Prerequisites: Part I: Lie Groups and Lie Algebras, familiarity with index notation for vectors and matrices, and special relativity at the level of knowing what a Lorentz boost is.
In the previous post we built the general machinery of Lie groups, i.e. generators, structure constants, representations and the exponential map. Now we use all of it on the group behind all of relativistic physics, the Lorentz group.
In a sense, the Lorentz group is the answer to the question of which linear transformations preserve the structure of spacetime. Everything that follows in quantum field theory (spinors, the Dirac equation, gauge theories) is built on its representation theory. So it is worth going slowly here, even if some of it looks familiar from special relativity.
Definition: the orthogonal group
Before going to spacetime, let me set up the general framework.
Take a space with coordinates . The group of transformations that leaves the quadratic form
invariant is called the orthogonal group .
When all signs are the same (), this is just the familiar rotation group , which preserves . The interesting physics happens when the signs are mixed, i.e. some coordinates enter with a plus and others with a minus. This indefinite signature is exactly what spacetime has.
Definition: the Lorentz group
The Lorentz group is the group of linear coordinate transformations
that leave the following quantity invariant,
which is the spacetime interval. We work in the mostly-minus convention throughout. Hence
The "" counts the spatial dimensions (which enter the quadratic form with a minus sign), and the "" counts time (which enters with a plus). Whether one writes or is a convention, and some authors write it the other way. What matters is the signature, one plus and three minuses.
Lorentz invariance of the Minkowski metric
Which condition does have to satisfy to be a Lorentz transformation? We require the spacetime interval to be invariant,
and substitute ,
where in the last step we relabelled the dummy indices so that the coefficients of can be compared. Since this has to hold for any , we can strip off the ‘s,
or, in matrix notation,
This is the defining equation of the Lorentz group. Compare it with the orthogonal group , whose defining equation is . The structure is the same, with the Minkowski metric in place of the Euclidean metric .
A note on index gymnastics: The metric is what raises and lowers indices, i.e. it converts contravariant (upper) indices into covariant (lower) ones and vice versa. In the equation above, can also change the indices on itself, as the contraction on the right-hand side shows.
How the Lorentz group splits up
The defining equation constrains in a way that splits the Lorentz group into four disconnected components. Let us see how.
By determinant: proper vs. improper
Taking the determinant of both sides of ,
and since and , we can divide through,
This gives two classes.
-
: Proper Lorentz transformations. These form the subgroup , where the "" stands for special, i.e. unit determinant, as in for rotations.
-
: Improper Lorentz transformations. These include parity, time reversal and their combinations. They cannot be continuously connected to the identity (there is no smooth way from to ), so they cannot be reached by exponentiating generators. There is still a continuum of them (parity followed by any rotation is improper), but each one is a discrete operation such as composed with an ordinary proper transformation.
By the 00-component: orthochronous vs. non-orthochronous
Next we look at the component of ,
which gives
Since , either or , with nothing in between. This is again a discrete split.
- : Orthochronous, the transformation preserves the direction of time.
- : Non-orthochronous, the transformation reverses the direction of time.
The four components
Combining the two binary choices, splits into four disconnected components,
| Component | Contains | Example | ||
|---|---|---|---|---|
| Identity | Rotations, boosts | |||
| Combined parity + time reversal | ||||
| Parity (spatial inversion) | ||||
| Time reversal |
Only , the proper orthochronous Lorentz group, is connected to the identity. This is the component that has a Lie algebra, and when physicists say “the Lorentz group” without further qualification they almost always mean it,
The other three components are obtained by applying the discrete transformations , or to elements of .
Non-orthochronous transformations
When , the transformation reverses the direction of time. Any non-orthochronous transformation can be written as an orthochronous one composed with a discrete inversion, and the relevant discrete operations are
- time reversal : , which has and ,
- the combination : , which has and .
Improper Lorentz transformations
Transformations with are improper. Any improper transformation can be written as a proper one () composed with a discrete transformation. Examples are
- Parity: , which flips all spatial coordinates, , orthochronous,
- Single-axis reflection: , which flips one spatial axis, , orthochronous,
- Time reversal: , which flips time, , non-orthochronous.
So parity is improper but orthochronous (), while time reversal is both improper and non-orthochronous. These are genuinely different operations and live in different disconnected components of the Lorentz group.
Lorentz group representations
Now we apply the representation theory from Part I. Recall that a set of objects (with ) transforms in an -dimensional representation of the Lorentz group if, under a Lorentz transformation,
Here is a matrix representation of the abstract group element. The are the Lorentz generators in the representation , which are matrices. The parameters are antisymmetric (), and an antisymmetric matrix has independent components, corresponding to three rotations and three boosts.
For infinitesimal transformations ( small) we expand the exponential to first order,
The pair labels which generator we mean (which rotation or boost), while are the matrix indices of that generator in the representation . The explicit form of as an matrix depends on the representation.
Let us go through the representations one at a time.
Scalar representation
For a scalar , the index takes only one value (), so this is a 1-dimensional representation, and each generator is a matrix, i.e. a single number for each pair .
A scalar field is invariant under Lorentz transformations,
(Invariant means the value does not change. In contrast, covariant means it transforms in a well-defined way under certain rules. All representations are covariant, but only the scalar is invariant.)
Since in this representation,
A representation in which all generators vanish is a valid solution of the Lie algebra (both sides are trivially zero), and it is called the trivial representation.
A typical Lorentz scalar in particle physics is the rest mass of a particle, on which all observers agree regardless of their frame.
Vector representation
A contravariant 4-vector transforms as
and a covariant 4-vector as
with satisfying the Lorentz condition . The spacetime coordinates and the four-momentum are the most important examples of contravariant 4-vectors.
This is a 4-dimensional representation, so each generator is a matrix, . Explicitly,
This is antisymmetric in , as it must be since , and it is the unique generator consistent with the infinitesimal form of .
To see this, take an infinitesimal Lorentz transformation ,
Comparing with the general formula and putting in the explicit generator, one can check that the two agree, .
Lorentz transformations of 4-vectors
Now the explicit transformations. A general Lorentz transformation depends on six parameters,
where are the three rapidity (boost) parameters and are the three rotation angles.
Boosts
A boost along the -axis can be written either with the velocity or with the rapidity. With the velocity (natural units, ) and ,
Since , we can write with the rapidity, and the same transformation becomes
The rapidity form is nicer for a couple of reasons. Rapidities add when collinear boosts are composed (velocities do not), and the hyperbolic functions make the analogy with rotations obvious, since the of rotations become for boosts.
Boosts in the and directions work the same way, with the and in the appropriate row and column.
Boost matrices
The explicit matrices for boosts along each axis are as follows.
Boost along (-form and -form):
Boost along :
Boost along :
The pattern is that a boost mixes the time component (row/column 0) with the spatial component along the boost, and leaves the other two spatial components alone. The sits on the diagonal and the off the diagonal, in the same positions where and sit for rotations.
Rotation matrices
Rotations do not touch the time component at all and act only on the spatial block.
Rotation about -axis:
Rotation about -axis:
Rotation about -axis:
These are the familiar rotation matrices placed in the lower-right block of a matrix, with the time-time component equal to 1 and all time-space components equal to 0.
Rotation and boost generators
We now extract the generators by differentiating the finite transformations and evaluating at the identity (all parameters zero).
Boost generators
Using and ,
The are symmetric matrices (the part in front of which the sits). This reflects the fact that boosts are not unitary transformations. The Lorentz group is non-compact, and boosts move you along a hyperbola instead of around a circle.
Rotation generators
Using and ,
The are antisymmetric (again the part after the ), which reflects that rotations are unitary, since the rotation group is compact.
The Lorentz algebra
The six generators satisfy the commutation relations
and each one has a clear physical meaning.
-
: The rotation generators close among themselves and satisfy the algebra. This just says that the are angular momenta and that rotations form a subgroup.
-
: The boosts transform as a vector under rotations. If you rotate the coordinate system, the boost generators rotate with it, which is what one would expect. A boost “in the -direction” should become a boost “in the -direction” under a rotation about .
-
: This is the most important one. The commutator of two boosts is a rotation, not another boost, so boosts do not form a subgroup. (The rotation obtained this way shows up physically as Thomas precession.) The minus sign, compared to , matters a lot. It is the reason the Lorentz group is non-compact, and the reason it has no finite-dimensional unitary representations apart from the trivial one. In Part III it will also be the reason left- and right-handed spinors are complex conjugates of each other.
If the third relation had a plus sign instead, , then would generate , a compact group with perfectly well-behaved finite-dimensional unitary representations. That one minus sign changes everything.
Connecting to the tensor notation
The six generators and can be packaged into the antisymmetric tensor via
A sign caveatIf you plug the boxed vector-representation formula into this, you get equal to minus the matrices we computed above, while the rotation generators agree. The mismatch comes from the sign convention for the rapidity. Differentiating as we did corresponds to writing , whereas has the opposite sign in the exponent. This is harmless for the algebra, since all three commutation relations are unchanged under . From here on we use the standard identification (as in Peskin & Schroeder).
The tensor form is useful because the Lorentz transformation can then be written compactly as , where the six independent components of are the three rotation angles and the three boost rapidities.
Connection to what comes next
We now have the complete Lorentz algebra, i.e. six generators, their explicit matrices and their commutation relations. The (vector) representation is only one possibility, though. The Lie algebra has infinitely many representations, and the physically most important ones are not the vector representation.
In the next post we complexify the Lorentz algebra by defining , which splits it into . This gives the spinor representations, the and representations that describe left- and right-handed Weyl fermions. From there we construct Dirac and Majorana spinors, and eventually get to the Poincaré group and the Wigner classification of particles by mass and spin.
The minus sign in will be behind all of it.
Next post: Part III: Representations of the Lorentz Group
This post is based on my own self-study notes that I created in 2022 in order to get a deeper understanding of all of this.