Chapter Overview
3.3 Rotations in three Dimensions
The standard methods to rotate vectors in three dimensions uses 3×3 rotation matrices:
Rx=
Analogous to the definition for SO(2), these matrices form a
ba-sis21for a group, called SO(3). This means an arbitrary element of 21This notion is explained in ap-pendix A.1.
SO(3)can be written as a linear combination of the above matrices.
If we want to rotate the vector
v=
around the z-axis22, we multiply it with the corresponding rotation 22A general, rotated vector is derived explicitly in appendix A.2.
To get a second description for rotations in three dimensions, the first thing we have to do is find a generalisation of complex numbers in higher dimensions. A first guess may be to go from 2-dimensional complex numbers to 3-dimensional complex numbers, but it turns out that there are no 3-dimensional complex numbers. Instead, we can find 4-dimensional complex numbers, called quaternions. The quaternions will prove to be the correct second tool to describe rota-tions in 3-dimensions and the fact that this tool is 4-dimensional re-veals something deep about the universe. We could have anticipated this result, because to describe an arbitrary rotation in 3-dimensions, 3parameters are needed. Four dimensional complex numbers, with
the constraint to unit quaternions23, have exactly 3 degrees of free- 23Remember that we used the con-straint to unit complex numbers in the two dimensional case, too.
dom.
3.3.1 Quaternions
The 4-dimensional complex numbers can be constructed analogous to the 2-dimensional complex numbers. Instead of just one complex
"unit" we introduce three, named i, j, k. These fulfil
i2=j2=k2= −1. (3.25) Then a 4-dimensional complex number, called a quaternion, can be written as
q=a1+bi+cj+dk. (3.26)
We now need multiplication rules for ij =? etc., because products like this will occur when one multiplies two quaternions. The extra condition
ijk= −1 (3.27)
suffices to compute all needed relations, for example ij=k follows from multiplying Eq. 3.27 with k:
ij kk
=−1
= −k→ij=k. (3.28)
The set of unit quaternions q=a1+bi+cj+dk are those satisfy-ing the condition24
24The symbol †, here is called "dagger"
and denotes transposition plus complex conjugation: a† = (a)T. The ordinary scalar product always includes a trans-position a·b=aTb, because matrix multiplication requires that we multiply a row with a column. In addition, for complex entities we include complex conjugation that makes sure we get something real, which is important if we want to interpret things in terms of length.
q†q=! 1
→ (a1−bi−cj−dk)(a1+bi+cj+dk) =a2+b2+c2+d2 !=1. (3.29) Exactly as the unit complex numbers formed a group under complex number multiplication, the unit quaternions form a group under quaternion multiplication.
Analogous to what we did for two-dimensional complex numbers, we represent each of the three complex units with a matrix. There are different possible ways of doing this, but one choice that does the job is as complex 2×2 matrices:
1=
1 0 0 1
, i=
0 1
−1 0
j=
0 −i
−i 0
, k=
i 0 0 −i
. (3.30)
You can check that these matrices fulfil the defining conditions in Eq. 3.25 and Eq. 3.27. A generic quaternion can then be written, using
these matrices, as
q=a1+bi+cj+dk=
a+di −b−ci b−ci a−di
. (3.31)
Furthermore, we have
det(q) =a2+b2+c2+d2. (3.32) Comparing this with Eq. 3.29 tells us that the set of unit quaternions is given by matrices of the above form with unit determinant. The unit quaternions, written as complex 2×2 matrices therefore fulfil the conditions
U†U=1 and det(U) =1. (3.33)
In addition, the matrices in Eq. 3.30 are linearly independent25and 25This notion is explained in ap-pendix A.1.
therefore form a basis for the group called SU(2). Take note that the way we define SU(2)here, is analogous to how we defined SO(2). The S denotes special, which means det(U) = 1 and U stands for
unitary, which means the property26U†U=1. Every unit quaternion 26For some more information about this, have a look at the appendix Sec. 3.10 at the end of this chapter.
can be identified with an element of SU(2).
Now, how is SU(2)related to rotations? Unfortunately, the map
between SU(2)and27SO(3)is not as simple as the one between U(1) 27Recall that SO(3)is the set of the usual rotation matrices acting on 3 dimensional vectors.
and SO(2).
In 2-dimensions the 2 parameters of a complex number z =a+ib could be easily identified with the two spatial axes, i.e. v = x+ iy. The restriction to unit complex numbers automatically makes
sure that the resulting matrix preserves the length of any vector28 28Recall that here R is a unit complex number, because complex numbers can be rotated by multiplication with unit complex numbers. Therefore we have RR=1, which is the defining condition for unit complex numbers.
(Rz)Rz = zRRz = zz. The quaternions have 4 parameters, so an identification with the 3 coordinates of a usual three-dimensional vector is not obvious. If we define
v≡xi+yj+zk (3.34)
we can compute, using the matrix representation of the quaternions det(v) =x2+y2+z2. (3.35) Therefore, if we want to consider transformations that preserve the length of the vector(x, y, z), we must use matrix transformations that preserve determinants. The restriction to unit quaternions means that we must restrict to matrices with unit determinant. Everything seems straight forward, but now comes a subtle point. A first guess would be that a unit quaternion u induces a rotation on v simply by multiplication. This is not the case, because the product of u and v
may not belong toRi+Rj+Rk. Therefore, the transformed vec-tor can have a component we are not able to interpret. Instead the transformation that does the job is given by
v=q−1vq. (3.36)
It turns out that by making this identification unit quaternions can describe rotations in 3-dimensions.
Let’s take a look at an explicit example: To make the connection to our example in two dimensions, we will define u as a unit vector in Ri+Rj+Rk and denote a unit quaternion with
t=cos(θ) +sin(θ)u. (3.37) Using Eq. 3.34 a generic vector can be written
v= (vx, vy, vz)T=vxi+vyj+vzk =
Eq. 3.31
ivz −vx−ivy
vx−ivy −ivz
. (3.38) With the identifications made above, we want to rotate, as an ex-ample, a vectorv = (1, 0, 0)Taround the z-axis. We will make a particular choice for the vector and for the quaternion representing the rotation and show that it works. We write, using quaternions in their matrix representation (Eq. 3.30)
v= (1, 0, 0)T →v=1i+0j+0k=
0 −1
1 0
(3.39)
Rz(θ) =cos(θ)1+sin(θ)k=
cos(θ) +i sin(θ) 0 0 cos(θ) −i sin(θ)
, (3.40) which we can rewrite using Euler’s formula29
29For a derivation have a look at
ap-pendix B.4.2. eix =cos(x) +i sin(x)
⇒Rz(θ) =
eiθ 0 0 e−iθ
. (3.41)
Inverting the quaternion rotation matrix yields
Rz(θ)−1 =
cos(θ) −i sin(θ) 0 0 cos(θ) +i sin(θ)
=
e−iθ 0 0 eiθ
. (3.42) The rotated vector is then using Eq. 3.36
v=Rz(θ)−1vRz(θ) =
e−iθ 0 0 eiθ
0 −1
1 0
eiθ 0 0 e−iθ
=
0 −e−i2θ e2iθ 0
=
0 −cos(2θ) +i sin(2θ) cos(2θ) +i sin(2θ) 0
(3.43) On the other hand, an arbitrary vector can be written in this quaternion notation (Eq. 3.38)
v=
ivz −vx−ivy vx−ivy −ivz
(3.44)
which we now compare with Eq. 3.43. This yields
vx=cos(2θ) , vy= −sin(2θ) , vz =0. (3.45) Therefore, written again in the conventional vector notation
→ v = (cos(2θ),−sin(2θ), 0)T. (3.46)
The identifications do indeed induce rotations30, but something 30See the example in Eq. 3.24 where we rotated the vector, using the conven-tional rotation matrix
needs our attention. We haven’t rotatedv byθ, but by 2θ. There-fore, we defineφ≡2θ because then φ really represents the angle we rotate and rewrite Eq. 3.37, which yields
t=cos(φ
2) +sin(φ
2)u. (3.47)
We can now see that the identifications we made are not one-to-one, but rather we have two unit-quaternions describing the same
rotation. For example31 31Because a rotation byπ is the same as
a rotation by 3π=2π+π for ordinary vectors, because 2π =360◦is a full rotation. In other words: We can see that two quaternions u and−u can be used to rotate a vector byπ.
tφ=π =u
((
tφ=3π = −u
uu
Vector Rotation by π2
This is the reason SU(2)is called the double-cover of SO(3). It is always possible to go unambiguously from SU(2)to SO(3)but not vice versa. One may think this is just a mathematical side-note, but we will understand later that groups which cover other groups are
indeed more fundamental32. 32To spoil the surprise: We will use
the double cover of the Lorentz group, instead of the Lorentz group itself, because otherwise we miss something important: Spin. Spin is some kind of internal momentum and one of the most important particle labels. This is discussed in detail in Sec. 4.5.4 and Sec. 8.5.5.
In order to reveal the group that covers a given group, we need to introduce the most important tool of Lie theory: Lie algebras, which is the topic of the next section.
Take note that the fact we had one quaternion parameter too much, may be interpreted as a hint towards relativity. One may argue that a more natural identification would have been, as in the
two-dimensional case, v =t1+xi+yj+zk. We see that pure math-ematics pushes us towards the idea of using a 4th component and what could it be, if not time? If we now want to describe rotations in 4-dimensions, because we know that the universe we live in is 4-dimensional, we have two choices:
• We could search for even higher dimensional complex numbers or
• we could again try working with quaternions.
From the last paragraph it may seem that quaternions have some-thing to say about rotations in 4 dimensions, too. An arbitrary ro-tation in 4 dimensions is described by336parameters. There is no
33Using ordinary matrices, we need in four dimensions 4×4 matrices. The two conditions OTO=1 and det(O) = 1 reduce the 16 components of an arbitrary 4×matrix to six independent components.
7-dimensional generalisation of complex numbers, which together with the constraint to unit objects would have 6 free parameters, but we see that two unit quaternions have exactly 6 free parameters.
Therefore, maybe it’s possible to describe a rotation in 4-dimensions by two quaternions? We will learn later that there is indeed a close connection between two copies of SU(2)and rotations in four dimen-sions.