CLASSICAL FIELD THEORY

Action and Lagrangian in Classical Mechanics

Before we study quantum field theory we need to learn about the classical fields and
the equations governing their dynamics. This dynamics is usually encoded in the action
functional, which generalizes a rather formal approach to classical mechanics. So let us start
with a quick review of the Lagrangian, the action, and the least action principle in classical
mechanics.

For simplicity, let’s start with a single dynamical variable q(t), for example a position of
a particle moving in one dimension, straight or curved. Consider any smooth “trajectory”
q(t) of this variable between times t1 and t2 — regardless of whether that trajectory obeys
some equations of motion or is completely random like a drunkard’s walk — and define the
action integral
Zt2
S[q(t)] = L(q, q̇) dt. (1)
t1

In this formula, L(q, q̇) is the Lagrangian, for example

m 2
L(q, q̇) = Ekin (q̇) − Epot (q) = q̇ − V (q). (2)
2

In the context of the integral (1), the Lagrangian is an ordinary function L of 2 variables
q and q̇ evaluated for q(t) and dq(t)/dt for the specific trajectory q(t) for each time t. On
the other hand, the action S is a functional of the entire trajectory q(t) — i.e., the entire
history of its motion — rather than its value of any particular time.

Now consider the set of all possible trajectories q(t) which begin at the same value
q1 at time t1 and end at the same value q2 at time t2 . The least action principle states
that trajectory which has the lowest action integral S[q(t)] among this set is the classical
trajectory — it happens to obey the classical equations of motion such as Newton Laws.

To see how this works, take any test trajectory q̄(t) and consider its infinitesimal varia-
tions q(t) = q̄(t) + δq(t). If the trajectory q̄(t) happens to minimize the action functional —

1
or even if its any other kind of it’s stationary point — then the first variation of the action
must vanish, δS = 0. So let’s calculate this

Zt2
δS = δL dt. (3)
t1

Obviously
∂L ∂L d
δL@(time t) = × δq(t) + × δq(t) (4)
∂q ∂ q̇ dt

where the derivatives of L are evaluated for q = q̄(t) and q̇ = dq̄(t)/dt. Consequently

Zt2  
∂L ∂L d
δS = × δq(t) + × δq(t) dt
∂q ∂ q̇ dt
t1
hh integrating the second term by parts ii (5)
Zt2   t2
∂L d ∂L ∂L
= δq(t) × − + δq × .
∂q dt ∂ q̇ ∂ q̇ t1
t1

Moreover, due to the initial and final conditions — fixed q(t1 ) = q1 and q(t2 ) = q2 — we
must have δq(t1 ) = δq(t2 ) = 0, which kills the boundary term in eq. (5). Thus, we are left
with
Z  
∂L d ∂L
δS[q] = δq × − dt. (6)
∂q dt ∂ q̇

Now, we want this variation to vanish for all δq(t), and the only way this can happen if the
expression inside (· · ·) vanishes for all times t. In this way, the least action principle leads
to the Euler–Lagrange equation of motion

∂L d ∂L
− = 0 (7)
∂q dt ∂ q̇

where the derivatives of L are evaluated for the would-be classical trajectory q̄(t). For

2
example, for the Lagrangian L = 21 mq̇ 2 − V (q), we have

∂L dV ∂L
= − = +F (q) while = mq̇, (8)
∂q dq ∂ q̇

so the Euler–Lagrange equation (7) becomes

d
F (q) − mq̇) = 0, (9)
dt

or equivalently the Second Law of Newton

dV
mq̈ = F (q) = − . (10)
dq

? ? ?

For the multiple dynamical variables q1 (t), . . . , qN (t) — for example, the 3D positions of
N/3 particles — the Lagrangian L(q1 , . . . , qN ; q̇1 , . . . q̇N ) is a function of 2N arguments — N
“positions” qi and N “velocities” q̇i . For N/3 particles with purely potential forces between
them we have
X mi
L = × q̇i2 − V (q1 , . . . , qn ) (11)
2
i

while more generally L may entangle all 2N of its arguments in a complicated fashion. In
any case, the action
Z
S[all qi (t)] = L(q1 , . . . , qN ; q̇1 , . . . q̇N ) dt (12)

is a functional of the combined history of all N dynamical variables at all times (in some
interval from t1 to t2 ). Similar to what we had for a single variable, the action functional is
defined for all kinds of variables’ trajectories regardless of any equations they might obey or
disobey. But the least action principle says that the classical trajectories — which obey the
Euler–Lagrange equations of motion — yield the lowest action among all other trajectories
which begin and end at the same values of the qi .

3
 The Euler–Lagrange equations for the multiple variables obtain from demanding zero
first variation of the action with respect to the independent infinitesimal variations δqi (t) of
all N variables. Since this exercise is very similar to what we have done for a single variable,
let me skip the intermediate steps and write down the end result: For each variable qi and
for every moment t, we must have

d ∂L ∂L
− = 0 (∀i = 1, . . . N ). (13)
dt ∂ q̇i ∂qi

For example, for the Lagrangian (11), the Euler Lagrange equations of motion become

d ∂V
(mi q̇i ) + = 0, (14)
dt ∂qi

or equivalently Newton equations for multiple bodies

∂V
mi q̈i = − = +Fi (q1 , . . . , qN ). (15)
∂qi

? ? ?

Normally, the Lagrangian L involves the dynamical variables qi themselves and their first
time derivatives q̇i but no higher derivatives; consequently, the Euler–Lagrange equations of
motion are second-order differential equations. If we include higher time derivatives into the
Lagrangian, the equation of motion would also be higher-order in d/dt. For example, let
L(q, q̇, q̈) involve the second time derivative; then the Euler–Lagrange equation becomes

d2 ∂L d ∂L ∂L
2
− + = 0, (16)
dt ∂ q̈ dt ∂ q̇ ∂q

which is of the 4th order in d/dt (assuming ∂L/∂ q̈ depends on the q̈). Mathematically, this
equation is equivalent to two second-order equations for two interacting variables q1 (t) and
q2 (t) (for example, q1 = q and q2 = q̈), and to pick a specific solution we need to fix 4 initial
...
conditions q(0), q̇(0), q̈(0), and q (0), or equivalently two variables q1 (0), q2 (0) and their first
derivatives q̇1 (0), q̇2 (0). Physically, this corresponds to two independent (but interacting)
degrees of freedom q1 (t) and q2 (t) rather than a single q(t).

4
 More generally, a dynamical degree of freedom is an independent variable obeying a
second-order (in d/dt) equation of motion which takes two initial conditions (the variable
itself and its velocity) to fix a specific solution. Thus, a variable obeying an equation of
motion of degree 2n describes n dynamical degrees of freedom rather than one, and it would
be more transparent to replace it with n variables obeying second-order equations. On the
other hand, a pair of first order equations for a pair of variables — such the coordinate and
the momentum — describes a single degree of freedom rather than two.

In the Lagrangian terms, if L depends on higher derivatives of some variable q, that

variable involves multiple degrees of freedom. If the highest derivative entering L is q (n) —
and if L is a non-linear function of q (n) — then q(t) describes n degrees of freedom. For
transparency’s sake, it is best to replace such q(t) with n variables which enter L with their
first derivatives only. On the other hand, if a variable enters the Lagrangian without any
time derivatives at all — i.e., L depends on a qi but not on q̇i , etc., — then the qi is not a
dynamical variable at all but rather a Lagrange multiplier enforcing some time-independent
constraint on the other variables. Finally, if L is linear on a time derivative q̇i — i.e., L
depends on the q̇i but ∂L/∂ q̇i does not depend on the q̇i itself or other velocities, — then
the qi (t) obeys a first-order equation and describes 1/2 of a degree of freedom; obviously,
there should be an even number of such variables.

Classical Fields and Lagrangian Densities

A classical field is a continuous family of dynamical variables Φx (t), one variable for each
space point x = (x, y, z); mathematically, this family is a function Φ(x, t) of all 4 spacetime
coordinates (x, y, z; t). Note that the space coordinates (x, y, z) here are not dynamical
variables as they do not describe the position of any physical body. Instead, it’s the field
which comprises dynamical variables Φx (t) at every point in space, and the coordinates
(x, y, z) = x merely label these variables.

Most generally, the Lagrangian L[Φ(x), Φ̇(x)] for a classical field Φ is a functional of the
entire space configuration of the field Φ(x) and its time derivative Φ̇(x) = ∂Φ(x, t)/∂t at
any particular time. However, in field theories which do not allow interactions at a distance

5
— such as all relativistic theories, — the Lagrangian L is a space integral

Z
L = d3 x L(Φ, ∇Φ, Φ̇) (17)

of the Lagrangian density L which is an ordinary local function of the field Φ and its space
and time derivatives ∇Φ and Φ̇. Consequently, the action is a 4D spacetime integral of the
Lagrangian density
Z Z
S[Φ(x, t)] = dt d3 x L(Φ, ∇Φ, Φ̇). (18)

Similarly to mechanics, the equations of motion for a classical field obtain from the
least action principle: Among all the time evolutions of a field with given initial and final
values (for all x), the evolution which obeys the field equations has the least action (18).
Thus, the classical field equations themselves obtain from infinitesimal variations of some
field configuration Φ(x, t) 7→ Φ(x, t) + δΦ(x, t) and demand δS = 0 for all δΦ(x, t). Under
such infinitesimal variations

∂L ∂L ∂δΦ ∂L
δL@(x, t) = × δΦ + × + · ∇δΦ (19)
∂Φ ∂ Φ̇ ∂t ∂(∇Φ)

˙
where the derivatives of L are taken for the Φ = Φ(x, t), Φ̇ = Φ(x, t), and ∇Φ = ∇Φ(x, t),
hence
Z Z  
3 ∂L ∂L ∂δΦ ∂L
δS = dt d x δΦ + + · ∇δΦ
∂Φ ∂ Φ̇ ∂t ∂(∇Φ)
hh integrating by parts ii
Z Z     
3 ∂L ∂ ∂L ∂L
= dt d x δΦ(x, t) − − ∇· @(x, t) (20)
∂Φ ∂t ∂ Φ̇ ∂(∇Φ)
Z   t2 Z ZZ
3 ∂L ∂L
+ d x δΦ + dt d2 Area · δΦ .
∂ Φ̇ t1 ∂(∇Φ)
boundary
of 3D space

On the last line here, the time boundary term vanishes because the δΦ(x, t) variation must
vanish at the initial and the final times t = t1 and t = t2 . As to the space boundary term, we

6
may eliminate it by assuming that for |x| → ∞, δΦ(x, t) → 0 fast enough that the surface
integral asymptotes to zero. Consequently, we are left with the
Z Z     
3 ∂L ∂ ∂L ∂L
δS = dt d x δΦ(x, t) − − ∇· @(x, t), (21)
∂Φ ∂t ∂ Φ̇ ∂(∇Φ)

which must vanish for any δΦ(x, t) which obeys the boundary conditions. Clearly, this
requires the expression inside the big (· · ·) to vanish at all x and t, hence the Euler–Lagrange
field equations
   
∂ ∂L ∂L ∂L
+ ∇· − = 0. (22)
∂t ∂ Φ̇ ∂(∇Φ) ∂Φ

For example, consider the free relativistic scalar field Φ with the Lagrangian

1 2 1 2 κ2 2
L = Φ̇ − (∇Φ) − Φ . (23)
2c2 2 2
For this Lagrangian

∂L 1 ∂L ∂L
= 2 Φ̇, = −∇Φ, = −κ2 Φ, (24)
∂ Φ̇ c ∂(∇Φ) ∂Φ

so the Euler–Lagrange equation (22) becomes

 
∂ 1 ∂Φ
−κ2 Φ ,



+ ∇ · −∇Φ − (25)
∂t c2 ∂t

which is the Klein–Gordon equation

1 ∂2
 
2 2
− ∇ + κ Φ(x, t) = 0 (26)
c2 ∂t2

for a free relativistic field of mass m = (h̄/c)κ.

Most field theories have multiple fields — i.e., multiple dynamical variables φ1 (x, t), . . . ,
φN (x, t) at each space point x. Such multiple fields can be different species of scalar fields, or
different components of a vector, tensor, etc., field, or any combination thereof: for example,
3 components of one vector field, 3 more components of another vector field, plus 5 scalar
fields for N = 11 fields altogether. For the moment, let me ignore the difference between the
components and the species and simply focus on having N independent φi (x, t) dynamical
variables at each space point.

7
 Assuming local interactions between the fields, the action has form
Z Z
S[φi (x, t) for all i at all x and all t] = dt d3 x L(φ1 , . . . , φN ; ∇φ1 , . . . , ∇φN ; φ̇1 , . . . , φ̇N ),
(27)
where the Lagrangian density L is a function of 5N variables: N fields φi themselves, N time
derivatives φ̇i , and 3N components of N space gradients ∇φi . Minimizing the action (27)
gives rise to N coupled Euler–Lagrange equations, one for each field:
   
∂ ∂L ∂L ∂L
∀i = 1, . . . , N : + ∇· − = 0. (28)
∂t ∂ φ̇i ∂(∇φi ) ∂φi

For example, for N relativistic scalar fields with Lagrangian density

N  
X 1 2 1 2
L = (φ̇i ) − (∇φi ) − V (φ1 , . . . , φN ) (29)
2c2 2
i=1

we have
∂L 1 ∂L ∂L ∂V
= 2 φ̇i , = −∇φi , = − , (30)
∂ φ̇i c ∂(∇φi ) ∂φi ∂φi

hence the Euler–Lagrange field equations (28) become

1 ∂2 2 ∂V
∀i = 1, . . . , N : φi − ∇ φi + = 0. (31)
c2 ∂t2 ∂φi

In general ∂V /∂φi depends on both the φi itself and on the other fields φj6=i , so eqs. (31) are
not independent but coupled to each other.

For another example, consider a non-relativistic complex (i.e., complex number valued)
scalar field ψ(x, t) describing the superfluid component of liquid helium II. Specifically, its
magnitude describes the superfluid density while the gradient of its phase describes the
superfluid’s velocity according to

h̄
ρ(x, t) = M |ψ(x, t)|2 ,


v(x, t) = ∇ phase ψ(x, t) (32)
M

where M is the mass of a helium atom. In the Landau–Ginzburg theory, of superfluidity,

8
the field equations for ψ(x, t) obtain from a local Lagrangian density

h̄2 λ
L = −h̄ Im(ψ ∗ ψ̇) − ∇ψ ∗ · ∇ψ + µ|ψ|2 − |ψ|4 . (33)
2M 2

A complex field ψ(x, t) is equivalent to two real fields Re ψ(x, t) and Im ψ(x, t), so we should
be getting two coupled field equations by independently varying δ Re ψ(x, t) and δ Im ψ(x, t).
However, it is easier to derive the field equation(s) in a manifestly complex form by treating
the δψ(x, t) and δψ ∗ (x, t) as independent variations, hence Euler–Lagrange equations
   
∂ ∂L ∂L ∂L
+ ∇· − = 0
∂t ∂ ψ̇ ∂(∇ψ) ∂ψ
and (34)
   
∂ ∂L ∂L ∂L
+ ∇· − = 0.
∂t ∂ ψ̇ ∗ ∂(∇ψ ∗ ) ∂ψ ∗

For the Landau–Ginzburg Lagrangian density (33)

− Im(ψ ∗ ψ̇) = i ∗
2ψ × ψ̇ − i
2ψ × ψ̇ ∗ , (35)

hence

∂L ih̄ ∗
∂L ih̄
= ψ , = − ψ, (36)
∂ ψ̇ 2 ∂ ψ̇ ∗ 2
2
∂L h̄ ∂L h̄2
= − ∇ψ ∗ , = − ∇ψ, (37)
∂(∇ψ) 2M ∂(∇ψ ∗ ) 2M
∂L ih̄
= − ψ̇ ∗ + (µ − λ|ψ|2 )ψ ∗ , (38)
∂ψ 2
∂L ih̄
= + ψ̇ + (µ − λ|ψ|2 )ψ, (39)
∂ψ ∗ 2

so eqs. (34) become

ih̄ ∗ h̄2 2 ∗ ih̄ ∗

ψ̇ − ∇ ψ + ψ̇ − (µ − λ|ψ|2 )ψ ∗ = 0, (40)
2 2M 2
ih̄ h̄2 2 ih̄
− ψ̇ − ∇ ψ − ψ̇ − (µ − λ|ψ|2 )ψ = 0, (41)
2 2M 2

9
or equivalently, a Schrödinger-like equation

∂ψ h̄2 2
ih̄ = − ∇ ψ + (λ|ψ|2 − µ)ψ (42)
∂t 2M

?
for the ψ field, and a complex-conjugate equation

∂ψ ∗ h̄2 2 ∗
−ih̄ = − ∇ ψ + (λ|ψ|2 − µ)ψ ∗ (43)
∂t 2M

for the conjugate field ψ ∗ . Note that differential eqs. (42) and (43) are first-order in the
time derivative (albeit second-order in space derivatives), so together ψ(x, t) and ψ ∗ (x, t) —
1
or equivalently the two real fields Re ψ(x, t) and Im ψ(x, t) — describe 2 × 2 = 1 degree of
freedom (at each x) rather than 2 degrees of freedom one naively expects from a complex
field.

In non-relativistic theories, the fields’ equations of motion often have more space deriva-
tives then time derivatives. Consequently, the Lagrangian density function may depend on
higher space derivatives of the fields, thus L(φ, ∇φ, ∇∇φ, . . . ; φ̇) or even
L(φ, ∇φ, ∇∇φ, . . . ; φ̇, ∇φ̇, . . .); all such Lagrangian density functions are OK as long as they
do not involve higher time derivatives. The general Euler–Lagrange equation for such a
higher-derivative Lagrangian density is

   
∂ ∂L ∂L ∂L ∂L ∂L
− ∇· + ··· − − ∇· + ∇i ∇j − ··· = 0.
∂t ∂ φ̇ ∂(∇φ̇) ∂φ ∂(∇φ) ∂(∇i ∇j φ)
(44)
Moreover, the Lagrangian L maybe non-local in space and involve multiple space integrals
rather than a single integral of the Lagrangian density L. For example, an improvement
of the Landau–Ginzburg theory which accounts for a finite-range two-body forces between

? Despite similarity, eq. (42) is not a true Schrödinger equation for a wave function. In particular, eq. (42)
has a non-linear term λ|ψ|2 × ψ, while a true Schrödinger equation is always linear, come hell or high
water. Also, the ψ(x, t) here is a classical field rather than a wave function of anything; indeed, a
true wave function of the superfluid would depend on a position of every helium atom in a superfluid
rather than on a single position x. Nevertheless, some old textbooks miscall ψ(x, t) a “macroscopic
wave function”; this is a very misleading terminology, so please be careful.

10
helium atoms has
!
Z 2
h̄
L = d3 x −h̄ Im(ψ ∗ ψ̇) − ∇ψ ∗ · ∇ψ + µ|ψ|2
2M
Z Z (45)
1
− d3 x d3 y V2 (x − y) × |ψ(x)|2 × |ψ(y)|2 .
2

where V2 (x − y) is the interaction potential for 2 helium atoms, one at x the other at y. The
field equations stemming from this Lagrangian are differential in time but integro-differential
in space, namely

h̄2 2
Z
∂
ih̄ ψ(x, t) = − ∇ ψ(x, t) − µψ(x, t) + ψ(x, t) × d3 y V2 (x − y) |ψ(y, t)|2 , (46)
∂t 2M

and the complex-conjugate equation for the ψ ∗ (x, t).

We shall return to such non-local-in-space classical (and quantum) field theories in 3–4
weeks. But for the next few lectures we shall focus on the relativistic field theories, which
are always local.

Relativistic Fields: Conventions and Notations

Following the Peskin & Schroeder textbook, I shall use Greek letters — especially the
mid-alphabet letters λ, µ, ν — for the 4D Lorentz indices and make them run from 0 to 3: 0
for time and 1,2,3 for space. I shall use the (+, −, −, −) signature for the metric,

 
+1 0 0 0
 0 −1 0 0 
 
gµν =   (47)
 0
 0 −1 0 

0 0 0 −1

thus spacetime coordinates xµ = (x0 = ct; x1 , x2 , x3 ) but xµ = (+ct, −x1 , −x2 , −x3 ). Like-
wise, for most 4–vectors Aµ with an upper index, the space components (A1 , A2 , A3 ) form
a 3–vector with the usual sign, while for a 4–vector Aµ with a lower index, the µ = 1, 2, 3
components have an opposite sign. Only the time component A0 = A0 has the usual sign

11
regardless of whether the index µ = 0 is upstairs or downstairs. For example, for the energy-
momentum 4–vector, P µ = (E/c; +p) while Pµ = (+E/c; −p). Likewise, for the electric
current 4–vector J µ = (cρ; +J) while Jµ = (+cρ; −J); for the electromagnetic 4–vector po-
tential Aµ = (Φ; +A) while Aµ = (+Φ, −A), etc., etc. The only exception from this rule is
the derivative 4-vector
   
∂ 1∂ µ ∂ 1∂
∂µ = = ; +∇ while ∂ = = ; −∇ . (48)
∂xµ c ∂t ∂xµ c ∂t

The reason for this exception is to have ∂µ xν = δµν and ∂ µ xν = δνµ without any extra signs.
On the other hand, ∂µ xν = gµν and ∂ µ xν = g µν involve the metric tensor.

Throughout this class, I shall use Einstein’s summation convention: When the same
Lorentz index appears twice within a product of 4–vectors or 4–tensors, once upstairs and
once downstairs, then such index is implicitly summed over. For example,
X
Aµ Bµ = Aµ Bµ = A0 B0 + A1 B1 + A2 B2 + A3 B3
µ=0,1,2,3
(49)
= A0 B 0 − A1 B 1 − A2 B 2 − A3 B 3
= A0 B 0 − A · B.
On the other hand, if the same index appears twice upstairs, or twice downstairs, or more
then twice, then this is probably a mistake of some kind. In very rare cases when this is not
a mistake, one should explicitly say whether such index should be summed over or not, but
either way the expression would break the Lorentz symmetry.

In relativistic notations, a classical field is φ(x) where x denotes the 4-vector xµ , its
space and time derivatives ∇φ and φ̇ combine into a 4–vector ∂µ φ(x), thus the Lagrangian
density L(φ, ∂µ φ), for example for a free relativistic scalar field

1 (mc/h̄)2 2
L = (∂µ φ)(∂ µ φ) − φ . (50)
2 2
The action of a relativistic field theory is a 4D spacetime integral of the Lagrangian density,
Z Z Z
3 1
S[φ(x)] = dt d x L = d4 x L. (51)
c
The 4D volume element d4 x is invariant under Lorentz transformations, so a Lorentz-
invariant action calls for a Lorentz-invariant (or rather Lorentz scalar) Lagrangian density L

12
such as (50). We shall see other examples of Lorentz-scalar Lagrangian densities next lecture
and in the homework set#1.

The relativistic form of the Euler–Lagrange field equation is

 
∂L ∂L
∂µ − = 0. (52)
∂(∂µ Φ) ∂Φ

For example, for the Lagrangian density (50),

∂L ∂L
= −(mc/h̄)2 φ while = +∂ µ Φ, (53)
∂Φ ∂(∂µ Φ)

— note how the index µ moves upstairs from downstairs, — hence eq. (52) becomes

∂µ ∂ µ Φ) + (mc/h̄)2 Φ = 0, (54)

or in a more compact form

[∂ 2 + (mc/h̄)2 ]Φ(x) = 0 (55)

where ∂ 2 = ∂µ ∂ µ is the Lorentz square of the derivative operator, also known as the
D’Alembert operator

2 1 ∂2
∂ = = 2 2 − ∇2 . (56)
c ∂t

For the multiple fields (or field components) φ1 (x), . . . , φN (x), the Lagrangian density L is
a function of all the φi and their first derivative 4–vectors ∂µ φi , thus

Z
1
S = d4 x L(φ1 , . . . , φN ; ∂µ φ1 , . . . , ∂µ φN ), (57)
c

and minimizing this action leads to N coupled Euler–Lagrange field equations

 
∂L ∂L
∀i = 1, . . . , N : ∂µ − = 0. (58)
∂(∂µ Φi ) ∂Φi

13
For example, for N scalar fields with Lagrangian density

N
X 1
L = (∂µ Φi )(∂ µ Φi ) − V (Φ1 , . . . , ΦN ), (59)
2
i=1

the field equations are

∂V
∀i = 1, . . . , N : ∂ 2 Φi + = 0. (60)
∂Φi

More interesting examples of multiple components of a vector or tensor field will be discussed
in the next set of my notes and in the homework set#1.

14