Matrix Representations of Symmetry Operations

Matrix representations are the computational backbone of every symmetry-based prediction made in modern chemistry, even when a working chemist never writes the matrices explicitly. Quantum chemistry software (Gaussian, ORCA, Psi4) uses point-group matrix representations internally to symmetrize molecular orbitals, block-diagonalize the Fock or Kohn–Sham matrix by irreducible representation, and dramatically cut computational cost — a calculation on benzene ($D_{6h}$) can be blocked into several much smaller symmetry sectors rather than diagonalized as one large matrix, because the Hamiltonian commutes with every $D(R)$ of the molecular point group. Vibrational spectroscopy software builds the $3N$-dimensional Cartesian representation exactly as sketched in this article (tracking which atoms are unmoved by each operation and how their displacement vectors transform) to predict which normal modes are IR- or Raman-active before any experiment is run, a technique used routinely in identifying isomers of coordination complexes and in interpreting the vibrational fingerprint region of organic and inorganic unknowns. Crystallographers use the matrix representations of space-group operations (an extension including translations) to generate the full set of atomic positions in a unit cell from a crystallographic asymmetric unit, a step performed automatically by every structure-refinement package (SHELX, Olex2) each time a `.cif` file is processed. In ligand field and molecular orbital theory, the matrices built on ligand $\sigma$- or $\pi$-orbital bases (as in the $\mathrm{XeF_4}$ and $\mathrm{NH_3}$ examples above) are the direct precursors to symmetry-adapted linear combinations that predict the splitting pattern of d-orbitals in octahedral and tetrahedral complexes, underlying the interpretation of UV-vis spectra, magnetism, and catalytic activity across transition-metal chemistry.

Ask the tutor: How do I turn a big reducible representation like $\Gamma_{3N}$ into the small set of irreducible representations that tell me which vibrations are IR-active?

A symmetry operation $R$ (rotation, reflection, inversion, improper rotation, or the identity) moves points in space. To do arithmetic with it, pick a basis — a set of vectors or functions $\{f_1, f_2, \dots, f_n\}$ that spans some space of interest (Cartesian axes, bond vectors, atomic orbitals, atomic Cartesian displacements). The operation $R$ sends each basis function to some linear combination of the basis functions:

The array of coefficients $D(R)_{ij}$ is the matrix representative of $R$ in that basis. Column $j$ of the matrix tells you where basis vector $f_j$ ends up, expressed in terms of the original basis. This convention — reading transformed basis vectors down columns — is the one used throughout; it is exactly analogous to how a rotation matrix's columns are the images of the standard unit vectors. For proper rotations $C_n(\theta)$ about the $z$-axis (with $\theta = 2\pi/n$ or an integer multiple thereof), elementary trigonometry gives the images of $\hat x, \hat y, \hat z$:

The $z$-axis is left invariant (it is the rotation axis), so the third row/column is trivially $(0,0,1)$. For a reflection $\sigma$ through a plane containing two of the basis axes, the matrix is diagonal with a $-1$ in the position of the axis perpendicular to the mirror. A reflection in the $xz$-plane (so $y \to -y$) is

Inversion through the origin sends every coordinate to its negative:

An improper rotation $S_n$ is a rotation by $2\pi/n$ followed by reflection in the plane perpendicular to the rotation axis ($\sigma_h$):

These four matrix forms — proper rotation, mirror reflection, inversion, improper rotation — are the entire toolkit; every point-group operation is one of them, expressed in whatever coordinate frame is natural for the molecule. Crucially, representations do not have to act on $(x,y,z)$. They can act on any basis that the point group permutes or mixes: the set of ligand $\sigma$-bond vectors in an octahedral complex, the $2p$ orbitals on a central atom, the three N–H bonds of ammonia, or all $3N$ Cartesian atomic displacement coordinates used in vibrational analysis. Choosing the basis is the modeler's decision; building the matrix is then pure geometry.

The defining algebraic property of a representation is that it turns group multiplication into matrix multiplication:

where: $D(R)$ and $D(S)$ are the matrices assigned to operations $R$ and $S$ in a fixed basis, and $RS$ denotes the group product (apply $S$ first, then $R$, matching the Symmetry Operations as a Mathematical Group convention). This equation says the matrices are not arbitrary — their multiplication table must mirror the operation table exactly, so $D$ is a genuine homomorphism, not just a bookkeeping device. Reading: if you know the matrices for the generating operations of a group, every other matrix is fixed by matrix multiplication; you never need to re-derive a matrix from geometry once enough generators are known. Every representation matrix must also be invertible, with inverse equal to the matrix of the inverse operation:

where: $R^{-1}$ is the inverse symmetry operation guaranteed to exist by the group axioms. For orthogonal transformations (rotations, reflections — all point-group operations preserve lengths and angles), $D(R)^{-1} = D(R)^{T}$, so representation matrices for point groups are always orthogonal matrices: $D(R)^T D(R) = \mathbb{1}$. Reading: this orthogonality is not an extra assumption — it follows because every point-group operation is a rigid motion (isometry) of 3-D space, hence its matrix (in an orthonormal Cartesian basis) automatically satisfies $D^TD=\mathbb 1$. Finally, if the basis is changed by an invertible matrix $P$ (new basis vectors are linear combinations of old ones, $f_i' = \sum_k P_{ki} f_k$), every representation matrix transforms by a similarity transformation:

where: $D(R)$ is the matrix in the old basis, $D'(R)$ the matrix in the new basis, and $P$ the (fixed, $R$-independent) change-of-basis matrix. The homomorphism property survives this transformation automatically, since $P^{-1}D(R)P \cdot P^{-1}D(S)P = P^{-1}D(R)D(S)P = P^{-1}D(RS)P$. Reading: individual matrix entries are basis-dependent and hence not physically special, but the trace of $D(R)$ is invariant under similarity transformations ($\mathrm{tr}(P^{-1}AP) = \mathrm{tr}(A)$), so the character $\chi(R) = \mathrm{tr}\,D(R)$ is the basis-independent fingerprint of an operation — the quantity that survives into character tables.

Example 1 — Foundational: the $S_4$ matrix and confirming $S_4^2 = C_2$. Take an $S_4$ improper rotation about $z$: $\theta = 90°$, so $\cos\theta = 0$, $\sin\theta = 1$. Using $S_n = \sigma_h C_n$:

Check this against the definitional identity $S_4^2 = C_2$ (two successive $S_4$ operations must equal a $C_2$ rotation, since two reflections in the same plane cancel while the rotations add to 180°). Multiply the matrix by itself:

This is exactly $D(C_2)$ derived earlier (Step 2 of the first derivation). The trace check: $\chi(S_4) = 0+0-1 = -1$, and $\chi(S_4^2) = -1-1+1=-1 = \chi(C_2)$, consistent. This example is deliberately asymmetric in structure (mixing rotation and reflection) precisely so that a sign error in either the rotation block or the $\sigma_h$ block would be caught — a purely diagonal example (like inversion alone) would not expose an error in the off-diagonal rotation terms.

Example 2 — Applied: representation of $C_{4v}$ acting on the four F ligand $\sigma$-orbitals of square-planar $\mathrm{XeF_4}$ under a non-principal operation. $\mathrm{XeF_4}$ belongs to $D_{4h}$, but restricting attention to its $C_{4v}$ subgroup of operations acting in the molecular plane is a common step when building the $\sigma$-only ligand representation for MO analysis. Label the four F $\sigma$ orbitals $f_1, f_2, f_3, f_4$ going anticlockwise, with $f_1$ along $+x$, $f_2$ along $+y$, $f_3$ along $-x$, $f_4$ along $-y$. Consider the operation $\sigma_v(xz)$: this mirror plane contains the $x$-axis and reflects $y \to -y$, so it fixes the ligands lying on the mirror ($f_1$ and $f_3$, both on the $x$-axis) and swaps the two off-axis ligands ($f_2 \leftrightarrow f_4$).

Read column by column: column 1 (image of $f_1$) is $(1,0,0,0)^T$ — $f_1$ is unchanged; column 2 (image of $f_2$) is $(0,0,0,1)^T$ — $f_2 \to f_4$; column 3 (image of $f_3$) is $(0,0,1,0)^T$ — unchanged; column 4 (image of $f_4$) is $(0,1,0,0)^T$ — $f_4 \to f_2$. Trace: $\chi(\sigma_v) = 1+0+1+0 = 2$. Compare with $C_4$ (90° rotation, $f_1\to f_2\to f_3\to f_4\to f_1$): every basis vector moves to a different one, so all diagonal entries are zero and $\chi(C_4) = 0$. This asymmetry between a mirror through two ligands (character 2, from the two fixed ligands) and a rotation through none (character 0) illustrates the general shortcut used later: the character of a permutation-type representation equals the number of basis objects left unmoved by the operation — a fact that follows immediately from the matrix construction (only unmoved basis vectors contribute a nonzero diagonal entry) and is exploited constantly when building $\Gamma_{\text{ligand}}$ or $\Gamma_{3N}$ representations in vibrational analysis.

Example 3 — Exam/competition level: verifying a similarity transformation changes matrix entries but not the trace, using a rotated basis in $C_{3v}$. Take the $D(C_3)$ matrix from the ammonia bond-vector derivation,

Now change to a new, non-orthonormal basis defined by

(these are exactly the symmetry-adapted-type combinations that will reappear when this representation is reduced). The change-of-basis matrix, whose columns express $g_i$ in terms of $r_1,r_2,r_3$, is

Its inverse (computed by cofactor expansion; $\det P = 1(1-0) -1(-1-1)+0 = 1+2 = 3$) is

Now compute $D'(C_3) = P^{-1} D(C_3) P$. First, $D(C_3) P$:

Then multiply by $P^{-1}$:

The entries have completely changed — $D'(C_3)$ is no longer a simple permutation matrix; the $g_1$ row/column has decoupled ($g_1$ transforms into itself alone, coefficient 1, since $g_1$ is the totally symmetric sum invariant under any relabeling of the bonds). But check the trace:

which exactly equals $\chi(C_3) = 0$ computed in the original bond basis. The similarity transformation scrambled every matrix entry yet left the character untouched, exactly as the invariance $\mathrm{tr}(P^{-1}AP)=\mathrm{tr}(A)$ guarantees — and it also reveals, as a bonus, that $g_1$ already sits in its own invariant 1-D block, an early hint of how Reducible and Irreducible Representations extracts the $A_1$ block from a larger reducible representation.

The matrix formalism used here draws on nineteenth-century group representation theory developed by mathematicians — most centrally Ferdinand Georg Frobenius and later Issai Schur, who built the theory of group representations and characters in the 1890s–1900s, decades before it was applied to molecules. The bridge into chemistry and spectroscopy was built by Hans Bethe (crystal field theory, 1929) and especially by Eugene Wigner, whose 1931 book Gruppentheorie und ihre Anwendung auf die Quantenmechanik der Atomspektren systematically applied representation theory to atomic and molecular quantum mechanics. Wigner showed that the matrices representing symmetry operations on a quantum-mechanical basis of wavefunctions dictate selection rules, degeneracies, and the block structure of the Hamiltonian — ideas absorbed directly into chemical group theory by the mid-twentieth century through workers like Robert Mulliken (who standardized much of the notation for irreducible representations used in character tables) and the widespread textbook synthesis of Cotton's Chemical Applications of Group Theory.