Wigner–Eckart Theorem

Spectroscopy confronts an intimidating bookkeeping problem. A rank- $k$ tensor operator $T^{(k)}_q$ — the dipole $\mathbf r$ , the quadrupole moment, a magnetic multipole — has $2k+1$ components and connects a level $j$ (with $2j+1$ orientations) to a level $j'$ (with $2j'+1$ ), so naively it hides $(2k+1)(2j+1)(2j'+1)$ independent matrix elements $\langle \alpha' j' m'|T^{(k)}_q|\alpha j m\rangle$ , one per way of pointing the states and the operator. The Wigner–Eckart theorem asserts this whole table collapses to a single orientation-independent number. All dependence on the magnetic quantum numbers $m,m',q$ is carried by a Clebsch–Gordan coefficient — a pure geometric object from a group-theory table — times one scalar, the reduced matrix element $\langle \alpha' j'\|T^{(k)}\|\alpha j\rangle$ . The reason is structural, not dynamical. The state $T^{(k)}_q|\alpha j m\rangle$ transforms exactly as a vector in the product representation $k\otimes j$ ; the Clebsch–Gordan series decomposes that product into irreducibles $j'$ with $|j-k|\le j'\le j+k$ , and overlapping with $\langle\alpha' j' m'|$ projects onto one. Schur's lemma then forces the overlap of two carriers of the same irrep to be a constant times the identity — so the geometry (which $j'$ , which $m'$ ) is fixed by the rotation group once and for all, and the dynamics survive only in that one constant. It is the cleanest example in physics of symmetry separating kinematics from dynamics, and it generalizes verbatim to any compact group — isospin, flavor $SU(3)$ , and crystal point groups all wear the same clothing.

Key takeaways

The matrix element of a spherical tensor operator factors as $\langle \alpha' j' m'|T^{(k)}_q|\alpha j m\rangle = \langle j\,m;k\,q|j'm'\rangle\,\langle\alpha' j'\|T^{(k)}\|\alpha j\rangle$ — a Clebsch–Gordan coefficient times a reduced matrix element.
The Clebsch–Gordan coefficient carries all the $m,m',q$ dependence and is fixed by the group $SO(3)$ / $SU(2)$ ; the reduced element holds the orientation-independent physics.
Selection rules fall out for free: the element vanishes unless $|j-k|\le j'\le j+k$ (triangle rule) and $m'=m+q$ ( $z$ -projection conservation).
One measured matrix element fixes the reduced element and hence every other component of the multiplet — line-intensity ratios need no dynamical calculation.
Special cases: the scalar ( $k=0$ ) is diagonal in $j,m$ ; the vector ( $k=1$ ) gives the projection theorem $\mathbf V\to g_V\mathbf J$ and the Landé $g$ -factor.
It generalizes to any compact group, with one reduced element per copy of $j'$ in $k\otimes j$ — the backbone of isospin and crystal-field selection rules.
Key idea: $T^{(k)}_q|\alpha j m\rangle$ lives in $k\otimes j$ ; overlapping with $\langle\alpha' j' m'|$ projects onto irrep $j'$ , and Schur's lemma makes that projection one number (the reduced matrix element) times a Clebsch–Gordan coefficient — geometry and dynamics separate cleanly.

The core insight

$T^{(k)}_q|\alpha j m\rangle$ transforms as a vector in $k\otimes j$ ; overlapping it with $\langle\alpha' j' m'|$ projects onto the irrep $j'$ , and Schur's lemma forces that projection to be a single reduced matrix element times a Clebsch–Gordan coefficient, cleanly separating group-fixed geometry from Hamiltonian-fixed dynamics.

Where this lives today

The theorem runs inside modern precision physics every day. In optical atomic clocks, the differential AC-Stark shifts and Zeeman sensitivities limiting accuracy follow from a handful of reduced elements cross-checked through Wigner–Eckart rather than thousands of measurements. In nuclear structure, the reduced transition probabilities $B(Ek)$ and $B(Mk)$ tabulated for every measured $\gamma$ -ray are exactly the squared reduced matrix elements the theorem isolates. Particle physics uses the $SU(2)$ isospin and $SU(3)$ flavor versions for cross-section ratios: the factor-of-nine ratio between $\pi^+p$ and $\pi^- p$ scattering through the $\Delta(1232)$ resonance is a pure isospin Clebsch–Gordan result, and the Gell-Mann–Okubo mass relations are Wigner–Eckart applied to $SU(3)$ . Trapped-ion quantum-information experiments pick "clock" and "cycling" transitions by reading selection rules straight off the theorem, and crystal-field spectroscopy of rare-earth and transition-metal ions relies on its finite-group analogue.

The physical picture

Picture a spherical tensor operator $T^{(k)}_q$ as a machine with $2k+1$ teeth that meshes with the angular-momentum gearbox. When it acts on $|j m\rangle$ it couples in an amount of angular momentum $k$ , exactly as if you had added a fictitious particle of spin $k$ . The result cannot be a single sharp $j'$ ; it is a superposition spread over the Clebsch–Gordan series $k\otimes j = |j-k|\oplus\cdots\oplus(j+k)$ . Overlapping with a definite bra $\langle\alpha' j' m'|$ then asks how much of that superposition points along irrep $j'$ — and the answer is zero unless $j'$ appears in the series. That single observation is the entire content of the selection rules.

Figure

A rank-1 operator acting on a j = 1 state behaves like coupling in a spin-1 object: the product 1 ⊗ 1 decomposes into irreps j′ = 0, 1, 2 obeying the triangle rule where j′ runs from j − k up to j + k. Overlapping with a final bra of definite j′ projects onto exactly one irrep, and Schur's lemma makes that projection a single number times a Clebsch–Gordan coefficient

Rotational invariance makes the collapse total. Physics cannot depend on the absolute orientation of your laboratory, so once you know one component between one pair of oriented states, rotating the apparatus reproduces every other component — no new physics in a rotation, only geometry. The Clebsch–Gordan table encodes that geometry universally; whatever is left after you strip it away is the reduced matrix element, the dynamical residue that knows about radial wavefunctions and interaction strength. The most familiar place this residue is measured is the Zeeman effect, where the projection theorem — the $k=1$ case — fixes every magnetic sublevel spacing of an entire multiplet from one number per level, the Landé $g$ -factor.

Figure

Weak-field anomalous Zeeman splitting of the sodium D lines: each level fans into 2j + 1 equally spaced sublevels with spacing g μ_B B, where the Landé factors are g = 2 for the ²S₁/₂ ground state, g = 2/3 for ²P₁/₂, and g = 4/3 for ²P₃/₂ — every spacing fixed by the projection theorem from one formula

The core equations

The theorem is one line, best read as a factorization. On the left, a matrix element depending a priori on five labels; on the right, everything geometric ( $m,m',q$ ) quarantined in a Clebsch–Gordan coefficient and everything dynamical in the reduced element. The two multiply, never mix.

\langle \alpha' j' m'|\,T^{(k)}_q\,|\alpha j m\rangle = \langle j\,m;\,k\,q\,|\,j'\,m'\rangle\;\langle \alpha' j'\|T^{(k)}\|\alpha j\rangle \qquad (1)

where: $T^{(k)}_q$ — the $q$ -th component of a rank- $k$ spherical tensor operator, $q=-k,\dots,k$ ; $|\alpha j m\rangle$ — a state of total angular momentum $j$ and projection $m$ , with $\alpha$ collecting all other (dynamical) quantum numbers; $\langle j m;k q|j' m'\rangle$ — a Clebsch–Gordan coefficient coupling $j\otimes k\to j'$ ; $\langle \alpha' j'\|T^{(k)}\|\alpha j\rangle$ — the reduced matrix element, independent of $m,m',q$ (the double bars are notation, not a norm). Reading it: fix the two levels and the rank, and one number times a table entry gives all $(2k+1)(2j+1)(2j'+1)$ matrix elements. A spherical tensor operator is defined by how it transforms; the practical test is a pair of commutators with the generators, saying the components mix under rotation exactly as the kets $|k q\rangle$ do.

[J_z, T^{(k)}_q] = \hbar q\,T^{(k)}_q,\qquad [J_\pm, T^{(k)}_q] = \hbar\sqrt{k(k+1)-q(q\pm1)}\;T^{(k)}_{q\pm1} \qquad (2)

where: $J_z,J_\pm$ — angular-momentum operators, $J_\pm=J_x\pm iJ_y$ ; $\hbar$ — reduced Planck constant. Reading it: the first commutator says $T^{(k)}_q$ raises the $J_z$ eigenvalue by $\hbar q$ (hence $m'=m+q$ ); the second shuttles you along the $2k+1$ components with the standard ladder coefficients — the fingerprint of rank $k$ . For tabulation one usually prefers the Wigner $3j$ symbol, a symmetrized, rescaled Clebsch–Gordan coefficient. It is symmetric under column permutations (up to a phase), exposing symmetries the CG form hides — at the cost of an extra $\sqrt{2j'+1}$ that migrates into the reduced element and causes endless convention disputes.

\langle \alpha' j' m'|\,T^{(k)}_q\,|\alpha j m\rangle = (-1)^{\,j'-m'}\begin{pmatrix} j' & k & j \\ -m' & q & m \end{pmatrix}\langle \alpha' j'\|T^{(k)}\|\alpha j\rangle \qquad (3)

where: $\begin{pmatrix} j' & k & j \\ -m' & q & m \end{pmatrix}$ — the Wigner $3j$ symbol, nonzero only when $-m'+q+m=0$ and the triangle rule holds; $(-1)^{j'-m'}$ — the phase relating it to the CG form. Reading it: because the $3j$ symbol absorbs a $\sqrt{2j'+1}$ , the reduced element in (3) differs from that in (1) by exactly that factor — always state which convention you mean. The theorem's most useful corollary is the selection rules: a matrix element is dead on arrival unless the Clebsch–Gordan coefficient is nonzero.

\langle \alpha' j' m'|T^{(k)}_q|\alpha j m\rangle = 0\quad\text{unless}\quad |j-k|\le j'\le j+k\ \ \text{and}\ \ m'=m+q \qquad (4)

where: $|j-k|\le j'\le j+k$ — the triangle inequality from the CG decomposition of $k\otimes j$ ; $m'=m+q$ — conservation of the $z$ -projection. Reading it: for the electric dipole $k=1$ this is $\Delta j=0,\pm1$ (with $0\to0$ forbidden, since $j'=j=0$ fails the triangle) and $\Delta m=0,\pm1$ . Within a single multiplet ( $j'=j$ ), the $k=1$ case becomes the projection theorem: every vector operator is proportional to $\mathbf J$ , the constant being a ratio of reduced elements written through the scalar $\mathbf J\cdot\mathbf V$ .

\langle \alpha j m'|V_q|\alpha j m\rangle = g_V\,\langle \alpha j m'|J_q|\alpha j m\rangle,\qquad g_V=\frac{\langle \mathbf{J}\cdot\mathbf{V}\rangle_{\alpha j}}{\hbar^2\,j(j+1)} \qquad (5)

where: $V_q$ — spherical components of a vector operator $\mathbf V$ ; $\mathbf J\cdot\mathbf V$ — the invariant scalar, whose expectation is $m$ -independent; $g_V$ — the projection constant. Reading it: inside a fixed- $j$ level you may replace $\mathbf V$ wholesale by $g_V\mathbf J$ — only the component of $\mathbf V$ along $\mathbf J$ survives. Applied to the magnetic moment $\boldsymbol\mu\propto\mathbf L+2\mathbf S=\mathbf J+\mathbf S$ , equation (5) delivers the Landé $g$ -factor of the anomalous Zeeman effect.

g_J = 1 + \frac{j(j+1)+s(s+1)-\ell(\ell+1)}{2\,j(j+1)} \qquad (6)

where: $g_J$ — Landé factor; $\ell,s,j$ — orbital, spin, and total angular-momentum quantum numbers. Reading it: this is (5) with $\mathbf V=\mathbf J+\mathbf S$ ; it gives $g_J=2$ for pure spin ( $\ell=0$ ) and $g_J=1$ for pure orbit ( $s=0$ ), the anchors every atomic-physics table must reproduce.

Derivation AProof of the theorem via Schur's lemma

The goal is to prove the matrix element must factor into a Clebsch–Gordan coefficient and an $m$ -independent constant, using nothing but how the objects transform under rotations. The strategy: show that $T^{(k)}_q|\alpha j m\rangle$ is a vector in the product representation $k\otimes j$ , couple it to states of definite total angular momentum, and invoke Schur's lemma — which controls all maps between irreducible representations — to collapse the overlap with the bra to a single number times a geometric coefficient. No dynamics enters; the whole proof is representation theory.

Step 1. Sandwich a rotation: since $U(R)^\dagger U(R)=1$ , $U(R)\,T^{(k)}_q|\alpha j m\rangle = \big(U T^{(k)}_q U^\dagger\big)\big(U|\alpha j m\rangle\big) = \sum_{q',m''} D^{(k)}_{q'q}(R)\,D^{(j)}_{m''m}(R)\,T^{(k)}_{q'}|\alpha j m''\rangle$ . Why: the operator carries $D^{(k)}$ and the ket carries $D^{(j)}$ , so the product transforms under $D^{(k)}\otimes D^{(j)}$ — literally the vector space $k\otimes j$ .

Step 2. Couple to good total angular momentum: $|(kj)JM\rangle \equiv \sum_{q,m}\langle k q;j m|J M\rangle\,T^{(k)}_q|\alpha j m\rangle$ . Why: the Clebsch–Gordan series decomposes $k\otimes j$ into irreducibles $J=|j-k|,\dots,j+k$ , and these combinations transform as a standard irrep $J$ — the pieces Schur's lemma can bite.

Step 3. Overlap with the bra $\langle\alpha' j' m'|$ , itself a carrier of irrep $j'$ : $\langle\alpha' j' m'|(kj)JM\rangle \propto \delta_{j'J}\,\delta_{m'M}$ . Why: this is an intertwiner between irreps $j'$ and $J$ ; Schur's lemma makes it vanish unless $j'=J$ , and then a constant multiple of the identity — independent of $M=m'$ .

Step 4. Name that constant: $\langle\alpha' j' m'|(kj)j' m'\rangle \equiv \langle\alpha' j'\|T^{(k)}\|\alpha j\rangle$ . Why: the Schur constant depends on the irrep labels $j',j,k$ and dynamical labels $\alpha',\alpha$ but not on $m'$ — this single number is the whole dynamical content.

Step 5. Invert the coupling with CG orthogonality, $T^{(k)}_q|\alpha j m\rangle=\sum_{JM}\langle j m;k q|JM\rangle\,|(kj)JM\rangle$ , overlap with $\langle\alpha' j' m'|$ , and keep the only surviving term $J=j',M=m'$ . Why: this returns exactly the theorem, $\langle\alpha' j' m'|T^{(k)}_q|\alpha j m\rangle=\langle j m;k q|j' m'\rangle\langle\alpha' j'\|T^{(k)}\|\alpha j\rangle$ — CG coefficient from geometry, constant from dynamics. Verified: at $k=0$ the CG coefficient is $\delta_{j'j}\delta_{m'm}/\sqrt{2j+1}$ , forcing a scalar to be diagonal.

Go deeper · Derivation B · 1 minThe projection theorem for vector operators

Derivation BThe projection theorem for vector operators

The goal is to show that inside a single $j$ -multiplet any vector operator $\mathbf V$ acts as if it were simply $g_V\mathbf J$ , and to pin the constant $g_V$ to the measurable scalar $\langle\mathbf J\cdot\mathbf V\rangle$ . The route is to apply the just-proven theorem twice — once to $\mathbf V$ and once to $\mathbf J$ , both rank-1 operators with the same initial and final $j$ — so their common Clebsch–Gordan coefficient cancels in a ratio, then to fix that ratio by contracting with $\mathbf J$ . This is the workhorse behind the Landé $g$ -factor and the anomalous Zeeman effect.

Step 1. Apply the theorem to both operators within $j'=j$ : $\langle\alpha j m'|V_q|\alpha j m\rangle=\langle j m;1 q|j m'\rangle\langle\alpha j\|V\|\alpha j\rangle$ and identically for $J_q$ . Why: $V_q$ and $J_q$ share the same rank ( $k=1$ ) and levels ( $j\to j$ ), hence the identical CG coefficient — only the reduced element can differ.

Step 2. Form the ratio component-by-component: $\dfrac{\langle\alpha j m'|V_q|\alpha j m\rangle}{\langle\alpha j m'|J_q|\alpha j m\rangle}=\dfrac{\langle\alpha j\|V\|\alpha j\rangle}{\langle\alpha j\|J\|\alpha j\rangle}\equiv g_V$ . Why: the shared CG coefficient cancels, leaving a single $m$ -independent number — $\mathbf V$ and $\mathbf J$ are proportional on the multiplet, the projection theorem in bare form.

Step 3. Fix $g_V$ by contracting with $\mathbf J$ : $\langle\mathbf J\cdot\mathbf V\rangle=\sum_q(-1)^q\langle J_{-q}V_q\rangle=g_V\langle\mathbf J\cdot\mathbf J\rangle=g_V\,\hbar^2 j(j+1)$ . Why: the invariant $\mathbf J\cdot\mathbf V$ has an $m$ -independent expectation; replacing $V_q\to g_V J_q$ and using $J^2=\hbar^2 j(j+1)$ isolates $g_V$ .

Step 4. Solve: $g_V=\dfrac{\langle\mathbf J\cdot\mathbf V\rangle}{\hbar^2 j(j+1)}$ , giving equation (5). Why: as a check, $\mathbf V=\mathbf J$ gives $g_V=1$ , and $\mathbf V=\mathbf J+\mathbf S$ gives the Landé factor derived next.

Go deeper · Derivation C · 1 minThe Landé g-factor from the projection theorem

Derivation CThe Landé g-factor from the projection theorem

The goal is to derive equation (6) — the Landé $g$ -factor that sets every anomalous-Zeeman spacing — from the projection theorem. The moment $\boldsymbol\mu=-\mu_B(\mathbf L+g_s\mathbf S)/\hbar\approx-\mu_B(\mathbf J+\mathbf S)/\hbar$ is a vector operator, so within a level its matrix elements are $g_J$ times those of $\mathbf J$ ; the task is to evaluate $g_J=\langle\mathbf J\cdot(\mathbf J+\mathbf S)\rangle/(\hbar^2 j(j+1))$ using the coupling identity $\mathbf L=\mathbf J-\mathbf S$ .

Step 1. Expand the scalar: $\mathbf J\cdot(\mathbf J+\mathbf S)=J^2+\mathbf J\cdot\mathbf S$ , and use $\mathbf J\cdot\mathbf S=\tfrac12(J^2+S^2-L^2)$ (from squaring $\mathbf L=\mathbf J-\mathbf S$ ). Why: this rewrites everything in Casimirs $J^2,S^2,L^2$ with known eigenvalues $\hbar^2 j(j+1)$ , $\hbar^2 s(s+1)$ , $\hbar^2\ell(\ell+1)$ — no dynamics needed.

Step 2. Take the expectation: $\langle\mathbf J\cdot(\mathbf J+\mathbf S)\rangle=\hbar^2 j(j+1)+\tfrac12\hbar^2\big[j(j+1)+s(s+1)-\ell(\ell+1)\big]$ . Why: each Casimir is diagonal, so the expectation is the sum of eigenvalues; the $m$ -dependence cancels because the scalar is invariant.

Step 3. Divide by $\hbar^2 j(j+1)$ per the projection theorem to get $g_J=1+\dfrac{j(j+1)+s(s+1)-\ell(\ell+1)}{2 j(j+1)}$ . Why: this is equation (6), the exact factor by which $\boldsymbol\mu$ projects onto $\mathbf J$ and hence the coefficient in $\Delta E=g_J\mu_B m_j B$ .

Step 4. Check the anchors: $\ell=0$ gives $g_J=2$ (pure spin), $s=0$ gives $g_J=1$ (pure orbit), and $^2P_{3/2}$ gives $g_J=1+\tfrac{15/4+3/4-2}{2\cdot 15/4}=\tfrac43$ . Why: these three values are tabulated to many digits in atomic spectroscopy and match the anomalous-Zeeman patterns of alkali doublets, confirming the derivation.

Limiting cases

The theorem contracts to a familiar identity in every special rank and situation:

Case	Result	Meaning
Scalar operator, $k=0$	$\langle \alpha' j' m'\vert T^{(0)}_0\vert \alpha j m\rangle = \delta_{j'j}\,\delta_{m'm}\,\dfrac{\langle\alpha' j\Vert T^{(0)}\Vert \alpha j\rangle}{\sqrt{2j+1}}$	A rotational scalar cannot change $j$ or $m$ : the CG coefficient is diagonal. This is why the Hamiltonian, a scalar, is block-diagonal in $j,m$ and why energy levels carry those good quantum numbers.
Vector operator, $k=1$	$\langle \alpha j m'\vert V_q\vert \alpha j m\rangle = \dfrac{\langle \mathbf{J}\!\cdot\!\mathbf{V}\rangle}{\hbar^2 j(j+1)}\,\langle \alpha j m'\vert J_q\vert \alpha j m\rangle$	Within a fixed- $j$ level, any vector is proportional to $\mathbf J$ (the projection theorem). Foundation of the Landé $g$ -factor and the anomalous Zeeman effect.
Selection rules	$\vert j-k\vert \le j' \le j+k,\quad m' = m + q$	The element vanishes unless the CG coefficient is nonzero. Electric dipole ( $k=1$ ): $\Delta j=0,\pm1$ (not $0\!\to\!0$ ), $\Delta m=0,\pm1$ , with a parity change.
Identical initial/final level, $J_q$	$\langle j\Vert J\Vert j\rangle=\hbar\sqrt{j(j+1)}$	The reduced element of angular momentum itself (CG convention) — the normalization every projection ratio is measured against.

Common trapthe reduced element is meaningless without its convention

The commonest Wigner–Eckart error is quoting a reduced matrix element without saying which convention produced it. Authors differ on three independent choices: (i) the CG form of equation (1) versus the $3j$ -symbol form of equation (3), which move a $\sqrt{2j'+1}$ in or out; (ii) whether the operator couples as $j\otimes k$ or $k\otimes j$ , which flips a phase $(-1)^{j+k-j'}$ ; and (iii) the state phase convention (Condon–Shortley versus others). Two textbooks can therefore report reduced elements differing by $\sqrt{2j'+1}$ or by a sign, and neither is wrong — the physical matrix elements agree. Never combine reduced elements, or plug one into another book's formula, without checking all three conventions: only full matrix elements and their ratios are convention-independent.

Worked examples

1 · The reduced matrix element of angular momentum (fully worked)

Verify the theorem by computing the diagonal matrix elements of $J_z$ within the $j=1$ multiplet and extracting $\langle 1\|J\|1\rangle$ , using the CG convention of equation (1) and $J_0=J_z$ (a rank-1 tensor).

Step 1. Write the exact left-hand sides. $J_z$ is diagonal with eigenvalue $\hbar m$ , so $\langle 1,1|J_0|1,1\rangle=\hbar$ , $\langle 1,0|J_0|1,0\rangle=0$ , $\langle 1,-1|J_0|1,-1\rangle=-\hbar$ — the numbers the theorem must reproduce.

Step 2. Look up the coefficients $\langle j m;1 0|j m\rangle=m/\sqrt{j(j+1)}$ ; for $j=1$ they are $1/\sqrt2$ , $0$ , $-1/\sqrt2$ . The middle one vanishes, consistent with $\langle 1,0|J_z|1,0\rangle=0$ .

Step 4. Cross-check against $\langle j\|J\|j\rangle=\hbar\sqrt{j(j+1)}$ (CG convention): for $j=1$ this is $\hbar\sqrt{2}$ — exactly what we extracted. One number, $\hbar\sqrt{j(j+1)}$ , encodes every component of $\mathbf J$ across the multiplet. In the $3j$ /Racah convention it would read $\hbar\sqrt{j(j+1)(2j+1)}=\hbar\sqrt6$ instead — same physics, the $\sqrt{2j+1}$ relocated, exactly the warning above.

2 · Landé g-factors and the sodium D-line Zeeman pattern (fully worked)

The sodium D doublet is $3\,^2P_{3/2}\to 3\,^2S_{1/2}$ (D2) and $3\,^2P_{1/2}\to 3\,^2S_{1/2}$ (D1). Find each level's Landé $g$ -factor and the weak-field Zeeman splitting of the ground state at $B=0.10\ \mathrm T$ .

Step 1. Apply equation (6) to each level. Ground $^2S_{1/2}$ ( $\ell=0,s=\tfrac12,j=\tfrac12$ ): $g_J=1+\dfrac{3/4+3/4}{3/2}=\mathbf 2$ , pure spin. Excited $^2P_{1/2}$ ( $j=\tfrac12$ ): numerator $\tfrac34+\tfrac34-2=-\tfrac12$ , so $g_J=1-\tfrac13=\mathbf{2/3}$ . Excited $^2P_{3/2}$ ( $j=\tfrac32$ ): numerator $\tfrac{15}{4}+\tfrac34-2=\tfrac{10}{4}$ over $\tfrac{30}{4}$ , so $g_J=1+\tfrac13=\mathbf{4/3}$ .

Step 2. Ground-state splitting: adjacent sublevels ( $m_j=\pm\tfrac12$ ) split by $\Delta E=g_J\mu_B B\,\Delta m_j$ . With $g_J=2$ , $\mu_B=5.788\times10^{-5}\ \mathrm{eV/T}$ , $B=0.10\ \mathrm T$ : $\Delta E=2\times(5.788\times10^{-5})\times0.10=\mathbf{1.16\times10^{-5}\ eV}$ .

Step 3. In frequency units, using $\mu_B/h=13.996\ \mathrm{GHz/T}$ : the ground-state sublevel gap is $g_J(\mu_B/h)B=2\times13.996\times0.10=\mathbf{2.80\ GHz}$ . Because $g$ differs between $^2S_{1/2}$ ( $2$ ), $^2P_{1/2}$ ( $2/3$ ), and $^2P_{3/2}$ ( $4/3$ ), the D1 and D2 lines split into different numbers of components with unequal spacings — the "anomalous" pattern that historically demanded electron spin and $g_s\approx2$ . And since both lines share one radial integral, their strengths differ only through the emitting-level degeneracy $2j+1$ , giving the famous $S(\mathrm{D2}):S(\mathrm{D1})=2:1$ ratio with no integral evaluated.

Physics of the portrait

The theorem is where group representation theory entered the daily practice of physics. It carries two names — Carl Eckart's 1930 application of group theory to atomic matrix elements and Eugene Wigner's systematic 1931 treatment — but its deepest champion was Wigner, who insisted throughout his career that symmetry, not force laws, is the organizing principle of quantum mechanics.

Origins · 1 minEugene Wigner and how this was discovered

Eugene Wigner

1902–1995 · symmetry and group theory in quantum mechanics

Eugene Paul Wigner was born in Budapest and trained as a chemical engineer before physics claimed him. In the late 1920s he did something the community initially resisted: he imported group representation theory, then a corner of pure mathematics, into the quantum theory of spectra. His 1931 book Gruppentheorie und ihre Anwendung auf die Quantenmechanik der Atomspektren laid out how the rotation group organizes angular momentum, and in it the theorem now bearing his and Eckart's names appears as a consequence of Schur's lemma: matrix elements of tensor operators must factor into a geometric coefficient and a dynamical constant. Colleagues joked about the "Gruppenpest" (the group-theory plague), but within two decades it had become the language of the field. Wigner also introduced the $3j$ and $6j$ symbols, proved the theorem on time-reversal antiunitarity, and classified the irreducible representations of the Poincaré group — the result that tells us a particle is a label $(m,s)$ of mass and spin. He shared the 1963 Nobel Prize for these symmetry principles. Giulio Racah later recast the apparatus in terms of irreducible tensor operators and reduced matrix elements — the "Racah algebra" spectroscopists still use — and A. R. Edmonds standardized the notation every modern textbook inherits.

History

1930

— Eckart applies group theory to the quantum dynamics of atomic matrix elements

1931

— Wigner's book states the theorem via the representation theory of the rotation group

1935

— Condon and Shortley fix the phase convention for angular-momentum coupling

1942

— Racah develops irreducible tensor operators and the reduced-matrix-element formalism

1957

— Edmonds systematizes the 3j and 6j symbols in Angular Momentum in Quantum Mechanics

Going further

Texts. Sakurai & Napolitano, Modern Quantum Mechanics, 3rd ed., Ch. 3 (standard derivation and projection theorem) · Edmonds, Angular Momentum in Quantum Mechanics (definitive on $3j/6j$ symbols) · Stone & Goldbart, Mathematics for Physics, group-representation chapter (Schur's-lemma proof) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed. · Hassani, Mathematical Physics · Messiah, Quantum Mechanics, Vol. 2, Appendix C.

Provenance. The theorem's CG and $3j$ forms, selection rules, and projection theorem follow Sakurai & Napolitano Ch. 3 and Edmonds; the Schur's-lemma proof follows Stone & Goldbart's group-representation chapter. A verifier should recompute: (i) the $k=0$ scalar limit using $\langle j m;0 0|j' m'\rangle=\delta_{j'j}\delta_{m'm}$ ; (ii) the CG– $3j$ relation with the $\sqrt{2j'+1}$ factor; (iii) the reduced element $\langle 1\|J\|1\rangle=\hbar\sqrt2$ from the $j=1$ $J_z$ matrix elements (Worked Example 1); (iv) the Landé factors $g(^2S_{1/2})=2$ , $g(^2P_{1/2})=2/3$ , $g(^2P_{3/2})=4/3$ from equation (6); (v) the D2:D1 line-strength ratio $2:1$ from the upper-level degeneracies. All numerical values are standard textbook results.

Prerequisites. Group Representations · Reducible and Irreducible Representations · Schur's Lemmas · Tensor Products and Clebsch–Gordan Decomposition · The Rotation Group SO(3). Leads to. Characters and Orthogonality Theorems · Lie Algebras and the Exponential Map · Character Tables and Decomposition of Representations.

Ask the tutor: Ask me to walk through the Schur's-lemma proof line by line, or to compute the reduced matrix element and full Zeeman pattern for any atomic multiplet you name.

What is a spherical tensor operator, in one sentence?

It is a set of $2k+1$ operators $T^{(k)}_{-k},\dots,T^{(k)}_{k}$ that rotate among themselves exactly the way the kets $|k q\rangle$ do, the practical test being the commutator pair in equation (2). Familiar examples: the identity (rank 0, scalar), the position or momentum vector (rank 1, with $T^{(1)}_{\pm1}=\mp(x\pm iy)/\sqrt2$ , $T^{(1)}_0=z$ ), and the quadrupole moment (rank 2). Building a rank- $k$ object from Cartesian tensors is just Clebsch–Gordan coupling done on operators instead of states.

Why does all the m-dependence live in the Clebsch–Gordan coefficient?

Because rotating your coordinate axes changes $m,m',q$ but cannot change any physics. Both the states and the operator carry known Wigner- $D$ matrices under a rotation; by the Schur's-lemma proof the reduced matrix element is a rotational invariant — what survives after averaging over all orientations — so it is blind to $m,m',q$ by construction. Everything that does respond to reorienting the system is geometry, exactly what the Clebsch–Gordan coefficient tabulates. The split is exact and forced by symmetry, not an approximation.

How is the theorem actually used to save labor?

You compute or measure one matrix element, divide out its Clebsch–Gordan coefficient to get the reduced element, and every other component of the multiplet follows by multiplying by a tabulated CG coefficient. For a dipole transition between $j=2$ and $j'=1$ that is $45$ potential matrix elements reduced to one number plus lookups. Better still, ratios within the multiplet need no dynamical input — the reduced element cancels — so relative line intensities are pure geometry, as in the sodium D2:D1 = 2:1 ratio.

What is the difference between the Clebsch–Gordan form and the 3j form?

They express the same theorem with a different geometric factor. The CG form (equation 1) uses $\langle j m;k q|j' m'\rangle$ , natural for coupling $j$ and $k$ to $j'$ . The $3j$ form (equation 3) uses the Wigner $3j$ symbol — the CG coefficient rescaled by a phase over $\sqrt{2j'+1}$ and symmetrized so permuting its columns only adds a known phase, which makes it preferred for tabulation. The price: the $\sqrt{2j'+1}$ is absorbed into the reduced element, so a reduced element in the $3j$ convention differs from the CG one by exactly that factor — the source of most convention confusion.

Does the theorem apply to groups other than SO(3)?

Yes — that is its real power. The proof used only three ingredients: states carry irreducible representations, the operator transforms in some irrep, and Schur's lemma governs maps between irreps. All three hold for any compact group, so for isospin $SU(2)$ , flavor $SU(3)$ , or a point group the matrix elements factor into that group's Clebsch–Gordan coefficients times reduced elements. One subtlety: when the target irrep appears more than once in the product (as in $SU(3)$ ), you need one reduced element per copy and a multiplicity label on the coefficient. For $SO(3)$ multiplicities are always one, which is why the rotational case looks so clean.

Can you measure the reduced matrix element directly?

Not by itself — only full matrix elements and their ratios are physical observables; the reduced element is a convention-dependent bookkeeping number. You determine it indirectly: measure a matrix element for one $(m,q,m')$ , divide by the tabulated Clebsch–Gordan coefficient in your convention, and the quotient predicts every other component. Because its value shifts with the CG-versus- $3j$ choice and coupling order, always publish the convention alongside it — a reduced element without its convention is like a coordinate without its axes.