Characters and Reducible vs Irreducible Representations

Take the three 2p orbitals on a nitrogen atom in ammonia. Ask how they transform under the six symmetry operations of the C₃ᵥ point group. Each operation shuffles and mixes the orbitals. You could in principle write down a 3×3 matrix for every single one. But chemists never do this by hand for real molecules. Instead they read off six numbers, one per operation. From these they instantly know that the 2p set splits into an A₁ orbital (pz) and a doubly degenerate E pair (px, py). Those six numbers are characters. The fact that a big tangled representation collapses into a short list of small, unmixable pieces is the entire content of reducible-to-irreducible decomposition. This matters because every physically meaningful question in molecular symmetry is answered this way. Which vibrations are IR or Raman active? Which orbitals can mix? Which integrals vanish by symmetry? How many independent normal modes of each symmetry type does a molecule have? Each question is answered by decomposing some representation — of atomic motions, of basis orbitals, of a Hamiltonian matrix element — into irreducible representations (irreps). Doing this decomposition operation-by-operation on full matrices would be correct but absurdly wasteful. The matrices carry far more information than symmetry actually needs. The character, defined as the trace of the matrix, throws away everything except one number per operation. That number must survive similarity transformation. That one number turns out to be exactly enough. The big picture is this: representations built from real chemical bases are almost always reducible. This includes bases like bond vectors, atomic orbitals, and Cartesian displacement vectors. They are block-diagonalizable into smaller pieces after a suitable change of basis. The irreducible representations are the "atoms" of symmetry. They form the finite list of transformation types a point group permits. These are tabulated once and for all in character tables. Characters are invariant under the change of basis that block-diagonalizes a matrix. This is precisely why they are the right bookkeeping device. You can identify the irreducible pieces hiding inside a big reducible representation. You never need to find the block-diagonalizing transformation explicitly. This article builds the machinery from the matrix representations you already have in hand. It defines the character rigorously as a trace. It proves why the character is basis-independent. It shows concretely how a reducible representation block-diagonalizes into irreducible blocks. It derives the reduction formula that extracts how many times each irrep appears. This is the single most-used equation in applied group theory. Three worked examples carry this forward step by step. They move from a toy 2×2 case, through the C₃ᵥ 2p problem, to a full C₂ᵥ vibrational analysis of water. Every number is checked by hand, not asserted.

Key takeaways

The character χ(R) of an operation R in a representation is the trace of its representation matrix, χ(R) = Σᵢ Γ(R)ᵢᵢ.
Characters are invariant under similarity transformations, so they don't depend on which basis you happened to choose — a deep consequence of Tr(S⁻¹AS) = Tr(A).
A representation is reducible if a single similarity transformation block-diagonalizes every one of its matrices simultaneously; the blocks are the irreducible representations.
Characters of a reducible representation are simply the sums of the characters of its irreducible components, operation by operation.
The reduction formula aᵢ = (1/h)Σ_R g(R)χ(R)χᵢ(R) extracts the multiplicity of each irrep directly from characters, with no matrix diagonalization needed.
Key idea: the trace survives basis changes that scramble a matrix's entries, so characters — not full matrices — are the currency in which symmetry decomposition is honestly and efficiently done.

Where this lives today

Vibrational spectroscopy is the most direct and universal chemical application: every FTIR or Raman spectrum of a molecule with any symmetry at all is interpreted by first constructing $\Gamma_\text{total}$ from Cartesian displacements exactly as in Example 3, subtracting translations and rotations, and reducing the remainder to identify which normal modes are IR-active (those transforming as $x, y,$ or $z$ ) versus Raman-active (those transforming as quadratic functions like $x^2, xy$ ). Instrument manufacturers and analytical chemists use these symmetry-derived selection rules daily to assign peaks in mixtures, catalysts, and polymorphic drug crystals, where a shift or a "forbidden" band appearing at low intensity can reveal a subtle symmetry-lowering distortion — a workhorse diagnostic in solid-state pharmaceutical characterization and forensic materials analysis. In electronic structure theory, character-based reduction determines which atomic or molecular orbitals are allowed to mix by symmetry before any energy calculation is performed, dramatically simplifying Hückel, extended-Hückel, and modern DFT calculations on symmetric molecules and clusters by block-diagonalizing the secular equations along irreducible-representation lines — the same block-diagonalization pictured in the first figure, now applied to a Fock or Hamiltonian matrix rather than a geometric transformation. Crystal-field and ligand-field theory for transition-metal complexes relies entirely on reducing the reducible representation of d orbitals under octahedral or tetrahedral symmetry into $e_g + t_{2g}$ (or $e + t_2$ ) pieces, which is the origin of the famous crystal-field splitting diagrams used throughout inorganic and bioinorganic chemistry, including active-site models for metalloenzymes. In materials science, phonon mode symmetry analysis in solids — essential for interpreting Raman spectra of catalytic supports, battery cathode materials, and semiconductor thin films — is the same reduction machinery applied to the space-group-compatible point group at each crystallographic site. Even in molecular design and photochemistry, Woodward–Hoffmann-style orbital-symmetry arguments for pericyclic reactions ultimately rest on tracking which irreducible representation a frontier orbital belongs to as a reaction coordinate evolves, connecting this bookkeeping directly to predicting whether a cyclization proceeds thermally or photochemically.

Ask the tutor: How do I find the actual symmetry-adapted linear combinations of orbitals once I know which irreps are present, using the projection operator built from full character-table data?

The physical picture — thorough prose

Start from where Matrix Representations of Symmetry Operations leaves off: for a chosen basis set (atomic orbitals, bond vectors, Cartesian displacement coordinates), every symmetry operation R of a point group is assigned a square matrix Γ(R) that tells you how the basis functions transform into linear combinations of one another. The set of all these matrices, one per group operation, obeys the same multiplication table as the group itself — that is what makes it a representation. The trouble is that the basis you first think to write down is rarely the "natural" one. If you build a representation from the three N–H bond vectors of ammonia, or the nine Cartesian displacement coordinates of a water molecule, the resulting matrices are generally not block-diagonal: applying a C₃ rotation to the three N–H bonds genuinely mixes all three vectors together, giving a full 3×3 matrix with no zero blocks. Such a representation is called reducible, meaning it can be transformed — by a single change of basis applied consistently to every matrix in the set — into a new set of matrices that are block-diagonal, with smaller matrices repeating down the diagonal. Those smaller, un-further-reducible blocks are the irreducible representations (irreps) of the group. Every point group has a small, finite list of irreps; C₂ᵥ has exactly four (A₁, A₂, B₁, B₂), C₃ᵥ has three (A₁, A₂, E), and so on. This is the chemist's version of decomposing a chord into its constituent pure tones. The character is the tool that makes finding this decomposition tractable without ever explicitly constructing the block-diagonalizing transformation matrix. Define the character of operation R in representation Γ as the trace of its matrix: the sum of the diagonal elements. Two matrices related by a similarity transformation — which is exactly the relationship between a reducible representation in an arbitrary basis and its block-diagonal form in the "natural" symmetry-adapted basis — have identical traces. This single fact (proved below) means that the character of a reducible representation, computed in any convenient basis whatsoever, already secretly contains the full information about which irreps are present and how many times each appears. You never need to diagonalize anything. Concretely: if Γ = a₁Γ₁ ⊕ a₂Γ₂ ⊕ ⋯ (a direct sum, meaning Γ block-diagonalizes into aᵢ copies of each irrep Γᵢ), then the character of Γ at any operation R is just

\chi(R) = \sum_i a_i \, \chi_i(R)

— characters add linearly under direct sums, because the trace of a block-diagonal matrix is the sum of the traces of its blocks. The whole machinery of point-group analysis in chemistry — counting IR-active modes, building symmetry-adapted linear combinations of orbitals, deciding which orbitals can have nonzero overlap — rests on inverting this one linear relationship to solve for the unknown multiplicities aᵢ.

Figure

Reducible representation block-diagonalizing into irreducible pieces under a similarity transformation

The figure is not a metaphor: this is literally the C₃ operation acting on the three N–H bonds of ammonia in one basis versus the symmetry-adapted basis, where cos120° = −0.5 appears as the diagonal entry of the doubly-degenerate E block and the fully symmetric combination gives the 1×1 A₁ block. Note also, as a sanity check that will recur, that χ(E) — the character of the identity operation — always equals the dimension of the representation, since the identity matrix is diagonal with every entry 1.

Figure

The 2p orbital set on nitrogen in C₃ᵥ ammonia: reducible Γ splits into A₁ ⊕ E

The core ideas — governing equations

The character of operation R in representation Γ, built on an n-dimensional basis, is the trace of its representation matrix:

\chi_\Gamma(R) \equiv \operatorname{Tr}\big[\Gamma(R)\big] = \sum_{i=1}^{n} \Gamma(R)_{ii}

where: $\Gamma(R)_{ii}$ are the diagonal entries of the $n\times n$ matrix representing operation $R$ in this basis; the sum runs over all $n$ basis functions. Reading: the character is a single real (or complex) number per symmetry operation — an enormous compression of the full matrix, and the object that character tables actually list. Invariance under similarity transformation is the load-bearing property:

\operatorname{Tr}\big[S^{-1}\Gamma(R)S\big] = \operatorname{Tr}\big[\Gamma(R)\big] \quad \text{for any invertible } S

where: $S$ is any change-of-basis matrix, $S^{-1}\Gamma(R)S$ is the same operation's matrix re-expressed in a new basis. Reading: no matter what basis you write your matrices in — natural, symmetry-adapted, or arbitrary — the character of each operation comes out the same, so characters are a property of the representation itself, not of the basis chosen to display it. When a reducible representation Γ decomposes as a direct sum of irreps,

\Gamma = a_1\Gamma_1 \oplus a_2\Gamma_2 \oplus \cdots \oplus a_k\Gamma_k

where: $a_i$ is the (non-negative integer) number of times irrep $\Gamma_i$ appears in the block-diagonal form of $\Gamma$ ; $k$ is the number of distinct irreps of the point group. Reading: this is the statement that Γ, after the right change of basis, literally becomes a block-diagonal matrix with $a_i$ copies of the $\Gamma_i$ block sitting along the diagonal, for each $i$ . Taking traces of both sides at each operation R gives the character decomposition rule:

\chi_\Gamma(R) = \sum_{i=1}^{k} a_i\, \chi_i(R)

where: $\chi_\Gamma(R)$ is the (easily computed) character of the reducible representation at operation $R$ ; $\chi_i(R)$ are the characters of irrep $\Gamma_i$ , read from the character table. Reading: the observed character at every operation is a fixed linear combination of tabulated irreducible characters, weighted by the unknown multiplicities — this is one linear equation per class of operations, and there are exactly as many classes as irreps, so the system is always exactly solvable. The reduction formula inverts this system using the orthogonality of irreducible characters (a consequence of the Great Orthogonality Theorem):

a_i = \frac{1}{h}\sum_{R} g(R)\,\chi(R)\,\chi_i(R)

where: $h$ is the order of the group (total number of symmetry operations); the sum runs over classes of operations; $g(R)$ is the number of operations in the class of $R$ ; $\chi(R)$ is the character of the reducible representation; $\chi_i(R)$ is the character of irrep $i$ (real characters assumed, as in all point groups relevant to real molecular vibrations and orbitals). Reading: this is literally a dot product between the "character vector" of your reducible representation and the character vector of irrep $i$ , normalized by the group order — it projects out exactly the multiplicity of that one irrep, discarding all others by orthogonality.

Derivation AWhy the trace is basis-independent: the cyclic property of Tr

Goal. Show rigorously that $\operatorname{Tr}[S^{-1}AS] = \operatorname{Tr}[A]$ for any square matrix $A$ and invertible $S$ , which is the entire justification for treating characters as intrinsic to a representation rather than artifacts of a chosen basis.

Step 1 — write the trace as an explicit sum. For any two matrices $B$ and $C$ of compatible size, the trace of their product is

\operatorname{Tr}[BC] = \sum_i (BC)_{ii} = \sum_i \sum_j B_{ij}C_{ji}

This double sum is symmetric in structure: it sums $B_{ij}C_{ji}$ over all pairs $(i,j)$ .

Step 2 — swap the order of multiplication. Compute $\operatorname{Tr}[CB]$ the same way:

\operatorname{Tr}[CB] = \sum_i (CB)_{ii} = \sum_i \sum_j C_{ij}B_{ji} = \sum_j \sum_i B_{ji}C_{ij}

Relabeling the dummy indices $i \leftrightarrow j$ in the last expression reproduces exactly the double sum from Step 1. Hence

\operatorname{Tr}[BC] = \operatorname{Tr}[CB]

This is the cyclic property of the trace: you may cyclically permute a product of matrices inside a trace without changing its value (note this is not the same as saying matrix multiplication commutes — $BC \ne CB$ in general, only their traces agree).

Step 3 — apply cyclic property to the similarity-transformed matrix. Set $B = S^{-1}$ and $C = AS$ :

\operatorname{Tr}[S^{-1}(AS)] = \operatorname{Tr}[(AS)S^{-1}]

The right-hand side simplifies using $SS^{-1} = \mathbb{1}$ :

\operatorname{Tr}[(AS)S^{-1}] = \operatorname{Tr}[A(SS^{-1})] = \operatorname{Tr}[A\cdot\mathbb{1}] = \operatorname{Tr}[A]

Step 4 — conclude and interpret. Therefore

\operatorname{Tr}[S^{-1}AS] = \operatorname{Tr}[A]

for any invertible $S$ , with no restriction on $A$ being symmetric, orthogonal, or anything else. Since every representation matrix $\Gamma(R)$ in a "wrong" basis is related to the matrix $\Gamma'(R)$ in the symmetry-adapted (block-diagonal) basis by exactly such a similarity transformation, $\Gamma'(R) = S^{-1}\Gamma(R)S$ , this proves $\chi(R) = \operatorname{Tr}[\Gamma(R)] = \operatorname{Tr}[\Gamma'(R)]$ . The character you compute from a clumsy, fully-mixed basis is identical, operation by operation, to the character you would get after finding the perfect symmetry-adapted basis and reading off the block-diagonal form directly. This is why characters — never the raw matrices — are what gets tabulated and used for decomposition.

Go deeper · Derivation B · 1 minThe reduction formula: extracting multiplicities aᵢ from characters alone

Derivation BThe reduction formula: extracting multiplicities aᵢ from characters alone

Goal. Derive $a_i = \frac{1}{h}\sum_R g(R)\chi(R)\chi_i(R)$ starting only from the character decomposition rule and the orthogonality of irreducible characters, so that multiplicities can be read off without ever diagonalizing a matrix.

Step 1 — state the character orthogonality theorem. The characters of the irreducible representations of a finite group satisfy a "great orthogonality"-derived relation (proved fully under The Great Orthogonality Theorem): treating $\chi_i(R)$ as a vector indexed by operations $R$ , weighted by class size,

\sum_{R} g(R)\,\chi_i(R)\,\chi_j(R) = h\,\delta_{ij}

where $\delta_{ij}=1$ if $i=j$ and $0$ otherwise. In words: the character vectors of different irreps are orthogonal, and each irrep's character vector has squared "length" exactly $h$ (the group order). This is taken here as the standing algebraic fact that licenses the projection trick below; it follows from the unitarity and irreducibility of the matrix representations.

Step 2 — start from the character decomposition of the reducible representation. From the direct-sum structure $\Gamma = \bigoplus_i a_i \Gamma_i$ , take characters at every operation $R$ :

\chi(R) = \sum_{j=1}^{k} a_j\,\chi_j(R)

Step 3 — multiply both sides by $g(R)\chi_i(R)$ and sum over all operations (equivalently, over classes).

\sum_R g(R)\,\chi_i(R)\,\chi(R) = \sum_R g(R)\,\chi_i(R)\sum_{j=1}^{k} a_j\,\chi_j(R)

Swap the order of summation on the right-hand side (sum over $R$ first, then over $j$ ):

\sum_R g(R)\,\chi_i(R)\,\chi(R) = \sum_{j=1}^{k} a_j \left[\sum_R g(R)\,\chi_i(R)\,\chi_j(R)\right]

Step 4 — apply orthogonality to collapse the bracket. By Step 1, the bracketed sum equals $h\delta_{ij}$ , which is zero unless $j=i$ , in which case it equals $h$ . The entire sum over $j$ therefore collapses to the single surviving term $j=i$ :

\sum_R g(R)\,\chi_i(R)\,\chi(R) = a_i\, h

Step 5 — solve for $a_i$ . Dividing through by $h$ :

a_i = \frac{1}{h}\sum_{R} g(R)\,\chi_i(R)\,\chi(R)

This is exactly the reduction formula. Interpret it as a projection: the orthogonality relation makes the irreducible characters into an orthogonal basis for "character space," and computing $a_i$ is nothing more than projecting the observed reducible character vector $\chi(R)$ onto the $i$ -th basis vector $\chi_i(R)$ , then normalizing by $h$ . Every $a_i$ obtained this way is guaranteed, by the underlying theorem, to come out as a non-negative integer if $\chi(R)$ genuinely arose from a representation — a built-in consistency check on any calculation.

Worked examples

Example 1 — Foundational: a 2×2 reducible representation of C₂ᵥ, decomposed entirely by hand. Consider a toy but completely explicit basis of two functions, $f_1$ and $f_2$ , transforming under C₂ᵥ ( $h = 4$ ; operations $E, C_2, \sigma_v, \sigma_v'$ , each its own class of size 1) according to the matrices

\Gamma(E) = \begin{pmatrix}1&0\\0&1\end{pmatrix},\quad \Gamma(C_2) = \begin{pmatrix}0&1\\1&0\end{pmatrix},\quad \Gamma(\sigma_v) = \begin{pmatrix}0&1\\1&0\end{pmatrix},\quad \Gamma(\sigma_v') = \begin{pmatrix}1&0\\0&1\end{pmatrix}

(this is, concretely, the representation generated by two equivalent N–H-type bonds swapped by $C_2$ and by one mirror, left alone by the other mirror and by $E$ — an asymmetric, off-diagonal-heavy case deliberately chosen so no shortcut hides an error). Reading off traces directly:

\chi(E) = 1+1 = 2,\qquad \chi(C_2) = 0+0 = 0,\qquad \chi(\sigma_v) = 0+0 = 0,\qquad \chi(\sigma_v') = 1+1 = 2

The C₂ᵥ character table gives:

	E	C₂	σᵥ	σᵥ'
A₁	1	1	1	1
A₂	1	1	−1	−1
B₁	1	−1	1	−1
B₂	1	−1	−1	1

Apply the reduction formula with $h=4$ , all classes of size $g(R)=1$ :

a_{A_1} = \frac{1}{4}\big[(1)(2) + (1)(0) + (1)(0) + (1)(2)\big] = \frac{1}{4}(4) = 1

a_{A_2} = \frac{1}{4}\big[(1)(2) + (1)(0) + (-1)(0) + (-1)(2)\big] = \frac{1}{4}(0) = 0

a_{B_1} = \frac{1}{4}\big[(1)(2) + (-1)(0) + (1)(0) + (-1)(2)\big] = \frac{1}{4}(0) = 0

a_{B_2} = \frac{1}{4}\big[(1)(2) + (-1)(0) + (-1)(0) + (1)(2)\big] = \frac{1}{4}(4) = 1

So $\Gamma = A_1 \oplus B_2$ . Check: the reconstructed character is $\chi_{A_1}(R) + \chi_{B_2}(R)$ , giving at $E$ : $1+1=2$ ✓, at $C_2$ : $1+(-1)=0$ ✓, at $\sigma_v$ : $1+(-1)=0$ ✓, at $\sigma_v'$ : $1+1=2$ ✓ — every entry matches the directly computed character exactly, and both multiplicities came out as non-negative integers, exactly as the theorem guarantees. Limiting case: if instead $\Gamma(C_2)$ and $\Gamma(\sigma_v)$ had been the identity matrix too (fully symmetric basis under every operation), you would find $\chi(R)=2$ for all four operations, giving $a_{A_1}=2$ and all others zero — the representation would be $2A_1$ , i.e., two independent totally symmetric functions, which is the sensible degenerate limit of a basis with no antisymmetric mixing at all.

Example 2 — Applied: the 2pₓ, 2p_y, 2p_z set on nitrogen in NH₃ (C₃ᵥ, h = 6). C₃ᵥ has classes $E$ (size 1), $2C_3$ (size 2, meaning $C_3$ and $C_3^2$ together), and $3\sigma_v$ (size 3). Place $p_z$ along the C₃ axis. Under $E$ , all three orbitals are unchanged: contribution to trace is $1+1+1=3$ . Under a $C_3$ rotation by 120°, $p_z$ is invariant (contributes $+1$ ), while the $(p_x,p_y)$ pair rotates into each other with rotation-matrix trace $2\cos120° = -1$ ; total $\chi(C_3) = 1 + (-1) = 0$ . Under a vertical mirror plane containing the $z$ -axis and one N–H bond, $p_z$ is unchanged ( $+1$ ), the in-plane $p$ -component along the mirror is unchanged ( $+1$ ), and the out-of-plane component flips sign ( $-1$ ); total $\chi(\sigma_v) = 1 + (1 - 1) = 1$ . So the reducible representation has character

\chi(E) = 3, \qquad \chi(C_3) = 0, \qquad \chi(\sigma_v) = 1

The C₃ᵥ character table:

	E	2C₃	3σᵥ
A₁	1	1	1
A₂	1	1	−1
E	2	−1	0

Apply the reduction formula with $h = 6$ :

a_{A_1} = \frac{1}{6}\big[(1)(1)(3) + (2)(1)(0) + (3)(1)(1)\big] = \frac{1}{6}(3+0+3) = \frac{6}{6} = 1

a_{A_2} = \frac{1}{6}\big[(1)(1)(3) + (2)(1)(0) + (3)(-1)(1)\big] = \frac{1}{6}(3+0-3) = 0

a_{E} = \frac{1}{6}\big[(1)(2)(3) + (2)(-1)(0) + (3)(0)(1)\big] = \frac{1}{6}(6+0+0) = 1

So $\Gamma(2p) = A_1 \oplus E$ , exactly matching the qualitative chemical picture: $p_z$ is the unique $A_1$ orbital pointing along the molecular axis, and $(p_x,p_y)$ form the degenerate $E$ pair — precisely what the second figure displayed by direct inspection. Check: reconstructed characters $1+2=3$ at $E$ ✓, $1+(-1)=0$ at $C_3$ ✓, $1+0=1$ at $\sigma_v$ ✓.

Example 3 — Exam-level: full vibrational analysis of H₂O (C₂ᵥ), separating translation, rotation, and genuine vibration. Water has 3 atoms, so 9 Cartesian degrees of freedom (3 per atom). Build $\Gamma_\text{total}$ , the representation of all 9 Cartesian displacement vectors, using the standard shortcut: only unmoved atoms contribute to the character (an atom that moves to a different position under $R$ contributes zero, since its displacement vectors map onto a different atom's coordinates, giving zero diagonal contribution), and each unmoved atom contributes according to the trace of the 3×3 Cartesian rotation/reflection matrix for that operation. Contribution per unmoved atom: $\chi_{xyz}(E) = 3$ ; $\chi_{xyz}(C_2) = -1$ (rotation by 180° about $z$ : matrix $\text{diag}(-1,-1,1)$ , trace $-1$ ); $\chi_{xyz}(\sigma_v) = 1$ (reflection: two coordinates unchanged, one flips sign, e.g. $\text{diag}(1,-1,1)$ , trace $1$ ); same for $\sigma_v'$ . Unmoved-atom counts for H₂O in C₂ᵥ: under $E$ , all 3 atoms are unmoved; under $C_2$ (through O, bisecting H–O–H), only O is unmoved (1 atom); under $\sigma_v$ (the molecular plane, containing both H's and O), all 3 atoms lie in the plane, so all 3 are unmoved; under $\sigma_v'$ (perpendicular to the molecular plane, containing only O), only O is unmoved (1 atom).

\chi_\text{total}(E) = 3\times 3 = 9

\chi_\text{total}(C_2) = 1\times(-1) = -1

\chi_\text{total}(\sigma_v) = 3\times 1 = 3 \qquad(\text{molecular plane})

\chi_\text{total}(\sigma_v') = 1\times 1 = 1 \qquad(\text{perpendicular plane, only O unmoved})

Reduce with $h=4$ , all class sizes $1$ :

a_{A_1} = \frac{1}{4}\big[(1)(9)+(1)(-1)+(1)(3)+(1)(1)\big] = \frac{1}{4}(12) = 3

a_{A_2} = \frac{1}{4}\big[(1)(9)+(1)(-1)+(-1)(3)+(-1)(1)\big] = \frac{1}{4}(4) = 1

a_{B_1} = \frac{1}{4}\big[(1)(9)+(-1)(-1)+(1)(3)+(-1)(1)\big] = \frac{1}{4}(12) = 3

a_{B_2} = \frac{1}{4}\big[(1)(9)+(-1)(-1)+(-1)(3)+(1)(1)\big] = \frac{1}{4}(8) = 2

(using the convention that $\sigma_v$ = molecular ( $xz$ ) plane transforms as $B_1$ , $\sigma_v'$ = perpendicular ( $yz$ ) plane as $B_2$ , standard for the $x$ -in-plane convention). So

\Gamma_\text{total} = 3A_1 \oplus A_2 \oplus 3B_1 \oplus 2B_2

Check on dimension: $3+1+3+2 = 9$ ✓, matching the 9 Cartesian degrees of freedom exactly. Now subtract translation and rotation, whose symmetries are read directly from the character table's last two columns: translations transform as $(A_1 + B_1 + B_2)$ (from $z, x, y$ ) and rotations as $(A_2 + B_1 + B_2)$ (from $R_z, R_x, R_y$ ). Removing these:

\Gamma_\text{vib} = (3A_1 + A_2 + 3B_1 + 2B_2) - (A_1+B_1+B_2) - (A_2+B_1+B_2) = 2A_1 \oplus B_1

Dimension check: $2+1 = 3$ vibrational modes, exactly matching $3N-6 = 9-6=3$ for a nonlinear triatomic ✓. This reproduces the textbook result for water: the symmetric stretch and bend are $A_1$ , the antisymmetric stretch is $B_1$ — all three IR-active (since $A_1, B_1$ both carry translational symmetry components $z$ and $x$ respectively) and all three Raman-active, consistent with water's well-known IR/Raman spectrum. Limiting case: if water were replaced by a hypothetical linear XY₂ (which it is not, but as a consistency check on method), the unmoved-atom analysis would need the $D_{\infty h}$ table instead and give the familiar $3N-5$ count with one bending mode doubly degenerate — the method generalizes, but the point-group input must change with geometry.

Limiting cases, scope & common traps

Situation	What happens to χ	Correct handling
Operation = identity, E	χ(E) = dimension of the basis (trace of identity matrix)	Always sanity-check: Σᵢaᵢ·(dim of Γᵢ) must equal χ(E)
Atom lies exactly on a rotation axis or mirror plane	Atom is "unmoved," contributes full 3×3 Cartesian trace	Count only truly unmoved atoms — an atom merely close to the axis still moves if it's off-axis
Degenerate irreps (E, T) present	χ can be non-integer-looking mid-calculation but combines to integers	Never round intermediate character values; keep exact surds (e.g. 2cos120° = −1)
Reducible representation happens to equal a single irrep already	All aᵢ = 0 except one, which = 1	The reduction formula still works — it just confirms Γ was already irreducible
Complex characters (some higher-symmetry, non-abelian groups with genuinely complex irreps)	Reduction formula needs χᵢ(R)* (complex conjugate)	For all real point groups of ordinary molecules with real character tables this subtlety doesn't arise, but state it for completeness
Using the wrong group order h	All aᵢ scaled incorrectly, often non-integer	h = total number of operations, not number of classes; double-count operations within a class properly via g(R)

Common trapχ(R) is not simply "the number of basis functions left unchanged"

It is tempting to think χ(R) just counts how many basis functions map to themselves. This is only true when the representation matrix is diagonal in that basis for that operation (e.g., the "unmoved atoms" trick works because each unmoved atom's local 3×3 block is diagonal-friendly and its trace is computed separately). In general, a basis function can be partially mixed into others by R — contributing a fractional or negative amount to the trace — and only the sum of diagonal entries, not a naive head-count, is the correct character. The $(p_x,p_y)$ rotation in Example 2, contributing $-1$ to $\chi(C_3)$ with neither orbital literally "unchanged," is the clearest illustration.

The chemist behind it

The systematic use of character theory in molecular science traces back to the mathematics of finite-group representation theory developed by Ferdinand Georg Frobenius in the 1890s, and its application to physics and chemistry was consolidated by Hermann Weyl and Eugene Wigner in the 1920s–30s, who showed that quantum-mechanical degeneracies and selection rules are dictated by the irreducible representations of a system's symmetry group. The translation of this abstract algebra into a practical laboratory tool for spectroscopists — character tables you can look up and apply to real molecules in minutes — is largely credited to the systematic tabulations popularized by chemists such as Friedrich Hund and later codified for chemistry curricula by figures like Frank Albert Cotton, whose textbook Chemical Applications of Group Theory made character-table reduction a standard part of the working chemist's toolkit rather than a specialist's mathematical curiosity.

Origins · 1 minEugene Paul Wigner and how this was discovered

Eugene Paul Wigner

1902–1995 · Central European émigré physicist, Princeton

Wigner's 1931 book Gruppentheorie und ihre Anwendung auf die Quantenmechanik der Atomspektren ("Group Theory and Its Application to the Quantum Mechanics of Atomic Spectra") was the decisive bridge between Frobenius's abstract character theory and its use as a working tool for predicting spectroscopic selection rules and degeneracies. He showed that whenever a Hamiltonian commutes with the symmetry operations of a group — as any Hamiltonian must commute with its molecule's or crystal's own point-group operations — its eigenfunctions must transform as irreducible representations of that group, and that matrix elements connecting states of different symmetry vanish identically. This theorem is the deep reason chemists care about reducing representations at all: it converts symmetry classification into hard selection rules for spectroscopy and bonding. Wigner shared the 1963 Nobel Prize in Physics for these and related contributions to the theory of the atomic nucleus and elementary particles via symmetry principles.

History

- 1896 — Frobenius introduces group characters and proves their orthogonality for finite groups.

- 1929 — Bethe applies group-theoretic splitting of atomic term symbols to crystal-field problems, an early chemistry-adjacent use of characters.

- 1931 — Wigner's book establishes character-based selection rules for quantum spectra as standard practice.

- 1963 — Wigner receives the Nobel Prize in Physics, partly for symmetry-principle contributions rooted in representation theory.

- 1971 — Cotton's Chemical Applications of Group Theory (3rd ed. era of popularity) cements character-table reduction as routine coursework for chemists.

Why is the character defined as a trace rather than, say, the determinant of the matrix?

The trace is the unique matrix invariant (up to scalar multiples and combinations) that is linear and additive under direct sums: the trace of a block-diagonal matrix is exactly the sum of the traces of its blocks, which is precisely the property needed so that χ(R) = Σᵢaᵢχᵢ(R) holds cleanly. The determinant of a block-diagonal matrix is instead the product of the block determinants, which does not decompose additively into simple multiplicities and would make the reduction formula far messier or impossible to invert linearly. Traces also survive similarity transformations exactly, as proven via the cyclic property, which determinants also do (det(S⁻¹AS)=det A) — but only the trace's additivity under direct sums makes it useful for counting irreducible components.

Do I ever need to find the actual block-diagonalizing matrix S in practice?

Almost never, for the purposes of identifying which irreps are present and how many times — that is exactly the point of using characters, since the reduction formula bypasses S entirely. You do need to construct the explicit symmetry-adapted linear combinations (which requires more than characters, using the full projection operator machinery built from complete representation matrices or character-table-derived projection formulas) when you want the actual symmetry-adapted orbitals or normal coordinates themselves, not just their symmetry labels and counts. For counting IR/Raman activity, predicting orbital mixing, or classifying vibrational modes, characters alone suffice completely.

What happens if my reduction formula gives a non-integer or negative aᵢ?

This is always a sign of an arithmetic or setup error, never a legitimate physical outcome — the underlying theorem guarantees non-negative integers whenever χ(R) genuinely came from summing traces of an actual representation. The most common culprits are: using the wrong class sizes g(R) (forgetting that 2C₃ means two operations, not one), misreading which mirror plane is σᵥ versus σᵥ', or an error in the "unmoved atoms" trace count (for example forgetting that the trace for a mirror reflection is +1, not −1, and for an improper rotation Sₙ it differs again). Recomputing χ(R) directly from a concrete small basis, as in the worked examples, is the fastest way to locate the mistake.

Is the character the same thing as the "trace of the Hamiltonian matrix" in perturbation theory?

No — they are related concepts but not the same object. The character here is the trace of a symmetry-operation representation matrix Γ(R), describing how a basis transforms geometrically; it has nothing to do, by itself, with energies. However, the two ideas connect deeply: because a molecular Hamiltonian commutes with every symmetry operation of its point group, its matrix representation in a symmetry-adapted basis is automatically block-diagonal, with blocks labeled by irreducible representations — this is the practical payoff of character-based reduction, since it tells you in advance which off-diagonal Hamiltonian matrix elements are forced to vanish by symmetry, before you ever compute an energy.

Why do χ(C₃) and χ(C₃²) both count as "2C₃" with the same character value?

Operations belonging to the same conjugacy class (in this case, rotation by +120° and by −120° about the same axis) are related to each other by conjugation with some other group operation, and characters are invariant under conjugation for exactly the same reason they are invariant under general similarity transformations — conjugation is a special case of a similarity transformation. This means all operations within one class automatically share the same character in every representation, which is precisely why character tables list one column per class rather than one column per individual operation, saving substantial space and computation.

Can a representation be "partially reducible" — reducible for some operations but not others?

No. Reducibility is a property of the entire set of matrices taken together under one single similarity transformation S that must simultaneously block-diagonalize every Γ(R) for every operation R in the group. It is not evaluated operation-by-operation; a representation is reducible only if one fixed S works for the whole group at once, which is guaranteed to exist whenever the character decomposition formula returns non-trivial (more than one nonzero, or one aᵢ greater than the irrep's own dimension) multiplicities.

How does this connect to counting IR and Raman active vibrational modes?

Once Γ_vib is reduced into irreps (as in the water example, giving 2A₁ ⊕ B₁), each mode's IR activity is determined by checking whether that irrep also appears in the character table's list of linear functions (x, y, z) — if so, the mode carries a nonzero transition dipole moment component and is IR-active. Raman activity is checked the same way against the quadratic functions (x², y², z², xy, xz, yz) listed in the table. For water, all three modes (2A₁ + B₁) match either x, y, z or a quadratic function, so all three are both IR- and Raman-active, matching experimental water vapor spectra.