2937 words
15 minutes
Lie Groups and Lie Algebras, Part IV: The Poincaré Group and the Classification of Particles

Prerequisites: Part I: Lie Groups and Lie Algebras, Part II: The Lorentz Group, Part III: Representations of the Lorentz Group. Familiarity with four-momentum and basic special relativity.


We have spent three posts on the Lorentz group and its representations. We learnt how fields transform under rotations and boosts, and we classified the building blocks (scalars, spinors, vectors) by pairs of half-integers (j+,j−)(j_+, j_-).

We left something out, though. The Lorentz group describes rotations and boosts, i.e. how to go between observers in relative motion. It does not contain translations, moving your experiment from one place to another, or waiting and doing it tomorrow. These are symmetries of nature too, and they matter. Conservation of energy and momentum follows from translational symmetry, not from Lorentz symmetry.

The full symmetry group of flat spacetime, Lorentz transformations plus translations, is the Poincaré group. Its representation theory answers one of the most basic questions in particle physics,

What is a particle?

Wigner’s answer is short: an elementary particle is an irreducible unitary representation of the Poincaré group. Every such representation is labelled by exactly two numbers, mass and spin, and this post works through how that comes about.


The Poincaré group#

Definition#

A general Poincaré transformation acts on spacetime coordinates as

xμ→x′μ=Λμν xν+aμx^\mu \to x'^\mu = \Lambda^\mu{}_\nu\, x^\nu + a^\mu

where Λ\Lambda is a Lorentz transformation and aμa^\mu is a constant translation four-vector. The Poincaré group is the set of all such transformations, and it is a 10-parameter group, six parameters for Λ\Lambda (three rotations, three boosts) and four for aμa^\mu (one time translation, three spatial translations).

It is a semi-direct product of the Lorentz group and the translation group,

Poincareˊ=R3,1⋊SO+(3,1)\text{Poincaré} = \mathbb{R}^{3,1} \rtimes SO^+(3,1)

“Semi-direct” (instead of direct) means that Lorentz transformations and translations do not simply commute. A Lorentz transformation acts on translations by rotating or boosting the translation vector, and this will show up in the algebra.

Generators#

The Poincaré group has 10 generators,

  • JμνJ^{\mu\nu}, the six Lorentz generators (three rotations JiJ^i, three boosts KiK^i), as before,
  • PμP^\mu, the four translation generators (the energy P0=HP^0 = H and the three-momentum PiP^i).

The Poincaré algebra#

The commutation relations are

[Jμν,Jρσ]=i(ηνρJμσ−ημρJνσ−ηνσJμρ+ημσJνρ)[J^{\mu\nu}, J^{\rho\sigma}] = i\left(\eta^{\nu\rho}J^{\mu\sigma} - \eta^{\mu\rho}J^{\nu\sigma} - \eta^{\nu\sigma}J^{\mu\rho} + \eta^{\mu\sigma}J^{\nu\rho}\right)[Jμν,Pρ]=i(ηνρPμ−ημρPν)[J^{\mu\nu}, P^\rho] = i\left(\eta^{\nu\rho}P^\mu - \eta^{\mu\rho}P^\nu\right)[Pμ,Pν]=0[P^\mu, P^\nu] = 0

and it is worth unpacking what each one says.

The first relation is the Lorentz algebra from Part II, now in fully covariant tensor notation. It contains [Ji,Jj]=iϵijkJk[J^i, J^j] = i\epsilon^{ijk}J^k, [Ji,Kj]=iϵijkKk[J^i, K^j] = i\epsilon^{ijk}K^k and [Ki,Kj]=−iϵijkJk[K^i, K^j] = -i\epsilon^{ijk}J^k in one compact formula.

The second relation says that PρP^\rho transforms as a 4-vector under Lorentz transformations. Physically this is obvious, since rotating the coordinate system rotates the components of the momentum. In terms of rotation and boost generators,

[Ji,Pj]=iϵijkPk(momentum is a 3-vector under rotations)[J^i, P^j] = i\epsilon^{ijk}P^k \qquad \text{(momentum is a 3-vector under rotations)}[Ji,P0]=0(energy is a scalar under rotations)[J^i, P^0] = 0 \qquad \text{(energy is a scalar under rotations)}[Ki,Pj]=−iP0δij(boosts mix energy and momentum)[K^i, P^j] = -iP^0\delta^{ij} \qquad \text{(boosts mix energy and momentum)}[Ki,P0]=−iPi(boosts mix energy and momentum)[K^i, P^0] = -iP^i \qquad \text{(boosts mix energy and momentum)}

The last two are exactly what special relativity tells us, a boost in the xx-direction mixes P0P^0 (energy) with PxP^x (momentum along xx).

The third relation says that translations commute, which is the statement that spacetime is flat. In general relativity spacetime is curved, translations along different directions do not commute, and the Poincaré group would have to be replaced by something more complicated. For the flat spacetime of particle physics, [Pμ,Pν]=0[P^\mu, P^\nu] = 0.


Casimir operators of the Poincaré group#

Recall from Part I that Casimir operators commute with all generators, take definite values on each irreducible representation (by Schur’s lemma), and so label the representations. The Poincaré group has two independent Casimirs.

First Casimir: P2P^2#

C1=PμPμ=P2C_1 = P^\mu P_\mu = P^2

This is the squared four-momentum, i.e. the mass squared,

P2=(P0)2−P2=m2P^2 = (P^0)^2 - \mathbf{P}^2 = m^2

One can check that [P2,Jμν]=0[P^2, J^{\mu\nu}] = 0 (since PμP^\mu is a vector, P2P^2 is a Lorentz scalar) and [P2,Pμ]=0[P^2, P^\mu] = 0 (since translations commute), so P2P^2 commutes with all ten generators.

On each irreducible representation P2P^2 takes a definite value m2m^2, the mass of the particle. That mass is a Casimir, i.e. a label of the representation and not a dynamical variable, is the reason all electrons have exactly the same mass. They all live in the same irreducible representation.

Second Casimir: W2W^2#

The second Casimir is built from the Pauli-Lubanski pseudovector,

Wμ=12ϵμνρσJνρPσW^\mu = \frac{1}{2}\epsilon^{\mu\nu\rho\sigma}J_{\nu\rho}P_\sigma

where ϵμνρσ\epsilon^{\mu\nu\rho\sigma} is the totally antisymmetric Levi-Civita tensor with ϵ0123=+1\epsilon^{0123} = +1. It has a few important properties.

  • WμPμ=0W^\mu P_\mu = 0, i.e. it is orthogonal to the four-momentum. This follows from the antisymmetry of ϵμνρσ\epsilon^{\mu\nu\rho\sigma} contracted with the symmetric product PσPμP_\sigma P_\mu.

  • WμW^\mu transforms as a four-vector under Lorentz transformations (it has one free index μ\mu), and it commutes with translations, [Wμ,Pν]=0[W^\mu, P^\nu] = 0.

  • The Casimir is C2=WμWμ=W2C_2 = W^\mu W_\mu = W^2, which commutes with all ten generators.

What W2W^2 means physically depends on whether the particle is massive or massless, as we will see shortly.


Wigner’s classification#

We now want all irreducible unitary representations of the Poincaré group. Eugene Wigner worked out the method in 1939.

Why unitary?#

In quantum mechanics, symmetry transformations have to be represented by unitary operators (or anti-unitary ones, for time reversal). This is Wigner’s theorem. Any symmetry of a quantum system, i.e. any map that preserves transition probabilities, is implemented by a unitary or anti-unitary operator on the Hilbert space.

For the Lorentz group alone, we saw in Part III that the finite-dimensional representations are not unitary (the boost generators are anti-Hermitian). This is not a contradiction. The finite-dimensional representations describe how field components transform, not how quantum states transform. Quantum states live in an infinite-dimensional Hilbert space, where unitary representations do exist.

So the Poincaré representations we are after are infinite-dimensional (they act on the space of one-particle states of definite momentum), and they are unitary.

The method of induced representations#

Wigner’s strategy goes as follows.

Step 1. Since P2=m2P^2 = m^2 is a Casimir, fix m2m^2 and consider the set of all four-momenta with p2=m2p^2 = m^2. This is a hyperboloid in momentum space (for m2>0m^2 > 0) or the light cone (for m2=0m^2 = 0).

Step 2. Pick a standard (reference) momentum kμk^\mu on that surface. Any other momentum pμp^\mu on the same surface can be reached from kμk^\mu by a Lorentz transformation, pμ=L(p)μν kνp^\mu = L(p)^\mu{}_\nu\, k^\nu, for some “standard boost” L(p)L(p).

Step 3. Find the little group, the subgroup of Lorentz transformations that leave kμk^\mu invariant,

Λμν kν=kμ\Lambda^\mu{}_\nu\, k^\nu = k^\mu

The little group depends on the choice of kμk^\mu, but different choices on the same mass shell give isomorphic little groups.

Step 4. Classify the irreducible representations of the little group. These give the “internal” degrees of freedom of the particle (spin, polarization, helicity).

Step 5. Build the full Poincaré representation by “boosting” the little group representation to all other momenta. This is the induced representation.

The result is that every irreducible unitary representation of the Poincaré group is fixed by the mass mm and an irreducible representation of the little group.

Let us carry this out for the two physical cases.


Massive representations (m2>0m^2 > 0)#

Standard momentum#

For a massive particle the natural standard momentum is the rest frame,

kμ=(m,0,0,0)k^\mu = (m, 0, 0, 0)

i.e. the momentum of a massive particle at rest. Every other momentum pμ=(E,p)p^\mu = (E, \mathbf{p}) on the mass shell p2=m2p^2 = m^2 is reached by boosting from the rest frame.

Little group: SO(3)SO(3)#

Which Lorentz transformations leave kμ=(m,0,0,0)k^\mu = (m, 0, 0, 0) invariant? Boosts certainly do not, since boosting a particle at rest gives it momentum. Rotations, though, leave the rest-frame momentum unchanged. A rotated particle at rest is still at rest, with the same energy mm and zero three-momentum.

So the little group for massive particles is the rotation group SO(3)SO(3), or equivalently its double cover SU(2)SU(2).

Spin#

The irreducible representations of SU(2)SU(2) are labelled by the spin j=0,12,1,32,2,…j = 0, \frac{1}{2}, 1, \frac{3}{2}, 2, \ldots, and have dimension 2j+12j + 1.

In the rest frame the Pauli-Lubanski vector simplifies a lot. Since kμ=(m,0)k^\mu = (m, \mathbf{0}),

W0=12ϵ0ijkJijPk=0(in the rest frame, Pk=0)W^0 = \frac{1}{2}\epsilon^{0ijk}J_{ij}P_k = 0 \quad \text{(in the rest frame, } P_k = 0\text{)}Wi=12ϵijk0JjkP0=−m 12ϵijkJjk=−m JiW^i = \frac{1}{2}\epsilon^{ijk0}J_{jk}P_0 = -m\,\frac{1}{2}\epsilon^{ijk}J^{jk} = -m\,J^i

(using ϵijk0=−ϵ0ijk=−ϵijk\epsilon^{ijk0} = -\epsilon^{0ijk} = -\epsilon^{ijk} and Jjk=JjkJ_{jk} = J^{jk} for two spatial indices). So in the rest frame Wμ=(0,−m J)W^\mu = (0, -m\,\mathbf{J}). Up to an overall sign that depends on conventions (many books define WμW^\mu with the opposite sign, or use ϵ0123=+1\epsilon_{0123} = +1), the Pauli-Lubanski vector is just the mass times the angular momentum. The sign drops out of the Casimir anyway,

W2=WμWμ=−W2=−m2 J2=−m2 j(j+1)W^2 = W^\mu W_\mu = -\mathbf{W}^2 = -m^2\,\mathbf{J}^2 = -m^2\,j(j+1)

This is the physical content of the second Casimir, W2W^2 measures the spin. A massive particle is characterized by

P2=m2,W2=−m2 j(j+1)\boxed{P^2 = m^2, \qquad W^2 = -m^2\,j(j+1)}

and a massive spin-jj particle has 2j+12j + 1 independent polarization states (the 2j+12j + 1 values of JzJ^z in the rest frame, mj=−j,−j+1,…,j−1,jm_j = -j, -j+1, \ldots, j-1, j). For example,

  • spin-0 (Higgs boson): 1 state,
  • spin-12\frac{1}{2} (electron): 2 states (spin up, spin down),
  • spin-1 (WW boson): 3 states (three polarizations),
  • spin-2 (hypothetical massive graviton): 5 states.

Massless representations (m2=0m^2 = 0)#

Standard momentum#

A massless particle has no rest frame, since it moves at the speed of light in every frame. The standard momentum is conventionally chosen along the zz-axis,

kμ=(ω,0,0,ω)k^\mu = (\omega, 0, 0, \omega)

for some energy ω>0\omega > 0, which satisfies k2=ω2−ω2=0k^2 = \omega^2 - \omega^2 = 0.

Little group: ISO(2)ISO(2)#

Which Lorentz transformations leave kμ=(ω,0,0,ω)k^\mu = (\omega, 0, 0, \omega) invariant?

Rotations about the zz-axis (the direction of motion) clearly do. They rotate the transverse xyxy-plane and leave kμk^\mu unchanged, which gives a U(1)U(1) subgroup generated by JzJ^z.

There are also less obvious ones, namely two further generators that leave kμk^\mu invariant. They are combinations of rotations and boosts in the transverse directions,

Π1=Kx−Jy,Π2=Ky+Jx\Pi_1 = K^x - J^y, \qquad \Pi_2 = K^y + J^x

From the commutators [Ji,Pj][J^i, P^j], [Ki,Pj][K^i, P^j], [Ki,P0][K^i, P^0] above, one can check directly that [Π1,2,Pμ][\Pi_{1,2}, P^\mu] vanishes when Pμ=kμP^\mu = k^\mu. So Π1,2\Pi_{1,2} acting on a state of momentum kk gives another state of momentum kk, i.e. these generators leave the standard momentum invariant. (The relative signs depend on the sign convention for KiK^i. With the opposite convention, Ki→−KiK^i \to -K^i, they read Kx+JyK^x + J^y and Ky−JxK^y - J^x, as in Weinberg.) Together with JzJ^z they satisfy

[Jz,Π1]=iΠ2,[Jz,Π2]=−iΠ1,[Π1,Π2]=0[J^z, \Pi_1] = i\Pi_2, \qquad [J^z, \Pi_2] = -i\Pi_1, \qquad [\Pi_1, \Pi_2] = 0

which is the algebra of ISO(2)ISO(2), the Euclidean group in two dimensions, i.e. rotations and translations of a 2D plane.

Compare this with the massive case,

Massive (m>0m > 0)Massless (m=0m = 0)
Standard momentum(m,0,0,0)(m, 0, 0, 0)(ω,0,0,ω)(\omega, 0, 0, \omega)
Little groupSO(3)≅SU(2)/Z2SO(3) \cong SU(2)/\mathbb{Z}_2ISO(2)ISO(2)
GeneratorsJx,Jy,JzJ^x, J^y, J^zJz,Π1,Π2J^z, \Pi_1, \Pi_2
Compact?YesNo (translations)

The little group has changed its topology, from the compact SO(3)SO(3) to the non-compact ISO(2)ISO(2). This is not a smooth deformation, there is no continuous way from one to the other, and this has physical consequences.

The translation generators of ISO(2)ISO(2)#

The representations of ISO(2)ISO(2) come in two types.

Faithful representations. The “translation” generators Π1,Π2\Pi_1, \Pi_2 have nonzero eigenvalues. These are the continuous spin representations. They are labelled by a continuous parameter ρ2=Π12+Π22>0\rho^2 = \Pi_1^2 + \Pi_2^2 > 0, and each momentum carries an infinite tower of helicity states λ=0,±1,±2,…\lambda = 0, \pm1, \pm2, \ldots (or all half-integers). They would describe particles with infinitely many polarization states for each momentum, and nothing like this has ever been observed in nature. They are discarded on physical grounds, since they lead to pathological behaviour (infinite degeneracy, problems with localizability, and conflict with the observed particle spectrum).

Unfaithful representations. We set Π1=Π2=0\Pi_1 = \Pi_2 = 0 and keep only JzJ^z. The little group then effectively reduces to SO(2)≅U(1)SO(2) \cong U(1), the rotations of the transverse plane.

Every physical massless particle belongs to the second type.

Helicity#

The irreducible representations of U(1)U(1) are one-dimensional and labelled by a single number, the helicity,

Jz ∣k,λ⟩=λ ∣k,λ⟩J^z \,|k, \lambda\rangle = \lambda\, |k, \lambda\rangle

where λ\lambda is the component of angular momentum along the direction of motion. Since JzJ^z generates rotations by an angle θ\theta via e−iθJze^{-i\theta J^z}, and a 2π2\pi rotation has to give ±1\pm 1 (for bosons and fermions respectively), λ\lambda must be an integer or half-integer, λ=0,±12,±1,±32,±2,…\lambda = 0, \pm\frac{1}{2}, \pm 1, \pm\frac{3}{2}, \pm 2, \ldots

A massless particle has only one polarization state for each λ\lambda. CPT symmetry (which holds in any local relativistic quantum field theory) requires that if a state with helicity +λ+\lambda exists, so does one with helicity −λ-\lambda. So physical massless particles come in pairs of helicity states, (+λ,−λ)(+\lambda, -\lambda).

The contrast with massive particles is clear,

Massive spin-jjMassless helicity λ\lambda
Polarization states2j+12j + 122 (for λ≠0\lambda \neq 0)
Spin-12\frac{1}{2}2 states2 states
Spin-13 states2 states
Spin-25 states2 states

For spin-12\frac{1}{2} the counts happen to agree (2 states in both cases). For higher spins, the massless case has fewer states. A massive spin-1 particle (like the WW boson) has 3 polarizations, while a massless spin-1 particle (the photon) has only 2. The “missing” longitudinal polarization of the photon is a direct consequence of the little group being ISO(2)ISO(2) and not SO(3)SO(3).

This is also why the Higgs mechanism is needed. When a gauge boson acquires a mass through the Higgs field, it has to pick up a third (longitudinal) polarization, which it “eats” from the Goldstone boson. Going from 2 polarizations to 3 is going from ISO(2)ISO(2) to SO(3)SO(3) as the little group.


The massive-massless discontinuity#

Now we can understand something that surprises a lot of people the first time they see it. The m→0m \to 0 limit of a massive representation is not the massless representation.

A massive spin-1 particle has 2(1)+1=32(1) + 1 = 3 polarization states. As m→0m \to 0, all three persist until mm is exactly zero, at which point the little group jumps from SO(3)SO(3) to ISO(2)ISO(2) and only 2 helicity states survive. So the counting of states is discontinuous.

Whether this discontinuity is physical depends on the spin. For a massive photon coupled to a conserved current it is not. The coupling of the longitudinal mode to the current vanishes as m→0m \to 0 (the kμkν/m2k^\mu k^\nu/m^2 term in the propagator drops out against ∂μJμ=0\partial_\mu J^\mu = 0), so the predictions of massive-photon electrodynamics go smoothly over to Maxwell theory, and the extra mode just decouples. For spin 2 it is different. In massive gravity (where one tries to give the graviton a small mass), the helicity-0 mode keeps coupling to the trace of the stress-energy tensor as m→0m \to 0, and the discontinuity is physical. The predictions of massive gravity do not approach those of general relativity smoothly as m→0m \to 0. This is the van Dam-Veltman-Zakharov (vDVZ) discontinuity, and it was a major puzzle in theoretical physics. (The resolution involves nonlinear effects, the Vainshtein mechanism, but the discontinuity at the linear level is real and a direct consequence of the little group structure we have described.)

The same change of little group explains why “taking the mass to zero” in the Proca equation (massive spin-1) is subtler than it looks. Setting m=0m = 0 formally gives Maxwell’s equations, but the massive theory has 3 degrees of freedom and no gauge invariance, while the massless theory has 2 degrees of freedom and needs gauge invariance. The Proca condition ∂μAμ=0\partial_\mu A^\mu = 0, which follows from the field equations for m≠0m \neq 0, becomes a gauge choice at m=0m = 0. So the gauge symmetry appears at m=0m = 0 precisely to remove the extra longitudinal mode. It is not put in by hand, it is forced by the change of little group.


What is a particle?#

Putting it all together, Wigner’s classification says

An elementary particle is an irreducible unitary representation of the Poincaré group, labeled by mass mm and spin (or helicity).

and the full classification is

m2m^2Little GroupLabelStatesExamples
m2>0m^2 > 0SO(3)SO(3)Spin jj2j+12j+1Electron (j=1/2j = 1/2), W±W^\pm (j=1j = 1), Higgs (j=0j = 0)
m2=0m^2 = 0ISO(2)→U(1)ISO(2) \to U(1)Helicity λ\lambda22Photon (λ=±1\lambda = \pm 1), graviton (λ=±2\lambda = \pm 2)
m2<0m^2 < 0SO(2,1)SO(2,1)——Tachyons (unphysical)

The case m2<0m^2 < 0 (tachyons) gives representations whose “mass shell” p2=m2p^2 = m^2 is a spacelike hyperboloid. Such particles would travel faster than light. They are not observed in nature, and their appearance in a theory usually signals an instability, i.e. the field sits at a local maximum of its potential and not a minimum. In the Standard Model, the Higgs field before electroweak symmetry breaking has a tachyonic mode, but this only means the field rolls down to its true vacuum, where the physical Higgs particle has m2>0m^2 > 0.


Looking back#

We started this series by asking what a Lie group is. Four posts later we can answer a much bigger question, namely what a particle is. The logical chain goes like this.

  1. Lie groups and algebras (Post I). Continuous symmetries are described by Lie groups, and their infinitesimal structure is encoded in Lie algebras (commutators of generators with structure constants).

  2. The Lorentz group (Post II). The symmetry of spacetime under rotations and boosts is SO+(3,1)SO^+(3,1). Its six generators satisfy a specific non-abelian algebra, with the minus sign in [Ki,Kj]=−iϵijkJk[K^i, K^j] = -i\epsilon^{ijk}J^k.

  3. Representations (Post III). After complexification, the Lorentz algebra decomposes into su(2)⊕su(2)\mathfrak{su}(2) \oplus \mathfrak{su}(2), which gives representations labelled by (j+,j−)(j_+, j_-). These are the scalars, Weyl spinors, vectors and higher-spin objects that quantum fields are built from.

  4. The Poincaré group and particles (Post IV). Adding translations gives the Poincaré group, and Wigner’s theorem classifies all its irreducible unitary representations by mass and spin. The little group (SO(3)SO(3) for massive particles, ISO(2)ISO(2) for massless ones) fixes the polarization content. That the photon has 2 polarizations, the electron 2 spin states and the WW boson 3 polarizations is not an empirical accident, it follows mathematically from the symmetry structure of spacetime.

The representation theory of the Poincaré group is the classification of everything that can exist as a free particle in flat spacetime. Describing how these particles interact needs gauge theory, Lagrangians and the full apparatus of quantum field theory. Which particles can exist in the first place, though, is fixed entirely by symmetry.


This concludes the series on Lie Groups, the Lorentz Group, and the Poincaré Group.


This post is based on my own self-study notes that I created in 2022 in order to get a deeper understanding of all of this.

Lie Groups and Lie Algebras, Part IV: The Poincaré Group and the Classification of Particles
https://rohankulkarni.me/posts/notes/poincare-group/
Author
Rohan Kulkarni
Published at
2026-04-06
License
CC BY-NC-SA 4.0
Share Post
Comments