Hamiltonian discovered the following Quaternion Algebra over reals.
Hamilton’s quaternions is the four dimensional algebra
with the relations
Here and hence all reals are assumed to commute with every element.
This algebra can also be seen as a two dimensional algebra over complex numbers
where is satisfies
But the complex numbers are not in the center.
We can also think of the algebra as a sum of scalar part and the vector part . The algebra now is defined by the relations
Scalar, vector parts are also called real and imaginary parts.
We have a conjugation map given by
The conjugation of a product reverses the order, that is
Norm is defined by
We also denote it by
This allows us to invert non-zero elements (both right and left inverse) and we have a division algebra.
Unit quaternions are the elements with norm
These unit quaternions precisely are the roots of . That is
.
Hence the name imaginary.
The norm is multiplicative, that is
which follows from writing and grouping terms.
Mutliplicativity of the norm already gives interesting information. For instance it’s useful to study representation of numbers as sums of four squares. The multiplicativity is equivalent to the following identity
$latex \displaystyle \left(a_{1}^{2}+\right.&\left.a_{2}^{2}+a_{3}^{2}+a_{4}^{2}\right)\left(b_{1}^{2}+b_{2}^{2}+b_{3}^{2}+b_{4}^{2}\right)=
\left(a_{1} b_{1}-a_{2} b_{2}-a_{3} b_{3}-a_{4} b_{4}\right)^{2}+\left(a_{1} b_{2}+a_{2} b_{1}+a_{3} b_{4}-a_{4} b_{3}\right)^{2} \quad +\left(a_{1} b_{3}-a_{2} b_{4}+a_{3} b_{1}+a_{4} b_{2}\right)^{2}+\left(a_{1} b_{4}+a_{2} b_{3}-a_{3} b_{2}+a_{4} b_{1}\right)^{2} &fg=000000$
We have the following matrix representations of the algebra
With complex matrices: The left action by an element on consider as the two dimensional complex vector space gives the map
Basically
Restrict to the unit quaternion we have an isomorphism to
With real matrices:
This corresponds to the left multiplication on the quaternions.
Similarly we have a representation corresponding to the right multiplication
Because the left and right actions commute, and so these matrices commute for all values of the parameters!
Three dimensional Geometry:
If we think in terms of scalar and vector parts of the quaternions, geometric operations can be defined in terms of the algebra.
Dot product of two vectors and
is given by scalar part of any of , , , .
The cross product is the vector part of . That is
In general, we have
In particular if the vectors are orthogonal, we have
and the product is anti-commutative. In fact if we , the vectors have to be orthogonal.
We have the general formulae
Reflections:
The action corresponds to the reflection in the hyperplane perpendicular to
To see this note that is sent to . And for any vector orthogonal to , we have the anti-commutativity which implies
So the hyperplane perpendicular to is fixed and is sent to , which proves that it’s a reflection.
3D Rotations:
The unit quaternions act by conjugation on the vector parts . This action corresponds to rotations in .
where is a unit quaternion.
From the multiplicativity of norms, we see that , so the action is a rotation.
We denote
So we can write every quaternion as
If we have
,
then the conjugate action corresponds to a rotation of about the axis .
The action fixes , so it has to be a rotation about .
To see that the conjugate action is rotation by angle precisely equal to , we need to get expression for rotations of by about .
By splitting the component of along and perpendicular to it, we get the Rodrigues formula
Let’s now compute with .
This matches with Rodrigues for rotation by , so we have checked that under conjugation rotates a vector by about the axis .
4D Rotations:
We can see that left and right multiplication by unit quaternions preserve the norm. Hence we can think of these actions as rotations on the 4-dimensional space. In fact the left and right actions commute, so they give a commuting pair of rotation of 4D space. In fact, every rotation is obtained this way!
So the left action with and right action with are both rotations of the 4-D space, but when you apply them together the vector part is invariant, and we get a rotation inside the 3-D space.
The left action by is a rotation in the 4-D space but it acts by the usual rule as a rotation of on the complex planes and .
In general the left action by is a rotation (counter clockwise) by for 3-D vectors in the plane perpendicular to and in the complex plane .
The right action by will also be a rotation by , but now the orientation is clockwise if we take vector perpendicular to .
Note that these left and right action act as rotation only on the planes perpendicular to . Other vectors could moved be away from the vector part and get a scalar part after the action. That is the left (and right) action won’t preserve the 3-D space.
But when we act by conjugation the whole 3-D space is invariant and we have a rotation by in the counter clockwise direction because of left action by and in the clockwise direction from the right action by , so in total a rotation of like we mentioned above.
If a rotation has two real eigenvalues, then they should be both , which implies that there are two lines which are fixed and hence which is almost fixed pointwise (modulo reflections). Such rotations are called simple rotation. If there are no real eigen values, there will be two orthogonal sets of 2-planes on which the rotation acts as 2-d rotation.
In terms of the quaternions, the rotation is simple iff the and have scalar parts that add to zero.
Also the left (right) action will correspond to “isoclinic” rotations – that is a rotation in which the angles of rotation on the two set of orthogonal planes are both equal (if the angles are different, we call them double rotations). Isoclinic rotations are special because they have infinitely many invariant planes- where the rotation acts shift through the same angle on all the planes.
If , by the matrix representation we have the rotation
On the other hand if we have
we can solve for the in terms of . Explicitly we compute
is of rank and can be uniquely (up to sign) written as
where . So we can recover and given the rotation .
To see that it is of rank compute every minor and show that it’s zero. For example the minor
Use the fact that to show that this is equal to zero.
It’s easy to check the factorization for infinitesimal elements near the identity. If
The factorization is
The existence of this factorization in terms of the quaternions is a very non-trivial property which is true only for 4-D and reflects special structure of the 4-dimensional geometry.