Symmetry Operations as a Mathematical Group

Every modern computational chemistry package (Gaussian, ORCA, VASP, CASTEP) begins electronic-structure calculations by determining the molecular or crystal point/space group and exploiting its group structure to block-diagonalize the secular equations — this is a direct industrial-scale descendant of the multiplication-table bookkeeping developed here, since a Hamiltonian matrix that commutes with every symmetry operation can be simplified enormously once the group's classes and irreducible representations are known. Vibrational spectroscopists use the group's multiplication table implicitly every time they apply the reduction formula to decide which normal modes are IR- or Raman-active in a newly synthesized coordination complex or metal-organic framework, a routine step in characterizing new catalysts and materials. In solid-state chemistry and crystallography, space groups (which include translations as well as point operations) are literally infinite groups built by the same closure/identity/inverse/associativity logic, just extended to include lattice translations — the classification of the 230 three-dimensional space groups used in every crystal structure solution (X-ray diffraction refinement software, protein crystallography, battery-material structure prediction) rests on exactly the group-theoretic skeleton developed here. In magnetic resonance and photophysics, selection rules for allowed electronic and vibrational transitions in transition-metal complexes (why some $d$–$d$ transitions are Laporte-forbidden, why certain IR bands vanish in highly symmetric metal carbonyls like $Cr(CO)_6$) are direct consequences of the class structure and orthogonality relations that only exist because the symmetry operations form a bona fide group. Even quantum computing error-correction codes borrow group-theoretic language (stabilizer groups) structurally identical to the molecular symmetry groups treated here.

Ask the tutor: How do you turn each symmetry operation into an explicit matrix once you know the multiplication table, and use those matrices to build a reducible representation for a real basis set (e.g., the three H 1s orbitals in NH₃)?

Picture a water molecule sitting in the $yz$-plane, oxygen at the origin, the two O–H bonds splayed symmetrically about the $z$-axis (the $C_2$ axis). Four rigid motions leave this geometric arrangement indistinguishable from where it started: doing nothing at all ($E$), rotating by 180° about the $z$-axis ($C_2$), reflecting through the molecular plane itself, $\sigma_v(xz)$... wait — convention matters here, so fix it precisely: let $\sigma_v(xz)$ be the plane containing the $C_2$ axis and perpendicular to the molecular plane, and $\sigma_v'(yz)$ be the actual molecular plane. Each of these four operations, applied to the three atoms, produces an arrangement of nuclei that exactly overlaps the original, atom for atom (accounting for the two equivalent hydrogens swapping). The crucial move is to stop thinking of these as four isolated instructions and start asking what happens when you do them one after another. If you rotate by $C_2$ and then reflect through $\sigma_v(xz)$, is the net effect the same as some single operation already in your list of four? For water, the answer is yes — always — and this is not a coincidence; it is forced by the fact that both $C_2$ and $\sigma_v$ are symmetries of the same rigid geometric object, so any sequence of them must also map the object onto itself, and there are only finitely many ways to do that (four, in this case). Working out which single operation results from a given sequence is exactly the content of a multiplication table. For ammonia, $NH_3$, belonging to point group $C_{3v}$, the operations are $E$, two rotations $C_3$ and $C_3^2$ (120° and 240° about the threefold axis), and three vertical mirror planes $\sigma_v$, $\sigma_v'$, $\sigma_v''$, each containing the $C_3$ axis and one N–H bond. Here the composition rule becomes genuinely interesting: rotating then reflecting is not the same as reflecting then rotating. This non-commutativity is not a defect — it is the signature of a "non-abelian" group, and it is precisely the feature responsible for ammonia having a doubly degenerate irreducible representation ($E$ symmetry, in the character-table sense foreshadowed at the end of this article), a phenomenon that never happens in the strictly abelian $C_{2v}$.

The four group axioms, stated for a set $G = \{E, A, B, C, \dots\}$ of symmetry operations of a fixed molecular geometry, with composition "$\circ$" meaning "apply the right-hand operation first, then the left-hand one":

where: $A$, $B$ are any two symmetry operations of the point group, and $AB$ denotes their composition (first $B$, then $A$); the statement says the result is also a symmetry operation already in the set — no new operations are ever generated.

where: $E$ is the "do nothing" operation; every point group contains it, and it commutes with everything by definition.

where: $A^{-1}$ is the operation that exactly undoes $A$; for a rotation $C_n^k$ the inverse is $C_n^{n-k}$, and every reflection and every $C_2$ is its own inverse.

where: grouping of a triple composition does not matter — this holds automatically because symmetry operations are functions (rigid motions) and function composition is always associative. Reading these together: a point group is completely specified once you list its elements and its multiplication table, because the table alone must satisfy closure, contain an identity row/column that reproduces the top row and left column, contain exactly one $E$ in every row and every column (a consequence of unique inverses), and be built from a composition rule that is automatically associative. This is why chemists can write down a $C_{3v}$ character table without re-deriving quantum mechanics every time: the table's internal arithmetic is guaranteed by the group axioms, and everything downstream (orthogonality of characters, reducibility formulas) inherits that guarantee. A second organizing relationship governs how operations sort into classes:

where: $X \sim A$ means $X$ is conjugate to $A$; $R$ ranges over every element of the group; a class is the set of all operations conjugate to a given one. Reading: two operations belong to the same class if some symmetry operation of the molecule can be used to relabel one into the other — geometrically, this means they are "the same kind of operation" viewed from equivalent directions (e.g., the two rotations $C_3$ and $C_3^2$ in $C_{3v}$ are conjugate to each other, and all three vertical mirrors are mutually conjugate). Classes, not individual operations, are what index the columns of a character table — this is the direct bridge to Character Tables Anatomy.

Example 1 (foundational) — verify closure by hand for a specific $C_{2v}$ triple product. Take $BeCl_2$ (linear, but treat its bent-molecule analogue $SO_2$, point group $C_{2v}$, for a genuine three-atom case). Using the matrices from the first derivation, compute $\sigma_v(xz)\cdot\big(C_2\cdot\sigma_v'(yz)\big)$ and check it matches a table entry. First, the inner product: $C_2\cdot\sigma_v'(yz)$ means apply $\sigma_v'(yz)$ first: $(x,y,z)\to(-x,y,z)$, then $C_2$: $(-x,y,z)\to(x,-y,z)$. Net map: $(x,y,z)\to(x,-y,z)$, which is $\sigma_v(xz)$. So $C_2\,\sigma_v'(yz) = \sigma_v(xz)$. Now the outer product: $\sigma_v(xz)\cdot \sigma_v(xz)$. Applying $\sigma_v(xz)$ twice: $(x,y,z)\to(x,-y,z)\to(x,y,z)$. This returns the identity:

This matches the table (every $C_{2v}$ operation is self-inverse) and confirms closure end-to-end for this specific triple: the full expression $\sigma_v(xz)\big(C_2\sigma_v'(yz)\big) = \sigma_v(xz)\sigma_v(xz) = E$, and $E$ is indeed one of the four allowed elements.

Example 2 (applied) — using non-commutativity to predict orbital degeneracy in $NH_3$. In $C_{3v}$ ammonia, the nitrogen $2p_x$ and $2p_y$ orbitals do not individually transform as one-dimensional representations, because $C_3$ mixes them into each other:

This is the same rotation matrix used in Derivation 2 — showing $p_x$ and $p_y$ mix under $C_3$, they cannot be assigned separately to one-dimensional irreducible representations; they must travel together as a 2-dimensional representation labeled $E$ (unrelated to the identity operation $E$; unfortunate but standard notation, always distinguished by context — italic $E$ for the operation, roman $E$ for the representation label). This is a direct, non-numerical consequence of the non-commuting structure derived above: an abelian group (like $C_{2v}$) can only have one-dimensional irreducible representations, because commuting operators can always be simultaneously diagonalized, while a non-abelian group is compelled to contain at least one representation of dimension $\geq 2$. Trace through the logic: the order of $C_{3v}$ is $h=6$; the sum of squares of the dimensions of irreps must equal $h$ (a fact proved via the group axioms in the theory of representations); with three classes there must be three irreps, and $1^2+1^2+2^2 = 1+1+4=6$ is the only combination that works, forcing exactly one doubly-degenerate irrep. Ammonia's degenerate N–H stretching and bending vibrations, and the double degeneracy of the relevant frontier orbitals, trace directly back to this group-theoretic necessity — not to any numerical coincidence in the Hamiltonian.

Example 3 (exam/competition level) — construct the multiplication table for $C_{3v}$ completely and locate a subgroup. List the six elements $\{E, C_3, C_3^2, \sigma_v, \sigma_v', \sigma_v''\}$ with $C_3$ a $120°$ rotation, $C_3^2$ a $240°$ rotation, and the three mirrors as in Derivation 2 ($\sigma_v$ through the $x$-axis, $\sigma_v'$ at $-60°$, i.e. $120°$, from $x$, and $\sigma_v''$ at $+60°$ from $x$). Using the matrix method of Derivation 2 systematically (multiply every pair of the six $2\times2$ matrices):

Read off the row for $\sigma_v$: $\sigma_v \cdot C_3 = \sigma_v'$, matching Step 3 of Derivation 2 exactly (recall $\sigma_v C_3 = \sigma_v'$ was computed there). Read off the row for $C_3$, column $\sigma_v$: $C_3\cdot\sigma_v = \sigma_v''$, matching Step 2 of Derivation 2 ($C_3\sigma_v = \sigma_v''$). The table is internally consistent with the explicit matrix computations. Now find a subgroup: the top-left $3\times3$ block, $\{E, C_3, C_3^2\}$, is closed under itself (every product of two rotations is a rotation, as seen directly in row/column intersections restricted to those three labels) — it is a subgroup of order 3, isomorphic to the cyclic group $C_3$. Lagrange's theorem (a consequence of the group axioms, not proven here but consistent with them) requires subgroup order to divide group order: $3 \mid 6$ ✓, and no subgroup of order 4 or 5 can exist since neither divides 6 — a fact the explicit table can be scanned to confirm has no such closed sub-block.

The systematic use of group theory in chemistry is usually credited less to a single "discoverer" and more to a convergence: crystallographers and mathematicians developed point-group and space-group classification in the 19th century for crystal morphology, while the application to molecular spectroscopy and bonding was driven in the 20th century largely by Hans Bethe (crystal-field splitting via group theory, 1929) and later systematized for chemists by F. Albert Cotton, whose textbook Chemical Applications of Group Theory (1963) became the standard route by which generations of chemists learned to build multiplication tables and character tables as working tools rather than abstract mathematics.