Reducible and Irreducible Representations

Vibrational spectroscopy remains the most direct application: infrared and Raman selection rules for every molecule — from simple diatomics to metal carbonyl clusters used to fingerprint CO-binding geometry — are read off by decomposing the $3N$ Cartesian representation into irreps and checking which ones share a symmetry label with $x,y,z$ (IR-active) or with the quadratic functions $x^2, xy$, etc. (Raman-active). This is routine in characterizing catalytic intermediates by IR, where the number and pattern of carbonyl stretching bands (e.g., $Fe(CO)_5$ showing two IR-active CO stretches, matching $A_2'' + E'$ symmetry in $D_{3h}$) directly diagnoses molecular geometry without crystallography. Ligand field and molecular orbital theory in coordination chemistry lean on the same machinery: decomposing ligand $\sigma$- and $\pi$-orbital sets into irreps ($A_{1g}+E_g+T_{1u}$ for octahedral $\sigma$, additional $T_{1g}+T_{2g}+T_{1u}+T_{2u}$ contributions for $\pi$-donor or $\pi$-acceptor ligand sets) explains why $\pi$-donor ligands lower $\Delta_o$ and $\pi$-acceptor ligands raise it, purely from which metal d-orbital irrep ($T_{2g}$) gains a symmetry-allowed partner. This underlies the spectrochemical series and the electronic-structure rationale behind catalysts and pigments alike. Crystal engineering and solid-state chemistry use the same reduction procedure for phonon modes in space groups (a generalization to infinite/periodic groups), predicting which lattice vibrations are Raman- or IR-active in materials characterization — essential in identifying polymorphs of pharmaceutical solids and in diagnosing defect structures in oxide catalysts and battery electrode materials via vibrational spectroscopy. Computational quantum chemistry packages exploit block-diagonalization directly for efficiency: symmetry-adapted linear combinations (SALCs), built exactly from the irreducible decomposition described here, are used to block-diagonalize the molecular Hamiltonian matrix before diagonalization, cutting a possibly enormous matrix diagonalization problem into much smaller independent blocks — one per irrep — which is why quantum chemistry software asks for (or automatically detects) point-group symmetry before an SCF or CI calculation, turning group theory into a genuine computational speedup rather than a purely qualitative tool. This same idea, formalized as the reduction formula built on the orthogonality theorem, is the natural next tool once reducible/irreducible representations themselves are understood.

Ask the tutor: How do I use the reduction formula itself, step by step, to pull the multiplicities $n_i$ out of a reducible representation's characters without guessing?

Consider a representation $\Gamma$ of a point group $G$: a rule assigning to each symmetry operation $g \in G$ a square matrix $\Gamma(g)$, acting on some chosen basis (three C–H bond vectors, five d-orbitals, nine Cartesian displacement coordinates — anything you like), such that the matrices multiply the same way the operations compose: $\Gamma(g_1)\Gamma(g_2) = \Gamma(g_1 g_2)$. This is Matrix Representations of Symmetry Operations, assumed known here. Now suppose that within the chosen basis there exists a smaller subset of basis vectors — a subspace — that maps into itself under every single operation of the group. That is, applying any $\Gamma(g)$ to a vector in this subspace produces another vector still inside the subspace; the group never kicks you out of it. Such a subspace is called invariant. If a proper (nontrivial, not the whole space) invariant subspace exists, the representation is reducible: you can choose a new basis — first spanning the invariant subspace, then completing it — in which every single matrix $\Gamma(g)$ simultaneously takes the block-diagonal form

for every $g$, with the same fixed matrix $S$ working for all operations at once. If no such invariant subspace exists (other than the trivial $\{0\}$ and the whole space), the representation is irreducible — there is no way to shrink it further; it is a genuine building block. This is exactly what happens physically. Take the two O–H bond vectors of water ($C_{2v}$ symmetry). Under $E$, $C_2$, $\sigma_v$, $\sigma_v'$, the two individual bond vectors get swapped and mixed — the naive "bond basis" representation is reducible. But the symmetric combination $r_1+r_2$ transforms into itself (times $\pm1$) under every operation, and so does the antisymmetric combination $r_1-r_2$. These two combinations are the block-diagonalizing basis: in this new basis, every $2\times2$ bond matrix collapses into two separate $1\times1$ numbers, one following the totally symmetric $A_1$ irrep, the other following $B_2$. Nothing approximate has happened — this is an exact change of coordinates that was always available, hiding inside the original bond-vector description.

Notice the check built into the figure: the trace of the original $2\times2$ block under $C_2$ is $0+0=0$ (pure off-diagonal swap matrix), and the trace of the block-diagonal form is $(+1)+(-1)=0$. These must agree — traces are basis-independent — and they do. This is the character-additivity property that makes the whole machinery usable without ever explicitly constructing $S$. The same idea scales up enormously. In an octahedral $ML_6$ complex, the six ligand $\sigma$-orbitals generate a six-dimensional reducible representation, $\Gamma_\sigma$. It decomposes as $A_{1g} + E_g + T_{1u}$ (dimensions $1+2+3=6$, matching). The five metal d-orbitals generate a five-dimensional representation that is already irreducible-adjacent: it splits as $E_g + T_{2g}$ ($2+3=5$). Symmetry alone, before any energetics, tells you the d-orbitals split into two sets and that only the $E_g$ metal d-orbitals share a symmetry label with (and hence can bond to) the ligand $\sigma$ combinations — the $T_{2g}$ set is symmetry-forbidden from $\sigma$-bonding and remains nonbonding (or engages only in $\pi$-bonding with ligand $\pi$ orbitals of $T_{2g}$ symmetry). The reducible/irreducible split is the qualitative molecular orbital diagram, extracted purely from geometry.

The formal statement, sometimes called the complete reducibility theorem (a consequence of Maschke's theorem for finite groups over the complex or real numbers, with unitary representations), is:

where:

• $\Gamma$ is any (finite-dimensional) representation of the group,
• $\Gamma_1,\ldots,\Gamma_k$ are all the group's inequivalent irreducible representations (a finite, fixed list — the rows of the character table),
• $n_i \ge 0$ is an integer, the multiplicity — how many times $\Gamma_i$ occurs inside $\Gamma$,
• $\oplus$ denotes direct sum: after an appropriate similarity transformation, every matrix $\Gamma(g)$ becomes block-diagonal with $n_1$ copies of $\Gamma_1(g)$-sized blocks, then $n_2$ copies of $\Gamma_2(g)$-sized blocks, and so on.

Reading it: any representation, however large or ugly, is entirely built from these standard irreducible pieces, each appearing some whole number of times, and once you know the multiplicities $n_i$ you know the representation completely (up to choice of basis). Because similarity transformations leave traces unchanged ($\mathrm{tr}(S^{-1}MS) = \mathrm{tr}(M)$ for any invertible $S$), and the trace of a block-diagonal matrix is the sum of the traces of its blocks, character additivity follows immediately:

where:

• $\chi(g)$ is the character (trace) of the reducible representation at operation $g$,
• $\chi_i(g)$ is the tabulated character of irrep $\Gamma_i$ at operation $g$ (found directly in the character table),
• $n_i$ are the same non-negative integer multiplicities as above, now to be determined from the characters alone — you never need to construct the matrices or the transformation $S$ explicitly.

Reading it: to decompose any reducible representation, all you actually need is its character (one number per class of operations), not its full matrix representation. This is what makes group theory tractable in practice; it is also the seed of the reduction formula introduced separately. A second core relationship pins down why the list of irreps is finite and fixed, via the dimensionality theorem:

where:

• $\ell_i = \chi_i(E)$ is the dimension of irrep $\Gamma_i$ (its character under the identity operation),
• $|G|$ is the order of the group (total number of symmetry operations),
• the sum runs over all inequivalent irreducible representations of the group.

Reading it: the "sizes" of the group's irreducible building blocks are not arbitrary — their squares must exactly tile the order of the group, which is why $C_{2v}$ (order 4) has four 1-dimensional irreps ($1^2\times4=4$) while $T_d$ (order 24) needs $1^2+1^2+2^2+3^2+3^2=24$, giving two 1-D, one 2-D, and two 3-D irreps ($A_1, A_2, E, T_1, T_2$).

Example 1 (foundational): the two O–H stretches of water under $C_{2v}$, done by hand. $C_{2v}$ has operations $\{E, C_2, \sigma_v(xz), \sigma_v'(yz)\}$ and four 1-D irreps: $A_1(1,1,1,1)$, $A_2(1,1,-1,-1)$, $B_1(1,-1,1,-1)$, $B_2(1,-1,-1,1)$ (using the convention that $\sigma_v$ is the molecular plane). Take the basis $\{r_1, r_2\}$, the two O–H bond-stretch coordinates. Under $E$, both bonds map to themselves: matrix is the $2\times2$ identity, trace $\chi(E) = 2$. Under $C_2$, the two H atoms swap, so $r_1 \leftrightarrow r_2$: the matrix is the swap matrix with zero diagonal, trace $\chi(C_2)=0$. Under $\sigma_v(xz)$ (the molecular plane containing both H's), both bonds map to themselves (this mirror plane doesn't move the H atoms since they lie in it), trace $\chi(\sigma_v) = 2$. Under $\sigma_v'(yz)$ (perpendicular plane, swapping the two H's), trace $\chi(\sigma_v')=0$. So $\Gamma_{\text{stretch}}$: $\chi(E,C_2,\sigma_v,\sigma_v') = (2,0,2,0)$. Reduce by inspection (this is literally $\chi = n_{A_1}\chi_{A_1} + n_{B_2}\chi_{B_2}$, checking against the additivity formula):

Matching the $E$ entry: $n_{A_1} + n_{B_2} = 2$. Matching the $C_2$ entry: $n_{A_1} - n_{B_2} = 0$, so $n_{A_1}=n_{B_2}=1$. Check $\sigma_v$: $1(1) + 1(-1) = 0 \ne 2$ — this fails, so try instead $n_{A_1}=1, n_{B_1}=0, n_{B_2}=1$ matched against $B_2$'s actual character $(1,-1,-1,1)$ evaluated at $\sigma_v$ gives $-1$, and $A_1$ gives $+1$, summing to $0$, still not $2$. The resolution: the correct combination is $A_1 + B_1$ ($B_1$ has character $(1,-1,1,-1)$, symmetric under $\sigma_v$, antisymmetric under $\sigma_v'$ — the assignment depends on whether $\sigma_v(xz)$ is chosen as the molecular plane or the perpendicular plane; with $\sigma_v(xz)$ as the H₂O plane, the totally symmetric stretch combination and its partner transform as $A_1 + B_2$ under the standard Mulliken convention used in most texts, where $B_2$ is symmetric under $\sigma_v(xz)$, antisymmetric under $C_2$ and $\sigma_v'(yz)$). Using that standard convention consistently: $\chi_{B_2} = (1,-1,1,-1)$ under $(E,C_2,\sigma_v,\sigma_v')$. Then $n_{A_1}(1,1,1,1) + n_{B_2}(1,-1,1,-1) = (n_{A_1}+n_{B_2},\, n_{A_1}-n_{B_2},\, n_{A_1}+n_{B_2},\, n_{A_1}-n_{B_2})$. Setting equal to $(2,0,2,0)$ gives $n_{A_1}+n_{B_2}=2$ and $n_{A_1}-n_{B_2}=0$, so $n_{A_1}=n_{B_2}=1$, and every entry now matches: $(2,0,2,0)$ exactly.

Dimension check: $1+1=2$, matching $\chi(E)=2$. Physically: $r_1+r_2$ (symmetric stretch, $A_1$, IR- and Raman-active) and $r_1-r_2$ (antisymmetric stretch, $B_2$, also IR-active) — exactly water's two observed O–H stretching bands ($\sim3657$ cm⁻¹ symmetric, $\sim3756$ cm⁻¹ antisymmetric), with no bending mode included here since only stretch coordinates were chosen as the basis.

Example 2 (applied): the σ-framework of a square planar $\text{PtCl}_4^{2-}$ ($D_{4h}$). Four ligand σ-orbitals, one per Cl, sit at the corners of a square around Pt. Build $\Gamma_\sigma$ by counting how many ligand orbitals are left unmoved (with weight $+1$, since a $\sigma$-orbital pointing at the metal transforms into itself under operations that fix its ligand in place) by each class of operation in $D_{4h} = \{E, 2C_4, C_2, 2C_2', 2C_2'', i, 2S_4, \sigma_h, 2\sigma_v, 2\sigma_d\}$.

• $E$: all 4 ligands fixed → $\chi=4$.
• $2C_4$ (90° rotation): no ligand stays in place (each moves to the next corner) → $\chi=0$.
• $C_2$ (180°, same axis as $C_4$): opposite ligands swap, none fixed → $\chi=0$.
• $2C_2'$ (through two ligands): those two ligands fixed, other two swap → $\chi=2$.
• $2C_2''$ (through edge midpoints, between ligands): no ligand fixed → $\chi=0$.
• $i$: all ligands move to opposite corners, none fixed → $\chi=0$.
• $2S_4$: none fixed → $\chi=0$.
• $\sigma_h$ (molecular plane, contains all 4 ligands): all 4 fixed → $\chi=4$.
• $2\sigma_v$ (containing 2 ligands each, and the $C_4$ axis, perpendicular to molecular plane... here taken through opposite ligand pairs): 2 fixed → $\chi=2$.
• $2\sigma_d$ (through edge midpoints): 0 fixed → $\chi=0$.

So $\Gamma_\sigma = (4,0,0,2,0,0,0,4,2,0)$ over $(E,2C_4,C_2,2C_2',2C_2'',i,2S_4,\sigma_h,2\sigma_v,2\sigma_d)$. Comparing against the $D_{4h}$ character table (order $|G|=16$), this reduces to

Dimension check: $1+1+2=4$ ✓. This is the textbook result for square-planar σ-bonding: the metal's $d_{x^2-y^2}$ orbital (which transforms as $B_{1g}$) is the only d-orbital that can σ-bond with the ligand set, exactly matching why $d_{x^2-y^2}$ is pushed to famously high energy in square-planar splitting diagrams, while $d_{z^2}(A_{1g})$ overlaps with the totally symmetric $A_{1g}$ ligand combination only weakly (mostly through the ligand's radial/axial character), and $d_{xy}, d_{xz}, d_{yz}$ ($B_{2g}, E_g$) have no $\sigma$-symmetry match at all and remain purely $\pi$-active or nonbonding.

Example 3 (exam/competition level): decomposing the 15-dimensional Cartesian representation of $NH_3$ ($C_{3v}$) and extracting the vibrational modes. Ammonia, $C_{3v}$, order $6$: operations $\{E, 2C_3, 3\sigma_v\}$; irreps $A_1(1,1,1)$, $A_2(1,1,-1)$, $E(2,-1,0)$. Assign 3 Cartesian axes ($x,y,z$) to each of the 4 atoms (N + 3 H), giving a 12-dimensional... but here we deliberately use the full atom count including all 4 atoms to build $\Gamma_{3N}$, $N=4$, dimension $3N=12$. Character of $\Gamma_{3N}$ at each operation is (number of atoms unmoved) × (contribution per unmoved atom, which is $1+2\cos\theta$ for proper rotation by angle $\theta$, or $-1+2\cos\theta$ for improper $S_n$/reflection, with $\sigma = S_1$ giving contribution $+1$):

• $E$: all 4 atoms unmoved, per-atom contribution $1+2\cos(0°) = 3$. $\chi(E) = 4 \times 3 = 12$.
• $2C_3$ (120°): only N unmoved (the 3 H's cycle), per-atom contribution $1+2\cos(120°) = 1 + 2(-0.5) = 0$. $\chi(C_3) = 1 \times 0 = 0$.
• $3\sigv$: N and one H (the one lying in that particular plane) unmoved — 2 atoms unmoved, per-atom reflection contribution $= +1$ (reflection is $-1+2\cos(0°)=1$). $\chi(\sigma_v) = 2 \times 1 = 2$.

So $\Gamma_{3N} = (12, 0, 2)$ over $(E, 2C_3, 3\sigma_v)$. Reduce using $n_i = \frac{1}{|G|}\sum_{\text{classes}} (\text{class size})\,\chi(g)\,\chi_i(g)$ (the reduction formula, used here as a tool though derived fully elsewhere):

So $\Gamma_{3N} = 3A_1 \oplus 1A_2 \oplus 4E$. Dimension check: $3(1)+1(1)+4(2) = 3+1+8=12$ ✓. Now remove translations ($T_x,T_y,T_z$ transform as $A_1 + E$ in $C_{3v}$) and rotations ($R_x,R_y,R_z$ transform as $A_2+E$):

Dimension check: $2(1)+2(2) = 2+4=6 = 3N-6 = 3(4)-6$ ✓, exactly ammonia's six vibrational degrees of freedom: the symmetric stretch and umbrella bend (both $A_1$, each nondegenerate) plus the doubly degenerate asymmetric stretch and doubly degenerate bend (each $E$, counted once as a 2-D irrep but contributing 2 to the mode count) — matching the four experimentally observed IR bands of NH₃ (ν₁, ν₂ nondegenerate; ν₃, ν₄ doubly degenerate), a classic worked case in vibrational spectroscopy.

The representation-theoretic backbone used throughout chemical symmetry analysis was not invented by chemists at all. Ferdinand Georg Frobenius developed the theory of group characters and representations in the 1890s, building on work connecting group theory to matrix algebra; Issai Schur, his student, supplied the lemma that makes "irreducible" a precise, checkable algebraic condition rather than an intuitive notion. Their mathematics sat largely dormant for chemistry until the 1930s, when spectroscopists needed a rigorous way to predict which vibrational and electronic transitions were allowed by molecular symmetry. The chemist most responsible for translating this apparatus into the character tables used in every inorganic chemistry course today is Robert S. Mulliken, whose 1933 notation for irreducible representations ($A_1, A_2, B_1, B_2, E, T_1, T_2$, and their gerade/ungerade subscripts) became the universal standard. Mulliken, trained as a physical chemist and later a Nobel laureate (1966, for molecular orbital theory), was uniquely positioned at the interface of the new quantum mechanics and the older, more empirical spectroscopy of molecular vibrations and electronic bands — precisely the discipline that needed group representation theory translated into laboratory-usable labels rather than abstract algebra.