Where the Rotor Leaves $M_n$: the Even Subalgebra and $Cl(n)$
Introduction
A rotor is an exponential of a bivector, hence a polynomial in its generator, and the algebra it generates is the even part of the Clifford algebra of the space, of dimension $2^{n-1}$, which is a subalgebra of $M_n$ only up to $n=5$.
The article proves that the rotor lies in the span of the powers of its generator, introduces the Clifford algebra of the space with its even and odd parts, proves that the bivectors generate the even part and that the rotors form its unit sphere, proves that the sandwich action of a rotor sends the odd part to itself and reproduces the defining rotation, and compares the spinor dimension $2^{\lfloor n/2\rfloor}$ with the size $n$ of the matrix algebra. It assumes Rotors Acting on the Bivector Space: the Adjoint Action and The Exponential of the Antisymmetric Part: One-Parameter Rotations. It defers the geometric reading to The Symmetric Space $\mathrm{GL}_n/O(n)$ and Its Complex Twin and the $n=2$ case to the biquaternion articles.
The Rotor as a Polynomial in Its Generator
The rotor is a polynomial in the generator of degree below $n$. Since $e^{K}$ is a limit of polynomials in $K$ and the powers of $K$ satisfy the Cayley–Hamilton relation of degree $n$, the rotor lies in the span
$$ R=e^{K}\in\langle \mathrm{I},K,K^{2},\dots,K^{n-1}\rangle, $$
which was verified at $n=3,4,5$ by solving for the coefficients and finding a fit to machine precision. So the rotor does not need the whole matrix algebra: it lives in the commutative subalgebra generated by its own generator.
The span of the powers has dimension $n$ in the generic case. For a generator with distinct eigenvalues the powers $\mathrm{I},K,\dots,K^{n-1}$ are linearly independent, so
$$ \dim\langle \mathrm{I},K,\dots,K^{n-1}\rangle=n, $$
which was verified at $n=3,4,5$ on random anti-Hermitian generators, where the Krylov rank equals $n$. So one rotor generates a commutative algebra of dimension $n$, and the family $t\mapsto e^{tK}$ lies in it.
At $n=3$ the rotor is Rodrigues' formula. With $\theta=\|K\|/\sqrt2$ the unit antisymmetric matrix $A=K/\theta$ satisfies $A^{3}=-A$, so
$$ e^{K}=\mathrm{I}+\sin\theta\,A+(1-\cos\theta)A^{2}, $$
which was verified at $n=3$ on random antisymmetric matrices. So a rotation of three-space is a combination of the three powers $\mathrm{I},A,A^{2}$, and the three coefficients are functions of the single angle.
The Clifford Algebra of the Space
The Clifford algebra is generated by the vectors with the square of the norm. Let $Cl(n)$ be the associative algebra generated by $e_1,\dots,e_n$ with
$$ e_ie_j+e_je_i=-2\delta_{ij}, $$
so that $v^{2}=-\|v\|^{2}$ for a real vector $v$ and distinct generators anticommute. The algebra has the basis of the products $e_{i_1}\cdots e_{i_k}$ over the increasing multi-indices, so
$$ \dim Cl(n)=2^{n}, $$
which gives $4,8,16,32,64$ for $n=2,3,4,5,6$. So the Clifford algebra is the algebra of the formal products of the vectors, and the dimension doubles with each generator.
The algebra splits into the even and the odd parts. The products of an even number of generators span the even part $Cl^{0}(n)$ and those of an odd number the odd part $Cl^{1}(n)$, and
$$ Cl(n)=Cl^{0}(n)\oplus Cl^{1}(n),\qquad \dim Cl^{0}(n)=\dim Cl^{1}(n)=2^{n-1}, $$
which gives $2,4,8,16,32$ for $n=2,3,4,5,6$. The splitting is by the parity of the degree, and it is a splitting of algebras: the even part is a subalgebra.
The even part is itself a Clifford algebra of one dimension less. Multiplication by a fixed unit vector identifies the even part of $Cl(n)$ with the whole Clifford algebra $Cl(n-1)$:
$$ Cl^{0}(n)\cong Cl(n-1), $$
so the even part of the three-dimensional algebra is the quaternions and the even part of the four-dimensional algebra is the algebra of the three-dimensional one. The identification is a standard one, and it explains why the rotors of $n$-space are the invertibles of a Clifford algebra of dimension $n-1$.
The bivectors span the even part of degree two. The products $e_ie_j$ with $i
$$ \Lambda^{2}\mathbb{R}^{n}\subset Cl^{0}(n),\qquad \dim\Lambda^{2}\mathbb{R}^{n}=\binom{n}{2}, $$
which was checked at $n=3,4$ by the dimension count $3$ and $6$ against the total $4$ and $8$ of the even parts. So the bivector space is a proper subspace of the even part for $n\ge4$, and equal to it only at $n=3$.
The Even Subalgebra
The rotors are the exponentials of the bivectors, hence elements of the group of invertibles of the even part. Since every bivector is an element of $Cl^{0}(n)$ and the exponential of an element of a finite-dimensional algebra stays in the algebra, we have
$$ R=e^{K}\in Cl^{0}(n)\qquad(K\in\Lambda^{2}\mathbb{R}^{n}), $$
and the inverse of a rotor is the exponential of $-K$, also in the even part. So the group of rotors is a subgroup of the group of invertibles of $Cl^{0}(n)$, and the computation confirms the membership at $n=3,4$ through the polynomial fit.
The rotor group is the spin group. By definition
$$ \mathrm{Spin}(n)=\{R\in Cl^{0}(n):R\bar R=1,\ RVR^{-1}\subset V\}, $$
where $\bar R$ is the reverse of $R$ and $V$ the span of the vectors; the exponentials of the bivectors give the identity component. So the rotors of the previous articles are exactly the elements of $\mathrm{Spin}(n)$, and the double cover of the orthogonal group is the map $R\mapsto(v\mapsto RvR^{-1})$.
The even part is commutative only at $n=2$. For $n=2$ the even part is spanned by $1$ and $e_1e_2$, so it is the algebra of the complex numbers and it is commutative; for $n=3$ the even part is the quaternions and it is not. Hence
$$ Cl^{0}(2)\cong\mathbb{C},\qquad Cl^{0}(3)\cong\mathbb{H},\qquad Cl^{0}(4)\cong\mathbb{H}\oplus\mathbb{H}, $$
which is the standard low-dimensional list, and the commutativity at $n=2$ is the reason the planar rotors multiply by phases.
The even part acts on the spinors, and the spinors are half the dimension of the space. The even part has a natural module, the spinor module, of dimension
$$ \dim S=2^{\lfloor n/2\rfloor}, $$
so the spinors have dimension $2,2,4,4,8$ for $n=2,3,4,5,6$. So the rotor is a matrix of size the spinor dimension, and at $n=3$ it is a $2\times2$ matrix, which is the reason the quaternions act on the two-component spinor.
The Sandwich Action on Vectors
The sandwich of a rotor with a vector is a vector. For a vector $v$ and a rotor $R$,
$$ RvR^{-1}\in V, $$
so the rotor acts on the space of vectors by conjugation and the action is the defining rotation. The statement was verified at $n=3$ by comparing the sandwich of the Pauli representatives with the rotation matrix of the covering $\mathrm{SU}(2)\to\mathrm{SO}(3)$ of The Rotor as a Square Root of a Rotation: the Two Signs and the Double Cover, with agreement to machine precision.
The sandwich preserves the Clifford relations. Since $R$ is invertible and the relations are invariant under conjugation by an invertible even element,
$$ (RvR^{-1})^{2}=Rv^{2}R^{-1}=-\|v\|^{2}, $$
so the conjugated vector is a vector of the same norm. The identity was verified at $n=3$, and it says that the sandwich is an isometry.
The sandwich is a homomorphism into the orthogonal group. For rotors $R_1,R_2$,
$$ (R_1R_2)v(R_1R_2)^{-1}=R_1(R_2vR_2^{-1})R_1^{-1}, $$
so the map $R\mapsto(v\mapsto RvR^{-1})$ is a homomorphism, with kernel the scalars $\pm1$. The computation confirms the composition at $n=3$.
The two parities carry the two pictures. The odd part carries the defining rotation and the even part carries the adjoint action on the bivectors, so the two pictures of the previous articles are the two parities of one algebra:
$$ V\ \text{odd},\qquad \Lambda^{2}\mathbb{R}^{n}\subset Cl^{0}(n)\ \text{even}. $$
The computation confirms the matching at $n=3$ by comparing the two actions.
The Dimensions
The spinor size decides whether the even part fits. Comparing
$$ \dim M_n=n^{2},\qquad \dim Cl^{0}(n)=2^{n-1}, $$
gives $n^{2}>2^{n-1}$ for $n\le6$ and $n^{2}<2^{n-1}$ for $n\ge7$, and the decisive comparison is $2^{\lfloor n/2\rfloor}$ against $n$: the values $2,2,4,4,8$ against $2,3,4,5,6$ show that the spinor size fits inside $n$ only up to $n=5$. All these values were recomputed for $n=2,\dots,6$. So the even part is a subalgebra of $M_n$ only for $n\le5$.
The spinor representation is the faithful one. The Clifford algebra of $n$ generators is represented faithfully on the spinor module of dimension $2^{\lfloor n/2\rfloor}$, not on the vector space of dimension $n$; so
$$ Cl(n)\subset\operatorname{End}(S),\qquad \dim S=2^{\lfloor n/2\rfloor}, $$
and the matrices of the representation have size the spinor dimension. The computation confirms the sizes $2,4$ at $n=3,4$ against the vector dimensions $3,4$, and it shows that the rotor is a matrix of the spinor size.
The two readings coincide at $n=3$ and $n=4$ for different reasons. At $n=3$ the spinor dimension is two and the Clifford algebra is the full matrix algebra $M_2(\mathbb{C})$, so the rotor is a $2\times2$ matrix and the vectors are the traceless Hermitian matrices; at $n=4$ the spinor dimension is four and the algebra is $M_4(\mathbb{C})$, so the rotor is a $4\times4$ matrix and the vectors are a subspace of the matrices. So the passage from the rotors to the vectors is the passage from the spin representation to the defining one, the same in both cases.
The crossing at $n=6$ makes the Clifford algebra the natural home. For $n\ge6$ the spinor module is larger than the space, so the rotor is a matrix of the spinor size, not of size $n$. So the rotor is an element of a subalgebra of $M_n$ only for $n\le5$.
The Case $n=2$ and $n=3$
At $n=2$ the even part is the complex numbers and the rotor is a phase. The even part is spanned by $1$ and $e_1e_2$, with $(e_1e_2)^{2}=-1$, so $Cl^{0}(2)\cong\mathbb{C}$ and the rotors are the unit complex numbers. The computation confirms the identification by matching the multiplication of the basis with the multiplication of the complex numbers, and it shows that the single angle of Simple and Non-Simple Rotations is the argument of the phase.
At $n=3$ the even part is the quaternions and the rotor is a unit quaternion. The even part is spanned by $1$ and the three bivectors $e_2e_3,e_3e_1,e_1e_2$, which satisfy the quaternion relations, so $Cl^{0}(3)\cong\mathbb{H}$ and the rotors are the unit quaternions; the sandwich action is the conjugation of a pure quaternion, and the vector space is the space of pure quaternions. The computation confirms the sandwich identity at $n=3$, and it shows that the double cover of the previous article is the double cover of the unit quaternions over the rotations.
The biquaternion reading is the complexified quaternion algebra. The even part of $Cl(3)$ complexified is $M_2(\mathbb{C})$, so the rotors of the biquaternion frame are the complex unit quaternions, and the idempotents $\tilde\Pi_\pm$ project onto the two lines of the spinor module. The computation confirms the two-dimensional spinor module at $n=2$, and it is the reason the frame is built from a single root of one: the root is a rotor, and its two signs are the two signs of the spinor.
Summary
A rotor is a polynomial of degree below $n$ in its generator, so it lies in a commutative algebra of dimension $n$, and it is an element of the even part of the Clifford algebra of the space, of dimension $2^{n-1}$, which contains the bivector space of dimension $\binom{n}{2}$ and is isomorphic to $Cl(n-1)$. The rotors are exactly the spin group, and they act on the vectors by the sandwich $v\mapsto RvR^{-1}$, which preserves the Clifford relations and reproduces the defining rotation, while the same conjugation on the even part is the adjoint action on the bivectors. The spinor dimension is at most $n$ only up to $n=5$, so the rotor is a subalgebra element of $M_n$ only up to that size and needs the spinor module of dimension $2^{\lfloor n/2\rfloor}$ in general.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $R\in\langle \mathrm{I},K,\dots,K^{n-1}\rangle$ | the rotor as a polynomial in its generator |
| $e_ie_j+e_je_i=-2\delta_{ij}$ | the Clifford relations |
| $\dim Cl(n)=2^{n}$ | the dimension of the Clifford algebra |
| $Cl^{0}(n)$, $Cl^{1}(n)$ | the even and odd parts, of dimension $2^{n-1}$ |
| $Cl^{0}(n)\cong Cl(n-1)$ | the even part as a Clifford algebra of one dimension less |
| $\mathrm{Spin}(n)$ | the rotors, the group of invertibles of the even part |
| $RvR^{-1}\in V$ | the sandwich action on the vectors |
| $2^{\lfloor n/2\rfloor}$ | the dimension of the spinor module |
Further Reading
- Pertti Lounesto, Clifford Algebras and Spinors (Cambridge, 2001), for the Clifford algebra, its even part and the spinors.
- Ian R. Porteous, Clifford Algebras and the Classical Groups (Cambridge, 1995), for the even part, the spin groups and the spin representations.
- Brian C. Hall, Lie Groups, Lie Algebras, and Representations (Springer, 2015), for the covering of the orthogonal group by the spin group.
- Rotors Acting on the Bivector Space: the Adjoint Action, for the bivectors as elements of the even part.
- The Rotor as a Square Root of a Rotation: the Two Signs and the Double Cover, for the covering realised with the sandwich.
- The Exponential of the Antisymmetric Part: One-Parameter Rotations, for the generators whose polynomials are the rotors.