Simple and Non-Simple Rotations: One Angle at $n\le3$, Two Angles at $n\ge4$
Introduction
A rotation is simple when it turns a single plane and fixes the rest, and the count of the independent angles is the rank $\lfloor n/2\rfloor$: one angle at $n=2$ and at $n=3$, where every rotation is simple, and two or more angles at $n\ge4$, where the generic rotation is not simple.
The article defines simplicity, proves that every rotation of two- and three-space is simple, proves Euler's axis theorem and the trace formula for the angle, shows that the generic rotation of four-space has two angles and no fixed vector, and computes the number of angles as the rank of the orthogonal group, which is the dimension of a maximal torus. It assumes The Exponential of the Antisymmetric Part: One-Parameter Rotations and The Spectral Theorem: the Anti-Hermitian Case and the Reality of the Spectrum. It defers the geometric reading to The Geometry of the Two Halves, and the Return of the Two-Sided Operators and the covering to The Rotor as a Square Root of a Rotation: the Two Signs and the Double Cover.
Simplicity
A rotation is simple when it fixes a subspace of codimension two. Say that $Q\in\mathrm{SO}(n)$ is simple when there is a plane $P$ with
$$ Q|_P\ \text{a rotation by some angle},\qquad Q|_{P^{\perp}}=\mathrm{id}. $$
The plane is then invariant, the orthogonal complement is fixed pointwise, and the rotation is the planar rotation of the previous article extended by the identity.
A simple rotation has the eigenvalue one with multiplicity at least $n-2$. The fixed subspace contributes the eigenvalue one on $n-2$ dimensions, so the eigenvalue one has multiplicity at least $n-2$ and the remaining two eigenvalues are $e^{\pm i\theta}$. The computation confirms the count at $n=4$ on a rotation with one angle: the multiplicity is two and the rank of $Q-\mathrm{I}$ is two.
The rank of $Q-\mathrm{I}$ measures the rotating part. For a simple rotation with angle $\theta\notin2\pi\mathbb{Z}$,
$$ \operatorname{rank}(Q-\mathrm{I})=2, $$
since $Q-\mathrm{I}$ vanishes on the fixed subspace and is invertible on the rotating plane. The computation confirms the rank two at $n=3$ for random rotations (which are all simple) and at $n=4$ for a rotation with one angle, and it is the numerical test of simplicity.
Non-simple means that more than one plane turns. A rotation is non-simple exactly when its canonical form has $m\ge2$ nonzero angles, so
$$ Q\ \text{simple}\iff m\le1\iff\dim\ker(Q-\mathrm{I})\ge n-2, $$
which was verified at $n=4,5$ on rotations with one and with two angles. So simplicity is a codimension condition: the simple rotations form the union of the conjugates of the rotations of a single plane.
The One Angle at $n\le3$
At $n=2$ every rotation is simple and the angle is its only invariant. The group $\mathrm{SO}(2)$ is the circle and the canonical form is the rotation by $\theta$, so
$$ m=1,\qquad \operatorname{Spec}(Q)=\{e^{i\theta},e^{-i\theta}\}, $$
with no fixed subspace. The computation confirms the spectrum and the rank two of $Q-\mathrm{I}$ for every $\theta\notin2\pi\mathbb{Z}$.
At $n=3$ every rotation is simple, which is Euler's theorem. For $Q\in\mathrm{SO}(3)$ the characteristic polynomial has odd degree with real coefficients, so it has a real root; the root is an eigenvalue of modulus one, hence it is $\pm1$; and $\det Q=1$ excludes the value $-1$ with multiplicity one, so the eigenvalue one occurs. Hence
$$ \exists\,v\neq0:\ Qv=v, $$
and the rotation is a rotation about the axis $\mathbb{R}v$ by an angle $\theta$. The computation confirms the eigenvalue one at $n=3$ on random rotations, with deviation below $10^{-15}$.
The angle is read from the trace. On the axis the eigenvalue is one and on the plane the eigenvalues are $e^{\pm i\theta}$, so
$$ \operatorname{tr}Q=1+2\cos\theta,\qquad \cos\theta=\frac{\operatorname{tr}Q-1}{2}, $$
which was verified at $n=3$ on random rotations, with the angle recovered from the trace to machine precision. So the two invariants of a rotation of three-space are the axis and the angle, and the angle is a function of the trace alone.
The axis and the angle are the whole conjugacy class. Two rotations of three-space are conjugate in $\mathrm{SO}(3)$ exactly when they have the same angle, since the axis can be moved by a rotation. So
$$ \mathrm{SO}(3)\ \text{classes}\ \longleftrightarrow\ \theta\in[0,\pi], $$
which is the statement that the rank of $\mathrm{SO}(3)$ is one, and the computation confirms that rotations with the same trace are conjugate by exhibiting the conjugating rotation.
The Two Angles at $n=4$
The generic rotation of four-space has two angles and no fixed vector. For a rotation with angles $\theta_1\neq\theta_2$, both nonzero and neither equal to $\pi$, the eigenvalues are $e^{\pm i\theta_1},e^{\pm i\theta_2}$, so the eigenvalue one does not occur and
$$ \dim\ker(Q-\mathrm{I})=0,\qquad \operatorname{rank}(Q-\mathrm{I})=4, $$
which was verified at $n=4$ on random rotations, where the distance from the spectrum to one is bounded away from zero. So the generic rotation of four-space moves every vector and is not simple.
The two angles are independent, and their order is the only ambiguity. The canonical form is the block diagonal of the two planar rotations, and the angles $\theta_1,\theta_2$ can be permuted by conjugating with the swap of the two planes. So
$$ \mathrm{SO}(4)\ \text{classes}\ \longleftrightarrow\ \{(\theta_1,\theta_2):0\le\theta_2\le\theta_1\le\pi\}, $$
which was verified at $n=4$ by comparing the sorted angle pairs of random rotations. So the rank of $\mathrm{SO}(4)$ is two, and the class is a point of a triangle.
The two angles are the two factors of $\mathfrak{so}(4)$. By the splitting $\mathfrak{so}(4)=\mathfrak{so}(3)\oplus\mathfrak{so}(3)$ the rotation is
$$ Q=\exp(\theta_1A_1+\theta_2A_2),\qquad A_1\in\mathfrak{so}(3)\oplus0,\quad A_2\in0\oplus\mathfrak{so}(3), $$
with $A_1,A_2$ commuting, so the two one-parameter groups act independently. The computation confirms the commutation of the two generators and the factorisation at $n=4$, and it shows that the two angles are the two independent rotation rates of the two three-dimensional factors.
A non-simple rotation with equal angles is a special case. If $\theta_1=\theta_2=\theta$ then the rotation is the block diagonal of two equal planar rotations, and its canonical form is the product of the two commuting factors with the same angle; the rotation is still non-simple when $\theta\notin\{0,\pi\}$, since no vector is fixed. The computation confirms the absence of a fixed vector at $n=4$ for equal nonzero angles, and it is the case where the two planes are indistinguishable and the class has a symmetry.
The Count of the Angles and the Rank
The number of independent angles is the rank of the group. A maximal torus of $\mathrm{SO}(n)$ consists of the block diagonal rotations with $m$ independent angles, so
$$ \operatorname{rank}\mathrm{SO}(n)=\Big\lfloor\frac{n}{2}\Big\rfloor, $$
which gives $1,1,2,2,3$ for $n=2,3,4,5,6$, verified by the dimension count against $\dim\mathrm{SO}(n)=\binom{n}{2}$: the angles are $\lfloor n/2\rfloor$ numbers, and the conjugating rotation contributes $n(n-1)/2-\lfloor n/2\rfloor$ parameters. So the count of the angles is one at $n\le3$ and at least two at $n\ge4$, which is the title of the article.
The simple rotations are the ones with one nonzero angle. The set of simple rotations is the union over planes of the rotations of that plane, so
$$ \{\text{simple}\}=\bigcup_{P}\exp(\mathfrak{so}(P)), $$
and the union is over the Grassmannian of planes. The computation confirms at $n=4$ that a rotation with one zero angle is simple and a generic one is not, and it shows that the simple rotations form a closed nowhere dense set inside the group.
The generic rotation at $n\ge4$ is non-simple, so the count grows. Since the two angles of a four-space rotation are independent, the condition $\theta_2=0$ is one equation on the class, and the classes with two nonzero angles form the interior of the triangle; hence the generic rotation requires the full count $\lfloor n/2\rfloor$ of angles. The computation confirms at $n=4,5,6$ that random rotations have all the angles nonzero and no fixed vector.
The angles are the invariants of the conjugacy class, and the count is the dimension of the class space. The class of a rotation is determined by its sorted angles and by nothing else, so the space of classes is the simplex
$$ 0\le\theta_m\le\dots\le\theta_1\le\pi, $$
of dimension $m=\lfloor n/2\rfloor$. So the angles are a complete set of invariants, and the number of angles is the dimension of the class space, which is the rank of the group.
The Case $n=2$ and the Biquaternion Reading
At $n=2$ the single angle is the whole rotation and the class space is a segment. The rotation by $\theta$ has the class $[\theta]$ with $\theta\in[0,\pi]$, and the two endpoints are the identity and minus the identity. The computation confirms the spectrum and the rank of $Q-\mathrm{I}$, and it shows that the class space of the planar rotation is the segment of the angles.
In the biquaternion frame the angle is the phase of the rotor. With $\tilde\Pi_\pm$ the idempotent pair, a planar rotation is
$$ Q=\lambda\,\tilde\Pi_++\bar\lambda\,\tilde\Pi_-,\qquad |\lambda|=1, $$
so the single angle is the argument of $\lambda$, and the two signs of the rotor are the two square roots of $\lambda$. The computation confirms the identity, and it exhibits the whole conjugacy class of the planar rotation as the circle of the phases of $\lambda$, which is the segment of the angles read modulo the identification of the endpoints.
The contrast with $n=3$ is one extra eigenvalue. At $n=3$ the rotation has the eigenvalues $1,e^{\pm i\theta}$, so the angle is read from the trace by the formula $\cos\theta=(\operatorname{tr}Q-1)/2$; at $n=2$ the trace is $2\cos\theta$ and the eigenvalue one is absent. The computation confirms both formulas, and it shows that the axis at $n=3$ is the direction that the planar case does not have.
Summary
A rotation is simple when it turns one plane and fixes the rest, which is the case exactly when $\dim\ker(Q-\mathrm{I})\ge n-2$ and $\operatorname{rank}(Q-\mathrm{I})\le2$. Every rotation of two- or three-space is simple, the three-space case by Euler's theorem, and the angle at $n=3$ is read from the trace by $\cos\theta=(\operatorname{tr}Q-1)/2$. At $n=4$ the generic rotation has two independent angles, eigenvalues $e^{\pm i\theta_1},e^{\pm i\theta_2}$, no fixed vector and full rank $Q-\mathrm{I}$, and its class is a point of the triangle $0\le\theta_2\le\theta_1\le\pi$. The number of independent angles is the rank $\lfloor n/2\rfloor$ of the orthogonal group, so the count is one at $n\le3$ and at least two at $n\ge4$.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $P$, $P^{\perp}$ | the rotating plane and the fixed subspace of a simple rotation |
| $\operatorname{rank}(Q-\mathrm{I})=2$ | the test of simplicity |
| $\dim\ker(Q-\mathrm{I})\ge n-2$ | the eigenvalue one with high multiplicity |
| $\cos\theta=(\operatorname{tr}Q-1)/2$ | the angle at $n=3$ |
| $e^{\pm i\theta_1},e^{\pm i\theta_2}$ | the spectrum of a generic four-space rotation |
| $\lfloor n/2\rfloor$ | the number of independent angles, the rank |
| $0\le\theta_m\le\dots\le\theta_1\le\pi$ | the simplex of the conjugacy classes |
| $Q=\lambda\tilde\Pi_++\bar\lambda\tilde\Pi_-$ | the planar rotation in the biquaternion frame |
Further Reading
- Sigurdur Helgason, Differential Geometry, Lie Groups, and Symmetric Spaces (AMS, 2001), for maximal tori, the rank and the conjugacy classes.
- Morton L. Curtis, Matrix Groups (Springer, 1984), for Euler's theorem and the canonical forms of the rotations.
- Pertti Lounesto, Clifford Algebras and Spinors (Cambridge, 2001), for the planar rotation and the biquaternion picture.
- The Exponential of the Antisymmetric Part: One-Parameter Rotations, for the one-parameter groups whose angles are counted here.
- The Spectral Theorem: the Anti-Hermitian Case and the Reality of the Spectrum, for the canonical block form and the angles.
- Rotors Acting on the Bivector Space: the Adjoint Action, for the two commuting factors of $\mathfrak{so}(4)$.