Clebsch-Gordan Coefficients

Addition of Angular Momenta proved the theorem that tells you which total angular momenta appear when two subsystems couple: every $J$ from $|j_1-j_2|$ to $j_1+j_2$ , exactly once. It even showed the mechanism on one example — starting from the stretched state of $1\otimes\tfrac12$ and lowering once. What it deliberately did not build is the machinery: the actual numbers $\langle j_1 m_1; j_2 m_2|J M\rangle$ that convert between the product basis and the coupled basis, for any pair of momenta, reproducibly, with signs everyone on Earth agrees on. This article is that machinery. The numbers are the Clebsch-Gordan coefficients, and three facts tame the infinite table of them. First, most vanish: $M=m_1+m_2$ exactly, and the triangle rule gates $J$ . Second, the survivors are generated by a two-term recursion — sandwich $\hat{J}_\pm=\hat{J}_{1\pm}+\hat{J}_{2\pm}$ between the right states and every coefficient follows from the one above it. Third, one convention — Condon-Shortley — nails every leftover phase, making the whole table real and reproducible from a single positivity rule. With those three facts you can regenerate any table in a textbook's endpapers, and this article does: the complete $1\otimes\tfrac12$ table (extending the previous article's single lowering step to all six entries) and the $1\otimes 1$ table, every entry machine-verified. The payoff reaches far beyond bookkeeping. These same coefficients are the geometric factor in the Wigner-Eckart theorem, the skeleton of every selection rule, and — symmetrized into 3-j symbols — the alphabet of atomic and nuclear structure. Learn to read a CG table here; the next four articles spend it.

Key takeaways

Clebsch-Gordan coefficients $\langle j_1 m_1; j_2 m_2|J M\rangle$ are the overlap amplitudes between the uncoupled product basis and the coupled $|J M\rangle$ basis — the entries of a real orthogonal change-of-basis matrix (in the Condon-Shortley convention).
Selection rules kill most of the table: $\langle\cdots\rangle=0$ unless $M=m_1+m_2$ and $|j_1-j_2|\le J\le j_1+j_2$ (the triangle rule, proven in Addition of Angular Momenta).
The generator: a two-term recursion from sandwiching $\hat{J}_\pm$ , which walks down each $J$ -ladder from the stretched state $|j_1 j_1\rangle|j_2 j_2\rangle$ .
Condon-Shortley: $\langle j_1\, j_1; j_2\, (J-j_1)|J\,J\rangle>0$ for every $J$ — this single rule makes all coefficients real and every table unique.
Orthogonality both ways: rows and columns of the CG matrix are orthonormal — completeness of either basis, in coefficient form.
Workhorse tables built and verified here: $1\otimes\tfrac12$ (all ten entries) and $1\otimes 1$ ; e.g. $\langle 1\,1;\tfrac12\,\tfrac12{-}\tfrac12|\tfrac32\,\tfrac12\rangle=1/\sqrt3$ and $\langle 1\,0;1\,0|1\,0\rangle=0$ .
Key idea: two conservation laws plus one recursion plus one phase convention generate every Clebsch-Gordan table — the coefficients are forced, not chosen.

The core insight

The coupled and uncoupled bases are two orthonormal eigenbases of the same Hilbert space $\mathcal{H}_{j_1}\otimes\mathcal{H}_{j_2}$ ; CG coefficients are the unitary (real, in Condon-Shortley) matrix elements relating them, constrained by $M=m_1+m_2$ and the triangle rule, and computed by ladder-operator descent plus orthogonalization.

Where this lives today

Every quantum-chemistry code, every nuclear-structure calculation, and every atomic-clock design query CG coefficients thousands of times per run — through library calls (`sympy.physics.quantum.cg`, `WignerSymbols.jl`, GSL) that implement exactly the recursion-plus-CS machinery of this article. In quantum computing, coupling maps qubit registers into total-spin sectors: the decoherence-free subspaces used in trapped-ion experiments are $J$ -ladders of many coupled spins, located by CG bookkeeping. And in particle physics, the same tables run isospin: the famous factor relating $\Delta^{++}\to p\pi^+$ to $\Delta^0\to n\pi^0$ rates is a ratio of two of the $\tfrac32$ -table entries you verified here — SU(2) does not care whether it rotates space or isospace.

The physical picture

Picture the product space $\mathcal{H}_{j_1}\otimes\mathcal{H}_{j_2}$ as a grid of cells, one per $(m_1,m_2)$ pair. The coupled state $|J M\rangle$ lives on the anti-diagonal $m_1+m_2=M$ : it is a superposition over exactly the cells on that diagonal, and the CG coefficients are the amplitudes it assigns to each. Conservation of $\hat{J}_z=\hat{J}_{1z}+\hat{J}_{2z}$ is what pins the state to its diagonal; how the amplitude distributes along the diagonal is what the recursion computes.

Figure

The CG coefficient lives on the M-diagonal of the (m₁, m₂) grid

Now stack the diagonals. The $M=j_1+j_2$ diagonal has one cell — the stretched state, shared verbatim by both bases. The next diagonal has two cells but must host two coupled states ( $J=j_1+j_2$ and $J=j_1+j_2-1$ ); the recursion fixes the first, orthogonality plus one sign convention fixes the second. Walking down diagonal by diagonal, the whole table is forced. That is worth internalizing: there is nothing adjustable in a CG table except one global phase per $J$ -ladder, and Condon-Shortley spends exactly that freedom.

Figure

Every table is forced: lower from the stretched state, orthogonalize, fix the sign

The object and its selection rules

Both bases span the same $(2j_1+1)(2j_2+1)$ -dimensional space, so either expands in the other. Writing the coupled states in the product basis defines the coefficients — this is just the expansion theorem of Basis Expansions and Components applied with intent:

|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|j_2 m_2\rangle \qquad (1)

where: the sum looks double but collapses to a single diagonal: applying $\hat{J}_z=\hat{J}_{1z}+\hat{J}_{2z}$ to both sides forces every surviving term to satisfy $M=m_1+m_2$ . The labels $(j_1 j_2)$ on the coupled ket are suppressed but real — a CG coefficient is meaningless until all four angular momenta $(j_1,j_2,J)$ and the magnetic labels are stated. The second gate is the triangle rule,

\langle j_1 m_1; j_2 m_2|J\,M\rangle = 0 \quad\text{unless}\quad |j_1-j_2|\;\le\; J \;\le\; j_1+j_2 \qquad (2)

where: this is not a new fact — it is the multiplicity theorem of Addition of Angular Momenta, consumed here as a vanishing rule: no coupled ladder with $J$ outside the window exists, so nothing to overlap with. Each allowed $J$ appears exactly once, which is why a CG coefficient needs no extra "which copy of $J$ " index for two-momentum coupling (three momenta break this — 3-j, 6-j and Racah Symbols takes over there). Because both bases are orthonormal and (with Condon-Shortley) the coefficients are real, the same numbers run the inverse expansion — reading a table sideways:

|j_1 m_1\rangle|j_2 m_2\rangle \;=\; \sum_{J=|j_1-j_2|}^{j_1+j_2} \langle j_1 m_1; j_2 m_2|J\,M\rangle\; |J\,M\rangle, \qquad M=m_1+m_2 \qquad (3)

where: the same coefficient appears — no transpose-conjugate bookkeeping survives once everything is real. Physically (3) answers the experimenter's question: prepare a product state, measure $\hat{J}^2$ — the probability of finding $J$ is $|\langle j_1 m_1; j_2 m_2|J M\rangle|^2$ . The unitarity behind this double duty is stated here and owned, as theory, by Coupled and Uncoupled Bases — the next article. Two orthogonality relations pack the change-of-basis consistency into coefficient form:

\sum_{m_1,m_2}\langle j_1 m_1; j_2 m_2|J M\rangle\langle j_1 m_1; j_2 m_2|J' M'\rangle=\delta_{JJ'}\delta_{MM'},\qquad \sum_{J,M}\langle j_1 m_1; j_2 m_2|J M\rangle\langle j_1 m_1'; j_2 m_2'|J M\rangle=\delta_{m_1 m_1'}\delta_{m_2 m_2'} \qquad (4)

where: the first says coupled states are orthonormal (rows), the second that product states are (columns). Both are verified exhaustively for $\tfrac12\otimes\tfrac12$ , $1\otimes\tfrac12$ and $1\otimes1$ in this article's verifier. They are also your fastest error detector: a hand-built table whose squared column doesn't sum to 1 has a wrong entry, guaranteed.

The recursion — where every table comes from

The stretched state costs nothing: at $M=j_1+j_2$ the diagonal has one cell, so

|J{=}j_1{+}j_2,\;M{=}j_1{+}j_2\rangle \;=\; |j_1\, j_1\rangle|j_2\, j_2\rangle \qquad (5)

where: its CG coefficient is $1$ — and even this hides a choice (why not $-1$ ?) that the Condon-Shortley convention formalizes. Everything below (5) follows from one identity. Sandwich the total lowering operator, acting left as $\hat{J}_-$ and right as $\hat{J}_{1-}+\hat{J}_{2-}$ , and the matrix elements of Ladder Operators and the Angular Momentum Spectrum (consumed, not re-derived) give the master recursion:

\sqrt{J(J{+}1)-M(M{-}1)}\;\langle m_1 m_2|J\,M{-}1\rangle = \sqrt{j_1(j_1{+}1)-m_1(m_1{+}1)}\;\langle m_1{+}1\, m_2|J M\rangle + \sqrt{j_2(j_2{+}1)-m_2(m_2{+}1)}\;\langle m_1\, m_2{+}1|J M\rangle \qquad (6)

where: the shorthand $\langle m_1 m_2|JM\rangle$ suppresses the fixed $j_1,j_2$ . Read it as a sourcing rule: the coefficient one rung down is fed by exactly two coefficients on the rung above — the cell to the $m_1$ -right and the cell to the $m_2$ -right. All square roots are the ladder normalizations, real and non-negative, so the recursion propagates reality: once the top of a ladder is real, the whole ladder is.

Derivation AThe two-term recursion that generates every CG table

We derive equation (6), the engine of the whole subject. The strategy is the standard sandwich: evaluate $\langle j_1 m_1; j_2 m_2|\hat{J}_-|J\,M\rangle$ twice — once letting $\hat{J}_-$ act rightward on the coupled ket, once splitting it as $\hat{J}_{1-}+\hat{J}_{2-}$ and acting leftward on the bra. Both evaluations consume the matrix-element theorem of Ladder Operators and the Angular Momentum Spectrum with its Condon-Shortley-fixed real positive coefficients; nothing new is derived about ladders here. The subtlety to watch is the leftward action: $\hat{J}_{1-}$ acting on a bra raises the bra's label.

Step 1. Rightward. $\hat{J}_-|J\,M\rangle=\hbar\sqrt{J(J{+}1)-M(M{-}1)}\,|J\,M{-}1\rangle$ , so the sandwich equals $\hbar\sqrt{J(J{+}1)-M(M{-}1)}\;\langle j_1 m_1; j_2 m_2|J\,M{-}1\rangle$ .

Step 2. Leftward. Split $\hat{J}_-=\hat{J}_{1-}+\hat{J}_{2-}$ and use $\langle j m|\hat{J}_-=\big(\hat{J}_+|j m\rangle\big)^\dagger$ : the bra $\langle j_1 m_1|$ hit by $\hat{J}_{1-}$ from the right becomes $\hbar\sqrt{j_1(j_1{+}1)-m_1(m_1{+}1)}\,\langle j_1\, m_1{+}1|$ , and likewise for the second factor. Why the labels go up: $\hat{J}_-$ between a bra and a ket is simultaneously a lowering operator for the ket and — by hermitian conjugation, since $\hat{J}_-^\dagger=\hat{J}_+$ — a raising operator for the bra. Missing this flip is the classic way to derive a wrong recursion with mismatched indices.

Step 3. Equate Step 1 and Step 2 and cancel $\hbar$ . The result is equation (6): each descent coefficient sourced by two parents. Starting data: the stretched coefficient (5) for the top ladder; for each lower ladder ( $J=j_1+j_2-1,\dots$ ), the top coefficient comes from orthogonality against all higher ladders on the same diagonal, with its overall sign fixed by Condon-Shortley (Derivation B). Why two terms and not more: $\hat{J}_-$ is a sum of exactly two subsystem operators, and each moves exactly one magnetic label by one. The recursion's sparseness is conservation of $M$ in action — it is also why CG tables can be generated by hand at all. *Verified: equation (6) checked symbolically on $1\otimes\tfrac12$ (at $J=\tfrac32$ ) and on $1\otimes1$ (at $J=2$ and $J=1$ ) against sympy's CG values — exact identities; full tables regenerated and matched entry-for-entry (article verifier).*

Condon-Shortley: one rule, every sign

The recursion (6) propagates real numbers down each ladder, but each ladder's top still enters with an arbitrary overall phase — orthogonality on a diagonal determines a unit vector only up to $\pm1$ (once the numbers are real). The Condon-Shortley convention spends this last freedom:

\langle j_1\, j_1;\; j_2\, (J-j_1)\;|\;J\,J\rangle \;>\; 0 \qquad\text{for every } J \qquad (7)

where: in words — in the top state of every $J$ -ladder, the component with the maximal $m_1$ is chosen positive. Combined with the real-propagating recursion, (7) makes every CG coefficient real, and every table on Earth identical. You have already met this convention twice, in other clothes: it is the same choice that makes the $\hat{J}_\pm$ matrix elements real and non-negative (Ladder Operators, its CS card), and the same $(-1)^m$ bookkeeping placement discussed for $P_\ell^m$ versus $Y_{\ell m}$ (Associated Legendre Functions). One convention, three faces.

Go deeper · Derivation B · 1 minWhy one positivity rule makes the whole table real and unique

Derivation BWhy one positivity rule makes the whole table real and unique

We prove that condition (7), plus the recursion, pins every coefficient. The strategy is induction down the diagonals: show that on each $M$ -diagonal, all coefficients are determined once the tops of all ladders are, and that each new top has exactly a sign ambiguity that (7) resolves. Watch how reality — not just uniqueness — rides along automatically.

Step 1. Top diagonal, $M=j_1+j_2$ : one cell, one coupled state; normalization forces coefficient $\pm1$ , and (7) with $J=j_1+j_2$ (where $J-j_1=j_2$ is the maximal $m_2$ too) picks $+1$ . This is the hidden choice in equation (5).

Step 2. Descend within ladders. Equation (6) computes every coefficient of $|J,M-1\rangle$ from the (already fixed, real) coefficients of $|J,M\rangle$ — the square-root sources are non-negative reals, so descent creates no new phases and no new choices.

Step 3. New ladder tops. On the diagonal $M=j_1+j_2-k$ , one new $J$ -value appears (as long as $J\ge|j_1-j_2|$ ; the diagonal has $k{+}1$ cells hosting $k{+}1$ states). Its top state must be a unit vector orthogonal to the $k$ already-fixed states on that diagonal: in a real $(k{+}1)$ -dimensional space, the orthogonal complement of $k$ independent real vectors is one-dimensional — determined up to $\pm1$ . Why the complement stays real: the fixed states have real components (Step 2), so orthogonality conditions are real linear equations; a real solution exists and any unit solution is $\pm$ it. Nothing forces complex numbers in — that is the deep convenience of SU(2) coupling.

Step 4. Spend (7): of the two sign choices, take the one making the maximal- $m_1$ component positive. One subtlety guards the rule: that component is never zero (it can be proven nonzero for the ladder-top of each allowed $J$ ), so the rule always bites. Descend by Step 2; repeat. The table is now unique — and wholly real. Verified: CS positivity swept over all ladder tops for $(j_1,j_2)\in\{(\tfrac12,\tfrac12),(1,\tfrac12),(1,1),(\tfrac32,1)\}$ — all stretched-side entries strictly positive; both orthogonality relations (4) exact on the three workhorse pairs (article verifier).

The workhorse tables

Running the machine on $1\otimes\tfrac12$ — an orbital $\ell=1$ electron picking up its spin, the doorway to fine structure — extends the single lowering step performed in Addition of Angular Momenta (its $|\tfrac32,\tfrac12\rangle$ mechanism exhibit, cited not re-run) to the complete table:

$\|J,M\rangle$	expansion in $\|1\,m_1\rangle\,\|\!\uparrow/\downarrow\rangle$
$\|\tfrac32,\pm\tfrac32\rangle$	$\|1\,{\pm1}\rangle\|\!\uparrow/\downarrow\rangle$ (stretched, coefficient $1$ )
$\|\tfrac32,+\tfrac12\rangle$	$\sqrt{\tfrac23}\,\|1\,0\rangle\|\!\uparrow\rangle+\sqrt{\tfrac13}\,\|1\,1\rangle\|\!\downarrow\rangle$
$\|\tfrac12,+\tfrac12\rangle$	$-\sqrt{\tfrac13}\,\|1\,0\rangle\|\!\uparrow\rangle+\sqrt{\tfrac23}\,\|1\,1\rangle\|\!\downarrow\rangle$
$\|\tfrac32,-\tfrac12\rangle$	$\sqrt{\tfrac23}\,\|1\,0\rangle\|\!\downarrow\rangle+\sqrt{\tfrac13}\,\|1\,{-1}\rangle\|\!\uparrow\rangle$
$\|\tfrac12,-\tfrac12\rangle$	$\sqrt{\tfrac13}\,\|1\,0\rangle\|\!\downarrow\rangle-\sqrt{\tfrac23}\,\|1\,{-1}\rangle\|\!\uparrow\rangle$

Read the signs with a physicist's eye: within each $J=\tfrac12$ state the two amplitudes carry opposite signs — that relative minus is what makes it orthogonal to its $J=\tfrac32$ partner on the same diagonal — and the Condon-Shortley rule decides which term gets the minus (never the maximal- $m_1$ one at the ladder top). Note the provenance anchor $\langle1\,1;\tfrac12\,{-\tfrac12}|\tfrac32\,\tfrac12\rangle=1/\sqrt3$ : squared, it says a stretched-orbital, flipped-spin product state is found in $J=\tfrac32$ one time in three. For $1\otimes1$ (two $p$ -electrons, or a rigid rotor absorbing a photon), the machine yields the $9=5+3+1$ decomposition. Highlights worth memorizing, all verified exactly:

\langle1\,0;1\,0|2\,0\rangle=\sqrt{\tfrac23},\qquad \langle1\,0;1\,0|1\,0\rangle=0,\qquad \langle1\,0;1\,0|0\,0\rangle=-\tfrac{1}{\sqrt3},\qquad \langle1\,1;1\,-1|1\,0\rangle=\tfrac{1}{\sqrt2} \qquad (8)

where: the zero is no accident. The exchange symmetry of CG coefficients, $\langle j_1 m_1;j_2 m_2|JM\rangle=(-1)^{j_1+j_2-J}\langle j_2 m_2;j_1 m_1|JM\rangle$ (verified in full sweep), makes the $J=1$ ladder of $1\otimes1$ antisymmetric under swapping the two subsystems — and the cell $(0,0)$ is symmetric, so its amplitude must vanish. Whole families of "mysterious" zero coefficients are this one symmetry argument in disguise.

Limiting cases

case	result	meaning
stretched state $M=J=j_1+j_2$	single coefficient $=1$	both bases share the corner state; no mixing possible
$j_2=0$	$\langle j_1 m_1;0\,0\|J M\rangle=\delta_{Jj_1}\delta_{Mm_1}$	coupling to nothing changes nothing — the identity table
$j_1=j_2$ , $J=$ odd-count ladder	swap-antisymmetric coefficients, symmetric cells vanish	e.g. $\langle1\,0;1\,0\|1\,0\rangle=0$ ; exchange symmetry $(-1)^{j_1+j_2-J}$
one $m$ maximal, one rung down	two-cell diagonal: coefficients $\sqrt{j_i/(j_1{+}j_2)}$ pattern	fastest hand-check of any table's second row
large $j_1,j_2$ (classical limit)	coefficients peak where vectors add head-to-tail	the vector-addition cartoon becomes exact; quantum spread around it shrinks

Common trapM is conserved, but m₁ and m₂ separately are not

The selection rule $M=m_1+m_2$ tempts a stronger, false reading — that a coupled state has definite $m_1$ and $m_2$ . It does not: $|J\,M\rangle$ is generically a superposition over all cells of its diagonal, and a measurement of $\hat{J}_{1z}$ on it gives several outcomes with CG-squared probabilities. Equally tempting and equally wrong: assuming $J$ is always $j_1+j_2$ . The whole window $|j_1-j_2|\le J\le j_1+j_2$ occurs, each once — which $J$ you have depends on the state, not on the ingredients. And when quoting any coefficient with its sign, state the convention: a table without "Condon-Shortley" attached is ambiguous up to phases per ladder.

Worked examples

Example 1 — reading the table as probabilities. An electron is prepared in the product state $|1\,1\rangle|\!\downarrow\rangle$ — stretched orbital, flipped spin. What does a measurement of total $\hat{J}^2$ give?

Step 1. Invert with equation (3) on the $M=\tfrac12$ diagonal: $|1\,1\rangle|\!\downarrow\rangle=\sqrt{\tfrac13}\,|\tfrac32,\tfrac12\rangle+\sqrt{\tfrac23}\,|\tfrac12,\tfrac12\rangle$ — the same two numbers as the table above, read down a column instead of across a row.

Step 2. Born rule: $P(J{=}\tfrac32)=\tfrac13$ , $P(J{=}\tfrac12)=\tfrac23$ , summing to 1 (column normalization — the second relation of (4) at work).

Step 3. Cross-check with the eigenvalue route: $\langle\hat{J}^2\rangle=\tfrac13\cdot\tfrac{15}{4}\hbar^2+\tfrac23\cdot\tfrac34\hbar^2=\tfrac{7}{4}\hbar^2$ . The identity $\hat{J}^2=\hat{J}_1^2+\hat{J}_2^2+2\hat{J}_1\!\cdot\!\hat{J}_2$ (from Addition of Angular Momenta) gives the same $\tfrac74\hbar^2$ directly since $\langle\hat{J}_1\cdot\hat{J}_2\rangle=\hbar^2\cdot1\cdot(-\tfrac12)\cdot\tfrac12\cdot2/2=-\tfrac14\hbar^2\cdot1$ — consistent. Answer: $J=\tfrac32$ with probability $\tfrac13$ ; $J=\tfrac12$ with probability $\tfrac23$ .

Example 2 — two p-electrons in the $|2,0\rangle$ state. Two electrons each with $\ell=1$ couple to total $L=2$ , $M_L=0$ (spin ignored here). What are the probabilities for the individual $(m_1,m_2)$ pairs?

Step 1. Expand on the $M=0$ diagonal of $1\otimes1$ with the verified coefficients: $|2,0\rangle=\tfrac{1}{\sqrt6}|1\,1\rangle|1\,{-1}\rangle+\sqrt{\tfrac23}\,|1\,0\rangle|1\,0\rangle+\tfrac{1}{\sqrt6}|1\,{-1}\rangle|1\,1\rangle$ .

Step 2. Square: $P(+1,-1)=P(-1,+1)=\tfrac16$ , $P(0,0)=\tfrac23$ . Total $=1$ . The symmetric weighting under swap is the $(-1)^{j_1+j_2-J}=+1$ exchange symmetry of the $J=2$ ladder, visible as physics.

Step 3. Contrast with $|1,0\rangle$ : there $P(0,0)=|\langle1\,0;1\,0|1\,0\rangle|^2=0$ — the antisymmetric ladder refuses the symmetric cell entirely. Same ingredients, different $J$ , dramatically different statistics. Answer: $(\tfrac16,\tfrac23,\tfrac16)$ for $|2,0\rangle$ ; the $(0,0)$ cell is strictly forbidden in $|1,0\rangle$ .

Origins · 1 minAlfred Clebsch and how this was discovered

Alfred Clebsch

1833–1872 · Karlsruhe, Giessen, Göttingen

Clebsch never saw a quantum. A Königsberg-trained mathematician, he worked through elasticity and hydrodynamics into the algebraic geometry of curves, and with Paul Gordan wrote the 1866 treatise on the invariant theory of binary forms — the algebra of how polynomial "forms" decompose when combined, whose expansion coefficients are, symbol for symbol, the coupling coefficients of SU(2). He co-founded Mathematische Annalen in 1868, built a school at Göttingen, and died of diphtheria at thirty-nine, sixty years before Wigner and Weyl found that his coefficients organize atomic spectra. The name "Clebsch-Gordan" was attached by the quantum generation reaching back for the right mathematics — a reminder that the recursion you just learned was first run on polynomials, not particles. Gordan's own, longer side of the story belongs to the next article.

History

1866

— Clebsch and Gordan publish Theorie der binären algebraischen Formen; the coupling series appears as pure invariant theory

1927

— Wigner's group-theory papers import the coefficients into quantum mechanics for atomic spectra

1931

— Wigner's Gruppentheorie book systematizes the tables; von Neumann-Wigner methods spread despite the "Gruppenpest" backlash

1935

— Condon and Shortley's The Theory of Atomic Spectra fixes the phase convention every modern table uses

1957

— Edmonds' Angular Momentum in Quantum Mechanics becomes the standard reference formulary

Going further

Sakurai & Napolitano, Modern Quantum Mechanics, Ch. 3 — the recursion (Eq. 3.8.49) and the coupling formalism at this article's register.
Griffiths & Schroeter §4.4 — the gentler on-ramp, with the $\tfrac12\otimes\tfrac12$ and $1\otimes\tfrac12$ tables worked longhand.
Edmonds, Angular Momentum in Quantum Mechanics (1957) — the standard formulary; Condon & Shortley (1935) for the convention's source.
Cohen-Tannoudji, Diu & Laloë, Vol. 2, Ch. X — complements with the full orthogonalization bookkeeping.

Provenance note: Selection rules, triangle inequality, dimension count, and the two-spin singlet/triplet are standard (Griffiths Sec. 4.4; Sakurai Ch. 3; Cohen-Tannoudji Vol. 2 Ch. X). The ladder recursion matches Sakurai Eq. (3.8.49); Condon-Shortley convention per Condon & Shortley (1935) and Edmonds (1957). Verifier should: (1) confirm $\sum_J(2J+1)=(2j_1+1)(2j_2+1)$ ; (2) regenerate the $1\otimes\tfrac12$ CG table and check $\langle1\,1;\tfrac12,-\tfrac12|\tfrac32,\tfrac12\rangle=1/\sqrt3$ ; (3) verify $S^2$ eigenvalues on singlet/triplet; (4) cross-check the 3-j relation phase $(-1)^{j_1-j_2+M}\sqrt{2J+1}$ against PDG tables. — All four verifier items are executed in this article's verify script except (3) and (4), which are owned, per the chapter map, by Singlet and Triplet States (the base's two-spin derivation is relocated there) and 3-j, 6-j and Racah Symbols (the 3-j phase audit) respectively. This article extends the one-step $1\otimes\tfrac12$ mechanism exhibit of Addition of Angular Momenta to the full table, as banked there.

Do I ever need complex CG coefficients?

Not for SU(2) coupling in the Condon-Shortley convention — the recursion propagates real numbers and CS fixes real ladder tops, so every coefficient is real. Complex phases can appear if you adopt a different convention, or in other groups' coupling coefficients. That is precisely why quoting a sign without naming the convention is meaningless: the physics (probabilities, which are squares) is convention-blind, but intermediate amplitudes are not.

Is the CG coefficient an amplitude or a matrix element?

Both, usefully. As written, $\langle j_1 m_1; j_2 m_2|J M\rangle$ is the overlap of two basis states — an amplitude whose square answers "prepare product, measure total" (or the reverse). Collect them all and you have the entries of the unitary matrix converting one basis to the other. The two readings are the same numbers doing two jobs; the matrix viewpoint is developed in Coupled and Uncoupled Bases.

Why does each J appear exactly once — could a coupling ever repeat a J?

For two momenta, never: the multiplicity theorem (proven in Addition of Angular Momenta by counting the $m$ -degeneracy staircase) gives each $J$ in the window exactly one ladder. Couple three or more momenta and repeats arrive immediately — e.g. three spin-½ give two distinct $J=\tfrac12$ doublets — and you need an extra label (the intermediate momentum) to tell them apart. That bookkeeping crisis is what 6-j symbols manage, in this chapter's closing article.

How do I quickly check a table I copied or half-remember?

Three fast audits: (1) squared columns and rows must sum to 1 — relations (4); (2) the stretched-side entry of every ladder top must be positive — CS; (3) exchange symmetry: swapping the two factors must multiply each coefficient by $(-1)^{j_1+j_2-J}$ . A wrong sign or value almost always fails one of these within seconds. For a definitive check, `sympy.physics.quantum.cg.CG` implements the CS convention exactly.

What is the difference between a CG coefficient and a 3-j symbol?

Content: none — each 3-j symbol is a CG coefficient times $(-1)^{j_1-j_2-m_3}/\sqrt{2j_3+1}$ . Ergonomics: everything. The 3-j symbol treats all three momenta democratically, so its symmetries (column permutations, $m\to-m$ ) are simple, where CG symmetries are cluttered with square-root ratios. For deriving physics use CG; for computing at scale use 3-j. The dictionary and its phase audit live in 3-j, 6-j and Racah Symbols.

Does the recursion actually terminate, or can it run off the table?

It self-terminates: whenever a source cell would need $m_1>j_1$ or $m_2>j_2$ , its ladder square root $\sqrt{j(j{+}1)-m(m{+}1)}$ vanishes, deleting that term. The recursion knows the table's edges through the same normalization factors that built the spectrum in Ladder Operators and the Angular Momentum Spectrum — no boundary conditions need imposing by hand.

When I couple ℓ and s, does the CG table know about the physics of spin-orbit coupling?

No — and that separation is the point. The table is pure kinematics: it knows only the group theory of rotations. Whether nature prefers the coupled basis (spin-orbit interactions diagonal in $J$ ) or the uncoupled one (strong external fields) is dynamics, decided by the Hamiltonian. The which-basis-when decision theory is the next article's; the fine-structure physics itself belongs to the spin-orbit article of the hydrogen chapters.

Are Clebsch and Gordan's polynomials really the same mathematics, or is the name honorary?

Genuinely the same. Binary forms of degree $2j$ carry the spin- $j$ representation of SU(2); multiplying two forms and re-expanding in irreducible pieces is precisely coupling $j_1\otimes j_2$ , and the 1866 expansion coefficients are the quantum table's entries up to normalization conventions. The quantum generation (Wigner, Weyl) recognized the isomorphism and reached for the existing name — the mathematics predates the physics by sixty years.

$\|J,M\rangle$	expansion in $\|1\,m_1\rangle\,\|\!\uparrow/\downarrow\rangle$
$\|\tfrac32,\pm\tfrac32\rangle$	$\|1\,{\pm1}\rangle\|\!\uparrow/\downarrow\rangle$ (stretched, coefficient $1$ )
$\|\tfrac32,+\tfrac12\rangle$	$\sqrt{\tfrac23}\,\|1\,0\rangle\|\!\uparrow\rangle+\sqrt{\tfrac13}\,\|1\,1\rangle\|\!\downarrow\rangle$
$\|\tfrac12,+\tfrac12\rangle$	$-\sqrt{\tfrac13}\,\|1\,0\rangle\|\!\uparrow\rangle+\sqrt{\tfrac23}\,\|1\,1\rangle\|\!\downarrow\rangle$
$\|\tfrac32,-\tfrac12\rangle$	$\sqrt{\tfrac23}\,\|1\,0\rangle\|\!\downarrow\rangle+\sqrt{\tfrac13}\,\|1\,{-1}\rangle\|\!\uparrow\rangle$
$\|\tfrac12,-\tfrac12\rangle$	$\sqrt{\tfrac13}\,\|1\,0\rangle\|\!\downarrow\rangle-\sqrt{\tfrac23}\,\|1\,{-1}\rangle\|\!\uparrow\rangle$