Quaternions and Lie groups (part 2)

The unit quaternions and three-dimensional rotations

(Disclaimer: This series assumes some familiarity with linear algebra and Lie groups.)

In the previous part of this series, we introduced quaternions as a natural extension of the real numbers. Now, we’ll present them explicitly and examine them more closely. Quaternions are the algebra spanned by four linearly independent numbers: $1,i,j,k$, subject to the relations

\[i^{2} = j^{2} = k^{2} = ijk = -1\,.\]

These rules, combined with the distributive and associative properties, uniquely determine the multiplication of any two quaternions.

In analogy with complex numbers, we define the real and imaginary quaternions as follows:

\[\operatorname{Re}(\mathbb{H}) = \lbrace a1 \in \mathbb{H}\colon a \in \mathbb{R} \rbrace\,,\] \[\operatorname{Im}(\mathbb{H}) = \lbrace ai + bj + ck \in \mathbb{H}\colon a,b,c \in \mathbb{R} \rbrace\,.\]

For any quaternion $w = a+bi+cj+dk$ with $a,b,c,d \in \mathbb{R}$, the conjugate is defined as

\[\overline{w} = a-bi-cj-dk\,.\]

The quaternion norm can be expressed as

\[\|w\|^2 = w \overline{w} = a^2 + b^2 + c^2 + d^2 \,.\]

Therefore, as a smooth manifold, the unit sphere within the quaternionsI.e. the locus of $\| w \| = 1$. is the three-dimensional sphere, $S^3$. When endowed with quaternion multiplication, this manifold becomes a Lie group called $\mathrm{Spin}(3)$. In particular, it’s a simply connected Lie group,All $S^n$ with $n \geq 2$ are simply connected. a property that will become relevant shortly.

Let’s now derive the Lie algebra of $\mathrm{Spin}(3)$. The tangent space to $\mathrm{Spin}(3)$ at $1$, $T_1 (\mathrm{Spin}(3))$, is naturally identified with the pure imaginary quaternions.To convince yourself of this, take a curve on $\mathrm{Spin}(3)$ passing through $1$ at $t=0$, $q(t) = a(t)+b(t)i+c(t)j+d(t)k$, with $a(0)=1$ and $b(0)=c(0)=d(0)=0$. Because the curve is on the manifold, it satisfies $\| q(t) \|^2 = 1$. Differentiating this condition and evaluating at $t=0$, we obtain simply $a'(0)=0$. Making the natural identification $T_1(\mathbb{H}) \simeq \mathbb{H}$ of the tangent space to a vector space at any point with the vector space itself, the condition before means that the tangent vector to the curve at $q(0)=1$ has zero component along $1$, implying it's pure imaginary. Since the choice of curve is arbitrary, the whole tangent space $T_1 (\mathrm{Spin}(3))$ is naturally identified with the pure imaginary quaternions. If we endow this space with the commutator $[x,y]=xy-yx$ for $x,y \in T_1 (\mathrm{Spin}(3))$, where quaternion multiplication is implied, then the following relations hold

\begin{equation} \label{commutation-relations} [i,j]=2k\,, \quad [j,k]=2i\,, \quad [k,i]=2j\,, \end{equation}

and the commutator is a Lie bracket. It's bilinear, antisymmetric and it satisfies the Jacobi identity. With this operation, the tangent space $T_1 (\mathrm{Spin}(3))$ becomes the Lie algebra of $\mathrm{Spin}(3)$, dubbed $\mathfrak{spin}(3)$. It’s simply the algebra of pure imaginary quaternions under their commutator.

Remarkably, from equation \eqref{commutation-relations} it’s immediately apparent that $\mathfrak{spin}(3)$ is isomorphic to $\mathfrak{su}(2)$, arguably the most conspicuous Lie algebra. The corresponding identifications are

\[i \mapsto -i\sigma_1\,, \quad j \mapsto -i\sigma_2\,, \quad k \mapsto -i\sigma_3\,,\]

where $\sigma_i$ are the Pauli matrices. Therefore, since $\mathrm{Spin}(3)$ and $\mathrm{SU}(2)$ are both simply connected Lie groups,Their underlying manifold is $S^3$ and their Lie algebras are isomorphic, the Lie correspondence guarantees that they are isomorphic as Lie groups.

An important implication of this isomorphism is that $\mathrm{Spin}(3)$, like $\mathrm{SU}(2)$, is a double cover of $\mathrm{SO}(3)$, the group of three-dimensional rotations. This is the defining property of the spin groups. In general, the group $\mathrm{Spin}(n)$ is a double cover of $\mathrm{SO}(n)$.For those who have familiarity with fiber bundles: an orientable vector bundle has an associated principal bundle of oriented orthonormal frames, whose structure group is $\mathrm{SO}(n)$. Under certain topological conditions, which are delicate and we will not delve into, the vector bundle can be equivariantly lifted to a spinor bundle with respect to the double cover $\rho : \mathrm{Spin}(n) \to \mathrm{SO}(n)$. The sections of spinor bundles are called spinor fields. The spinor fields are heavily used in quantum mechanics and quantum field theory. They are a fundamental ingredient in the description of fermions, the particles which obey the Pauli exclusion principle. In addition, understanding spinor bundles is crucial for a proper formulation of quantum field theory on curved spacetimes.

This remarkable property makes quaternions useful to compute three-dimensional rotations of vectors, because multiplication of numbers is much easier to handle computationally than composition of matrices. This is critical in computer graphics, computer vision, robotics, and many other applied fields, where three-dimensional rotations are intensively performed.

Let’s examine how this works in practice. A unit quaternion can be expressed as:

\[q = \cos(\theta/2) + \sin(\theta/2) (u_x i + u_y j + u_z)\,,\]

which represents a rotation by an angle $\theta$ around the unit vector $ \mathbf{u} = (u_x,u_y,u_z) \in \mathbb{R}^3$. Indeed, if we represent a vector $ \mathbf{v} = (v_x, v_y, v_z) \in \mathbb{R}^3$ as a pure imaginary quaternion $v_x i + v_y j + v_z k$, the rotation of $\mathbf{v}$ by an angle $\theta$ around $\mathbf{u}$ is calculated as

\[v' = q v \overline{q}\,,\]

where $q$ is the unit quaternion given above. This is a compact way to encode rotations, requiring only four numbers, as opposed to the nine needed by rotation matrices. Notice that $q$ and $-q$ represent the same rotation, reflecting the fact that the covering of $\mathrm{SO}(3)$ by $\mathrm{Spin}(3)$ is double.

Unlike Euler angles, which have singularities and are susceptible to gimbal lock when rotation axes align, quaternions don’t have singularities and solve the gimbal lock problem.

Furthermore, practical applications often require interpolating between rotations, for instance, to animate transitions, for aircraft navigation, and more. Quaternions make this easy through the slerp technique (spherical linear interpolation). Given two unit quaternions, $q_1$ and $q_2$, the shortest path between them on the unit quaternion sphere is easily parameterized as

\[q(t) = \frac{\sin((1-t)\theta)}{\sin(\theta)} q_1 + \frac{\sin(t\theta)}{\sin(\theta)} q_2\,,\]

with $\theta$ being the angle between $q_1$ and $q_2$. This path is an arc of great circleTo ensure it's the shortest path between the rotations represented by $q_1$ and $q_2$, $\cos \theta$ must be non-negative. Otherwise, $q_2$ has to be replaced by $-q_2$. and it’s traversed at constant speed, leading to very natural-looking transitions. The absence of gimbal lock further guarantees that the interpolation is free from sudden jumps, regardless of the proximity of the interpolated path to singular points in alternative representations. Altogether, these advantages demonstrate how quaternions are useful and why they are preferred in a broad range of practical problems.

In the next part, we’ll introduce the compact symplectic groups, the quaternionic analogues to the unitary and orthogonal groups, and we’ll discuss the fundamental role they have in the classification of simple Lie groups.