Angular Momentum is Rotation

Put this down and try it with an actual book, or your phone. Lying flat, spine toward you: tip it 90° forward (rotate about the horizontal axis running left-right), then spin it 90° on the table (rotate about the vertical axis). Note which way the cover faces. Now start over and do the same two moves in the other order. The cover ends up facing a different way.

This has nothing to do with quantum mechanics yet — it is a fact about space itself, as true for a thrown brick as for an electron, and most people have never noticed it because we rarely rotate things twice about different axes and compare. Chapter 5 found that time-translations are generated by energy; rotations, it turns out, are generated by something too — and the something is not commutative, which is going to matter enormously. Watch the mismatch appear, and watch where its own axis points as the angle grows:

As in Chapter 5, name the generator: a rotation by angle $\theta$ about axis $\hat{n}$ is produced by a Hermitian operator ${\hat{J}}_{\hat{n}}$,

called the angular momentum along $\hat{n}$ — the same name classical mechanics uses, and for good reason: for orbital motion $\hat{J}$ will turn out to be exactly $\hat{\vec{r}}\times \hat{\vec{p}}$. What algebra must ${\hat{J}}_x,{\hat{J}}_y,{\hat{J}}_z$ obey? Ordinary 3×3 rotation matrices settle it. Write $R_i(\theta )=e^{\theta K_i}$ for the classical rotation about axis $i$ generated by the real antisymmetric matrix $K_i$; direct multiplication of the familiar $3\times 3$ generators gives, cleanly and checkably by hand,

Rotating a physical system by $R_1$ then $R_2$ must give the same result whether the system is a book, a brick, or an electron — so the quantum operators $\hat{U}(R)$ must compose exactly the way the classical $R$ do. Expand $\hat{U}$ to second order in a small angle $\epsilon$ and match terms against the geometric fact that composing a small rotation about $x$ with one about $y$ differs, to order ${\epsilon }^2$, from the reverse order by precisely a small rotation about $z$ — exercise 1 walks the algebra — and the $\hslash /i$ needed to keep $\hat{J}$ Hermitian while $\hat{U}$ stays unitary lands exactly here:

This is the noncommutativity you felt with the book, now the defining law of an entire algebra. Everything else in this chapter is a corollary.

Chapter 7 solved one commutator, $[\hat{a},{\hat{a}}^{†}]=1$, with a ladder. Equation (9.3) is three commutators at once, but the same trick works — it just needs a referee to keep track of which axis is “energy” this time. ${\hat{J}}^2$ commutes with every component (check it: each term individually fails to commute, but the failures cancel in the sum), so pick ${\hat{J}}_z$ as the observable to diagonalize alongside it, and build raising and lowering operators exactly as before:

${\hat{J}}_{+}$ raises the ${\hat{J}}_z$ eigenvalue by $\hslash$, exactly as ${\hat{a}}^{†}$ raised the oscillator rung — but now the ladder is bounded on both ends (angular momentum along one axis cannot exceed the total), so there must be a top state $|j,j⟩$ with ${\hat{J}}_{+}|j,j⟩=0$ and a bottom state $|j,-j⟩$ with ${\hat{J}}_{-}|j,-j⟩=0$. Apply ${\hat{J}}_{-}{\hat{J}}_{+}$ to the top state using (9.4) and the answer is forced:

Climbing from $-j$ to $j$ in unit steps only closes up if $2j$ is a non-negative integer: $j=0,\frac{1}{2},1,\frac{3}{2},2,\dots$. That half-integer option is new — the oscillator's ladder had no such twin bound and only ever gave one family. Here is where the mathematics first admits the possibility of spin, before we have said one physical word about it.

For a particle moving in ordinary space, angular momentum really is $\hat{L}=\hat{\vec{r}}\times \hat{\vec{p}}$, and in position representation the operator that generates rotation about $z$ is simply ${\hat{L}}_z=-i\hslash \partial /\partial \phi$. Its eigenfunctions are $e^{im\phi }$ — but a wavefunction must return to the same value after a full turn, $\phi \to \phi +2\pi$, and $e^{im(\phi +2\pi )}=e^{im\phi }$ only if $m$ is an integer. Since $m$ reaches all the way up to $j$ itself, orbital $l$ (the traditional letter) is stuck at $0,1,2,\dots$ — half-integers are mathematically legal but physically forbidden for orbital motion, by the sheer requirement that a wave close up on itself.

Spin has no wavefunction winding around a physical angle — it is an internal degree of freedom, obeying the same algebra (9.3) with no such constraint attached. Nature is not shy about using the other option: electrons, protons, quarks — spin ½; photons and gravitons — spin 1 and 2 (integer, like orbital motion, and for a directly related reason we will meet in Part IV). One algebra, and space itself decides which representations get to describe motion through it, while something else entirely — Nature's particle catalogue — decides which get to describe the particles.

Chapter 2 handed you a rule — $P={\cos }^2(\Theta /2)$ — and a promise: turn a spin-1/2 all the way around and it comes back wrong. We can finally show it. For $j=1/2$, the spin operators are ${\hat{J}}_i=\frac{\hslash }{2}{\sigma }_i$, and because $(\hat{n}\cdot \vec{\sigma }{)}^2=1$, the rotation operator sums to closed form — no infinite series required:

These spin-1/2 rotation operators have a name that recurs throughout the rest of the book. Each one is a $2\times 2$ matrix that is unitary (it preserves probability) and has determinant 1; the set of all of them is the group SU(2) — “Special Unitary, 2×2.” These are precisely the rotations of a spin-1/2 spinor, the two-component object a qubit lives in. The ordinary rotations of 3D space — turning a book, orienting a compass — form a different group, SO(3) (the $3\times 3$ orthogonal matrices of determinant 1). The link between them is the punchline of this section: SU(2) is a double cover of SO(3). “Double cover” means the map from SU(2) down to physical rotations is exactly two-to-one: for every physical rotation there are two SU(2) matrices, $\hat{U}$ and $-\hat{U}$, that both realize it. Trace a single 360° turn in real space and SU(2) travels only halfway around its own loop, arriving at $-1$; only a second 360° completes the SU(2) loop back to $+1$. That is the whole reason a spinor needs 720° to come home. The two panels below make it concrete — first the experiment that measures a spin at all, then the sign flip itself:

Set $\theta =2\pi$: $\cos \pi =-1$, $\sin \pi =0$, so $\hat{U}(\hat{n},2\pi )=-1$. A full turn returns every spin-1/2 ket to minus itself. Only at $\theta =4\pi$ does $\hat{U}=+1$. Compare with a $j=1$ system: ${\hat{J}}_z$'s eigenvalues are the integers $-1,0,1$, so $e^{-i2\pi m}=1$ for every one of them — integer-$j$ systems return home at the ordinary $2\pi$. The qubit's mystery was never a property of qubits; it is a property of the representation $j=1/2$ of the rotation group, which spin-1/2 particles happen to carry. Watch it directly:

This is not bookkeeping without teeth: a real interferometer measures it, because interference (Chapter 1's whole subject) is exactly the tool that turns an invisible overall phase into a visible fringe shift when you compare a rotated beam against an untouched one.

Chapter 6 caged a wave between two walls and got a ladder of standing waves, $\sin (n\pi x/L)$. Cage a wave on the surface of a sphere instead — demand that it be single-valued and finite everywhere, including the poles — and the same idea returns with the same name: standing waves that fit the boundary. They are the spherical harmonics $Y_l^m(\theta ,\phi )$, the simultaneous eigenfunctions of ${\hat{L}}^2$ and ${\hat{L}}_z$:

The two factors split the two angles between them. The $e^{im\phi }$ is the same azimuthal winding around the polar axis we have already met; the associated-Legendre function $P_l^m(\cos \theta )$ carries the polar profile — how the wave rises, falls, and wiggles as you travel along a meridian from the north pole down to the south. It is nothing more exotic than the standing-wave shape of the box (Chapter 6) bent to fit between the two poles, and every latitude at which it changes sign is one nodal circle.

The count of nodes tells the whole story, exactly as it did for the box: $l-|m|$ nodal circles of latitude, and — in the real part the viewer below draws — $|m|$ nodal planes containing the $z$-axis. Count them carefully: those are $|m|$ nodal great circles through the poles, which is $2|m|$ meridians in the geographic sense, because each great circle is two meridians joined at the poles. (The full complex $Y_l^m$ has no azimuthal nodes at all — only its real and imaginary parts do, a point the density map is about to make loudly.) More curvature, more ${\hat{L}}^2$. Play with real ones:

Look once more at the pair of maps for any $m\ne 0$. The real part ripples $|m|$ times around every line of latitude — a traveling wave circling the pole. The density is dead flat in $\phi$. This is Chapter 5's stationary-state lesson wrapped around a sphere: a definite-$m$ state has a sharp rate of phase winding but no preferred location on the equator — asking “where on the ring is the electron?” has no more of an answer than asking where in time a stationary state's clock hand points. Only a superposition of different $m$'s — a wave packet in $\phi$ — would localize and precess, the direct sequel to Chapter 7's “single rungs are frozen, superpositions swing.”

The lab Rust-QP/ch09-angmom constructs the $(2j+1)\times (2j+1)$ matrices ${\hat{J}}_x,{\hat{J}}_y,{\hat{J}}_z$ directly from the ladder formulas of equation (9.4) — no differential equation, no calculus, just the raising/lowering matrix elements written down and assembled:

1/// Build Jx, Jy, Jz for angular momentum j, given as twice_j = 2j (an
2/// integer, so half-integer j is represented exactly).
3fn build_j(twice_j: i32) -> (Mat, Mat, Mat) {
4    let j = twice_j as f64 / 2.0;
5    let n = (twice_j + 1) as usize;
6    // basis ordered m = j, j-1, ..., -j  (index 0 = top of the ladder)
7    let m_of = |i: usize| j - i as f64;
8
9    let mut jz = Mat::zeros(n);
10    for i in 0..n {
11        jz.set(i, i, c(m_of(i), 0.0));
12    }
13    // J+ |j,m> = sqrt(j(j+1)-m(m+1)) |j,m+1>  -- raises m, so moves index i -> i-1
14    let mut jp = Mat::zeros(n);
15    for i in 1..n {
16        let m = m_of(i); // starting state
17        let coeff = (j * (j + 1.0) - m * (m + 1.0)).sqrt();
18        jp.set(i - 1, i, c(coeff, 0.0));
19    }
20    // J- is the transpose of J+ (real, since our basis and phases are real)
21    let mut jm = Mat::zeros(n);
22    for i in 0..n - 1 {
23        let m = m_of(i);
24        let coeff = (j * (j + 1.0) - m * (m - 1.0)).sqrt();
25        jm.set(i + 1, i, c(coeff, 0.0));
26    }
27
28    let jx = jp.add(&jm).scale(c(0.5, 0.0));
29    let jy = jp.sub(&jm).scale(c(0.0, -0.5)); // (J+ - J-) / (2i)
30    (jx, jy, jz)
31}

Then it settles Section 9.5's claim without any approximation at all. Because ${\hat{J}}_z$ is already diagonal by construction, exponentiating it costs nothing — the rotation operator $R_z(\theta )$ is exact to floating-point precision, no truncated series, no numerical integration. But be honest about what that buys: once build_j has written the half-integers onto the diagonal, exponentiating a diagonal matrix cannot give any answer but $-1$. That check can catch a coding slip; it cannot catch a wrong physical claim. So the lab computes the same two rotations a second time along a route that can fail — a scaling-and-squaring Taylor series for $e^{-i\theta {\hat{J}}_x}$ on the non-diagonal ${\hat{J}}_x$, which lands on $-1$ at $2\pi$ only if the ladder coefficients give ${\hat{J}}_x$ the same spectrum $m=-j,\dots ,j$. Break one coefficient and that referee breaks; the diagonal one never notices:

1// R_z(theta) = exp(-i theta Jz): Jz is diagonal, so this is EXACT
2    let r_z = |theta: f64| -> Mat {
3        let mut m = Mat::zeros(n);
4        for i in 0..n {
5            let mval = jz.get(i, i).re;
6            m.set(i, i, c(0.0, -theta * mval).exp());
7        }
8        m
9    };
10    let expected_sign = if twice_j % 2 == 0 { 1.0 } else { -1.0 };
11    let rz_2pi = r_z(2.0 * PI);
12    let rz_4pi = r_z(4.0 * PI);
13    let rz_2pi_err = rz_2pi.dist(&Mat::identity(n).scale(c(expected_sign, 0.0)));
14    let rz_4pi_err = rz_4pi.dist(&Mat::identity(n));
15
16    // The same two rotations again, along a route that can actually fail:
17    // exp(-i theta Jx) summed as a series on the NON-diagonal Jx. Getting
18    // -1 at 2*pi requires Jx's eigenvalues to be exactly m = -j..j, which
19    // is a statement about the ladder coefficients, not about arithmetic on
20    // a diagonal we ourselves wrote down.
21    let rx_2pi = expm(&jx.scale(c(0.0, -2.0 * PI)));
22    let rx_4pi = expm(&jx.scale(c(0.0, -4.0 * PI)));
23    let rx_2pi_err = rx_2pi.dist(&Mat::identity(n).scale(c(expected_sign, 0.0)));
24    let rx_4pi_err = rx_4pi.dist(&Mat::identity(n));
25
26    // …
27
28        Referee {
29            name: "R_z(2π) − (±1)·1, exact diagonal route".into(),
30            value: worst_rz_2pi,
31            tol: 1e-15,
32            pass: worst_rz_2pi < 1e-15,
33        },
34        Referee {
35            name: "R_z(4π) − 1, exact diagonal route".into(),
36            value: worst_rz_4pi,
37            tol: 2e-15,
38            pass: worst_rz_4pi < 2e-15,
39        },
40        Referee {
41            name: "exp(−i2πJx) − (±1)·1, series route".into(),
42            value: worst_rx_2pi,
43            tol: 6e-15,
44            pass: worst_rx_2pi < 6e-15,
45        },
46        Referee {
47            name: "exp(−i4πJx) − 1, series route".into(),
48            value: worst_rx_4pi,
49            tol: 1e-14,
50            pass: worst_rx_4pi < 1e-14,
51        },

A second, independent piece of the lab builds $Y_l^m$ from the associated-Legendre three-term recurrence and checks it by quadrature over the sphere — Gauss-Legendre in $\cos \theta$ (exact for these polynomial integrands) and the periodic rectangle rule in $\phi$ — testing normalization, orthogonality, and the ${\hat{L}}_z$ eigenvalue via a finite-difference $\phi$-derivative, catching any stray sign or off-by-one that the construction might have smuggled in. Every number below, tolerances included, is read live from the JSON the run just wrote:

1. (F) Fill in the ε² derivation sketched in 9.2: expand ${\hat{U}}_x(\epsilon ){\hat{U}}_y(\epsilon )$ and ${\hat{U}}_y(\epsilon ){\hat{U}}_x(\epsilon )$ to second order, subtract, and match the result against ${\hat{U}}_z({\epsilon }^2)-1$ to recover equation (9.3) with its ħ intact.
2. (F) Verify, from (9.4) alone, that ${\hat{J}}_{+}|j,m⟩=\hslash \sqrt{j(j+1)-m(m+1)}|j,m+1⟩$. Then use it to find every matrix element of ${\hat{J}}_x$ for $j=1$ and check $[{\hat{J}}_x,{\hat{J}}_y]=i\hslash {\hat{J}}_z$ by hand on the resulting 3×3 matrices.
3. (C) In the rotation widget, find the smallest slider setting at which the two orders disagree by more than 30°, and say whether the mismatch grows faster or slower than θ itself at small θ. Then check the small-angle story where it is supposed to break: the widget also reports the tilt of the residual rotation's own axis away from $\hat{z}$. It is a few degrees for a few degrees of θ, but by θ = 90° it is well over 50°, and the printed θ² estimate has by then overshot the true mismatch badly. Now push the slider all the way to 180°: the mismatch collapses to zero. Explain that by writing $R_x(\pi )$ and $R_y(\pi )$ as 3×3 matrices and asking whether those two particular rotations commute. Which of the two claims — “the mismatch is a rotation about z” or $[{\hat{J}}_x,{\hat{J}}_y]=i\hslash {\hat{J}}_z$ — is an exact statement, and which is only the $\theta \to 0$ limit of one?
4. (C) In the spherical-harmonics viewer, set l=3. On the $\mathrm{Re}(Y)$ map count the nodal circles of latitude at m=0, m=1, m=3 and verify $l-|m|$ each time; then count the sign changes as you travel once around a line of latitude and check that you get $2|m|$ of them — $|m|$ nodal planes through the z-axis, each crossed twice. Which (l,m) gives the “fattest doughnut around the equator, no polar caps” shape, and why does that match $m=\pm l$?
5. (F, hard) Prove the single-valuedness argument rigorously in reverse: show that if you allowed half-integer orbital $l$, the wavefunction $e^{i\phi /2}$ would be double-valued on the sphere (differs by sign after$\phi \to \phi +2\pi$), and explain precisely why that is fatal for a position-space probability amplitude but not fatal for an abstract spin ket (which has no φ-dependence to be double-valued in).
6. (F, hard) Derive equation (9.6) from scratch: expand $e^{-i\theta \hat{n}\cdot \vec{\sigma }/2}$ as a power series, use $(\hat{n}\cdot \vec{\sigma }{)}^2=1$ to collapse it into even and odd powers of $\theta /2$, and resum both as sine and cosine series. Then apply your formula to $\hat{n}=\hat{z}$ and recover Chapter 2's equation (2.2) directly — the half-angle law was this equation all along.
7. (P) Extend the Rust lab to compute the full Wigner rotation matrix $d_{m^{\prime }m}^j(\beta )=⟨j,m^{\prime }|e^{-i\beta {\hat{J}}_y/\hslash }|j,m⟩$ for j=1/2 and j=1 via the matrix exponential (reuse the exact-diagonal trick by first diagonalizing ${\hat{J}}_y$ — it shares eigenvalues with ${\hat{J}}_z$). Verify $d_{1/2,1/2}^{1/2}(\beta )=\cos (\beta /2)$ against equation (9.6).
8. (P, hard) Simulate a crude “neutron interferometer”: prepare two copies of $|+x⟩$, rotate one by θ about z using your Rust lab's exact $R_z(\theta )$, and compute the recombined intensity $|⟨+x|(1+{\hat{U}}_z(\theta ))|+x⟩{|}^2$ as a function of θ from 0 to 4π. Confirm the period is 4π, not 2π, and locate the dip that a naive “spin is just a little arrow” picture would never predict.