ i + E− Hi |Eβ i = (βi − 2mαi ) E− |Eβ i. (5.19)
It therefore has root equal to β − 2mα = β − 2(α · β)/(α · α)α. Similarly, if m < 0 one may
2|m|
act 2|m| times with the raising operator to produce a non-vanishing state E+ |Eβ i with root
β + 2|m|α = β − 2(α · β)/(α · α)α.
32
We conclude that the roots of any compact Lie algebra g form a root system Φ with rank r equal
to the rank of g, i.e. the dimension of its Cartan subalgebra. Although we will not prove this in
detail, any root system Φ characterizes a unique compact Lie algebra.
Fact 5.12. Compact Lie algebras g are in 1-to-1 correspondence with root systems Φ.
is an integer. Since 0 < θαβ < π, this quantity must be an integer smaller than 4. Therefore there
are only a couple of possibilities for the angle θαβ and the length ratio
s
|α| 2(α · β) . 2(α · β)
= (5.21)
|β| β·β α·α
Definition 5.13 (Positive root). A root α = (α1 , . . . , αr ) is called positive if its first non-
zero component is positive, and negative otherwise.
Definition 5.14 (Simple root). A simple root of g is a positive root that cannot be obtained
as the sum of two positive roots.
Simple roots will be highlighted with a hat on top of the root (α̂) in the sequel.
Lemma 5.15. The simple roots satisfy the following important properties:
(i) A root system of rank r has precisely r simple roots, which form a basis of Rr .
(ii) All other roots can be obtained from successive Weyl reflections of simple roots (see
property (iv) of the root system).
(iii) If α̂ and β̂ are simple roots then α̂ − β̂ is not a root.
(iv) If α̂ and β̂ are different simple roots, then their scalar product is nonpositive, α̂ · β̂ ≤ 0.
33
Proof of Lemma 5.15(i) . We first prove that the simple roots are linearly independent, i.e. we
need to show that the only solution to ai α̂i = 0 occurs for ai = 0. To this end suppose that
another solution exists. Let us write ai = bi − ci in such a way that bi = ai and ci = 0 when ai > 0,
while bi = 0 and ci = −ai when ai < 0. Then bi α̂i = ci α̂i and therefore (bi α̂i ) · (ci α̂i ) = |bi α̂i |2 =
|bi α̂i |2 > 0. On the other hand, by expanding the inner product
We skip the proof of (ii). The two remaining parts are easily deduced as follows.
Proof of Lemma 5.15(iii) . Assume that α̂ and β̂ are simple rootsand α̂ − β̂ is a root.
Then
either
α̂ − β̂ or β̂ − α̂ is a positive root. But then either α̂ = α̂ − β̂ + β̂ or β̂ = β̂ − α̂ + α̂ can
be written as the sum of two positive roots. This contradicts the assumption that α̂ and β̂ are
simple.
Proof of Lemma 5.15(iv) . From (iii) it follows that E−α̂ |Eβ̂ i = |[E−α̂ , Eβ̂ ]i = 0 since β̂ − α̂ cannot
be a root. Hence |β̂i is a lowest-weight state in the su(2)-representation of α̂, but then its E3 -
eigenvalue α̂ · β̂/(α̂ · α̂) cannot be positive.
The information about the r simple roots α̂1 , . . . , α̂r is conveniently stored in the Cartan ma-
trix.
Definition 5.16 (Cartan matrix). Given the r simple roots of a simple, compact Lie
algebra, the r × r Cartan matrix A is defined by the matrix elements
2α̂i · α̂j
Aij = . (5.23)
|α̂j |2
Example 5.17 (Cartan matrix of su(3)). The two simple roots are
√ √
α̂1 = 21 , 23 , α̂2 = 12 , − 23 . (5.25)
34
Evaluating the scalar products of the roots gives
Substituting these values into the definition of the Cartan matrix (5.23) gives
2 −1
A= . (5.27)
−1 2
35
6 Irreducible representations of su(3) and the quark model
After SU(2), whose representations we have already studied, SU(3) is arguably the most important
(simple) compact Lie group in particle physics. The simplest application is the quark model that
describes mesons and baryons in terms of their constituent quarks. The main idea is that the three
lightest quarks, the up, down and strange quark, are related by SU(3) flavour symmetry, spanning
a 3-dimensional vector space that transforms under the defining representation (or fundamental
representation). Their anti-particles transform accoding to the complex conjugate of this represen-
tation (or anti-fundamental representation) to be defined below. Any composite states of quarks
and anti-quarks live in tensor products of these representations, which necessarily decompose into
irreducible representations of SU(3). The similarities between the masses of certain mesons and
baryons can be explained by recognizing the particles with similar masses as living in the same
irreducible representation. To understand this grouping let us start by classifying the irreducible
representations.
Remark 6.1. This approximate flavour-SU(3) symmetry is unrelated to the exact color-SU(3)
symmetry of the Standard Model gauge group SU(3) × SU(2) × U(1) that was discovered later.
A basis of Hermitian traceless 3 × 3 matrices, analogous to the Pauli matrices of su(2), is given by
the Gell-Mann matrices
0 1 0 0 −i 0 1 0 0
λ1 = 1 0 0 , λ 2 = i 0 0 , λ3 = 0 −1 0 ,
0 0 0 0 0 0 0 0 0
0 0 1 0 0 −i 0 0 0
λ4 = 0 0 0 , λ5 = 0 0 0 , λ 6 = 0 0 1 , (6.1)
1 0 0 i 0 0 0 1 0
0 0 0 1 0 0
λ7 = 0 0 −i , λ8 = √13 0 1 0 .
0 i 0 0 0 −2
They are normalized such that tr(λa λb ) = 2δab . It is customary to take the generators in the
fundamental representation to be
The structure constants fabc are completely antisymmetric and the only nonvanishing constants
(with indices in ascending order) are
√
1 3
f123 = 1, f147 = f246 = f257 = f345 = −f156 = −f367 = , f458 = f678 = . (6.3)
2 2
Any other set of Hermitian matrices X1 , . . . , X8 satisfying
therefore determines a representation of su(3) (and of SU(3) via the exponential map).
36
6.2 Cartan subalgebra and roots
By examining the structure constants (6.3) one may observe that at most two generators can be
found to commute, meaning that the Lie algebra su(3) is of rank 2. The Cartan subalgebra is
conventionally taken to be spanned by the generators H1 = iξ3 = 12 λ3 and H2 = iξ8 = 12 λ8 , which
in the defining representation are diagonal. These generators have clear physical interpretations
in the quark model, since
1 2 1 1
I3 = H1 = λ3 , Y = √ H2 = √ λ8 , Q = I3 + Y (6.5)
2 3 3 2
are respectively the isospin, hypercharge and electrical charge generators of the quarks.
The remaining generators can be combined into raising and lowering operators Eα ,
i i i
E±α1 = √ (ξ1 ± iξ2 ), E±α2 = √ (ξ4 ± iξ5 ), E±α3 = √ (ξ6 ± iξ7 ) (6.6)
2 2 2
with corresponding roots
√ √
α1 = (1, 0), α2 = ( 21 , 12 3), α3 = (− 12 , 12 3). (6.7)
6.3 Representations
Let us assume X1 , . . . , X8 are N -dimensional skew-Hermitian matrices satisfying (6.4) and let
us assume they determine an irreducible representation of su(3). The corresponding matrices
Ta = iXa are Hermitian, and H1 = iX3 = T3 and H2 = iX8 = T8 are the Cartan generators.
Recall from (5.12) that for each root αk , the operators E±αk and αik Hi form an su(2)-subalgebra
inside su(3),
[Eαk , E−αk ] = αik Hi , [αik Hi , E±αk ] = ±E±αk , (6.8)
which may or may not be reducible (note that |αk | = 1). In any case our knowledge about
su(2)-representations implies that the eigenvalues of αik Hi are half-integers. These eigenvalues
are precisely the inner products αk · µ of the roots with the weights µ = (µ1 , µ2 ) ∈ R2 of the
representation. Hence, the weights must lie on the vertices of the triangular lattice depicted in the
figure below.
If µ ∈ R2 is a weight, meaning that Hi |Ψi = µi |Ψi for some state |Ψi, then
Hi E±αk |Ψi = (E±αk Hi ± αik E±αk )|Ψi = (µi ± αik )E±αk |Ψi, (6.9)
so E±αk |Ψi is a state with weight µ ± αk provided E±αk |Ψi = 6 0. Using a similar reasoning as
for the roots in Section 5.3 it follows that if α · µ > 0 then all of µ, µ − αk , µ − 2αk , . . . ,
k
µ − 2(αk · µ)αk are also weights. In particular, the set of weights is invariant under the Weyl
reflections µ → µ − 2(αk · µ)αk , i.e. reflections in the planes orthogonal to the roots α1 , α2 and
α3 .
37
Moreover, any two states with different weights can be obtained from each other by actions of
the operators E±αk , because the collection of all states accessible from some initial state span an
invariant subspace, which has to be the full vector space because of irreducibility. Therefore the
weights must differ by integer multiples of αk . Combining these observations we conclude that the
set of weights necessarily occupy a convex polygon that is invariant under Weyl reflections (thus
either a hexagon or a triangle), like in the following figure.
in Example 5.17.
Definition 6.1 (Highest weight). A state |Ψi that is annihilated by the raising operators
Eα̂1 and Eα̂2 is called a highest weight state. Its weight is called the highest weight.
Any irreducible representation can be shown to have a unique highest weight with multiplic-
ity one. In the weight diagram above it corresponds to the right-most weight (i.e. largest H1 -
eigenvalue).
The irreducible representations of su(3) satisfy the following properties, which we do not prove
here.
• Each label (p, q), p, q = 0, 1, 2, . . ., corresponds to a unique irreducible representation of su(3).
• The set of weights of the representation with label (p, q) can be determined as follows. By
inverting (6.11) the highest weight is
µ = 21 (p + q), 2√
1
3
(p − q) . (6.12)
The Weyl reflections of µ determine the corners of a convex polygon P , that is either a
hexagon or a triangle. The complete set of weights then corresponds to those vectors of the
form µ + n1 α̂1 + n2 α̂2 with n1 , n2 ∈ Z that are contained in P .
38
• The multiplicities of the weights (i.e. the number of linearly independent states with that
weight) can be deduced from the diagram. The weights occur in nested “rings” that are
either hexagonal (the outer ones) or triangular (the inner ones). The weights on the outer
ring have multiplicity 1. Each time one moves inwards from a hexagonal ring to the next
ring the multiplicity increases by 1. Inside triangular rings the multiplicities do not change
anymore.
Examples of weight diagrams with small labels are shown below. The highest weights are circled
(in blue), and multiplicities are indicated on the weights when they are larger than 1.
As one may verify for these examples, the dimension of the representation is generally given
by
1
N = (p + 1)(q + 1)(p + q + 2). (6.13)
2
For p 6= q the two irreducible representations (p, q) and (q, p) are not equivalent but related by
complex conjugation. To see this, let Ta be the generators of the irreducible (p, q)-representation.
Since [Ta , Tb ] = ifabc Tc with real structure constants fabc we have that
showing that the matrices −Ta∗ also generate a representation, which is automatically irreducible
too. All eigenvalues of H1 = T3 and H2 = T8 have changed sign, thus changing the set of weights
to those of the (q, p) representation. The irreducible representations are often denoted by their
dimension (6.13) in bold face N instead of their label (p, q), where if p > q the representation N
corresponds to (p, q) and N̄ to (q, p).
Let us discuss the interpretation of some of these representations in the quark model.
• 3, (p, q) = (1, 0): This is the defining representation or fundamental representation (6.2) of
su(3) as can be easily checked by determining the eigenvalues of H1 = 21 λ3 and H2 = 12 λ8 .
The three basis states, |ui, |di, |si correspond to the up, down and strange quark respectively.
Their arrangement determines the values of the isospin and charge via (6.5).
• 3̄, (p, q) = (0, 1): the complex conjugate representation or anti-fundamental representation.
Since the corresponding states have opposite quantum numbers as compared to the funda-
mental representation, it stands to reason that they correspond to the anti-quarks |ūi, |d̄i, |s̄i.
39
• 8, (p, q) = (1, 1): this is the octet or adjoint representation, since the non-zero weights are
exactly the roots of su(3). It appears in the decomposition 3 ⊗ 3̄ = 8 ⊕ 1 of the tensor
product of the fundamental and anti-fundamental representation, and therefore describes
bound states of a quark with an anti-quark, which are mesons. It also appears in the
decomposition 3 ⊗ 3 ⊗ 3 = 10 ⊕ 8 ⊕ 8 ⊕ 1 of the tensor product of three fundamental
representations, describing bound states of three quarks, which are baryons. The baryons in
the adjoint representation have spin 1/2.
• 10, (p, q) = (3, 0): the decuplet representation also appears in the decomposition of 3 ⊗ 3 ⊗ 3
and describes the spin-3/2 baryons.
We close this section with several remarks concerning the quark model.
• The existence of quarks was unknown in the early sixties. It is on the basis of the nice
grouping (the Eightfold way) of the discovered mesons and baryons into octets and decu-
plets that Gell-Mann and Zweig (independently in 1964) postulated the existence of more
fundamental particles living in the fundamental representation of su(3) (dubbed “quarks”
by Gell-Mann). At that time the spin-3/2 baryon Ω− had not been detected yet, but on
the basis of symmetry its approximate mass and quantum numbers were predicted and soon
confirmed.
40
• The su(3) flavor symmetry is only an approximate symmetry of the standard model (contrary
to the su(3) color symmetry of QCD) because, among other reasons, the quark masses are
quite different (mu = 2.3 MeV, md = 4.8 MeV, ms = 95 MeV) although a lot lighter than
the three other quarks (mc = 1270 MeV, mb = 4 GeV, mt = 172 GeV). Since the difference
in masses between the up and down quark is much smaller than between the up and strange
quark, one would expect that the symmetry relating the up and down quarks is more accurate.
This smaller symmetry corresponds precisely to su(2)-subrepresentation associated to the
root α1 = (1, 0), which has the isospin I3 as Cartan generator. Indeed, the masses of the
mesons and baryons in the corresponding su(2)-multiplets (horizontal lines in the diagrams)
are quite close, e.g. the proton and neutron among the spin-1/2 baryons.
41
7 Classification of compact, simple Lie algebras
7.1 Simple Lie algebras
Recall (Definition 4.5) that a Lie algebra is called compact if its Cartan-Killing metric is negative
definite. An important result, that we will not prove, is that any compact Lie algebra g decomposes
as a direct sum of compact, simple Lie algebras
g = g1 ⊕ · · · ⊕ g2 . (7.1)
Hence, in order to classify compact Lie algebras it is sufficient to classify compact, simple Lie
algebras and then consider all possible compositions. In order to introduce the concept of simple
Lie algebras, we need a few definitions.
Recall that a subspace h ⊂ g is called a subalgebra of g if [h, h] ⊂ h. A stronger condition is the
following.
[h, g] ⊂ h. (7.2)
For a Lie algebra g with a basis X1 , . . . , Xn of generators and corresponding structure constants
fabc the k-dimensional subspace spanned by X1 , . . . , Xk is a subalgebra precisely when
fijα = 0 for 1 ≤ i, j ≤ k, k < α ≤ n, (7.3)
while it is invariant if in addition
fiβα = 0 for 1 ≤ i ≤ k, k < α, β ≤ n. (7.4)
Definition 7.2 (Simple Lie algebra). A Lie algebra g is simple if it has no invariant
subalgebras (other than {0} and g itself).
Equivalently, a Lie algebra is simple precisely when its adjoint representation is irreducible. Indeed,
h is an invariant subalgebra precisely when X|Y i = |[X, Y ]i ∈ h for any X ∈ g and Y ∈ h. This is
precisely the condition for h ⊂ g to be an invariant subspace for the adjoint representation.
Definition 7.3 (Dynkin diagram). The Dynkin diagram describing a Lie algebra is con-
structed as follows:
- simple roots are represented by nodes •
- max{|Aij |, |Aji |} gives the number of lines connecting the circles representing the roots
α̂i and α̂j .
- if the roots connected by lines have different length, one adds an arrow pointing towards
42
the shorter root, i.e. if Aij = −2 or Aij = −3 then the arrow points from α̂i to α̂j (hint:
think of the arrow as a “greater than” symbol >)
In the exercises it is shown that the Dynkin diagram of a simple compact Lie algebra is connected.
Otherwise the Dynkin diagram consists of a number of connected components corresponding to
the simple compact Lie algebras in to which it decomposes as a direct sum.
From the definition it follows that there is only one Dynkin diagram of rank 1 and that there are
four diagrams of rank 2:
Four of these correspond to Lie algebras that are familiar from Chapter 2, while G2 is the first
exceptional Lie group that we encounter.
If we go to higher rank, then not every graph built from three or more nodes connected by (multiple)
links corresponds to the Dynkin diagram of a compact Lie algebra. Indeed, there have to exist r
linearly independent vectors α̂1 , . . . , α̂r , whose inner products give rise to the links in the diagram.
Luckily this is the only criterion that a diagram has to satisfy to qualify as a Dynkin diagram. As
we will see now it puts severe restrictions on the types of graphs that can appear.
Lemma 7.4 (Dynkin diagrams of rank 3). The only Dynkin diagrams of rank 3 are
Proof. This results from the fact that the sum of angles between 3 linearly independent vectors
must be less than 360◦ . Computing the angle in the first diagram, one has 120◦ +120◦ +90◦ = 330◦ .
The analogous computation for the second diagram yields 120◦ + 135◦ + 90◦ = 345◦ . This also
implies that the diagrams shown below are not valid, since the angles add up to 360◦ :
Adding even more lines of course even increases the sum of angles.
This has important consequences for diagrams of higher rank, because of the following.
Lemma 7.5 (Subsets of a Dynkin diagram). A connected subset of nodes from a Dynkin
diagram together with the links connecting them is again a Dynkin diagram.
43
Proof. Taking a subset of simple roots preserves the corresponding lengths and angles, and the
vectors are still linearly independent.
In particular, a triple line cannot occur in diagrams with three or more nodes, because otherwise
there would be a subset of rank 3 with a triple line. Hence, we have:
Lemma 7.6 (No triple lines). G2 is the only Dynkin diagram with a triple line.
Lemma 7.7 (Contraction of a single line). If a Dynkin diagram contains two nodes
connected by a single line, then shrinking the line and merging the two nodes results again in
a Dynkin diagram.
Proof. Suppose that α̂ and β̂ are connected by a single line. We claim that replacing the pair of
vectors α̂ and β̂ by the single vector α̂ + β̂ preserves the necessary properties. By Lemma 7.4 and
Lemma 7.5 each other node γ̂ is connected to at most one of α̂ or β̂ (otherwise there would be a
triangle subdiagram). If γ̂ was connected to α̂ then γ̂ · (α̂ + β̂) = γ̂ · α̂, while if γ̂ was connected
to β̂ then γ̂ · (α̂ + β̂) = γ̂ · β̂. Since also |α̂ + β̂|2 = |α̂|2 = |β̂|2 , all lines other than the contracted
one are preserved. The resulting set of vectors is also still linearly independent.
Proof. By Lemma 7.4 a loop must always contain a single edge, which can be contracted by Lemma
7.7. Hence, every loop can be contracted to a triangle, in contradiction with Lemma 7.4.
Lemma 7.9 (At most one double line). A Dynkin diagram can contain at most one
double line.
Proof. If there are two or more double lines, then one can always contract single lines until two
double lines becomes adjacent. But this is not allowed by Lemma 7.4.
At this point we know that Dynkin diagrams always have the shape of a tree with at most one of
its links being a double line. To further restrict branching of the tree the following result is quite
useful.
Lemma 7.10 (Merging two single lines). If a Dynkin diagram has a node with (at least)
two dangling single lines, then merging them into a double line gives another Dynkin diagram.
In other words, the following is a valid operation on Dynkin diagrams (with the gray blob
representing the remaining part of the diagram):
Proof. Replacing α̂ and β̂ by α̂+ β̂, linear independence of the resulting set of vectors is automatic.
Hence it remains to check the inner products between γ̂ and α̂ + β̂. Using that |α̂|2 = |β̂|2 = |γ̂|2 ,
44
α̂ · β̂ = 0, and 2α̂ · γ̂/|γ̂|2 = 2β̂ · γ̂/|γ̂|2 = −1, it follows indeed that
Lemma 7.11 (At most one junction or double line). A Dynkin diagram contains one
double line, or one junction, or neither. A junction is a node with more than two neighbours.
Any junction must have the form
,
i.e. three neighbours connected by single lines.
Proof. By repeated contraction (Lemma 7.7) and merging (Lemma 7.10) any diagram with a
different type of junction (with more than three neighbours or involving double lines) can be
reduced to a diagram with two adjacent double lines, which is not allowed. The same is true if a
diagram has two junctions, or both a junction and a double line.
The remaining step is to get rid of some remaining diagrams on a case-by-case basis, by showing
that they cannot satisfy linear independence.
Lemma 7.12 (Diagrams failing linear independence). The following diagrams cannot
be realized by linear independent vectors:
Proof. The labels µ1 , . . . , µr indicated on the diagram are such that rj=1 µj α̂j = 0. To see this
P
one may straightforwardly compute | j=1 µj α̂j |2 = ri,j=1 µi µj α̂i · α̂j = 0 using the inner products
Pr P
45
α̂i · α̂j as prescribed by the diagram. Such a linear relation implies that the vectors α̂j are not
linearly independent.
This is all we need to deduce the final classification theorem. In general, connected Dynkin
diagrams are denoted by a capital letter denoting the family and a subscript denoting the rank,
e.g. B2 for the Dynkin diagram of the rank-2 Lie algebra so(5). The same notation is also
sometimes confusingly used to specify the corresponding Lie algebra and/or Lie group.
Theorem 7.13 (Classification of simple, compact Lie algebras). Any simple, compact
Lie algebra has a Dynkin diagram equal to a member of one of the four infinite families An ,
Bn , Cn , Dn , n ≥ 1, or to one of the exceptional Lie algebras G2 , F4 , E6 , E7 , or E8 . In other
words its Dynkin diagram appears in the following list:
As one may note by looking at the infinite families for small n, there are some overlaps between
the families. Since there is only one rank-1 diagram it makes sense to label it by A1 , and require
n ≥ 2 for the remaining families. However, B2 = C2 and D3 = A3 , while D2 consisting of two
disconnected nodes does not correspond to a simple Lie algebra. Hence, to make the classification
unique one may require n ≥ 1 for An , n ≥ 2 for Bn , n ≥ 3 for Cn , and n ≥ 4 for Dn .
Finally, going back to the common Lie groups that we identified in Chapter 6 it is possible to
obtain the following identification:
46
Cartan label Lie algebra dimension rank
An su(n + 1) (n + 1)2 − 1 n≥1
Bn so(2n + 1) (2n + 1)n n≥2
Cn sp(n) (2n + 1)n n≥3
Dn so(2n) (2n − 1)n n≥4
G2 14 2
F4 52 4
E6 78 6
E7 133 7
E8 248 8
From the overlaps described above it follows that we have the following isomorphisms between Lie
algebras,
Cartan label Lie algebra
A1 , B1 , C1 so(3) ' su(2) ' sp(1)
, (7.6)
B2 , C2 so(5) ' sp(2)
A3 , D3 so(6) ' su(4)
as well as the nonsimple so(4) = su(2) ⊕ su(2) (see exercises).
which forms the basis of the famous Georgi-Glashow GUT based on SU(5).
47