MathLabs

Algebra

Representation theory

Studies abstract algebraic structures by representing their elements as matrices acting on vector spaces.

IntuitionTurning abstract symmetry into matrices you can compute with

A group GG can be entirely abstract — symbols with a multiplication rule — but every time it acts as symmetries of something (a triangle, a molecule, a physical system), those symmetries can be written down as matrices acting on a vector space VV. A representation is exactly this: a map ρ:G→GL(V)\rho: G \to \mathrm{GL}(V) satisfying ρ(g1g2)=ρ(g1)ρ(g2)\rho(g_1 g_2) = \rho(g_1)\rho(g_2), turning the abstract multiplication of GG into ordinary matrix multiplication. This is a change of viewpoint with enormous payoff: hard questions about an abstract group become linear-algebra questions about matrices, which we know how to compute with. The simplest nontrivial example is a rotation: the cyclic group Z/nZ\mathbb{Z}/n\mathbb{Z} acts on C\mathbb{C} by ρ(k)=e2πik/n\rho(k) = e^{2\pi i k/n}, sending the abstract generator to a rotation of the plane by one nn-th of a full turn.

Point on the unit circle at angle theta, illustrating a one-dimensional representation of a cyclic group by roots of unity.
The point e2πik/ne^{2\pi i k/n} traces the 11-dimensional representation ρ(k)=e2πik/n\rho(k) = e^{2\pi i k/n} of the cyclic group Z/nZ\mathbb{Z}/n\mathbb{Z} — every irreducible complex representation of a finite abelian group looks exactly like this: a root of unity.

SchoolRepresentations, characters, and reducibility

Definition: Representation and character

A representation of a group GG on a vector space VV over a field kk is a homomorphism ρ:G→GL(V)\rho: G \to \mathrm{GL}(V); dim⁡V\dim V is the dimension of the representation. The character χρ\chi_\rho records just the trace of each matrix: χρ(g)=tr⁡(ρ(g))\chi_\rho(g) = \operatorname{tr}(\rho(g)). Remarkably, for representations over C\mathbb{C} of a finite group, the character alone determines the representation up to isomorphism — an entire matrix-valued function collapses to a single number per group element without losing information.

ρ:G→GL(V),ρ(g1g2)=ρ(g1)ρ(g2)for all g1,g2∈G\rho: G \to \mathrm{GL}(V), \qquad \rho(g_1 g_2) = \rho(g_1)\rho(g_2) \quad \text{for all } g_1, g_2 \in G

A subspace W⊆VW \subseteq V is **GG-invariant** if ρ(g)W⊆W\rho(g)W \subseteq W for every g∈Gg \in G; a representation with no invariant subspace other than 00 and VV is irreducible. Every representation of a finite group decomposes into a direct sum of irreducibles — think of irreducibles as the "prime numbers" of representation theory, the indivisible building blocks every representation is built from.

Reducible versus irreducible representations
PropertyReducibleIrreducible
Invariant subspaceshas a proper nonzero oneonly 00 and VV
Example for G=Z/nZG=\mathbb{Z}/n\mathbb{Z}regular representation on Cn\mathbb{C}^neach 11-dimensional piece ρ(k)=e2πijk/n\rho(k)=e^{2\pi i jk/n}
Role in the theorydecomposes furtherbuilding block (indecomposable and simple)

UndergraduateComplete reducibility, Schur's lemma, and orthogonality

Let GG be a finite group and VV a representation over a field kk with char⁡(k)\operatorname{char}(k) not dividing ∣G∣|G| (in particular, always true over C\mathbb{C}). If W⊆VW \subseteq V is a GG-invariant subspace, then there is a GG-invariant complement W′W' with V=W⊕W′V = W \oplus W'. Consequently every such representation decomposes as a direct sum of irreducible representations.

Why is it true?

Linear algebra alone guarantees some complement to WW, but a generic complement will not be GG-invariant. Maschke's theorem shows we can always average an arbitrary projection over the group to force it to commute with the GG-action, at the cost of dividing by ∣G∣|G| — which is exactly why the theorem needs char⁡(k)∤∣G∣\operatorname{char}(k) \nmid |G|, and exactly why modular representation theory (where this division is impossible) behaves so differently.

Proof

Step 1 (start with any complement). By ordinary linear algebra, choose any vector-space complement UU to WW in VV (not necessarily GG-invariant), and let π:V→W\pi: V \to W be the corresponding projection (π∣W=id\pi|_W = \mathrm{id}, ker⁡π=U\ker\pi = U).

Step 2 (average over the group). Define πˉ=1∣G∣∑g∈Gρ(g) π ρ(g)−1\bar\pi = \dfrac{1}{|G|}\sum_{g \in G} \rho(g)\, \pi\, \rho(g)^{-1}. This makes sense exactly because char⁡(k)∤∣G∣\operatorname{char}(k) \nmid |G|, so ∣G∣|G| is invertible in kk.

Step 3 (πˉ\bar\pi is still a projection onto WW). For w∈Ww \in W, since WW is GG-invariant, ρ(g)−1w∈W\rho(g)^{-1}w \in W, so π(ρ(g)−1w)=ρ(g)−1w\pi(\rho(g)^{-1}w) = \rho(g)^{-1}w, giving ρ(g)π(ρ(g)−1w)=w\rho(g)\pi(\rho(g)^{-1}w) = w for every term; averaging these ∣G∣|G| copies of ww gives πˉ(w)=w\bar\pi(w) = w. For general v∈Vv \in V, π(ρ(g)−1v)∈W\pi(\rho(g)^{-1}v) \in W always (image of π\pi), so ρ(g)π(ρ(g)−1v)∈ρ(g)W=W\rho(g)\pi(\rho(g)^{-1}v) \in \rho(g)W = W by invariance, and averaging stays in WW; hence πˉ:V→W\bar\pi: V \to W is a projection onto WW.

Step 4 (πˉ\bar\pi commutes with the action, so its kernel is GG-invariant). For any h∈Gh \in G, ρ(h)πˉρ(h)−1=1∣G∣∑gρ(hg)πρ(hg)−1=πˉ\rho(h)\bar\pi\rho(h)^{-1} = \dfrac{1}{|G|}\sum_{g} \rho(hg)\pi\rho(hg)^{-1} = \bar\pi since g↦hgg \mapsto hg merely permutes the sum; so ρ(h)πˉ=πˉρ(h)\rho(h)\bar\pi = \bar\pi\rho(h). Setting W′=ker⁡πˉW' = \ker\bar\pi, for v∈W′v \in W' we get πˉ(ρ(h)v)=ρ(h)πˉ(v)=ρ(h)(0)=0\bar\pi(\rho(h)v) = \rho(h)\bar\pi(v) = \rho(h)(0) = 0, so ρ(h)v∈W′\rho(h)v \in W': W′W' is GG-invariant, and V=W⊕W′V = W \oplus W' since πˉ\bar\pi is a projection onto WW. Iterating this splitting on W′W' (finite dimension forces termination) decomposes VV into a direct sum of irreducibles.

(Schur's Lemma) If V,WV, W are irreducible complex representations of GG and f:V→Wf: V \to W is a GG-equivariant linear map, then either f=0f = 0 or ff is an isomorphism; if V=WV = W, then f=λIf = \lambda I for some scalar λ∈C\lambda \in \mathbb{C}. (Character orthogonality) Consequently, for irreducible characters χi,χj\chi_i, \chi_j of a finite group GG, ⟨χi,χj⟩=δij\langle \chi_i, \chi_j \rangle = \delta_{ij}, where ⟨χi,χj⟩=1∣G∣∑g∈Gχi(g)χj(g)‾\langle \chi_i, \chi_j \rangle = \dfrac{1}{|G|}\sum_{g \in G} \chi_i(g)\overline{\chi_j(g)}.

Why is it true?

Schur's Lemma says irreducible representations are as rigid as possible: the only maps between them that respect the group action are either zero or invertible, and self-maps are just scalars. This rigidity is exactly what forces the orthogonality relation, which in turn gives an extremely practical tool — a dot-product test on a finite table of numbers (the character table) that instantly tells you which representations are irreducible and how any representation decomposes.

Proof

Step 1 (Schur's Lemma, kernel and image). Let f:V→Wf: V \to W be GG-equivariant, i.e. f(ρV(g)v)=ρW(g)f(v)f(\rho_V(g)v) = \rho_W(g)f(v) for all g,vg,v. Then ker⁡f⊆V\ker f \subseteq V is a GG-invariant subspace (if f(v)=0f(v)=0 then f(ρV(g)v)=ρW(g)f(v)=0f(\rho_V(g)v) = \rho_W(g)f(v) = 0), and since VV is irreducible, ker⁡f\ker f is either 00 or all of VV.

Step 2 (conclude f=0f=0 or ff injective, then surjective). If ker⁡f=V\ker f = V, then f=0f = 0. Otherwise ker⁡f=0\ker f = 0, so ff is injective; the image f(V)⊆Wf(V) \subseteq W is also GG-invariant (by equivariance), and since WW is irreducible and f(V)≠0f(V) \ne 0 (as f≠0f\ne0), f(V)=Wf(V) = W, making ff bijective, hence an isomorphism.

Step 3 (V=WV=W forces f=λIf=\lambda I). Since C\mathbb{C} is algebraically closed, ff has an eigenvalue λ\lambda with eigenspace Eλ=ker⁡(f−λI)≠0E_\lambda = \ker(f-\lambda I) \ne 0. Because ff is GG-equivariant, so is f−λIf - \lambda I, so EλE_\lambda is a nonzero GG-invariant subspace of the irreducible VV, forcing Eλ=VE_\lambda = V; hence f=λIf = \lambda I.

Step 4 (orthogonality from averaging). Fix bases and let f:Vj→Vif: V_j \to V_i be an arbitrary linear map; form the equivariant map fˉ=1∣G∣∑gρi(g)fρj(g)−1\bar f = \dfrac{1}{|G|}\sum_{g} \rho_i(g) f \rho_j(g)^{-1} exactly as in Maschke's theorem. By Steps 1–3, fˉ=0\bar f = 0 if Vi≇VjV_i \not\cong V_j, and fˉ=λI\bar f = \lambda I (with λ\lambda computable via trace) if Vi=VjV_i = V_j. Choosing ff to be elementary matrix units and comparing matrix entries of fˉ\bar f on both sides of this identity yields, after summing over the diagonal to extract traces, exactly 1∣G∣∑g∈Gχi(g)χj(g)‾=δij\dfrac{1}{|G|}\sum_{g \in G} \chi_i(g)\overline{\chi_j(g)} = \delta_{ij}.

⟨χi,χj⟩=1∣G∣∑g∈Gχi(g)χj(g)‾=δij\langle \chi_i, \chi_j \rangle = \dfrac{1}{|G|}\sum_{g \in G} \chi_i(g)\overline{\chi_j(g)} = \delta_{ij}

UndergraduateReal-World Applications and Worked Examples

Representation theory is the mathematics of exploiting symmetry to simplify computation: chemists use character tables to predict which molecular vibrations are visible in infrared and Raman spectra without solving any differential equation directly; physicists classify elementary particles and quantum states by which irreducible representation of a symmetry group they belong to; and modern number theory (the Langlands program) is organized entirely around representations of arithmetic groups.

Example: Predicting infrared-active vibrations of water

The water molecule H2O\mathrm{H_2O} has symmetry group C2vC_{2v} of order 44 (identity, a 180°180° rotation, and two mirror reflections), with 44 one-dimensional irreducible representations. The 99-dimensional representation of atomic displacements has character χ(g)\chi(g) equal to 9,−1,1,39, -1, 1, 3 on the four group elements respectively. Using the reduction formula ni=1∣G∣∑g∈Gχ(g)χi(g)‾n_i = \dfrac{1}{|G|}\sum_{g\in G} \chi(g)\overline{\chi_i(g)}, determine how many times each irreducible representation of C2vC_{2v} appears in this 99-dimensional representation.

Solution

Step 1: recall the C2vC_{2v} character table. Its four irreducible characters A1,A2,B1,B2A_1, A_2, B_1, B_2 take values (1,1,1,1)(1,1,1,1), (1,1,−1,−1)(1,1,-1,-1), (1,−1,1,−1)(1,-1,1,-1), (1,−1,−1,1)(1,-1,-1,1) respectively on the ordered elements (identity, C2C_2, σv\sigma_v, σv′\sigma_v').

Step 2: apply the reduction formula for each irreducible. For A1A_1: nA1=14(9⋅1+(−1)⋅1+1⋅1+3⋅1)=14(12)=3n_{A_1} = \tfrac14(9\cdot1 + (-1)\cdot1 + 1\cdot1 + 3\cdot1) = \tfrac14(12) = 3. For A2A_2: nA2=14(9−1−1−3)=14(4)=1n_{A_2} = \tfrac14(9 -1 -1 -3) = \tfrac14(4) = 1. For B1B_1: nB1=14(9+1+1−3)=14(8)=2n_{B_1} = \tfrac14(9+1+1-3) = \tfrac14(8) = 2. For B2B_2: nB2=14(9+1−1+3)=14(12)=3n_{B_2} = \tfrac14(9+1-1+3) = \tfrac14(12) = 3.

Step 3: sanity-check dimensions. 3(1)+1(1)+2(1)+3(1)=93(1) + 1(1) + 2(1) + 3(1) = 9, matching the total dimension, and each nin_i came out a nonnegative integer, confirming the computation is consistent.

Step 4: interpret physically. After removing 33 translations and 33 rotations (which also transform according to specific irreducibles of C2vC_{2v}) from this 99-dimensional representation, the remaining 33 genuine vibrational modes are identified, and group theory alone — via characters, no differential equations — determines exactly which are infrared-active (those transforming like x,y,zx,y,z, i.e. like A1A_1 or B1,B2B_1,B_2 here) versus silent.

Example: Angular momentum as SU(2)SU(2) representations

In quantum mechanics, rotational symmetry means the Hamiltonian commutes with the action of SU(2)SU(2) (the double cover of SO(3)SO(3)), so every energy eigenspace is a representation of SU(2)SU(2). The irreducible representations of SU(2)SU(2) are indexed by a spin l∈{0,12,1,32,…}l \in \{0, \tfrac12, 1, \tfrac32, \ldots\} and have dimension 2l+12l+1. Explain why an electron's spin states form a 22-dimensional space, and why a pp-orbital (angular momentum l=1l=1) has exactly 33 degenerate states in the absence of external fields.

Solution

Step 1: identify the representation-theoretic dictionary. Each physically observed "degenerate multiplet" of states related by rotation is, mathematically, exactly one irreducible representation of SU(2)SU(2); the number of degenerate states is its dimension 2l+12l+1.

Step 2: electron spin. An electron's spin corresponds to l=12l=\tfrac12, the smallest nontrivial (half-integer) representation of SU(2)SU(2); its dimension is 2(12)+1=22(\tfrac12)+1 = 2, matching the two observed states "spin up" and "spin down."

Step 3: the pp-orbital. A pp-orbital carries angular momentum l=1l=1; the corresponding irreducible representation of SU(2)SU(2) has dimension 2(1)+1=32(1)+1=3.

Step 4: why they are degenerate absent a field. Without an external field, the Hamiltonian is exactly rotationally symmetric, so by Schur's Lemma the Hamiltonian acts as a scalar on each irreducible representation (it is a GG-equivariant self-map of an irreducible space) — meaning every state within the same irreducible representation shares exactly the same energy eigenvalue, i.e. the 33 states of the pp-orbital are forced to be degenerate purely by symmetry, with no dynamical calculation needed; turning on an external field (e.g. a magnetic field) breaks the SU(2)SU(2) symmetry and splits this degeneracy (the Zeeman effect).

What condition must a map ρ:G→GL(V)\rho: G \to \mathrm{GL}(V) satisfy to be a representation of GG?

By Schur's Lemma, a GG-equivariant linear map f:V→Vf: V \to V between an irreducible complex representation and itself must be:

A chemist computes the character of a 99-dimensional atomic-displacement representation and wants to know how many independent vibrational modes are infrared-active. Which tool from representation theory directly answers this, without solving any differential equation?

What is the dimension of the irreducible representation of SU(2)SU(2) corresponding to spin l=1l=1?

References

  1. Jean-Pierre Serre (1977). Linear Representations of Finite Groups
  2. William Fulton, Joe Harris (1991). Representation Theory: A First Course
  3. Dennis Gaitsgory, Sam Raskin, et al. (2024). The Proof of the Geometric Langlands Conjecture · arXiv:2405.03599
  4. Gordon D. James (1978). The Representation Theory of the Symmetric Groups