Rotation Operators and SU(2)

Three articles ago the algebra $[\hat J_i,\hat J_j]=i\hbar\,\epsilon_{ijk}\hat J_k$ was derived and then set loose: the ladder built its spectrum, the spherical harmonics realized the integer rungs on the sphere, and the azimuthal article evicted the half-integers from orbital motion with a promise — the real resolution is group theory. This article keeps that promise. Exponentiating the generators turns the algebra into actual finite rotations, $\hat U(\hat{\mathbf n},\theta)=e^{-i\theta\,\hat{\mathbf n}\cdot\hat{\mathbf J}/\hbar}$ , and the collection of these operators is a group — the same multiplication table as physical rotations, almost. The "almost" is the payoff. For spin- $\tfrac12$ the rotation operators are the $2\times2$ unitary matrices of determinant one, the group $SU(2)$ , and there are exactly two of them for every physical rotation: $U$ and $-U$ turn vectors identically. The map $SU(2)\to SO(3)$ is two-to-one — a double cover. Rotate a spin- $\tfrac12$ state by a full $2\pi$ and it comes back as minus itself; only $4\pi$ restores it. That minus sign is not bookkeeping. Split a neutron beam, precess the spin in one arm through $2\pi$ , recombine — the interference fringes invert, and they return only after $4\pi$ . The experiment was done in 1975, and the field-path integral that buys one full fringe period comes out of this article's equations at about $1.5\times10^{-4}$ T·m. Mastering this buys the modern currency of rotation: why integer $\ell$ lives on spheres while half-integer $j$ needs spinors, why every single-qubit gate is an $SU(2)$ element, and the axis-angle-composition machinery the Wigner D-matrices will industrialize next.

Key takeaways

Finite rotations are exponentials of the angular momentum generators: $\hat U(\hat{\mathbf n},\theta)=e^{-i\theta\,\hat{\mathbf n}\cdot\hat{\mathbf J}/\hbar}$ ; unitary because $\hat{\mathbf J}$ is Hermitian.
For spin- $\tfrac12$ the closed form is $U=\cos\tfrac{\theta}{2}\,\mathbb 1-i\sin\tfrac{\theta}{2}\,(\hat{\mathbf n}\cdot\boldsymbol\sigma)$ — the algebraic identity owned by Spin-1/2 and the Pauli Matrices; here it becomes the group $SU(2)$ , geometrically the 3-sphere $S^3$ .
Conjugation $U(\hat{\mathbf v}\cdot\boldsymbol\sigma)U^\dagger=(R\hat{\mathbf v})\cdot\boldsymbol\sigma$ defines a map $SU(2)\to SO(3)$ that is onto with kernel $\{\pm\mathbb 1\}$ : exactly two-to-one, the double cover.
On any $|j,m\rangle$ multiplet a $2\pi$ rotation gives $(-1)^{2j}$ : invisible for integer $j$ , a sign flip for half-integer $j$ — measurable as a $4\pi$ -periodic interference pattern (neutrons, 1975).
Orbital wavefunctions carry single-valued $SO(3)$ representations, hence integer $\ell$ only; spinors carry $SU(2)$ representations, where half-integers are legal.
Differentiating the group at the identity returns the algebra: group and algebra encode the same local structure, but topology separates them globally.
Key idea: the angular momentum algebra exponentiates to two different groups' worth of physics — $SO(3)$ for space, its double cover $SU(2)$ for state vectors — and the factor-two mismatch is where spin lives.

The core insight

Angular momentum generates rotations, $\hat U(\hat{\mathbf n},\theta)=e^{-i\theta\,\hat{\mathbf n}\cdot\hat{\mathbf J}/\hbar}$ ; because $SU(2)$ covers $SO(3)$ twice, the same algebra produces both single-valued integer- $j$ rotations and spinor half-integer- $j$ rotations that need $4\pi$ to come home.

Where this lives today

Every quantum computer executes this article constantly: a single-qubit gate is an $SU(2)$ element, implemented exactly as Example 1's neutron — a controlled Larmor precession through a chosen axis and angle. The standard gates are points of $S^3$ : the Hadamard is a $\pi$ turn about $(\hat{\mathbf x}+\hat{\mathbf z})/\sqrt2$ (equal to $-iH$ , the $-i$ the ever-present global phase), the phase and $\pi/8$ gates partial $z$ -turns. Compilers synthesize an arbitrary $U$ from a discrete gate set by navigating $S^3$ — quaternion geometry as engineering. The same bookkeeping runs flight controllers and game engines (gimbal-lock-free attitude via unit quaternions); Berry's phase, the continuum cousin of the $\pm U$ ambiguity, is measured routinely in interferometry.

The physical picture

A rotation operator is a machine assembled from infinitesimal parts: rotating by $\mathrm d\theta$ about $\hat{\mathbf n}$ barely disturbs a state — $\hat U\approx\mathbb 1-\tfrac{i}{\hbar}\mathrm d\theta\,\hat{\mathbf n}\cdot\hat{\mathbf J}$ — and a finite turn is this step compounded, which is what an exponential is. The generator doctrine is Unitary Operators and Transformations' theorem; Orbital Angular Momentum Operators showed the orbital instance, $e^{-i\alpha\hat L_z/\hbar}$ shifting $\phi\to\phi-\alpha$ . What is new here is taking the collection of these operators seriously as a group and asking what shape it has. The shape question has a tabletop answer. Hold a plate flat on your palm and rotate it by $2\pi$ : the plate is back, your arm is twisted. A further $2\pi$ in the same direction — the part that surprises everyone once — untwists it completely. Loops of rotations come in exactly two kinds: continuously undoable ( $4\pi$ , or none) and not ( $2\pi$ ). Vectors are blind to the difference. Spinors ride the loop and report which kind it was, with a sign.

Figure

The double cover — two SU(2) points over every rotation, and the 2π loop that cannot be undone

The picture to carry: as a shape, $SU(2)$ is the three-dimensional sphere $S^3$ — every element is $U=a_0\mathbb 1-i\,\mathbf a\cdot\boldsymbol\sigma$ with $a_0^2+|\mathbf a|^2=1$ , four real numbers on a unit sphere. (Hamilton's quaternion coordinates; Angular Momentum Commutation Relations exhibited the units $-i\sigma_k$ as his $i,j,k$ .) $SO(3)$ is what you get by gluing each point of $S^3$ to its antipode, since $U$ and $-U$ rotate vectors identically. That gluing is precisely what makes the $2\pi$ loop unshrinkable downstairs — it is an honest closed loop in $SO(3)$ , but upstairs it is an open path from $U$ to $-U$ . A spinor state, which transforms by $U$ rather than by $R$ , feels the open endpoint: the sign.

From generator to group

The construction of the exponential from infinitesimal steps is the generator doctrine of Unitary Operators and Transformations, and the series grammar was fixed back in Canonical Commutation Relation for translations. Applied to rotations it reads: a rotation by $\theta$ about the fixed axis $\hat{\mathbf n}$ is $N$ compounded rotations by $\theta/N$ ,

\hat U(\hat{\mathbf n},\theta)=\lim_{N\to\infty}\Big(\mathbb 1-\tfrac{i}{\hbar}\tfrac{\theta}{N}\,\hat{\mathbf n}\cdot\hat{\mathbf J}\Big)^{N}=\exp\!\Big(-\tfrac{i}{\hbar}\,\theta\,\hat{\mathbf n}\cdot\hat{\mathbf J}\Big)

where: $\hat{\mathbf n}\cdot\hat{\mathbf J}=n_x\hat J_x+n_y\hat J_y+n_z\hat J_z$ is the generator of turns about $\hat{\mathbf n}$ , and the limit is legitimate because a single fixed generator commutes with itself — no ordering subtleties arise until you change axes mid-rotation, which is exactly where the group structure will bite. Hermiticity of $\hat{\mathbf J}$ makes $\hat U$ unitary: probabilities survive the turn, as a symmetry demands. Two consequences come free. About the quantization axis it is diagonal on any multiplet,

\hat U(\hat{\mathbf z},\theta)\,|j,m\rangle=e^{-im\theta}\,|j,m\rangle

where: each rung just picks up the winding phase $e^{-im\theta}$ — the operator face of the azimuthal factor $e^{im\phi}$ that Eigenfunctions of L_z and the Azimuthal Quantization diagonalized in space. And because $[\hat{\mathbf J}^2,\hat J_i]=0$ (the Casimir card of Angular Momentum Commutation Relations), every $\hat U$ commutes with $\hat{\mathbf J}^2$ : rotations never change $j$ , they only mix the $2j+1$ values of $m$ . How they mix them, element by element, is the next article's business — the Wigner D-matrices. For spin- $\tfrac12$ the exponential collapses to a closed form. The algebraic identity

e^{-i(\theta/2)\,\hat{\mathbf n}\cdot\boldsymbol\sigma}=\cos\tfrac{\theta}{2}\,\mathbb 1-i\sin\tfrac{\theta}{2}\,(\hat{\mathbf n}\cdot\boldsymbol\sigma)

where: the series splits on $(\hat{\mathbf n}\cdot\boldsymbol\sigma)^2=\mathbb 1$ — is derived and owned by Spin-1/2 and the Pauli Matrices; this article consumes it as the explicit matrix of $\hat U(\hat{\mathbf n},\theta)$ at $j=\tfrac12$ and reads off its group-theoretic credentials instead. Unitarity is manifest. The determinant is $\cos^2\tfrac\theta2+\sin^2\tfrac\theta2\,|\hat{\mathbf n}|^2=1$ . So the spin- $\tfrac12$ rotation operators are exactly the unitary $2\times2$ matrices of unit determinant: the group $SU(2)$ , coordinatized as

U=a_0\,\mathbb 1-i\,\mathbf a\cdot\boldsymbol\sigma,\qquad a_0=\cos\tfrac{\theta}{2},\quad \mathbf a=\hat{\mathbf n}\,\sin\tfrac{\theta}{2},\qquad a_0^2+|\mathbf a|^2=1

where: the constraint says the group manifold is the unit sphere in four dimensions, $S^3$ — simply connected, every loop shrinkable. Note what the half-angle does to the coordinates: as the physical angle $\theta$ runs $0\to2\pi$ , the point $(a_0,\mathbf a)$ travels only half way around a great circle of $S^3$ , from $(1,\mathbf 0)$ to $(-1,\mathbf 0)$ . The full physical turn lands on $-\mathbb 1$ , not $\mathbb 1$ .

The two-to-one theorem

Now the central claim: every $SU(2)$ element is a rotation of space, and exactly two of them are each rotation.

Derivation AConjugation rotates vectors and the kernel is plus or minus the identity

The goal is the covering homomorphism: acting on the Pauli vector by conjugation, $V\mapsto UVU^\dagger$ , sends each $U\in SU(2)$ to a genuine $SO(3)$ rotation $R(U)$ , the map respects group multiplication, it reaches every rotation, and exactly the two elements $\pm\mathbb 1$ act trivially. The strategy is to encode 3-vectors as traceless Hermitian matrices — where lengths hide inside determinants — and let unitarity do the work. Watch how the two-to-one conclusion needs nothing but the question "which $U$ commutes with all of $\boldsymbol\sigma$ ?".

Step 1. Encode. To the real vector $\mathbf v$ associate $V=\mathbf v\cdot\boldsymbol\sigma$ . This is a bijection between $\mathbb R^3$ and traceless Hermitian $2\times2$ matrices, and it stores the geometry: $\det V=-|\mathbf v|^2$ .

Step 2. Conjugate. For $U\in SU(2)$ set $V'=UVU^\dagger$ . Then $V'$ is again Hermitian and traceless (trace is cyclic), so $V'=\mathbf v'\cdot\boldsymbol\sigma$ for some real $\mathbf v'$ — a linear function of $\mathbf v$ . Write $\mathbf v'=R(U)\mathbf v$ ; taking components with the trace inner product gives the explicit matrix $R_{ik}(U)=\tfrac12\mathrm{Tr}\big[\sigma_i\,U\sigma_k U^\dagger\big]$ .

Step 3. It is a rotation. $\det V'=\det U\det V\det U^\dagger=\det V$ , so $|\mathbf v'|=|\mathbf v|$ : $R(U)$ is orthogonal. Continuity settles the determinant sign: at $U=\mathbb 1$ , $R=\mathbb 1$ with $\det R=+1$ , and $SU(2)\cong S^3$ is connected, so $\det R(U)=+1$ throughout. Hence $R(U)\in SO(3)$ . Why conjugation and not left multiplication: states transform by one-sided action $|\chi\rangle\to U|\chi\rangle$ , but operators — and $\mathbf v\cdot\boldsymbol\sigma$ is an operator — transform by conjugation. The vector of expectation values then rotates by $R(U)$ ; the spinor itself rides $U$ .

Step 4. It is a homomorphism onto. $R(U_1U_2)=R(U_1)R(U_2)$ follows by inserting $U_2U_2^\dagger$ inside the conjugation. Ontoness is explicit: $U(\hat{\mathbf n},\theta)$ maps to the rotation about $\hat{\mathbf n}$ by $\theta$ — checking the $z$ -axis case with the trace formula gives exactly the standard $R_z(\theta)$ , columns $(\cos\theta,\sin\theta,0)$ , $(-\sin\theta,\cos\theta,0)$ , $(0,0,1)$ .

Step 5. The kernel. $R(U)=\mathbb 1$ means $UVU^\dagger=V$ for every traceless Hermitian $V$ , i.e. $U$ commutes with all three Pauli matrices. Since $\{\mathbb 1,\boldsymbol\sigma\}$ spans all $2\times2$ matrices, $U$ commutes with everything, so $U=\lambda\mathbb 1$ ; unitarity and $\det U=1$ force $\lambda=\pm1$ .

Step 6. Conclusion. The preimage of each rotation is a pair $\{U,-U\}$ : the map is exactly two-to-one, and $SU(2)$ — simply connected — is the universal double cover of $SO(3)\cong S^3/\{\pm1\}\cong\mathbb{RP}^3$ . Verified: numerically on 20 random SU(2) elements — R(U) orthogonal, det +1, R(−U)=R(U), and R(U₁U₂)=R(U₁)R(U₂), all to 1e-13; the z-axis case reproduces the standard R_z(θ) exactly (verify script, section 4).

The kernel is the whole story of spin in miniature. Downstairs, $U$ and $-U$ are the same physical rotation; upstairs they are different operators. Any object that transforms by $R$ — positions, fields, expectation values — cannot tell them apart. Any object that transforms by $U$ — a two-component spinor — can.

The sign theorem: what a full turn does

Go deeper · Derivation B · 2 minA 2π rotation multiplies a multiplet by (−1) to the power 2j

Derivation BA 2π rotation multiplies a multiplet by (−1) to the power 2j

The goal is the cleanest theorem in the chapter: on any angular momentum multiplet $|j,m\rangle$ , a full $2\pi$ rotation about any axis is the scalar $(-1)^{2j}$ — the identity for integer $j$ , minus the identity for half-integer $j$ . The strategy is disarmingly short: diagonalize along the rotation axis and read the phases; the only subtlety is why the answer cannot depend on the axis, which is one sentence of group theory. This card generalizes the spin- $\tfrac12$ fact that Spin Angular Momentum stated and scopes it: the sign is a property of the representation label $j$ , nothing else.

Step 1. Rotate about the quantization axis. On the rung $|j,m\rangle$ ,

e^{-i\,2\pi\hat J_z/\hbar}\,|j,m\rangle=e^{-i2\pi m}\,|j,m\rangle.

Step 2. Read the phase through the spectrum. The ladder theorem fixed $m\in\{-j,\dots,+j\}$ in unit steps, so every $m$ in the multiplet is an integer when $j$ is an integer, and a half-odd-integer when $j$ is. Hence $e^{-i2\pi m}=+1$ for all $m$ (integer $j$ ) or $-1$ for all $m$ (half-integer $j$ ):

e^{-i\,2\pi\hat J_z/\hbar}\Big|_{\,j}=(-1)^{2j}\,\mathbb 1.

Why every rung gets the same sign: the $m$ 's within one multiplet differ by integers, and $e^{-i2\pi(\Delta m)}=1$ — the sign is a class property of the multiplet, which is exactly what lets it escape as an overall scalar.

Step 3. Any axis. A rotation about $\hat{\mathbf n}$ is conjugate to one about $\hat{\mathbf z}$ : $\hat U(\hat{\mathbf n},2\pi)=\hat W\,\hat U(\hat{\mathbf z},2\pi)\,\hat W^\dagger$ for the unitary $\hat W$ that turns $\hat{\mathbf z}$ into $\hat{\mathbf n}$ . Conjugating a scalar changes nothing, so the $(-1)^{2j}$ holds for every axis. For $j=\tfrac12$ the closed form shows it directly: $\cos\pi\,\mathbb 1-i\sin\pi\,(\hat{\mathbf n}\cdot\boldsymbol\sigma)=-\mathbb 1$ , axis dependence gone.

Step 4. The dichotomy, stated as representation theory. Integer- $j$ multiplets represent $SO(3)$ honestly — the $2\pi$ loop acts as the identity, matching the physical rotation it is. Half-integer multiplets represent only $SU(2)$ : the $2\pi$ loop acts as $-\mathbb 1$ , distinguishing the two elements upstairs over the identity rotation. On rays (states up to global phase) the half-integer representations still project consistently — which is why nature is allowed to use them at all. Verified: numerically, e^{−i2πJ_z/ħ} computed by matrix exponential equals (−1)^{2j}·𝟙 for j = 1/2, 1, 3/2, 2, each to 1e-12; the j=1/2 any-axis forms at 2π and 4π check exactly (verify script, sections 3).

This theorem closes the chapter's ledger. Eigenfunctions of L_z and the Azimuthal Quantization showed orbital wavefunctions — functions on physical space, which only knows $SO(3)$ — must be single-valued, killing half-integer $\ell$ ; its "deep resolution" line pointed here. Spin Angular Momentum claimed the evicted half-integers for internal doublets and stated the $2\pi$ sign as spin's badge. Now both have their reason: which group a state's home space represents decides which $j$ 's are allowed. Sphere wavefunctions represent $SO(3)$ : integer $\ell$ , $(-1)^{2\ell}=+1$ always. Spinors represent $SU(2)$ : $j=\tfrac12$ legal, $(-1)^{2j}=-1$ its signature.

Figure

What a spinor reports as the physical angle winds — overlap with its initial self, from 0 to 4π

Closing the loop: the algebra from the group

The chapter opened with the algebra and built upward; the group can also be walked back down, and doing it once makes clear that nothing was smuggled in.

Go deeper · Derivation C · 1 minDifferentiating the group at the identity recovers the commutation relations

Derivation CDifferentiating the group at the identity recovers the commutation relations

The goal is to show the rotation group implies the angular momentum algebra: composing two infinitesimal rotations about different axes, forwards then backwards, does not cancel — and the leftover, at second order, is precisely $[\hat J_x,\hat J_y]=i\hbar\hat J_z$ . This closes the loop with Angular Momentum Commutation Relations, which owns the algebra itself: the commutator is not an extra postulate on top of rotations, it is the statement that rotations about different axes fail to commute by a rotation about the third. Watch the bookkeeping order: first-order terms cancel identically, so the expansion must be kept honestly to second order.

Step 1. Expand each factor to second order, $\hat U_i(\epsilon)=\mathbb 1-\tfrac{i\epsilon}{\hbar}\hat J_i-\tfrac{\epsilon^2}{2\hbar^2}\hat J_i^2+O(\epsilon^3)$ , and form the group commutator $\hat U_x^{-1}\hat U_y^{-1}\hat U_x\hat U_y$ .

Step 2. Multiply and collect. All first-order terms cancel in the forward–backward pattern; the squared terms cancel against the cross terms except for one antisymmetric survivor:

\hat U_x^{-1}\hat U_y^{-1}\hat U_x\hat U_y=\mathbb 1-\frac{\epsilon^2}{\hbar^2}\,[\hat J_x,\hat J_y]+O(\epsilon^3).

Step 3. Identify the survivor geometrically. Composing the same four rotations on space leaves a net rotation about $z$ by the second-order angle $\epsilon^2$ (a fact of $3\times3$ geometry, checkable by multiplying the matrices). For the operators to represent the group, the survivor must equal $\hat U_z(\epsilon^2)=\mathbb 1-\tfrac{i\epsilon^2}{\hbar}\hat J_z+\dots$ , forcing

[\hat J_x,\hat J_y]=i\hbar\,\hat J_z

and its cyclic siblings. The structure constants of the algebra are the second-order failure of rotations to commute. Why this direction matters: it certifies that "angular momentum" and "generator of rotations" are the same office. Anything that generates rotations — orbital, spin, total, isospin analogues — inherits the algebra wholesale, which is exactly the abstraction pivot the algebra article executed. Verified: symbolically at j=1/2 — the group commutator expanded to O(ε²) equals 𝟙 − ε²[J_x,J_y] with [S_x,S_y]=iS_z exact (verify script, section 6).

Limiting cases

case	result	meaning
$\theta\to0$	$\hat U\approx\mathbb 1-\tfrac{i}{\hbar}\theta\,\hat{\mathbf n}\cdot\hat{\mathbf J}$	the generator reappears as the derivative at the identity — group and algebra agree locally
$j=\tfrac12$ , any axis	$U=\cos\tfrac\theta2\,\mathbb 1-i\sin\tfrac\theta2\,\hat{\mathbf n}\cdot\boldsymbol\sigma$	all of $SU(2)$ , one formula; period $4\pi$
$\theta=2\pi$	$\hat U=(-1)^{2j}\,\mathbb 1$	integer $j$ blind, half-integer $j$ flips sign — the double cover made audible
about $\hat{\mathbf z}$	$\hat U\|j,m\rangle=e^{-im\theta}\|j,m\rangle$	pure winding phases; the azimuthal quantization in operator dress
expectation values	$\langle\boldsymbol\sigma\rangle\to R(U)\langle\boldsymbol\sigma\rangle$	whatever the spinor does, the measured arrow turns by the honest $SO(3)$ matrix

Common trapThe minus sign is a global phase, so it can never matter

Half right, and the half matters. For an isolated system rotated by $2\pi$ , the $-1$ multiplies the entire state — a global phase, unobservable, which is why half-integer representations are physically admissible at all. But rotate part of a superposition and the sign becomes a relative phase between branches: $|\text{arm I}\rangle+|\text{arm II}\rangle\to|\text{arm I}\rangle-|\text{arm II}\rangle$ , and the interference pattern inverts. That is precisely the neutron experiment below. The correct statement is: $-1$ is unobservable on a ray, observable in an interferometer — same theorem, two readings.

Worked examples

Example 1 — the 4π fringe: how much field-path buys one full period? Task: in a neutron interferometer, one arm passes through a transverse magnetic field region (field $B$ , path length $l$ ). Find the field-path integral $Bl$ for one full period of the interference contrast, for thermal neutrons of wavelength $\lambda=1.8$ Å. ( $|\gamma_n|=1.832\times10^8\ \mathrm{rad\,s^{-1}T^{-1}}$ .)

Step 1. Beam speed from the de Broglie relation: $v=h/(m_n\lambda)=6.626\times10^{-34}/(1.675\times10^{-27}\times1.8\times10^{-10})=\mathbf{2.20\times10^{3}\ m/s}$ .

Step 2. In the field the spin precesses at the Larmor rate $\omega=|\gamma_n|B$ for a transit time $t=l/v$ , accumulating the precession angle $\theta=|\gamma_n|Bl/v$ . The rotated-arm spinor is multiplied by $U(\hat{\mathbf n},\theta)$ , and for an unpolarized beam the interference term averages to $\cos(\theta/2)$ — the spinor half-angle, straight from the closed form.

Step 3. One full contrast period therefore needs $\theta=4\pi$ , not $2\pi$ : at $\theta=2\pi$ the overlap is $\cos\pi=-1$ (fringes inverted), and only $\theta=4\pi$ restores $+1$ . Solve for the field-path product:

Bl=\frac{4\pi v}{|\gamma_n|}=\frac{4\pi\times2198}{1.832\times10^{8}}=\mathbf{1.51\times10^{-4}\ T\,m}\;(=151\ \mathrm{G\,cm}).

Step 4. Reality check: Rauch's 1975 beam ran at $\lambda=1.82$ Å, where the formula gives $149\ \mathrm{G\,cm}$ — the experiment reported a measured period of $149\pm5\ \mathrm{G\,cm}$ . The double cover is not philosophy; it is a dial setting on a magnet. $Bl_{4\pi}\approx1.5\times10^{-4}$ T·m.

Example 2 — composing rotations the SU(2) way. Task: rotate by $90^\circ$ about $\hat{\mathbf x}$ , then $90^\circ$ about $\hat{\mathbf z}$ . What single rotation results?

Step 1. Multiply the $SU(2)$ elements (apply $x$ first, so it sits on the right): with $c=s=\cos45^\circ=1/\sqrt2$ ,

U=U_z\,U_x=(c-is\sigma_z)(c-is\sigma_x)=\tfrac12\,\mathbb 1-\tfrac{i}{2}(\sigma_x+\sigma_y+\sigma_z),

using $\sigma_z\sigma_x=i\sigma_y$ for the cross term — the algebra doing the geometry.

Step 2. Read the answer off the $S^3$ coordinates: $a_0=\cos\tfrac\theta2=\tfrac12$ gives $\theta=120^\circ$ , and $\mathbf a\propto(1,1,1)$ normalizes to $\hat{\mathbf n}=(1,1,1)/\sqrt3$ . Two quarter-turns about perpendicular axes compose into a $120^\circ$ turn about the cube diagonal.

Step 3. Cross-check downstairs: multiplying the $3\times3$ matrices $R_z(90^\circ)R_x(90^\circ)$ and comparing with the axis-angle formula for $120^\circ$ about $(1,1,1)/\sqrt3$ gives the same matrix, trace $1+2\cos120^\circ=0$ . The trace of the composite $2\times2$ was all we needed to find the angle — this "multiply two small matrices, read the half-angle from the trace" routine is why flight software and graphics engines store rotations as quaternions, i.e. as $SU(2)$ points.

Origins · 1 minFelix Klein and how this was discovered

Felix Klein

1849–1925 · Göttingen

The two-to-one shadow between $SU(2)$ and the rotations was mapped decades before quantum mechanics needed it. In the 1870s Klein — fresh from reorganizing geometry around group actions — studied the finite rotation groups of the Platonic solids and their preimages under exactly this article's covering map: working with the unit quaternions (Hamilton's $S^3$ , our $SU(2)$ ), he found each rotation group lifts to a binary group upstairs with twice the elements, where a $2\pi$ turn is a genuinely new element of order two. His 1884 Lectures on the Icosahedron built a whole theory — quintic equations, modular functions — on the binary icosahedral group of order 120 over the icosahedron's 60 rotations. That the doubled elements he catalogued would one day be occupied — by electrons, whose states ride $SU(2)$ rather than $SO(3)$ — he could not have guessed; when spin arrived in 1925–27, the double cover's mathematics was on the shelf, fifty years ripe. (His work on spinning tops belongs to the mechanics track and is told there.)

History

1843

— Hamilton carves the quaternion relations: the coordinates of SU(2), eighty years early

1876

— Klein studies the binary polyhedral groups — finite rotation groups lifted through the double cover

1884

— Klein's Lectures on the Icosahedron systematize the two-to-one correspondence

1925

— Spin is proposed; half-integer angular momentum demands the covering group

1927

— Pauli's spin matrices realize SU(2) in physics; Weyl's group-theory text names the structure for quantum mechanics

1975

— Rauch and Werner independently observe the 4π spinor periodicity in neutron interferometers

Going further

References. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., Ch. 3 (Theory of Angular Momentum) · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 12 & 15 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. 1, Ch. VI & Complement · Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., Ch. 4 · Townsend, A Modern Approach to Quantum Mechanics, 2nd ed., Ch. 3 · Tung, Group Theory in Physics, Ch. 7–8 · Rauch et al., Phys. Lett. A 54, 425 (1975); Werner et al., Phys. Rev. Lett. 35, 1053 (1975)

Provenance. All equations are standard results from the cited graduate texts. A verifier should: (1) confirm $U=\cos(\theta/2)\mathbb{1}-i\sin(\theta/2)\hat{\mathbf n}\cdot\boldsymbol\sigma$ by exponentiating using $(\hat{\mathbf n}\cdot\boldsymbol\sigma)^2=\mathbb{1}$ (Sakurai §3.3, Shankar Ch. 12); (2) verify $U(2\pi)=-\mathbb{1}$ , $U(4\pi)=+\mathbb{1}$ by direct substitution; (3) reproduce the $\mathfrak{su}(2)$ relations from the group commutator (Sakurai §3.1); (4) check the two-to-one covering via the adjoint action on Pauli matrices (Tung Ch. 8). Neutron-interferometry references: Rauch et al., Phys. Lett. A 54, 425 (1975); Werner et al., Phys. Rev. Lett. 35, 1053 (1975). No fabricated numerical values are used; all numbers ( $\cos\pi=-1$ , half-angle factor, $2\pi/4\pi$ periods) are exact. Edition note: this article's neutron worked example computes the field-path integral from CODATA constants ( $v=h/m_n\lambda$ , $|\gamma_n|=1.832\times10^8\ \mathrm{rad\,s^{-1}T^{-1}}$ ); every number was recomputed in the accompanying verification script.

Upstream: Angular Momentum Commutation Relations (the algebra and its quaternion shadow) · Spin-1/2 and the Pauli Matrices (the exponential identity consumed here) · Ladder Operators and the Angular Momentum Spectrum (the $m$ -spectrum the sign theorem reads) · Unitary Operators and Transformations (the generator doctrine) · Orbital Angular Momentum Operators (the orbital rotation instance). Downstream: Wigner D-Matrices (the matrix elements of $\hat U$ on each multiplet, next) · Spinors and Two-Component Wavefunctions (the carrier space and its ray reading) · Addition of Angular Momenta (tensor products of these representations). One line for later tracks: the Lorentz group repeats this story with $SL(2,\mathbb C)$ over $SO(3,1)$ — relativistic spinors live on that double cover.

Is the −1 under a 2π rotation actually measurable, or is this a convention?

Measurable — but only interferometrically. On an isolated system the $-1$ is a global phase, unobservable; that is why half-integer representations are allowed at all. But split a particle into two coherent branches and rotate the spin in only one: the $-1$ becomes a relative phase and the recombined intensity inverts. Neutron interferometers saw exactly this in 1975, with the fringe period at $4\pi$ of precession angle. The sign is physical; it needs a reference branch to show against.

If U and −U are the same rotation, why doesn't quantum mechanics just pick one and forget SU(2)?

Because no continuous choice exists. Select one matrix per rotation and follow a $2\pi$ loop: your selection arrives back as its negative — the loop upstairs is an open path from $U$ to $-U$ . Any one-representative rule must jump discontinuously somewhere, wrecking the composition law. The clean options: represent $SO(3)$ honestly (integer $j$ only), or accept the double-valued bookkeeping on $SU(2)$ — nature does both, in different sectors.

Where exactly does the half-angle θ/2 come from?

From the size of the spin- $\tfrac12$ generator. The exponent is $\theta\,\hat{\mathbf n}\cdot\hat{\mathbf S}/\hbar$ , and $\hat{\mathbf S}=\tfrac\hbar2\boldsymbol\sigma$ , so the $\hbar$ cancels and $\theta/2$ multiplies $\boldsymbol\sigma$ . The deeper reading: the eigenvalues of $\hat{\mathbf n}\cdot\hat{\mathbf S}$ are $\pm\hbar/2$ , so the two components of the spinor wind as $e^{\mp i\theta/2}$ — half a cycle per physical turn each. Vectors, built bilinearly from spinors (two half-angles = one full angle), recover the full $\theta$ .

Does the 2π sign have anything to do with fermions and the exclusion principle?

Connected, but not by this theorem alone. The spin-statistics theorem — relativistic quantum field theory — says half-integer-spin particles must carry antisymmetric statistics, and its proofs use the $(-1)^{2j}$ rotation behavior as an ingredient. In nonrelativistic quantum mechanics, spin and statistics are independent inputs. The double cover makes fermions possible and supplies the sign; making antisymmetry mandatory takes relativity — the identical-particles and relativistic chapters' story.

Is SU(2) the symmetry group of the physical world or is SO(3)?

Space rotates by $SO(3)$ — laboratory turntables know nothing of $\pm U$ . But quantum states are rays, and symmetries need only be represented up to phase. The theorem (Wigner, Bargmann) is that projective representations of $SO(3)$ are ordinary representations of its universal cover, $SU(2)$ . So the honest answer: the physical symmetry is $SO(3)$ ; quantum mechanics implements it through $SU(2)$ because rays tolerate the sign. Half-integer spin is the physical sector that exercises this tolerance.

How do I go from an SU(2) matrix back to the axis and angle?

Read the $S^3$ coordinates. Write $U=a_0\mathbb 1-i\,\mathbf a\cdot\boldsymbol\sigma$ ; then $a_0=\tfrac12\mathrm{Tr}\,U=\cos(\theta/2)$ fixes the angle, and $\hat{\mathbf n}=\mathbf a/|\mathbf a|$ (with $\mathbf a$ extracted from the traceless part, $a_k=\tfrac{i}{2}\mathrm{Tr}(U\sigma_k)$ ) fixes the axis. Worked Example 2 runs this in reverse gear: multiply two known rotations as $2\times2$ matrices, read $\cos(\theta/2)=\tfrac12$ from the trace, and the composite axis falls out of three more traces. This is the quaternion "attitude extraction" used in navigation software.