Clebsch–Gordan Coefficients · Elementary Particle Physics

Two angular momenta combine in one way only, fixed before any force is named. A table of pure numbers therefore predicts a measured ratio that no dynamics was allowed to influence.

🎯 Why this matters

A prediction with no adjustable content fails informatively. When a ratio like 9 : 2 : 1 comes out wrong there is no coupling to retune and no model to blame — the symmetry it assumed is simply absent, which is how isospin’s limits were mapped.

Appendix 4 is one page of numbers with no explanation whatsoever, and it is used in Chapters 3, 4, 6 and 9. This page explains what those numbers are, gives you a calculator that produces them for any pair of angular momenta, and then shows the single most convincing thing they do: predicting a measured cross-section ratio of 9 : 2 : 1 from nothing but bookkeeping.

📐 Physics you need first — why “adding” angular momenta is not adding

Two systems each carry an angular momentum — spins, orbital momenta, or (as here) isospins. Combine them and the total is not simply

j_1 + j_2

, because the two vectors can point in different relative directions and quantum mechanics only lets that relative orientation take discrete values. The allowed totals are

J = |j_1 - j_2|,\ |j_1 - j_2| + 1,\ \ldots,\ j_1 + j_2

— the triangle rule . A spin-1 combined with a spin-½ can be $3/2$ or $1/2$ , nothing else. Meanwhile the projections on a chosen axis do just add: $M = m_1 + m_2$ , exactly.

So you have two ways to label the same set of states: by the individual projections $(m_1, m_2)$ , or by the total $(J, M)$ . Both are complete. The Clebsch–Gordan coefficients are the dictionary between them. Nothing more mysterious than that.

The definition

\htmlClass{t-JM}{|J, M\rangle} = \sum_{m_1 + m_2 = M} \htmlClass{t-cg}{\langle j_1 m_1;\, j_2 m_2 | J M \rangle}\; \htmlClass{t-prod}{|j_1 m_1\rangle |j_2 m_2\rangle}

(R.4)

One state of definite total angular momentum, written out in the basis where the two parts are labelled separately.

Every symbol, one at a time

Hover or tap a symbol above — it lights up in the equation and its meaning, units and type appear here.

💡 What this really says — a rotation of coordinates in state space, not a computation

There are exactly

(2j_1+1)(2j_2+1)

states either way, so no information is gained or lost — this is a rotation of coordinates in state space, not a computation. Reading the table left to right expands a total-J state in the product basis; reading it top to bottom does the reverse. Because the matrix is orthogonal, the second direction is just the transpose. The calculator below checks that numerically for whatever

j_1, j_2

you pick.

⚙️ Engineer’s bridge

This is tensor-product decomposition, and it is linear algebra you have done before under other names.

$|j_1 m_1\rangle|j_2 m_2\rangle$ is a Kronecker product of two vectors: a $(2j_1+1)(2j_2+1)$ -dimensional space built from two smaller ones.
The rotation operator acting on that space is a Kronecker product of two matrices — and it is not block-diagonal in the product basis.
The Clebsch–Gordan table is the orthogonal matrix $U$ that block-diagonalises it: $U^{\mathsf T} (D^{j_1}\otimes D^{j_2}) U$ comes out as blocks of size $2J+1$ , one per allowed $J$ . Each block is an irreducible piece — a subspace that rotations mix internally but never leak out of.
$\mathbf{3}\otimes\mathbf{2} = \mathbf{4}\oplus\mathbf{2}$ is exactly the dimension bookkeeping $6 = 4 + 2$ .

Switch the calculator to as a matrix and you can see the blocks. Sorting both bases by $M$ is what makes them contiguous, because $M = m_1 + m_2$ is conserved cell by cell — the same trick as permuting a sparse matrix into block form before you factor it.

Where it breaks: unlike a generic change of basis you might pick for convenience, this one is forced. The blocks are the irreducible representations of the rotation group, and no finer decomposition exists. That uniqueness is why the same table appears in every textbook and why a coefficient is a physical prediction, not a choice.

The calculator

Set $j_1$ and $j_2$ and you get any table you need — including the three the book prints on p. 501. Every coefficient here comes from Racah’s closed formula and was checked entry by entry, signs included, against the printed page.

j₁j₂

$1 \otimes \tfrac{1}{2} = \mathbf{4} \oplus \mathbf{2}\quad (6 = 4 + 2)$

$m_1$	$m_2$	$\tfrac{3}{2},\,+\tfrac{3}{2}$	$\tfrac{3}{2},\,+\tfrac{1}{2}$	$\tfrac{1}{2},\,+\tfrac{1}{2}$	$\tfrac{3}{2},\,-\tfrac{1}{2}$	$\tfrac{1}{2},\,-\tfrac{1}{2}$	$\tfrac{3}{2},\,-\tfrac{3}{2}$
$+1$	$+\tfrac{1}{2}$	$1$
$+1$	$-\tfrac{1}{2}$		$\sqrt{1/3}$	$\sqrt{2/3}$
$+0$	$+\tfrac{1}{2}$		$\sqrt{2/3}$	$-\sqrt{1/3}$
$+0$	$-\tfrac{1}{2}$				$\sqrt{2/3}$	$\sqrt{1/3}$
$-1$	$+\tfrac{1}{2}$				$\sqrt{1/3}$	$-\sqrt{2/3}$
$-1$	$-\tfrac{1}{2}$						$1$

Dimension check: 6 = 4 + 2 = 6 ✓ · rows orthonormal ✓ — the table is an orthogonal matrix, so reading it backwards is just its transpose.

Aside — the sign convention

Signs are a convention (Condon–Shortley: the coefficient with

m_1

maximal is taken positive), but you must use one convention consistently, because relative signs inside a sum are physical. The book states the symmetry that relates the two orderings:

\langle j_1 m_1;\, j_2 m_2 | J M \rangle = (-1)^{J - j_1 - j_2}\, \langle j_2 m_2;\, j_1 m_1 | J M \rangle .

Swapping which particle you call “first” can flip a sign — and for $j_1 = j_2 = 1/2$ , the $J = 0$ combination picks up exactly that minus sign. That is the antisymmetric singlet, and its antisymmetry is why two identical fermions can share a spatial state only in the spin-0 combination. A convention with physical consequences.

The payoff: pion–nucleon scattering, 9 : 2 : 1

Here is the classic use, and it is worth following in full because it is the whole method in miniature. The pion is an isospin triplet ( $I = 1$ ), the nucleon an isospin doublet ( $I = 1/2$ ). The strong interaction conserves isospin, so a scattering amplitude can depend only on the total $I$ — there are exactly two independent amplitudes, $A_{3/2}$ and $A_{1/2}$ , and everything else is bookkeeping.

Read the coefficients off the $1 \otimes 1/2$ table:

\begin{aligned} |\pi^+ p\rangle &= \left|\tfrac32, +\tfrac32\right\rangle \\ |\pi^- p\rangle &= \sqrt{\tfrac13}\left|\tfrac32, -\tfrac12\right\rangle - \sqrt{\tfrac23}\left|\tfrac12, -\tfrac12\right\rangle \\ |\pi^0 n\rangle &= \sqrt{\tfrac23}\left|\tfrac32, -\tfrac12\right\rangle + \sqrt{\tfrac13}\left|\tfrac12, -\tfrac12\right\rangle \end{aligned}

π⁺ on a proton is a pure $I = 3/2$ state — there is simply no other way to make $I_z = +3/2$ . The other two are mixtures.

🔢 Worked example — the Δ(1232) resonance

At the

\Delta^{++}(1232)

peak the

I = 3/2

amplitude is resonant and dwarfs the other, so set

A_{1/2} \approx 0

. Then each cross-section is the squared overlap through the

I = 3/2

channel alone:

\begin{aligned} \sigma(\pi^+ p \to \pi^+ p) &\propto |1 \cdot 1|^2 = 1 \\ \sigma(\pi^- p \to \pi^0 n) &\propto \left|\sqrt{\tfrac23}\sqrt{\tfrac13}\right|^2 = \tfrac{2}{9} \\ \sigma(\pi^- p \to \pi^- p) &\propto \left|\sqrt{\tfrac13}\sqrt{\tfrac13}\right|^2 = \tfrac{1}{9} \end{aligned}

\boxed{\;\sigma(\pi^+p \to \pi^+p) : \sigma(\pi^-p \to \pi^0n) : \sigma(\pi^-p \to \pi^-p) = 9 : 2 : 1\;}

The sharpest test is the ratio of the two totals. Everything a π⁻ can do to a proton goes through those two channels, so

\frac{\sigma_{\text{tot}}(\pi^+ p)}{\sigma_{\text{tot}}(\pi^- p)} = \frac{1}{2/9 + 1/9} = 3 .

At the resonance peak the measured totals are roughly 200 mb and 70 mb — a ratio of about 2.9, against a prediction of exactly 3, obtained from a table of square roots and no dynamics whatsoever. §4.2 shows the actual excitation curves.

Reproduce it

from math import sqrt
# Decompositions read off the 1 (x) 1/2 table (Appendix 4, p. 501),
# keyed by total isospin I.
pip_p = {1.5: 1.0,          0.5: 0.0}          # pi+ p = |1,+1>|1/2,+1/2>
pim_p = {1.5: sqrt(1/3),    0.5: -sqrt(2/3)}   # pi- p = |1,-1>|1/2,+1/2>
pi0_n = {1.5: sqrt(2/3),    0.5:  sqrt(1/3)}   # pi0 n = |1, 0>|1/2,-1/2>

def amp(final, initial, A32, A12):
    return final[1.5]*initial[1.5]*A32 + final[0.5]*initial[0.5]*A12

A32, A12 = 1.0, 0.0        # on the Delta(1232): only I = 3/2 resonates
el_p = amp(pip_p, pip_p, A32, A12)**2
ce   = amp(pi0_n, pim_p, A32, A12)**2
el_m = amp(pim_p, pim_p, A32, A12)**2
print(f"pi+p -> pi+p      : {el_p:.6f}")
print(f"pi-p -> pi0n (CE) : {ce:.6f}")
print(f"pi-p -> pi-p      : {el_m:.6f}")
print(f"ratio             : {el_p/el_m:.1f} : {ce/el_m:.1f} : 1")

prints

pi+p -> pi+p      : 1.000000
pi-p -> pi0n (CE) : 0.222222
pi-p -> pi-p      : 0.111111
ratio             : 9.0 : 2.0 : 1

💡 What this really says — symmetry alone fixed a measurable ratio, with no force law anywhere

No force law was used. No potential, no coupling constant, no propagator. The only input was the strong interaction cannot tell a proton from a neutron, plus a table of geometric factors. Symmetry alone fixed a ratio of measurable quantities — and the measurement agreed. This is why Chapter 3 spends its time on symmetries before Chapter 5 writes down a single Lagrangian.

The converse is the real power: a measured ratio that violates 9 : 2 : 1 would prove isospin is broken. Selection rules are falsifiable.

⚙️ Engineer’s bridge — selection rules are a type system

A coefficient of zero is not a small number; it is a forbidden transition. The triangle rule and

M = m_1 + m_2

are compile-time constraints: states that do not typecheck have amplitude exactly zero, before any dynamics is considered. In the matrix view of the calculator, every cell outside a block is structurally zero — not “small”, zero.

That is why physicists reach for symmetry first. It prunes the space of possible outcomes for free, in the same way a type system rules out whole classes of programs before you run one.

Where it breaks: a type system is sound — it rejects only programs that would genuinely fail — and selection rules are sound only relative to the interaction assumed. A rule derived from angular-momentum conservation holds always; one derived from parity or isospin holds for the strong interaction and is broken by the weak one, so the same “type error” that forbids a strong decay merely makes a weak decay slow. Pruning the search space is free; deciding which rules are in force is the part that requires physics.

🔑 If you remember only three things

The printed table gives squared coefficients. Take the square root and restore the sign from the heading, or every interference term comes out wrong.
The order of $j_1$ and $j_2$ is not free. Swapping them multiplies the coefficient by $(-1)^{j_1+j_2-J}$ , which is invisible in a rate and fatal in an amplitude.
A zero in the table is a selection rule, not a small number. It says the final state cannot be reached at all, and no amount of energy changes that.

Where this is used

§3.9 Summing isospins derives the isospin machinery this page assumes and works more examples.
§4.2 The 3/2⁺ baryons uses exactly the Δ analysis above.
Appendix 5 supplies the other half of the angular toolkit: the spherical harmonics and d-functions that turn a spin hypothesis into an angular distribution you can fit.

✅ Check yourself — adding angular momenta

0/5 answered · 0 correct

1.Combine $j_1 = 3/2$ with $j_2 = 1$ . Which total angular momenta are allowed, and does the dimension count work out?
Hint: Triangle rule, then check $(2j_1+1)(2j_2+1) = \sum_J (2J+1)$ .
2.In the calculator's matrix view, most cells are exactly zero and the non-zero ones form blocks. What forces that structure?
3.The $\pi^- p$ state is $\sqrt{1/3}\,|3/2,-1/2\rangle - \sqrt{2/3}\,|1/2,-1/2\rangle$ . On the Δ(1232) resonance, what fraction of the $\pi^+ p$ elastic cross-section does $\pi^- p \to \pi^- p$ have?
4.Which statement about the Clebsch–Gordan table is correct?
5.An experiment measures $\sigma(\pi^+p) : \sigma(\pi^-p) = 3.0 : 1.0$ at the Δ peak but 1.8 : 1.0 well above it. What is the natural reading?