Characters and Orthogonality Theorems

A representation $D$ of a finite group $G$ hands every element $g$ a matrix $D(g)$ , but those matrices are written in a chosen basis, and a change of basis reshuffles all their entries at once. The one quantity that survives every reshuffling is the trace — which has a name of its own: the character, $\chi(g)=\mathrm{tr}\,D(g)$ . It is a single complex number per element, far simpler than the full matrix, yet it retains all that matters: two representations with equal characters are equivalent, so it is a complete fingerprint free of coordinate noise. Because the trace is cyclic, $\chi(hgh^{-1})=\chi(g)$ , the character is constant on each conjugacy class — it is a class function. Class functions live in a space whose dimension equals the number of conjugacy classes, and this article's central discovery is that the irreducible characters form an orthonormal basis of that space. Every question about representations then collapses into linear algebra: how a representation decomposes, whether it is irreducible, and how many irreducible pieces exist are answered by inner products of characters. The engine is Schur's Lemma, which forces averaged intertwining operators to be scalars or zero; from it flows the Great Orthogonality Theorem, whose trace is the orthonormality of characters. In physics these relations split degenerate levels, enforce selection rules, build symmetry-adapted bases, and decompose tensor products — and carry verbatim to compact Lie groups once the group sum becomes a Haar integral.

Key takeaways

The character is the trace $\chi(g)=\mathrm{tr}\,D(g)$ ; because the trace is conjugation-invariant, $\chi$ is a class function, constant on each conjugacy class.
The Great Orthogonality Theorem makes the matrix elements of inequivalent unitary irreps orthogonal over the group; tracing it gives the row orthogonality of characters, $\frac{1}{|G|}\sum_g\chi^{(\alpha)}(g)^*\chi^{(\beta)}(g)=\delta_{\alpha\beta}$ .
The dual column orthogonality forces the character table to be square: the number of irreps equals the number of conjugacy classes.
Any representation decomposes as $\bigoplus_\alpha m_\alpha\rho_\alpha$ with $m_\alpha=\langle\chi,\chi^{(\alpha)}\rangle$ , and it is irreducible if and only if $\langle\chi,\chi\rangle=1$ .
Decomposing the regular representation gives the dimension sum rule $\sum_\alpha d_\alpha^2=|G|$ — for $S_3$ this is $1^2+1^2+2^2=6$ , fixing the irrep dimensions $1,1,2$ .
Key idea: the trace erases the basis-dependence of representation matrices, leaving class functions; the irreducible characters are an orthonormal basis of class-function space, so projecting any character onto them reads off its decomposition into irreps.

The core insight

The trace erases the basis-dependence of representation matrices, leaving class functions; the irreducible characters form an orthonormal basis of class-function space, so projecting any character onto them reads off its decomposition into irreducible representations.

Where this lives today

Character orthogonality runs quietly underneath a great deal of working physics. In molecular vibrational spectroscopy it decides which normal modes appear in infrared and Raman spectra: the $3N-6$ vibrations form a reducible representation of the point group, and projecting its character onto the irreps sorts the modes by symmetry — ammonia ( $C_{3v}$ , isomorphic to the $S_3$ used above) splits into $2A_1\oplus2E$ , fixing how many lines a spectroscopist sees. In ligand-field theory the same arithmetic splits atomic levels: a $d$ -shell in an octahedral field decomposes as $e_g\oplus t_{2g}$ , the multiplets whose gap sets the color of transition-metal complexes and the high-spin/low-spin distinction. Solid-state band theory carries the idea to the 230 space groups, labeling electronic bands at high-symmetry points of the Brillouin zone by irreps — the basis of the "symmetry indicators" and topological quantum chemistry programs now driving high-throughput searches for topological insulators. And in particle physics the compact-group version organizes the hadron spectrum through $SU(3)$ flavor multiplets and the Clebsch–Gordan tables of angular-momentum coupling. Wherever a symmetry acts, its character table is the lookup key.

The physical picture

Forget the matrices and keep only their traces. Each irreducible representation becomes a short list of numbers — one per conjugacy class — and these lists behave like perpendicular unit vectors. The character table is just these lists stacked into a square array, and the orthogonality theorems say it is essentially unitary: rows orthonormal, columns orthogonal. The smallest non-abelian example, $S_3$ (equivalently $C_{3v}$ ), already shows every feature.

Figure

Character table of S₃ (equivalently C₃ᵥ): three irreducible representations A₁, A₂, E across three conjugacy classes e, flips, rotations. The dimension sum rule reads 1² + 1² + 2² = 6 = group order, and every pair of rows is orthonormal under the class-weighted inner product, e.g. ⟨A₁,E⟩ = (1·1·2 + 3·1·0 + 2·1·(−1))/6 = 0

The second picture is geometric. With $\langle f,h\rangle=\frac{1}{|G|}\sum_g f(g)^*h(g)$ making function-space a Hilbert space, the class functions form a subspace whose dimension is the number of classes, and the irreducible characters are its orthonormal axes $\hat e_{A_1},\hat e_{A_2},\hat e_E$ . Decomposing a representation is just resolving its character vector along these axes, the component along $\hat e_\alpha$ being the multiplicity $m_\alpha$ . The permutation character $(3,1,0)$ resolves into one part along $\hat e_{A_1}$ and one along $\hat e_E$ — trivial plus standard, nothing else.

Figure

Class functions of S₃ as a 3D Hilbert space: the irreducible characters χ_A₁, χ_A₂, χ_E are orthonormal axes. The permutation character χ = (3,1,0) is the vector A₁ ⊕ E with coordinates (1,0,1), so ⟨χ,χ⟩ = 1² + 0² + 1² = 2 counts two irreducible pieces

The two relations are dual faces of one square array: rows (irreps) and columns (classes) are both perpendicular, forcing the number of irreps to equal the number of classes. Physically the self-overlap $\langle\chi,\chi\rangle=\sum_\alpha m_\alpha^2$ counts the irreducible multiplets in a state space, each squared — it equals $1$ exactly when the level is a single, symmetry-protected, unsplittable multiplet.

The core equations

The definition is disarmingly simple: sum the diagonal of the representation matrix and discard the rest. What remains is basis-independent — the character keeps only what a change of basis cannot touch.

\chi^{(\alpha)}(g)=\mathrm{tr}\,D^{(\alpha)}(g)=\sum_{i=1}^{d_\alpha} D^{(\alpha)}(g)_{ii} \qquad (1)

where: $D^{(\alpha)}(g)$ — the matrix of irrep $\alpha$ for element $g$ ; $d_\alpha$ — the dimension of irrep $\alpha$ ; $\chi^{(\alpha)}$ — its character, a function on $G$ . Reading it: at the identity $\chi^{(\alpha)}(e)=\mathrm{tr}\,\mathbb{1}=d_\alpha$ , so the identity column reads off the dimensions. The trace is cyclic, so conjugating $g$ leaves its character unchanged — demoting the character from a function on $|G|$ elements to one on the far smaller set of conjugacy classes.

\chi(hgh^{-1})=\mathrm{tr}\,[D(h)D(g)D(h)^{-1}]=\mathrm{tr}\,[D(h)^{-1}D(h)D(g)]=\chi(g) \qquad (2)

where: $h,g$ — any group elements. Reading it: a character is a class function, one value per conjugacy class, so the table has exactly one column per class. The Great Orthogonality Theorem is the master statement from which everything follows. It treats each matrix entry $D^{(\alpha)}(g)_{ij}$ as a vector (a function of $g$ ) in $|G|$ -dimensional space and asserts these are mutually orthogonal.

\sum_{g\in G} D^{(\alpha)}(g)_{ij}\,D^{(\beta)}(g)^{*}_{kl}=\frac{|G|}{d_\alpha}\,\delta_{\alpha\beta}\,\delta_{ik}\,\delta_{jl} \qquad (3)

where: $|G|$ — the order of the group; $D^{(\alpha)},D^{(\beta)}$ — inequivalent unitary irreps unless $\alpha=\beta$ ; $i,j,k,l$ — matrix indices. Reading it: the entry-functions of all irreps form an orthogonal set in $|G|$ -dimensional space, so $\sum_\alpha d_\alpha^2\le|G|$ , later an equality. Tracing the theorem — setting $j=i$ , $l=k$ and summing — collapses the four indices onto the characters, the workhorse relation of the subject.

\langle\chi^{(\alpha)},\chi^{(\beta)}\rangle\equiv\frac{1}{|G|}\sum_{g\in G}\chi^{(\alpha)}(g)^{*}\,\chi^{(\beta)}(g)=\frac{1}{|G|}\sum_{i}|C_i|\,\chi^{(\alpha)}(C_i)^{*}\,\chi^{(\beta)}(C_i)=\delta_{\alpha\beta} \qquad (4)

where: $\langle\cdot,\cdot\rangle$ — the normalized inner product on class functions; $C_i$ — the $i$ -th conjugacy class; $|C_i|$ — its size. Reading it: irreducible characters are orthonormal unit vectors; the class-sum form on the right — one term per class, weighted by $|C_i|$ — is what you compute. The dual relation runs down the columns: fixing two classes and summing over all irreps gives a second orthogonality, making the number of irreps equal the number of classes.

\sum_{\alpha}\chi^{(\alpha)}(C_i)^{*}\,\chi^{(\alpha)}(C_j)=\frac{|G|}{|C_i|}\,\delta_{ij} \qquad (5)

where: $\sum_\alpha$ — the sum over all inequivalent irreps. Reading it: distinct columns are orthogonal and a column's self-overlap is $|G|/|C_i|$ ; for $S_3$ the identity column gives $1^2+1^2+2^2=6$ . Decomposing the regular representation — where $G$ acts on itself by left multiplication — pins down every irrep dimension at once, since it contains each irrep exactly $d_\alpha$ times.

\chi^{\mathrm{reg}}(g)=|G|\,\delta_{g,e}\quad\Longrightarrow\quad \sum_{\alpha} d_\alpha^{2}=|G|,\qquad \#\{\text{irreps}\}=\#\{\text{classes}\} \qquad (6)

where: $\chi^{\mathrm{reg}}$ — the regular character, nonzero only at the identity, where it equals $|G|$ . Reading it: finding all irrep dimensions of a small group reduces to solving $\sum d_\alpha^2=|G|$ in positive integers. Finally, orthogonality turns any reducible character into its multiplicities by a single projection, and gives a one-line irreducibility test.

m_\alpha=\langle\chi,\chi^{(\alpha)}\rangle=\frac{1}{|G|}\sum_i |C_i|\,\chi(C_i)\,\chi^{(\alpha)}(C_i)^{*},\qquad \langle\chi,\chi\rangle=\sum_\alpha m_\alpha^2 \qquad (7)

where: $\chi$ — the character of a general representation; $m_\alpha$ — the non-negative integer multiplicity of irrep $\alpha$ . Reading it: $\langle\chi,\chi\rangle=1$ forces a single unit multiplicity (irreducible); any larger value counts and squares the pieces.

Derivation AThe Great Orthogonality Theorem from Schur's Lemma

The goal is to prove equation (3): the matrix elements of inequivalent unitary irreps are orthogonal functions on the group with a definite normalization. The strategy is Schur's — manufacture an operator that intertwines two irreps by averaging an arbitrary map over the group, use Schur's Lemma to force it to zero or a scalar, fix the scalar by tracing, and read the result off component by component using unitarity. Character orthonormality, the dimension sum rule, and decomposition all follow as corollaries.

Step 1. Let $D^{(\alpha)},D^{(\beta)}$ be unitary irreps of dimensions $d_\alpha,d_\beta$ . For an arbitrary $d_\alpha\times d_\beta$ matrix $X$ , form the group average $M=\frac{1}{|G|}\sum_{g}D^{(\alpha)}(g)\,X\,D^{(\beta)}(g)^{-1}$ . Why: averaging symmetrizes $X$ into exactly the kind of intertwiner Schur's Lemma constrains.

Step 2. Check intertwining: $D^{(\alpha)}(h)\,M=M\,D^{(\beta)}(h)$ for every $h$ , which follows by substituting $g\to hg$ , a relabeling of the group. Why: this twisted commutation is precisely the hypothesis of Schur's Lemma.

Step 3. Apply Schur. Lemma I forces $M=0$ when $\alpha\neq\beta$ ; Lemma II forces $M=\lambda\mathbb{1}$ when $\alpha=\beta$ . Why: irreducibility leaves no room between zero and a pure scalar.

Step 4. Fix the scalar by tracing. For $\alpha=\beta$ , $\lambda\,d_\alpha=\mathrm{tr}\,M=\frac{1}{|G|}\sum_g\mathrm{tr}\,X=\mathrm{tr}\,X$ , so $\lambda=\mathrm{tr}\,X/d_\alpha$ . Why: the trace is conjugation-invariant, so every term shares the trace $\mathrm{tr}\,X$ and the sum collapses.

Step 5. Choose $X=E_{jl}$ , with a single $1$ in position $(j,l)$ , so $\mathrm{tr}\,X=\delta_{jl}$ . Reading the $(i,k)$ entry and using unitarity $D^{(\beta)}(g)^{-1}_{lk}=D^{(\beta)}(g)^{*}_{kl}$ gives equation (3). Why: single-entry matrices extract each component independently, and unitarity converts the inverse into a conjugate so both matrix elements are compared under the same inner product. *Why it closes: setting $i=j,\ k=l$ and summing recovers character orthonormality, and the bound $\sum_\alpha d_\alpha^2\le|G|$ is saturated by $S_3$ , where $1^2+1^2+2^2=6$ .*

Go deeper · Derivation B · 1 minFrom matrix elements to character orthonormality and decomposition

Derivation BFrom matrix elements to character orthonormality and decomposition

The goal is to descend from the Great Orthogonality Theorem to the two facts a physicist actually uses: that irreducible characters are orthonormal, and that projecting any character onto them returns the multiplicities of its irreducible constituents. This is where the theory becomes practical — a reducible representation is reduced to a list of integers by a few inner products.

Step 1. In equation (3) set $j=i$ and $l=k$ and sum over $i,k$ . The left side becomes $\sum_g\big(\sum_i D^{(\alpha)}(g)_{ii}\big)\big(\sum_k D^{(\beta)}(g)^{*}_{kk}\big)=\sum_g\chi^{(\alpha)}(g)\chi^{(\beta)}(g)^{*}$ . Why: summing the diagonal is taking the trace, so contracting the free indices turns matrix elements into characters.

Step 2. The right side becomes $\frac{|G|}{d_\alpha}\delta_{\alpha\beta}\sum_i 1=|G|\,\delta_{\alpha\beta}$ ; dividing by $|G|$ gives $\langle\chi^{(\alpha)},\chi^{(\beta)}\rangle=\delta_{\alpha\beta}$ , equation (4). Why: the double delta forces $i=k$ , and $\sum_i 1=d_\alpha$ cancels the $1/d_\alpha$ exactly, leaving orthonormality with no leftover factor.

Step 3. Any finite-group representation is completely reducible, $D=\bigoplus_\alpha m_\alpha D^{(\alpha)}$ , and traces add block by block, so $\chi(g)=\sum_\alpha m_\alpha\chi^{(\alpha)}(g)$ . Why: a block-diagonal trace is the sum of block traces, so a character is a non-negative integer combination of irreducible characters.

Step 4. Project: $\langle\chi,\chi^{(\beta)}\rangle=\sum_\alpha m_\alpha\delta_{\alpha\beta}=m_\beta$ , and $\langle\chi,\chi\rangle=\sum_\alpha m_\alpha^2$ . Why: orthonormality collapses each projection onto a single multiplicity, and the self-overlap becomes a sum of squares of integers.

Step 5. The irreducibility test follows: $\sum_\alpha m_\alpha^2=1$ has only one solution in non-negative integers — a single $m_\alpha=1$ — so a representation is irreducible if and only if its character has unit norm. Why: a sum of squared integers equals one only when exactly one term is present. *Why it works: the $S_3$ permutation character $(3,1,0)$ on classes of sizes $(1,3,2)$ gives $\langle\chi,\chi\rangle=\frac{1}{6}(9+3+0)=2$ ; the projections $m_{A_1}=1,\ m_{A_2}=0,\ m_E=1$ identify it as trivial $\oplus$ standard.*

Go deeper · Derivation C · 1 minThe regular representation, the dimension sum rule, and column orthogonality

Derivation CThe regular representation, the dimension sum rule, and column orthogonality

The goal is to prove the two counting statements that make the character table square — the dimension sum rule $\sum_\alpha d_\alpha^2=|G|$ and the equality of the number of irreps with the number of conjugacy classes — together with the column orthogonality relation (5) that expresses this squareness. The lever is the regular representation, in which $G$ permutes a basis labeled by its own elements — its character is starkly simple, and decomposing it forces every irrep to appear with multiplicity equal to its dimension.

Step 1. In the regular representation $g$ sends $|h\rangle$ to $|gh\rangle$ ; the diagonal entry at $h$ is nonzero only if $gh=h$ , i.e. $g=e$ . Hence $\chi^{\mathrm{reg}}(g)=|G|\,\delta_{g,e}$ . Why: a permutation matrix has a $1$ on the diagonal only at fixed points, and left multiplication by a non-identity element fixes nothing.

Step 2. Project: $m_\alpha=\langle\chi^{\mathrm{reg}},\chi^{(\alpha)}\rangle=\chi^{(\alpha)}(e)^{*}=d_\alpha$ . Why: the delta collapses the sum to the identity term, where the character equals the dimension, so each irrep sits inside the regular representation exactly $d_\alpha$ times.

Step 3. Evaluate the decomposed character at $e$ : $|G|=\chi^{\mathrm{reg}}(e)=\sum_\alpha m_\alpha\,\chi^{(\alpha)}(e)=\sum_\alpha d_\alpha^2$ . Why: two expressions for the same number must agree — this is the dimension sum rule.

Step 4. Count dimensions. The irreducible characters are orthonormal vectors in class-function space, whose dimension is the number of classes; they also span it, so the number of irreps equals the number of classes. Why: an orthonormal set cannot exceed the dimension of its space, and completeness upgrades the bound to equality, making the table square.

Step 5. Squareness lets the array be read down its columns. Forming $U_{\alpha i}=\sqrt{|C_i|/|G|}\,\chi^{(\alpha)}(C_i)$ , row orthonormality is $UU^\dagger=\mathbb{1}$ , so $U^\dagger U=\mathbb{1}$ too, which written out is equation (5). Why: for a square matrix left inverse equals right inverse, so orthonormal rows force orthogonal columns — column orthogonality is the transpose of the first. *For $S_3$ the column norms confirm equation (5): identity column $1^2+1^2+2^2=6$ ; flips column $1^2+(-1)^2+0^2=2=6/3$ ; rotations column $1^2+1^2+(-1)^2=3=6/2$ ; cross-column overlaps vanish.*

Limiting cases

Special cases turn the machinery into something checkable on a napkin, and each is a sanity test the general theory must pass.

Case	Result	Meaning
Abelian group	$d_\alpha=1$ for all $\alpha$ , $\#\text{irreps}=\vert G\vert$	every element is its own class; all irreps one-dimensional, homomorphisms $G\to U(1)$
Trivial representation	$\chi^{(\mathrm{triv})}(g)=1$ for all $g$	$\langle\chi,\chi^{(\mathrm{triv})}\rangle$ = dimension of the invariant subspace
Regular representation	$\chi^{\mathrm{reg}}(g)=\vert G\vert\,\delta_{g,e}$	contains each irrep $d_\alpha$ times, giving $\sum_\alpha d_\alpha^2=\vert G\vert$
Compact Lie group	$\dfrac{1}{\vert G\vert}\sum_g\ \longrightarrow\ \displaystyle\int_G\mathrm{d}\mu(g)$	group average becomes the normalized Haar integral; orthonormality survives
$SU(2)$ spin- $j$	$\chi_j(\theta)=\dfrac{\sin[(j+\tfrac12)\theta]}{\sin(\theta/2)}$	orthonormal under $\tfrac{1}{\pi}\int_0^{2\pi}\sin^2(\tfrac{\theta}{2})\,\mathrm{d}\theta$ ; Chebyshev polynomials, products give Clebsch–Gordan

Common trapDropping the conjugate, the class weight, or misreading χ(e)

Three slips ruin character arithmetic. First, the inner product carries a conjugate on the first factor and a $1/|G|$ average, $\langle\chi,\psi\rangle=\frac{1}{|G|}\sum_g\chi(g)^*\psi(g)$ ; omitting the conjugate breaks orthogonality for complex (non-real) characters like those of $C_3$ . Second, regrouping the element sum into a class sum weights each class by its size, $\sum_g(\cdots)=\sum_i|C_i|(\cdots)$ ; forgetting $|C_i|$ gives wrong multiplicities — the most common numerical error. Third, $\chi^{(\alpha)}(e)=d_\alpha$ is the dimension, not $1$ — reading the identity column as all ones wrecks the dimension sum rule.

Worked examples

1 · Building the S₃ character table from nothing (fully worked)

Construct the entire character table of $S_3$ (order $6$ , three conjugacy classes: the identity $e$ of size $1$ , the three transpositions of size $3$ , and the two 3-cycles of size $2$ ) using only the counting theorem, the dimension sum rule, and orthogonality — never writing down a single matrix.

Step 1. Fix the shape of the table. The counting theorem says the number of irreps equals the number of conjugacy classes, which is $3$ . So the table is $3\times3$ : three irreps, three columns. There are exactly three irreducible representations.

Step 2. Find the dimensions. The dimension sum rule needs $d_1^2+d_2^2+d_3^2=6$ in positive integers with at least one $d=1$ (the trivial rep); the only solution is $1^2+1^2+2^2=6$ , so the dimensions are $1,1,2$ — two one-dimensional irreps ( $A_1,A_2$ ) and one two-dimensional ( $E$ ).

Step 3. Write the identity column and the trivial row. Every $\chi^{(\alpha)}(e)=d_\alpha$ , so the first column is $(1,1,2)$ ; the trivial rep has $\chi=1$ everywhere: row $A_1=(1,1,1)$ .

Step 4. Get $A_2=(1,a,b)$ by orthogonality to $A_1$ : with class sizes $(1,3,2)$ , $\frac{1}{6}(1+3a+2b)=0$ . Since a one-dimensional rep maps to $U(1)$ and a transposition squares to the identity, $a=\pm1$ ; the non-trivial choice $a=-1$ gives $b=1$ . Row $A_2=(1,-1,1)$ — the sign representation, $-1$ on odd permutations (transpositions), $+1$ on even ones.

Step 5. Get $E=(2,x,y)$ from column orthogonality. The identity column $(1,1,2)$ orthogonal to the transpositions column forces $1-1+2x=0\Rightarrow x=0$ ; orthogonal to the rotations column forces $1+1+2y=0\Rightarrow y=-1$ . Row $E=(2,0,-1)$ .

Step 6. Verify: $\langle E,E\rangle=\frac{1}{6}(4+0+2)=1$ , so $E$ is irreducible. The completed table $A_1=(1,1,1)$ , $A_2=(1,-1,1)$ , $E=(2,0,-1)$ is derived entirely from $|G|=6$ , three classes, and orthogonality — no matrices.

2 · Decomposing a reducible representation

The natural 3-dimensional permutation representation of $S_3$ (each permutation as a $3\times3$ permutation matrix acting on three objects) has character equal to the number of objects a permutation leaves fixed. Decompose it into irreducibles.

Step 1. Read off the character. The identity fixes all three objects, a transposition fixes exactly one, and a 3-cycle fixes none, so on the classes $(e,\text{transpositions},\text{3-cycles})$ the character is $\chi=(3,1,0)$ .

Step 2. Test irreducibility. $\langle\chi,\chi\rangle=\frac{1}{6}(1\cdot3^2+3\cdot1^2+2\cdot0^2)=\frac{1}{6}(9+3+0)=2$ . Since $2\ne1$ , the representation is reducible, and $\sum_\alpha m_\alpha^2=2$ means two inequivalent irreps each appear once.

Step 3. Project onto each irrep with equation (7). $m_{A_1}=\frac{1}{6}(1\cdot3\cdot1+3\cdot1\cdot1+2\cdot0\cdot1)=\frac{1}{6}(3+3+0)=1$ ; $m_{A_2}=\frac{1}{6}(1\cdot3\cdot1+3\cdot1\cdot(-1)+2\cdot0\cdot1)=\frac{1}{6}(3-3+0)=0$ ; $m_E=\frac{1}{6}(1\cdot3\cdot2+3\cdot1\cdot0+2\cdot0\cdot(-1))=\frac{1}{6}(6+0+0)=1$ .

Step 4. Assemble and check. The decomposition is $\Gamma_{\mathrm{perm}}=A_1\oplus E$ , confirmed by adding characters: $(1,1,1)+(2,0,-1)=(3,1,0)=\chi$ . The trivial piece is the symmetric combination $|1\rangle+|2\rangle+|3\rangle$ ; the standard 2-dimensional piece is its orthogonal complement — the general fact that " $N$ objects carry one totally symmetric mode plus an $(N-1)$ -dimensional standard representation".

3 · A selection rule from a triple product

A transition amplitude $\langle f|\hat O|i\rangle$ vanishes unless the trivial representation appears in $\rho_f^*\otimes\rho_O\otimes\rho_i$ . With both states transforming as $E$ and the operator as $A_2$ , is it allowed?

Step 1. Tensor characters multiply pointwise: $\chi=\chi_E\,\chi_{A_2}\,\chi_E$ gives $(2\cdot1\cdot2,\ 0,\ (-1)\cdot1\cdot(-1))=(4,0,1)$ . Counting trivial copies, $m_{A_1}=\frac{1}{6}(1\cdot4+3\cdot0+2\cdot1)=1\ge1$ , so the transition is symmetry-allowed with one independent amplitude — a zero here would forbid it identically, the essence of a selection rule.

Physics of the portrait

Character theory was born not from physics but from a number-theoretic puzzle about factoring a determinant — yet the functions Frobenius invented to solve it, the traces of representation matrices, became within a generation the standard language of quantum symmetry.

Origins · 1 minFerdinand Georg Frobenius and how this was discovered

Ferdinand Georg Frobenius

1849–1917 · the invention of group characters

Frobenius created the character theory of finite groups almost single-handedly between 1896 and 1900, initially to answer a question Dedekind had raised about factoring the "group determinant". In his 1896 Über die Gruppencharaktere he defined characters as class functions satisfying orthogonality relations; only afterward — with the matrix representations he and Issai Schur developed — did it become clear that a character is simply the trace, $\chi(g)=\mathrm{tr}\,D(g)$ . He proved both orthogonality relations, the dimension sum rule $\sum_\alpha d_\alpha^2=|G|$ , and the equality of the number of irreducible characters with the number of conjugacy classes. His student Schur streamlined the edifice around 1905 by deriving orthogonality from what is now Schur's Lemma, and Burnside reached much of it independently in England. When Weyl extended it to compact Lie groups through the Haar integral and Wigner brought it to bear on atomic spectra in the late 1920s, the characters Frobenius had defined to factor a determinant became the instrument for classifying every degeneracy and selection rule in quantum mechanics.

History

1896

— Frobenius introduces group characters and proves their orthogonality relations

1897

— Frobenius recasts characters as traces of matrix representations

1905

— Schur streamlines the theory, deriving orthogonality from his lemma

1911

— Burnside's second edition brings representation theory to a wide audience

1927

— Peter and Weyl extend character orthogonality to compact Lie groups via Haar measure

1931

— Wigner applies characters to atomic spectra and selection rules

Going further

Texts. Stone & Goldbart, Mathematics for Physics, group representation theory · Arfken, Weber & Harris, Mathematical Methods for Physicists, group-theory chapter · Hassani, Mathematical Physics, group representations · Tinkham, Group Theory and Quantum Mechanics, ch. 3 · Hamermesh, Group Theory and Its Application to Physical Problems · Byron & Fuller, Mathematics of Classical and Quantum Physics. All numerical results use the $S_3=C_{3v}$ table $A_1=(1,1,1)$ , $A_2=(1,-1,1)$ , $E=(2,0,-1)$ with class sizes $(1,3,2)$ , which satisfies both orthogonality relations, the dimension sum rule, and $\Gamma_{\mathrm{perm}}=A_1\oplus E$ . The $SU(2)$ character is orthonormal under $\tfrac{1}{\pi}\int_0^{2\pi}\sin^2(\theta/2)\,\mathrm{d}\theta$ (the Chebyshev relation).

Prerequisites. Group Representations · Reducible and Irreducible Representations · Schur's Lemmas · Conjugacy Classes. Leads to. Character Tables and Decomposition · Tensor Products and Clebsch–Gordan · The Rotation Group SO(3) · SU(2) and the Spin Double Cover.

Ask the tutor: Bring me any group whose character table you want built or any representation you want reduced, and we'll turn the trace into a decomposition together.

Why take the trace and throw the rest of the matrix away — isn't that a loss of information?

It is a loss of redundant information. A representation matrix depends on an arbitrary basis, so most entries encode only the coordinate frame you picked, not the representation; the trace is the largest piece insensitive to that choice, since $\mathrm{tr}(SDS^{-1})=\mathrm{tr}\,D$ . Remarkably this residue is not merely convenient but complete: two representations with equal characters are equivalent. The character discards the coordinate noise and keeps the physics — which is why one number per class can classify an object built from $d_\alpha\times d_\alpha$ matrices.

Why does the number of irreducible representations equal the number of conjugacy classes?

Because the irreducible characters are an orthonormal basis for class-function space, which has one dimension per conjugacy class. Orthonormality alone caps their number at the number of classes; the harder fact that they also span the space pins the count to exact equality. The table is then square, and row-orthonormality forces column-orthogonality automatically — the two relations are two sides of one square unitary-like matrix.

Why the complex conjugate and the factor 1/|G| in the inner product?

The $1/|G|$ normalizes irreducible characters to unit norm, keeping the multiplicity formula free of stray factors. The complex conjugate is essential whenever a representation has non-real character — which happens as soon as the group has an element not conjugate to its inverse; $C_3$ already has two complex one-dimensional irreps with characters involving $e^{\pm2\pi i/3}$ . Without the conjugate these would not come out orthogonal and the projection machinery would fail. The conjugated, averaged form is the Hermitian inner product that makes class-function space a Hilbert space.

How does all of this survive when the group is infinite, like SO(3) or SU(2)?

For a compact Lie group the only change is that $\frac{1}{|G|}\sum_g$ becomes the integral $\int_G\mathrm{d}\mu(g)$ against the normalized Haar measure; every structural statement carries over — characters orthonormal, representations decomposed by inner products, irreducibility still the unit-norm test. For $SU(2)$ the classes are labeled by rotation angle $\theta$ , the spin- $j$ character is $\chi_j(\theta)=\sin[(j+\tfrac12)\theta]/\sin(\theta/2)$ , and orthonormality holds under $\frac{1}{\pi}\sin^2(\theta/2)\,\mathrm{d}\theta$ — Chebyshev-polynomial orthonormality in disguise. Projecting $\chi_{j_1}\chi_{j_2}$ onto $\chi_j$ reproduces $j_1\otimes j_2=\bigoplus_{j=|j_1-j_2|}^{j_1+j_2}j$ . Non-compact groups need more care, but the compact case is a clean generalization.

Can two different groups have the same character table?

Yes — the character table does not always determine the group. The classic example is the dihedral group $D_4$ and the quaternion group $Q_8$ : both have order $8$ , five conjugacy classes, four one-dimensional irreps and one two-dimensional irrep, and identical character tables entry for entry, yet they are not isomorphic ( $Q_8$ has a unique element of order two, $D_4$ has several). So the character table is a powerful invariant but not a complete one: it captures the whole representation theory while losing some finer group structure.

Does orthogonality also give the actual invariant subspaces, not just their multiplicities?

Multiplicities come from characters alone, but recovering the subspaces needs the projection operator $P^{(\alpha)}=\frac{d_\alpha}{|G|}\sum_g\chi^{(\alpha)}(g)^*\,D(g)$ — built from the character but applied to the full matrices $D(g)$ . It projects any vector onto the part transforming as irrep $\alpha$ , producing the symmetry-adapted basis (molecular orbitals, normal modes) that block-diagonalizes the Hamiltonian. So characters tell you how many of each irrep you have; the projection operators tell you which vectors span each piece — developed in the article on character tables and decomposition.