Representations of Lorentz and Poincaré groups

Joseph Maciejko
CMITP and Department of Physics, Stanford University, Stanford, California, 94305

I. LORENTZ GROUP

We consider first the Lorentz group O(1, 3) with infinitesimal generators J µν and the
associated Lie algebra given by

[J µν , J ρσ ] = i(g νρ J µσ − g µρ J νσ − g νσ J µρ + g µσ J νρ ) (1)

In quantum field theory, we actually consider a subgroup of O(1, 3), the proper orthochronous
or restricted Lorentz group SO+ (1, 3) = {Λ ∈ O(1, 3)| det Λ = 1 and Λ0 0 ≥ 0}, which ex-
cludes parity and time-reversal transformations that are thus considered as separate, discrete
operations P and T . A generic element Λ of the Lorentz group is given by exponentiating
the generators together with the parameters1 of the transformation, Λ = exp(−iωµν J µν /2).
The Lorentz group has both finite-dimensional and infinite-dimensional representations.
However, it is non-compact, therefore its finite-dimensional representations are not unitary
(the generators are not Hermitian). The generators of the infinite-dimensional representa-
tions can be chosen to be Hermitian.

A. Finite-dimensional representations

We first study the finite-dimensional representations of SO+ (1, 3). These representations
act on finite-dimensional vector spaces (the base space). Elements of these vector spaces are
said to transform according to the given representation.
Trivial representation. In the trivial representation, we have the one-dimensional
representation J µν = 0. Hence any Lorentz transformation Λ is represented by 1. This
representation acts on a one-dimensional vector space whose elements are 1-component ob-
jects called Lorentz scalars. One can thus say that the trivial representation implements a
Λ
Lorentz transformation Λ on a scalar φ by the rule φ → 1 · φ = φ. The trivial representation
is denoted by (0, 0).
 2

Vector representation. In the vector or 4-vector representation, each generator J µν is

represented by a 4 × 4 matrix (J µν )ρ σ , which acts on a four-dimensional vector space whose
elements are 4-component objects called Lorentz four-vectors. The vector representation is
thus four-dimensional. The explicit form of the matrices is (J µν )ρ σ = i(g µρ δ ν σ − g νρ δ µ σ ).
Λ
A Lorentz transformation Λ is now implemented on a 4-vector V ρ by the rule V ρ →
µν /2
(e−iωµν J )ρ σ V σ where the J µν in the argument of the exponential is now a matrix, so
that we exponentiate a matrix to get another matrix which multiplies the 4-vector to give
the transformed 4-vector. Since the elements of the Lorentz group SO+ (1, 3) are actually
matrices2 , it is easily seen that the matrices of the vector representation are nothing but
µν /2
the SO+ (1, 3) matrices themselves, i.e. (e−iωµν J )ρ σ = Λρ σ where Λρ σ is the usual Lorentz
transformation matrix. A representation of a matrix group which is given by the elements
(matrices) of the group itself is called the fundamental representation. Hence the vector
representation is the fundamental representation of the Lorentz group. The vector represen-
tation is denoted by ( 21 , 12 ).
Tensor representations. Tensor representations are given by the direct (tensor) prod-
uct of copies of the vector representation. They act on the set of tensors of a given rank,
which is indeed a linear vector space. These tensors are called Lorentz 4-tensors. For exam-
ple, consider (2, 0) tensors, that is, tensors with two contravariant (upper) indices T ρσ . An
element of the Lorentz group Λ will be represented by a 16 × 16 matrix Λρσ ρ0 σ0 ≡ Λρ ρ0 Λσ σ0 ,
which is clearly seen to be the direct product of two 4 × 4 matrices of the vector repre-
Λ 0 0 0 0
sentation. As a result, a tensor T ρσ will transform as T ρσ → Λρσ ρ0 σ0 T ρ σ = Λρ ρ0 Λσ σ0 T ρ σ ,
which reproduces the usual Lorentz transformation law for a tensor. Representations for
higher-rank tensors are constructed in the same way, with additional copies of the vector
representation in the direct product.
Adjoint representation. The Lie algebra of Eq. (1) can be written as [J µν , J ρσ ] =
if µνρσ αβ J αβ where f µνρσ αβ are the structure constants. If we define 16 × 16 matri-
ces (J µν )ρσ αβ ≡ −if µνρσ αβ , it is possible to show from the Jacobi identity [A, [B, C]] +
[B, [C, A]] + [C, [A, B]] = 0 satisfied by any matrices A, B, C that the matrices (J µν )ρσ αβ
satisfy the Lorentz algebra. We have thus constructed a representation of the generators
of the Lorentz group from the structure constants of the group: this is called the adjoint
representation. Its dimension is the number of generators of the group. However, since the
generators J µν of the Lorentz group are antisymmetric J µν = −J νµ , there are actually only
 3

4 · (4 − 1)/2 = 6 independent3 generators (corresponding to the 3 rotations and 3 boosts)

so that the adjoint representation of the Lorentz group is six-dimensional. In other words,
we could write the Lie algebra as [J˜a , J˜b ] = if ab c J˜c where a, b, c = 1, . . . , 6, by defining
(J˜1 , J˜2 , J˜3 ) = J where J i ≡ 12 ijk J jk are the generators of rotations, and (J˜4 , J˜5 , J˜6 ) = K
where K i ≡ J 0i are the generators of boosts. With this notation, the adjoint representation
is composed of 6 × 6 matrices (J˜a )b c ≡ −if ab c .
Spinorial representations. Spinorial representations of the Lie group SO(n, m) are
given by representations of the double cover4 of SO(n, m) called the spin group Spin(n, m).
It is possible to show that the double cover of the restricted Lorentz group SO+ (1, 3) is
Spin+ (1, 3) = SL(2, C) where SL(2, C) is the set of complex 2 × 2 matrices with unit deter-
minant. We can thus construct a first spinorial representation of SO+ (1, 3) by the matrices
M of SL(2, C) themselves, i.e. the fundamental representation of SL(2, C) is a spinorial
representation of SO+ (1, 3) called ( 21 , 0). The vector space upon which this representation
acts is the set of two-component objects (complex 2 × 1 column vectors) called spinors,
or more precisely left-handed Weyl spinors for the fundamental representation. If we take
the complex conjugated matrices M ∗ , this defines another inequivalent representation of
SL(2, C) called the anti-fundamental representation. We can thus construct a second spino-
rial representation of SO+ (1, 3) called (0, 21 ). Objects that are acted upon in this represen-
tation are now called right-handed Weyl spinors. These two representations are thus seen
to be 2-dimensional. By taking the direct sum ( 12 , 0) ⊕ (0, 12 ) of the two representations,
we obtain a 4-dimensional (reducible) representation of the Lorentz group which acts upon
four-component objects called Dirac spinors.
In nonrelativistic quantum mechanics, invariance under Lorentz boosts is not required
so that only SO(3), the rotational subgroup of SO+ (1, 3), is relevant. In this case, the
double cover of SO(3) is Spin(3) = SU(2) so that the relevant spinorial representations are
representations of SU(2), leading to the usual spinors of quantum mechanics. Since SO(3)
and SU(2) are compact5 Lie groups, their generators can be chosen to be Hermitian.
Summary of finite-dimensional representations:
1. For a scalar φ, element of a 1-dimensional vector space R, we have

Λ
φ → ΛS φ = φ

where ΛS = exp(−iωµν J µν /2) = 1 with J µν = 0 are trivial 1 × 1 matrices.

 4

2. For a vector V ρ , element of a 4-dimensional (ρ = 0, 1, 2, 3) vector space R4 , we have

Λ
V ρ → (ΛV )ρ σ V σ

where ΛV = exp(−iωµν J µν /2) with (J µν )ρ σ = i(g µρ δ ν σ − g νρ δ µ σ ) are 4 × 4 matrices.

3. For a left-handed Weyl spinor ψα , element of a 2-dimensional (α = 1, 2) vector space
C2 , we have
Λ
ψα → (ΛL )αβ ψβ

where ΛL = exp(−iωµν S µν /2) with S µν = J µν are 2 × 2 matrices such that S ij ≡ 21 ijk σ k

and S 0i ≡ − 2i σ i .
4. For a right-handed Weyl spinor6 ψα , we have similarly

Λ
ψα → (ΛR )αβ ψβ

where ΛR = exp(−iωµν S µν /2) with S ij ≡ 21 ijk σ k again but now S 0i ≡ 2i σ i .

5. For a Dirac spinor Ψa , element of a 4-dimensional (a = 1, 2, 3, 4) vector space C4 , we
have
Λ
Ψa → (ΛD )ab Ψb (2)

where ΛD = exp(−iωµν S µν /2) but now the S µν are 4 × 4 matrices given by S µν = 4i [γ µ , γ ν ].

They are actually the direct sum of the 2 × 2 S µν matrices for the left-handed and right-
handed Weyl spinors:
   
k i
1 σ 0 1 σ 0 
S ij = ijk   ≡ 1 ijk Σk ; S 0i = −  ,
2 0 σk 2 2 0 −σ i
 
ψL
so that a Dirac spinor (also called bispinor ) transforms like   where ψL/R is a
ψR
left/right-handed Weyl spinor.

B. Infinite-dimensional representations

Field representations. So far we have been dealing only with finite-dimensional rep-
resentations, that acted on finite-dimensional vector spaces whose elements were scalars,
vectors, tensors, spinors, giving us respectively the scalar, vector, tensor and spinorial rep-
resentations. These objects are however only ‘constants’: in quantum field theory, we deal
 5

with fields, which are functions of spacetime. Therefore, a generic multicomponent field Φa
will not only transform as
Λ
Φa → Mab (Λ)Φb

where M (Λ) corresponds to the ΛS,V,L,R,D matrices of the finite-dimensional representations,

but since it is a function of coordinates Φa (x) and these coordinates (being Lorentz 4-vectors)
Λ
are affected by the Lorentz transformations as xµ → Λµ ν xν where Λµ ν ≡ (ΛV )µ ν , then we
will actually have
Λ
Φa (x) → Mab (Λ)Φb (Λ−1 x) (3)

In other words, the contours of the function Φa (x) are ‘boosted’ as well by the Lorentz
transformation7 . It is not difficult (by considering an infinitesimal Lorentz transformation,
for instance) to check that this transformation of coordinates in a generic field ψ(x) can be
implemented by
µν /2
ψ(Λ−1 x) = e−iωµν L ψ(x)

where now Lµν is a differential operator defined as

Lµν ≡ i(xµ ∂ ν − xν ∂ µ ) (4)

i.e. we implement the transformation on coordinates by a (exponential-resummed) Taylor

expansion. It can be checked that the Lµν satisfy the Lorentz algebra Eq. (1). Since Lµν acts
on a space of functions ψ(x) which is an infinite-dimensional vector space, it corresponds to
an infinite-dimensional representation of the Lorentz algebra.
So we can put together both contributions to the transformation of the field Φa (x) under
a Lorentz transformation, and we obtain

Λ µν /2 µν /2
Φa (x) → (e−iωµν S )ab e−iωµν L Φb (x) (5)

where for simplicity, we will henceforth denote generically by S µν all the finite-dimensional
representations of the infinitesimal Lorentz generators J µν encountered in the previous sec-
tion (S µν , J νν , ...) such that

• for a scalar field, S µν = 0;

• for a vector field, (S µν )ρ σ ≡ (J µν )ρ σ = i(g µρ δ ν σ − g νρ δ µ σ );

• for a spinor (Weyl or Dirac) field, S µν is given in 3.4.5. of the previous section.
 6

Now, since all the S µν matrices are finite-dimensional and constant, we can put the two
factors of Eq. (5) into a single exponential, and write

Λ µν /2
Φa (x) → (e−iωµν J )ab Φb (x)

where
J µν = Lµν + S µν (6)

is now an infinite-dimensional representation of the Lie algebra of the Lorentz group. While
the S µν have different forms corresponding to the different types of fields (scalars, vectors,
spinors, ...), Lµν always has the form given in Eq. (4). The representations of Eq. (6) act
on the space of fields Φa (x) and are called the field representations for the generators of the
Lorentz group.
Representations on 1-particle Hilbert space. So now we have representations of the
Lorentz group on fields, which are multicomponent functions of spacetime Φa (x). However,
there has actually been no mention of whether these fields are classical or quantum. In
other words, we use the representation theory of the Lorentz group to construct a Lorentz-
invariant Lagrangian for a given set of fields. However, this is independent of whether or not
we quantize the resulting field theory. If we quantize the theory, then we can construct still
other representations of the Lorentz group, formed by the set of unitary operators acting on
the quantum states belonging to the 1-particle Hilbert space of our quantum field theory.
Indeed, a famous theorem by Wigner, at the heart of the application of group theory to
quantum mechanics, asserts that the symmetry group of the Hamiltonian (or the Lagrangian,
better suited for a relativistic theory) can be represented by a group of unitary8 symmetry
operators acting on the Hilbert state space. Correspondingly, the algebra of infinitesimal
generators (in the case of a continuous symmetry) is represented by an algebra of Hermitian
operators. In the case of the Lorentz group, to each element of the Lorentz group Λ we assign
a unitary operator U (Λ) which implements this transformation on the 1-particle states of
p
the free field theory. For example, consider a 1-fermion state |p, si ≡ 2Ep as†p |0i with

momentum p and spin s. We require that9

2EΛp as†
p
U (Λ)|p, si ≡ |Λp, si = Λp |0i

Since the vacuum is Lorentz invariant U (Λ)|0i = |0i, this implies that the cre-
 7

ation/annihilation operators transform as

s s
−1 EΛp s† EΛp s
U (Λ)as†
pU (Λ) = a ; U (Λ)asp U −1 (Λ) = a .
Ep Λp Ep Λp

To write down an algebra of Hermitian operators to represent the abstract Lorentz algebra
Eq. (1), it is easier to work with the equivalent six generators J˜a defined in the section on
the adjoint representation, namely J i ≡ 12 ijk J jk and K i ≡ J 0i . The corresponding algebra
of Hermitian operators is defined by the following commutation relations, which all follow
directly from Eq. (1):

[J i , J j ] = iijk J k ; [J i , K k ] = iijk K k ; [K i , K j ] = −iijk J k .

If we define the combinations J+ = 12 (J + iK) and J− = 12 (J − iK), it is straightforward to

show that both sets (J±1 , J±2 , J±3 ) satisfy an independent su(2) algebra,

[J±i , J±j ] = iijk J±k ; [J±i , J∓j ] = 0, (7)

so the Lie algebra of the Lorentz group is isomorphic to su(2) × su(2).

The operators J2+ and J2− commute with all the generators: such operators are called
Casimir operators or Casimir invariants. As follows from the su(2) commutation relations,
they have eigenvalues j± (j± +1) with j± = 0, 21 , 1, 32 , . . . If we now consider finite-dimensional
representations of the two su(2) algebras of Eq. (7), i.e. the usual spinorial representations
of quantum mechanics, we see that we recover the finite-dimensional representations of the
Lorentz algebra, and we also see that they can be labeled by the pair of integers/half-integers
(j+ , j− ). This explains the notation we have been using: (0, 0) for scalars, ( 21 , 21 ) for vectors,
( 21 , 0) and (0, 12 ) for Weyl spinors, ( 12 , 0) ⊕ (0, 21 ) for Dirac spinors.

II. POINCARÉ GROUP

In the Poincaré group, one adds spacetime translations to the set of (homogeneous)
Lorentz transformations:
(Λ,a)
xµ −→ Λµ ν xν + aµ

where aµ ∈ R4 . The Poincaré group is thus also called the inhomogeneous Lorentz group.
The proper orthochronous or restricted Poincaré group ISO+ (1, 3) is a subgroup of the full
 8

Poincaré group that contains only proper orthochronous Lorentz transformations. It is given
by a semidirect product,
ISO+ (1, 3) ∼
= SO+ (1, 3) o R4

where the translation subgroup R4 is a normal subgroup10 . To generate translations, one

has to add a new generator P µ to the Lorentz algebra Eq. (1) to form the Poincaré algebra
which reads
[J µν , J ρσ ] = i(g νρ J µσ − g µρ J νσ − g νσ J µρ + g µσ J νρ )

[P µ , P ν ] = 0

[P µ , J ρσ ] = i(g µρ P σ − g µσ P ρ )

A correct relativistic quantum field theory has to be not only Lorentz covariant, but Poincaré
covariant as well.
Field representations. The field representations of the Poincaré algebra are

J µν = Lµν + S µν = i(xµ ∂ ν − xν ∂ µ ) + S µν

as previously, and
P µ = i∂ µ

for the generator of translations.

Representations on 1-particle Hilbert space. On 1-particle Hilbert space, the gen-
erators of the above Poincaré algebra are represented by Hermitian operators (hence the
elements of the Poincaré group by unitary operators). In order to do that, it is easier to
work with the equivalent set of generators J i , K i , P i and P 0 . As explained before, J i and
K i have a clearer physical meaning than the abstract generators J µν ; J i are the generators
of rotations and K i are the generators of boosts. In addition, P i are the generators of
space translations and P 0 is the generator of time translations. It follows that P will be
represented by the total momentum operator P, P 0 will be represented by the Hamiltonian
operator H, and J will be represented by an angular momentum operator J.
A convenient way to label representations consisting of operators acting on Hilbert space
is by the eigenvalues of Casimir operators (or Casimir invariants), which are operators that
commute with all the generators of the algebra. In the case of the Poincaré algebra, there
are two Casimir operators P µ Pµ and W µ Wµ where
1
W µ = − µνρσ Jνρ Pσ
2
 9

is the Pauli-Lubanski pseudovector. We now consider two cases, depending on whether the
particle in the 1-particle state |pi ∼ a†p |0i is massive or massless.

• Massive representations: on a massive 1-particle state, P µ Pµ = P 2 has eigenvalue

p2 = Ep2 − |p|2 = m2 where m is the mass of the particle, and W µ Wµ has eigenvalue
−m2 j(j + 1) with j the spin of the particle. The 1-particle states are actually |p, ji =
p
2Ep aj†
p |0i in this case. The representations can be labeled by two labels (m, j) where

j = 0, 21 , 1, 23 , . . . Another way to see this is that if we bring P µ to the form (m, 0) by a

Lorentz transformation, P µ is unchanged by arbitrary spatial rotations. This subgroup
of the Poincaré group is called the little group. In this case, it is the group of rotations
SO(3); however, to allow for spinorial representations, we choose the little group to be
the double cover SU(2). It is well-known that the irreducible representations of SU(2)
are labeled by the spin quantum number j. Thus to completely specify a massive
representation of the Poincaré group, we have to provide two labels: the mass m, and
the spin j which indexes the representation of the little group SU(2), also called the
small representation. Each spin-j representation is (2j + 1)-dimensional; therefore a
massive particle of spin j has 2j + 1 degrees of freedom.

• Massless representations: on a massless 1-particle state, obviously P 2 has eigenvalue

0. In this case, P µ can be reduced to P µ = (ω, 0, 0, ω) where ω is the eigenvalue
of P 0 . The transformations that leave this P µ invariant are rotations in the (x, y)
plane, that is, elements of SO(2). We choose the little group to be the double cover
U(1) which has only one generator. Its 1-dimensional representations are labeled by
a single index h which corresponds to the helicity of the particle. In principle, each
massless particle has thus only one degree of freedom. However, for massless particles
which participate only in parity-conserving interactions such as the photon (h = 1)
and graviton (h = 2), we group the +h and −h representations together since parity
transforms positive to negative helicity and vice-versa. Thus the photon and graviton
have two degrees of freedom.

1 The antisymmetric tensor of parameters ωµν is given from the usual rotation θ = (θ1 , θ2 , θ3 )
and boost β = (β1 , β2 , β3 ) parameters by ωij = ijk θk and ω0i = βi .
 10

2 The Lie groups O(n, m), O(n) and their various subgroups SO(n, m), SO(n) and others are
actually all subgroups of a matrix group, the general linear group GL(n, R) which is nothing
but the set of real invertible n × n matrices. For example, O(1, 3) is a subgroup of GL(4, R).
3 The number of independent components of a second-rank tensor is n(n + 1)/2 for a symmetric
tensor S µν = S νµ and n(n − 1)/2 for an antisymmetric tensor Aµν = −Aνµ , where n is the
dimension, i.e. µ, ν = 1, . . . , n.
4 For n > 2, Spin(n, m) is simply connected so that it is also the universal covering group of
SO(n, m).
5 For example, SU(2) is homeomorphic to the 3-sphere S 3 = {(x, y, z, w) ∈ R4 |x2 + y 2 + z 2 + w2 =
1} which is a compact manifold; SO(3) is homeomorphic to the real projective space RP 3 which
is also a compact manifold (think of RP 3 as a cube [0, 1]×[0, 1]×[0, 1] with antiperiodic boundary
conditions in all three directions). In general, RP n ∼
= S n /Z2 where Z2 is the cyclic group of
order 2 and S n is the double cover of RP n . One can thus write SO(3) ∼
= SU(2)/Z2 which makes
clear the meaning of ‘double cover’: SU(2) ‘covers’ the operations of SO(3) twice, so that we
have to remove the distinction between two different elements of SU(2) (such as R(2π) = −1
and R(4π) = 1) which give the same element of SO(3). To do that, we ‘divide’ by Z2 = {1, −1}.
In the same spirit we can write SO+ (1, 3) ∼
= SL(2, C)/Z2 . In analogy with RP n , we see that
viewed as a manifold, SO+ (1, 3) is a projective space and we write SO+ (1, 3) ∼
= PSL(2, C) where
PSL(2, C) = SL(2, C)/Z2 is the projective special linear group.
6 A technical detail following up on our discussion of the representations of SL(2, C). A left-handed
Weyl spinor ψα transforms according to the matrices M of the fundamental representation of
SL(2, C) as ψα = Mα β ψβ . However, a right-handed Weyl spinor is usually rather denoted by χ̄α̇
(dotted spinor) and will transform according to the complex conjugated matrices M ∗ , which
belong to the (inequivalent) anti-fundamental representation of SL(2, C), as χ̄α̇ = χ̄β̇ (M ∗−1 )β̇ α̇ .
7 This is the usual Wigner’s convention for symmetry operators in quantum mechanics: to each
R
coordinate transformation R that acts on the coordinates as x → Rx, we assign a symmetry
R
operator PR that acts on quantum states as |ψi → PR |ψi such that hr|PR |ψi = hR−1 r|ψi, or
R
equivalently on functions as ψ(x) → PR ψ(x) ≡ ψ(R−1 x).
8 If we include time-reversal symmetry, then we must allow for antiunitary operators as well.
However we restrict our discussion here to orthochronous Lorentz transformations as explained
earlier.
 11

9 Note that we could have chosen U (Λ)|p, si ≡ |Λ−1 p, si to stick blindly to our previous con-
vention. However, our previous convention applied to fields which are functions Φa (x) of some
coordinates x. While we can still consider a 1-particle state as an abstract function of some
coordinates p, it actually describes a physical particle with momentum p wandering around in
vacuum. Thus applying a Lorentz transformation U (Λ) to such a state means to rotate and boost
Λ
this particle’s momentum according to Λ: p → Λp. The resulting state thus has a particle with
momentum Λp, i.e. U (Λ)|p, si = |Λp, si. The only drawback of this approach is that when we
quantize our field Ψa (x), the associated field operator Ψ̂a (x) that is expanded in creation and an-
nihilation operators will transform in a way opposite to the field itself. For example, the operator
Λ
for a free Dirac field will transform as Ψ̂a (x) → U (Λ)Ψ̂a (x)U −1 (Λ) = (Λ−1
D )ab Ψ̂b (Λx) whereas we
Λ
know from Eqs. (2) and (3) that the field itself transforms as Ψa (x) → (ΛD )ab Ψb (Λ−1 x). How-
ever, the relative orientation of the spinor transformation ΛD and the coordinate transformation
Λ is correct.
10 The translation subgroup {(0, a)} is normal because (Λ0 , a0 )−1 (0, a)(Λ0 , a0 ) = (0, Λ0−1 a) is still a
translation for any Poincaré transformation (Λ0 , a0 ). However, the Lorentz subgroup {(Λ, 0)} is
not normal because (Λ0 , a0 )−1 (Λ, 0)(Λ0 , a0 ) = (Λ0−1 ΛΛ0 , (Λ0−1 Λ − Λ0−1 )a0 ) which is not a homo-
geneous Lorentz transformation. When only one of the factors in the group product is normal,
one speaks of a semidirect group product o instead of a direct group product ×.