A group is a set with a defined product. This product is associative. The group is closed under this product. We have an identity element and each element has an inverse. Formally,

Definition (1.10.1)
A group is a (nonempty) set \(G\) with a binary operation (a product) \(G \times G \rightarrow G\) satisfying the following properties:
  1. The group is closed under the operation.
  2. The product is associative. For all \(a, b, c \in G\), we have \((ab)c = a(bc)\).
  3. There is an identity element \(e \in G\) with property that for all \(a \in G\), \(ea = ae = a\).
  4. For each element \(a \in G\), there is an element \(a^{-1}\ \in G\) satisfying \(aa^{-1} = a^{-1}a = e\) (The inverse).


What are examples of the groups?

  1. \(\mathbf{R}\) with the addition operation.
  2. \(\mathbf{R}^{x} = \mathbf{R} - \{0\}\) with the multiplication operation.
  3. \(\mathbf{Z}\) with the addition operation.
  4. The set of invertible \(n \times n\) matrices with entries in \(\mathbf{R}\) with matrix multiplication as the product.
  5. Any vector space is a group if you forget about the scalar multiplication.
  6. For any set \(T\), define the set of all bijections \(g: T \rightarrow T\). The set of all bijections is the symmetric group of \(T\). The operation here is the composition of these maps.


Example

Suppose we define the group \((\mathbf{R}, \ast)\) where \(\ast\) is defined as

$$ \begin{align*} x \ast y = \sqrt[3]{x^3 + y^3} \end{align*} $$

We can further check that this product satisfies the three axioms of a group.

Example

Let \(F\) be a field (for example \(\mathbf{R}\) or \(\mathbf{C}\)). Let \(Mat_{n \times n}(F)\) be the set of matrices with entries in \(F\). Then define \(GL_n(F)\) (General Linear Group) as a subset of \(A \in M_{n \times n}(F)\) where \(A\) is an invertible matrix.


The set \(GL_n(F)\) equipped with matrix multiplication is a group. It satisfies the three axioms

  1. The group is closed under multiplication because multiplying two invertible matrices is another invertible matrix. \((AB)^{-1} = B^{-1}A^{-1}\)
  2. The identity element is the identity matrix.
  3. Matrix multiplication is associative.


Rotations in Plane

We can define a plane rotation by an angle counter clockwise by the left multiplication by the following rotation matrix

$$ \begin{align*} Rot(\theta) &= \begin{pmatrix} \cos\theta & -\sin\theta \\ \sin\theta & \cos\theta \end{pmatrix} \end{align*} $$

The rotation matrix has some properties

  1. \(\text{Rot}(0) = I\) is the identity matrix.
  2. \(\text{Rot}(\alpha)Rot(\beta) = \text{Rot}(\alpha + \beta)\).
  3. \(\text{Rot}(\alpha)^{-1} = \text{Rot}(-\alpha)\).
  4. \(\text{Rot}(\alpha + 2\pi n) = \text{Rot}(\alpha)\) where \(n \in \mathbf{Z}\).

From this we see that the collection of rotation matrices forms a group with matrix multiplication.

Rotations in Space

$$ \begin{align*} \text{Rot}_{e_3}(\theta) &= \begin{pmatrix} \cos\theta & -\sin\theta & 0 \\ \sin\theta & \cos\theta & 0 \\ 0 & 0 & 1 \end{pmatrix} \end{align*} $$

This defines a rotation around the \(z-\)axis by angle \(\theta\) counter clockwise viewed from the head of the vector \(e_3\) where \(e_3 = (0,0,1)\). But what if we wanted to rotate around a different vector? let \(u \in \mathbf{R}^3\) such that \(\lVert {u} \rVert = 1\) and let \(\text{Rot}_u(\theta)\) be a rotation matrix around the axis through \(u\) by angle \(\theta\) counterclockwise when viewed from the head of \(u\). How do we compute such a matrix?



How Do you Compute The Rotation Matrix?

Recall from Linear Algebra that

Definition
An orthogonal matrix is a square matrix \(P \in Mat_{n \times n}(\mathbf{R})\) such that \(P^{T}P = I\) or \(P^{-1} = P^{T}\).


Equivalently the columns of \(P\) are an orthonormal basis of \(\mathbf{R}^n\) and the rows of \(P\) are also orthonormal basis of \(\mathbf{R}^n\). Based on this, we have a useful formula where if we’re given \(u \in \mathbf{R}^3\) such that \(\lVert u \rVert = 1\), \(\theta \in \mathbf{R}\) and \(P\) an orthogonal matrix, then \(P \text{Rot}_u(\theta) P^{-1}\) is actually a rotation matrix around the vector \(Pu\). So

$$ \begin{align*} \text{Rot}_{Pu}(\theta) = P \text{Rot}_u(\theta) P^{-1} \end{align*} $$

Why is this a useful formula? Given \(\lVert u \rVert = 1\) and some angle \(\theta\), construct an orthonormal basis using Gram-Schmidt by setting \(u_3 = u\) and finding two more perpendicular vectors (normalized) \(u_1, u_2 \in \mathbf{R}^3\). Let \(P = [u_1 \quad u_2 \quad u_3]\). This means that \(Pe_3 = u_3 = u\). Therefore

$$ \begin{align*} \text{Rot}_{u}(\theta) = P \text{Rot}_{e_3}(\theta) P^{T} \end{align*} $$

So we’re taking the standard rotation around the \(z\)-axis or \(e_3\) and changing the basis from the standard vectors \(e_1, e_2, e_3\) to \(u_1, u_2, u_3\) so now the rotation is actually around \(u_3\) instead.



Example

So now suppose we want to rotate around \(\frac{(e_1 + e_2)}{\sqrt{2}}\) (which is a unit vector) by \(\theta = \frac{\pi}{3}\). Then, we’ll need two more orthonormal vectors in addition to \(\frac{(e_1 + e_2)}{\sqrt{2}}\) to construct an orthonormal basis. We can use Gram-Schmidt to come up with the following orthonormal vectors and set them to be the column vectors of \(P\) as follows

$$ \begin{align*} P &= \begin{pmatrix} 0 & \frac{1}{\sqrt{2}} & \frac{1}{\sqrt{2}} \\ 0 & -\frac{1}{\sqrt{2}} & \frac{1}{\sqrt{2}} \\ 1 & 0 & 0 \end{pmatrix} \end{align*} $$

The rotation \(\text{Rot}_{e_3}\) is as follows

$$ \begin{align*} \text{Rot}_{e_3}(\frac{\pi}{3}\big) &= \begin{pmatrix} \frac{1}{2} & -\frac{\sqrt{3}}{2} & 0 \\ \frac{\sqrt{3}}{2} & \frac{1}{2} & 0 \\ 0 & 0 & 1 \end{pmatrix} \end{align*} $$

Therefore,

$$ \begin{align*} \text{Rot}_{\frac{e_1 + e_2}{\sqrt{2}}}\big(\frac{\pi}{3}\big) &= P \text{Rot}_{e_3}(\frac{\pi}{3}\big) P^{T} \\ &= \begin{pmatrix} \frac{3}{4} & \frac{1}{4} & \frac{\sqrt{3}}{8} \\ \frac{3}{4} & \frac{3}{4} & -\frac{\sqrt{3}}{8} \\ -\frac{\sqrt{3}}{8} & \frac{\sqrt{3}}{8} & \frac{1}{2} \end{pmatrix} \end{align*} $$


Rotations in Space

From last lecture, we saw the symmetries of the square. These symmetries can be represented with rotation matrices. We have the identity rotation \(I\). We can also define the rotation around the axis coming through the centroid of the face (\(z-\)axis in the lecture) as \(R = \text{Rot}_{e_3}(\frac{\pi}{2})\). Therefore, \(R^2 = \text{Rot}_{e_3}(\pi)\) and \(R^3 = \text{Rot}_{e_3}(\frac{3\pi}{2})\).

What about the \(180\) degrees rotation around the \(x-axis\) labeled as \(a\) from last time? We can define it as \(A = \text{Rot}_{e_1}(\pi)\). Similarly, \(B = \text{Rot}_{e_2}(\pi)\), \(C = \text{Rot}_{\frac{e_1 - e_2}{\sqrt{2}}}(\pi)\) and \(D = \text{Rot}_{\frac{e_1 + e_2}{\sqrt{2}}}(\pi)\)

Rotations in space have also the following properties / identities

  1. \(\text{Rot}_u(0) = I\) is the identity matrix where \(u\) is any unit vector.
  2. \(\text{Rot}_u(\theta + 2\pi n) = \text{Rot}_u(\theta)\) where \(n \in \mathbf{Z}\).
  3. \(\text{Rot}_u(\theta) = \text{Rot}_{-u}(-\theta)\).
  4. \(\text{Rot}_u(\theta)^{-1} = \text{Rot}_{u}(-\theta) = \text{Rot}_{-u}(\theta)\).
  5. \(\text{Rot}_u(\alpha)Rot_v(\beta) = a\) is a rotation matrix.
  6. \(\text{Rot}_u(\alpha)Rot_u(\beta) = Rot_u(\alpha + \beta)\).

But why is the last statement true? To figure it out, we need to introduce another definition

Definition
A special orthogonal matrix is an \(A \in Mat_{n \times n}(\mathbf{R})\) such that
  1. \(A^TA = I\). (The orthogonal property)
  2. \(\det(A) = \pm 1\). (This is as a result of the fact that \(det(A^{-1}) = \frac{1}{\det(A)}\) and that \(\det(A) = \det(A^T)\)


And now we have the following proposition

Proposition
\(A \in Mat_{3 \times 3}(\mathbf{R})\) is a rotation matrix if and only if it is special orthogonal.


We denote these matrices special matrices with

$$ \begin{align*} SO(n) = \{ A \in Mat_{n \times n} \text{ special orthogonal} \} \subseteq O(n) \subseteq GL_n(\mathbf{R}) \end{align*} $$

Observation: \(A, B \in SO(n)\) implies that \(AB \in SO(n)\). To show that this is true, we need to show that properties of special orthogonal matrices hold. To see that, observe that

  • \( (AB)^TAB = B^TA^TAB = B^TB = I \)
  • \( \det(AB) = \det(A)\det(B) = 1 \)


Also observe that the identity matrix is in \(SO(n)\) and that for any \(A \in SO(n)\), \(A^{-1} = A^{T} \in SO(n)\). Therefore \(SO(n)\) is a group with matrix multiplication.

So now we know that the collection of special orthogonal matrices is a group. Therefore, if the proposition we introduced earlier holds (where we said that \(A \in Mat_{3 \times 3}(\mathbf{R})\) is a rotation matrix if and only if it is special orthogonal), then we can also conclude that the product of two rotation matrices is also a rotation matrix. So let’s sketch the proof of the proposition



Proof
\(\Rightarrow\) (\(\text{Rot}_u(\theta) \in SO(3)\)):
Let \(\text{Rot}_u(\theta) = P\text{Rot}_{e_3}(\theta)P^{-1}\) for some chosen \(P \in O(3)\). We want to show that \(P\text{Rot}_{e_3}(\theta)P^{-1} \in SO(3)\). To do so we need to show that if \(A \in SO(n)\) and \(P \in O(n)\), then \(PAP^{-1} = PAP^{T} \in SO(3)\). We can easily show this by verifying the two properties of special orthogonal matrices. For example,

$$ \begin{align*} (PAP^{T})^{T}PAP^{T} &= PA^TP^TPAP^{T} \\ &= PA^TAP^{T} \quad \text{ (because $P \in O(n)$)} \\ &= PP^{T} = I. \end{align*} $$

The verification that the determinat is 1 is also the same. So now what’s left is to show that \(\text{Rot}_{e_3}(\theta)\) is in \(SO(3)\). We can verify this because we know the exact matrix of \(\text{Rot}_{e_3}(\theta)\) so we can computationally verify that it is a special orthogonal matrix. Finally, since \(P \in O(n)\) and we just showed that \(\text{Rot}_{e_3}(\theta) \in SO(3)\), then \(P\text{Rot}_{e_3}(\theta)P^{T}\) is in \(SO(3)\) as we wanted to show.

\(\Leftarrow\) (\(A \in SO(3) \implies A = \text{Rot}_u(\theta)\)):
For this we’ll use the fact that if \(A \in SO(3)\), then \(1\) is an eigenvalue of \(A\). So let \(u\) be the eigenvector of \(A\) corresponding to \(\lambda = 1\) so \(Au = u\). Choose \(\lVert u \rVert = 1\). Now let \(u_3 = u\) and form an orthonormal basis \(\{u_1, u_2, u_3\}\). Now Let \(P = [u_1 \quad u_2 \quad u_3]\) and let \(B = P^{-1}AP = P^{T}AP\). We know that \(Pe_3 = u_3\) but \(u_3\) is an eigenvector of \(A\).

$$ \begin{align*} Be_3 = P^{-1}APe_3 = P^{-1}Au_3 = P^{-1}u_3 = e_3 \end{align*} $$

So \(e_3\) is an eigenvector of \(B\) … then by the fact we used in the left direction, we can conclude that \(B \in SO(3)\) (Why?). So this means that \(B\) has orthonormal columns since it’s in \(SO(3)\) and the third column specifically is \(e_3\). So

$$ \begin{align*} B &= \begin{pmatrix} a & b & 0 \\ c & d & 0 \\ 0 & 0 & 1 \end{pmatrix} \end{align*} $$

By an algebraic argument we can show that \(a = d = \cos\theta\) and \(c = -b = \sin\theta\). So \(B\) must be \(\text{Rot}_{e_3}\). Therefore, \(A = P\text{Rot}_{e_3}P^{T}\)….



References