Physics 411 Lecture 9

Relativistic Mechanics
Lecture 9

Physics 411
Classical Mechanics II

September 17th, 2007

We have developed some tensor language to describe familiar physics – we

reviewed orbital motion from the Lagrangian and Hamiltonian points of
view, and learned how to write the equations of motion generically in terms
of a metric and its derivatives (in the special combination Γαµν ). All of
the examples involved were in flat, three-dimensional space, and the most
we have done is introduce curvilinear coordinates, i.e. there has been no
fundamental alteration of Euclidean geometry.
Our next step might be to discuss dynamics, or at least, parametrized “mo-
tion” in a truly curved space, and for that, we would need a digression
into what curvature really is. We will do that, of course, but for now, I’d
like to introduce four-dimensional space-time, a slightly different alteration
of geometry. The space-time of special relativity is the basic framework
for general relativity, so looking at four-dimensional Minkowski geometry,
while still flat, allows us to introduce some of the ideas that will be useful
when we take away the metric entirely. This is also a good time to talk
about the meaning of parameters in equations of motion when time itself is
a coordinate (also true in GR). Finally, we can look at the Killing vectors as-
sociated with the non-dynamical metric, and discuss the role of infinitesimal
generators once again, this time for special relativistic mechanics.

9.1 Minkowski Metric

A defining feature of special relativity, that there is a universal speed mea-

sured to have the same value in any frame (any “laboratory”, moving w.r.t.
another or not), is expressed mathematically as:

−c2 dt2 + dx2 + dy 2 + dz 2 = −c2 dt̄2 + dx̄2 + dȳ 2 + dz̄ 2 , (9.1)

1 of 7
9.2. LAGRANGIAN Lecture 9

where dxα =(c

˙ dt, dx, dy, dz) is the coordinate differential in one frame, and
dx̄α =(c
˙ dt̄, dx̄, dȳ, dz̄) is the coordinate differential in another frame. The
two frames are related via a Lorentz transformation: Λαβ . In fact, we can
view the Lorentz transformation as defined by the above – it is the linear
transformation that makes the above true – if we take dx̄µ = Λµν dxν , then
dx̄µ ηµν dx̄ν = Λµα Λνβ ηµν dxα dxβ , (9.2)
where we define the Minkowski metric for Cartesian coordinates as
 
−1 0 0 0
 0 1 0 0 
ηµν =  0 0 1 0 .
˙   (9.3)
0 0 0 1
The requirement of (9.1), written in matrix form, is ΛT η Λ = η. This is very
much analagous to RT I R = I for rotations (where the metric “matrix” is
just the identity).
The interesting new element of the Lorentz transformations is the “boost”,
a mixing of temporal and spatial coordinates. We typically describe a frame
Ō moving at speed v relative to O along a shared x̂ axis as in Figure 9.1 –
then the Lorentz boost takes the form:
 
γ −γ β 0 0
 −γ β
Λµν =
˙ 
γ 0 0   γ≡p 1 β≡ .
v
(9.4)
 0 0 1 0  1−β 2 c
0 0 0 1

We also have all the rotation structure from Euclidean space built in –
consider a transformation defined by Λµν expressed in block form:
 
µ 1 0
Λ ν=˙ , (9.5)
0 R
(the upper entry is 1×1, while the lower right is a 3×3 matrix) for any matrix
R with RT R = I – then we again preserve dxα ηαβ dxβ = dx̄α ηαβ dx̄β .

9.2 Lagrangian

We have two basic problems to address in generating the (a) Lagrangian

appropriate to dynamics in special relativity. The first is the form of the free
particle Lagrangian, and the second is the parametrization of free particle
motion.

2 of 7
9.2. LAGRANGIAN Lecture 9

ŷ ȳ
O Ō

x̄
x̂

Figure 9.1: Coordinate systems related by a Lorentz boost – a frame O sees

the Ō moving to the right with speed v.

9.2.1 Length Extremization

To get the form, we can motivate from classical mechanics. For our usual
free particle Lagrangian in three spatial dimensions with time playing the
role of curve parameter, we have L = 12 mv 2 . If we think of this in terms of
the infinitesimal motion of a particle along a curve, we can write:
1 dxµ dxν
L= m gµν (9.6)
2 dt dt
with gµν the metric appropriate to the coordinates we are using. For con-
creteness, take dxµ =(dx,
˙ dy, dz), i.e. Cartesian coordinates – then gµν can
be represented by the identity matrix. It is clear that the above is directly
related to:
dr2 = dxµ gµν dxν , (9.7)
or what we would call the square of an infinitesimal distance, implicitly,
“along the curve parametrized by t”. If we make this explicit, associating
dxµ dxµ dxν
xµ (t) with a curve, µ 2 2
R then dx = dt , and dr = dt gµν dt dt . Think of
the action, S = Ldt – for L proportional to the length (squared) of the
curve, we are, roughly speaking, minimizing the length of the curve itself
(a.k.a. making a “line”). This is not entirely clear here, so we might ask
what happens if we really take a Lagrangian proportional to dr – suppose

3 of 7
9.2. LAGRANGIAN Lecture 9

we started with the action:

Z Z
p
S= dr = ẋµ gµν ẋν dt. (9.8)

p
In two dimensions, which suffices, we have the integrand L = ẋ2 + ẏ 2 , and
the variation gives us:

ẋ (ẋ ÿ − ẏ ẍ) ẏ (ẏ ẍ − ẋ ÿ)

0= 3/2
0 = 2 , (9.9)
2
(ẋ + ẏ ) 2 (ẋ + ẏ 2 )3/2

and these are degenerate equations – the solution is, for example

y(t) = A + B x(t), (9.10)

an infinite family of extremal paths! Or is it? For any A and B, what we are
really doing is writing y as a linear function of x, that defines a line as the
curve in the x − y plane, regardless of x(t). We have, in effect, parametrized
a line with x itself. That is a consequence of the manifest reparametrization
invariance of this action, which is directly proportional to the length along
the curve. We will see the same sort of action (by construction) for special
relativity.

9.2.2 “Length” Extremization

On the relativistic side, we have a line element that is indefinite:

ds2 = −c2 dt2 + dx2 + dy 2 + dz 2 = dxµ ηµν dxν . (9.11)

Now suppose we parametrize our curve in D = 3 + 1 via some ρ, to which we

attach no physical significance. The goal of (special) relativistic mechanics
is to find xα (ρ), that is: t(ρ), x(ρ), etc. So we can once again write the ds2
along the ρ-parametrized curve as:

dxµ (ρ)
ds2 = ẋµ ηµν ẋν dρ2 ẋµ ≡ (9.12)
dρ

Motivated by the above action (9.8) for a purely spatial geometry, take:
Z Z
p
S=α ds = α −ẋµ ηµν ẋν dρ, (9.13)

4 of 7
9.2. LAGRANGIAN Lecture 9

1 and call this the relativistic, free particle action. We again expect straight
lines as the free particle motion, and we have gained reparametrization in-
variance as before. Explicitly, suppose we have in mind a function of ρ call
it τ (ρ), and we want to rewrite this action in terms of τ – by the chain rule,
we have
dxµ (τ (ρ)) dxµ dτ
ẋµ = = , (9.14)
dρ dτ dρ
so that the action is
r
dxµ dxν dρ
Z
dτ
S=α − ηµν dτ
dρ dτ dτ dτ
Z r (9.15)
dxµ dxν
=α − ηµν dτ,
dτ dτ
and we have no way of establishing the difference between two different
curve parametrizations from the action integrand alone. This gives us the
freedom to define parametrizations of interest. In particular, we know that
for any instantaneous velocity of a particle, it is possible to develop a Lorentz
transformation to the local rest frame – that is, we can always generate
a boost matrix that takes us from a “lab” in which a particle moves in
time and space to the frame in which the particle moves only in time (use
the instantaneous velocity to define the boost parameter γ – this is shown
diagramatically in Figure 9.2).
This local rest frame, in which the particle moves only in a time-like direction
serves to define the proper time τ of the particle. Since a generic trajectory
will not have uniform velocity, τ changes along the trajectory, and we can
define it only in an infinitesimal sense – we know there exists a “barred”
frame (the local rest frame) in which a lab measurement of (dt, dx, dy, dz)
is purely temporal:

−c2 dt2 + dx2 + dy 2 + dz 2 = −c2 dt̄2 ≡ −c2 dτ 2 . (9.16)

Using this defining relation, we can write the derivative of τ w.r.t t and vice
versa:
"  2  2  2 !#
2 2 1 dx dy dz
c dt 1 − 2 + + = c2 dτ 2 (9.17)
c dt dt dt
1
The minus sign under the square root is an artefact of the signature of our metric –
for a particle at rest, moving through time only, we want a real action.

5 of 7
9.2. LAGRANGIAN Lecture 9

ct particle light

ct̄

x̄

Figure 9.2: For a material particle trajectory in space time (D = 1 + 1

here), we can define a Lorentz boost to the frame (c t̄, x̄) shown such that
the particle is instantaneously at rest (the c t̄ axis is tangent to the particle
trajectory) – physically, this is the frame in which the particle moves only
in time, and defines the “proper time” of the particle.

where the term involving the squares of coordinate-time-derivatives is what

we would call the instantaneous lab velocity of the particle. This provides
a physical connection between the parameter τ and measurements made in
the lab. We can write the above differential relation as
1
ṫ2 = 2 . (9.18)
1 − vc2

We started by suggesting that a natural Lagrangian for special relativity

would be one that was proportional to the generalized length along the dy-
namical free particle curve. In order to see that this is a consistent descrip-
tion, we can take the relativistic Lagrangian (the integrand of the action)
in coordinate time parametrization and see what it reduces to in the low-
velocity limit where we should recover the classical free particle Lagrangian
(the kinetic energy).
Since S is reparametrization invariant, it takes the same form for any pa-
rameter, in particular, for t parametrization:
Z r
dxµ dxν
S=α − ηµν dt (9.19)
dt dt
and we would call the Lagrangian (the integrand of the action)
r
p v2 1 v2
L = α c2 − v 2 = α c 1 − 2 ∼ α c − α . (9.20)
c 2 c

6 of 7
9.2. LAGRANGIAN Lecture 9

If we demand that this be the kinetic energy in the low velocity limit, then
we must set α = −m c. We then have an approximate Lagrangian that
differs from T = 12 m v 2 only by a constant, and that will not change the
equations of motion, it is ignorable under variation.
This leaves us with the final form for a free particle relativistic action and
associated Lagrangian:
Z r
dxµ dxν
S = −m c − ηµν dτ (9.21)
dτ dτ
where we use the proper time as the parameter, and we can, at any point,
connect this to coordinate time via (9.18). The advantage to proper time as
a parameter is that the action is manifestly a scalar. We know that the four
µ
velocity dx µ
dτ is a contravariant four-vector (since dx is, and dτ is clearly a
scalar), the metric ηµν is a covariant second rank tensor, so the term inside
the square root is clearly a scalar.
Now we can begin the same sorts of analysis we did for the classical La-
grangian, using the above action as our starting point. In particular, it will
be interesting to find the canonical infinitesimal generators associated with
constants of the motion, although we already know basically what these are
(Lorentz transformations, after all, have ΛT η Λ = η).

7 of 7