Yu
Yu
FIELD
THOMAS YU
Abstract. This paper will, given some physical assumptions and experimen-
tally verified facts, derive the equations of motion of a charged particle in
an electromagnetic field and Maxwell’s equations for the electromagnetic field
through the use of the calculus of variations.
Contents
1. Introduction 1
2. Preliminaries 2
3. Derivation of the Electromagnetic Field Equations 8
4. Concluding Remarks 15
References 15
1. Introduction
In introductory physics classes students obtain the equations of motion of free
particles through the judicious application of Newton’s Laws, which agree with em-
pirical evidence; that is, the derivation of such equations relies upon trusting that
Newton’s Laws hold. Similarly, one obtains Maxwell’s equations from the applica-
tion of Coulomb’s Law, special relativity, and other ancillary laws that agree with
empirical evidence. Arriving at such equations through an exploration of various
laws and relationships is usually the main goal of introductory electromagnetism
classes. However, with the calculus of variations, one can derive all of these equa-
tions neatly with a few physical assumptions and a single variational principle: the
principle of stationary action.
With regard to modeling physical phenomena, the functionals used to describe
systems in nature usually consist of an integral over time of a function called the
Lagrangian. In the words of John Baez, a noted mathematical physicist, “The
Lagrangian measures something we could vaguely refer to as the ’activity’ or ’live-
liness’ of the system.”[4] The arguments of the Lagrangian are those functions we
are interested in for use in modeling the behavior of the system; for instance, in
the modeling of the behavior of a free particle, the Lagrangian’s arguments con-
sist of the particle’s position and velocity functions. When we attempt to find
extrema of these functionals with respect to the arguments, in the case of the free
particle, we are locating from the set of all possible position and velocity functions
those particular position and velocity functions which lead to zero variation in the
functional(this will be explained later; for now, think of these particular functions
as akin to critical points in differential calculus). The principle of stationary ac-
tion expresses the experimental fact that systems in nature tend to favor behavior,
expressed in terms of the argument functions, that are critical points of the func-
tional describing them. In terms of John Baez’s words, we can intuitively express
this fact with the phrase“Nature is lazy”; in essence, nature acts to minimize its
’activity’ or liveliness’ as expressed in terms of the functional and the Lagrangian.
Through the calculus of variations we will define these notions rigorously and apply
them to electromagnetism with the aim of deriving the equations of motion in an
electromagnetic field as well as Maxwell’s equations.
2. Preliminaries
Definition 2.1. A functional is a function which maps functions to R. Let C be
a vector space of functions with norm kk. We say ϕ[h], where h ∈ C, is a linear
functional if ∀h ∈ C
Example 2.6. Let y ∈ C 1 [a, b]. The arc length of the graph of y between the
points a and b is
Zb p
(2.7) 1 + (y 0 )2 dx
a
(2.10)
Z
J= L(x0 , . . . , xn , u1 , . . . , um , ∂x0 u1 , . . . , ∂xn u1 , . . . , ∂x0 um , . . . , ∂xn um )dx0 . . . dxn
X
The crux of the calculus of variations lies in analyzing functionals and those
functions which minimize or maximize the value of the functionals through the
variation of the functional. For example, equipped with the concept of variation, it
is possible to find the function which gives the shortest distance in the functional
above(a line as one might imagine), or finding the function for the tunnel through
two given points on the Earth which minimizes transit time.
To that end, we shall now explore the concept of the variation of a functional.
Remark 2.11. We can think of the increment function h as an analogue to the h in
the definition of the derivative, limh→0 f (x+h)−f
h
(x)
.
If the increment of J can be expressed as
Construction 2.15. We will now derive the variation of functionals of the form
Z
(2.16) J[u] = L(x0 , . . . , xn , u, ∂x0 u, . . . , ∂xn u)dx0 . . . dxn
X
Z
(2.20) J[u] = L(→
−
x , u, ∇u)d→
−
x
R
where R is the variable region over which the functional is defined.
We now consider a set of transformations
(2.21) x∗ = Φ (→
i
−
x , u, ∇u; ) i
(2.22) u = Ψ(→
∗ −
x , u, ∇u; )
which depend on the parameter , are differentiable with respect to infinitely
many times, and are equal to the identity transformation when = 0. Let ∼ signify
equality except for terms of order higher than 1 relative to ; the reason for this
will become clear momentarily.
x∗i = Φi (→
−
x , u, ∇u; 0) + ∂ Φi (→
−
x , u, ∇u; )|=0 + O(2 )
(2.26) = xi + φi + O()
u = Ψ(→
∗ −
x , u, ∇u; 0) + ∂ Ψ(→
−
x , u, ∇u; )|=0 + O(2 )
(2.27) = u + ψ + O()
where O(2 ) represents higher order terms and
φ = ∂ Φ (→
i
−x , u, ∇u; )|
i =0
ψ = ∂ Ψi (→
−
x , u, ∇u; )|=0
Thus, if we eliminate the higher order terms with respect to of the transforma-
tions of xi , then we can write the Jacobian of that set of transformations as
LAGRANGIAN FORMULATION OF THE ELECTROMAGNETIC FIELD 5
1 + ∂φ∂x
0
∂φ
∂x0
1
··· ∂φ
∂x0
n
∂φ0 0 ∂φ1
∂x1 1 + ∂x1 ··· ∂φ n
∂x1
(2.28) B= .. .. ..
···
. . .
∂φ0 ∂φ1
∂x n
∂x n
··· 1 + ∂φ n
∂xn
∂φ0 ∂φn
det(B) ∼ (1 + ) . . . (1 + )
∂x0 ∂xn
n
X ∂φi
(2.29) ∼1+
i=0
∂xi
The above follows from the definition of the determinant as a sum of products
and discarding the higher order terms.
Thus, we can write
n
−
→
Z X ∂φi
(2.30) ∆J ∼ [(L(x∗ , u∗ , ∇∗ u∗ )(1 + ) − L(→
−
x , u, ∇u)]d→
−
x
R i=0
∂xi
−
→
We now use Taylor’s Theorem to expand L(x∗ , u∗ , ∇∗ u∗ ) about (→
−
x , u, ∇u) and
write
n n
X X ∂L
(2.31) L∗ − L ∼ (∂xi L)δxi + (∂u L)δu + ( ∂u )δ(∂xi u)
i=0 i=0
∂ ∂xi
where δxi , δu, δ(∂xi u) represent the significant terms, with respect to , of x∗i −
x, u∗ − u, (∂x∗i u∗ ) − (∂xi u). From (2.26) and (2.27), it is clear that δxi = φi and
δu = ψ.
Now consider the increment ∆u = u∗ (x) − u(x) where we see the change in the
two functions at the same coordinates. It can be shown from the definition of δu
that
n
X
(2.32) δu = δu + (∂xi u)δxi
i=0
(2.33) δu = ψ
Pn
where ψ = ψ − i=0 (∂xi u)φi .
Analogously, it can be shown that
n
X
(2.34) δ(∂xi u) = δ(∂xi u) − (∂xi ∂xk u)δxk
k=0
The equations above follow intuitively due to similarities with the chain rule,
but can be proved rigorously through some manipulation of the terms; for example,
∆u = u∗ (x) − u(x) = (u∗ (x) − u∗ (x∗ )) + (u∗ (x∗ ) − u(x)). Expanding the first term
around x, using (2.27) for the second term, and getting rid of negligible resulting
terms, we arrive at (2.32).
6 THOMAS YU
Using our expressions in (2.26), (2.27), (2.33), and (2.34), we can now write
(2.31) as
(2.35)
Z X n n n
X X ∂L
δJ = (∂xi L)δxi + (∂u L)δu + (∂u L) (∂xi u)δxi + ( ∂u )(∂xi δu)
R i=0 i=0 i=0
∂ ∂xi
n n
∂L
∂xi (δxi )d→
−
X X
+ ( ∂u
)(∂x k
∂ xi
u)δxk + L x
i,k=0
∂ ∂x i i=0
Z n
∂L
))ψd→
−
X
δJ = (∂u L − ∂xi ( ∂u
x
R i=0
∂ ∂xi
n
Z X
∂L
(2.36) + ∂xi ( ∂u
)ψ + Lφi )d→
−
x
R i=0 ∂ ∂x i
Pn
Remark 2.37. ∂u L − i=0 ∂xi ( ∂∂L
∂u ) = 0 are called the Euler-Lagrange Equations;
∂xi
in functional problems with fixed endpoints, boundaries, etc. the variation of the
functional reduces to the first term of (2.38). We know that in order for a function
to be an extremal of a functional, the variation must be identically zero for all
admissible incrementP functions. In this case, that would be ψ. Since ψ is arbitrary,
n
it follows that ∂u L− i=0 ∂xi ( ∂∂L →
−
∂u ) = 0 for all x if a function is to be an extremal.
∂xi
The above reasoning is formalized for functions of one variable in the following
lemma; note the boundary conditions on the increment functions:
Lemma 2.38. Let f : [a, b] ⊂ R → R be a k-times continuously differentiable
function. Suppose that for all h ∈ C0k [a, b]
Z b
(2.39) f (x)h(x)dx = 0
a
Z b
(2.41) f (x)h(x)dx = 0
a
(2.43)
m
Z X n n
Z X m
∂L ∂L
))ψj ]d→
− ψ + Lφi )d→
−
X X
δJ = [(∂uj L − ∂xi ( ∂u
x + ∂xi ( ∂uj j
x
R j=1 i=0 ∂ ∂xji R i=0 j=1 ∂ ∂xi
where
n
X
(2.44) ψj = ψj − (∂xi uj )φi
i=0
∂L
∂x L = ∂t
∂ ∂x
∂t
∂L
∂y L = ∂t ∂y
∂ ∂t
∂L
∂z L = ∂t ∂z
∂ ∂t
These equations correspond to
mx00 (t) = −∂x U
my 00 (t) = −∂y U
mz 00 (t) = −∂z U
Since free particles are only under the influence of conservative forces, and con-
servative forces can be written as the negative of the spatial derivative of a scalar
potential energy function, the above equations reduce to
ma = F
Which is just Newton’s second law in vector form.
In the next section, we will derive the equations of motion for a particle in an
electromagnetic field in more detail as a stepping stone to deriving the field equa-
tions.
The integrand of the action is called the Lagrangian of the system, L if the
integration is with respect to t. If the action is an integral over all of space
time(t, x, y, z), then the integrand is called the Lagrangian density, L. In the above
example, J would be the action of the system and T − U would be the Lagrangian.
We note that finding Lagrangians that accurately model physical situations is not
a simple process but rather one of guessing and checking using some reasonable
physical assumptions.
In this case, the trajectory of the system consists of x since we are deriving the
equations of motion for a charged particle in a given electromagentic field. Thus,
we know that x0 , x1 , x2 , x3 are functions of time, or xc0 ;
Let us now fit our general expression for the variation of a functional in (2.43)
to our current situation:
In this case, we only have one independent variable, x0 , and 4 functions of one
variable, which correspond to the uj . Thus (2.43) reduces to
Z t1 3
X ∂L
δJ = ((∂xi L − ∂x0 ( ∂xi ))ψi )dx0
t0 i=0
∂ ∂x0
Z t1 3
X ∂L
(3.7) + ∂x0 ( ψ + Lφ)dx0
∂xi i
t0 i=0
∂ ∂x0
Z t1 3 3
X ∂L X ∂L t1
(3.9) ∂x0 ( ∂u
)ψi + Lφ)dx0 = ( ∂ui
)ψi + Lφ)t0 = 0
t0 i=0 ∂ ∂x0j i=0
∂ ∂x0
by (3.8).
Therefore, we know that all physically accurate trajectories must satisfy
Z t1 3
X ∂L
(3.10) ((∂xi L − ∂x0 ( ∂xi ))ψi )dx0 = 0
t0 i=0
∂ ∂x0
10 THOMAS YU
3
X ∂L
(3.11) ((∂xi L − ∂x0 ( ∂xi )) = 0
i=0
∂ ∂x0
We now substitute our Lagrangian
3
1 X
L = [ m(∂x0 x) · (∂x0 x) + e Ai (∂x0 xi )]
2 i=0
for a particle in an electromagnetic field into (3.11):
1 2
Since the first term of (3.6) only contains the terms 2 m(∂x0 xi ) and we know
Ai = Ai (x) we can see that
3
X
(3.12) ∂xi L = e (∂xi Aj )(∂x0 xj )
j=0
∂L
(3.13) ∂x0 ( ∂
) = [m(∂x20 xi ) + e(∂x0 Ai )]
∂ ∂xxi
0
3
X
m(∂x20 xi ) = e( [(∂xi Aj )(∂x0 xj )] − ∂x0 Ai )
j=0
3
X
(3.14) =e (∂xi Aj − ∂xj Ai )(∂x0 xj )
j=0
P3
where we used that ∂x0 Ai = j=0 (∂xj Ai )(∂x0 xj ), which follows from the chain
rule.
Note that the left hand side of (3.14) is a rough analogue of the net Newtonian
force (F = ma), where a = ∂x20 xi . We say roughly since we are in the realm of
special relativity, and thus must deal with 4 vectors for force and momentum. In any
case, (3.14) represents the equations of motion for a particle in an electromagnetic
field in the ith coordinate; for all four equations, we merely need to sum fromi = 0
toi = 3 as expressed in (3.11). These equations express the Lorentz Force Law,
which consists of a contribution from the electric Field and the magnetic Field,
encoded through A. Thus, we have demonstrated how variational principles can
be used to derive fundamental equations of motions.
However, the particularly important aspect of this derivation was to derive
(3.15) Fij = (∂xi Aj − ∂xj Ai )
Since we sum over i, j = 0 toi, j = 3, it is clear that Fij represents a 4x4
matrix. More importantly, Fij is called the covariant electromagnetic field tensor;
its importance lies in the fact that it allows for a compact, covariant forumlation
of Maxwell’s equations. We will use Fij in order to calculate a Lagrangian density
that will allow us to obtain Maxwell’s equations.
LAGRANGIAN FORMULATION OF THE ELECTROMAGNETIC FIELD 11
We can represent Fij in matrix form using the relations between A, E, and B
expressed in (3.3) and (3.4) as follows:
E1 E2 E3
0 c c c
− E1 0 −B3 B2
(3.16) Fij = c
− E2 B3 0 −B1
c
− Ec3 −B2 B1 0
Derivation of Maxwell’s Equations 3.17. Maxwell’s equations given in differ-
ential form with respect to our variables in x are as follows:
ρ
(3.18a) ∇·E=
0
(3.18b) ∇·B=0
(3.18c) ∇ × E = c(∂x0 B)
→
− ∂x E
(3.18d) ∇ × B = µ0 J + 0
c
→
−
where µ0 and 0 are physical constants, J = (J1 , J2 , J3 ) is called the current
density, and ρ is the volume charge density.
It is easy to see that (3.18b) and (3.18c) are direct consequences of our formu-
lation of E and B in terms of A; indeed, it is precisely these laws that motivated
such a construction.
Using the vector calculus identities that the divergence of a curl and the curl
of a gradient are both identically zero, it is easy to see that (3.3) and (3.4) imply
(3.18b) and (3.18c).
Thus, we need only derive (3.18a) and (3.18d) using variational principles in
order to complete a comphrehensive formulation of electromagnetism through the
calculus of variations.
Definition 3.19. We define the 4 current J
→
−
(3.20) J = (J0 , J1 , J2 , J3 ) = (cρ, J )
Now, we need a suitable Lagrangian density. Since we have our covariant elec-
tromagnetic tensor Fij , let us raise indices to create the contravariant tensor F ij .
Readers not familiar with tensors must accept that this new tensor is given by
0 − Ec1 − Ec2 − Ec3
E1 0 −B3 B2
(3.21) F ij = c
E2
c B3 0 −B1
E3
c −B2 B1 0
We now examine F ij Fij .
E·E
(3.22) F ij Fij = 2( − B · B)
c2
This equality, easily verified from our constructions of the tensors as well as the
definition of inner products for tensors, is a Lorenz invariant; that is, it is a quantity
that is invariant under the Lorenz transformations of special relativity. This quality
is one of the reasons this inner product will be used in the Lagrangian density for
12 THOMAS YU
We now write the Lagrangian density for the electromagnetic field given a certain
charge density and current density.
3
1 ij X
(3.23) L=− F Fij − Ai Ji
4µ0 i=0
The second term of the Lagrangian density is analogous to the potential energy
term of the Lagrangian of the charged particle.
In this case, our trajectory for the system are the possible values or configurations
of A which in turn correspond to the possible configurations of E and B through
their relations. Since charge and current densities are given, we are examining
the four functions A0 , A1 , A2 , A3 , which are functions of x. In addition, we are
integrating over all of spacetime.
Thus, we need to generalize (2.43) to our system of four functions each relying
on 4 variables, where the Ai correspond to the ui . Changing the j index of (2.43) to
match our current relativistic notation, we have that the variation of our functional
is
(3.24)
3
Z X 3 3
Z X 3
X ∂L X ∂L
δJ = ((∂Aj L − ∂xi ( ∂A
))ψj )dx + ∂xi ( ψ + LLφi )dx
∂Aj j
R j=0 i=0 ∂ ∂xij R i=0 j=0 ∂ ∂xi
The second term is now an integral over ∂R, which is the boundary of all of
spacetime. We now invoke a physical principle that the magnitude of the electro-
magnetic field goes to zero sufficiently ”fast” as one moves towards the boundary
of spacetime(infinity) so that the second term of (3.24) is essentially zero
With respect to the first term, we invoke the same arguments as in the derivation
of the electromagnetic tensor so that we may conclude that any physically meaning-
ful configuration of the electromagnetic field must be such that the Euler-Lagrange
equations for this system hold.
Thus, we conclude that
3 3
X X ∂L
(3.26) (∂Aj L − ∂xi ( ∂Aj )) = 0
j=0 i=0 ∂ ∂xi
Calculating F ij Fij explicity in terms of A using our relations for E, B, and A,
we get that:
F ij Fij = 2[(∂x1 A0 − ∂x0 A1 )2 + (∂x2 A0 − ∂x0 A2 )2 + (∂x3 A0 − ∂x0 A3 )2
(3.27) − (∂x2 A3 − ∂x3 A2 )2 − (∂x3 A1 − ∂x1 A3 )2 − (∂x1 A2 − ∂x2 A1 )2 ]
Calculating the j = 0 term yields
LAGRANGIAN FORMULATION OF THE ELECTROMAGNETIC FIELD 13
and
3
X ∂L 1
∂xi ( ∂A0
) = − [0 + ∂x1 (∂x1 A0 − ∂x0 A1 ) + ∂x2 (∂x2 A0 − ∂x0 A2 ) + ∂x3 (∂x3 A0 − ∂x0 A3 )]
i=0
∂ ∂xi µ0
1 Ex Ex Ex
=− (∂x0 (0) + ∂x1 ( 1 ) + ∂x2 ( 2 ) + ∂x3 ( 3 )
µ0 c c c
1
= − (∂x0 (F 00 ) + ∂x1 (F 10 ) + ∂x2 (F 20 ) + ∂x3 (F 30 )
µ0
(3.29)
3
1 X
=− ∂xk F k0
µ0
k=0
3
X ∂L 1
∂xi ( ∂A1
) = − [−∂x0 (∂x1 A0 − ∂x0 A1 ) + ∂x1 0 + ∂x2 (∂x1 A2 − ∂x2 A1 ) − ∂x3 (∂x3 A1 − ∂x1 A3 )]
i=0
∂ ∂xi µ0
1
=− [∂x (∂x A1 − ∂x1 A0 ) + ∂x1 0 + ∂x2 (∂x1 A2 − ∂x2 A1 ) + ∂x3 (∂x1 A3 − ∂x3 A1 )]
µ0 0 0
1 −Ex1
= − [∂x0 ( ) + ∂x1 0 + ∂x2 (Bx3 ) + ∂x3 (−Bx2 )]
µ0 c
1
= − [∂x0 (F 01 ) + ∂x1 (F 11 + ∂x2 (F 21 ) + ∂x3 (F 31 )]
µ0
(3.32)
3
1 X
=− ∂xk F k1
µ0
k=0
3
X
(3.33) ∂xk F k1 = µ0 J1
k=0
14 THOMAS YU
It is now easy to see that we can generalize the entire set of Euler-Lagrange
equations to
3
X
(3.34) ∂xk F ki = µ0 Ji
k=0
for i = 0, 1, 2, 3.
We only need one more physical fact before we derive (3.18a) and (3.18d), com-
pleting our derivation of Maxwell’s Equations from variational principles. The
physical constants 0 , µ0 , and c are related in the following way:
1
(3.35) c= √
0 µ0
Let us now derive (3.18a). From the definition of the 4 current J, J0 = cρ.
3
X Ex1 Ex Ex
∂xk F k0 = ∂x0 (0) + ∂x1 ( )) + ∂x2 ( 2 ) + ∂x3 ( 1 )
c c c
k=0
∇·E
(3.36) =
c
Thus, it follows from (3.35) and (3.36) that
∇ · E = c2 µ0 ρ
µ0
= ρ
0 µ0
ρ
(3.37) =
0
just as in (3.18a).
(3.18d) comes from examining the i = 1, 2, 3 terms of (3.34):
Ex1
∂x0 (− )) + ∂x1 (0) + ∂x2 (Bx3 ) + ∂x3 (−Bx2 ) = µ0 J1
c
Ex
∂x0 (− 2 )) + ∂x1 (−Bx3 ) + ∂x2 (0) + ∂x3 (Bx1 ) = µ0 J2
c
Ex3
∂x0 (− )) + ∂x1 (Bx2 ) + ∂x2 (−Bx1 ) + ∂x3 (Bx1 ) = µ0 J3
c
Shifting the E terms to the right side, we obtain
E x1
[∂x2 (Bx3 ) − ∂x3 (Bx2 )] = ∂x0 ()) + µ0 J1
c
Ex
[∂x3 (Bx1 ) − ∂x1 (Bx3 )] = ∂x0 ( 2 )) + µ0 J2
c
E x3
[∂x1 (Bx2 ) − ∂x2 (Bx1 )] = ∂x0 ( )) + µ0 J3
c
this set of three equations is equivalent to the vector equation
→
− ∂x E
(3.38) ∇ × B = µ0 J + 0
c
which is exactly (3.18d).
LAGRANGIAN FORMULATION OF THE ELECTROMAGNETIC FIELD 15
4. Concluding Remarks
Using the calculus of variations, we developed the necessary mathematical tools
such as the general variation of a relevant functional as well as the Euler Lagrange
equations for a variational construction of modeling physical phenomena. We then
applied these tools specifically to electromagnetism in the context of special relativ-
ity, deriving the equations of motion in an electromagnetic field as well as Maxwell’s
Equations.
References
[1] L.D. Landau, E.M. Lifschitz, Mechanics. Butterworth-Heinemann, Oxford, 3rd Edition, 1993.
[2] L.D. Landau, E.M. Lifschitz, The Classical Theory of Fields. Butterworth-Heinemann, Oxford,
3rd Edition, 1993.
[3] Herbert Goldstein, Classical Mechanics. Addison-Wesley Press, MA, 1st Edition, 1950.
[4] John C. Baez, Lectures on Classical Mechanics. 2005.
[5] I.M. Gel’fand, S.V. Fomin, Calculus of Variations. Prentice-Hall, NJ, 1963