The Rotation Group SO(3)

The rotation group $SO(3)$ is what "rotational symmetry in three dimensions" means once you write it down as mathematics. A proper rotation is any linear map of $\mathbb{R}^3$ that keeps lengths, angles, and handedness intact, and the entire content of that sentence is captured by two matrix conditions: $R^{\mathsf{T}}R=\mathbb{1}$ (lengths and angles preserved) and $\det R=+1$ (orientation preserved, reflections excluded). Everything physical about rotations — that they compose, refuse to commute, and place a rigid body's orientation at a single point of a space — is already latent in those two lines. Three structural facts organize the whole article. First, $SO(3)$ is a compact, connected, three-dimensional Lie group: near the identity it is a copy of its Lie algebra $\mathfrak{so}(3)$ , the antisymmetric matrices, whose generators close on the angular-momentum algebra $[J_a,J_b]=i\epsilon_{abc}J_c$ — the same commutators that run all of quantum angular momentum. Second, its global topology is nontrivial: the group manifold is a solid ball of radius $\pi$ with antipodal surface points glued, so $\pi_1(SO(3))=\mathbb{Z}_2$ ; a $2\pi$ rotation is a loop you cannot shrink to a point, while a $4\pi$ rotation is. Third, that topology forces a double cover by the simply connected group $SU(2)$ , and the split between contractible and non-contractible loops is exactly the split between integer spin (true representations of $SO(3)$ ) and half-integer spin (spinors, living only on $SU(2)$ ).

Key takeaways

$SO(3)$ is the group of real $3\times3$ matrices with $R^{\mathsf{T}}R=\mathbb{1}$ and $\det R=+1$ : the proper rotations of $\mathbb{R}^3$ about axes through the origin.
It is a compact, connected, three-dimensional Lie group; every element is an axis–angle pair $(\hat n,\theta)$ with $\theta\in[0,\pi]$ , filling a solid ball of radius $\pi$ .
Its Lie algebra $\mathfrak{so}(3)$ is the antisymmetric matrices; the generators obey $[J_a,J_b]=i\epsilon_{abc}J_c$ , the angular-momentum commutators, and exponentiate to finite rotations via Rodrigues' formula.
The topology is nontrivial: $\pi_1(SO(3))=\mathbb{Z}_2$ , so $SO(3)\cong SU(2)/\mathbb{Z}_2$ is doubly covered by $SU(2)$ , and a spinor picks up a factor $-1$ under a $2\pi$ turn, $+1$ under $4\pi$ .
Irreducible representations are labelled by $\ell=0,1,2,\dots$ with dimension $2\ell+1$ ; characters are $\chi^{(\ell)}(\theta)=\sin[(\ell+\tfrac12)\theta]/\sin(\theta/2)$ and tensor products decompose by the Clebsch–Gordan series.
Key idea: $SO(3)$ is a compact connected three-dimensional Lie group whose generators satisfy $[J_a,J_b]=i\epsilon_{abc}J_c$ , and whose topology $\pi_1=\mathbb{Z}_2$ forces a double cover by $SU(2)$ , splitting representations into integer (true $SO(3)$ ) and half-integer (spinorial) sectors.

The core insight

$SO(3)$ is a compact connected three-dimensional Lie group whose generators satisfy the angular momentum algebra $[J_a,J_b]=i\epsilon_{abc}J_c$ ; its nontrivial topology $\pi_1=\mathbb{Z}_2$ forces a double cover by $SU(2)$ , splitting representations into integer (true $SO(3)$ ) and half-integer (spinorial) sectors.

Where this lives today

The $\mathbb{Z}_2$ topology of $SO(3)$ is not a curiosity — it has been measured. In 1975 Rauch and collaborators in Vienna, and independently Werner and colleagues, sent neutrons through a perfect-crystal interferometer and rotated one path's spin with a magnetic field. The interference pattern reappeared only after the spin had turned through $4\pi$ , not $2\pi$ : direct confirmation that a spin- $\tfrac12$ state changes sign under a single $2\pi$ rotation, as the double cover demands. The same $SU(2)$ / $SO(3)$ structure is now everyday engineering. A qubit's state lives on the Bloch sphere and its gates are elements of $SU(2)$ ; a control pulse is literally a rotation, and the sign difference between a $2\pi$ and a $4\pi$ pulse must be tracked to keep phases coherent. Away from the quantum world, spacecraft and robot-arm attitude control represent orientations by unit quaternions (points of $SU(2)\cong S^3$ ) to dodge the coordinate singularities — gimbal lock — that afflict Euler-angle charts on the $SO(3)$ manifold, and Rodrigues' formula is the exponential map every graphics and robotics library uses to interpolate rotations.

The physical picture

Give every rotation an address. Point an arrow along the axis $\hat n$ and give it length equal to the angle $\theta$ . Because turning by more than $\pi$ about $\hat n$ equals turning by less than $\pi$ about $-\hat n$ , no arrow need be longer than $\pi$ , and all of $SO(3)$ fits inside a solid ball of radius $\pi$ : the identity at the centre, a half-turn on the surface. There is one subtlety, and it is the whole story: a half-turn by $+\pi$ about $\hat n$ and by $\pi$ about $-\hat n$ are the same rotation, so the antipodal surface points $\pi\hat n$ and $-\pi\hat n$ must be glued together. That single gluing twists the topology.

Figure

The parameter space of SO(3): every rotation is the point θ n̂ inside a solid ball of radius π, so the rotation angle satisfies 0 ≤ θ ≤ π. Diametrically opposite surface points π n̂ and −π n̂ are the same rotation and are glued together — the identification that makes π₁ of SO(3) equal to ℤ₂

Now look up close. An infinitesimal rotation is the identity plus a tiny antisymmetric matrix, $R\approx\mathbb{1}+\delta\theta\,(\hat n\cdot\mathbf J)$ ; antisymmetry is orthogonality read to first order. Do two small rotations in one order, then in reverse, and subtract: the leftover is itself a small rotation, and that closure is the commutator $[J_a,J_b]=i\epsilon_{abc}J_c$ . The topology has a tactile demonstration — the belt trick. Rotate a buckle by $2\pi$ and the belt is twisted; you cannot untwist it without rotating the buckle back. Rotate by a further $2\pi$ (total $4\pi$ ) and the twist slides out with the buckle held fixed. The $2\pi$ loop is the nontrivial element of $\mathbb{Z}_2$ , the $4\pi$ loop is trivial, and only objects on the double cover $SU(2)$ — spinors — register the difference between $0$ and $2\pi$ , as the second figure makes quantitative.

Figure

Spinor versus vector under rotation about a fixed axis: the spin-½ overlap cos(θ/2) flips to −1 after a full 2π turn and returns to +1 only after 4π, while a vector observable cos θ is already periodic in 2π. The 4π period of the blue curve is the visible signature of the double cover SU(2) onto SO(3)

The core equations

The definition is two conditions on a matrix. Orthogonality preserves the Euclidean inner product; the unit determinant discards reflections and improper rotations. Together they carve $SO(3)$ out of all $3\times3$ real matrices.

SO(3)=\{\,R\in M_3(\mathbb{R})\;:\;R^{\mathsf{T}}R=\mathbb{1},\ \det R=+1\,\} \qquad (1)

where: $R$ — a real $3\times3$ rotation matrix; $R^{\mathsf{T}}$ — its transpose; $\mathbb{1}$ — the $3\times3$ identity; $\det R$ — the determinant, fixed to $+1$ to exclude reflections. Reading it: orthogonality $R^{\mathsf{T}}R=\mathbb{1}$ means $\langle Rx,Ry\rangle=\langle x,y\rangle$ for all vectors, so lengths and angles survive. Differentiating the orthogonality constraint at the identity gives the Lie algebra: the tangent space to $SO(3)$ at $\mathbb{1}$ is the set of antisymmetric matrices. Its dimension is three, matching the three independent entries above the diagonal and the three-dimensionality of the group.

\mathfrak{so}(3)=\{\,A\in M_3(\mathbb{R})\;:\;A^{\mathsf{T}}=-A\,\},\qquad \dim\mathfrak{so}(3)=3 \qquad (2)

where: $A$ — a generator, a real antisymmetric $3\times3$ matrix. Reading it: near the identity a rotation is $\mathbb{1}+A$ , and antisymmetry is orthogonality linearised — $(\mathbb{1}+A)^{\mathsf{T}}(\mathbb{1}+A)=\mathbb{1}$ to first order forces $A^{\mathsf{T}}+A=0$ . Choosing Hermitian generators $J_a$ (the physicist's convention, carrying an explicit $i$ ) turns the bracket of the algebra into the angular-momentum commutation relations. The structure constants are the Levi-Civita symbol; this single equation is the reason $\mathfrak{so}(3)$ and quantum angular momentum are the same subject.

[J_a,J_b]=i\,\epsilon_{abc}\,J_c \qquad (3)

where: $J_a$ — Hermitian generator of rotations about the $a$ -axis; $\epsilon_{abc}$ — Levi-Civita symbol, the structure constants of $\mathfrak{so}(3)$ ; $i$ — imaginary unit, present because the $J_a$ are Hermitian. Reading it: the generators do not commute, so $SO(3)$ is non-abelian. Exponentiating a generator recovers a finite rotation. The axis–angle form writes any rotation as the exponential of $-i\theta\,\hat n\cdot\mathbf J$ , the group-theoretic statement that finite rotations are generated by infinitesimal ones.

R(\hat n,\theta)=\exp\!\big(-i\,\theta\,\hat n\cdot\mathbf{J}\big) \qquad (4)

where: $\hat n$ — unit vector along the axis; $\theta\in[0,\pi]$ — rotation angle; $\mathbf J=(J_1,J_2,J_3)$ — the vector of generators. Reading it: the exponential map is surjective because $SO(3)$ is compact and connected — every rotation really is some exponential. In the defining (vector) representation the exponential closes into an elementary form. Rodrigues' formula gives the explicit $3\times3$ rotation matrix from the axis and angle, without any infinite series.

R(\hat n,\theta)=\mathbb{1}+\sin\theta\,K+(1-\cos\theta)\,K^2,\qquad K_{ij}=-\epsilon_{ijk}n_k \qquad (5)

where: $K$ — the antisymmetric "cross-product matrix" with $Kx=\hat n\times x$ ; $n_k$ — components of the axis $\hat n$ . Reading it: because $K^3=-K$ the exponential series collapses onto just $\mathbb{1}$ , $K$ , and $K^2$ , and the surviving coefficients are exactly $\sin\theta$ and $1-\cos\theta$ . The irreducible representations are labelled by $\ell$ , and their characters — traces, which depend only on the rotation angle — are given by a geometric sum in closed form. The character is the workhorse of representation theory: it counts multiplicities and decomposes tensor products.

\chi^{(\ell)}(\theta)=\sum_{m=-\ell}^{\ell}e^{im\theta}=\frac{\sin\!\big[(\ell+\tfrac12)\theta\big]}{\sin(\theta/2)},\qquad \dim=2\ell+1 \qquad (6)

where: $\chi^{(\ell)}$ — trace of the spin- $\ell$ representation matrix; $\ell$ — non-negative integer (or half-integer on $SU(2)$ ); $\theta$ — the class parameter, the rotation angle. Reading it: at $\theta\to0$ the character returns the dimension, $\chi^{(\ell)}(0)=2\ell+1$ , the trace of the identity. Finally, the topology and its consequence. The fundamental group is $\mathbb{Z}_2$ , and the simply connected double cover is $SU(2)$ ; the two facts are the same fact told twice.

\pi_1\big(SO(3)\big)=\mathbb{Z}_2,\qquad SO(3)\cong SU(2)/\{\pm\mathbb{1}\} \qquad (7)

where: $\pi_1$ — the first homotopy group (closed loops up to deformation); $SU(2)\cong S^3$ — the simply connected universal cover. Reading it: because a $2\pi$ loop is nontrivial, half-integer spin is not single-valued on $SO(3)$ and lives only on $SU(2)$ ; representations then combine by the Clebsch–Gordan series $D^{(j_1)}\otimes D^{(j_2)}=\bigoplus_{J=|j_1-j_2|}^{j_1+j_2}D^{(J)}$ .

Derivation ABuilding the generators and the angular momentum algebra

The goal is to turn the abstract statement "the tangent space at the identity is the antisymmetric matrices" into a concrete, computable basis, and then prove that this basis closes on the commutators $[J_a,J_b]=i\epsilon_{abc}J_c$ — the angular-momentum algebra — using nothing but the Levi-Civita symbol and the epsilon–delta identity. We write the coordinate generators, convert to the Hermitian convention, and grind out one representative bracket in index notation.

Step 1. Define the three real antisymmetric generators by $(\tilde J_a)_{bc}=-\epsilon_{abc}$ , so that $\tilde J_3$ has nonzero entries only in the $1$ – $2$ block and generates rotations in that plane, and $\tilde J_1,\tilde J_2$ do the same cyclically. Why: these are precisely the derivatives $\mathrm dR/\mathrm d\theta$ at $\theta=0$ for rotations about each coordinate axis, so they span the tangent space $\mathfrak{so}(3)$ .

Step 2. Pass to the Hermitian (physicist's) convention $J_a=i\tilde J_a$ , giving matrix elements $(J_a)_{bc}=-i\,\epsilon_{abc}$ . Why: the extra $i$ makes the $J_a$ Hermitian, so their eigenvalues are real and they can serve directly as quantum-mechanical angular-momentum operators.

Step 3. Write the commutator in components, $\big([J_a,J_b]\big)_{cd}=(J_a)_{ce}(J_b)_{ed}-(J_b)_{ce}(J_a)_{ed}$ , summing over the repeated index $e$ . Why: a commutator of matrices is computed entry by entry, and index notation keeps the bookkeeping honest.

Step 4. Insert the generators. Each factor of $(J_a)_{ce}=-i\epsilon_{ace}$ brings an $i$ , so the product carries $(-i)^2=-1$ , giving $\big([J_a,J_b]\big)_{cd}=-\big(\epsilon_{ace}\epsilon_{bed}-\epsilon_{bce}\epsilon_{aed}\big)$ . Why: this reduces the whole computation to contracting products of two Levi-Civita symbols.

Step 5. Apply the identity $\epsilon_{ace}\epsilon_{bed}=\epsilon_{eac}\epsilon_{ebd}=\delta_{ab}\delta_{cd}-\delta_{ad}\delta_{cb}$ to each term. The pieces symmetric in $(a,b)$ cancel between the two terms, and what remains is $\delta_{ad}\delta_{cb}-\delta_{bd}\delta_{ca}$ repackaged as $i\,\epsilon_{abe}(J_e)_{cd}$ . Why: the epsilon–delta identity is what converts a quadratic-in- $\epsilon$ expression back into a single generator, exposing the structure constants.

Step 6. Read off the result: $\big([J_a,J_b]\big)_{cd}=i\,\epsilon_{abe}(J_e)_{cd}$ , i.e. $[J_a,J_b]=i\,\epsilon_{abc}J_c$ . Specialising to $a=1,b=2$ gives the familiar $[J_x,J_y]=iJ_z$ . Why: the surviving antisymmetric combination is exactly $i\epsilon_{abe}J_e$ , so the structure constants of $\mathfrak{so}(3)$ are the Levi-Civita symbol — the defining relation of angular momentum.

Go deeper · Derivation B · 1 minRodrigues' formula from the exponential of a generator

Derivation BRodrigues' formula from the exponential of a generator

The goal is to prove the closed-form Rodrigues rotation, $R(\hat n,\theta)=\mathbb{1}+\sin\theta\,K+(1-\cos\theta)K^2$ , directly from the exponential $R=e^{\theta K}$ , where $K$ implements the cross product $Kx=\hat n\times x$ . The engine is a single algebraic identity, $K^3=-K$ , which holds for any unit axis and truncates the otherwise infinite exponential to three terms; grouping the surviving powers reconstructs the series of $\sin$ and $\cos$ .

Step 1. Establish $K^3=-K$ for a unit axis. Since $K$ acts as $Kx=\hat n\times x$ , one finds $K^2 x=\hat n\times(\hat n\times x)=(\hat n\cdot x)\hat n-x$ , and applying $K$ once more gives $K^3x=-\hat n\times x=-Kx$ . Why: this identity means every power $K^n$ with $n\ge 3$ reduces to a multiple of $K$ or $K^2$ , so the exponential can involve only $\mathbb{1}$ , $K$ , and $K^2$ .

Step 2. Expand the exponential and sort by powers. Using $K^3=-K$ , the odd powers all point along $K$ and the even powers ( $\ge 2$ ) along $K^2$ : $e^{\theta K}=\mathbb{1}+\big(\theta-\tfrac{\theta^3}{3!}+\tfrac{\theta^5}{5!}-\cdots\big)K+\big(\tfrac{\theta^2}{2!}-\tfrac{\theta^4}{4!}+\cdots\big)K^2$ . Why: the recursion $K^{2m+1}=(-1)^mK$ and $K^{2m}=(-1)^{m-1}K^2$ separates the series into two alternating scalar series times fixed matrices.

Step 3. Recognise the two scalar series as $\sin\theta$ and $1-\cos\theta$ , giving $R=\mathbb{1}+\sin\theta\,K+(1-\cos\theta)K^2$ . Why: the coefficient of $K$ is exactly the Taylor series of $\sin\theta$ , and the coefficient of $K^2$ is $1-\cos\theta$ , so the finite rotation is trigonometric, not transcendental in the axis and angle.

Step 4. Sanity-check the endpoints. At $\theta=0$ , $R=\mathbb{1}$ ; at $\theta=\pi$ , $R=\mathbb{1}+2K^2$ , a half-turn that flips the plane perpendicular to $\hat n$ and fixes $\hat n$ itself (since $K\hat n=0$ ). Why: the axis is the eigenvector of $R$ with eigenvalue $+1$ , confirming $\hat n$ is genuinely the rotation axis for every $\theta$ . *As a concrete check: $\hat n=\tfrac{1}{\sqrt3}(1,1,1)$ with $\theta=2\pi/3$ yields the cyclic permutation matrix with rows $(0,0,1),(1,0,0),(0,1,0)$ , satisfying $\det R=+1$ and $R^{\mathsf T}R=\mathbb1$ .*

Go deeper · Derivation C · 1 minThe two-to-one cover of SU(2) onto SO(3)

Derivation CThe two-to-one cover of SU(2) onto SO(3)

The goal is to construct the covering homomorphism explicitly and prove it is two-to-one with kernel $\{\pm\mathbb{1}\}$ , from which $SO(3)\cong SU(2)/\mathbb{Z}_2$ and $\pi_1(SO(3))=\mathbb{Z}_2$ follow immediately. The trick is to identify a vector in $\mathbb{R}^3$ with a traceless Hermitian $2\times2$ matrix via the Pauli matrices, so that conjugation by an $SU(2)$ element acts on the vector as a rotation while preserving its length through the determinant.

Step 1. Map $\mathbf x\in\mathbb{R}^3$ to $X=\mathbf x\cdot\boldsymbol\sigma$ , the most general traceless Hermitian $2\times2$ matrix, with $\det X=-|\mathbf x|^2$ . Why: this dictionary turns geometric data (a vector and its length) into linear-algebraic data (a matrix and its determinant) that $SU(2)$ can act on cleanly.

Step 2. Conjugate: $X'=UXU^{\dagger}$ for $U\in SU(2)$ . Then $X'$ is again traceless and Hermitian, so it is $\mathbf x'\cdot\boldsymbol\sigma$ for some $\mathbf x'=R(U)\mathbf x$ . Why: conjugation by a unitary preserves both the tracelessness and the Hermiticity, so it maps vectors to vectors — it defines a linear operator $R(U)$ on $\mathbb{R}^3$ .

Step 3. Show $R(U)$ preserves length: $\det X'=\det(UXU^{\dagger})=\det X$ because $\det U=1$ , so $|\mathbf x'|=|\mathbf x|$ and $R(U)\in O(3)$ . Continuity of $U\mapsto R(U)$ from the connected group $SU(2)$ then forces $\det R(U)=+1$ , i.e. $R(U)\in SO(3)$ , and $R(U_1U_2)=R(U_1)R(U_2)$ . Why: length preservation is orthogonality, and connectedness rules out the reflection component, so the map lands in the identity component $SO(3)$ and respects group multiplication.

Step 4. Compute the kernel. $UXU^{\dagger}=X$ for all traceless Hermitian $X$ means $U$ commutes with every such matrix, hence with all of $M_2(\mathbb{C})$ , so $U=\pm\mathbb{1}$ . Why: only the centre of $SU(2)$ acts trivially, so exactly two elements $\pm U$ map to each rotation — the map is two-to-one and surjective.

Step 5. Conclude the topology: $SO(3)\cong SU(2)/\{\pm\mathbb{1}\}$ , and since $SU(2)\cong S^3$ is simply connected, quotienting by the two-element group makes $\pi_1(SO(3))=\mathbb{Z}_2$ . Why: a simply connected cover with a two-element deck group hands its $\mathbb{Z}_2$ straight to the fundamental group of the base, which is the belt trick made rigorous. *As a concrete check: $U=e^{-i\theta\sigma_3/2}=\mathrm{diag}(e^{-i\theta/2},e^{i\theta/2})$ conjugates $\mathbf x\cdot\boldsymbol\sigma$ into a rotation by $\theta$ about the $z$ -axis, and $U(\theta{=}2\pi)=-\mathbb1$ maps to $R=\mathbb1$ , confirming the two-to-one nature of the cover.*

Limiting cases

Specialisations and relatives of $SO(3)$ expose its structure from every side:

Case	Result	Meaning
Fix the axis	$SO(2)\cong U(1)\subset SO(3)$	Rotations about one axis form the abelian one-parameter subgroup, a circle generated by a single $J_z$ , with one-dimensional irreps labelled by $m$ .
Include reflections	$O(3)=SO(3)\rtimes\mathbb{Z}_2$	Adjoining parity $P=-\mathbb{1}$ ( $\det P=-1$ ) builds the full orthogonal group; $SO(3)$ is its identity component.
Universal cover	$SU(2)/\{\pm\mathbb{1}\}\cong SO(3)$	The simply connected $SU(2)$ covers $SO(3)$ two-to-one; integer- $\ell$ irreps descend, half-integer ones remain only on $SU(2)$ .
Turn off boosts	$SO(3)\subset SO(3,1)^{\uparrow}$	$SO(3)$ is the rotation subgroup of the proper orthochronous Lorentz group; zeroing the three boost rapidities leaves the three rotations.
High spin, small angle	$\chi^{(\ell)}(\theta)\to 2\ell+1$	As $\theta\to0$ every character returns the dimension $2\ell+1$ , recovering the degeneracy of a rotational multiplet.

Common trapA 2π rotation is the identity, so spin-½ ought to be an ordinary representation of SO(3)

The first clause is true and the conclusion is false, and the gap between them is the whole point of the group's topology. In $SO(3)$ the matrix produced by a $2\pi$ rotation is indeed $\mathbb{1}$ — but the path traced through the group as the angle sweeps from $0$ to $2\pi$ is a closed loop that cannot be contracted to a point. A spin- $\tfrac12$ "representation" would have to assign the number $-1$ to that $2\pi$ loop, which is single-valued on the double cover $SU(2)$ (where the $2\pi$ and $4\pi$ elements are genuinely different) but double-valued on $SO(3)$ itself. So spinors are not representations of $SO(3)$ at all; they are representations of $SU(2)$ . A second, cheaper trap: rotations do not commute, so $e^Ae^B\neq e^{A+B}$ in general and the order of finite rotations matters — never treat $SO(3)$ as if its exponentials add like numbers.

Worked examples

1 · A 120° rotation permutes the coordinate axes (fully worked)

Use Rodrigues' formula to compute the rotation by $\theta=120^\circ=2\pi/3$ about the body diagonal $\hat n=\tfrac{1}{\sqrt3}(1,1,1)$ , and interpret the result geometrically.

Step 1. Build the cross-product matrix $K$ with $Kx=\hat n\times x$ . With $n_1=n_2=n_3=1/\sqrt3$ ,

K=\frac{1}{\sqrt3}\begin{pmatrix}0&-1&1\\1&0&-1\\-1&1&0\end{pmatrix}\equiv\frac{1}{\sqrt3}M.

Squaring the inner matrix, $M^2=\begin{pmatrix}-2&1&1\\1&-2&1\\1&1&-2\end{pmatrix}$ , so $K^2=\tfrac13 M^2$ .

Step 2. Insert the trigonometric coefficients for $\theta=2\pi/3$ : $\sin\theta=\tfrac{\sqrt3}{2}$ and $1-\cos\theta=1-(-\tfrac12)=\tfrac32$ . Then $\sin\theta\,K=\tfrac{\sqrt3}{2}\cdot\tfrac{1}{\sqrt3}M=\tfrac12 M$ and $(1-\cos\theta)K^2=\tfrac32\cdot\tfrac13 M^2=\tfrac12 M^2$ .

Step 3. Assemble $R=\mathbb{1}+\tfrac12 M+\tfrac12 M^2$ . Adding entry by entry, every diagonal term is $1+0-1=0$ and the off-diagonal terms combine to a single $+1$ in each row:

R=\begin{pmatrix}0&0&1\\1&0&0\\0&1&0\end{pmatrix}.

Step 4. Interpret. Acting on the basis vectors, $R\hat x=\hat y$ , $R\hat y=\hat z$ , $R\hat z=\hat x$ : the rotation cyclically permutes the coordinate axes. This is exactly what a $120^\circ$ turn about the cube's body diagonal must do, since it maps the three edges meeting at a corner into one another. Checks: $\det R=+1$ and $R^{\mathsf{T}}R=\mathbb{1}$ , so $R\in SO(3)$ , and $R^3=\mathbb{1}$ as three $120^\circ$ turns must give. The rotation matrix is the cyclic permutation $\hat x\to\hat y\to\hat z\to\hat x$ , with $\det R=+1$ and $R^3=\mathbb{1}$ .

2 · Decomposing 3 ⊗ 3 with characters

Two spin-1 systems are combined. Use the character (6) to show $D^{(1)}\otimes D^{(1)}=D^{(0)}\oplus D^{(1)}\oplus D^{(2)}$ , and verify the dimension count.

Step 1. Write the spin-1 character. From (6), $\chi^{(1)}(\theta)=e^{-i\theta}+1+e^{+i\theta}=1+2\cos\theta$ . The character of a tensor product is the product of characters, so the combined system has character $\chi^{(1)}(\theta)^2=(1+2\cos\theta)^2$ .

Step 2. Expand the square using $\cos^2\theta=\tfrac12(1+\cos2\theta)$ : $(1+2\cos\theta)^2=1+4\cos\theta+4\cos^2\theta=3+4\cos\theta+2\cos2\theta$ .

Step 3. Match against a sum of irreducible characters. With $\chi^{(0)}=1$ , $\chi^{(1)}=1+2\cos\theta$ , and $\chi^{(2)}=1+2\cos\theta+2\cos2\theta$ , their sum is $\chi^{(0)}+\chi^{(1)}+\chi^{(2)}=3+4\cos\theta+2\cos2\theta$ — identical to Step 2. Because the irreducible characters are orthogonal, this match is a proof, not a coincidence: $D^{(1)}\otimes D^{(1)}=D^{(0)}\oplus D^{(1)}\oplus D^{(2)}$ .

Step 4. Count dimensions at $\theta=0$ : the left side is $3\times3=9$ , the right side $1+3+5=9$ . Physically this is the decomposition of a Cartesian rank-2 tensor $T_{ij}$ into its scalar trace ( $\ell=0$ ), antisymmetric vector part ( $\ell=1$ ), and symmetric-traceless quadrupole ( $\ell=2$ ). Both sides give dimension 9: $\mathbf{3}\otimes\mathbf{3}=\mathbf{1}\oplus\mathbf{3}\oplus\mathbf{5}$ .

Physics of the portrait

No one did more to turn the representation theory of rotations into a physical instrument than Hermann Weyl.

Origins · 1 minHermann Weyl and how this was discovered

Hermann Weyl

1885–1955 · representation theory of the rotation group in quantum mechanics

Hermann Weyl was the mathematician who taught physics how to use symmetry. Trained under David Hilbert at Göttingen, he ranged across relativity, gauge theory, and the foundations of mathematics, but his most enduring gift here was the representation theory of compact Lie groups — of which $SO(3)$ is the archetype. In the 1920s Weyl proved the completeness and orthogonality of characters for compact groups (the Peter–Weyl theorem, 1927) and derived the character formula whose $SO(3)$ specialisation, $\chi^{(\ell)}(\theta)=\sin[(\ell+\tfrac12)\theta]/\sin(\theta/2)$ , is equation (6) here. His 1928 book Gruppentheorie und Quantenmechanik argued — against physicists who dismissed the "Gruppenpest" — that the multiplet structure of atomic spectra, the $2\ell+1$ degeneracies, and the selection rules are all consequences of rotational symmetry. Working in the same years, Eugene Wigner built the explicit angular-momentum machinery (the Wigner $D$ -matrices and $3j$ symbols) on the same foundation, while Élie Cartan's earlier discovery of spinors (1913) supplied the half-integer representations that live on the double cover. Weyl's synthesis is why a physicist can say "this state is spin-2" and mean a precise, computable statement about how it transforms under $SO(3)$ .

History

1775

— Euler proves every rigid rotation fixes an axis (Euler's rotation theorem)

1840

— Rodrigues gives the axis–angle formula and the composition law of rotations

1843

— Hamilton's quaternions encode 3D rotations, foreshadowing SU(2)

1913

— Cartan discovers spinors, the representations that need the double cover

1927

— Weyl and Wigner build quantum angular momentum on SO(3) and SU(2) representation theory

Going further

Texts. Stone & Goldbart, Mathematics for Physics (2009), Ch. 16–17 · Arfken & Weber, Mathematical Methods for Physicists, 7th ed., Ch. 17 · Hassani, Mathematical Physics, 2nd ed., Ch. 27–29 · Nakahara, Geometry, Topology and Physics, 2nd ed., §5.6 · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 10 · Sakurai & Napolitano, Modern Quantum Mechanics, Ch. 3 · Tung, Group Theory in Physics, Ch. 7–8. The key self-consistency checks for this article are: (1) the commutator $[J_a,J_b]=i\epsilon_{abc}J_c$ follows from $(J_a)_{bc}=-i\epsilon_{abc}$ and the epsilon–delta identity; (2) Rodrigues' formula follows from $K^3=-K$ , and the worked $120^\circ$ example gives the cyclic permutation matrix with $\det=+1$ and $R^3=\mathbb1$ ; (3) the character $\chi^{(\ell)}(\theta)=\sin[(\ell+\tfrac12)\theta]/\sin(\theta/2)$ has limit $2\ell+1$ as $\theta\to0$ , and the identity $(1+2\cos\theta)^2=\chi^{(0)}+\chi^{(1)}+\chi^{(2)}$ underlies $\mathbf3\otimes\mathbf3=\mathbf1\oplus\mathbf3\oplus\mathbf5$ ; (4) the kernel of the $SU(2)\to SO(3)$ map is $\{\pm\mathbb{1}\}$ .

Prerequisites. Groups, Subgroups, and Cosets · Lie Groups · Lie Algebras and the Exponential Map · Conjugacy Classes and Normal Subgroups. Leads to. SU(2) and the Spin Double Cover · Group Representations · Tensor Products and Clebsch–Gordan Decomposition · Characters and Orthogonality Theorems.

Ask the tutor: Ask me to walk from the two defining conditions to why an electron needs two full turns to come back — or to work any rotation through Rodrigues' formula step by step.

Why is SO(3) three-dimensional as a manifold?

Because it takes exactly three numbers to specify a rotation. In the algebra picture, $\mathfrak{so}(3)$ is the antisymmetric $3\times3$ matrices, and such a matrix has three independent entries above the diagonal, so the tangent space — and hence the group — is three-dimensional. In the geometric picture, a rotation needs an axis (two angles to point a unit vector) plus one rotation angle, again three. Both agree with the axis–angle ball: a solid three-dimensional region of radius $\pi$ .

What is the difference between SO(3) and SU(2), if they have the same Lie algebra?

They share a Lie algebra — the same commutators $[J_a,J_b]=i\epsilon_{abc}J_c$ — so locally, near the identity, they are indistinguishable. They differ globally. $SU(2)$ is the three-sphere $S^3$ , simply connected: every loop shrinks to a point. $SO(3)$ is $SU(2)$ with antipodal points identified, which introduces a non-contractible loop and makes $\pi_1=\mathbb{Z}_2$ . The map $SU(2)\to SO(3)$ is two-to-one, sending both $U$ and $-U$ to the same rotation. The extra room in $SU(2)$ is what half-integer spin needs: it can tell $2\pi$ apart from $0$ , whereas $SO(3)$ cannot.

Why do opposite points on the surface of the parameter ball get glued?

A half-turn ( $\theta=\pi$ ) about $\hat n$ flips every vector in the plane perpendicular to $\hat n$ and leaves $\hat n$ fixed. The same half-turn about $-\hat n$ flips the same plane in the same way and fixes the same axis line — it is literally the same matrix. But its two addresses in the ball are the antipodal surface points $\pi\hat n$ and $-\pi\hat n$ , so those points must be identified. That antipodal gluing on the boundary sphere is the geometric origin of $\pi_1(SO(3))=\mathbb{Z}_2$ .

How does the character formula encode the dimension of a representation?

The character $\chi^{(\ell)}(\theta)$ is the trace of the rotation matrix in the spin- $\ell$ representation, and depends only on $\theta$ because trace is invariant under the basis change that reorients the axis. At $\theta=0$ the rotation is the identity, whose trace is the number of basis states — the dimension. The sum $\sum_{m=-\ell}^{\ell}e^{im\theta}$ at $\theta=0$ gives $2\ell+1$ ones, and the closed form tends to the same $2\ell+1$ by l'Hôpital. So "put $\theta=0$ " is the standard trick for reading a dimension off a character.

Why does the belt or plate trick have anything to do with group theory?

The trick is a physical loop in $SO(3)$ . Rotating a buckle continuously by $2\pi$ traces a closed path in the rotation group, and the twist in the belt records whether that path can be undone with the endpoints held fixed — that is, whether the loop is contractible. A $2\pi$ loop leaves a twist you cannot remove: the nontrivial element of $\pi_1(SO(3))=\mathbb{Z}_2$ . A $4\pi$ loop untwists, because $\mathbb{Z}_2$ addition gives $1+1=0$ . So the belt is a homotopy detector, its "two twists cancel" behaviour the group $\mathbb{Z}_2$ acted out by hand.

Where do half-integer spins actually come from if SO(3) forbids them?

They come from the double cover $SU(2)$ , which physics may use because the symmetry of quantum mechanics is realised only up to a phase. States are rays in Hilbert space, so a representation need only be single-valued up to an overall factor — a projective representation of $SO(3)$ . Wigner's theorem shows these are exactly the genuine representations of $SU(2)$ , which include the half-integer spins. Nature uses all of them: electrons, protons, and neutrons are spin- $\tfrac12$ , and their $2\pi$ sign flip is seen directly in neutron interferometry. Integer spins descend to honest $SO(3)$ representations; half-integer spins live one level up.