Addition of Angular Momenta

Single angular momenta are a luxury of non-interacting problems. The moment anything couples two of them — orbit to spin, spin to spin, nucleus to electron cloud — the individual vectors stop being conserved and stop labeling energy levels. What survives is the sum $\hat{\mathbf J}=\hat{\mathbf J}_1+\hat{\mathbf J}_2$ , and the question of this article: given definite $j_1$ and $j_2$ , which total $J$ values exist, and how do the new eigenstates relate to the old? The answer is pure algebra. Because $\hat{\mathbf J}_1$ and $\hat{\mathbf J}_2$ live on different factors of a tensor-product space, they commute — and their sum then satisfies the same commutation relations they do. The total is a legitimate angular momentum, so the ladder machinery applies unchanged; running it through a counting argument yields the triangle rule: $J$ takes every value from $|j_1-j_2|$ to $j_1+j_2$ in integer steps, each exactly once. Quantum vector addition is classical vector addition with the continuum of angles replaced by a discrete menu of lengths. Here multiplet bookkeeping is born: $(2j_1+1)(2j_2+1)$ product states regroup into total- $J$ families without gaining or losing a dimension, and the stretched-state-and-lower algorithm exhibited here is the seed of every Clebsch–Gordan table. Fine structure, singlet–triplet magnetism, hyperfine splitting and the 21 cm line, selection rules — all downstream customers of the counting on this page.

Key takeaways

If $[\hat J_{1i},\hat J_{2j}]=0$ and each obeys the angular-momentum algebra, so does $\hat{\mathbf J}=\hat{\mathbf J}_1+\hat{\mathbf J}_2$ : the total is an angular momentum, eigenvalues $J(J+1)\hbar^2$ and $M\hbar$ .
Triangle rule: the product $j_1\otimes j_2$ decomposes into one multiplet for each $J=|j_1-j_2|,\dots,j_1+j_2$ — each value exactly once, proven by counting $M$ -degeneracies.
Dimensions balance exactly: $\sum_J(2J+1)=(2j_1+1)(2j_2+1)$ — regrouping creates and destroys nothing.
Two complete address books: uncoupled $|j_1m_1\rangle|j_2m_2\rangle$ vs coupled $|JM\rangle$ . They are incompatible bookkeeping — $[\hat J^2,\hat J_{1z}]\ne0$ — related by the unitary Clebsch–Gordan bridge, with $M=m_1+m_2$ exact.
The build algorithm: start at the stretched state (a pure product), lower with $\hat J_-=\hat J_{1-}+\hat J_{2-}$ , seed the next multiplet by orthogonality. Condon–Shortley signs make every coefficient real.
$\hat{\mathbf J}_1\cdot\hat{\mathbf J}_2=\tfrac12(\hat J^2-\hat J_1^2-\hat J_2^2)$ is diagonal in the coupled basis — why coupled states tame spin–orbit-type interactions.
Key idea: two commuting angular momenta sum to an angular momentum, and the product space splits into total- $J$ multiplets $|j_1-j_2|\le J\le j_1+j_2$ , each appearing exactly once.

The core insight

Because $\hat{\mathbf J}_1$ and $\hat{\mathbf J}_2$ commute and each obeys the angular-momentum algebra, their sum is again an angular momentum; the product space $j_1\otimes j_2$ decomposes into total- $J$ multiplets $J=|j_1-j_2|,\dots,j_1+j_2$ , with Clebsch–Gordan coefficients as the unitary bridge between uncoupled and coupled bases.

Where this lives today

Addition is daily infrastructure across scales. Atomic clocks run on hyperfine multiplets — cesium's ground state splits into $F=3$ and $F=4$ by coupling $I=\tfrac72$ to $J=\tfrac12$ , and the 9.19 GHz gap defines the second. Radio astronomy maps galactic hydrogen through the singlet–triplet flip at 1420.4 MHz. Nuclear and particle physics reuse the algebra as isospin multiplets and quark-model baryons ( $\tfrac12\otimes\tfrac12\otimes\tfrac12$ ); every decay table's selection rules are triangle-rule statements. And two-qubit quantum computing speaks exactly this article's coupled-vs-uncoupled language — the singlet is the workhorse entangled state. The counting argument of 1925 never aged; it changed laboratories.

The physical picture

Think of two spinning tops mounted on one shaft. Each has a fixed spin magnitude — $\sqrt{j_1(j_1+1)}\,\hbar$ and $\sqrt{j_2(j_2+1)}\,\hbar$ — but quantum-uncertain orientation. Aligned as well as the cones allow, they add up to the longest total, $J=j_1+j_2$ ; opposed, they cancel down to $J=|j_1-j_2|$ ; and because projections change in whole units of $\hbar$ , the totals in between come in integer steps. Classical vector addition survives, with the continuum of relative angles replaced by a short discrete menu of allowed lengths.

The cleanest way to prove which lengths appear is not geometry but inventory. The uncoupled basis has $(2j_1+1)(2j_2+1)$ product states, and each carries a definite total projection $M=m_1+m_2$ . Tally how many states share each $M$ and you get a staircase: one state at the top ( $M=j_1+j_2$ ), two at the next step, climbing by one per step until it plateaus, then descending symmetrically. Every multiplet of total $J$ must contribute exactly one state at each $M$ from $-J$ to $J$ — so the staircase is the multiplet content, readable by differences. That counting argument, run in full below, is the entire triangle rule.

Figure

The M-staircase for 1 ⊗ 1 — degeneracy counts 1,2,3,2,1 decode into multiplets J = 2, 1, 0

One more picture: coupled states are eigenstates of $\hat J^2$ , and $\hat J^2$ contains the cross term $2\hat{\mathbf J}_1\cdot\hat{\mathbf J}_2$ , which trades a unit of $m_1$ for one of $m_2$ and back. So, except at the stretched extremes, no coupled state can be a single product: a definite total length requires a definite relative orientation, incompatible with definite individual projections. Superposition is forced by the geometry of addition.

Two angular momenta, one algebra

Everything begins with the sum and the claim that it deserves the name "angular momentum". The claim needs proof: sums of operators do not automatically inherit algebraic structure.

\hat{\mathbf J} = \hat{\mathbf J}_1 + \hat{\mathbf J}_2, \qquad [\hat J_{1i},\,\hat J_{2j}] = 0

where: $\hat{\mathbf J}_1$ acts on the first tensor factor and $\hat{\mathbf J}_2$ on the second (strictly $\hat J_{1i}\otimes\mathbb 1+\mathbb 1\otimes\hat J_{2i}$ ), which is why every component of one commutes with every component of the other. That vanishing cross-commutator is the entire hypothesis of the construction.

Derivation AThe total inherits the algebra

The goal is the statement licensing all chapter machinery on the composite: $[\hat J_i,\hat J_j]=i\hbar\,\epsilon_{ijk}\hat J_k$ for the sum, given each piece satisfies the algebra of Angular Momentum Commutation Relations (consumed, not re-derived) and the pieces commute. The strategy is bilinearity: expand, watch the cross terms die, watch the like-terms reassemble the sum. The punchline is structural — once the total is an angular momentum, the ladder theorem applies verbatim, eigenvalues $J(J+1)\hbar^2$ and $M\hbar$ included, no new work.

Step 1. Expand by bilinearity of the commutator:

[\hat J_i,\hat J_j] = [\hat J_{1i}+\hat J_{2i},\,\hat J_{1j}+\hat J_{2j}] = [\hat J_{1i},\hat J_{1j}] + [\hat J_{2i},\hat J_{2j}] + [\hat J_{1i},\hat J_{2j}] + [\hat J_{2i},\hat J_{1j}].

Step 2. Kill the cross terms. The last two commutators vanish because the operators act on different tensor factors — subsystem 1's operators cannot fail to commute with subsystem 2's. Why this hypothesis is not decoration: for two angular momenta of the same particle that do not commute, the sum would not close into the algebra. Independence of the two carriers is physics, not pedantry.

Step 3. Reassemble with each subsystem's own algebra:

[\hat J_i,\hat J_j] = i\hbar\,\epsilon_{ijk}\hat J_{1k} + i\hbar\,\epsilon_{ijk}\hat J_{2k} = i\hbar\,\epsilon_{ijk}\hat J_k.

Step 4. Harvest. $\hat{\mathbf J}$ is a bona fide angular momentum, so Ladder Operators and the Angular Momentum Spectrum applies wholesale: eigenvalues $J(J+1)\hbar^2$ , $M=-J,\dots,J$ , ladders $\hat J_\pm=\hat J_{1\pm}+\hat J_{2\pm}$ with standard matrix elements. Which $J$ occur is the next derivation; that they are ladder-legal is settled here. Verified: the full algebra checked numerically for the composites $\tfrac12\otimes\tfrac12$ and $1\otimes\tfrac12$ (Kronecker-built matrices, machine precision), including all nine vanishing cross-commutators.

Two address books that cannot both be used

The composite space supports two natural complete labelings, and their tension is the subject's engine. The uncoupled basis $|j_1m_1\rangle|j_2m_2\rangle$ diagonalizes $\{\hat J_1^2,\hat J_{1z},\hat J_2^2,\hat J_{2z}\}$ — each top described separately. The coupled basis $|JM\rangle$ diagonalizes $\{\hat J_1^2,\hat J_2^2,\hat J^2,\hat J_z\}$ — the basis interactions prefer. Both quartets are complete commuting sets per Simultaneous Diagonalization and Complete Sets of Commuting Observables (whose two-spin $m=0$ tie previewed exactly this choice). But you cannot merge them:

[\hat J^2,\,\hat J_{1z}] = 2i\hbar\,(\hat J_{1x}\hat J_{2y}-\hat J_{1y}\hat J_{2x}) \ne 0

where: the obstruction comes from $\hat J^2=\hat J_1^2+\hat J_2^2+2\hat{\mathbf J}_1\cdot\hat{\mathbf J}_2$ — only the cross term fails to commute with $\hat J_{1z}$ , by exactly this amount (machine-checked on $\tfrac12\otimes\tfrac12$ ). What does survive: $\hat J_z=\hat J_{1z}+\hat J_{2z}$ commutes with $\hat J^2$ , so $M$ is a good label — and being a sum, $M=m_1+m_2$ holds term by term in any superposition. The unitary bridge between the two books is the subject of its own article; here we only name its entries. Expanding a coupled state over the uncoupled basis,

|j_1j_2;JM\rangle = \sum_{m_1+m_2=M} \langle j_1m_1\,j_2m_2|JM\rangle\;|j_1m_1\rangle|j_2m_2\rangle

where: the numbers $\langle j_1m_1\,j_2m_2|JM\rangle$ are the Clebsch–Gordan coefficients — real in the Condon–Shortley convention, nonzero only when $M=m_1+m_2$ and the triangle rule holds. Their recursions, orthogonality and tables are owned by Clebsch-Gordan Coefficients; the change-of-basis theory by Coupled and Uncoupled Bases. This article establishes which $J$ -blocks the bridge connects and exhibits the mechanism generating every entry.

The triangle rule, by counting

Go deeper · Derivation B · 2 minThe triangle rule from highest weights

Derivation BThe triangle rule from highest weights

The goal is the decomposition theorem: in $j_1\otimes j_2$ the total takes each value $J=|j_1-j_2|,\dots,j_1+j_2$ exactly once — no gaps, no repeats, nothing below the floor. The strategy is pure inventory: let $n(M)$ count uncoupled states with projection $M$ and $N(J)$ count multiplets; a $J$ -multiplet contributes one state at every $M\in[-J,J]$ , so the censuses are rigidly linked and differencing the staircase extracts $N(J)$ . Watch how little is used — no dynamics, no Clebsch–Gordan values, only the shape of a sum of two integer-spaced ranges.

Step 1. Census. $M=m_1+m_2$ with $|m_i|\le j_i$ (take $j_1\ge j_2$ ). The extreme $M=j_1+j_2$ is reached only by $(j_1,j_2)$ : $n=1$ . Each unit step down adds one more split, so $n(M)$ climbs by one per step until $M=j_1-j_2$ , where the shorter range is exhausted; it plateaus, then descends mirror-symmetrically. For $1\otimes1$ : $n=1,2,3,2,1$ across $M=2,\dots,-2$ (the staircase figure).

Step 2. Link the censuses. A multiplet of total $J$ has one state at each $M$ with $|M|\le J$ , so

n(M) = \sum_{J\ge|M|} N(J).

Why this is the load-bearing line: it says the staircase is a cumulative count of multiplets, read from the top.

Step 3. Difference to isolate. Subtracting consecutive censuses telescopes away every $J>M$ :

N(J) = n(M{=}J) - n(M{=}J{+}1).

Step 4. Evaluate on the staircase. From Step 1, $n$ increases by exactly one per step from $M=j_1+j_2$ down to $j_1-j_2$ , so each difference is $1$ : one multiplet each at $J=j_1+j_2,\ j_1+j_2-1,\ \dots,\ j_1-j_2$ . On the plateau ( $|M|<j_1-j_2$ ) the difference is $0$ : nothing new appears below the floor. Hence

j_1\otimes j_2 = (j_1{+}j_2)\ \oplus\ (j_1{+}j_2{-}1)\ \oplus\ \cdots\ \oplus\ |j_1{-}j_2|,

each exactly once — the decomposition is multiplicity-free, a special kindness of the rotation algebra that composite labels elsewhere in physics (flavor symmetries, for instance) do not enjoy.

Step 5. Audit the dimensions. The multiplet dimensions must resum to the product. The $J$ -range is an arithmetic series of $2j_2+1$ terms with mean $2j_1+1$ :

\sum_{J=j_1-j_2}^{j_1+j_2}(2J+1) = (2j_2+1)(2j_1+1).

Nothing gained, nothing lost — regrouping is a change of basis, not of space. Verified: the staircase census run programmatically for $(j_1,j_2)=(1,\tfrac12),(1,1),(\tfrac32,1)$ — every $N(J)=1$ on the range, $0$ below it; the dimension identity proven symbolically (sympy) for general $j_1\ge j_2$ and audited numerically ( $9=5+3+1$ , $12=6+4+2$ , $10=6+4$ ).

Building the multiplets: the descent mechanism

The counting says what exists; the ladder builds it. The algorithm needs one free entry point — and the census provides it: the top state $M=j_1+j_2$ is unique, so no superposition is possible and the stretched state is a pure product,

|J{=}j_1{+}j_2,\,M{=}j_1{+}j_2\rangle = |j_1j_1\rangle|j_2j_2\rangle

where: its Clebsch–Gordan coefficient is exactly $1$ (normalization allows $\pm1$ ; Condon–Shortley fixes $+$ ). From here, $\hat J_-=\hat J_{1-}+\hat J_{2-}$ walks down the top multiplet, and orthogonality seeds each new one. Watch the mechanism once, in full, on the smallest case with a genuine superposition.

Go deeper · Derivation C · 1 minOne rung down from the stretched state — how every table is born

Derivation COne rung down from the stretched state — how every table is born

The goal is the explicit pair of $M=\tfrac12$ coupled states of $1\otimes\tfrac12$ — the smallest instance exhibiting all three moves of the general algorithm: lower the stretched state, normalize against the total's ladder coefficient, seed the next multiplet by orthogonality with the Condon–Shortley sign. The stepping coefficients of Ladder Operators and the Angular Momentum Spectrum are consumed as given; nothing is re-derived.

Step 1. Lower the left side as a total. On $|J,M\rangle=|\tfrac32,\tfrac32\rangle$ the master formula gives

\hat J_-\big|\tfrac32,\tfrac32\big\rangle = \hbar\sqrt{\tfrac32\cdot\tfrac52-\tfrac32\cdot\tfrac12}\;\big|\tfrac32,\tfrac12\big\rangle = \hbar\sqrt3\;\big|\tfrac32,\tfrac12\big\rangle.

Step 2. Lower the right side as a sum. $\hat J_-=\hat J_{1-}+\hat J_{2-}$ acts on the product $|1\,1\rangle|\tfrac12\,\tfrac12\rangle$ one factor at a time: $\hat J_{1-}|1\,1\rangle=\hbar\sqrt2\,|1\,0\rangle$ and $\hat J_{2-}|\tfrac12\,\tfrac12\rangle=\hbar\,|\tfrac12,-\tfrac12\rangle$ , so

\hat J_-\,|1\,1\rangle|\tfrac12\,\tfrac12\rangle = \hbar\big(\sqrt2\,|1\,0\rangle|\tfrac12\,\tfrac12\rangle + |1\,1\rangle|\tfrac12,-\tfrac12\rangle\big).

Why both terms: the total lowering operator does not know which subsystem to lower, so it lowers each — the superposition of the coupled world is generated by the sum structure of $\hat J_-$ .

Step 3. Match and normalize. Equating the two expressions and dividing by $\hbar\sqrt3$ :

\big|\tfrac32,\tfrac12\big\rangle = \sqrt{\tfrac23}\,|1\,0\rangle|\tfrac12\,\tfrac12\rangle + \sqrt{\tfrac13}\,|1\,1\rangle|\tfrac12,-\tfrac12\rangle.

Both coefficients real and positive — Condon–Shortley inherited automatically from the ladder's phase card.

Step 4. Seed the next multiplet by orthogonality. The $M=\tfrac12$ subspace is two-dimensional, and the counting theorem says exactly one more multiplet lives here: $J=\tfrac12$ . Its top state is the combination orthogonal to Step 3, with overall sign fixed by the convention that the coefficient of the maximal- $m_1$ component ( $m_1=j_1=1$ ) be positive:

\big|\tfrac12,\tfrac12\big\rangle = -\sqrt{\tfrac13}\,|1\,0\rangle|\tfrac12\,\tfrac12\rangle + \sqrt{\tfrac23}\,|1\,1\rangle|\tfrac12,-\tfrac12\rangle.

Step 5. Read the general algorithm off the instance: lower each multiplet from its top; at every new $M$ -level one fresh orthogonal direction opens (the staircase climbs by one), and orthogonality plus the sign convention pins its top state. Iterating fills the whole table — how Clebsch-Gordan Coefficients generates its tables; that article owns the machinery, and this exhibit is its seed. Verified: all four coefficients match sympy's Condon–Shortley CG values exactly ( $\sqrt{2/3},\sqrt{1/3},-\sqrt{1/3},+\sqrt{2/3}$ ); the two states confirmed numerically as $\hat J^2$ eigenvectors with eigenvalues $\tfrac{15}4\hbar^2$ and $\tfrac34\hbar^2$ and $\hat J_z=\tfrac\hbar2$ ; stretched-state coefficient $=1$ exact.

The same mechanism on $\tfrac12\otimes\tfrac12$ produces the most famous decomposition in physics, $\tfrac12\otimes\tfrac12=1\oplus0$ : a three-state triplet and a one-state singlet, audit $4=3+1$ . The explicit construction — symmetric/antisymmetric combinations, exchange symmetry, ortho/para matter — is deliberately left to Singlet and Triplet States, which owns it; here the pair stands as the triangle rule's smallest ledger entry.

The coupling term: why the coupled basis earns its keep

Interactions between two angular momenta almost always enter through their dot product, and the dot product is a disguised difference of Casimirs. Square the sum:

\hat{\mathbf J}_1\cdot\hat{\mathbf J}_2 = \tfrac12\left(\hat J^2 - \hat J_1^2 - \hat J_2^2\right)

where: all three operators on the right are diagonal in the coupled basis, so the interaction is too — its eigenvalue on $|JM\rangle$ is $\tfrac{\hbar^2}2[J(J+1)-j_1(j_1+1)-j_2(j_2+1)]$ , independent of $M$ as rotational invariance demands. This one line is why nature's couplings ( $\hat{\mathbf L}\cdot\hat{\mathbf S}$ spin–orbit, $\hat{\mathbf S}_1\cdot\hat{\mathbf S}_2$ exchange, $\hat{\mathbf I}\cdot\hat{\mathbf J}$ hyperfine) all select the coupled book: in it they are numbers, not matrices. Two instances carry most of atomic physics. For orbit and spin, $\ell\otimes\tfrac12=(\ell{+}\tfrac12)\oplus(\ell{-}\tfrac12)$ — every $\ell\ge1$ level is a doublet, with coupling values $+\tfrac{\ell}{2}\hbar^2$ ( $j=\ell+\tfrac12$ ) and $-\tfrac{\ell+1}{2}\hbar^2$ ( $j=\ell-\tfrac12$ ): splitting $\tfrac{2\ell+1}{2}\hbar^2$ . That count is the origin of fine-structure doublets — sodium's D pair at 589.0/589.6 nm ( $\ell=1$ , 2.1 meV) — while the radial physics that turns the count into electron-volts belongs to Spin-Orbit Coupling and Fine Structure. For two spins, $\hat{\mathbf S}_1\cdot\hat{\mathbf S}_2$ takes $+\tfrac14\hbar^2$ (triplet) and $-\tfrac34\hbar^2$ (singlet) — the switch behind exchange magnetism and the hydrogen hyperfine pair whose $F=1\to0$ flip radiates the 21 cm line ( $1420.4$ MHz; $c/\nu=21.1$ cm).

Limiting cases

case	result	meaning
two spin- $\tfrac12$	$\tfrac12\otimes\tfrac12=1\oplus0$	triplet + singlet; audit $4=3+1$ ; construction owned by Singlet and Triplet States
spin–orbit, $\ell\ge1$	$\ell\otimes\tfrac12=(\ell{+}\tfrac12)\oplus(\ell{-}\tfrac12)$	every level a doublet — fine-structure pairs
couple to $j_2=0$	$j\otimes0=j$	a spinless partner changes nothing; the triangle collapses
stretched state	$\|j_1{+}j_2,j_1{+}j_2\rangle=\|j_1j_1\rangle\|j_2j_2\rangle$	the unique pure-product coupled state; CG $=1$
equal spins, opposed	$J_{\min}=0$ exists iff $j_1=j_2$	perfect cancellation needs equal lengths
large $j_1,j_2$	$J\approx\sqrt{j_1^2+j_2^2+2j_1j_2\cos\theta_{12}}$	the discrete menu densifies into classical vector addition

Common trapM = m₁ + m₂ means the coupled state is the product state with those m's

Two confusions, one root. First: $M=m_1+m_2$ is exact — but it constrains which products can appear, not which single product the state equals. Except at the stretched extremes, $|JM\rangle$ is a superposition of all products sharing that $M$ (the $J^2$ cross terms force it). Second: a product like $|1\,0\rangle|\tfrac12\,\tfrac12\rangle$ has definite $M=\tfrac12$ but no definite $J$ — measuring $\hat J^2$ gives $\tfrac{15}4\hbar^2$ with probability $\tfrac23$ , $\tfrac34\hbar^2$ with $\tfrac13$ (squared descent coefficients). Never read $J=j_1+j_2$ off one pair of $m$ 's: the maximal multiplet is a menu option, not an identity.

Worked examples

Example 1 — two p-electrons (confidence builder). Task: two electrons occupy $\ell=1$ orbitals. What total orbital angular momenta $L$ can the pair carry, what is the dimension audit, and what does $\hat{\mathbf L}_1\cdot\hat{\mathbf L}_2$ equal in each block?

Step 1. Triangle rule: $1\otimes1$ gives $L=2,1,0$ — the $D$ , $P$ , $S$ terms of atomic spectroscopy.

Step 2. Dimension audit: $(2{\cdot}1{+}1)^2=9$ product states regroup as $5+3+1=9$ . Balanced.

Step 3. Coupling term per block: $\hat{\mathbf L}_1\cdot\hat{\mathbf L}_2=\tfrac12[L(L+1)-2-2]\hbar^2$ evaluates to $\boldsymbol{+\hbar^2}$ ( $L=2$ ), $\boldsymbol{-\hbar^2}$ ( $L=1$ ), $\boldsymbol{-2\hbar^2}$ ( $L=0$ ) — verified as the exact eigenvalue spectrum (multiplicities 5, 3, 1) of the Kronecker-built operator. Any interaction $\propto\hat{\mathbf L}_1\cdot\hat{\mathbf L}_2$ splits the nine states into exactly these three shelves.

Example 2 — a d-electron's spin–orbit ledger (exam level). Task: one electron with $\ell=2$ , $s=\tfrac12$ . Find the allowed $j$ , the state counts, and the values and splitting of $\hat{\mathbf L}\cdot\hat{\mathbf S}$ .

Step 1. Menu: $2\otimes\tfrac12$ gives $j=\tfrac52$ and $j=\tfrac32$ — a doublet, as for every $\ell\ge1$ .

Step 2. Counts: $2j+1$ gives $6$ and $4$ ; audit $6+4=10=(2{\cdot}2{+}1)(2)=5\times2$ . Balanced.

Step 3. Coupling values: $\langle\hat{\mathbf L}\cdot\hat{\mathbf S}\rangle=\tfrac{\hbar^2}2[j(j{+}1)-6-\tfrac34]$ gives $\boldsymbol{+\hbar^2}$ for $j=\tfrac52$ and $\boldsymbol{-\tfrac32\hbar^2}$ for $j=\tfrac32$ — matching the general $\ell\otimes\tfrac12$ formulas $+\tfrac\ell2\hbar^2$ and $-\tfrac{\ell+1}2\hbar^2$ at $\ell=2$ .

Step 4. Splitting: $\Delta=\tfrac{2\ell+1}{2}\hbar^2=\tfrac52\hbar^2$ . Multiplied by the radial factor $\xi_{n\ell}$ , this is the fine-structure gap of every $d$ -level; the heavier the atom, the larger $\xi$ , which is why $5d$ metals wear their spin–orbit splitting on their chemistry. The counting is universal; only the prefactor is element-specific.

Origins · 1 minHenry Norris Russell and how this was discovered

Henry Norris Russell

1877–1957 · Princeton University Observatory

The coupling scheme taught in every quantum course was invented by an astronomer, before quantum mechanics could justify it. Henry Norris Russell ran Princeton's observatory and co-invented the Hertzsprung–Russell diagram — a story for astrophysics. His spectroscopy problem was harsher: tens of thousands of unassigned lines from iron, titanium, calcium, defying order. In 1925, with Frederick Saunders of Harvard, he proposed that in light atoms the orbital momenta first sum to a total $L$ , the spins separately to $S$ , and only then do $L$ and $S$ couple to $J$ . LS coupling (Russell–Saunders, as every term symbol still silently credits) turned spectral chaos into a filing system: term symbols, multiplicities, interval rules. When the Schrödinger equation arrived a year later the scheme survived intact — the astronomers' bookkeeping was exactly this article's triangle-rule decomposition, waiting for its algebra. Russell spent two more decades reading the Sun's composition with it; his school's multiplet tables still calibrate stellar spectra.

History

1913

— Landé begins organizing anomalous Zeeman multiplets with empirical vector-addition rules

1922

— Landé's g-formula and interval rule expose vector-model structure with half-integers nobody can yet explain

1925

— Russell and Saunders publish LS coupling for two-valence-electron spectra; Uhlenbeck and Goudsmit supply spin

1926

— Heisenberg treats helium: two-electron symmetric and antisymmetric states meet exchange

1928

— Clebsch–Gordan mathematics of the 1860s, built for invariant theory, is recognized as the coupling calculus of the new quantum mechanics

1935

— Condon and Shortley's Theory of Atomic Spectra fixes the phase conventions still in force

Going further

References. Griffiths, 3rd ed., §4.4 · Sakurai & Napolitano, 2nd ed., §3.8 · Shankar, 2nd ed., Ch. 15 · Cohen-Tannoudji, Diu & Laloë, Vol. 2, Ch. X · Townsend, 2nd ed., Ch. 11 · Edmonds, Angular Momentum in Quantum Mechanics (Clebsch–Gordan tables).

Provenance. All structural results follow from the angular momentum algebra; a verifier should recompute: (1) the dimension identity $\sum(2J+1)=(2j_1+1)(2j_2+1)$ ; (2) the spin- $\tfrac12$ singlet/triplet states by lowering and orthogonality (Griffiths §4.4); (3) $\langle\mathbf{L}\cdot\mathbf{S}\rangle$ via $\tfrac12(J^2-L^2-S^2)$ (Griffiths §7.3); (4) $\mathbf{S}_1\cdot\mathbf{S}_2$ eigenvalues $+\tfrac14\hbar^2,-\tfrac34\hbar^2$ . Numbers cited: sodium D lines 589.0/589.6 nm and the hydrogen 21 cm line are standard spectroscopic values (NIST). Cross-check Sakurai §3.8, Cohen-Tannoudji Vol. 2 Ch. X, Edmonds for Clebsch–Gordan conventions (Condon–Shortley). For this edition, items (1), (3), (4) and every number were machine-verified (sympy/numpy, 42 checks): the composite algebra and both CSCOs numerically, the staircase census vs the theorem for three $(j_1,j_2)$ pairs, the dimension telescope symbolically, the $1\otimes\tfrac12$ descent coefficients against sympy's Condon–Shortley CG values, both worked examples as exact spectra. Ownership note: item (2), the singlet/triplet construction in the base, belongs per the chapter map to Singlet and Triplet States and is deliberately not reproduced — this article establishes $\tfrac12\otimes\tfrac12=1\oplus0$ by counting and hands the build downstream.

Upstream: Angular Momentum Commutation Relations (the algebra, consumed whole) · Ladder Operators and the Angular Momentum Spectrum (stepping coefficients + Condon–Shortley card, consumed in the descent) · Spin Angular Momentum / Spin-1/2 and the Pauli Matrices (the spin- $\tfrac12$ factors) · Simultaneous Diagonalization and Complete Sets of Commuting Observables (the two-CSCO grammar). Downstream: Clebsch-Gordan Coefficients (the bridge machinery) · Coupled and Uncoupled Bases (basis-change theory) · Singlet and Triplet States (the ½⊗½ build + exchange) · Spin-Orbit Coupling and Fine Structure / Hyperfine Structure (paying customers) · Wigner-Eckart Theorem.

Is J = j₁ + j₂ always the state the system is actually in?

No — the triangle rule lists the possible totals; which is realized depends on the state or, for energy eigenstates, the Hamiltonian. A positive $\hat{\mathbf J}_1\cdot\hat{\mathbf J}_2$ coupling puts the smallest $J$ lowest (its eigenvalue grows with $J(J+1)$ ) — why antiferromagnets pair spins into singlets. The stretched $J=j_1+j_2$ is just the top of the menu, special only as the unique product state.

Why does each J appear exactly once? Group theory books make this sound deep.

The elementary reason is the staircase: $n(M)$ climbs by exactly one per unit step, so consecutive differences are all $1$ . The deep-sounding name is that SU(2) tensor products are multiplicity-free — a genuine luxury. Three or more momenta already break it: $\tfrac12\otimes\tfrac12\otimes\tfrac12$ contains $J=\tfrac12$ twice, and distinguishing the copies needs the recoupling schemes and 6j symbols of a later article.

Do j₁ and j₂ remain good quantum numbers after coupling?

Yes — $\hat J_1^2$ and $\hat J_2^2$ commute with everything in sight, including $\hat J^2$ and $\hat J_z$ (machine-checked), so the coupled states carry all four labels $|j_1j_2;JM\rangle$ . What is sacrificed is only the projections $m_1,m_2$ . Physically: coupling reorients the two vectors relative to each other; it never changes their lengths.

How do I know which basis to use for a given problem?

Follow the Hamiltonian. If it contains $\hat{\mathbf J}_1\cdot\hat{\mathbf J}_2$ (spin–orbit, exchange, hyperfine), the coupled basis diagonalizes the interaction and is the only sane choice. If each subsystem couples separately to an external probe — say $B(a\hat J_{1z}+b\hat J_{2z})$ with different gyromagnetic factors — the uncoupled basis wins. When both terms coexist, the basis choice tracks the stronger coupling, and the crossover (weak-field Zeeman vs Paschen–Back) is itself physics.

Where exactly does the Condon–Shortley convention enter the construction?

Twice, both times as a choice of sign, never of physics. The stretched state's coefficient is fixed to $+1$ (normalization alone would allow $\pm1$ ); and each new multiplet seeded by orthogonality gets its overall sign fixed by demanding the coefficient of the maximal- $m_1$ component be positive — that is why $|\tfrac12,\tfrac12\rangle$ above reads $-\sqrt{1/3},+\sqrt{2/3}$ and not the negatives. Lowering preserves these choices because the ladder coefficients are themselves real and non-negative in this convention.

Does adding angular momenta have anything to do with entanglement?

Everything. Every coupled state except the stretched ends is entangled — the descent derivation manufactures superpositions that cannot factor; the two-spin singlet is the workhorse of Bell tests. The dictionary runs both ways: interactions conserving total angular momentum naturally generate entanglement, and the coupled basis is its natural language.

My two angular momenta belong to the same particle (its L and its S). Do the theorems still apply?

Yes — commutation is what matters, not particle count. $\hat{\mathbf L}$ acts on the spatial factor, $\hat{\mathbf S}$ on the spin factor, so $[\hat L_i,\hat S_j]=0$ and the whole construction goes through verbatim; $\ell\otimes\tfrac12$ is the textbook case. The theorems fail only for non-commuting "angular momenta", which the hypothesis excludes.