Rotations Lorentz Transformations
Rotations Lorentz Transformations
Rotations Lorentz Transformations
ARTHUR JAFFE
Contents
I. Lorentz Transformations 2
I.1. Points in R4 and Lorentz Transformations L 2
I.2. Components of L 3
I.3. The Components of L 4
II. Restricted Lorentz Transformations 5
II.1. Proper Rotations 5
II.2. Pure Lorentz Boost: 6
II.3. The Structure of Restricted Lorentz Transformations 7
III. 2 2 Matrices and Points in R4 7
III.1. R4 and H 2 8
III.2. Determinants and Minkowski Geometry 9
III.3. Irreducible Sets of Matrices 9
III.4. Unitary Matrices are Exponentials of Anti-Hermitian Matrices 9
III.5. Strictly-Positive Hermitian Matrices are Exponentials of Hermitian Matrices 10
IV. The Group SL(2, C) Acting on H2 10
V. Structure of Restricted Lorentz Transformations 13
VI. Complex-Conjugate Representations in SU (2) and SL(2, C) 16
VI.1. The Group SU (2) 16
VI.2. The Group SL(2, C) 16
1
2 ARTHUR JAFFE
I. Lorentz Transformations
I.1. Points in R4 and Lorentz Transformations L. Let us designate a point in R4 by the
4-vector with real coordinates x = (x0 , x1 , x2 , x3 ) = (ct, ~x) R4 . In other words, we use units
for which the four components of x all have dimension length, and we say that the four-vector x
comprises a time component and a 3-vector spatial part. A homogeneous Lorentz transformation
is a 44 real matrix that acts on x R4 that preserves the Minkowski length x2M = x20 x21 x22 x23
of every 4-vector x. Let L denote the set of all such Lorentz transformation matrices.
More explicitly, let us denote a Lorentz transformation x 7 x0 by
3
X
x0 = x , with x0 = x ,
=0
02
with the property xM = x2M . The Minkowski square can be written in terms of the Minkowski-space
metric
1 0 0 0
0 1 0 0
g= .
0 0 1 0
0 0 0 1
Denote by h , i the standard scalar product on Euclidean space, hx, yi = 3=0 x y . We take
P
the Minkowski scalar product to equal
3
X
hx, yiM = hx, gyi = x g y = x0 y0 ~x ~y . (I.1)
,=0
I.2. Components of L. We classify the matrices L, the set of all Lorentz transformations,
into four distinct connected components.1 We will show that there are four connected components of
L that are determined by the sign of det and the sign of 00 . Those transformations L with
det = 1 are caller proper Lorentz transformations, and one denotes the set of such transformations
as L+ . Those transformations L with 00 > 0 are called isochronous Lorentz transformations,
and one denotes the set of such transformations as L .
Proposition I.2. The Lorentz transformations L fall into four disconnected, disjoint components
according to the sign of det = 1, and the sign of 00 for which |00 | > 1.
Proof. We have seen in the proof of Proposition I.1 that det = 1. As det is a polynomial in the
matrix elements ij , it depends continuously on these matrix elements. Hence L has disconnected
components according to the sign of det .
The relation (I.2) also shows that |00 | > 1. In fact the 00 matrix element of the identity (I.2)
requires that
3 3
!1/2
X X
200 2i0 = 1 , namely |00 | = 1 + 2i0 >1. (I.6)
i=1 i=1
0 0 tr
In Proposition I.1 we saw that if L, then L. Thus from (I.6) we infer
3
!
0 2 X
00 = 1+ 0j02 > 1 . (I.7)
j=1
1One says that a set LX L of Lorentz transformations is a connected component, if one can find a continuous
trajectory of matrices between any two given 1 , 2 LX . This means that for a parameter 0 6 s 6 1, there
is a family of Lorentz transformations (s) LX , with matrix elements ij (s), such that (s = 0) = 1 and
(s = 1) = 2 .
4 ARTHUR JAFFE
We now see that sgn ((0 )00 ) = sgn 00 sgn 000 . In fact matrix multiplication yields
3
X
0
( )00 = 00 000 + 0j 0j0 . (I.8)
j=1
If 00 and 000 have the same sign, the first term on the right of (I.8) is positive, and the bound
in (I.9) yields (always for positive square roots),
3
!1/2 3
!1/2 3
!1/2 3
!1/2
X X X X
(0 )00 > 1 + 20i 1+ 0j02 20i 0j0 >0. (I.10)
i=1 j=1 i=1 j=1
If 00 and 000have the opposite sign, one has similarly (with positive square roots)
3
!1/2 3
!1/2 3
!1/2 3
!1/2
X X X X
(0 )00 6 1+ 20i 1+ 0j02 + 20i 0j0 60. (I.11)
i=1 j=1 i=1 j=1
0 0
As |( )00 | > 1 for any L, we infer that the components determined by 00 are disjoint.
Table 2. The Multiplication Table for the Components of the Lorentz Group L
L+ L L+ L
L+ L+ L L+ L
L L L+ L L+
L+ L+ L+ L+ L
L L L+ L L+
We use the letter R to denote the 3 3 orthogonal matrix with determinant 1 that implements
the rotation three-vectors ~x. Technically, R SO(3), the group of real, orthogonal, 3 3 matrices
with determinant one.
Each pure rotation matrix R SO(3) is specified by an axis, namely a unit vector ~n in 3-
space, and an angle of rotation about this axis. One writes R = R(~n, ). The rotation leaves ~n
unchanged, and acts in the plane orthogonal to ~n. For example, a rotation by angle about the
third axis ~n = ~e3 is given by the matrix
cos sin 0
R(~e3 , ) = sin cos 0 , (II.2)
0 0 1
Here the sign of the angle is a convention: one can interpret the rotation as rotating the coordinate
system (one sign) or rotating the space in a fixed coordinate system (the other sign). The former
is called a passive transformation, while the latter is called an active transformation.
II.2. Pure Lorentz Boost: A restricted Lorentz transformation L+ is a pure boost in the
direction ~n (here ~n is a unit vector in 3-space), if it leaves unchanged any vectors in 3-space in the
plane orthogonal to ~n. Such a pure boost in the direction ~n depends on one more real parameter
R that determines the magnitude of the boost. By choosing the direction of the boost to be
~n, we can restrict the parameter to be non-negative, 0 6 .
For example the pure Lorentz boost along the first coordinate axis is = B(~e1 , ), where
~e1 denotes a unit vector for the Cartesian direction x1 , and with parameter . This Lorentz
transformation is given by the real, symmetric matrix
cosh sinh 0 0
sinh cosh 0 0
B(~e1 , ) = . (II.4)
0 0 1 0
0 0 0 1
Clearly B = B(~e1 , ) satisfies the defining relation B tr gB = g of a Lorentz transformation, and also
det B = +1. Furthermore B00 = cosh > 1. Thus B L+ . One has the transformed point
x00 = x0 cosh + x1 sinh , and x01 = x0 sinh + x1 cosh , x02 = x2 , and x03 = x3 . (II.5)
LORENTZ TRANSFORMATIONS, ROTATIONS, AND BOOSTS 7
It is interesting to use another parameterization. Define = tanh [0, 1), and so using the
relation cosh2 sinh2 = 1, we see that
1
cosh = p , and sinh = p . (II.6)
1 2 1 2
So in the example,
1 0 0
1 2 1 2
1 0 0
B(~e1 , ) = 1 2 1 2 . (II.7)
0 0 1 0
0 0 0 1
Thus in this example we can write x0 = B(~e1 , )x, expressed above as (II.5), in the familiar form
x0 + x 1 x0 + x1
x00 = p , and x01 = p , x02 = x2 , and x03 = x3 . (II.8)
1 2 1 2
This is a standard form of the pure Lorentz boost along the first axis, where x0 = ct (to give x0 and
x1 the same dimension) and = v/c is the dimensionless velocity of the boost. Here the velocity is
characterized by 0 6 v < c. (Of course one could also consider negative v or if one wished.) Note
0 0
that x02 x12 = x20 x21 .
II.3. The Structure of Restricted Lorentz Transformations. The following describes the
structure of a restricted Lorentz transformation in L+ . We state these properties now in a propo-
sition. We show that this proposition is true toward the end of these notes. Meanwhile we develop
some properties relating points on R4 to hermitian matrices, which we need to analyze Lorentz
transformations in a simple way.
Proposition II.1 (Restricted Lorentz Transformations). Every restricted Lorentz transfor-
mation L+ has a unique decomposition as a product of a pure rotation R followed by a pure
boost B,
= BR . (II.9)
The rotation has the form (II.1) with R SO(3); the boost is a symmetric matrix B SO(1, 3)+ .
Remark II.2. One could write this decomposition of a restricted transformation in the reverse
order = RB,e with a pure boost B e = R1 BR along a different direction. Alternatively there is a
pair of pure rotations R1 , R2 , and a pure boost B(~e1 , ) of the form (II.4), such that
= R1 B(~e1 , )R2 . (II.10)
This decomposition is also unique, unless the Lorentz transformation is a pure rotation.
III.1. R4 and H 2 . There is a 1-1 transformation between points in R4 and the space of 2 2
hermitian matrices that we denote by H 2 . We consider the map
!
x0 + x3 x1 ix2
x = (x0 , ~x) = (x0 , x1 , x2 , x3 ) x
b= . (III.1)
x1 + ix2 x0 x3
We use this correspondence in order to analyze the structure of a general Lorentz transformation.
It is natural to define a scalar product on 2 2 matrices as
1
hA, Bi = Tr(A B) . (III.2)
2
The trace of a matrix, denoted Tr, is the sum of the diagonal entries. It is convenient to introduce
four hermitian matrices for = 0, 1, 2, 3, as a basis for H 2 . Let
! ! ! !
1 0 0 1 0 i 1 0
0 = , 1 = , 2 = , 3 = . (III.3)
0 1 1 0 i 0 0 1
Note that one can write the 4-vector = (I, ~ ). The zero component 0 is the identity matrix;
the three-vector with components i for i = 1, 2, 3 are equal to the 2 2 Pauli matrices. Then
1 2 = i3 = 2 1 , and similarly for cyclic permutation of 1, 2, 3. The matrices are orthonormal
in this scalar product,
h , i = , for , = 0, 1, 2, 3 .
In other words, the are an orthonormal basis for the space of 2 2 hermitian matrices. We
summarize the properties of x
b that follow immediately from this observation:
Proposition III.1. Consider the transformation x 7 x b from R4 to H 2 defined by
!
x0 + x3 x1 ix2
x
b = x0 0 + x1 1 + x2 2 + x3 3 = . (III.4)
x1 + ix2 x0 x3
b 7 x from H 2 to R4 is
The corresponding inverse transformation x
1
x = h , x
bi = Tr ( x
b) . (III.5)
2
Both transformations are linear and 1 to 1.
Proof. The fact that (III.4) and (III.1) agree follows from our choice of . The inverse transforma-
tion is a consequence of the orthonormal property of the s in the scalar product h , i. Finally
for x, y R4 and R, the linearity of the correspondence between x and x b follows from
D E
(x\+ y) = x y , and (x + y) = , x\
b + b + y = h , x
bi + h , ybi = x + y . (III.6)
Thus x x
b is linear and 1-to-1.
LORENTZ TRANSFORMATIONS, ROTATIONS, AND BOOSTS 9
III.2. Determinants and Minkowski Geometry. The link between the mappings of (III.4)
(III.5) and Minkowski geometry comes from observing that the determinant of the hermitian matrix
b is the Minkowski length squared of the four-vector x = (x0 , ~x). In particular defining x2M = x20 ~x 2 ,
x
one has
b = (x0 + x3 )(x0 x3 ) |x1 ix2 |2 = x20 ~x 2 = x2M .
det x
Now we study transformations x b0 that are linear and that preserve the determinant of x
b 7 x b.
III.3. Irreducible Sets of Matrices.
Definition III.2. A set of matrices A is irreducible, if any matrix C that commutes with every
matrix in A must be a multiple of the identity matrix.
Proposition III.3. Any two of the three Pauli matrices {1 , 2 , 3 } are an irreducible set of 2 2
matrices.
!
a b
Proof. Take C = for arbitrary complex numbers a, b, c, d. Explicitly
c d
! ! ! ! ! !
a b 0 1 b a 0 1 a b c d
C1 = = , and 1 C = = .
c d 1 0 d c 1 0 c d a b
Hence C1 = 1 C means that b = c and a = d. Likewise
! ! ! ! ! !
a b 0 i ib ia 0 i a b ic id
C2 = = , and 2 C = = .
c d i 0 id ic i 0 c d ia ib
Hence C2 = 2 C means that b = c and a = d. And finally
! ! ! ! ! !
a b 1 0 a b 1 0 a b a b
C3 = = , while 3 C = = .
c d 0 1 c d 0 1 c d c d
Hence C3 = 3 C means that b = c = 0.
Inspecting the consequences of any two of the three conditions Cj = j C, we infer b = c = 0
and a = d. In other words, C commuting with any two Pauli matrices shows that
! !
a b 1 0
C= =a = aI .
c d 0 1
Thus any two of the Pauli matrices form an irreducible set of 2 2 matrices.
III.4. Unitary Matrices are Exponentials of Anti-Hermitian Matrices. The matrix H is
hermitian, H = H , if and only if the matrix A = iH is anti-hermitian, A = A .
Proposition III.4. Unitary matrices have the following properties:
iH
(1) Every unitary N N matrix U can be written U = Pe3 , where H is hermitian.
(2) In case N = 2, the matrix H can be written H = j=0 j j for real constants j .
(3) In case N = 2 and det U = 1, the matrix H can be written H = 3j=1 j j , namely 0 = 0.
P
10 ARTHUR JAFFE
Proof. Every unitary matrix is normal, so there is another unitary V so that V U V = D is diagonal.
The diagonal entries Djj = dj of D are the eigenvalues of U , and they have absolute value 1.
Therefore dj = eij for some real j . Hence D = ei where is the diagonal hermitian matrix with
real entries ij = j ij . Hence H = V V is hermitian and U = V DV = V ei V = eiV V = eiH
has the form claimed in (1).
In case N = 2, we know that the j for j = 0, 1, 2, 3 are a basis for hermitian matrices,
that is
Porthonormal in the scalar product (III.2). Therefore the given H has a unique expansion
3
H = j=0 j j . In case that det U = 1 as well, one has
Thus one needs to choose 0 = n for integer n. But all these choices yield the same U , so we
choose n = 0.
Proof. The proof is very similar to the proof of Proposition III.4. One significant difference is that
for hj > 0, we can find real j for which hj = ej . Also det H = e20 ; so det H = 1 ensures
0 = 0.
((AB)x) = AB x \ A = ((A)(B)x) .
b B A = A (B)x
Hence for any x, one can invert the map to hermitian matrices to show (AB)x = (A) (B)x. As
a consequence, the identity (AB) = (A) (B) also holds.
In order to see that the Lorentz transformation (A) is an element of L+ , we argue that both
det (A) and (A)00 are continuous functions of the matrix A. For every A SL(2, C) we can find
a continuous path A() from A(0) = I to A(1) = A, which we construct as follows. The matrix A
has a unique polar decomposition A = HU , where 0 < H is a positive hermitian matrix, and U is
a unitary matrix. Every hermitian H and unitary U has an orthonormal basis of eigenvectors, so
we can raise H and U to an arbitrary fractional power [0, 1]. Define
The values of the matrix elements (A) are the coefficients of x in the relation (IV.6). Thus
(A) = h , A A i = 21 Tr( A A ). Also
1 1 1
(A) = Tr( A A ) = Tr (A A ) = Tr ( A A ) = (A) ,
2 2 2
so the matrix elements of (A) are real.
As (A) is a representation, (A1 ) = (A)1 . In case that A = U is unitary, (U )00 =
1
2
Tr(I) = 1. Also (U )0i = 12 Tr(U i U ) = 21 Tr(i ) = 0, and similarly (U )i0 = 0. Furthermore
1 1 1
Tr ( U U ) = Tr U 1 U 1 = Tr U 1 U 1 = (U 1 )
(U ) =
2 2 2
1
= (U ) . (IV.7)
One takes H as the positive square root of the hermitian matrix AA , namely H = (AA )1/2 , and
defines U = H 1 A = (AA )1/2 A. One then sees that U is unitary, for
Along the properties established in Corollary IV.2, this completes the proof of Proposition II.1.
We now find out in detail what are the transformations (H) and (U ).
Proposition V.2 (Identification of (U ) and (H)). If U = ei 2 ~n~ , then (U ) = R(~n, )
rotates by angle about the axis ~n. If H = e 2 ~n~ , then (H) = B(~n, ) is a pure boost along the
axis ~n by velocity vc = = tanh .
Proof. In order
P3 to identify the rotation arising from U = ei 2 ~n~ , note that U commutes with the
matrix nb = j=1 nj j . Thus U nb U = n b0 = n
b, so n b, and consequently (U ) leaves ~n unchanged. In
other words it is a rotation about the axis ~n.
In order to analyze the angle by which one rotates, it is sufficient to choose ~n = ~e3 . Then we
claim that
1 0 0 0
0 cos sin 0
(U ) = (ei 2 3 ) = (ei 2 3 ) = , (V.4)
0 sin cos 0
0 0 0 1
which is just the rotation by angle about the third axis found in (II.3). We compute the matrix
elements of (U ) and show that they equal the matrix elements in (V.4). We have already shown
in Corollary IV.2 that (U )00 = 1 and (U )0i = (U )i0 = 0, for i = 1, 2, 3. We also have
!
i
1 i 3 i 3 1 1 e 0 1 i i
= Tr 12 ei 3 = Tr
(U )11 = Tr 1 e 2 1 e 2 = e + e = cos ,
2 2 2 0 ei 2
14 ARTHUR JAFFE
1 i 3 i 3 1 1
= Tr 1 2 ei 3 = Tr i3 ei 3
(U )12 = Tr 1 e 2 2 e 2
2 ! 2 2
i
1 e 0 1 i
e ei = sin ,
= i Tr i
=
2 0 e 2i
1
1 1
(U )13 = Tr 1 ei 2 3 3 ei 2 3 = Tr (1 3 ) = Tr (i2 ) = 0 ,
2 2 2
1 i 3 i 3 1
= Tr 2 1 ei 3 = (U )12 = sin ,
(U )21 = Tr 2 e 2 1 e 2
2 2
!
i
1 i 3 i 3 1 1 e 0
= Tr 22 ei 3 = Tr
(U )22 = Tr 2 e 2 2 e 2 = cos ,
2 2 2 0 ei
1
1 1
(U )23 = Tr 2 ei 2 3 3 ei 2 3 = Tr (2 3 ) = Tr (i1 ) = 0 ,
2 2 2
1
1 1
Tr 3 ei 2 3 1 ei 2 3 = Tr 3 1 ei3 = Tr i2 ei3
(U )31 =
2 ! 2!! 2 !
1 0 1 e i
0 1 0 ei
= Tr = Tr =0,
2 1 0 0 ei 2 ei 0
1 i 3 i 3 1 1
= Tr 3 2 ei3 = Tr i1 ei3
(U )32 = Tr 3 e 2 2 e 2
2 ! 2 !! 2 !
1 0 i e i
0 1 0 iei
= Tr = Tr =0,
2 i 0 0 ei 2 iei 0
and
1 i 3 i 3 1 1
(U )33 = Tr 3 e 2 3 e 2 = Tr 32 = Tr(I) = 1 .
2 2 2
Putting together all these matrix elements shows that the rotation (U ) just equals (II.3), as
claimed.
Similarly we can work out the matrix (H), for
1 1
H = e 2 ~n~ = e 2 ~n~ , (V.5)
where ~n is a unit vector and is a real parameter. As in the case of the rotation, we include the
factor 12 in the exponent, so that everything works out nicely.
For i, j = 1, 2, 3, we have
i j = j i + 2ij I , for i, j = 1, 2, 3 .
Assume that y = (0, ~y ) where ~y R3 is orthogonal to ~n, namely ~y ~n = 0. Then
(~y ~ ) (~n ~ ) = (~n ~ ) (~y ~ ) ,
and it follows from (V.5) that
~y ~ H = H 1 ~y ~ .
LORENTZ TRANSFORMATIONS, ROTATIONS, AND BOOSTS 15
Then
\ = H~y ~ H = ~y ~ = yb .
(H)y (V.6)
So (H) leaves ~y ~n unchanged. This is the definition of a pure Lorentz boost along ~n. We denote
this transformation B(H).
Let us compute the components of B(H) for H or the form (V.5) and with ~n = e1 . We claim
that in this case,
cosh sinh 0 0
sinh cosh 0 0
B(H) = . (V.7)
0 0 1 0
0 0 0 1
In other words, we claim that B(H) is exactly the boost B(~e1 , ) given in the example (II.4).
In the case that A = H is hermitian, we have shown in Corollary IV.2 that the corresponding
Lorentz boost B(H) is always a real, symmetric matrix, so in order to determine B(H), we only
need to find B(H) for 6 . These components are
!
1
1 1 cosh sinh
B(H)00 = Tr 0 e 2 1 0 e 2 1 = Tr (e 1 ) = Tr = cosh ,
2 2 2 sinh cosh
!
1
1 1 sinh cosh
B(H)01 = Tr 0 e 2 1 1 e 2 1 = Tr (1 e 1 ) = Tr = sinh ,
2 2 2 cosh sinh
1
1
B(H)02 = Tr 0 e 2 1 2 e 2 1 = Tr (2 ) = 0 ,
2 2
and
1
1
B(H)03 = Tr 0 e 2 1 3 e 2 3 = Tr (3 ) = 0 .
2 2
Also
1
1
B(H)11 = Tr 1 e 2 1 1 e 2 1 = Tr (e 1 ) = cosh ,
2 2
1
1 1
B(H)12 = Tr 1 e 2 1 2 e 2 1 = Tr (1 2 ) = Tr (i3 ) = 0 ,
2 2 2
1
1 1
B(H)13 = Tr 1 e 2 1 3 e 2 1 = Tr (1 3 ) = Tr (i2 ) = 0 ,
2 2 2
1
1
B(H)22 = Tr 2 e 2 1 2 e 2 1 = Tr (I) = 1 ,
2 2
1
1 1
B(H)23 = Tr 2 e 2 1 3 e 2 1 = Tr (2 3 ) = Tr (i1 ) = 0 ,
2 2 2
and
1
1
B(H)33 = Tr 3 e 2 1 3 e 2 1 = Tr (I) = 1 .
2 2
So these all agree with (V.7).
16 ARTHUR JAFFE
Here we use cyclicity of the trace, namely Tr(AB) = Tr(BA). Also Tr(Y ) = T r(Y ). Therefore if
X and X are unitarily equivalent, then Tr(X) = Tr(X) must be real.
Hence one way to show that a general matrix A SL(2, C) is not unitarily equivalent to its
complex conjugate is to find an SL(2, C) matrix whose trace is complex. In fact, one can take the
diagonal 2 2 matrix2
!
+i
e 0
A = e3 ei3 = SL(2, C) , with , R . (VI.8)
0 ei
Then
Tr(A) = 2 cosh cos + 2i sinh sin . (VI.9)
This is real only if = 0 or = n. So if we choose , so sinh cos 6= 0, then A is not unitarily
equivalent to A.
Remark VI.1 (Particle/Anti-Particle Symmetry). In quantum theory, complex conjugation
is the symmetry that has the physical interpretation of transforming particle states to corresponding
anti-particle states. (This is a slight oversimplification.)
Furthermore the two-dimensional matrices in SU (2) or SL(2, C) are involved in the interpre-
tation of states of spin- 12 particles, such as electrons and positrons, or protons and anti-protons.
So one can interpret the difference between rotations and Lorentz transformations as saying that
rotational symmetry does not mix spin- 12 fermions from their anti-particles. However relativistic
symmetry can mix the two.
P.A.M. Dirac discovered this fact, when he found his relativistic wave equation for the elec-
tron, the equation we know as the Dirac equation. The Dirac equation is an equation for a
4-component wave function. In order to obtain this equation, one combines the 2-dimensional
standard representation of SL(2, C) (which is the familiar Pauli equation involving the Pauli matri-
ces in non-relativistic quantum theory) with the 2-dimensional complex-conjugate representation,
SL(2, C). Putting them together they make up the 4 components of the relativistic Dirac wave
function. The Weyl equation is a two-component relativistic wave equation that describes particles
or anti-particles. The Dirac equation is a four-component relativistic wave equation that describes
both.
2Our example does not need the trace as a test, for we just observe that the set of eigenvalues of A, namely the
diagonal entries, do not coincide with those of A. However it is instructive to look at invariants of A under unitary
transformations, such as Tr(A) and det(A) which must be the same for unitary equivalent matrices. These serve as
useful tests when A is not diagonal.