Tensor Products and Clebsch–Gordan Decomposition

Two physical systems, brought together, are described by the tensor product of their state spaces — and the moment a symmetry acts on the pair, the product carries a representation of the group. Rotate a two-particle system and both particles rotate at once; that diagonal action on $V_1\otimes V_2$ is the tensor-product representation $D^{(1)}\otimes D^{(2)}$ . The central fact of this article is that even when the two factors are irreducible, their product almost never is: the combined space silently breaks into invariant blocks, each transforming irreducibly on its own. Naming those blocks and their multiplicities is the Clebsch–Gordan decomposition, and the unitary matrix rotating the uncoupled basis into the block (coupled) basis is the array of Clebsch–Gordan coefficients. What makes the decomposition tractable is that it is dictated entirely by characters. Because the trace of a Kronecker product factorizes, the character of $D^{(1)}\otimes D^{(2)}$ is the pointwise product $\chi^{(1)}\chi^{(2)}$ , and one line of character orthogonality reads off how many copies of each irreducible appear. The most important instance is the addition of angular momenta: two spins $j_1$ and $j_2$ combine into total $j$ from $|j_1-j_2|$ to $j_1+j_2$ , one multiplet each. The same machinery organizes atomic term symbols, the singlet–triplet chemistry of two electrons, the meson octet of flavour $SU(3)$ , and — through the Wigner–Eckart theorem next — every selection rule in spectroscopy.

Key takeaways

The tensor product $V_1\otimes V_2$ carries a representation $D^{(1)}\otimes D^{(2)}$ whose character is the ordinary product of characters, $\chi^{(1\otimes2)}=\chi^{(1)}\chi^{(2)}$ .
That product is generically reducible; the Clebsch–Gordan series writes it as $\bigoplus_\lambda m_\lambda D^{(\lambda)}$ , with multiplicities $m_\lambda$ read off by character orthogonality.
For $SU(2)$ the rule is $j_1\otimes j_2=\bigoplus_{j=|j_1-j_2|}^{j_1+j_2}j$ , each total $j$ exactly once, checked by $(2j_1+1)(2j_2+1)=\sum_j(2j+1)$ .
The Clebsch–Gordan coefficients $\langle j_1 m_1\,j_2 m_2|j\,m\rangle$ are the unitary change of basis from the uncoupled product states to the coupled total- $j$ states.
They vanish unless $m=m_1+m_2$ and the triangle $|j_1-j_2|\le j\le j_1+j_2$ holds, and are real in the Condon–Shortley convention.
Key idea: a tensor product of irreducibles splits uniquely into a direct sum of irreducibles whose multiplicities come from products of characters, and the unitary matrix relating the uncoupled basis to the coupled basis is exactly the array of Clebsch–Gordan coefficients.

The core insight

A tensor product of irreducible representations decomposes uniquely into a direct sum of irreducibles whose multiplicities are computed from products of characters; the unitary matrix relating the product (uncoupled) basis to the irreducible (coupled) basis is the array of Clebsch–Gordan coefficients.

Where this lives today

The coupling of angular momenta is live infrastructure. In quantum computing, the singlet–triplet qubit encodes one logical bit in exactly the $\tfrac12\otimes\tfrac12=1\oplus0$ split of two electron spins in a semiconductor double quantum dot: the logical $0$ is the singlet $|0,0\rangle$ and the logical $1$ a triplet state, and gates are controlled exchange interactions $J\,\mathbf{S}_1\cdot\mathbf{S}_2$ whose eigenstates are the coupled basis. In precision metrology, optical-lattice atomic clocks use Clebsch–Gordan coefficients to compute the tensor light shifts and Zeeman structure that must be nulled to reach $10^{-18}$ accuracy, and the Wigner–Eckart theorem reduces every transition rate to a reduced matrix element times a coefficient. In nuclear and particle physics the same algebra survives at higher symmetry: hadron spectroscopy is organized by $SU(3)$ tensor products (the meson octet and singlet, the baryon decuplet), and lattice-QCD calculations project multi-particle operators onto irreducible representations of the lattice rotation group using Clebsch–Gordan tables. Every general-purpose scientific library — from GSL to SymPy to the atomic-structure codes behind astrophysical spectra — ships a Clebsch–Gordan or $3j$ routine, because the moment two quantum systems meet, this decomposition is the first thing you compute.

The physical picture

Picture two independent spinning tops, each carrying its own multiplet of orientation states. Bolt them onto one rigid frame and rotate the whole assembly: the rotation acts on both tops at once. That single diagonal action is the tensor-product representation, and the assembly has a sharp total angular momentum only along special combinations of the individual states. Finding those combinations — repackaging the product basis into blocks that each rotate as a clean multiplet — is the entire content of the decomposition. The cheapest sanity check rides along for free: no states are created or destroyed, so the dimensions balance, $(2j_1+1)(2j_2+1)=\sum_j(2j+1)$ .

Figure

Two spin-½ systems: the four product states resort by m = m₁+m₂ into a j=1 triplet and a j=0 singlet, so 2 × 2 = 3 + 1

The one label that survives the product intact is the additive one, $m=m_1+m_2$ , because $J_z=J_{1z}+J_{2z}$ . This is why the coefficients are block-diagonal in $m$ , and why the stretched state $|j_1 j_1\rangle\otimes|j_2 j_2\rangle$ is automatically the highest-weight state of the largest multiplet $j=j_1+j_2$ . Lowering operators walk down that ladder, and orthogonality within each fixed- $m$ subspace builds the smaller multiplets underneath — a taller version of the same staircase, as $1\otimes1$ shows below.

Figure

The 1 ⊗ 1 decomposition: nine product states with m = m₁+m₂ regroup into j = 2, 1, 0 ladders of 5, 3, 1 states, so 3 × 3 = 9 = 5 + 3 + 1

The core equations

The starting object is the tensor-product representation itself: the group element acts on both factors at once, so its matrix on $V_1\otimes V_2$ is the Kronecker product of the two factor matrices. Nothing here is optional — this diagonal action is what "rotating the whole assembly" means.

\big(D^{(1)}\otimes D^{(2)}\big)(g)\,(v_1\otimes v_2)=\big(D^{(1)}(g)v_1\big)\otimes\big(D^{(2)}(g)v_2\big) \qquad (1)

where: $D^{(i)}(g)$ — representation matrices of $g\in G$ on $V_i$ ; $v_i$ — a vector in $V_i$ ; $\otimes$ — the tensor (Kronecker) product. Reading: the same $g$ turns both factors at once, so a single conserved total charge labels the result. Because the trace of a Kronecker product factorizes, the character of the product is simply the product of the characters — the single fact that makes the decomposition computable.

\chi^{(1\otimes 2)}(g)=\chi^{(1)}(g)\,\chi^{(2)}(g) \qquad (2)

where: $\chi^{(i)}(g)=\mathrm{Tr}\,D^{(i)}(g)$ — the character of representation $i$ . Reading: multiply two known character rows pointwise and you hold the character of the product, ready to project onto irreducibles. Every representation of a compact or finite group is completely reducible, so the product is a direct sum of irreducibles with nonnegative integer multiplicities — the Clebsch–Gordan series.

D^{(1)}\otimes D^{(2)}=\bigoplus_\lambda m_\lambda\,D^{(\lambda)} \qquad (3)

where: $D^{(\lambda)}$ — the irreducible representations of $G$ ; $m_\lambda$ — the nonnegative integer multiplicity of $D^{(\lambda)}$ . Reading: the product is a list of which irreducibles appear and how often. The multiplicities are extracted by projecting the product character against each irreducible character via the first orthogonality theorem.

m_\lambda=\frac{1}{|G|}\sum_{g\in G}\chi^{(\lambda)}(g)^{*}\,\chi^{(1)}(g)\,\chi^{(2)}(g) \qquad (4)

where: $|G|$ — the order of a finite group; $\chi^{(\lambda)}(g)^{*}$ — the complex conjugate of the irreducible character. Reading: for a compact group replace $\tfrac{1}{|G|}\sum_g$ by $\int\mathrm{d}\mu(g)$ with normalized Haar measure; otherwise identical. For angular momentum the coupled eigenstates are assembled from the uncoupled product states, and the expansion coefficients are, by definition, the Clebsch–Gordan coefficients.

|j\,m\rangle=\sum_{m_1,m_2}\langle j_1 m_1\,j_2 m_2\,|\,j\,m\rangle\;|j_1 m_1\rangle\otimes|j_2 m_2\rangle \qquad (5)

where: $|j\,m\rangle$ — a coupled total-angular-momentum eigenstate; $\langle j_1 m_1\,j_2 m_2|j\,m\rangle$ — a Clebsch–Gordan coefficient; $|j_1 m_1\rangle\otimes|j_2 m_2\rangle$ — an uncoupled product state. Reading: the coefficients are the overlaps $\langle\text{uncoupled}|\text{coupled}\rangle$ , entries of a unitary matrix. Specializing (4) to $SU(2)$ collapses to the triangle rule: every total $j$ from $|j_1-j_2|$ to $j_1+j_2$ appears exactly once.

j_1\otimes j_2=\bigoplus_{j=|j_1-j_2|}^{j_1+j_2} j \qquad (6)

where: $j_1,j_2$ — the angular momenta of the two factors; $j$ — the total, stepping in integers. Reading: the range runs between the extremes $|j_1-j_2|$ and $j_1+j_2$ with no repeats — the $SU(2)$ product is multiplicity-free. The reorganization conserves the number of states, giving a dimension identity — the fastest check on any decomposition.

(2j_1+1)(2j_2+1)=\sum_{j=|j_1-j_2|}^{j_1+j_2}(2j+1) \qquad (7)

where: $2j+1$ — the dimension of the multiplet $j$ . Reading: if the coupled dimensions do not sum to the product dimension, the decomposition is wrong. Two constraints kill most coefficients up front: the additive $m$ -rule and the triangle inequality.

\langle j_1 m_1\,j_2 m_2|j\,m\rangle=0 \quad\text{unless}\quad m=m_1+m_2 \ \text{and}\ |j_1-j_2|\le j\le j_1+j_2 \qquad (8)

where: $m$ — the total magnetic quantum number; $m_1,m_2$ — the factor magnetic quantum numbers. Reading: only terms with $m_1+m_2=m$ contribute, so the coefficient array is block-diagonal in $m$ . Finally, because the change of basis is unitary, the coefficients obey an orthogonality (completeness) relation.

\sum_{m_1,m_2}\langle j\,m|j_1 m_1\,j_2 m_2\rangle\langle j_1 m_1\,j_2 m_2|j'm'\rangle=\delta_{jj'}\delta_{mm'} \qquad (9)

where: $\delta$ — the Kronecker delta. Reading: the coupled basis is orthonormal, and inverting (5) transposes the same real numbers back onto the product states.

Derivation AThe character product law and the multiplicity formula

The goal is to justify the two equations that drive every decomposition — that the character of a tensor product is the product of characters, and that character orthogonality then extracts each multiplicity — from nothing but the definition of the Kronecker product and the trace. Once these two lines are in hand, decomposing any product for a finite group is mechanical arithmetic on a character table, and the argument transplants unchanged to compact groups by swapping the group sum for a Haar integral.

Step 1. Index the product space by pairs: the matrix element of $D^{(1)}\otimes D^{(2)}$ between $(a,b)$ and $(c,d)$ is $\big(D^{(1)}\otimes D^{(2)}\big)_{(ab),(cd)}=D^{(1)}_{ac}\,D^{(2)}_{bd}$ . Why: the two factors act on independent indices, so the combined matrix is their outer product — the definition of the Kronecker product.

Step 2. Take the trace by setting $c=a$ , $d=b$ and summing: $\chi^{(1\otimes2)}(g)=\sum_{a,b}D^{(1)}_{aa}D^{(2)}_{bb}=\big(\sum_a D^{(1)}_{aa}\big)\big(\sum_b D^{(2)}_{bb}\big)=\chi^{(1)}(g)\,\chi^{(2)}(g)$ . Why: the double diagonal sum factorizes because $a$ and $b$ run independently — equation (2), and the reason products are computable.

Step 3. Expand the (completely reducible) product in irreducibles, $\chi^{(1)}\chi^{(2)}=\sum_\lambda m_\lambda\chi^{(\lambda)}$ , since characters of a direct sum add. Why: complete reducibility guarantees such an expansion exists with nonnegative integer $m_\lambda$ ; the character is a fingerprint that determines the representation up to equivalence.

Step 4. Project with the first orthogonality theorem $\tfrac{1}{|G|}\sum_g\chi^{(\mu)*}\chi^{(\lambda)}=\delta_{\mu\lambda}$ to isolate one coefficient, giving equation (4), $m_\lambda=\tfrac{1}{|G|}\sum_g\chi^{(\lambda)*}\chi^{(1)}\chi^{(2)}$ . Why: orthonormal characters act like an orthonormal basis, so the multiplicity is the inner product of the product character with $\chi^{(\lambda)}$ ; for compact groups $\tfrac{1}{|G|}\sum_g\to\int\mathrm{d}\mu(g)$ and the identical projection holds.

Go deeper · Derivation B · 1 minThe SU(2) coupling rule by multiplying characters

Derivation BThe SU(2) coupling rule by multiplying characters

The aim is to prove the triangle rule (6) — that $j_1\otimes j_2$ contains every total $j$ from $|j_1-j_2|$ to $j_1+j_2$ exactly once — directly from the $SU(2)$ character, without computing a single coefficient. The strategy is to write each character as a finite sum of weight phases, multiply, and watch the double sum reorganize into a telescoping stack of single-multiplet characters; the range and unit multiplicity both fall out of the bookkeeping.

Step 1. On $SU(2)$ every element is conjugate to a rotation by angle $\theta$ , with character $\chi_j(\theta)=\sum_{m=-j}^{j}e^{im\theta}=\dfrac{\sin\big((2j+1)\theta/2\big)}{\sin(\theta/2)}$ . Why: the character depends only on the rotation angle (a class function), and the sum of weight phases is a geometric series closing into the sine ratio.

Step 2. Multiply the two characters into a double sum over uncoupled weights, $\chi_{j_1}\chi_{j_2}=\sum_{m_1=-j_1}^{j_1}\sum_{m_2=-j_2}^{j_2}e^{i(m_1+m_2)\theta}$ . Why: this product is, by equation (2), exactly the character of $j_1\otimes j_2$ , now displayed as a grid of phases indexed by $(m_1,m_2)$ — the same grid drawn in the figure above.

Step 3. Collect terms by total weight $M=m_1+m_2$ : the coefficient of $e^{iM\theta}$ is $n(M)$ , the number of grid cells on the diagonal $m_1+m_2=M$ , which grows by one per step from $M=j_1+j_2$ inward and forms a symmetric plateau. Why: this staircase of degeneracies is the raw weight content that must be reassembled into whole multiplets.

Step 4. Peel off multiplets from the top: the largest weight $M=j_1+j_2$ occurs once, forcing a $j=j_1+j_2$ multiplet; subtract its character $\chi_{j_1+j_2}$ , the new top weight is $j_1+j_2-1$ , and so on down to $|j_1-j_2|$ . Why: each subtraction removes one unit from the height of the remaining staircase, so every $j$ in the range appears with multiplicity one — establishing (6) and the multiplicity-free property.

Go deeper · Derivation C · 1 minSinglet and triplet from two spin-½ systems

Derivation CSinglet and triplet from two spin-½ systems

The purpose is to turn the abstract rule $\tfrac12\otimes\tfrac12=1\oplus0$ into the four explicit coupled states and read the Clebsch–Gordan coefficients straight off — the smallest nontrivial case, and the template for larger couplings. Write $|{\uparrow}\rangle=|\tfrac12,\tfrac12\rangle$ and $|{\downarrow}\rangle=|\tfrac12,-\tfrac12\rangle$ ; the total lowering operator is $J_-=J_{1-}+J_{2-}$ with $J_-|j\,m\rangle=\sqrt{j(j+1)-m(m-1)}\,|j\,m{-}1\rangle$ .

Step 1. The unique state of maximal weight $m=1$ is $|{\uparrow\uparrow}\rangle$ , so it must be the highest-weight state of the largest multiplet: $|1,1\rangle=|{\uparrow\uparrow}\rangle$ . Why: nothing else has $m=1$ , and the stretched product state always tops the $j=j_1+j_2$ ladder.

Step 2. Apply $J_-$ to both sides. The left gives $J_-|1,1\rangle=\sqrt{2}\,|1,0\rangle$ ; the right gives $(J_{1-}+J_{2-})|{\uparrow\uparrow}\rangle=|{\downarrow\uparrow}\rangle+|{\uparrow\downarrow}\rangle$ . Why: lowering the coupled state and the product state must agree because $J_-$ is the same operator written in two bases.

Step 3. Divide by $\sqrt2$ to identify $|1,0\rangle=\tfrac{1}{\sqrt2}\big(|{\uparrow\downarrow}\rangle+|{\downarrow\uparrow}\rangle\big)$ , so $\langle\tfrac12\,\tfrac12;\tfrac12\,{-}\tfrac12|1\,0\rangle=\langle\tfrac12\,{-}\tfrac12;\tfrac12\,\tfrac12|1\,0\rangle=\tfrac{1}{\sqrt2}$ . Why: the symmetric combination of the two $m=0$ product states is the middle rung of the triplet, with equal weights forced by the lowering step.

Step 4. One more lowering gives $|1,-1\rangle=|{\downarrow\downarrow}\rangle$ , and the state orthogonal to $|1,0\rangle$ in the two-dimensional $m=0$ subspace is $|0,0\rangle=\tfrac{1}{\sqrt2}\big(|{\uparrow\downarrow}\rangle-|{\downarrow\uparrow}\rangle\big)$ . Why: the antisymmetric $m=0$ combination is annihilated by $J_+$ and $J_-$ , so it is a $j=0$ singlet; the minus-sign placement is fixed by the Condon–Shortley convention, and the dimension check $2\times2=3+1$ confirms the four states span the product space.

Limiting cases

The decomposition specializes cleanly in several regimes, and each limit is worth memorizing as a check on the general machinery:

Case	Result	Meaning
Coupling with the trivial rep	$j\otimes 0=j$	tensoring with the one-dimensional trivial representation changes nothing — it is the identity of the representation ring
Two spin-½ systems	$\tfrac12\otimes\tfrac12=1\oplus0$	symmetric triplet ( $j=1$ , 3 states) plus antisymmetric singlet ( $j=0$ , 1 state); $2\times2=3+1$
Two equal momenta	$j\otimes j\supset 0$	always contains a rotational invariant, the singlet $\tfrac{1}{\sqrt{2j+1}}\sum_m(-1)^{j-m}\|j\,m\rangle\|j\,{-}m\rangle$
Orbital plus spin	$\ell\otimes\tfrac12=(\ell+\tfrac12)\oplus(\ell-\tfrac12)$	the two fine-structure levels $j=\ell\pm\tfrac12$ of a one-electron atom
Quark–antiquark, $SU(3)$	$\mathbf{3}\otimes\bar{\mathbf{3}}=\mathbf{8}\oplus\mathbf{1}$	the meson octet plus singlet; dimension $3\times3=8+1$ , the larger-group analogue of the same idea

Common trapTreating a product of irreducibles as a single big irreducible

The single most common error is to imagine that $j_1\otimes j_2$ is itself one irreducible representation of dimension $(2j_1+1)(2j_2+1)$ carrying "total spin $j_1+j_2$ ." It is not. The product is reducible, and $j_1+j_2$ is only its largest piece: the space splits into every $j$ from $|j_1-j_2|$ up to $j_1+j_2$ , each a genuine invariant subspace. Two spin-1 particles do not make a single spin-2 object — they make a spin-2, a spin-1, and a spin-0, all at once ( $3\times3=5+3+1$ ). A second, related trap is to forget the additive rule: a Clebsch–Gordan coefficient vanishes unless $m=m_1+m_2$ , so summing over the wrong magnetic quantum numbers, or expecting a nonzero overlap when $m\ne m_1+m_2$ , produces nonsense. When a decomposition looks wrong, check the dimension identity $(2j_1+1)(2j_2+1)=\sum_j(2j+1)$ first — it catches almost every slip.

Worked examples

1 · Coupling j₁ = 1 and j₂ = ½ (fully worked)

A p-orbital electron has $\ell=1$ and spin $s=\tfrac12$ . Build the coupled total-angular-momentum states of the $m=\tfrac12$ sector and read off the Clebsch–Gordan coefficients, using the ladder method.

Step 1. The rule (6) gives $1\otimes\tfrac12=\tfrac32\oplus\tfrac12$ , and the dimension check is $3\times2=6=4+2$ — a $j=\tfrac32$ quartet and a $j=\tfrac12$ doublet. Start from the stretched state $|\tfrac32,\tfrac32\rangle=|1,1\rangle|\tfrac12,\tfrac12\rangle$ , the unique $m=\tfrac32$ state.

Step 2. Apply $J_-=J_{1-}+J_{2-}$ . The coupled side gives $J_-|\tfrac32,\tfrac32\rangle=\sqrt{\tfrac{15}{4}-\tfrac32\cdot\tfrac12}\,|\tfrac32,\tfrac12\rangle=\sqrt{3}\,|\tfrac32,\tfrac12\rangle$ . The product side gives $J_{1-}|1,1\rangle=\sqrt2\,|1,0\rangle$ and $J_{2-}|\tfrac12,\tfrac12\rangle=|\tfrac12,-\tfrac12\rangle$ , so the result is $\sqrt2\,|1,0\rangle|\tfrac12,\tfrac12\rangle+|1,1\rangle|\tfrac12,-\tfrac12\rangle$ .

Step 3. Divide by $\sqrt3$ : $|\tfrac32,\tfrac12\rangle=\sqrt{\tfrac23}\,|1,0\rangle|\tfrac12,\tfrac12\rangle+\sqrt{\tfrac13}\,|1,1\rangle|\tfrac12,-\tfrac12\rangle$ . The Clebsch–Gordan coefficients are $\langle1,0;\tfrac12,\tfrac12|\tfrac32,\tfrac12\rangle=\sqrt{2/3}\approx\mathbf{0.816}$ and $\langle1,1;\tfrac12,-\tfrac12|\tfrac32,\tfrac12\rangle=\sqrt{1/3}\approx\mathbf{0.577}$ .

Step 4. The $m=\tfrac12$ subspace is two-dimensional, spanned by the same two product states, so $|\tfrac12,\tfrac12\rangle$ is the orthogonal, normalized combination with the Condon–Shortley sign (largest- $m_1$ coefficient positive): $|\tfrac12,\tfrac12\rangle=\sqrt{\tfrac23}\,|1,1\rangle|\tfrac12,-\tfrac12\rangle-\sqrt{\tfrac13}\,|1,0\rangle|\tfrac12,\tfrac12\rangle$ . Check: the overlap with $|\tfrac32,\tfrac12\rangle$ is $\sqrt{\tfrac13}\sqrt{\tfrac23}-\sqrt{\tfrac23}\sqrt{\tfrac13}=0$ , norm $\tfrac23+\tfrac13=1$ . These are exactly the tabulated $1\times\tfrac12$ coefficients, and they turn radial wavefunctions into the spinor spherical harmonics of the Dirac hydrogen atom.

2 · Decomposing 1 ⊗ 1 and its invariant singlet

Decompose the product of two spin-1 systems, verify the dimension count, classify the pieces by exchange symmetry, and write the rotational invariant explicitly.

Step 1. By (6), $1\otimes1=2\oplus1\oplus0$ , matching the second figure. Dimension check: $3\times3=9=5+3+1$ . The product is not a single spin-2 object — it is a spin-2 quintet, a spin-1 triplet, and a spin-0 singlet simultaneously.

Step 2. Classify by exchange of the two identical factors: $j=2$ is symmetric (5 states), $j=1$ antisymmetric (3), $j=0$ symmetric (1). This symmetric/antisymmetric/symmetric alternation as $j$ steps down is a general feature of $j\otimes j$ , and via the Pauli principle it dictates which combined states identical particles may occupy.

Step 3. Write the invariant singlet from the general formula $|0,0\rangle=\tfrac{1}{\sqrt{2j+1}}\sum_m(-1)^{j-m}|j\,m\rangle|j\,{-}m\rangle$ with $j=1$ : $|0,0\rangle=\tfrac{1}{\sqrt3}\big(|1,1\rangle|1,-1\rangle-|1,0\rangle|1,0\rangle+|1,-1\rangle|1,1\rangle\big)$ . This state is annihilated by every component of $\mathbf{J}=\mathbf{J}_1+\mathbf{J}_2$ , so it is the unique rotational invariant in $1\otimes1$ — the abstract cousin of the scalar $\mathbf{A}\cdot\mathbf{B}$ of two vectors.

Physics of the portrait

The Clebsch–Gordan coefficients are a rare case of pure nineteenth-century algebra that reappeared, decades later, as the backbone of quantum mechanics — carrying with them the names of two mathematicians who never met an electron.

Origins · 1 minClebsch–Gordan coefficients and how this was discovered

Clebsch–Gordan coefficients

invariant theory 1872 → quantum angular momentum 1931 · the coupling algebra of SU(2) and its irreducible representations

The coefficients are named for Alfred Clebsch (1833–1872) and Paul Gordan (1837–1912), who introduced them not for physics but for the invariant theory of binary forms — homogeneous polynomials in two variables under linear substitutions. In the 1860s and early 1870s they worked out how a product of two such forms decomposes into irreducible pieces, and the numerical coefficients in that expansion are what we now call Clebsch–Gordan coefficients. For half a century they lived quietly inside algebra. Then quantum mechanics arrived: the states of a rotating system are the irreducible representations of the rotation group, and combining two systems means decomposing a product of representations — the very problem Clebsch and Gordan had solved. Eugene Wigner made the connection explicit in his 1931 book Gruppentheorie und ihre Anwendung auf die Quantenmechanik der Atomspektren, turning the coefficients into the standard tool for adding angular momenta and later introducing the symmetric $3j$ symbols. Condon and Shortley fixed the universal sign convention in their 1935 Theory of Atomic Spectra, and Racah built the recoupling algebra (the $6j$ and $9j$ symbols) for three or more momenta. What began as a device for classifying polynomials is now the machinery behind atomic spectra, nuclear structure, and the selection rules of every spectroscopy.

History

1868

— Gordan proves the finiteness theorem for invariants of binary forms

1872

— Clebsch's theory of binary algebraic forms gives the series for products

1931

— Wigner's book brings the coefficients into quantum angular momentum

1935

— Condon and Shortley fix the standard sign convention

1942

— Racah builds the recoupling algebra of tensor operators

Going further

Texts. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., Ch. 3 (addition of angular momenta and Clebsch–Gordan coefficients) · Stone & Goldbart, Mathematics for Physics, Ch. on group representations and angular momentum · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., group-theory chapter · Tinkham, Group Theory and Quantum Mechanics, §5 · Georgi, Lie Algebras in Particle Physics, 2nd ed. (the $SU(3)$ tensor products) · Edmonds, Angular Momentum in Quantum Mechanics (the definitive coefficient tables and phase conventions).

Provenance. The tensor-product representation, the character product law $\chi^{(1\otimes2)}=\chi^{(1)}\chi^{(2)}$ , and the multiplicity formula via character orthogonality are standard finite-group results (Tinkham; Arfken–Weber–Harris) — verify the multiplicity formula against any character table. The $SU(2)$ character $\chi_j(\theta)=\sin((2j+1)\theta/2)/\sin(\theta/2)$ and the triangle rule follow from the elementary trigonometric product law derived above. The spin-½ singlet/triplet states, the $1\times\tfrac12$ coefficients $\sqrt{2/3}$ and $\sqrt{1/3}$ , and the $1\otimes1$ singlet are the tabulated Condon–Shortley values (Sakurai Ch. 3; Edmonds), reproduced here by the ladder method. The dimension identity $(2j_1+1)(2j_2+1)=\sum_j(2j+1)$ is a direct algebraic check, and the $SU(3)$ decompositions are from Georgi.

Prerequisites. Group Representations · Reducible and Irreducible Representations · Characters and Orthogonality Theorems · SU(2) and the Spin Double Cover · Schur's Lemmas. Leads to. The Wigner–Eckart Theorem · Character Tables and Decomposition of Representations · The Rotation Group SO(3).

Ask the tutor: Ask me to hand-build the coupled states for any $j_1\otimes j_2$ by the ladder method, or to show how one Clebsch–Gordan coefficient turns into a spectroscopic selection rule through the Wigner–Eckart theorem.

Why does a product of two irreducible representations fail to be irreducible?

Irreducibility means a single space has no invariant subspaces, but the product $V_1\otimes V_2$ is large and generically has many. Concretely, the highest-weight state $|j_1 j_1\rangle|j_2 j_2\rangle$ generates a whole $j=j_1+j_2$ ladder under lowering operators, yet that ladder does not exhaust the space — leftover states at each intermediate weight generate their own smaller ladders. Each ladder is an invariant subspace, so the product is a direct sum of several irreducibles rather than one. Only trivial cases, such as tensoring with the one-dimensional trivial representation, leave an irreducible irreducible.

What is the difference between the coupled and uncoupled bases?

The uncoupled basis $|j_1 m_1\rangle|j_2 m_2\rangle$ diagonalizes the commuting set $J_1^2, J_{1z}, J_2^2, J_{2z}$ — you know each subsystem's projection separately. The coupled basis $|j\,m\rangle$ diagonalizes $J_1^2, J_2^2, J^2, J_z$ — you know the total angular momentum and its projection, but not the individual $m_1, m_2$ . Both are complete orthonormal bases of the same space, connected by the Clebsch–Gordan matrix. Which you use is dictated by the physics: uncoupled states when the subsystems are independent, coupled states when an interaction like spin–orbit coupling depends on the total.

Why is m = m₁ + m₂ but j is not simply j₁ + j₂?

Because $J_z=J_{1z}+J_{2z}$ is a sum of commuting operators, its eigenvalues add: $m=m_1+m_2$ exactly. But $J^2=(\mathbf{J}_1+\mathbf{J}_2)^2=J_1^2+J_2^2+2\mathbf{J}_1\cdot\mathbf{J}_2$ contains the cross term $2\mathbf{J}_1\cdot\mathbf{J}_2$ , whose components do not commute with $J_{1z}$ or $J_{2z}$ individually. The magnitudes therefore do not simply add; instead the total $j$ ranges from $|j_1-j_2|$ to $j_1+j_2$ , corresponding to the two subsystem vectors anywhere from anti-aligned to aligned. The additive quantum number is special precisely because it comes from a commuting sum.

Are Clebsch–Gordan coefficients always real?

In the standard Condon–Shortley convention, yes — every $SU(2)$ Clebsch–Gordan coefficient is real, which is why the same numbers appear whether you expand coupled states in uncoupled ones or the reverse. This is a convention choice, not a theorem: it fixes the arbitrary phases of the coupled states by requiring that the coefficient with the largest $m_1$ in each stretched-down state be real and positive. Condon–Shortley is now universal, so tabulated coefficients and library routines agree. For larger groups like $SU(3)$ the analogous coefficients can be complex, and extra labels (isoscalar factors, multiplicity indices) are needed.

How does this connect to the Wigner–Eckart theorem?

Directly — the Wigner–Eckart theorem is the payoff. A spherical tensor operator $T^{(k)}_q$ transforms under rotations like an angular-momentum- $k$ multiplet, so its matrix element $\langle j'm'|T^{(k)}_q|j m\rangle$ is an overlap between the coupled product $|j m\rangle\otimes|k q\rangle$ and $|j'm'\rangle$ . The theorem states this factorizes as a Clebsch–Gordan coefficient $\langle j m\,k q|j'm'\rangle$ times a reduced matrix element $\langle j'\|T^{(k)}\|j\rangle$ independent of $m, m', q$ . All the orientation dependence — every selection rule, every relative intensity within a multiplet — lives in the coefficient, while the reduced element holds the dynamics. It is the single most useful consequence of the decomposition.

Does the decomposition still work for non-compact or infinite groups?

The clean statement — complete reducibility into irreducibles with finite multiplicities — relies on the group being compact (or finite), where averaging with an invariant measure guarantees every representation breaks into a discrete sum of finite-dimensional irreducibles. For non-compact groups like the Lorentz or Poincaré group, unitary irreducibles are infinite-dimensional and tensor products generally decompose into direct integrals rather than direct sums, with continuous labels. The spirit survives — products still decompose, and Clebsch–Gordan-type coefficients still exist — but the analysis is far more delicate. For the rotation groups $SU(2)$ and $SO(3)$ that dominate atomic and nuclear physics, the compact, discrete picture of this article is exact.