Schur's Lemmas

A representation turns an abstract group into concrete matrices: to each element $g$ it assigns an invertible operator $D(g)$ on a vector space $V$ , faithfully reproducing the multiplication table. The representation is irreducible when $V$ contains no proper subspace that every $D(g)$ leaves in place — no smaller arena on which the group already acts. Schur's Lemma is the pair of statements that pin down exactly which operators are compatible with such an indivisible structure, and it is the single most consequential result in the whole theory despite a proof that fits in a paragraph. The first lemma looks at one irreducible representation on a complex space and asks which operators $A$ commute with all of it; the answer is austere — only scalar multiples of the identity, $A=\lambda\,\mathbb{1}$ . The second lemma looks at a map $A$ between two irreducibles that intertwines their actions, and finds only two survivors: the zero map, or an isomorphism forcing the two representations to be the same one in disguise. The engine behind both is a single observation applied three ways — to eigenspaces, to kernels, and to images: each of these is automatically an invariant subspace, and an invariant subspace of an irreducible representation is either nothing or everything. From that squeeze flow the Great Orthogonality Theorem, character orthogonality, the projection operators of spectroscopy, and the statement that a Casimir invariant is constant across a multiplet. Whenever a physical operator commutes with a full symmetry group, Schur collapses its action on each irreducible sector to a single number — the cleanest expression of how symmetry organizes physics.

Key takeaways

Schur I: any operator that commutes with every matrix of a complex irreducible representation is a scalar multiple of the identity, $A=\lambda\,\mathbb{1}$ .
Schur II: an operator intertwining two irreducibles is either zero (inequivalent case) or invertible (equivalent case) — nothing in between.
Combined over ℂ, the space of intertwiners between irreps $\mu$ and $\nu$ has dimension $\delta_{\mu\nu}$ : one line if equivalent, zero otherwise.
This single dimension count is the engine of the Great Orthogonality Theorem, character orthogonality, and the block-diagonal reduction of any representation.
Physically it means a symmetry-invariant Hamiltonian is constant on each irreducible multiplet — the origin of exact degeneracies and good quantum numbers like $j(j+1)$ .
The scalar conclusion needs an algebraically closed field; over ℝ the commutant can be ℝ, ℂ, or the quaternions ℍ.
Key idea: an invariant subspace of an irrep is either trivial or the whole space, so the eigenspaces, kernels, and images of commuting or intertwining operators are forced to be all-or-nothing — which pins those operators to zero or a scalar.

The core insight

An invariant subspace of an irreducible representation is either trivial or the whole space; applying this to the eigenspaces, kernels, and images of commuting or intertwining operators forces those operators to be either zero or proportional to the identity.

Where this lives today

Schur's Lemma is quietly at work wherever symmetry meets computation. In molecular and solid-state spectroscopy the projection operators of equation (6) build symmetry-adapted linear combinations of atomic orbitals, block-diagonalizing the secular determinant so that a hopeless dense eigenvalue problem factors into small blocks labelled by irreps — the practical engine behind every point-group and space-group analysis of normal modes and molecular orbitals. In particle physics the classification of elementary particles is the statement that a single species is an irreducible representation of the Poincaré group times an internal gauge group, with mass and spin the Casimir invariants that Schur guarantees are constant across the multiplet; the quark model's $SU(3)$ multiplets and the selection rules of the Wigner–Eckart theorem are direct descendants. Condensed-matter theory uses the same dichotomy to diagnose topological degeneracies and to constrain which perturbations can split a level. And in modern computational chemistry and lattice gauge theory, exploiting Schur block-diagonalization is a standard route to reducing the cost of large symmetric eigenproblems. An active frontier is the representation theory of the infinite and continuous symmetry groups relevant to conformal field theory and to error-correcting codes, where Schur-type commutant arguments (the Schur–Weyl duality between $SU(d)$ and the symmetric group) underlie the design of decoherence-free subspaces in quantum computing.

The physical picture

An irreducible representation is a block with no internal seam. There is no way to carve $V$ into smaller pieces that the group respects, so any operator that commutes with the whole group has nowhere finer to act — it cannot single out a sub-collection of directions and treat them differently, because no invariant sub-collection exists. The only action left that is blind to every direction is multiplication by one number. The moment the representation does have a seam — when it splits into inequivalent irreducible blocks — commuting operators regain freedom, but only a rigid, block-by-block scalar freedom. The first figure contrasts the two situations and states the sharp count that separates them.

Figure

On an irreducible block the only operator commuting with the whole group is a scalar, so dim Hom_G(V,V) = 1; on a reducible block D^(μ) ⊕ D^(ν) built from inequivalent irreps the commutant is block-scalar, one independent number per block

The physical payoff is degeneracy. If a Hamiltonian commutes with every symmetry operation, then on each irreducible multiplet Schur forces $H=E_\mu\,\mathbb{1}$ — one energy for the whole multiplet, exactly degenerate until something breaks the symmetry. The same logic runs on the Casimir operator of a Lie group: because $\mathbf{J}^2$ commutes with every rotation generator, it is a single number on each spin sector, and that number $j(j+1)$ is precisely the label physicists call angular momentum. The second figure shows the rotation-invariant spectrum this produces, with the Casimir eigenvalue tagging each level.

Figure

A rotation-invariant Hamiltonian is scalar on each SU(2) multiplet, so every spin-j level is exactly (2j+1)-fold degenerate and carries the Casimir label J² = j(j+1)ħ²: j = 0 gives 1 state at 0, j = 1/2 gives 2 states at (3/4)ħ², j = 1 gives 3 states at 2ħ²

The core equations

Schur's Lemma comes in two halves. The first constrains operators that commute with a single irreducible representation; over the complex numbers the answer is as rigid as possible.

A\,D(g)=D(g)\,A\quad\forall g\in G\ \Longrightarrow\ A=\lambda\,\mathbb{1} \qquad (1)

where: $D(g)$ — operator of an irreducible representation $D$ at element $g$ ; $A$ — any linear operator on the representation space $V$ ; $\lambda$ — a complex scalar (an eigenvalue of $A$ ); $\mathbb{1}$ — the identity on $V$ . Read it as: the commutant of an irrep is one-dimensional, spanned by the identity alone. The second half concerns a map between two irreducibles that respects both actions. Such an intertwiner has no middle ground — it is either trivial or an isomorphism.

A\,D^{(1)}(g)=D^{(2)}(g)\,A\quad\forall g\ \Longrightarrow\ A=0\ \ \text{or}\ \ A\ \text{invertible} \qquad (2)

where: $D^{(1)},D^{(2)}$ — irreducible representations on $V_1,V_2$ ; $A:V_1\to V_2$ — an intertwining (equivariant) linear map. An invertible $A$ means $D^{(2)}(g)=A\,D^{(1)}(g)\,A^{-1}$ , i.e. the two representations are equivalent. Packaging both halves gives the count that everything downstream rests on: how many independent intertwiners connect two irreps.

\dim\,\mathrm{Hom}_G\!\left(V^{(\mu)},V^{(\nu)}\right)=\delta_{\mu\nu} \qquad (3)

where: $V^{(\mu)},V^{(\nu)}$ — carrier spaces of irreps labelled $\mu,\nu$ ; $\delta_{\mu\nu}$ — Kronecker delta, $1$ when the irreps are equivalent and $0$ otherwise. This one integer is the seed of every orthogonality relation below. Feeding equation (3) an averaged rank-one operator produces the Great Orthogonality Theorem — the statement that the matrix elements of the irreps form an orthogonal set of functions on the group.

\frac{1}{|G|}\sum_{g\in G} D^{(\mu)}(g)^{*}_{ij}\,D^{(\nu)}(g)_{kl}=\frac{1}{d_\mu}\,\delta_{\mu\nu}\,\delta_{ik}\,\delta_{jl} \qquad (4)

where: $|G|$ — order of the finite group; $d_\mu$ — dimension of irrep $\mu$ ; $D^{(\mu)}(g)_{ij}$ — a matrix element in a unitary basis; the star is complex conjugation. The reciprocal dimension $1/d_\mu$ is exactly the trace normalization that Schur's constant forces. Taking traces on both index pairs collapses the matrix elements to characters, giving the workhorse relation of every character table.

\frac{1}{|G|}\sum_{g\in G}\chi^{(\mu)}(g)^{*}\,\chi^{(\nu)}(g)=\delta_{\mu\nu} \qquad (5)

where: $\chi^{(\mu)}(g)=\mathrm{tr}\,D^{(\mu)}(g)$ — the character of irrep $\mu$ , a class function. The characters of distinct irreps are orthonormal under the group average, so a character table's rows form an orthonormal basis. The same averaging, applied to a general (reducible) representation, builds the projector onto its $\mu$ -isotypic part.

P^{(\mu)}=\frac{d_\mu}{|G|}\sum_{g\in G}\chi^{(\mu)}(g)^{*}\,D(g) \qquad (6)

where: $P^{(\mu)}$ — projector onto the piece of $V$ carrying copies of irrep $\mu$ ; $D(g)$ — an arbitrary representation. These operators block-diagonalize a secular problem and build symmetry-adapted bases. Finally, promoted to Lie groups, Schur says any invariant (Casimir) operator is a constant on each irrep — the constant being the invariant label of that irrep.

[C,\,D(g)]=0\ \ \forall g\ \Longrightarrow\ C=c_\mu\,\mathbb{1}\ \ \text{on each irrep }\mu \qquad (7)

where: $C$ — an operator commuting with the whole representation (e.g. the quadratic Casimir $\mathbf{J}^2$ ); $c_\mu$ — its scalar value on irrep $\mu$ , such as $j(j+1)$ for $SU(2)$ .

Derivation AProof of Schur's Lemma Part One

The goal is to show that over the complex numbers any operator $A$ commuting with every matrix of an irreducible representation must be a scalar multiple of the identity, using nothing beyond two facts: that every operator on a finite-dimensional complex space has at least one eigenvalue, and that a nonzero invariant subspace of an irrep must be the whole space. The strategy is to build from $A$ a companion operator whose kernel is guaranteed nonzero, show that kernel is invariant, and let irreducibility swallow it.

Step 1. Over ℂ the characteristic polynomial of $A$ has a root, so there is an eigenvalue $\lambda$ and a nonzero eigenvector: $A\,v=\lambda\,v$ with $v\neq0$ . Why: algebraic closure of ℂ is exactly what guarantees an eigenvalue exists, and this is the one place the field matters.

Step 2. Define the shifted operator $B\equiv A-\lambda\,\mathbb{1}$ . Since $\mathbb{1}$ commutes with everything, $B$ inherits the commutation property: $B\,D(g)=D(g)\,B$ for all $g$ . By construction $B\,v=0$ , so $\ker B$ is nonzero. Why: subtracting the eigenvalue manufactures an operator with a guaranteed nontrivial kernel while preserving equivariance.

Step 3. Show $\ker B$ is $G$ -invariant. If $B\,w=0$ , then $B\,(D(g)\,w)=D(g)\,B\,w=D(g)\,0=0$ , so $D(g)\,w\in\ker B$ too. The kernel is closed under every $D(g)$ . Why: equivariance turns " $w$ is killed by $B$ " into " $D(g)w$ is killed by $B$ ", which is precisely invariance of the kernel.

Step 4. Apply irreducibility. The subspace $\ker B$ is invariant and nonzero, and an irrep admits no invariant subspace other than $\{0\}$ and $V$ ; being nonzero, it must be all of $V$ . Hence $B=0$ on $V$ , i.e. $A=\lambda\,\mathbb{1}$ . Why: this is the squeeze — irreducibility leaves the kernel no room to be a proper subspace, forcing $B$ to vanish everywhere.

Go deeper · Derivation B · 1 minProof of Schur's Lemma Part Two

Derivation BProof of Schur's Lemma Part Two

The goal is to prove that a linear map $A:V_1\to V_2$ intertwining two irreducible representations, $A\,D^{(1)}(g)=D^{(2)}(g)\,A$ , is either the zero map or an invertible one, with the invertible case forcing the two irreps to be equivalent. The strategy exploits the same all-or-nothing dichotomy twice: once on the kernel of $A$ inside $V_1$ , and once on the image of $A$ inside $V_2$ , because equivariance makes both of them invariant subspaces.

Step 1. The kernel is invariant. If $v\in\ker A$ , then $A\,D^{(1)}(g)\,v=D^{(2)}(g)\,A\,v=D^{(2)}(g)\,0=0$ , so $D^{(1)}(g)\,v\in\ker A$ . Thus $\ker A$ is a $D^{(1)}$ -invariant subspace of $V_1$ . Why: the intertwining identity converts group action inside the domain into group action that still lands in the kernel.

Step 2. By irreducibility of $V_1$ , either $\ker A=V_1$ (then $A=0$ , and we are done) or $\ker A=\{0\}$ (then $A$ is injective). Why: an invariant subspace of an irrep is trivial or everything, so the kernel has only these two sizes.

Step 3. The image is invariant. For any $w=A\,v$ in $\mathrm{im}\,A$ , $D^{(2)}(g)\,w=D^{(2)}(g)\,A\,v=A\,D^{(1)}(g)\,v\in\mathrm{im}\,A$ . So $\mathrm{im}\,A$ is a $D^{(2)}$ -invariant subspace of $V_2$ . Why: the same identity read the other way keeps the image closed under $D^{(2)}$ .

Step 4. By irreducibility of $V_2$ , either $\mathrm{im}\,A=\{0\}$ (again $A=0$ ) or $\mathrm{im}\,A=V_2$ ( $A$ is surjective). For a nonzero $A$ we already have injectivity from Step 2, and now surjectivity, so $A$ is a bijection with $A^{-1}$ existing. Then $D^{(2)}(g)=A\,D^{(1)}(g)\,A^{-1}$ : the representations are equivalent. Why: injective plus surjective is invertible, and an invertible intertwiner is exactly a change of basis relating the two irreps.

Go deeper · Derivation C · 1 minThe Great Orthogonality Theorem from Schur

Derivation CThe Great Orthogonality Theorem from Schur

The goal is to derive the Great Orthogonality Theorem, equation (4), directly from the two lemmas by averaging an arbitrary matrix over the group to manufacture an intertwiner, then using Schur to collapse that intertwiner to zero or a scalar and fixing the scalar by a trace. This is the canonical demonstration that orthogonality is not an independent axiom but a corollary of the indivisibility of irreps.

Step 1. Pick any fixed matrix $X:V^{(\mu)}\to V^{(\nu)}$ and average it against the two representations: $A=\frac{1}{|G|}\sum_{g} D^{(\nu)}(g)\,X\,D^{(\mu)}(g)^{-1}$ . A short relabelling $g\to hg$ shows $A\,D^{(\mu)}(h)=D^{(\nu)}(h)\,A$ , so $A$ is an intertwiner by construction. Why: group-averaging is the standard trick that turns any operator into an equivariant one, the raw material Schur needs.

Step 2. Apply the lemmas. If $\mu\neq\nu$ the irreps are inequivalent, so by Part Two $A=0$ ; if $\mu=\nu$ then by Part One $A=c\,\mathbb{1}$ . Both cases are captured by $A=\delta_{\mu\nu}\,c\,\mathbb{1}$ . Why: Schur reduces an entire group average to a single Kronecker delta times one unknown constant.

Step 3. Fix the constant with a trace. Taking the trace of $A=c\,\mathbb{1}$ on the $d_\mu$ -dimensional space gives $c\,d_\mu=\mathrm{tr}\,A=\mathrm{tr}\,X$ (the average preserves the trace), so $c=\tfrac{1}{d_\mu}\,\mathrm{tr}\,X$ . Why: the trace is basis-independent and cyclic, so it survives the averaging untouched and delivers the normalization $1/d_\mu$ .

Step 4. Choose $X=E_{jl}$ , the matrix with a single unit entry in row $j$ , column $l$ . Reading off the $(i,k)$ matrix element and using a unitary basis so $D(g)^{-1}=D(g)^{\dagger}$ turns the scalar statement into $\frac{1}{|G|}\sum_g D^{(\mu)}(g)^{*}_{ij}D^{(\nu)}(g)_{kl}=\frac{1}{d_\mu}\delta_{\mu\nu}\delta_{ik}\delta_{jl}$ . Why: probing with each unit matrix in turn extracts the full index structure of equation (4) one entry at a time. *As a consistency check, setting $\mu=\nu$ and $X=\mathbb{1}$ in the averaged operator gives $\sum_g|\chi^{(\mu)}(g)|^2=|G|$ , the standard norm-squared relation for characters.*

Limiting cases

Schur's Lemma is sharp: its conclusion changes decisively with the field and with whether the representation is truly irreducible.

Case	Result	Meaning
Complex irrep	$\mathrm{Hom}_G(V,V)\cong\mathbb{C}$	Over ℂ the commutant is one-dimensional — only scalar multiples of $\mathbb{1}$ commute with the whole irrep.
Inequivalent complex irreps	$\mathrm{Hom}_G(V_1,V_2)=\{0\}$	No nonzero intertwiner exists; this vanishing is the foundation of every orthogonality relation.
Real irreps	$\mathrm{Hom}_G(V,V)\in\{\mathbb{R},\,\mathbb{C},\,\mathbb{H}\}$	Over ℝ the commutant is a real division algebra (real, complex, or quaternionic type); the scalar conclusion can fail.
Reducible, multiplicity-free	$A=\bigoplus_\mu\lambda_\mu\,\mathbb{1}_{d_\mu}$	Distinct irreducible blocks each get their own scalar; the commutant is block-scalar, one number per block.
Reducible with repeats	$A=\bigoplus_\mu(M_\mu\otimes\mathbb{1}_{d_\mu})$	If irrep $\mu$ appears $m_\mu$ times, a full $m_\mu\times m_\mu$ matrix $M_\mu$ on the multiplicity space commutes — the commutant is a matrix algebra.

Common trapthe scalars-only conclusion needs an algebraically closed field

Part One says a commuting operator is a scalar, but that rests entirely on Step 1 of its proof — the existence of an eigenvalue — which is guaranteed only over an algebraically closed field such as ℂ. Over the reals an irreducible representation can have a commutant equal to ℝ, ℂ, or the quaternions ℍ (the real, complex, and quaternionic types classified by the Frobenius–Schur indicator), so a commuting operator need not be a real scalar. A companion trap is confusing irreducible with indecomposable: over fields of positive characteristic, or for non-semisimple algebras, a representation can be indecomposable (not a direct sum of smaller pieces) yet still contain a proper invariant subspace, and Schur's clean dichotomy no longer applies. Always confirm you are over ℂ and genuinely irreducible before quoting " $A=\lambda\mathbb{1}$ ".

Worked examples

1 · Decomposing the permutation representation of S₃ (fully worked)

The symmetric group $S_3$ has order $6$ and three conjugacy classes: the identity $\{e\}$ (1 element), the two 3-cycles $\{(123),(132)\}$ (2 elements), and the three transpositions (3 elements). Its three irreducibles are the trivial $A_1$ , the sign $A_2$ , and the 2-dimensional standard rep $E$ , with characters below. Let $D$ be the natural 3-dimensional representation that permutes three coordinate axes. Decompose $D$ into irreducibles.

Step 1. Record the character table. On the classes $(e,\ \text{3-cycle},\ \text{transposition})$ the irreps read $\chi^{A_1}=(1,1,1)$ , $\chi^{A_2}=(1,1,-1)$ , $\chi^{E}=(2,-1,0)$ . As a sanity check the dimensions satisfy $d_{A_1}^2+d_{A_2}^2+d_E^2=1+1+4=6=|G|$ , so the list of irreps is complete. The three irreps exhaust the group.

Step 2. Find the character of $D$ . A permutation matrix's trace is its number of fixed points, so $\chi^D(e)=3$ (all axes fixed), $\chi^D(\text{3-cycle})=0$ (a 3-cycle fixes nothing), and $\chi^D(\text{transposition})=1$ (a swap fixes one axis). Thus $\chi^D=(3,0,1)$ on the three classes.

Step 3. Apply the multiplicity formula $a_\mu=\frac{1}{|G|}\sum_{g}\chi^{(\mu)}(g)^{*}\chi^D(g)$ , weighting each class by its size $(1,2,3)$ :

a_{A_1}=\tfrac{1}{6}\big[1\cdot1\cdot3+2\cdot1\cdot0+3\cdot1\cdot1\big]=\tfrac{1}{6}(3+0+3)=1

a_{A_2}=\tfrac{1}{6}\big[1\cdot1\cdot3+2\cdot1\cdot0+3\cdot(-1)\cdot1\big]=\tfrac{1}{6}(3+0-3)=0

a_{E}=\tfrac{1}{6}\big[1\cdot2\cdot3+2\cdot(-1)\cdot0+3\cdot0\cdot1\big]=\tfrac{1}{6}(6+0+0)=1

Step 4. Read off the result: $D\cong A_1\oplus E$ , with the sign representation absent. The dimensions check, $1+2=3=\dim D$ , and the physical meaning is transparent — the $A_1$ piece is the fully symmetric "sum of the three axes" direction, and $E$ is the complementary 2-dimensional plane. The permutation representation is $A_1 \oplus E$ ; every step used character orthogonality, itself a Schur corollary.

2 · The SU(2) Casimir as a Schur scalar (fully worked)

The generators of $SU(2)$ in the spin- $\tfrac12$ representation are $J_a=\tfrac{\hbar}{2}\sigma_a$ with $\sigma_a$ the Pauli matrices. Show directly that the Casimir $\mathbf{J}^2=J_x^2+J_y^2+J_z^2$ is a scalar, and check its value against the general formula $j(j+1)\hbar^2$ .

Step 1. Use $\sigma_a^2=\mathbb{1}$ for each Pauli matrix. Then $\mathbf{J}^2=\tfrac{\hbar^2}{4}(\sigma_x^2+\sigma_y^2+\sigma_z^2)=\tfrac{\hbar^2}{4}(\mathbb{1}+\mathbb{1}+\mathbb{1})=\tfrac{3}{4}\hbar^2\,\mathbb{1}$ . So the Casimir is manifestly $\tfrac34\hbar^2$ times the identity — a scalar, exactly as Schur demands for an operator commuting with all generators.

Step 2. Compare with $j(j+1)\hbar^2$ for $j=\tfrac12$ : $j(j+1)=\tfrac12\cdot\tfrac32=\tfrac34$ , giving $\mathbf{J}^2=\tfrac34\hbar^2\,\mathbb{1}$ . The two agree. The spin-½ Casimir is (3/4)ħ²·𝟙.

Step 3. Confirm the mechanism. $\mathbf{J}^2$ commutes with each $J_a$ (the defining property of a Casimir), hence with every $D(g)=e^{-i\theta\hat n\cdot\mathbf{J}/\hbar}$ . Since the spin- $j$ representation is irreducible, Part One forces $\mathbf{J}^2$ to be a single scalar on it — there is no room for it to distinguish the $2j+1$ states. That is why $j(j+1)$ is a good quantum number: it is constant across the whole multiplet.

Step 4. Sanity-check a second value. For the adjoint (spin- $1$ ) representation, $j(j+1)=1\cdot2=2$ , so $\mathbf{J}^2=2\hbar^2\,\mathbb{1}$ on the 3-dimensional multiplet — matching the figure in the physical picture. Each irrep is tagged by one Casimir number, and Schur is what makes that number well defined.

Physics of the portrait

Schur's Lemma is named for the mathematician who made it the cornerstone of representation theory, turning a technical observation about commuting matrices into the organizing principle of symmetry in physics.

Origins · 1 minIssai Schur and how this was discovered

Issai Schur

1875–1941 · representation theory of finite and compact groups

Issai Schur was born in Mogilev in the Russian Empire and became the central figure, alongside his teacher Ferdinand Georg Frobenius, in building the representation theory of groups in early twentieth-century Berlin. Frobenius had introduced group characters in 1896 through the mysterious "group determinant"; Schur's contribution, crystallized in his 1905 paper Neue Begründung der Theorie der Gruppencharaktere, was to rebuild the entire theory on a cleaner and more powerful foundation — and the lemma that now bears his name was the keystone of that reconstruction. From it he derived the orthogonality relations directly, without Frobenius's determinantal machinery, and the resulting toolkit (characters, orthogonality, the multiplicity formula) is essentially the one still taught today. With Frobenius he introduced in 1906 the indicator that classifies irreducible representations into real, complex, and quaternionic types, delineating exactly where the "scalars only" conclusion holds over the reals. Schur was a revered lecturer in Berlin whose seminar trained a generation of algebraists, but as a Jewish professor he was stripped of his position under the Nazi race laws in 1935, forced out of the Prussian Academy, and emigrated to Palestine in 1939, dying in Tel Aviv in 1941. His lemma outlived that catastrophe to become indispensable: every time a physicist says a symmetry-invariant operator is constant on a multiplet, they are invoking Schur.

History

1896

— Frobenius introduces group characters via the group determinant

1905

— Schur rebuilds representation theory and states his lemma as its keystone

1906

— Frobenius and Schur classify real, complex, and quaternionic irreps

1927

— Peter and Weyl extend orthogonality and completeness to compact Lie groups

1931

— Wigner applies irreducible representations systematically to atomic spectra

Going further

Texts. Tinkham, Group Theory and Quantum Mechanics, ch. 3 (the cleanest physicist's proof) · Hamermesh, Group Theory and Its Application to Physical Problems, ch. 3–4 · Stone & Goldbart, Mathematics for Physics, ch. on group representations · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., ch. on group theory · Georgi, Lie Algebras in Particle Physics, ch. 1–2 (Casimir and Schur in the Lie setting) · Serre, Linear Representations of Finite Groups (the rigorous algebraist's account). Key reference values: the $S_3$ character table has $\chi^{A_1}=(1,1,1)$ , $\chi^{A_2}=(1,1,-1)$ , $\chi^E=(2,-1,0)$ with class sizes $(1,2,3)$ , and the 3-dimensional permutation representation $\chi^D=(3,0,1)$ decomposes into $A_1\oplus E$ via $a_\mu=|G|^{-1}\sum_g\chi^{(\mu)*}\chi^D$ . The $SU(2)$ Casimir takes the value $\mathbf{J}^2=\tfrac{3}{4}\hbar^2\mathbb{1}$ for $j=\tfrac{1}{2}$ and $2\hbar^2\mathbb{1}$ for $j=1$ , consistent with the general formula $j(j+1)\hbar^2$ . No nonstandard numbers are introduced.

Prerequisites. Group Representations · Reducible and Irreducible Representations · Homomorphisms and Quotient Groups. Leads to. Characters and Orthogonality Theorems · Character Tables and Decomposition of Representations · Tensor Products and Clebsch–Gordan Decomposition · SU(2) and the Spin Double Cover.

Ask the tutor: Ask me to walk through why a symmetry-invariant Hamiltonian is forced to be constant on a multiplet, and we will build the Schur argument from a concrete character table together.

What is the difference between Schur's Lemma Part One and Part Two?

They cover two different setups. Part One is about a single irreducible representation and an operator $A$ that commutes with all of it; the conclusion is $A=\lambda\mathbb{1}$ , a scalar. Part Two is about two irreducibles and an operator $A$ that intertwines them, $A\,D^{(1)}=D^{(2)}\,A$ ; the conclusion is that $A$ is either zero or an isomorphism. Part One is really the special case of Part Two where the two representations coincide, combined with the extra fact that over ℂ an eigenvalue exists. Together they yield the master statement $\dim\mathrm{Hom}_G(V^\mu,V^\nu)=\delta_{\mu\nu}$ , which is what powers the orthogonality relations.

Why does the proof insist on complex scalars?

Because Part One needs $A$ to have at least one eigenvalue, so that $A-\lambda\mathbb{1}$ has a nonzero kernel to feed to irreducibility. Over an algebraically closed field like ℂ every operator has an eigenvalue — the characteristic polynomial always factors — so the argument goes through. Over the reals the characteristic polynomial may have no real root (think of a rotation matrix), and then no real $\lambda$ makes $A-\lambda\mathbb{1}$ singular. The commutant can consequently be larger: real, complex, or quaternionic. This is not a defect of the physics but a genuine structural fact, catalogued by the Frobenius–Schur indicator, and it is why quantum mechanics is naturally phrased over ℂ.

How does Schur's Lemma explain degeneracy in quantum mechanics?

If a Hamiltonian $H$ commutes with every operator $D(g)$ of a symmetry group, then on any irreducible multiplet $H$ is an operator commuting with an irrep, so Part One forces $H=E_\mu\mathbb{1}$ there — a single energy for the whole multiplet. Every state in that multiplet therefore has exactly the same energy: the degeneracy is protected by symmetry, not by accident, and it persists precisely until a perturbation breaks the symmetry and lets the level split. This is the group-theoretic reason a hydrogen level of given orbital angular momentum has its $2\ell+1$ states degenerate, and why turning on a magnetic field (breaking rotational symmetry) splits them.

What exactly is an intertwiner, and why do we care about counting them?

An intertwiner between representations $D^{(1)}$ and $D^{(2)}$ is a linear map $A$ that commutes with the group action in the sense $A\,D^{(1)}(g)=D^{(2)}(g)\,A$ for all $g$ — a map that respects symmetry. The dimension of the space of such maps, $\dim\mathrm{Hom}_G$ , measures how much the two representations "overlap". Schur says this number is $0$ for inequivalent irreps and $1$ for equivalent ones. Counting intertwiners is powerful because it converts questions about representations into simple integer arithmetic: the multiplicity of an irrep in a big representation is an intertwiner count, and so is the vanishing of a matrix element that gives a selection rule.

Does Schur's Lemma still hold for continuous groups like SU(2)?

Yes, with mild functional-analytic care. For compact groups such as $SU(2)$ or $SO(3)$ one replaces the finite average $\frac{1}{|G|}\sum_g$ by an integral against the invariant Haar measure, and finite-dimensional irreducible representations again have one-dimensional commutants. The Peter–Weyl theorem packages the resulting orthogonality and completeness. The main new subtlety appears for non-compact groups (like the Poincaré group), where irreps are infinite-dimensional and one must use a version of Schur's Lemma for unitary representations on Hilbert space; there it still delivers that Casimir invariants act as scalars, which is exactly what defines a particle's mass and spin.

If two equivalent irreps have an invertible intertwiner, is it unique?

It is unique up to a single overall scalar. Part Two hands you one invertible intertwiner $A$ . Suppose $A'$ is another intertwiner between the same pair. Then $A^{-1}A'$ commutes with the first irrep (a short check using the intertwining identities), so Part One makes it a scalar: $A^{-1}A'=\lambda\mathbb{1}$ , i.e. $A'=\lambda A$ . Hence the whole space of intertwiners is the one-dimensional line spanned by $A$ , consistent with $\dim\mathrm{Hom}_G=\delta_{\mu\nu}=1$ . Physically this is why a change of basis relating two copies of the same multiplet is fixed once you set one overall phase and normalization.