Lagrangian and Hamiltonian Formulations of

Relativistic Mechanics
Quinton Westrich
December 2, 2006
Abstract
Lagrangian and Hamiltonian mechanics are modern formulations of mechanics equivalent
to the mechanics of Newton. That is, all three formulations of classical mechanics yield
the same equations of motion for the same physical systems. However, Lagrangian and
Hamiltonian mechanics have the advantages of oering insight into the physical system
not aorded by Newtonian mechanics. It is also often much easier to obtain the equations
of motion for more complicated systems in Lagrangian and Hamiltonian mechanics. One
more advantage of Hamiltonian and Lagrangian mechanics is that they lend themselves to
a straightforward generalization to quantum mechanics via a canonical transformation of
their Lie algebra structures. This motivates us to nd relativistic Lagrangian and Hamil-
tonian formulations. This can be done quite easily for the case of continuum mechanics.
However, it remains dicult to carry this through for discrete degrees of freedom, as is
attempted herein. We begin by constructing a Lagrangian formulation and proceed to con-
struct a Hamiltonian formulation. The approach follows closely that of Goldstein, Poole,
and Safko [3]. An appendix of some of the mathematics used is included, although much
of the complexity of the rigorous mathematics is omitted in the body of the paper in hopes
of increasing readability as well as avoiding distraction away from the insight of physical
intuition.
Contents
1 Lagrangian Formulations of Relativistic Mechanics 2
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The Ad Hoc Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Covariant Lagrangian Formulations . . . . . . . . . . . . . . . . . . . . . . 5
2 Hamiltonian Formulations of Relativistic Mechanics 9
2.1 A Brief Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 The Ad Hoc Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 The Covariant Hamiltonian Formulation . . . . . . . . . . . . . . . . . . . 10
3 Appendix: Some Mathematics 13
1
1 Lagrangian Formulations of Relativistic Mechanics
1.1 Introduction
There are two ways in which a Lagrangian formulation of relativistic mechanics has been
attempted:
1. The rst concentrates on reproducing, for some particular Lorentz frame, the spatial
part of the equation of motion
F
i

dp
i
dt
(1)
where p
i
is some relativistic generalization of the Newtonian momentum that reduces
to mv
i
in the small limit of , the simplest being the four-momentum dened by p =
mu. The forces F
i
may or may not be suitably related to a covariant Minkowski force.
The basis for this method is often quite shaky, especially considering that relativistic
forces are not well formulated (cf. Goldstein-Poole-Safko 297-300). However, quite
often this method will produce equations of motion which, while not being manifestly
covariant, are relativistically correct for some Lorentz frame.
2. The second sets out to obtain a covariant Hamiltons principle and ensuing Lagranges
equations which treat space and time on a common footing as the generalized coordi-
nates in a four-dimensional conguration space. This seems to be clearly the proper
approach, but it quickly runs into problems that are a bit tricky to solve, even for a
single particle. In fact, it breaks down right at the start for systems involving more
than one particle!
No satisfactory formulation for an interacting multiparticle system exists in classical
relativistic mechanics except for a few special cases.
1.2 The Ad Hoc Method
We shall simply look for a Lagrangian that will yield the correct relativistic equations of
motion for some particular inertial system.
Note that we cannot approach this problem from dAlemberts principle,

i
(F
appl
i
p
i
) r
i
= 0. (2)
The principle itself is still true in any given Lorentz frame, but the derivation of the
Lagrangian from dAlemberts principle is based on the equation p
i
= m
i
v
i
, which is no
longer true in relativistic mechanics.
We shall proceed by trying to nd a suitable Lagrangian L that satises Hamiltons
principle,

_
t
2
t
1
L dt = 0, (3)
2
and for which the Euler-Lagrange equations reproduce the known relativistic equations of
motion (1). We make the following ansatz for the Lagrangian of a single particle acted on
by conservative forces, which are independent of the velocity:
L = mc
2
_
1
2
V (r), (4)
where V is the potential and
2
=
v
2
c
2
, and v is the speed of the particle in the Lorentz
frame under consideration.
This is the correct Lagrangian! Indeed, when we put it in the Euler-Lagrange equations,
we obtain
d
dt
_
L
v
i
_

L
x
i
= 0

d
dt
_
mv
i
_
1
2
_
+
V
x
i
= 0

d
dt
(p
i
) =
V
x
i
= F
i
.
Remarks:
1. The Lagrangian is no longer L = T V but it still holds that
L
v
i
= p
i
.
2. We can easily extend this Lagrangian to many particle systems and change from
Cartesian spatial coordinates to generalized coordinates q
i
of the system.
3. By dening the canonical momentum as usual,
p
i

L
q
i
, (5)
we retain the connection between cyclic coordinates and conservation of the corre-
sponding momenta.
4. If L does not contain the time explicitly, there exists a constant of the motion
h = q
i
p
i
L. (6)
3
In fact, h is the total energy:
h = q
i
_
L
q
i
_
+ mc
2
_
1
2
+ V (r)
= v
i
_
mv
i
_
1
2
_
+ mc
2
_
1
2
+ V (r)
=
mv
i
v
i
_
1
2
+ mc
2
_
1
2
+ V (r)
=
m(v
i
v
i
+ c
2
(1
2
))
_
1
2
+ V (r)
=
m(v
i
v
i
+ c
2
v
i
v
i2
)
_
1
2
+ V (r)
=
mc
2
_
1
2
+ V (r)
= mc
2
+ V (r)
= E
5. We can introduce velocity-dependent potentials exactly as before in the case of non-
relativistic mechanics. For example, the Lagrangian of a single particle of charge e
in an electromagnetic eld is
L = mc
2
_
1
2
e + eA v (7)
and as before the canonical momentum is no longer mu, but involves some extra
terms p
i
= mu
i
+ eA
i
.
Some of the problems associated with the ad hoc formulation are listed in the table
below. This motivates our transition to a covariant Lagrangian formulation of relativistic
mechanics.
4
Problem Principle Resolution
1. No eort has
been made to
keep the ideal of
a four-dimensional
covariant form for
all the laws of me-
chanics, i.e. time
has been treated as
a parameter entirely
distinct from the
spatial coordinates.
A covariant formu-
lation would require
that space and time
be considered as en-
tirely similar coordi-
nates in Minkowski
(world) space.
So we must nd some
invariant parameter
instead of t to trace
the path of the sys-
tem in conguration
space.
2. There are ex-
amples of Lagrangian
functions (such as
that of a single par-
ticle in an electro-
magnetic eld listed
above) that possess
no Lorentz transfor-
mation properties.
Hamiltons principle
must be manifestly
covariant.
So the action inte-
gral must be a world
scalar. (If the param-
eter of integration is
a Lorentz invariant,
then the Lagrangian
function itself must
be a world scalar in
any covariant formu-
lation.)
3. The Lagrangian is
a function of nonrel-
ativistic coordinates
and velocities.
The Lagrangian
should be a function
of the coordinates
in Minkowski space
and their derivatives
with respect to an
invariant parameter.
So change the La-
grangian to be a
function of the
Minkowski space
coordinates and
their derivatives with
respect to the new
invariant parameter.
Therefore we must nd the invariant parameter and the Lagrangian must be a function
of the Minkowski spacetime coordinates and their derivatives with respect to the parameter.
1.3 Covariant Lagrangian Formulations
We now seek a covariant formulation of Lagrangian mechanics. Consider a one particle
system. We need two things
1. an invariant parameter of the path in conguration space, and
2. to put the Lagrangian L as a function of the Minkowski space coordinates and their
derivatives with respect to the parameter.
5
The natural choice for the parameter is , the proper time, but
u u = u

= c
2
. (8)
So the generalized velocities in Minkowski space are not independent in this case.
Dene a monotonic function of time (t) such that is Lorentz invariant. We shall use
the notation
(x

dx

d
(9)
and
x

dx

dt
. (10)
A sucient covariant Hamiltons principle is
I
_

1
(x

, (x

)d, (11)
where is a world scalar Lagrangian function and (x

, (x

) means a function of any or

all of the Minkowski space coordinates and respective generalized velocities. The corre-
sponding Euler-Lagrange equations are
d
d
_

(x

= 0. (12)
Thus we have given our demands a mathematical language and the problem is now reduced
to nding a Lagrangian such that the Euler-Lagrange equations Eqs.(12) reduce to the
equations of motion Eqs.(1) for some Lorentz frame.
To nd such a , we shall start with the old action integral
_
t
2
t
1
L dt (13)
and attempt to transform the integral over t to an integral over . We should then treat
time in the old Lagrangian L as a generalized coordinate instead of a parameter. Thus,
dx
i
dt
=
dx
i
d
d
dt
=
_
x
i
_

1
dt
d
= c
_
x
i
_

1
d(ct)
d
= c
(x
i
)

(x
0
)

(14)
and
dt =
dt
d
d =
d (ct)
d
d
c

(x
0
)

c
d (15)
implies
I =
_
t
2
t
1
L
_
x
j
, t, x
j
_
dt =
_
(t
2
)
2
(t
1
)
1
L
_
x

, c
(x
i
)

(x
0
)

_
(x
0
)

c
d. (16)
6
Apparently a suitable covariant Lagrangian function is
(x

, (x

) =
(x
0
)

c
L
_
x

, c
(x
i
)

(x
0
)

_
(17)
Remarks:
1. is a homogeneous function of the generalized velocities in the rst degree, i.e.
(x

, a (x

) = a(x

, (x

). (18)
This is because we used time as a generalized coordinate and another parameter to
describe the path in phase-space. is often called a homogeneous Lagrangian. It
requires a special treatment in the variational calculus.
2. h = 0. This requires a theorem of Euler:
Theorem 1.1 (Eulers theorem on homogeneous functions) Let be a func-
tion homogeneous to rst degree in (x

. Then
= (x

(x

. (19)
3. is homogeneous and obeys the Euler-Lagrange equations Eqs.(12) imply
_
d
d
_

(x

_
(x

= 0. (20)
So, if any three of the Euler-Lagrange equations are satised by a homogeneous
function , the fourth Euler-Lagrange equation is also satised.
We now consider as an example the free particle. From above, the relativistic (but
noncovariant) Lagrangian for a free particle was, from Eq.(4),
L = mc
2
_
1
2
= mc
_
c
2
x
i
x
i
.
Using Eq.(17), we obtain
(x

, (x

) =
(x
0
)

c
mc

c
2

_
c
(x
i
)

(x
0
)

__
c
(x
i
)

(x
0
)

_
= m
_
x
0
_

c
2
_
1
_
(x
i
)

(x
0
)

__
(x
i
)

(x
0
)

__
= mc

_
(x
0
)

_
2
_
1
_
(x
i
)

(x
0
)

__
(x
i
)

(x
0
)

__
= mc

_
_
(x
0
)

_
2
(x
i
)

(x
i
)

= mc

_
(x

(x

.
7
Thus, the Euler-Lagrange equations become
d
d
_

(x

_
= 0

d
d
_
_
mc (x

_
(x

(x

_
_
= 0
.
Now, maps to t monotonically and t maps to monotonically. The composition of two
monotonic functions is monotonic. Hence,
(x

dx

d
=
dx

d
d
d
= u

d
d

d
d
_
mcu

_
=
d
d
(mu

) = 0
since u u = u

= c
2
. So the Euler-Lagrange equations reduce to the form
K

dp

d
= 0, (21)
where K

can be recognized as the familiar Minkowski force, a four-vector which is just

a generalization of Newtons second law in the sense that it obeys
lim
0
K
i
= F
i
.
Let us now consider a method of arriving at a covariant Lagrangian formulation of
relativistic mechanics devised by P.A.M. Dirac (1902-1984). Diracs contribution was to
give the problem a dierent perspective. By noticing that our previous dismissal of as
an appropriate covariant parameter was based not on its restriction of the dynamics of the
system in time, but on the geometry of the possible states in conguration space to some
hyperplane, Dirac labeled this geometric constraint a weak constraint and suggested that
we wait until the last step of the derivation to impose this restriction. Thus we can use
as a parameter of the motion in conguration space.
Also noteworthy is that Eq.(17) is a sucient covariant Lagrangian, but is not neces-
sarily the only covariant Lagrangian. We may choose a Lagrangian such that the action
integral has dierent values when dierent parameters are used. Thus it is not impos-
sible that we nd a covariant Lagrangian that is not homogeneous to rst degree in the
velocities.
In considering multiple particle systems a few problems arise that dont easily go away.
Firstly, which particles should we use? One idea that seems hopeful is to consider
the center of mass . This would at least treat all particles on an unbiased mathemat-
ical ground. Particles in an external eld are no problem as long as the eld equations
are covariant also. Most daunting perhaps is what happens when one tries to consider
8
particle-particle interactions. The famous action-at-a-distance interactions requie a bit
of thought since signals cant travel faster than light in special relativity. Contact forces
seem to be the only ones allowed and in fact, it has been shown that a covariant action-
at-a-distance permitting theory would violate the conservation of total linear momentum.
Some attempts to get around this noninteraction theorem have required approximately
covariant Lagrangians and modied Hamiltons principles.
2 Hamiltonian Formulations of Relativistic Mechan-
ics
2.1 A Brief Introduction
We shall now attempt a Hamiltonian formulation of relativistic mechanics. Just as in the
case of the Lagrangian formulations, there will be two approaches. Each will follow directly
in the footsteps of the corresponding formulation above.
2.2 The Ad Hoc Method
The motivation for the rst approach is this: If the Lagrangian that leads to the Hamil-
tonian is itself based on a relativistically invariant eld theory, then the corresponding
Hamiltonian picture should also lead to the correct (relativistic) equations of motion.
As before, consider a single particle (Eq.(4)):
L = mc
2
_
1
2
V.
Now, we had found that h = E above. But we can identify h as the Hamiltonian H in
disguise so that also
H = T + V. (22)
From relativistic dynamics, it is well known that (with a suitable renaming)
T
2
= m
2
c
4
+ p
2
c
2
. (23)
Taking the square-root and substituting this into our equation for H we have
H =
_
m
2
c
4
+ p
2
c
2
+ V. (24)
Consider the following example as an illustration of how this Hamiltonian formulation
might be used. Of course, we may only speak of a single particle. Let that particle be in
some electrogmagnetic eld. Since we arent able to exploit the covariance of the elds we
use the nonrelativistic notation of the elds, although they transform just as good as ever.
The Lagrangian from the Ad Hoc formulation above (Eq.(7))was
L = mc
2
_
1
2
e + eA v.
9
By mere observation of the forms of the Lagrangian and Eq.(24), we may note that the
linear term in v does not appear in the Hamiltonian. The rst term of L gives the form of
the kinetic energy term in Eq.(24) so that we may posit
H = T + q. (25)
Also from above, the canonical momentum was
p
i
= mu
i
+ eA
i
. (26)
If we denote the mechanical momentum by the vector

i
=mu
i
, (27)
then we may write
p
i
=
i
+ eA
i

i
= p
i
eA
i
.
It follows that
T
2
= m
2
c
4
+
2
c
2
T
2
= m
2
c
4
+ (p eA)
2
c
2

H =
_
m
2
c
4
+ (p eA)
2
c
2
+ q,
where p is the vector of the canonical momenta p
i
which are the conjugate momenta of
the Cartesian coordinates of the particles position in space.
2.3 The Covariant Hamiltonian Formulation
As before, let be a parameter of the path (but this time in phase-space). Then
(q
i
, (q
i
)

, t, t

) = t

L(q
i
,
(q
i
)

, t), (28)
where the prime notation means a (total) derivative with respect to . Lets explore what
happens with this Lagrangian. Denoting the momentum conjugate of t by p
t
, we obtain
p
t
=

(t

)
= L + t

L
(t

)
. (29)
Using derivatives like fractions, we write
q =
q

10
so that
p
t
= L +
q

i
t

L
q
i
= L q
i
L
q
i
= H. (30)
Now since x
0
= ct,
p
0
=
H
c
=
E
c
,
as desired.
The problem remains that h = H = 0, since is homogeneous to rst degree in the
velocties. However, we can use another covariant Lagrangian such as
(x

, u

) =
1
2
mu

(31)
for a free particle. This does in fact give the correct equations of motion. We can also
employ the method of Dirac with
H
covariant
H
c
=

2m
(32)
for a free particle.
Lets see how our single charge e behaves in an electromagnetic eld in this formulation
of relativistic Hamiltonian mechanics. Now our covariant Lagrangian is
(x

, u

) =
1
2
mu

+ eu

(x

) (33)
with canonical momenta
p

= mu

+ eA

+ eA

. (34)
So
H
c
=
(p

eA

) (p

eA

)
2m
. (35)
There are eight equations of motion:
dx

d
=
H
c
g

(36)
and
dp

d
=
H
c
g

, (37)
where the g

are the entries of the Minkowski space metric,

g

=
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
(38)
11
It should be noted that not all x

and p

are independent and, hence, these eight equations

can be reduced to the six which correspond to the spatial part.
Indeed, when we consider a charge in an electromagnetic eld, the zero component of
the velocity is just
u
0
=
H
c
g
0
p

=
H
c
g
00
p
0
=
H
c
p
0
=

p
0
_
(p

eA

) (p

eA

)
2m
_
=
p
0
eA
0
m
or
p
0
=
T + e
c
=
H
c
c
(39)
as was found previously. The second equation with = 0 gives
dp
0
d
=
dp
0
dt
=
H
c
g
00
x
0
=
1
c
H
c
t
or
d
dt
_
H
c
c
_
=
1
c
H
c
t

dH
c
dt
=
1

H
c
t
.
Problems arise if one cannot nd a suitable covariant potential to describe the forces
other than electromagnetism. Also, the action-at-a-distance diculty remains for mul-
tiparticle interacting systems. At the current development of these formulations, it is
obvious that we have much better tools for practical applications. Nevertheless, it remains
a mathematical curiosity and there is much room for further development of the theories.
But perhaps trying to make a theory that inherently treats time as a parameter of the
dynamical variable adhere to relativistic egalitarianism is really stretching old mechanical
ideas to the point of tearing them apart.
12
3 Appendix: Some Mathematics
We shall consider some notations, denitions, and theorems necessary for the derivation.
These are taken from Gelfand and Fomin [2].
Notation 3.1 We denote by C
1
(a, b) the normed function space which has elements y(x)
that are continuous functions of x on the closed interval [a, b] possessing continuous rst
order derivatives which we denote by y

(x). By addition of the elements of C

1
(a, b) we
mean the ordinary addition of functions and by multiplication of elements of C
1
(a, b) by
numbers we mean ordinary multiplication of functions by numbers. The norm of C
1
(a, b)
is given by the formula
y
1
= max
axb
|y(x)| + max
axb
|y

(x)| .
Denition 3.2 The functional J [y] is said to be continuous at the point y C
1
(a, b)
i > 0,

> 0 s.t y C
1
(a, b),
y y
1
<

|J [y] J [ y]| < .

Denition 3.3 The functional J [y] is said to be continuous on C
1
(a, b) if it is contin-
uous at every point in C
1
(a, b).
Denition 3.4 Let [h] be a functional dened on C
1
(a, b). Then [h] is said to be
linear i
1. h C
1
(a, b), R, [h] = [h]
2. h
1
, h
2
C
1
(a, b), [h
1
+ h
2
] = [h
1
] + [h
2
]
Example 3.5 Let [h] =
_
b
a
(x)h(x) dx be s.t. (x) is xed in C
1
(a, b). Then [h]
denes a continuous linear functional on C
1
(a, b).
Example 3.6 Let [h] =
_
b
a
[(x)h(x) + (x)h

(x)] dx be s.t. (x) and (x) are xed in

C
1
(a, b). Then [h] denes a continuous linear functional on C
1
(a, b).
Notation 3.7 We denote by C(a, b) the normed function space which has elements y(x)
that are continuous functions of x on the closed interval [a, b] with addition and multipli-
cation the same as for the space C
1
(a, b). The norm of C(a, b) is given by the formula
y = max
axb
|y(x)| .
Lemma 3.8 If (x) is continuous on [a, b] and if
_
b
a
(x)h(x)dx = 0 for all h C(a, b)
s.t. h(a) = h(b) = 0, then (x) = 0 for all x [a, b].
Lemma 3.9 If (x) is continuous on [a, b] and if
_
b
a
(x)h

(x)dx = 0 for all h C

1
(a, b)
s.t. h(a) = h(b) = 0, then (x) = c for all x [a, b], where c is a constant.
13
Lemma 3.10 If (x) and (x) are continuous on [a, b] and if
_
b
a
[(x)h(x) + (x)h

(x)] dx = 0
for all h C
1
(a, b) s.t. h(a) = h(b) = 0, then (x) is dierentiable and

(x) = (x) for

all x [a, b].
Denition 3.11 Let J [y] be a fuctional dened on C
1
(a, b). We dene the increment
of J [y] to be
J [h] = J [y + h] J [y] .
If y is held xed then J [h] is, in general, a nonlinear functional of h. If there exists a
linear functional [h] s.t.
J [h] = [h] + h
1
where 0 as h
1
0, then we say that J [y] is dierentiable and we call the linear
functional J [h] [h] the variation or dierential of J [y].
Theorem 3.12 The dierential of a dierentiable functional is unique.
Denition 3.13 The functional J [y] has a (relative) extremum for y = y if J [y] J [ y]
does not change its sign in some neighborhood of the curve y = y(x). The functional J [y]
has a weak extremum for y = y if there exists an > 0 such that J [y] J [ y] has the
same sign for all y in the domain of denition of the functional which satisfy the condition
y y
1
< . The functional J [y] has a strong extremum for y = y if there exists an
> 0 such that J [y] J [ y] has the same sign for all y in the domain of denition of the
functional which satisfy the condition y y < .
Theorem 3.14 A necessary condition for the dierentiable functional J [y] to have an
extremum for y = y is that its variation vanish for y = y, i.e., that J[h] = 0 for y = y
and all admissible h.
14
References
[1] Arfken and Weber. Mathematical Methods for Physicists. 5th Ed. Harcourt Academic
Press, 2001.
[2] Fomin, S.V. and Gelfand, I.M. Calculus of Variations. Dover Publications, 2000.
[3] Goldstein, Poole, Safko. Classical Mechanics, 3rd Ed. Pearson Education Inc., 2002
[4] Landau, L.D. and Lifshitz, E.M. Mechanics, 3rd Ed. Course of Theoretical Physics,
Vol. I. Butterworth-Heinemann, 1999.
15