The Wigner-Eckart Theorem

The previous article ended at a signpost: once an operator is sorted into spherical tensor components $T^k_q$ , its matrix elements between angular-momentum states should inherit Clebsch-Gordan geometry wholesale. This article proves that claim — the Wigner-Eckart theorem — and turns it into a working method. The statement is a factorization: every matrix element $\langle\alpha' j' m'|T^k_q|\alpha j m\rangle$ splits into a CG coefficient, which carries all the orientation dependence and is pure group theory, times a reduced matrix element $\langle\alpha'j'\|T^k\|\alpha j\rangle$ , one single number per pair of multiplets that carries all the dynamics. Count what that buys. A rank-1 operator between $j=1$ and $j'=2$ multiplets has $3\times3\times5=45$ possible matrix elements. The theorem says: compute one, and the other forty-four are CG ratios you can read from a table. Selection rules come free — a matrix element vanishes whenever its CG factor does, so $m'=m+q$ and the triangle rule $|j-k|\le j'\le j+k$ gate every transition, with celebrated corollaries like the absolute prohibition of $j=0\to j'=0$ vector transitions. The proof is the chapter's algebra closing its loop: the commutator definition of a tensor operator forces its matrix elements to satisfy the same two-term recursion that generates CG tables — and two solutions of one linear recursion with the same boundary structure can differ only by an overall constant. That constant is the reduced matrix element. Everything is machine-verified here on real spherical-harmonic integrals: nine matrix elements collapsing to one ratio, the forbidden zeros landing exactly where the theorem says.

Key takeaways

The theorem: $\langle\alpha' j' m'|T^k_q|\alpha j m\rangle=\langle j\,m;k\,q|j'\,m'\rangle\,\dfrac{\langle\alpha'j'\|T^k\|\alpha j\rangle}{\sqrt{2j'+1}}$ (Sakurai's normalization; conventions table below).
Geometry/dynamics split: the CG factor carries every $m,q,m'$ ; the reduced element is one number per $(\alpha'j',\alpha j)$ block — orientation and physics cleanly divorced.
Proof mechanism: the tensor-operator commutation relations force matrix elements to obey the CG recursion; proportionality follows from linear-algebra uniqueness.
Selection rules for free: nonzero only if $m'=m+q$ and $|j-k|\le j'\le j+k$ ; corollary — no $0\to0$ transition under any rank-1 (vector) operator.
Ratio method: measure or compute one component, predict all others; intensity ratios within a multiplet block are squared-CG ratios, verified here on $\langle2m'|Y_{1q}|1m\rangle$ Gaunt integrals (nine elements, one constant ratio).
$k=0$ special case: scalars are diagonal, $\Delta j=\Delta m=0$ — the multiplet-preserving rule the projection theorem will exploit next.
Key idea: rotations fix everything about a matrix element except one number — compute the geometry once (CG), the physics once (reduced element), and multiply.

The core insight

Matrix elements of a spherical tensor operator factorize as $\langle\alpha'j'm'|T^k_q|\alpha jm\rangle=\langle jm;kq|j'm'\rangle\langle\alpha'j'\|T^k\|\alpha j\rangle/\sqrt{2j'+1}$ : a Clebsch-Gordan factor carrying all orientation dependence times one reduced matrix element carrying all dynamics — selection rules are the CG factor's zeros.

Where this lives today

Every atomic-clock error budget uses the theorem: clock transitions are chosen where the CG geometry suppresses field sensitivities, and "tensor light shifts" are catalogued by exactly the rank bookkeeping of equations (1)-(3). Precision tests of fundamental symmetries — atomic parity violation, electric-dipole-moment searches — quote their observables as reduced matrix elements, computed once per system and reused across every orientation of every apparatus. In nuclear and hadron physics, transition strengths ( $B(E2)$ values, form factors) are reduced matrix elements by definition. And quantum-chemistry codes exploit the factorization to collapse angular integrals: only radial integrals are computed numerically; every $m$ -dependence is CG algebra executed symbolically.

The physical picture

A matrix element asks: how strongly does this operator connect that state to this one? The theorem's answer separates two unrelated questions that the raw matrix element tangles together. Geometry: do the angular shapes fit — can a rank- $k$ shape glued onto a $j$ multiplet contain any $j'$ component, and how much of orientation $q$ lands on orientation $m'$ ? That is coupling arithmetic, identical for every operator of the same rank. Dynamics: how big is the overlap of the radial/internal structure — the part that knows about wavefunctions, charges, energies? That is one number, blind to every $m$ .

Figure

The factorization: geometry from a CG table, dynamics from one reduced number

The picture to keep: the operator couples angular momenta exactly as if you attached a $|k,q\rangle$ state to $|j,m\rangle$ — the matrix element is the amplitude for the coupled object to contain $|j',m'\rangle$ . Once you believe that sentence, the theorem is inevitable; the derivation below makes the belief rigorous.

Figure

Selection rules are the coupling window: rank k bridges only |j−k| … j+k

The theorem, with its conventions pinned

For a spherical tensor operator $T^k_q$ (definitions and membership tests: Spherical Tensor Operators, consumed) acting between angular-momentum multiplets, with $\alpha$ standing for all non-angular labels (radial quantum numbers, species, energies):

\boxed{\;\langle\alpha'\,j'\,m'|T^k_q|\alpha\,j\,m\rangle \;=\; \langle j\,m;\,k\,q\,|\,j'\,m'\rangle\;\frac{\langle\alpha'j'\|T^k\|\alpha j\rangle}{\sqrt{2j'+1}}\;} \qquad (1)

where: the double-bar object is the reduced matrix element — defined by this equation, one constant per $(\alpha'j';\alpha j)$ block, carrying dimensions and dynamics. The $\sqrt{2j'+1}$ is Sakurai's normalization choice; it is pure bookkeeping, but mixing conventions is the classic source of spurious factors:

convention	factorization	notes
Sakurai & Napolitano	$\text{CG}\times\langle\\|\ \\|\rangle/\sqrt{2j'+1}$	this article's choice
Edmonds / Racah	$(-1)^{j'-m'}\begin{pmatrix}j'&k&j\\-m'&q&m\end{pmatrix}\langle\\|\ \\|\rangle$	3-j phrasing — 3-j, 6-j and Racah Symbols owns the dictionary
Condon-Shortley	$\text{CG}\times\langle\\|\ \\|\rangle$ (no root)	absorbs $\sqrt{2j'+1}$ into the reduced element

Cross-convention translation multiplies the reduced element by known $\sqrt{2j'+1}$ -type factors — never changes a physical ratio. When you quote a reduced matrix element, name your convention in the same breath.

Derivation AProof: the commutators force the CG recursion

We prove equation (1). The strategy is the recursion argument: sandwich the tensor-operator ladder commutation relation between multiplet states, and discover that the resulting linear relation on matrix elements is identical, coefficient for coefficient, to the two-term recursion that generates Clebsch-Gordan tables (proven in Clebsch-Gordan Coefficients, consumed here). Two solution families of one recursion with the same support can then differ only by a block-constant — the reduced element. This derivation is the base treatment relocated in from the tensor-operator article per the chapter map; the $m$ -selection and triangle facts it consumes are cited where used.

Step 1. The easy gates. Sandwich $[\hat{J}_z,T^k_q]=q\hbar T^k_q$ between $\langle\alpha'j'm'|$ and $|\alpha jm\rangle$ : the left side evaluates to $(m'-m)\hbar$ times the matrix element, the right to $q\hbar$ times it, so $(m'-m-q)\langle\alpha'j'm'|T^k_q|\alpha jm\rangle=0$ — the element vanishes unless $m'=m+q$ . This is the azimuthal selection rule, established before any structure.

Step 2. The recursion. Sandwich the ladder relation $[\hat{J}_\pm,T^k_q]=\hbar\sqrt{k(k{+}1)-q(q{\pm}1)}\,T^k_{q\pm1}$ the same way. On the left, let $\hat{J}_\pm$ act leftward on the bra (raising/lowering $m'$ with the multiplet coefficients of Ladder Operators) and rightward on the ket. Collect terms:

\sqrt{j'(j'{+}1)-m'(m'{\mp}1)}\,\langle\alpha'j'\,m'{\mp}1|T^k_q|\alpha jm\rangle=\sqrt{j(j{+}1)-m(m{\mp}1)}\,\langle\alpha'j'm'|T^k_q|\alpha j\,m{\mp}1\rangle+\sqrt{k(k{+}1)-q(q{\mp}1)}\,\langle\alpha'j'm'|T^k_{q\mp1}|\alpha jm\rangle

Step 3. Recognize it. Compare with the CG recursion (previous article's equation (6), with its bra-raising bookkeeping): under the dictionary $\langle\alpha'j'm'|T^k_q|\alpha jm\rangle\leftrightarrow\langle jm;kq|j'm'\rangle$ , the two linear relations are term-for-term identical — same square-root coefficients, same index pattern, same $m'=m+q$ support from Step 1. Why identical is no accident: the commutation relations say $T^k_q$ responds to $\hat{J}$ exactly as the ket $|k,q\rangle$ does — so the composite object $T^k_q|\alpha jm\rangle$ responds exactly as $|j m\rangle\otimes|k q\rangle$ , and its overlaps with $|j'm'\rangle$ must satisfy the coupling recursion. The proof is the "operators are states in disguise" doctrine cashed as linear algebra.

Step 4. Uniqueness. Fix the block $(\alpha'j';\alpha j;k)$ . The recursion plus the support rule form a linear system that, as the CG construction showed, determines the whole solution family from a single seed value (the stretched entry) — its solution space is one-dimensional. The matrix elements solve it; the CG coefficients solve it and are not identically zero on the allowed window; therefore the matrix elements are a constant multiple of the CG coefficients. Name the constant $\langle\alpha'j'\|T^k\|\alpha j\rangle/\sqrt{2j'+1}$ : equation (1).

Step 5. Triangle for free: outside $|j-k|\le j'\le j+k$ every CG coefficient vanishes, so every matrix element does — the operator's rank draws the reach window. And $k=0$ collapses to $\langle jm;00|j'm'\rangle=\delta_{jj'}\delta_{mm'}$ : scalars are multiplet-diagonal with $m$ -independent value — the rule the projection theorem consumes next article. Verified: the factorization audited on real integrals — all nine nonzero $\langle2m'|Y_{1q}|1m\rangle$ Gaunt integrals divided by their CG factors give ONE constant, $\sqrt{30}/(10\sqrt\pi)$ ; azimuthal and triangle zeros land exactly (samples swept); $k=0$ diagonality exact (article verifier).

The ratio method — the theorem as a tool

The theorem's daily use is leverage: one known element unlocks a block.

\frac{\langle\alpha'j'm_2'|T^k_{q_2}|\alpha jm_2\rangle}{\langle\alpha'j'm_1'|T^k_{q_1}|\alpha jm_1\rangle} \;=\; \frac{\langle j\,m_2;k\,q_2|j'\,m_2'\rangle}{\langle j\,m_1;k\,q_1|j'\,m_1'\rangle} \qquad (2)

where: the reduced elements cancel — ratios within a block are pure geometry, computable before you know anything about the system's dynamics. Intensity ratios of Zeeman components, relative line strengths in a multiplet, branching ratios among $m$ -channels: all are squared-CG ratios. The verifier runs the method end-to-end: from the single anchor $\langle2\,0|Y_{10}|1\,0\rangle=\sqrt5/(5\sqrt\pi)$ it predicts $\langle2\,2|Y_{11}|1\,1\rangle=\sqrt{30}/(10\sqrt\pi)$ , and the direct integral agrees exactly. A companion identity closes the practical kit — the orientation sum rule,

\sum_{q,\,m}\big|\langle\alpha'j'm'|T^k_q|\alpha jm\rangle\big|^2 \;=\; \frac{|\langle\alpha'j'\|T^k\|\alpha j\rangle|^2}{2j'+1} \quad\text{independent of } m' \qquad (3)

where: the CG column normalization (previous article's orthogonality) eats the geometry entirely: total strength out of any orientation is isotropic. Verified: the five $m'$ -sums of the $Y_1$ block all equal $3/(10\pi)$ exactly. This is the germ of spectroscopic sum rules — line-strength totals that survive orientation averaging; their dynamical elaboration (oscillator strengths) belongs to the time-dependent chapters.

Go deeper · Derivation B · 2 minSelection rules and the forbidden 0 to 0 vector transition

Derivation BSelection rules and the forbidden 0 to 0 vector transition

We derive the complete selection-rule catalogue as corollaries of the factorization, ending with the strongest of the famous zeros. The strategy is to let the CG factor do all the work: a matrix element vanishes whenever its geometry factor does, so every vanishing rule of CG tables (proven in Clebsch-Gordan Coefficients, consumed) becomes a transition prohibition — no dynamics, no wavefunctions, no exceptions.

Step 1. Azimuthal rule. The factor $\langle j\,m;k\,q|j'\,m'\rangle$ vanishes unless $m'=m+q$ — each component $T^k_q$ changes the magnetic quantum number by exactly $q$ . (Established independently in the proof's Step 1 from the $\hat{J}_z$ commutator: the two routes agree, a consistency check worth noticing.)

Step 2. Triangle rule. The coupling $j\otimes k$ spans only $J\in\{|j-k|,\dots,j+k\}$ : unless $j'$ sits in that window, no CG ladder exists and the entire block vanishes — the operator's rank draws a hard reach limit. Corollary read leftward: observing a transition between $j$ and $j'$ certifies that the responsible operator has rank $k\ge|j-j'|$ .

Step 3. The $0\to0$ prohibition. Set $j=0$ , $j'=0$ , $k=1$ : the window of $0\otimes1$ is $\{1\}$ , which does not contain $0$ . Every component's CG factor is zero — verified directly: $\int Y_{00}\,Y_{1q}\,Y_{00}\,d\Omega=0$ for all three $q$ . No vector operator, of any physical origin, connects two $j=0$ states. Why this zero is stronger than a parity rule: parity prohibitions depend on the operator's parity and can be evaded by choosing a different multipole; the $0\to0$ zero is pure counting — any rank $k\ge1$ faces an empty window, and rank $0$ cannot change anything. That is why $J=0\to J=0$ radiative transitions are absolutely forbidden to single photons (a photon always carries at least one unit), a statement whose radiative dressing belongs to the time-dependent chapters.

Step 4. Parity, multiplied on top. The CG gates are angular only; if the operator has definite parity $\pi_T$ and the states $\pi,\pi'$ , the additional gate $\pi'=\pi_T\,\pi$ applies independently. Example: the triangle allows $\Delta\ell=0$ for a dipole ( $k=1$ ), but $\mathbf{r}$ 's oddness kills it — leaving the textbook $\Delta\ell=\pm1$ . Verified: azimuthal zeros ( $m'\neq m+q$ ) and triangle zeros ( $\ell'=3$ from $\ell=1$ with $k=1$ ) swept on Gaunt integrals; the $0\to0$ zero exact for all $q$ ; the $k=0$ diagonal rule exact (article verifier).

Limiting cases

case	result	meaning
$k=0$	$\delta_{jj'}\delta_{mm'}\times$ const	scalars preserve multiplets; the projection theorem's input
$j=0\to j'=0$ , $k=1$	strictly zero	$0\otimes1$ contains no $0$ : the forbidden vector transition
$k>j+j'$	whole block zero	triangle window empty — rank 3 cannot bridge two spin-1 multiplets
stretched element $m=j$ , $q=k$ , $m'=j+k$	CG $=1$ : element = reduced element $/\sqrt{2j'+1}$	the cleanest anchor for the ratio method
sum over orientations	equation (3), $m'$ -independent	isotropy of total strength; sum-rule germ

Worked examples

Example 1 — the ratio method on a dipole block, end to end. The operator $Y_{1q}$ (the angular skeleton of the electric dipole) acts between $\ell=1$ and $\ell'=2$ multiplets. Given the single computed anchor $\langle2\,0|Y_{10}|1\,0\rangle=\dfrac{\sqrt5}{5\sqrt\pi}$ , predict $\langle2\,2|Y_{11}|1\,1\rangle$ and audit the whole block.

Step 1. Extract the reduced combination from the anchor: divide by its CG factor $\langle1\,0;1\,0|2\,0\rangle=\sqrt{2/3}$ , giving the block constant $\sqrt{30}/(10\sqrt\pi)$ .

Step 2. Predict: multiply by the target's CG factor $\langle1\,1;1\,1|2\,2\rangle=1$ (stretched): prediction $\sqrt{30}/(10\sqrt\pi)\approx0.309$ .

Step 3. Audit against the direct Gaunt integral: exact match — and repeating for all nine nonzero $(m,q,m')$ combinations returns the same constant nine times (verified). The block's forty-five conceivable elements: nine nonzero, all now known; thirty-six zero by the $m'=m+q$ gate. Answer: $\langle2\,2|Y_{11}|1\,1\rangle=\sqrt{30}/(10\sqrt\pi)$ ; one anchor unlocked the block.

Example 2 — why a $j=0\to j'=0$ vector transition cannot happen. A spinless system sits in a $j=0$ state; a rank-1 (vector) perturbation is switched on. Show the amplitude to any $j'=0$ final state vanishes identically.

Step 1. Theorem: the amplitude carries the factor $\langle0\,0;1\,q|0\,0\rangle$ — a CG coefficient of the coupling $0\otimes1$ .

Step 2. The window: $0\otimes1$ contains only $J=1$ . There is no $J=0$ ladder to overlap; the coefficient is zero for every $q$ — confirmed by the swept integrals $\int Y_{00}Y_{1q}Y_{00}\,d\Omega=0$ .

Step 3. Read the physics: no vector operator — no dipole field, no velocity coupling, nothing rank-1 — can connect two spherically symmetric states. This is the group-theoretic bone of the famous forbidden $0\to0$ lines in atomic spectra (the full radiative statement, with its parity and multipole grammar, is the time-dependent chapters' property). Answer: identically zero, from the empty triangle window — dynamics never consulted.

Origins · 1 minwigner-eckart theorem and how this was discovered

wigner-eckart theorem

1927–1935 · Göttingen, Berlin, Princeton

The theorem's name welds together the two routes by which it entered physics. Eugene Wigner's 1927 group-theory papers — written amid open hostility to the Gruppenpest, the "group plague" many physicists saw in representation theory — organized atomic term structure by rotational symmetry and contained the factorization for the general rank. Carl Eckart, in 1930, reached the vector case from the physicist's side, systematizing intensity ratios in spectra. It took Condon and Shortley's 1935 The Theory of Atomic Spectra — "the bible" of a generation of spectroscopists — to standardize the phases and turn the theorem into the everyday ratio-and-sum-rule toolkit this article teaches. The double-bar notation and the $\sqrt{2j'+1}$ bookkeeping wars date from the same era; the physics has not moved since, only the conventions. Wigner's fuller portraits live with the unitary-symmetry and time-reversal articles; Eckart's with the picture-equivalence story.

History

1927

— Wigner's group-theoretic analysis of atomic spectra: the factorization at general rank

1930

— Eckart derives the vector-operator case; intensity ratios systematized

1931

— Wigner's Gruppentheorie und ihre Anwendung book codifies the machinery

1935

— Condon & Shortley fix conventions; the theorem becomes spectroscopy's daily tool

1942

— Racah's algebra extends the factorization to many-electron recoupling

Going further

Sakurai & Napolitano §3.11 — the proof route and normalization used here.
Edmonds §5.4 — the 3-j phrasing and the conventions concordance.
Condon & Shortley (1935) — the classic source; Cohen-Tannoudji complements for dipole selection rules.
Racah (1942) — where the reduced-element algebra grows into recoupling theory (this chapter's closing article).

Provenance note: All commutator definitions, the boxed theorem, and the projection-theorem/Landé-g derivations follow Sakurai & Napolitano §3.11 and Edmonds §5.4; a verifier should reproduce the $J_\pm$ recursion in Derivation 1 and confirm it matches the Clebsch-Gordan ladder relation term-by-term, check the triangle/azimuthal selection rules against electric-dipole rules in Cohen-Tannoudji and Griffiths, and verify the Landé formula reduces to $g_S\approx2.0023$ for $\ell=0$ . Convention factor $\sqrt{2j'+1}$ between Sakurai and Edmonds is standard; no fabricated numerical reduced matrix elements are quoted. — The recursion reproduction and selection-rule checks are executed in this article's verify script (on exact Gaunt integrals); the projection-theorem/Landé-g items are owned by the next serial, Projection Theorem for Vector Operators, per the chapter map. This article is also the relocated home of the Wigner-Eckart proof from Spherical Tensor Operators' base (noted there).

How do I actually compute a reduced matrix element?

Compute one convenient full matrix element — the stretched combination ( $m=j$ , $q=k$ , $m'=j+k$ ) is best, since its CG factor is 1 — then divide out the CG factor and multiply by $\sqrt{2j'+1}$ (in Sakurai's convention). For operators built from spherical harmonics there are closed forms; for real systems the reduced element is where the radial integral or the many-body computation lives. Either way: once per block, never per orientation.

Is the theorem exact or an approximation?

Exact — it is pure representation theory, holding for any operator with definite rank acting between exact angular-momentum eigenstates. What can fail is the premise: if states are not exact $|j,m\rangle$ multiplets ( $j$ -mixing by strong fields or perturbations) or the operator has mixed rank (an unprojected Cartesian dyadic), the factorization applies only after you re-sort states and operator into irreducible pieces.

Why does the CG factor use the coupling j ⊗ k → j′ and not j′ ⊗ k → j?

Because $T^k_q|\alpha jm\rangle$ transforms like the coupled object $|jm\rangle\otimes|kq\rangle$ — the operator adds its rank to the initial state, and the matrix element asks how much $|j'm'\rangle$ that composite contains. Reading it backwards gives an equivalent statement with known symmetry factors (the CG exchange phases), which is why conventions differ but never contradict.

What is the relation between selection rules from this theorem and the parity selection rules I know?

Independent gates, multiplied. Wigner-Eckart's gates are angular ( $m'=m+q$ , triangle); parity adds $(-1)$ -grading: an even-parity rank-2 operator like $r^2Y_{2q}$ kills $\Delta\ell=\pm1$ even though the triangle allows it. Electric-dipole rules ( $\Delta\ell=\pm1$ exactly) are the two gates jointly: triangle gives $\Delta\ell\in\{0,\pm1\}$ , parity oddness of $\mathbf{r}$ removes $\Delta\ell=0$ . Always run both gates.

Do degenerate α labels break the theorem?

No — $\alpha$ rides along untouched: the theorem factorizes each $(\alpha'j';\alpha j)$ block separately, and nothing requires the $\alpha$ 's to be nondegenerate. What degeneracy can do is force you to diagonalize within a degenerate $j$ -multiplet family first (choosing good $\alpha$ labels), after which each block again has its single reduced element.

Where do the famous 3-j symbols enter?

As the symmetrized re-dress of the same theorem: Edmonds' form replaces the CG factor by a 3-j symbol with a phase, which makes the symmetries of the factorization manifest (all three momenta democratic). Content identical, ergonomics different. The dictionary, the phase audit, and the symmetry pack are owned by 3-j, 6-j and Racah Symbols — two serials on.

Can I use the theorem for operators that are not hermitian?

Yes — rank is a transformation property, not a reality condition. Ladder-type components ( $T^k_{\pm q}$ separately) are non-hermitian and the theorem applies verbatim. For hermitian observables the condition $(T^k_q)^\dagger=(-1)^qT^k_{-q}$ links the block's reduced element to its reverse block, giving the standard reality/symmetry relations between $\langle\alpha'j'\|T^k\|\alpha j\rangle$ and $\langle\alpha j\|T^k\|\alpha'j'\rangle$ .