Characters as Matrix Traces

Characters are not a historical curiosity confined to textbooks — they are the computational backbone of every symmetry-based prediction made in modern chemistry and materials science. Vibrational spectroscopy (IR and Raman activity of normal modes) is decided entirely by whether a given irreducible representation's character matches the character of $x, y, z$ (IR-active) or of quadratic functions like $x^2, xy$ (Raman-active) — a determination that always starts from computing $\chi$ of the full reducible representation of atomic motion, exactly as demonstrated for the N–H stretches above, before reduction. Crystal-field and ligand-field theory, used to rationalize the color and magnetism of transition-metal complexes, relies on characters of the rotation group restricted to point-group operations to predict how d-orbitals split in octahedral, tetrahedral, or square-planar environments. In modern computational chemistry, quantum-chemical software (Gaussian, ORCA, and similar packages) automatically exploits point-group symmetry by first constructing reducible representations of the atomic-orbital or molecular-orbital basis and computing their characters, which lets the program block-diagonalize huge Fock or Hamiltonian matrices into small symmetry-adapted blocks — turning what would be an intractable diagonalization into a set of small, fast ones. Materials scientists use character-based selection rules to interpret Raman and IR spectra of crystals and to predict which phonon modes couple to light. Even outside chemistry, the trace-invariance idea that gives characters their power reappears throughout physics: it is the same mathematical fact used to define invariant quantities in particle physics representation theory and in the classification of quantum-mechanical angular momentum states. Every one of these applications ultimately depends on the two theorems proven here: that the trace does not care about your basis, and that it does not care which member of a conjugacy class you picked.

Ask the tutor: Can you walk through how the character of the full 3N-dimensional Cartesian representation for a bent triatomic molecule like H₂O is computed operation-by-operation?

Start with something concrete: a matrix representation of a point group assigns, to every symmetry operation $\hat R$, a square matrix $D(\hat R)$ that tells you how a chosen set of basis functions transforms. If your basis is the three 2p orbitals on a central atom, $D(\hat R)$ is a 3×3 matrix; if your basis is the twelve Cartesian displacement coordinates of a four-atom molecule, $D(\hat R)$ is a 12×12 matrix. These matrices multiply together exactly the way the operations compose — that is what "representation" means (it is a group homomorphism into matrices). The catch is that the matrix itself depends on an arbitrary choice: which orbitals you numbered first, which Cartesian axes you drew, whether you used $p_x, p_y, p_z$ or some rotated combination. Chemists want a description of the symmetry content of a basis that does not depend on such bookkeeping choices. The trace supplies exactly that. If you change basis with an invertible matrix $S$, the new matrix is $D'(\hat R) = S^{-1} D(\hat R) S$, and the trace of $D'(\hat R)$ equals the trace of $D(\hat R)$ — always, for any $S$. So while the full matrix carries basis-dependent detail, the single number $\chi(\hat R) = \mathrm{Tr}\,D(\hat R)$ is a genuine, basis-free property of the operation acting on that representation. A second and equally important redundancy is built into the group itself. Symmetry operations that "look the same" from different vantage points — the three C₃ rotations about the same axis in NH₃, or the three σᵥ mirror planes — are related to each other by conjugation: $\hat R' = \hat X\hat R\hat X^{-1}$ for some other operation $\hat X$ in the group. Conjugate operations form a class. It turns out that if $\hat R$ and $\hat R'$ are conjugate, their representation matrices are related by a similarity transformation (built from $D(\hat X)$), and so — by the very same trace argument — they share the identical character. This is the reason a character table lists one column per class (like "$2C_3$" or "$3\sigma_v$") rather than one column per individual operation: the character genuinely does not distinguish between conjugate operations, so listing them separately would just repeat the same number.

The picture above is worth dwelling on because it is the cleanest possible illustration of the "unmoved-atom" rule that later becomes indispensable for building reducible representations of molecular motion. In the basis of the three N–H bond vectors, the matrix for $E$ is the 3×3 identity (trace 3, the dimension of the basis). The matrix for $C_3$ is a pure permutation matrix — it sends $r_1\to r_2\to r_3\to r_1$ — and a permutation matrix with no fixed points has all zeros on the diagonal, so its trace is 0. The matrix for $\sigma_v$ fixes one bond vector and swaps the other two, giving exactly one nonzero diagonal entry, so its trace is 1. Every one of these traces was computed without ever writing down which of $C_3$'s three "copies" ($C_3$ or $C_3^2$) you used, or which of the three mirror planes — and indeed all three $\sigma_v$ planes give trace 1, all consistent with the class-invariance argument.

This second figure demonstrates the invariance numerically rather than only abstractly: a $120°$ rotation matrix in an orthonormal 2D basis has trace $2\cos 120° = -1.0$... but here we deliberately chose a non-orthogonal, skewed second basis to make the point sharply — even though the individual matrix entries change completely (0.5, −0.87, 0.87, 0.5 become 1.0, 0.0, 0.32, 0.0), the trace is unchanged. That resilience is not a coincidence of this particular numeric example; it is a theorem, proved below.

The defining relationship of this article is simply:

where:

• $D(\hat R)$ is the $n\times n$ matrix representing operation $\hat R$ in some chosen basis of $n$ functions,
• $D(\hat R)_{ii}$ is the $i$-th diagonal entry of that matrix,
• $n$ is the dimension of the representation (number of basis functions),
• $\chi(\hat R)$, the character, is a single complex (usually real, often integer) number attached to $\hat R$ in this representation.

Reading it: the character discards all the off-diagonal "mixing" information in the matrix — which basis function turns into which linear combination of others — and keeps only the diagonal sum, i.e., how much each basis function maps back "onto itself" (in coefficient, not necessarily literally unchanged), summed over the whole basis. The two invariance properties that make this number useful are:

where:

• $A$ is any square matrix (here, $D(\hat R)$ in the original basis),
• $S$ is the change-of-basis matrix,
• $S^{-1}AS$ is $A$ re-expressed in the new basis.

Reading it: whatever basis you pick to build your representation, the trace — and hence the character — comes out the same. Characters are properties of the operation and the representation, not of your coordinate choices.

where:

• $\hat X$ is any operation in the point group,
• $\hat R'$ is the operation conjugate to $\hat R$ via $\hat X$ (they belong to the same class),
• the implication follows because $D(\hat R') = D(\hat X)D(\hat R)D(\hat X)^{-1}$ is a similarity transform of $D(\hat R)$.

Reading it: all operations within one conjugacy class — e.g. all three $C_3$-type rotations in NH₃, or all three vertical mirrors — carry exactly the same character, in every representation. This is precisely why a character table needs only one column per class.

Example 1 — Foundational: characters of the identity and a C₂ rotation on a 2D orbital basis. Consider a basis of two orbitals, $p_x$ and $p_y$, on a central atom sitting on a $C_2$ axis (take the axis along $z$). The identity operation obviously leaves both basis functions completely unchanged, so

This illustrates the general rule $\chi(E) = n$ (the dimension), since $D(E)$ is always the $n\times n$ identity matrix regardless of basis, and the sum of $n$ ones is $n$. Now apply $C_2$ (180° rotation about $z$) to $(p_x, p_y)$. A 2D rotation by angle $\theta$ acting on $(x,y)$ has matrix

So $\chi(C_2) = -1 + (-1) = -2$. Check with the general rotation-character formula $\chi(C_n) = 2\cos(360^\circ/n)$ for a 2D basis transforming like $(x,y)$: with $n=2$, $\chi(C_2) = 2\cos180^\circ = 2(-1) = -2$, matching exactly. As a limiting-case sanity check, setting $n=1$ (i.e., the identity, $\theta=0$) gives $\chi = 2\cos 0^\circ = 2$, correctly reproducing $\chi(E)=2$ found above by direct matrix construction.

Example 2 — Applied: the reducible representation of NH₃'s three N–H bond stretches under all of C₃ᵥ. Take the basis $\{r_1, r_2, r_3\}$ of the three N–H bond-stretch coordinates in ammonia (point group $C_{3v}$, operations $E, 2C_3, 3\sigma_v$). We compute the character of each class using the "unmoved basis function" shortcut, justified because any basis function that is moved to a different position contributes a zero to that diagonal entry (it maps entirely onto a different basis vector, contributing nothing to its own coefficient), while an unmoved bond vector contributes exactly $+1$ (bond order/length is preserved, so the coefficient mapping it to itself is $1$). For $E$: all three bonds are fixed in place (trivially), so

For $C_3$ (or $C_3^2$, same class, same character by the theorem proved above): the rotation cyclically permutes $r_1 \to r_2 \to r_3 \to r_1$, leaving no bond vector fixed, so every diagonal entry of the corresponding permutation matrix is 0:

For $\sigma_v$ (a mirror plane containing the $C_3$ axis and one N–H bond, say bond 1): that bond lies in the mirror and is left fixed, while the mirror swaps the other two bonds, $r_2 \leftrightarrow r_3$ (again mapping each entirely onto a different vector, contributing 0 to their own diagonal terms):

The resulting reducible representation is summarized as $\Gamma_{\text{stretch}}: \chi(E)=3,\ \chi(C_3)=0,\ \chi(\sigma_v)=1$. As a limiting-case check, note that $\chi(E)=3$ matches the general rule that the character of the identity equals the dimension of the basis (three bond vectors), confirming internal consistency before this reducible representation is ever decomposed into irreducible pieces (a task for Reducible and Irreducible Representations).

Example 3 — Exam/competition level: characters of a doubly-degenerate E representation in C₄ᵥ from an explicit non-round similarity transformation. This example is designed to make the invariance theorem bite on numbers that are not simply integers, so that no student can suspect the result was rigged by symmetry of the inputs. Suppose in $C_{4v}$ (operations $E, C_4, C_4^3, C_2, 2\sigma_v, 2\sigma_d$) a two-dimensional representation $E$ acts on the orthonormal basis $(p_x, p_y)$. The matrix for the class-representative rotation $C_4$ (90°) is the standard 2D rotation matrix:

Now deliberately re-express this same operation in a skewed, non-orthonormal basis to test the invariance theorem with asymmetric numbers. Choose the change-of-basis matrix

The inverse is

Compute $D(C_4)S$ first:

Then

Working the top-left entry: $0.5542(0.6) + (-0.0964)(1.7) = 0.3325 - 0.1639 = 0.1686$. Working the bottom-right entry: $0.1446(-2.3) + 0.4096(0.4) = -0.3326 + 0.1638 = -0.1688$. So

matching $\chi(C_4)=0$ found in the clean orthonormal basis, with the tiny residual (−0.0002) attributable purely to rounding at 4 decimal places rather than any real discrepancy — confirming the similarity-invariance theorem numerically even for a thoroughly asymmetric, non-orthogonal $S$ with no special angles or round entries. As the required limiting-case check: if instead $S$ were chosen as the identity matrix ($S=\mathbb 1$), the whole calculation trivially collapses to $D'(C_4) = D(C_4)$ and $\chi=0$ exactly with no rounding at all — verifying that the general procedure correctly reduces to the trivial "no change of basis" case.

The formalism of representation theory and characters was not built by chemists at all, but imported wholesale from pure mathematics developed decades before its first application to molecules. The pivotal figure is Georg Frobenius, whose work on group characters in the 1890s laid the entire algebraic foundation — orthogonality relations, class functions, the character table itself as an object — that chemists would later apply to point groups of molecules. Frobenius worked in an era of pure abstract algebra with no thought of vibrational spectra or crystal-field splittings; the "character" of a representation was, for him, a tool for classifying representations of abstract finite groups. The translation of this machinery into chemistry is largely credited to Hans Bethe, whose 1929 treatment of crystal-field splitting in transition-metal ions used group-theoretical characters to predict how atomic d-orbital levels split in octahedral and tetrahedral environments, and to Eugene Wigner, whose applications of group theory to quantum mechanics (including molecular and atomic spectra) cemented characters as the working tool of the field. By the time character tables became a standard part of physical-inorganic-chemistry curricula (through texts refining the Bethe–Wigner program in the mid-20th century), the character had become exactly what this article defines it as: the trace of a representation matrix, tabulated once per class, and used to decode everything from orbital splitting patterns to infrared/Raman selection rules.