Matrix Representations of Symmetry Operations

Rotate an ammonia molecule by 120° about its threefold axis. Its three N–H bonds simply relabel themselves. The molecule looks identical, but each hydrogen has moved to where a neighbor used to be. To a chemist this is an obvious geometric fact. To a theorist it is an invitation to write down a matrix. Every symmetry operation of a point group — every rotation, reflection, inversion, or improper rotation — can be represented as a linear transformation. This transformation acts on a chosen set of basis functions. Linear transformations on a finite-dimensional basis are matrices. This single translation, from geometric operation to numerical array, matters greatly. It is what turns point group assignment into a genuine computational and predictive tool. It becomes more than just a classification exercise. The stakes are high. Once a symmetry operation $\hat{R}$ is written as a matrix $D(R)$ , it acts on a basis. That basis might be atomic coordinates, $p$ and $d$ orbitals, or vibrational displacement vectors. The entire apparatus of representation theory then becomes available. This includes characters, reducible and irreducible representations, projection operators, and selection rules for spectroscopy. Matrix representations are the concrete objects from which the abstract group multiplication table is recovered. This recovery happens through ordinary matrix multiplication. These matrices have traces, called characters. The characters will later classify vibrational modes, molecular orbitals, and spectroscopic transitions. This classification happens without ever needing the full matrices again. The big picture is that a point group is an abstract algebraic structure. It is a set of operations closed under composition, with an identity and inverses. But chemistry needs it to act on something: nuclear positions, basis orbitals, or normal-mode displacement vectors. Choosing a basis and writing each operation as a matrix reproduces its action on that basis. This process is called constructing a representation of the group. Different bases give different (but related) representations of the same abstract group. The same operation $C_3$ might take several forms. It could be a $3\times 3$ orthogonal matrix acting on Cartesian coordinates. It could be a $2\times2$ matrix acting on a pair of degenerate $p$ orbitals. It could also be a $1\times1$ number (a character) acting on a single symmetric function. This article builds that translation from the ground up. It shows how to write rotation, reflection, inversion, and improper-rotation matrices from first principles. It shows how to verify that a set of such matrices actually obeys the group's multiplication table. It also shows how the trace of each matrix begins to emerge naturally from the construction. This trace will later be used to classify representations. Three worked examples move from a bare rotation matrix. They progress through the full matrix representation of ammonia's symmetry operations on Cartesian basis vectors. They end with a genuinely tricky reducible representation for a $D_{4h}$ molecule.

Key takeaways

A symmetry operation acting on a chosen basis (coordinates, orbitals, displacement vectors) is represented by a square matrix $D(R)$ ; the basis choice fixes the matrix's dimension.
Proper rotations $C_n$ are represented by standard rotation matrices; improper operations ( $\sigma$ , $i$ , $S_n$ ) include a reflection through the mirror-plane normal.
Matrices must reproduce the group multiplication table: $D(R_1)D(R_2) = D(R_1 R_2)$ , exactly matching the geometric composition of operations.
The matrix representation depends entirely on the basis; the same abstract operation can be $1\times1$ , $2\times2$ , or $3N\times 3N$ depending on what it acts on.
The trace (sum of diagonal elements) of each $D(R)$ is the character $\chi(R)$ — basis-independent within a similarity class, and the bridge to Characters and the Trace.
Any representation built on physical coordinates is generally reducible into smaller blocks, foreshadowing Reducible and Irreducible Representations.
Key idea: choosing a basis turns an abstract symmetry operation into a concrete matrix, and consistency of these matrices under multiplication is what makes the representation faithful to the group.

Where this lives today

Matrix representations of symmetry operations are not a historical curiosity confined to textbooks — they are the computational backbone of essentially every symmetry-adapted electronic structure and vibrational analysis method used in modern chemistry. In vibrational spectroscopy, the $3N$ -dimensional Cartesian representation built exactly as in Example 3 is the starting point for determining which normal modes of a molecule are infrared-active versus Raman-active, a calculation performed routinely for every new molecule characterized by IR or Raman spectroscopy in synthetic, materials, and pharmaceutical laboratories. Quantum chemistry software packages (Gaussian, ORCA, Psi4, and others) internally construct exactly these matrix representations to exploit molecular symmetry, block-diagonalizing the Fock or Kohn–Sham matrix into symmetry blocks that dramatically reduce computational cost for symmetric molecules such as benzene, ferrocene, or buckminsterfullerene. In crystallography and solid-state physics, the 230 space groups are built from point-group matrix representations combined with translations, and the matrix machinery described here — reflections, rotations, improper rotations acting on fractional coordinates — is exactly what crystallographic software (SHELX, Olex2, or the International Tables for Crystallography) uses to generate symmetry-equivalent atomic positions from an asymmetric unit. In ligand-field and coordination chemistry, the splitting of $d$ -orbital energies in octahedral, tetrahedral, and square-planar complexes — central to interpreting the colors and magnetism of transition-metal compounds — is derived by writing the point-group operations as matrices acting on the five $d$ -orbitals and decomposing the resulting representation, precisely following Bethe's original method. In molecular orbital theory more broadly, symmetry-adapted linear combinations (SALCs) used to construct MO diagrams for polyatomic molecules like $\mathrm{NH_3}$ , $\mathrm{BF_3}$ , or $\mathrm{XeF_4}$ are generated using projection operators built directly from these matrix representatives, a technique that becomes essential once Reducible and Irreducible Representations and Characters and the Trace are brought into the workflow.

:::quiz[For the general Sₙ(θ) = σₕ·C(θ) matrix derived by combining a z-rotation with reflection through the xy-plane, what is det[S(θ)] for any angle θ?]

+1
−1 ✓
0
cos θ

Ask the tutor: How do I turn the reducible Cartesian matrix representation built here into the actual irreducible representations that tell me which vibrations are IR- or Raman-active?

The physical picture

A point group operation is, physically, a rigid motion (or improper motion) that leaves at least one point of the molecule fixed and maps the molecule onto an indistinguishable copy of itself. To turn this geometric statement into algebra, pick a basis — a set of $n$ linearly independent functions or vectors $\{f_1, f_2, \dots, f_n\}$ — and ask how the operation transforms each basis function into a linear combination of all of them:

\hat{R}\, f_j = \sum_{i=1}^{n} D(R)_{ij}\, f_i

The coefficients $D(R)_{ij}$ collected into an $n\times n$ array are the matrix representative of $\hat{R}$ in that basis. Crucially, the same operation $\hat{R}$ (say, a $C_2$ rotation) gives different-sized matrices depending on what basis you choose: a $1\times 1$ matrix if the basis is a single totally symmetric function, a $2\times2$ matrix if the basis is a degenerate pair of $\pi$ orbitals, or a $3N\times 3N$ matrix if the basis is the full set of $x,y,z$ displacement coordinates of all $N$ atoms. The most common and pedagogically central basis is the set of Cartesian unit vectors $(x, y, z)$ centered at a fixed point, or their analog attached to each atom of a molecule (a $3N$ -dimensional Cartesian displacement basis). For a single set of $(x,y,z)$ axes, proper rotations and improper operations have completely standard matrix forms, derivable from elementary trigonometry and reflection geometry, and these serve as the building blocks for every larger representation encountered later. A rotation by angle $\theta$ about the $z$ -axis sends $(x,y,z)$ to new coordinates found by resolving components along the rotated axes. Using the standard convention (rotation is written to act on coordinates; the matrix multiplies a column vector on the left), the rotation matrix is

C(\theta) = \begin{pmatrix} \cos\theta & -\sin\theta & 0 \\ \sin\theta & \cos\theta & 0 \\ 0 & 0 & 1 \end{pmatrix}

A reflection through the $xz$ -plane (i.e., $\sigma_v(xz)$ , which leaves $x$ and $z$ unchanged but flips the sign of $y$ ) is

\sigma(xz) = \begin{pmatrix} 1 & 0 & 0 \\ 0 & -1 & 0 \\ 0 & 0 & 1 \end{pmatrix}

Inversion through the origin flips the sign of every coordinate, giving simply $i = -\mathbf{1}_3$ (minus the $3\times3$ identity), and an improper rotation $S_n(\theta) = \sigma_h \, C_n(\theta)$ is the ordinary product of the rotation matrix with a reflection matrix through the plane perpendicular to the rotation axis. These four matrix forms — proper rotation, reflection, inversion, improper rotation — are the entire vocabulary needed to write down the matrix representative of any point-group operation in a Cartesian basis, because every point-group operation is one of these four types (the identity $E$ being $C_1$ , the trivial rotation by $0°$ ).

Figure

Cartesian rotation by θ about the z-axis, resolved into matrix elements

Beyond bare Cartesian vectors, chemists frequently need the matrix representative on a basis of atomic displacement vectors — three Cartesian axes attached to every atom in the molecule, giving a $3N$ -dimensional basis. In this basis, an operation's matrix has a special block structure: if atom $A$ is moved to the position of atom $B$ by the operation, then the $3\times3$ block coupling atom $A$ 's displacement to atom $B$ 's new displacement is nonzero, while blocks connecting atoms that are not interchanged are entirely zero. This is the origin of the well-known shortcut that only atoms left unmoved by an operation contribute to its trace — but that shortcut is a consequence, not an assumption, of writing out the full matrix honestly.

Figure

Ammonia (C3v): matrix action of C3 permuting the three N–H directions

Notice already, in the ammonia figure, the seed of the character concept: $C_3$ moves every hydrogen away from its own position, so the $3\times3$ permutation matrix on the $(H_1,H_2,H_3)$ basis has zero diagonal entries and trace $0$ — while $E$ (which moves nothing) trivially has trace $3$ . This diagonal-block bookkeeping, generalized to full $3N$ -dimensional Cartesian bases including rotational contributions from off-diagonal sine/cosine terms, is exactly the technique used later to build reducible Cartesian representations for vibrational analysis.

The core ideas

The defining property that any candidate set of matrices must satisfy to be called a representation of a group is that matrix multiplication must reproduce the group's operation table:

D(R_1)\,D(R_2) = D(R_1 R_2)

where: $R_1, R_2$ are any two operations of the point group, $R_1 R_2$ is their product as defined by the group's multiplication table (apply $R_2$ first, then $R_1$ , matching operator composition convention), and $D(R)$ denotes the matrix representative of operation $R$ in the chosen basis.

Reading: this equation says the map $R \mapsto D(R)$ is a homomorphism from the abstract group to matrices under multiplication — the essential test that a set of matrices genuinely represents the group, not just a coincidental collection of numbers. If this identity fails for even one pair of operations, the matrices do not constitute a valid representation. Every representation must also contain the identity matrix for $E$ , and every matrix must be invertible, with

D(R^{-1}) = D(R)^{-1}

where: $R^{-1}$ is the inverse operation (the operation that undoes $R$ ; for rotations, rotation by $-\theta$ ; for reflections and inversion, each operation is its own inverse).

Reading: the matrix of the inverse operation is literally the matrix inverse — a strong consistency check, since improper operations that are self-inverse (like any $\sigma$ or $i$ ) must satisfy $D(R)^2 = \mathbf{1}$ , i.e. $D(R)$ must be its own inverse, which is easily checked for reflection and inversion matrices above. For orthogonal Cartesian-type bases (rotations and reflections acting on real, orthonormal coordinate axes), every representative matrix is an orthogonal matrix:

D(R)^{\mathsf T} D(R) = \mathbf 1

where: $D(R)^{\mathsf T}$ is the transpose of $D(R)$ and $\mathbf 1$ is the identity matrix of the same dimension.

Reading: orthogonality encodes the physical fact that symmetry operations preserve lengths and angles — they are rigid motions (or rigid motions combined with a mirror flip), never stretches or shears. This guarantees $\det D(R) = \pm 1$ : $+1$ for proper rotations (including $E$ ), $-1$ for improper operations ( $\sigma$ , $i$ , $S_n$ ), which is itself a fast diagnostic for classifying an operation from its matrix alone. Finally, the trace of the matrix representative — the character — is defined as

\chi(R) = \operatorname{Tr}\big[D(R)\big] = \sum_{i=1}^n D(R)_{ii}

where: the sum runs over the diagonal elements of $D(R)$ in whatever basis was chosen.

Reading: although $D(R)$ itself depends heavily on the basis (and even on the arbitrary orientation of basis vectors within a degenerate subspace), the trace $\chi(R)$ is invariant under any change of basis related by a similarity transformation — this invariance is exactly why characters, not full matrices, become the workhorse of representation theory once one moves past constructing the representation itself.

Derivation ABuilding the 3×3 matrix for a general Cₙ rotation and verifying group closure

Goal. Construct the explicit $3\times3$ matrix for rotation by $2\pi/n$ about the $z$ -axis and confirm that repeated application $n$ times returns the identity, as required by the definition of $C_n$ .

Step 1 — Set up the geometric transformation. Consider a rotation by angle $\theta$ counterclockwise about $z$ , viewed as an active transformation of the coordinate vector (the axes stay fixed; the point moves). A point originally at polar angle $\phi$ and radius $r$ in the $xy$ -plane is described by $x = r\cos\phi$ , $y = r\sin\phi$ . After rotation by $\theta$ , the new angle is $\phi + \theta$ , so

x' = r\cos(\phi+\theta) = r\cos\phi\cos\theta - r\sin\phi\sin\theta = x\cos\theta - y\sin\theta

y' = r\sin(\phi+\theta) = r\sin\phi\cos\theta + r\cos\phi\sin\theta = x\sin\theta + y\cos\theta

and $z' = z$ trivially, since the rotation axis is $z$ .

Step 2 — Write the matrix. Collecting these three linear relations into matrix form, with the column vector $(x,y,z)^{\mathsf T}$ multiplied on the left,

C(\theta) = \begin{pmatrix} \cos\theta & -\sin\theta & 0 \\ \sin\theta & \cos\theta & 0 \\ 0 & 0 & 1 \end{pmatrix}

For the specific operation $C_n$ , set $\theta = 2\pi/n$ .

Step 3 — Confirm orthogonality. Compute $C(\theta)^{\mathsf T} C(\theta)$ directly. The upper-left $2\times2$ block is

\begin{pmatrix}\cos\theta & \sin\theta\\ -\sin\theta & \cos\theta\end{pmatrix}\begin{pmatrix}\cos\theta & -\sin\theta\\ \sin\theta & \cos\theta\end{pmatrix} = \begin{pmatrix}\cos^2\theta+\sin^2\theta & -\cos\theta\sin\theta+\sin\theta\cos\theta \\ -\sin\theta\cos\theta+\cos\theta\sin\theta & \sin^2\theta+\cos^2\theta\end{pmatrix} = \begin{pmatrix}1&0\\0&1\end{pmatrix}

confirming $C(\theta)$ is orthogonal for any $\theta$ , with $\det C(\theta) = \cos^2\theta + \sin^2\theta = 1$ , correctly identifying it as a proper rotation.

Step 4 — Verify group closure: applying $C_n$ successively $n$ times. Because rotation matrices about the same axis commute and compose by adding angles (a direct consequence of the angle-addition formulas used in Step 1), applying $C(\theta)$ twice gives $C(\theta)C(\theta) = C(2\theta)$ , and by induction $C(\theta)^k = C(k\theta)$ . Setting $\theta = 2\pi/n$ and $k=n$ ,

C\!\left(\frac{2\pi}{n}\right)^{\!n} = C(2\pi) = \begin{pmatrix}\cos 2\pi & -\sin2\pi & 0\\ \sin2\pi & \cos2\pi & 0 \\ 0&0&1\end{pmatrix} = \begin{pmatrix}1&0&0\\0&1&0\\0&0&1\end{pmatrix} = D(E)

exactly reproducing the group-theoretic requirement that $(C_n)^n = E$ . This is the matrix-level proof that the representation is faithful to the abstract cyclic structure generated by a single $n$ -fold rotation axis.

Go deeper · Derivation B · 1 minConstructing D(σᵥ) and D(S_n) from reflection and rotation building blocks, and checking σᵥ² = E

Derivation BConstructing D(σᵥ) and D(S_n) from reflection and rotation building blocks, and checking σᵥ² = E

Goal. Derive the reflection matrix for a general vertical mirror plane containing the $z$ -axis at angle $\alpha$ to the $xz$ -plane, and combine reflection with rotation to build the improper rotation matrix $S_n$ , checking self-consistency at $n=2$ ( $S_2 \equiv i$ ).

Step 1 — Reflection through the plane $xz$ . By definition, $\sigma(xz)$ leaves any vector's $x$ and $z$ components unchanged and reverses the $y$ component (the component perpendicular to the mirror plane):

\sigma(xz) = \begin{pmatrix}1&0&0\\0&-1&0\\0&0&1\end{pmatrix}

Its determinant is $\det \sigma(xz) = 1\cdot(-1)\cdot1 = -1$ , correctly flagging it as an improper operation, and $\sigma(xz)^2 = \mathbf 1$ directly by squaring the diagonal matrix, matching the physical requirement that reflecting twice through the same plane restores the original configuration.

Step 2 — Reflection through a mirror at general angle $\alpha$ . A vertical mirror plane rotated by angle $\alpha$ from the $xz$ -plane is obtained by conjugating $\sigma(xz)$ with the rotation that carries the $xz$ -plane onto the new plane:

\sigma(\alpha) = C(\alpha)\,\sigma(xz)\,C(\alpha)^{-1} = C(\alpha)\,\sigma(xz)\,C(-\alpha)

Carrying out this matrix product (using $C(-\alpha) = C(\alpha)^{\mathsf T}$ by orthogonality) yields, after simplification,

\sigma(\alpha) = \begin{pmatrix}\cos2\alpha & \sin2\alpha & 0\\ \sin2\alpha & -\cos2\alpha & 0\\ 0&0&1\end{pmatrix}

which correctly reduces to $\sigma(xz)$ at $\alpha=0$ , since $\cos 0 =1,\ \sin 0=0$ .

Step 3 — Build $S_n$ as the product of a rotation and $\sigma_h$ . An improper rotation $S_n(\theta)$ (rotate by $\theta$ about $z$ , then reflect through the $xy$ -plane, $\sigma_h$ , whose matrix is $\operatorname{diag}(1,1,-1)$ ) is the ordinary matrix product

S(\theta) = \sigma_h\, C(\theta) = \begin{pmatrix}1&0&0\\0&1&0\\0&0&-1\end{pmatrix}\begin{pmatrix}\cos\theta & -\sin\theta & 0\\ \sin\theta&\cos\theta&0\\0&0&1\end{pmatrix} = \begin{pmatrix}\cos\theta & -\sin\theta & 0\\ \sin\theta & \cos\theta & 0\\ 0&0&-1\end{pmatrix}

with determinant $\det S(\theta) = (\cos^2\theta+\sin^2\theta)\times(-1) = -1$ , confirming its improper character regardless of $\theta$ .

Step 4 — Check $S_2 = i$ . Set $\theta = 2\pi/2 = \pi$ in $S(\theta)$ : $\cos\pi = -1$ , $\sin\pi=0$ , giving

S(\pi) = \begin{pmatrix}-1&0&0\\0&-1&0\\0&0&-1\end{pmatrix} = -\mathbf 1_3 = i

exactly matching the independently known inversion matrix, and confirming the standard group-theory identity $S_2 \equiv i$ at the level of explicit matrices rather than by definition alone. As a further self-consistency check, $i^2 = (-\mathbf 1_3)(-\mathbf 1_3) = \mathbf 1_3 = D(E)$ , correctly reproducing that inversion is its own inverse.

Worked examples

Example 1 — Foundational: matrix for C₃ acting on a single (x,y) pair, with an off-round angle check. Consider the operation $C_3$ (rotation by $120°$ ) acting on the Cartesian pair $(x,y)$ , ignoring $z$ for a planar problem. Using $\theta = 120° = 2\pi/3$ , so $\cos120° = -\tfrac12$ and $\sin120° = \tfrac{\sqrt3}{2}$ :

D(C_3) = \begin{pmatrix} -\tfrac12 & -\tfrac{\sqrt3}{2} \\[2pt] \tfrac{\sqrt3}{2} & -\tfrac12 \end{pmatrix}

Apply this to a deliberately asymmetric point, $(x,y) = (2.0,\,0.7)$ (not a special angle, not equal coordinates), to verify the transformation makes geometric sense:

x' = (-\tfrac12)(2.0) + (-\tfrac{\sqrt3}{2})(0.7) = -1.0 - 0.6062 = -1.6062

y' = (\tfrac{\sqrt3}{2})(2.0) + (-\tfrac12)(0.7) = 1.7321 - 0.35 = 1.3821

Check length preservation (a required property of any orthogonal rotation matrix): original length $\sqrt{2.0^2+0.7^2} = \sqrt{4.49} = 2.1190$ ; new length $\sqrt{(-1.6062)^2+(1.3821)^2} = \sqrt{2.5799+1.9102} = \sqrt{4.4901} = 2.1190$ . The lengths agree to four significant figures, confirming orthogonality holds even for this non-special point, exactly as required by $D(C_3)^{\mathsf T}D(C_3)=\mathbf1$ .

Limiting case check: setting $\theta \to 0$ (i.e., considering $C_1 = E$ instead of $C_3$ ) gives $\cos0=1,\sin0=0$ , so $D(E) = \begin{pmatrix}1&0\\0&1\end{pmatrix}$ , the identity — the point $(2.0,0.7)$ maps to itself exactly, as it must for the identity operation.

Example 2 — Applied: full 3×3 Cartesian-basis matrices for the C₃v operations of ammonia, and verifying D(C₃)D(σᵥ) = D(σᵥ′). Ammonia belongs to $C_{3v}$ , with operations $\{E, C_3, C_3^2, \sigma_v, \sigma_v', \sigma_v''\}$ . Take the $z$ -axis along the $C_3$ axis, and place one mirror plane, $\sigma_v$ , as the $xz$ -plane. Using the results derived above:

D(E) = \begin{pmatrix}1&0&0\\0&1&0\\0&0&1\end{pmatrix}, \qquad D(C_3) = \begin{pmatrix}-\tfrac12 & -\tfrac{\sqrt3}{2} & 0\\ \tfrac{\sqrt3}{2} & -\tfrac12 & 0 \\ 0&0&1\end{pmatrix}

D(\sigma_v) = \begin{pmatrix}1&0&0\\0&-1&0\\0&0&1\end{pmatrix}

The three mirror planes of $C_{3v}$ are separated by $120°$ ; using the general mirror formula from the derivation above at $\alpha = 120°$ ( $2\alpha = 240°$ , $\cos240° = -\tfrac12$ , $\sin240° = -\tfrac{\sqrt3}{2}$ ) gives the matrix for the mirror plane $\sigma_v'$ obtained by rotating $\sigma_v$ by $C_3$ :

D(\sigma_v') = \begin{pmatrix}-\tfrac12 & -\tfrac{\sqrt3}{2} & 0\\ -\tfrac{\sqrt3}{2} & \tfrac12 & 0\\ 0&0&1\end{pmatrix}

Now verify the group-closure test central to this article: the product $D(C_3)D(\sigma_v)$ should equal $D(\sigma_v')$ (rotating the mirror plane by $C_3$ produces the adjacent mirror plane — a standard $C_{3v}$ relation). Multiply the upper-left $2\times2$ blocks:

\begin{pmatrix}-\tfrac12 & -\tfrac{\sqrt3}{2} \\ \tfrac{\sqrt3}{2} & -\tfrac12\end{pmatrix} \begin{pmatrix}1&0\\0&-1\end{pmatrix} = \begin{pmatrix}-\tfrac12\cdot1 + (-\tfrac{\sqrt3}{2})\cdot0 & -\tfrac12\cdot0+(-\tfrac{\sqrt3}{2})\cdot(-1) \\ \tfrac{\sqrt3}{2}\cdot1+(-\tfrac12)\cdot0 & \tfrac{\sqrt3}{2}\cdot0+(-\tfrac12)(-1)\end{pmatrix}

= \begin{pmatrix}-\tfrac12 & \tfrac{\sqrt3}{2} \\ \tfrac{\sqrt3}{2} & \tfrac12\end{pmatrix}

This does not immediately look like $D(\sigma_v')$ written above (signs differ), which is a useful, deliberately non-round check that exposes a subtlety: the labeling convention for which mirror plane is " $\sigma_v'$ " (rotated by $+120°$ vs. reached by $C_3^2$ ) must be fixed consistently. Redoing the comparison against the mirror plane obtained by conjugation $D(C_3)D(\sigma_v)D(C_3)^{-1}$ — the mathematically unambiguous way to ask "which mirror plane does $C_3$ carry $\sigma_v$ into" — gives

D(C_3)D(\sigma_v)D(C_3)^{-1} = D(C_3)D(\sigma_v)D(C_3^2)

Using $D(C_3^2)$ (rotation by $240°$ , $\cos240°=-\tfrac12,\sin240°=-\tfrac{\sqrt3}{2}$ ) and multiplying through confirms the result equals $D(\sigma_v')$ as written, with the upper-left block $\begin{pmatrix}-\tfrac12&-\tfrac{\sqrt3}{2}\\-\tfrac{\sqrt3}{2}&\tfrac12\end{pmatrix}$ reproduced exactly. The lesson: matrix representation checks are only as good as a fixed, explicit convention for operation order and axis placement — precisely why textbook character tables fix a standard orientation once and for all.

Limiting case check: at $C_3^0 = E$ (no rotation), $D(E)D(\sigma_v)D(E)^{-1} = D(\sigma_v)$ trivially, confirming a mirror plane conjugated by the identity maps to itself, as required.

Example 3 — Exam-level: reducible Cartesian representation matrix for the C₄ operation in a D₄ₕ-type XeF₄ Cartesian-displacement basis, extracting the character by inspection. Square-planar $\mathrm{XeF_4}$ belongs to $D_{4h}$ ; consider only the $C_4$ operation (rotation by $90°$ about the principal axis, taken along $z$ ) acting on the $3N = 15$ -dimensional Cartesian displacement basis (Xe plus 4 F atoms, three coordinates each). Rather than writing the full $15\times15$ matrix, exploit the block-diagonal structure emphasized earlier: the trace only receives contributions from atoms that are not moved to a different position by the operation, since any atom sent elsewhere contributes an off-diagonal block with all-zero diagonal entries. Under $C_4$ (rotation by $90°$ about the $z$ -axis through Xe), the four fluorines cyclically permute ( $F_1\to F_2\to F_3\to F_4\to F_1$ ), so none of them stays fixed; only the central Xe atom is unmoved. The $3\times3$ block for the unmoved Xe atom is exactly the rotation matrix $C(90°)$ derived above, with $\cos90°=0,\ \sin90°=0$ ... more precisely $\cos90°=0$ and $\sin90°=1$ :

D_{\mathrm{Xe}}(C_4) = \begin{pmatrix}0 & -1 & 0\\ 1& 0 & 0\\ 0&0&1\end{pmatrix}

Its trace is $\operatorname{Tr}\big[D_{\mathrm{Xe}}(C_4)\big] = 0+0+1 = 1$ . Since all four fluorine blocks lie entirely off the main block-diagonal (each F's post-rotation position coincides with a different F's original position, contributing to an off-diagonal $3\times3$ block rather than a diagonal one), they contribute exactly zero to the total trace. Hence the character of the full $15$ -dimensional Cartesian representation under $C_4$ is

\chi_{3N}(C_4) = \chi_{\mathrm{Xe}}(C_4) + 4\times 0 = 1

This matches the standard shortcut formula for Cartesian representations, $\chi(R) = N_{\text{unmoved}}\times(1+2\cos\theta)$ , checked explicitly here: with $N_{\text{unmoved}}=1$ (only Xe) and $\theta=90°$ so $2\cos90°=0$ , the formula gives $1\times(1+0)=1$ , exactly reproducing the trace computed directly from the $3\times3$ block above — an independent algebraic confirmation that the block-diagonal bookkeeping and the general trace formula agree.

Limiting case check: applying the same logic to $E$ (the identity, $\theta=0$ , $N_{\text{unmoved}}=5$ since every atom stays put) gives $\chi_{3N}(E) = 5\times(1+2\cos0°) = 5\times3 = 15$ , correctly equal to the total dimension of the basis ( $3N=15$ ), as it must be — the identity operation is always represented by the identity matrix, whose trace equals its dimension.

Limiting cases, scope & common traps

Situation	What happens to $D(R)$	Why
$R = E$	$D(E) = \mathbf 1_n$ , always, regardless of basis	Identity operation changes nothing, so every basis vector maps to itself
Proper rotation $C_n$ , $\theta\to0$	$D(C_n)\to \mathbf1_n$	Continuity of $\cos\theta\to1$ , $\sin\theta\to0$ recovers the identity smoothly
Self-inverse improper operation ( $\sigma$ , $i$ )	$D(R)^2 = \mathbf1_n$	Reflecting or inverting twice restores the original configuration
Degenerate basis (e.g. two $\pi_x,\pi_y$ orbitals)	$D(R)$ is genuinely $2\times2$ and mixes the basis functions, not diagonal	A $C_n$ rotation rotates $\pi_x,\pi_y$ into linear combinations of each other
Atom left unmoved by $R$	Contributes its full local $3\times3$ block to the diagonal of a Cartesian representation	Only unmoved atoms can have nonzero diagonal entries, hence nonzero trace contribution
Atom moved to another atom's site	Contributes an off-diagonal block; diagonal contribution is exactly zero	The new position vector belongs to a different basis vector entirely
Choice of basis orientation within a degenerate pair	$D(R)$ changes (different numerical entries)	Basis change is a similarity transformation; only the trace $\chi(R)$ stays invariant

Common trapTwo persistent misconceptions

First, students often assume the matrix representative $D(R)$ is a fixed, universal object attached to the operation itself. It is not: $D(R)$ depends entirely on the chosen basis, and the same operation $C_3$ is a $1\times1$ number ( $\chi=1$ for a totally symmetric function), a $2\times2$ matrix (for a degenerate $e$ -orbital pair), or part of a $15\times15$ matrix (for a full Cartesian displacement basis) — only the character $\chi(R)$ within each representation, not the matrix entries themselves, is basis-independent (up to similarity transformation). Second, in constructing Cartesian displacement representations, it is tempting to assume every atom contributes to the trace. Only atoms whose position is unchanged by the operation contribute — an atom permuted elsewhere contributes an off-diagonal block that is entirely invisible to the trace, however large the block's individual matrix entries might be.

The chemist behind it

The systematic use of matrices to represent symmetry operations traces to the broader mathematical development of group representation theory in the late nineteenth and early twentieth centuries, later imported wholesale into chemistry and molecular spectroscopy.

Origins · 1 minHans Bethe and how this was discovered

Hans Bethe

1906–2005 · Theoretical physics, Munich and later Cornell/Los Alamos

Hans Bethe's 1929 paper on crystal-field splitting in ionic crystals is widely credited as the first systematic application of point-group representation theory to a chemistry-adjacent problem — determining how a central ion's $d$ -orbitals split under the symmetry of surrounding ligands. Bethe built exactly the machinery this article describes: writing the point-group operations of an octahedral or tetrahedral environment as matrices acting on the five $d$ -orbitals, then using their traces (characters) to decompose the orbital set into symmetry-adapted combinations. This "Bethe method" became the direct ancestor of ligand-field theory and, more broadly, of the standard practice — now universal in inorganic and physical chemistry — of building a reducible representation on a chosen chemical basis and reducing it using the character table. Bethe's later career moved into nuclear and astrophysics (he received the 1967 Nobel Prize in Physics for the theory of stellar nucleosynthesis), but this early group-theoretic work quietly set the template every modern spectroscopy course still follows.

History

- 1854 — Arthur Cayley formalizes abstract group theory and the notion of a group multiplication table, the algebraic skeleton later populated by matrices.

- 1896–1897 — Georg Frobenius develops the theory of group characters and representations in full generality, providing the mathematical foundation (orthogonality of characters, irreducibility) used ever since.

- 1929 — Hans Bethe applies point-group representation theory to crystal-field splitting, the first major chemistry-facing use of matrix representations of symmetry operations.

- 1930s — Eugene Wigner systematizes the application of group representation theory to quantum mechanics, formalizing the connection between symmetry operators and matrix representatives on Hilbert-space bases.

- 1963 — F. Albert Cotton's textbook "Chemical Applications of Group Theory" standardizes the pedagogical presentation (matrices, characters, reduction formula) still used in chemistry curricula today.

Why does the same symmetry operation sometimes appear as a 1×1, 2×2, or larger matrix?

The matrix representative's dimension is fixed entirely by the chosen basis, not by any intrinsic property of the operation. A totally symmetric function (like an s orbital) is unchanged by every operation, so its matrix representative is always the $1\times1$ number $1$ . A pair of degenerate $p_x,p_y$ orbitals forms a two-dimensional basis, so operations that mix $p_x$ and $p_y$ (like a rotation) require $2\times2$ matrices. A basis of $3N$ Cartesian displacement coordinates for an $N$ -atom molecule requires $3N\times3N$ matrices. The abstract operation ( $C_3$ , say) is a single geometric fact, but its numerical matrix representative always answers the question "acting on what?" first.

Do matrix representations have to use orthogonal (Cartesian-type) bases?

No. Orthogonality of $D(R)$ specifically reflects the fact that Cartesian coordinates form an orthonormal basis and that symmetry operations preserve lengths and angles in real space. If the chosen basis functions are not orthonormal — for example, a set of non-normalized atomic orbitals — the matrix representative can have a more general (non-orthogonal) form, though it will still satisfy the fundamental homomorphism property $D(R_1)D(R_2)=D(R_1R_2)$ . In practice, chemists almost always work with orthonormalized bases (Cartesian axes, orthonormalized SALCs) specifically so that the clean orthogonality and determinant properties hold.

How is the matrix for an improper rotation Sₙ related to the matrices for Cₙ and σₕ separately?

By direct construction, $S_n(\theta) = \sigma_h\, C_n(\theta)$ as an ordinary matrix product — rotate first, then reflect through the plane perpendicular to the rotation axis (in the convention used throughout this article, matrices act on column vectors from the left, so the rightmost matrix acts first). Because $\sigma_h = \operatorname{diag}(1,1,-1)$ only modifies the $z$ -row, the resulting $S_n(\theta)$ matrix has exactly the same upper-left $2\times2$ rotation block as $C_n(\theta)$ , with the $(3,3)$ entry flipped from $+1$ to $-1$ . This single sign flip is what changes the determinant from $+1$ (proper) to $-1$ (improper).

Why does only the trace of D(R), not the full matrix, end up mattering for most chemistry applications?

Chemists ultimately want basis-independent, physically meaningful numbers — quantities that do not depend on an arbitrary choice of coordinate orientation within a degenerate orbital set, for instance. The trace (character) of $D(R)$ has exactly this invariance property under similarity transformations, while the individual matrix elements do not. This is precisely why the character table — a compact record of just the traces $\chi(R)$ for each class of operation, for each irreducible representation — replaces the far bulkier full matrices in almost every practical chemistry application, from vibrational analysis to molecular-orbital symmetry classification.

Can two completely different-looking matrices represent the same operation correctly?

Yes, and this is expected rather than an error. Two matrices $D(R)$ and $D'(R) = P^{-1}D(R)P$ related by any invertible similarity transformation $P$ (corresponding to a change of basis) represent the same abstract operation equally validly — they simply describe it in different coordinate systems or different choices of basis functions within a degenerate subspace. What must stay fixed across any valid choice is the group-multiplication consistency $D(R_1)D(R_2)=D(R_1R_2)$ and the trace $\chi(R)$ ; the individual numerical entries are basis artifacts.

How does this connect to constructing symmetry-adapted molecular orbitals?

Once a full matrix representation is built on a chosen atomic-orbital basis (for example, the four sp³-like fluorine orbitals pointing at a central atom in $\mathrm{XeF_4}$ ), the trace of each operation's matrix gives a reducible representation $\Gamma$ . Standard reduction formulas — built from the character table and requiring no further matrix algebra — decompose $\Gamma$ into irreducible representations, whose labels immediately identify which symmetry-adapted linear combinations (SALCs) of atomic orbitals will mix correctly with central-atom orbitals of matching symmetry. This entire downstream workflow, covered under Reducible and Irreducible Representations, is only possible because the matrix representation was constructed correctly and consistently at the outset.

What is the fastest way to check that a proposed matrix for an operation is correct?

Three checks catch essentially every construction error. First, confirm $\det D(R) = +1$ for proper rotations (including $E$ ) and $-1$ for improper operations ( $\sigma$ , $i$ , $S_n$ ) — an immediate red flag if violated. Second, confirm orthogonality, $D(R)^{\mathsf T}D(R) = \mathbf1$ , which must hold for any operation acting on a real orthonormal Cartesian-type basis. Third, and most powerfully, confirm at least one nontrivial multiplication relation from the group's known structure (for example, $D(C_n)^n = D(E)$ , or $D(\sigma)^2=D(E)$ , or a specific product like $D(C_3)D(\sigma_v)$ landing on the matrix for the correct adjacent mirror plane) — this is the test that catches sign errors invisible to the determinant and orthogonality checks alone.