2010 words
10 minutes
Lie Groups and Lie Algebras, Part I

Prerequisites: Linear algebra (matrix exponentials, commutators), some exposure to group theory (what a group is, what the identity element is), and basic quantum mechanics (angular momentum).


If you have ever tried to learn quantum field theory, you have probably run into a sentence like “Consider the Lie algebra of the Lorentz group…”, after which every textbook assumes you already know what that means.

This post is the first in a series that builds up the language of Lie groups and their representations from the ground up. It started as self-study notes I wrote in 2022 to get a deeper understanding of all this. I am not going to be rigorous in the way a mathematician would want (mathematicians, please look away). What I do want is to be honest, i.e. to say what each object is, why we care about it, and where the subtleties are.


Definition: Lie algebra#

A Lie group is a group whose elements depend continuously on a set of parameters. Rotations are the standard example. You can rotate by any angle θ\theta, and as θ\theta varies smoothly, so does the rotation. This is in contrast to discrete groups (like the group {+1,−1}\{+1, -1\} under multiplication), whose elements can simply be listed.

Since the group elements depend on continuous parameters, we can do calculus with them. In particular, we can expand a group element near the identity and extract the generators, the infinitesimal building blocks of the group. The commutation relations between the generators define the Lie algebra,

[Ta,Tb]=ifabc Tc[T^a, T^b] = i f^{ab}{}_c \, T^c

where Ta,TbT^a, T^b are generators and the fabcf^{ab}{}_c are called the structure constants.

There are a few points here I want to stress.

  • The structure constants do not depend on the representation. This one matters. However you choose to represent the generators (as 2×22\times2 matrices, 3×33\times3 matrices, differential operators, etc.), the structure constants fabcf^{ab}{}_c are always the same. They belong to the algebra itself.

  • The relation is exact, not an approximation. When we derive the commutator from the group multiplication law (we will do so shortly), it appears at second order in the expansion parameters. You might worry that higher-order terms modify the algebra. They do not, because the Lie algebra captures the full local structure of the group near the identity. The higher-order terms in the Baker-Campbell-Hausdorff expansion are entirely fixed by repeated commutators, so they add nothing beyond what [Ta,Tb]=ifabcTc[T^a, T^b] = if^{ab}{}_c T^c already contains.

  • The structure constants define the Lie algebra. Two Lie groups can look very different globally and still have the same Lie algebra. For instance, SU(2)SU(2) and SO(3)SO(3) share the same algebra, even though SU(2)SU(2) is a double cover of SO(3)SO(3). The thing to remember is that the algebra captures the local structure, not the global topology.

Put differently, finding all matrix representations of a Lie algebra is the algebraic problem of finding all matrix solutions TRaT^a_R of [TRa,TRb]=ifabcTRc[T^a_R, T^b_R] = if^{ab}{}_c T^c_R.


Abelian Lie groups#

A group is called abelian if all of its elements commute,

g1⋅g2=g2⋅g1for all g1,g2g_1 \cdot g_2 = g_2 \cdot g_1 \quad \text{for all } g_1, g_2

Since the elements commute, so do the generators,

[Ta,Tb]=0for all a,b[T^a, T^b] = 0 \quad \text{for all } a, b

and so for an abelian Lie group all structure constants vanish, fabc=0f^{ab}{}_c = 0.

Representations of abelian Lie groups#

The representation theory of abelian Lie groups is very simple.

Any dd-dimensional abelian Lie algebra is isomorphic to the direct sum of dd one-dimensional abelian Lie algebras.

Equivalently,

All irreducible representations of abelian groups are one-dimensional.

Why? In an abelian group every generator commutes with every other one, so in an irreducible representation Schur’s lemma forces each generator to act as a multiple of the identity. But if every generator is a scalar, then every subspace is invariant, so an irreducible representation can only be one-dimensional, labelled by the eigenvalues of the generators. (This holds for complex representations, which is what quantum mechanics uses. Over the reals, rotations of the plane SO(2)SO(2) give an irreducible 2-dimensional representation.)

The classic example is U(1)U(1), the group of phase rotations eiαe^{i\alpha}. It has one generator, and every irreducible representation is labelled by a single number, the charge qq. In quantum electrodynamics this is the electric charge.


Non-abelian Lie groups#

When the generators do not all commute (i.e. at least some fabc≠0f^{ab}{}_c \neq 0), the group is non-abelian. This is where Lie theory gets interesting, and honestly where most of the physics is. The Standard Model is built on the non-abelian groups SU(3)×SU(2)×U(1)SU(3) \times SU(2) \times U(1).

Definition: Casimir operators#

In a non-abelian algebra the generators do not commute with each other. We can still ask whether there are combinations of generators that commute with all of them.

Operators constructed from the generators TaT^a that commute with all the generators are called Casimir operators.

In other words, a Casimir operator CC satisfies

[C,Ta]=0for all a[C, T^a] = 0 \quad \text{for all } a

These operators have two properties we will use all the time.

  1. In each irreducible representation, a Casimir is proportional to the identity matrix. This follows directly from Schur’s lemma, since any operator that commutes with all generators in an irreducible representation has to be a multiple of the identity.

  2. The proportionality constants label the representation. Since a Casimir takes a definite value on each irreducible representation, these values act as quantum numbers that identify the representation.

The number of independent Casimir operators equals the rank of the Lie algebra, i.e. the dimension of its maximal abelian subalgebra (the Cartan subalgebra). For SU(2)SU(2) the rank is 1, so there is one Casimir, and for SU(3)SU(3) the rank is 2, so there are two.

Example: angular momentum#

The angular momentum algebra is

[Ji,Jj]=iϵijkJk[J^i, J^j] = i\epsilon^{ijk} J^k

which is the Lie algebra of SU(2)SU(2) (equivalently so(3)\mathfrak{so}(3)). Its single Casimir operator is

J2=(Jx)2+(Jy)2+(Jz)2\mathbf{J}^2 = (J^x)^2 + (J^y)^2 + (J^z)^2

and one can check that [J2,Ji]=0[\mathbf{J}^2, J^i] = 0 for every ii, i.e. it commutes with all three generators.

On an irreducible representation J2\mathbf{J}^2 takes the value

J2=j(j+1) I\mathbf{J}^2 = j(j+1)\,\mathbf{I}

where j=0,12,1,32,2,…j = 0, \tfrac{1}{2}, 1, \tfrac{3}{2}, 2, \ldots labels the representation, and each value of jj gives a (2j+1)(2j+1)-dimensional irreducible representation. This is exactly the story of angular momentum quantization from quantum mechanics, and we now see that it is not a peculiarity of angular momentum but a structural consequence of the SU(2)SU(2) Lie algebra.


Definition: linear representation#

So far I have used the word “representation” loosely, so let me be precise.

A (linear) representation RR of a group assigns a linear operator DR(g)D_R(g) to each abstract group element gg.

Properties#

The map g↦DR(g)g \mapsto D_R(g) has to respect the group structure.

  1. Identity. DR(e)=1D_R(e) = \mathbf{1}, where ee is the identity element of the group. (For a Lie group with continuous parameter θ\theta we have g(0)=eg(0) = e, so DR(g(0))=1D_R(g(0)) = \mathbf{1}.)

  2. Homomorphism. DR(g1)DR(g2)=DR(g1g2)D_R(g_1) D_R(g_2) = D_R(g_1 g_2), i.e. the map preserves the group multiplication law.

So a representation is a homomorphism from the group to the group of linear operators on some vector space. It does not have to be injective (one-to-one), and several group elements may map to the same operator. When the map is injective, the representation is called faithful.

Basis of the representation#

The vector space on which the operators DR(g)D_R(g) act is called the basis (or carrier space) of the representation RR.

In a matrix representation the DR(g)D_R(g) are n×nn \times n matrices (DR(g))ij(D_R(g))^i{}_j acting on an nn-dimensional vector space, and the indices i,j=1,…,ni, j = 1, \ldots, n label the components of vectors in that space.

Reducible vs. irreducible representations#

A representation is reducible if there is a basis in which every DR(g)D_R(g) is block-diagonal at the same time,

DR(g)=(DR1(g)00DR2(g))D_R(g) = \begin{pmatrix} D_{R_1}(g) & 0 \\ 0 & D_{R_2}(g) \end{pmatrix}

for all group elements gg. The representation then decomposes as a direct sum, R=R1⊕R2R = R_1 \oplus R_2, and the vector space splits into invariant subspaces that the group action never mixes.

A representation is irreducible if no such decomposition exists, i.e. there is no invariant subspace other than {0}\{0\} and the whole space. Irreducible representations (often called irreps) are the building blocks of representation theory. Under fairly general conditions, any representation can be decomposed into a direct sum of irreducible ones, much like a vector can be decomposed into components along basis vectors.

Equivalent representations#

Two representations DR(g)D_R(g) and DR′(g)D_{R'}(g) are equivalent if they are related by a similarity transformation,

DR′(g)=S DR(g) S−1D_{R'}(g) = S \, D_R(g) \, S^{-1}

for one fixed invertible matrix SS and all gg. Physically, equivalent representations are the same representation written in a different basis. They contain the same physics, just in different coordinates on the vector space.


Where [Ta,Tb][T^a, T^b] comes from#

We stated the Lie algebra relation [Ta,Tb]=ifabcTc[T^a, T^b] = if^{ab}{}_c T^c at the beginning, but where does it actually come from? Here is a quick derivation.

Take a Lie group element near the identity, parameterized by a small parameter αa\alpha^a,

g(α)≈1+iαaTa+12(iαaTa)2+…g(\alpha) \approx \mathbf{1} + i\alpha^a T^a + \frac{1}{2}(i\alpha^a T^a)^2 + \ldots

and consider the combination g(α) g(β) g(α)−1 g(β)−1g(\alpha)\,g(\beta)\,g(\alpha)^{-1}\,g(\beta)^{-1}. For an abelian group this is just the identity. For a non-abelian group it is not, and how it deviates from the identity tells us about the commutator.

Expanding each factor to the relevant order and multiplying out, the first-order terms cancel (because g⋅g−1=1g \cdot g^{-1} = \mathbf{1}), and at second order we find

g(α) g(β) g(α)−1 g(β)−1≈1−αaβb[Ta,Tb]+…g(\alpha)\,g(\beta)\,g(\alpha)^{-1}\,g(\beta)^{-1} \approx \mathbf{1} - \alpha^a \beta^b [T^a, T^b] + \ldots

The left-hand side is a group element (the group is closed under multiplication) close to the identity, so it must be expressible in terms of generators,

≈1+iγcTc+…\approx \mathbf{1} + i\gamma^c T^c + \ldots

for some parameters γc\gamma^c. Comparing the two expressions gives

[Ta,Tb]=ifabcTc[T^a, T^b] = if^{ab}{}_c T^c

where the structure constants fabcf^{ab}{}_c are defined through γc=−fabc αaβb\gamma^c = -f^{ab}{}_c \,\alpha^a \beta^b. So the commutator of two generators is again a linear combination of generators, i.e. the algebra closes.

If you remember one line from this post, I would make it this one: the Lie algebra is the infinitesimal version of the group multiplication law.


Example: deriving a generator (rotations about the zz-axis)#

All of this is quite abstract, so let us see it in a concrete case. Take a wavefunction ψ(θz)\psi(\theta^z) and rotate it about the zz-axis by an infinitesimal angle δθ0\delta\theta_0,

ψ(θz+δθ0)=ψ(θz)+dψ(θz)dθz δθ0+…\psi(\theta^z + \delta\theta_0) = \psi(\theta^z) + \frac{d\psi(\theta^z)}{d\theta^z}\,\delta\theta_0 + \ldots

Recall that the angular momentum operator along zz is

Jz=−iddθzJ^z = -i\frac{d}{d\theta^z}

and substituting it gives

ψ(θz+δθ0)=ψ(θz)+iJz ψ(θz) δθ0=(1+iJz δθ0) ψ(θz)\psi(\theta^z + \delta\theta_0) = \psi(\theta^z) + iJ^z\,\psi(\theta^z)\,\delta\theta_0 = (1 + iJ^z\,\delta\theta_0)\,\psi(\theta^z)

which is the infinitesimal rotation (1+iJzδθ0)(1 + iJ^z \delta\theta_0) acting on the wavefunction.

Building a finite rotation#

To rotate by a finite angle θ0\theta_0, we compose NN infinitesimal rotations of size δθ0=θ0/N\delta\theta_0 = \theta_0/N each,

ψ(θz+θ0)=lim⁡N→∞(1+iJzθ0N)Nψ(θz)=eiJzθ0 ψ(θz)\psi(\theta^z + \theta_0) = \lim_{N\to\infty}\left(1 + iJ^z\frac{\theta_0}{N}\right)^N \psi(\theta^z) = e^{iJ^z\theta_0}\,\psi(\theta^z)

This is the exponential map, which takes us from the Lie algebra (the generator JzJ^z) to the Lie group (the finite rotation eiJzθ0e^{iJ^z\theta_0}).

Sign convention#

For a general rotation with angle vector θ=(θx,θy,θz)\boldsymbol{\theta} = (\theta^x, \theta^y, \theta^z), the rotation operator is conventionally written as

R(θ)=exp⁡(−i J⋅θ)R(\boldsymbol{\theta}) = \exp(-i\,\mathbf{J}\cdot\boldsymbol{\theta})

where J=(Jx,Jy,Jz)\mathbf{J} = (J^x, J^y, J^z).

We just derived e+iJzθ0e^{+iJ^z\theta_0}, and now there is a minus sign. This is not a mistake, only a choice of convention. The sign depends on whether we think of the rotation as acting on the coordinates (passive) or on the physical system (active).

  • Active transformation (rotating the physical state). The operator acting on a state ∣ψ⟩|\psi\rangle in the Hilbert space is e−iJ⋅θe^{-i\mathbf{J}\cdot\boldsymbol{\theta}}.
  • Passive transformation (rotating the coordinate axes). The wavefunction transforms as ψ→e+iJ⋅θψ\psi \to e^{+i\mathbf{J}\cdot\boldsymbol{\theta}}\psi.

The two are inverses of each other, related by θ→−θ\boldsymbol{\theta} \to -\boldsymbol{\theta}. Most QFT textbooks (Schwartz, Peskin & Schroeder, Weinberg) use the active convention with the minus sign, and so do Lancaster & Blundell. Pick one, state it clearly, and be consistent throughout.


Connection to what comes next#

Everything we have built here (generators, commutation relations, representations, the exponential map) is the language we need for the Lorentz group.

In the next post we define the Lorentz group O(3,1)O(3,1) as the group of transformations that preserve the spacetime interval, extract its generators (three rotations JiJ^i and three boosts KiK^i), and find that these six generators satisfy a very specific algebra. Finite transformations are then recovered with the exponential map from this post,

Λ=exp⁡ ⁣(−i2 ωμν Jμν)\Lambda = \exp\!\left(-\tfrac{i}{2}\,\omega_{\mu\nu}\,J^{\mu\nu}\right)

The fact that this algebra is non-abelian, and non-compact, has big consequences for its representation theory, and it leads us to spinors, the Dirac equation and the classification of all particles by mass and spin.


Next post: Part II: The Lorentz Group: Definition, Structure, and Representations


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 I
https://rohankulkarni.me/posts/notes/lie-groups-lie-algebras/
Author
Rohan Kulkarni
Published at
2026-04-03
License
CC BY-NC-SA 4.0
Share Post
Comments