Académique Documents
Professionnel Documents
Culture Documents
Field Theory
Sanjeev S. Seahra
Department of Physics
University of Waterloo
May 11, 2000
Abstract
We discuss the path integral formulation of quantum mechanics and use it to derive
the S matrix in terms of Feynman diagrams. We generalize to quantum field theory,
and derive the generating functional Z[J] and n-point correlation functions for free
scalar field theory. We develop the generating functional for self-interacting fields
and discuss 4 and 3 theory.
Introduction
Thirty-one years ago, Dick Feynman told me about his sum over histories version of quantum mechanics. The electron does anything it likes,
he said. It goes in any direction at any speed, forward and backward in
time, however it likes, and then you add up the amplitudes and it gives
you the wavefunction. I said to him, Youre crazy. But he wasnt.
F.J. Dyson1
When we write down Feynman diagrams in quantum field theory, we proceed with
the mind-set that our system will take on every configuration imaginable in traveling
from the initial to final state. Photons will split in to electrons that recombine
into different photons, leptons and anti-leptons will annihilate one another and the
resulting energy will be used to create leptons of a different flavour; anything that
can happen, will happen. Each distinct history can be thought of as a path through
the configuration space that describes the state of the system at any given time.
For quantum field theory, the configuration space is a Fock space where each vector
represents the number of each type of particle with momentum k. The key to
the whole thing, though, is that each path that the system takes comes with a
probabilistic amplitude. The probability that a system in some initial state will end
up in some final state is given as a sum over the amplitudes associated with each path
connecting the initial and final positions in the Fock space. Hence the perturbative
expansion of scattering amplitudes in terms of Feynman diagrams, which represent
all the possible ways the system can behave.
But quantum field theory is rooted in ordinary quantum mechanics; the essential difference is just the number of degrees of freedom. So what is the analogue of
this sum over histories in ordinary quantum mechanics? The answer comes from
the path integral formulation of quantum mechanics, where the amplitude that a
particle at a given point in ordinary space will be found at some other point in the
future is a sum over the amplitudes associated with all possible trajectories joining
the initial and final positions. The amplitude
R associated with any given path is
just eiS , where S is the classical action S = L(q, q)
dt. We will derive this result
from the canonical formulation of quantum mechanics, using, for example, the timedependent Schrodinger equation. However, if one defines the amplitude associated
with a given trajectory as eiS , then it is possible to derive the Schr
odinger equation2 .
We can even derive the classical principle of least action from the quantum amplitude eiS . In other words, one can view the amplitude of traveling from one point
to another, usually called the propagator, as the fundamental object in quantum
theory, from which the wavefunction follows. However, this formalism is of little
1
2
To motivate our use of the path integral formalism in quantum field theory, we
demonstrate how path integrals arise in ordinary quantum mechanics. Our work
is based on section 5.1 of Ryder [1] and chapter 3 of Baym [2]. We consider a
quantum system represented by the Heisenberg state vector |i with one coordinate
degree of freedom q and its conjugate momentum p. We adopt the notation that
the Schrodinger representation of any given state vector |i is given by
|, ti = eiHt |i,
(1)
where H = H(q, p) is the system Hamiltonian. According to the probability interpretation of quantum mechanics, the wavefunction (q, t) is the projection of |, ti
onto an eigenstate of position |qi. Hence
(q, t) = hq|, ti = hq, t|i,
(2)
(3)
(4)
Z
hq|i =
or
(5)
dq 0 |q 0 ihq 0 |.
(6)
Z
1=
q
t
(qf ,tf )
(q i ,ti )
t = t1
Figure 1: The various two-legged paths that are considered in the calculation of
hqf , tf |qi , ti i
0
Multiplying by eiHt on the left and eiHt on the right yields that
Z
1 = dq 0 |q 0 , t0 ihq 0 , t0 |.
(7)
(8)
The quantity hqf , tf |qi , ti i is called the propagator and it represents the probability
amplitudes (expansion coefficients) associated with the decomposition of (qf , tf )
in terms of (qi , ti ). If (qi , ti ) has the form of a spatial delta function (q0 ), then
(qf , tf ) = hqf , tf |q0 , ti i. That is, if we know that the particle is at q0 at some time
ti , then the probability that it will be later found at a position qf at a time tf is
P (qf , tf ; q0 , ti ) = |hqf , tf |qi , t0 i|2 .
(9)
It is for this reason that we sometimes call the propagator a correlation function.
Now, using completeness, it is easily seen that the propagator obeys a composition equation:
Z
hqf , tf |qi , ti i =
(10)
This can be understood by saying that the probability amplitude that the position
of the particle is qi at time ti and qf at time tf is equal to the sum over q1 of the
probability that the particle traveled from qi to q1 (at time t1 ) and then on to qf .
In other words, the probability amplitude that a particle initially at qi will later
be seen at qf is the sum of the probability amplitudes associated with all possible
3
A
2
1
(11)
The presence of the double-slit ensures that the integral in (10) reduces to the twopart sum in (11). When the probability |h2|1i|2 is calculated, interference between
the h2|AihA|1i and h2|BihB|1i terms will create the classic intensity pattern on the
screen.
There is no reason to stop at two-legged paths. We can just as easily separate
the time between ti and tf into n equal segments of duration = (tf ti )/n. It
then makes sense to relabel t0 = ti and tn = tf . The propagator can be written as
Z
hqn , tn |q0 , t0 i = dq1 dqn1 hqn , tn |qn1 , tn1 i hq1 , t1 |q0 , t0 i.
(12)
We take the limit n to obtain an expression for the propagator as a sum over
infinite-legged paths, as seen in figure 3. We can calculate the propagator for small
time intervals = tj+1 tj for some j between 1 and n 1. We have
hqj+1 , tj+1 |qj , tj i = hqj+1 |eiHtj+1 e+iHtj |qj i
4
t0
t1
t2
t3
t4
t5
qf
qi
n
OO
qi
qf
Figure 3: The continuous limit of a collection of paths with a finite number of legs
= hqj+1 |(1 iH + O( 2 )|qj i
= (qj+1 qj ) i hqj+1 |H|qj i
Z
1
i
=
dp eip(qj+1 qj )
hqj+1 |p2 |qj i
2
2m
i hqj+1 |V (q)|qj i,
(13)
p2
+ V (q).
2m
(14)
Z
2
hqj+1 |p |qj i =
(15)
1
hq|pi = eipq ,
2
hp|p0 i = (p p0 ).
(16)
(17)
where we should point out that p2 is a number, not an operator. Working on the
other matrix element in (13), we get
hqj+1 |V (q)|qj i = hqj+1 |qj iV (qj )
= (qj+1 qj )V (qj )
Z
1
=
dp eip(qj+1 qj ) V (qj ).
2
1
hqj+1 , tj+1 |qj , tj i =
dp eip(qj+1 qj ) 1 i H(p, qj ) + O( 2 )
2
Z
qj
1
dp exp i p
H(p, qj ) ,
=
2
Z
n1
n1
Y dqi dpi
X qj
hqn , tn |q0 , t0 i = dp0
exp i
pj
H(pj , qj ) .
2
i=1
(18)
j=0
tn
dt,
t0
and
qj
dq
= q,
dt
Z
hqn , tn |q0 , t0 i =
dp0
n
Y
dqi dpi
i=1
Z
[dq] [dp] exp i
tn
[dq] [dp],
dt [p q H(p, q)] .
(19)
(20)
t0
The notation [dq] [dp] is used to remind us that we are integrating over all possible
paths q(t) and p(t) that connect the points (q0 , t0 ) and (qn , tn ). Hence, we have
succeed in writing the propagator hqn , tn |q0 , t0 i as a functional integral over the
all the phase space trajectories that the particle can take to get from the initial
to the final points. It is at this point that we fully expect the reader to scratch
their heads and ask: what exactly is a functional integral? The simple answer is a
quantity that arises as a result of the limiting process we have already described.
The more complicated answer is that functional integrals are beasts of a rather
vague mathematical nature, and the arguments as to their standing as well-behaved
entities are rather nebulous. The philosophy adopted here is in the spirit of many
mathematically controversial manipulations found in theoretical physics: we assume
that everything works out alright.
The argument of the exponential in (20) ought to look familiar. We can bring
this out by noting that
( "
#)
h
i
Z
q
1
m qi 2
1
i pi i H(pi ,qi )
dpi e
=
exp i
V (qi )
2
2
2
"
#
Z
mqi 2
i
p
dpi exp
2m
( "
#)
m 1/2
m qi 2
=
exp i
V (qi )
.
2i
2
"
#
2
n1
m n/2 Z n1
Y
X
m qj
hqn , tn |q0 , t0 i =
dqi exp i
V (qj )
2i
2
i=1
j=0
Z tn
Z
1 2
N [dq] exp i
mq V (q) ,
(21)
dt
2
t0
where the limit is taken, as usual, for n and 0. Here, N is an infinite
constant given by
m n/2
N = lim
.
(22)
n 2i
We wont worry too much about the fact that N diverges because we will later
normalize our transition amplitudes to be finite. Recognizing the Lagrangian L =
T V in equation (21), we have
Z tn
Z
Z
hqn , tn |q0 , t0 i = N [dq] exp i
L(q, q)
dt = N [dq] eiS[q] ,
(23)
t0
qf
q (t )
q ( t ) + d q ( t)
qi
Figure 4: Neighbouring particle trajectories. If the action evaluated along q(t) is
stationary (i.e. S = 0), then the contribution of q(t) and its neighbouring paths
q(t) + q(t) to the propagator will constructively interfere and reconstruct the classical trajectory in the limit ~ 0
.
satisfying S[q]|q=q0 ; i.e. for paths connected by classical trajectories determined by
Newtons 2nd law. We have hence seen how the classical principle of least action
can be understood in terms of the path integral formulation of quantum mechanics
and a corresponding principle of stationary phase.
tn
exp i
t0
Z
V (x, t) dt = 1 i
tn
t0
1
V (x, t) dt
2!
tn
2
V (x, t) dt + .
(25)
t0
The K0 term is
Z
K0 = N
1
2
[dx] exp i
mx dt .
2
(27)
If we turned off the potential, the full propagator would reduce to K0 . It is for this
reason that we call K0 the free particle propagator, it represents the amplitude that
a free particle known to be at x0 at time t0 will later be found at xn at time tn .
Going back to the discrete expression:
n1
m n/2 Z n1
Y
X
im
dxi exp
(xj+1 xj )2 .
(28)
K0 = lim
n 2i
2
i=1
j=0
m n/2 1 2i (n1)/2
im(xn x0 )2
.
exp
n 2i
m
2n
n1/2
K0 = lim
(29)
m
K0 (xn x0 , tn t0 ) =
2i(tn t0 )
1/2
im(xn x0 )2
,
exp
2(tn t0 )
tn > t0 .
(30)
Here, weve noted that the substitution n = (tn t0 ) is only valid for tn > t0 . In
fact, if K0 is non-zero for tn > t0 it must be zero for t0 > tn . To see this, we note
that the calculation of K0 involved integrations of the form:
Z
Z
1 ix2
2
eix dx =
e
dx
2 0
Z
i1/2 i es
=
ds
4
s1/2
0
Z
i1/2 i
es
(is) 1/2 ds,
=
4
s
i
where sign( ) = sign(tn t0 ). Now, we can either choose the branch of s1/2
to be in either the left- or righthand part of the complex s-plane. But, we need
to complete the contour in the lefthand plane if > 0 and the righthand plane if
< 0. Hence, the integral can only be non-zero for one case of the sign of . The
choice we have implicitly made is the the integral is non-zero for (tn t0 ) > 0,
hence it must vanish for tn < t0 . When we look at equation (8) we see that K0
is little more than a type of kernel for the integral solution of the free-particle
Schrodinger equation, which is really a statement about Huygens principle. Our
9
choice of K0 obeys causality in that the configuration of the field at prior times
determines the form of the field in the present. We have hence found a retarded
propagator. The other choice for the boundary conditions obeyed by K0 yields
the advanced propagator and a version of Huygens principle where future field
configurations determine the present state. The moral of the story is that, if we
choose a propagator that obeys casuality, we are justified in writing
h m i1/2
imx2
exp
.
(31)
K0 (x, t) = (t)
2it
2t
Now, we turn to the calculation of K1 :
Z
Z
Z
1
2
K1 = iN [dx] exp i
mx dt
dt V (x(t), t).
2
(32)
Z
K1 = i
n/2
dx1 dxn1
im
exp
2
n1
X
(xj+1 xj )
j=0
n1
X
V (xi , ti ),
(33)
i=1
where = m/2i and the limit n is understood. Lets take the sum over i
(which has replaced the integral over t) in front of the spatial integrals. Also, lets
split up the sum over j in the exponential to a sum running from 0 to i 1 and a
sum running from i to n 1. Then
Z
i1
n1
X
X Z
im
(xj+1 xj )2 V (xi , ti )
K1 = i
dxi i/2 dx1 dxi1 exp
2
j=0
i=1
Z
n1
X
im
(ni)/2 dxi+1 dxn1 exp
(xj+1 xj )2 .
(34)
2
j=i
n1
X
(35)
i=1
R tn
Pn1
Now, we can replace
i=1 by t0 dt and ti t in the limit n . Since
K0 (x x0 , t t0 ) = 0 for t < t0 and K0 (xn x, tn t) for t > tn , we can extend the
limits on the time integration to . Hence,
Z
K1 = i dx dt K0 (xn x, tn t)V (x, t)K0 (x x0 , t t0 ).
(36)
10
Z
n1
X
(i)2 n/2
im
K2 =
(37)
j=0
n1
X
i=1
V (xi , ti )
n1
X
V (xk , tk ).
(38)
k=1
We would like to play the same trick that we did before by splitting the sum over
j into three parts with the potential terms sandwiched in between. We need to
construct the middle j sum to go from an early time to a late time in order to replace
it with a free-particle propagator. But the problem is, we dont know whether ti
comes before or after tk . To remedy this, we split the sum over k into a sum from
1 to i 1 and then a sum from i to n 1. In each of those sums, we can easily
determine which comes first: ti or tk . Going back to the continuum limit:
Z t1
Z
Z tn
(i)2
dx1 dx2
dt1
dt2 K0 (xn x1 , tn t1 )
K2 =
2!
t0
t0
V (x1 , t1 )K0 (x1 x2 , t1 t2 )V (x2 , t2 )K0 (x2 x0 , t2 t0 )
Z tn
dt2 K0 (xn x2 , tn t2 )V (x2 , t2 )K0 (x2 x1 , t2 t1 )
+
t1
(40)
Higher order contributions to the propagator follow in a similar fashion. The general
j th order correction to the free propagator is
Z
j
Kj = (i)
dx1 . . . dxj dt1 . . . dtj K0 (xn xj , tn tj )
V (xj ) V (x1 )K0 (x1 x0 , t1 t0 ).
(41)
As t , we assume the potential goes to zero, which models the fact that the
particle is far away from the scattering region in the distant past and the distant
future. We go over from one to three dimensions and write
Z
(xf , tf ) =
dxi K0 (xf xi , tf ti )(xi , ti )
Z
i dxi dx dt K0 (xf x, tf t)
V (x, t)K0 (x xi , t ti )(xi , ti ) +
(42)
We push ti into the distant past, where the effects of the potential may be ignored,
and take the particle to be in a plane wave state:
1
in (xi , ti ) = eipi xi ,
V
(43)
where we have used a box normalization with V being the volume of the box and
pi x = Ei ti pi xi . The in label on the wavefunction is meant to emphasize that
it is the form of before the particle moves into the scattering region. We want to
calculate the first integral in (42) using the 3D generalization of (31):
K0 (x, t) = i(t)
3/2
2
ex ,
(44)
3/2
i
V
Z
2
eiEi ti dxi e(xf xi ) +ipi xi .
(45)
This integral reduces to in (xf , tf ) as should have been expected, because K0 is the
free particle propagator and must therefore propagate plane waves into the future
without altering their form. We also push tf into the infinite future where the effects
of the potential can be ignored. Then,
Z
+ (xf , tf ) = in (xf , tf ) i dxi dx dt K0 (xf x, tf t)
V (x, t)K0 (x xi , t ti )in (xi , ti ) +
(46)
The + notation on is there to remind us that + is the form of the wave function
after it interacts with the potential. What we really want to do is Fourier analyze
+ (xf , tf ) into momentum eigenstates to determine the probability amplitude for
a particle of momentum pi becoming a particle of momentum pf after interacting
12
with the potential. Defining out (xf , tf ) as a state of momentum pf in the distant
future:
1
(47)
out (xf , tf ) = eipf xf ,
V
we can write the amplitude for a transition from pi to pf as
Sf i = hout | + i.
(48)
R
Inserting the unit operator 1 = dxf |xf , tf ihxf , tf | into (48) and using the propagator expansion (46), we obtain
Z
(49)
13
xf , tf
xf , tf
xf , tf
x2, t2
Sfi =
x1, t1
+
x1, t1
xi , ti
xi , ti
xi , ti
We now consider a alteration of the system Lagrangian that models the presence
of a time-dependent source. Our discussion follows section 5.5 of Ryder [1] and
chapters 1 and 2 of Brown [3]. In this context, we call any external agent that
may cause a non-relativistic system to make a transition from one energy eigenstate
to another a source. For example, a time-dependent electric field may induce a
charged particle in a one dimensional harmonic oscillator potential to go from one
eigenenergy to another. In the context of field theory, a time-dependent source may
result in spontaneous particle creation4 . In either case, the source can be modeled
by altering the Lagrangian such that
L(q, q)
L(q, q)
+ J(t)q(t).
(50)
14
J=0
-d
J=0
t1
J=0
-d
t2
Te-id
Figure 6: The rotation of the time axis needed to isolate the ground state contribution to the propagator
=
mn
where we have introduced a basis of energy eigenstates H|ni = En |ni and energy
eigenfunctions n (q, t) = eiEn t hq|mi with n (q) = hq|ni. The J subscripts on the
propagators remind us that the source is to be accounted for. It is important to
note that n (q) is only a true eigenfunction for times when the source is not acting;
i.e. prior to t1 and later than t2 . The integral on the last line can be thought of
as a wavefunction, n (q1 , t1 ), that is propagated through the time when the source
is acting by hq2 , t2 |q1 , t1 iJ , and is then dotted with a wavefunction m (q2 , t2 ). But,
n (q1 , t1 ) and m (q2 , t2 ) are energy eigenfunctions for times before and after the
source, respectively. Hence, the integral is the amplitude that an energy eigenstate
|ni will become an energy eigenstate |mi through the action of the source. Now,
lets perform a rotation of the time-axis in the complex plane by some small angle
( > 0), as shown in figure 6. Under such a transformation
T1 T1 + i|T1 |
(51)
T2 T2 i|T2 |,
(52)
where we have chosen the axis of rotation to lie between T1 and T2 . We see
that the exponential term ei(En T2 Em T1 ) will acquire a damping that goes like
e(En |T2 |+Em |T1 |) . As we push T1 and T2 , the damping will become
infinite for each term in the sum, except for the ground state which we can set to
have an energy of E0 0. Therefore,
lim
T ei
(53)
> 0,
the form of 0 (q) in the distant past will still have the form of 0 (q) in the distant
future. In other words, it is the ground-to-ground state transition amplitude, which
we denote by
h0, |0, iJ lim hQ2 , T |Q1 , T iJ ,
(54)
T ei
Z
1
2
+ JQ + iq
h0, |0, iJ [dQ] exp i
dt L(Q, Q)
.
(55)
2
Finally, want to normalize this result such that if the source is turned off, the
amplitude h0, |0, i is unity. Defining
n R
h
io
R
+ JQ + 1 iQ2
[dQ] exp i dt L(Q, Q)
2
n R
h
io ,
Z[J] =
(56)
R
1
+ iQ2
[dQ] exp i dt L(Q, Q)
2
we have
h0, |0, iJ = Z[J].
(57)
Before moving on the the next section, we would like to establish a result that will
prove very useful later when we consider field theories. We first define the functional
derivative of Z[J] with respect to J(t0 ). Essentially, the functional derivative of a
functional f [y], where y = y(x), is the derivative of the discrete expression with
respect to the value of y at a given x. For example, the discrete version of Z[J] is
Z
n1
Xh
Z ( J(t0 ), J(t1 ) . . . J(tn1 ))
exp i
L(Qj , Q j )
j=0
) n1
Y
1
dQi ,
(58)
+ J(tj )Qj + iQ2j
2
i=1
where we have indicated that the discrete version of Z[J] is an ordinary function of n
variables J(tj ) and omitted the normalization factor. We have explicitly included
5
An alternative procedure for singling out the ground state contribution comes from considering
t to be purely imaginary, i.e. consideration of Euclidean space. This is discussed in the next section.
16
the weighting factor with each of the discrete variables to account for the fact
that as n , each J(tk ) covers a smaller and smaller portion of the integration
interval. The functional derivative of Z[J] with respect to J(tk ) is then the partial
derivative of the discrete expression with respect to J(tk ). Going back to the
continuum limit, we write the functional derivative of Z[J] with respect to Q(t1 ) as:
Z
Z
Z[J]
1
2
i [dQ]Q(t1 ) exp i
dt L(Q, Q) + JQ + iQ
.
(59)
J(t1 )
2
[dQ]Q(t1 ) Q(tn )
Z
1
2
exp i
dt L(Q, Q) + JQ + iQ
.
2
(60)
We notice a similarity between this expression and the expression from statistical
mechanics that gives the average value of a microscopic variable in the canonical
ensemble. We argue that equation (60) gives the exact same thing: the expectation
value of Q(t1 ) Q(tn ). There is one wrinkle, however, which we now proceed to
outline. Consider, with tk > tk0 ,
Z
hqf , tf |q(tk )q(tk0 )|qi , ti i =
dq1 dqn1 hqf , tf |qn1 , tn1 i
hqk , tk |q(tk )|qk1 , tk1 i hqk0 , tk0 |q(tk0 )|qk0 1 , tk0 1 i
hq1 , t1 |qi , ti i
Z
=
dq1 dqn1 q(tk )q(tk0 )hqf , tf |qn1 , tn1 i
hq1 , t1 |qi , ti i.
(61)
(63)
which follows from an argument similar to the one we used to calculate the matrix
element h0, |0, iJ with t1 = tk0 , t2 = tk and J = 0. Just as before, the
17
rotation of the time axis can be achieved by adding iq 2 /2 to the Lagrangian. This
calculation cannot be repeated for the case tk > tk0 because the order of q(tk )q(tk0 )
in hqf , tf |q(tk )q(tk0 )|qi , ti i cannot be switched without introducing terms involving
the commutator of H and q. But, in order to perform the decomposition (61), we
need the late q operator appearing to the left of the earlier q operator. What this
means is that we must be considering the time-ordered product of q(tk ) and q(tk0 ).
Putting all of this together along with the expression for the functional derivatives
of Z[J], we get
2 Z[J]
J=0
n Z[J]
(65)
We have demanded strict equality in these expression to ensure that the n = 0 case
returns h0|0i = 1. This is a very important formula for what follows.
We now move on to the quantum field theory of a scalar field (x). In this section
we draw on sections 6.1 and 6.3 of Ryder [1], chapter 2 of Popov [4] and section 3.2
of Brown [3]. The classical field is assumed to satisfy the Klein-Gordon equation
( + m2 ) = 0.
(66)
1
Z[J] =
Z0
with
1 2
4
[d] exp i d x L() + J(x) + i
,
2
Z
Z0 =
1 2
4
[d] exp i d x L() + i
.
2
(67)
(68)
(69)
In this expressions, the measure [d] is meant to convey an integration over all field
configurations, which can be achieved in practice by dividing spacetime into N 4
points (tm , xi , yj , zk ), with m, i, j, k = 1 . . . N . We can schematically merge all of
these indices into a single one n = 1 . . . N 4 . The field is considered to be a collection
RQ
of N 4 independent variables, and the functional integration [d] becomes
i di .
The functional Z[J] represents the fundamental object in the theory. Knowledge of
the form of Z[J] allows us to derive all the results that we could hope to obtain from
18
1
1
2
2
4
Z[J] =
(m i) + J
.
(70)
[d] exp i d x
Z0
2
Integrating the term by parts and using Gauss theorem to discard the
boundary term (assuming 0 at infinity) gives
Z
Z
1
1
4
2
[d] exp i d x
Z[J] =
( + m i) J
.
(71)
Z0
2
Regarding as an integration variable, we can change variables according to
(x) (x) + 0 (x).
Noting that
(72)
Z
4
d x 0 =
d4 x 0
Z
1
1
4
Z[J] =
[d] exp i d x
( + m2 i)
Z0
2
1
2
2
+( + m i)0 + 0 ( + m i)0 ( + 0 )J
.
2
(73)
(74)
(75)
We can solve this equation by introducing the Feynman propagator, which satisfies
( + m2 i)F (x) = 4 (x).
Hence,
(76)
Z
0 =
F (x y)J(y) d4 y.
(77)
Z
1
i
4
4
Z[J] =
exp
J(x)F (x y)J(y) d x d y
Z0
2
Z
Z
i
2
4
[d] exp
( + m i) d x .
2
19
(78)
Z
Z
i
2
4
Z0 = [d] exp
( + m i) d x .
2
This leads to our final expression:
Z
i
4
4
Z[J] = exp
J(x)F (x y)J(y) d x d y .
2
(79)
(80)
Before we move forward, we would like to make a comment on the inclusion of the
term in our expression for Z[J]. The reader will recall that the reason that we
added this term was to simulate the effects of rotating the time axis by a small angle
in the complex plane. Instead of adding the -term, we can instead rotate the
t axis by /2 so that t = i . The metric becomes = diag(1, 1, 1, 1),
which means that we are considering a Euclidean, not Lorentzian, manifold. In that
case, the vacuum-to-vacuum transition amplitude is
1
1
2
2
2 2
( ) + + m J
,
(81)
Z[J] =
[d] exp d dx
Z0
2
i2 /2
,
(2)4
2 + m2
(82)
(83)
Im(k0)
Im(k0)
Re(k0)
-w
+w
-w
Re(k 0)
+w
1
n Z[J]
(85)
We recall from the canonical formulation of field theory that h0|T [(x)(y)]|0i is
the amplitude for the creation of a particle at y and its later destruction at x (or
vice versa, depending on the times associated with x and y). Using (80) and our
previously mentioned notions of functional differentiation, we have
Z
1
Z[J] = Z[J] dy F (x1 y)J(y).
(86)
i J(x1 )
The second order derivative is
Z
Z
1 1
Z[J] = Z[J] dy1 F (x1 y1 )J(y1 ) dy2 F (x2 y2 )J(y2 )
i J(x2 ) i J(x1 )
+ iF (x1 x2 )Z[J].
Continuing,
1 1 1
Z[J] =
i J(x3 ) i J(x2 ) i J(x1 )
Z
iF (x1 x2 )Z[J] dy3 F (x3 y3 )J(y3 )
Z
iF (x1 x3 )Z[J] dy2 F (x2 y2 )J(y2 )
Z
iF (x2 x3 )Z[J] dy1 F (x1 y1 )J(y1 )
21
Z
Z[J]
(87)
(88)
(89)
(90)
We call h0|T [(x1 ) (xn )]|0i an n-point correlation function. Generalizing the
pattern above to T products of more field operators, we find that if n is odd, the
n-point function vanishes. On the other hand, if n is even, the correlation function reduces to the sum all possible permutations of products of 2-point functions
iF (x y) with x, y {x1 , . . . , xn }, x 6= y. When this result is derived from
canonical methods, it is know as Wicks Theorem.
It is interesting to interpret these results in terms of the Taylor series expansion
of Z[J] about J = 0. To make sense of this object, recall that the Taylor series
expansion of a function of a finite number of variables is
F (y1 , . . . , yk ) =
X
k
X
n=0 i1 =1
k
X
nF
1
yi1 . . . yik
.
n!
yi1 . . . yin yi =0
(91)
in =1
This is expansion is taken about the zero of all of the independent variables. Assuming the variables are weighted appropriately, when we go to the continuum limit
k we obtain that
1
nF
.
(92)
F [y] =
dx1 dxn y(x1 ) y(xn )
n!
y(x1 ) . . . y(xn ) y=0
n=0
Z[J] = 1 + x1
x2 +
x1
x2
x3
x4
x1
x2
x1
x2
+
x3
x4
x3
x4
Z
i2
1+
dx1 dx2 J(x1 )J(x2 )iF (x1 x2 )
2!
Z
i4
dx1 dx2 dx3 dx4 J(x1 )J(x2 )J(x3 )J(x4 )
4!
(93)
While free field theory has a certain amount of elegance to it, it is not terribly
interesting. In this section, we consider a self-interacting field whose Lagrangian is
23
given by
1
1
L() = m2 2 + Lint () = L0 () + Lint ().
(94)
2
2
The discussion follows section 6.4 of Ryder [1]. Here, Lint () is the Lagrangian
describing the self interaction. The generating functional is
R
R
[d] exp iS + i d4 x J
R
Z[J] =
,
(95)
[d]eiS
where S is the classical action
Z
S = d4 x [L0 () + Lint ()] = S0 + Sint .
(96)
We have dropped the i2 /2 term used to single out the ground state, which we can
rationalize by the rotation of the time axis. In this section, we will write the free
field propagator as
R
R
[d] exp iS0 + i d4 x J
R
.
(97)
Z0 [J] =
[d]eiS0
We would like to write Z[J] in a form particularly useful for calculations. We
cant really reproduce the manipulations of the last section because the Lint in the
action S introduces difficulties when the shift + 0 is performed. We will
instead derive a differential equation satisfied by Z[J] and then solve it in terms
of J(x) and the Feynman propagator. The result will be something that we might
have guessed intuitively. What is the equation satisfied by the free field propagator?
Now, we know that
Z
1
Z0 [J] = Z0 [J] dy F (x y)J(y).
(98)
i J(x)
We operate on both sides with x +m2 and use the defining relation for the Feynman
propagator (76) to get:
(x + m2 )
1
Z0 [J] = J(x)Z0 [J].
i J(x)
(99)
The is the differential equation satisfied by Z0 [J]. In order to find the differential
equation satisfied by Z[J], let us define the functional
iS
=R e
Z[]
.
[d] eiS
Then
Z
Z[J] =
exp i d4 x J ,
[d] Z[]
24
(100)
(101)
differentiate Z[]
with respect to using
Z
1
1 2 2
4
S =
d x
m + Lint ()
2
2
Z
1
2
4
( + m ) Lint ,
(102)
= d x
2
where Gauss theorem has been used. The superiority of the latter form is that we
Z[]
L0int ()Z[],
= ( + m2 )(x)Z[]
(x)
where
(103)
Lint
(104)
[d]eiS .
(106)
L0int () =
R
and = x . We multiply both sides of (103) by exp[i J(y)(y) d4 y] and integrate
over , i.e. we take the Fourier transform. The RHS of (103) becomes
Z
Z
1
2
4
RHS =
( + m ) [d](x) exp iS + i J(y)(y) d y
Z0
Z
Z
1
0
4
Z
Z0 =
Z
4
J(y)(y) d y ,
which leads to
Z
Z
1
1
n
4
Z[J] =
[d] (x) exp iS + i J(y)(y) d y .
i J(x)
Z0
(107)
(108)
Now, we assume that L0int () possesses a Taylor series expansion in . We can then
reproduce the second term in (105) by adding together a series of contributions of
the form of (108). Hence,
1 Z[J]
1
RHS = ( + m2 )
L0int
Z[J].
(109)
i J(x)
i J(x)
25
Z()
4
LHS = i [d]
exp i J(y)(y) d y
(x)
Z
= iZ()
exp i J(y)(y) d4 y
i [d]Z()
exp i J(y)(y) d y
(x)
Z
Z
4
= J(x)Z[J].
(110)
In the second line we have performed a functional integration by parts, which follows
from the fact that the functional derivative satisfies the product rule and functional
integral obeys the fundamental theorem of calculus. The boundary term must vanish
as because if it didnt, the integral for Z[J] would diverge. Putting together
our formulae for the LHS and RHS of the Fourier transform of (103):
1 Z[J]
1
( + m2 )
L0int
Z[J] = J(x)Z[J].
(111)
i J(x)
i J(x)
This is the differential equation satisfied by Z[J]. We see that if the interacting
Lagrangian is set to zero, the result reduces to (99).
We will assume a solution of the differential equation of the form
1
4
Z0 [J]
(112)
Z[J] = N exp i d x Lint
i J
partly because this is what we might expect from L = L0 +Lint if we replace i/J
with , as we usually do, partly because we know its the right answer. As usual,
N is a normalizing factor. Lets first establish and identity:
h R
i
h R
1
= iik .
(114)
Ji ,
i Jk
The continuum limit (n ) of this is
1
J(x),
= i(x y).
i J(y)
26
(115)
Now, its not hard to see that if A and B are operators whose commutator is a
number a, then
[A, B] = a,
[A, B 2 ] = 2aB,
[A, B 3 ] = 3aB 2 ,
..
.
[A, B n ] = naB n1 .
Hence,
J(x),
1
i J(y)
= i(x y)n
1
i J(y)
n1
.
(116)
X 1
1
cn n .
Lint () = c0 + c1 + c2 2 + =
2!
n!
(117)
n=0
n #
n1
X
X
1
1
1
1
J(x),
cn
= i(x y)
cn
.
n!
i J(y)
(n 1)!
i J(y)
n=0
(118)
n=1
But
L0int ()
X
X
1
n
n1
cn
=
cn n1 .
=
n!
(n 1)!
(119)
n=1
n=0
Z
1
1
4
0
J(x), i d y Lint
= Lint
.
i J(y)
i J(x)
(120)
where we have scaled Lint by i. Finally, note that the Hausdorff formula gives
eB AeB = A + [A, B]
(121)
when
[A, B] is a number. Putting (120) into (121) with A = J(x) and B =
R
i d4 y Lint (i/J(y)) yields (113).
Using our assumed form for Z[J] (112) and the identity (113) we get
Z
1
4
J(x)Z[J] = N J(x) exp i d y Lint
Z0 [J]
i J(y)
Z
1
1
= N exp i d4 y Lint
J(x) L0int
Z0 [J]
i J(y)
i J(x)
27
1
1
4
= N exp i d y Lint
(x + m2 )
Z0 [J]
i J(y)
i J(x)
1
1
L0int
N exp i d4 y Lint
Z0 [J]
i J(x)
i J(y)
1
2 1 Z[J]
0
= (x + m )
Lint
Z[J].
i J(x)
i J(x)
In going from the second to the third line, we have used the differential equation
satisfied by Z0 [J] (99). This result is just the differential equation (111), which
confirms that we can write
h R
i
h R
i
Z[J] =
(122)
where we have
i
Z0 [J] = exp
2
Z
4
J(x)F (x y)J(y) d x d y .
(123)
With these two equations, we have succeeding in writing down the generating functional entirely in terms of the source J and the Feynman propagator F .
4 and 3 theory
We would like to demonstrate the calculation of the generating functional Z[J] and
some n-point functions in the case of self-interacting 4 and 3 theory. We follow
section 6.5 of Ryder [1] and chapter 2 of Popov [4]. We first consider 4 theory,
which has the interacting Lagrangian:
Lint () = 4 .
4!
(124)
It goes without saying that is small. The generating functional Z4 [J] for 4 is
R 4 1 4
i
exp 4! d z i J(z)
Z0 [J]
Z4 [J] =
(125)
4
R
1
i
4
exp 4! d z i J(x)
Z0 [J]
J=0
4
Z
i
1
num Z4 [J] = 1
d4 z
+ Z0 [J].
4!
i J(z)
28
(126)
i
4
2
d z 3F (0) + Z0 [J].
denom Z4 [J] = num Z4 [J]
= 1
4!
J=0
Putting the two results together yields, to order , we have:
(
"
Z
2
Z
i
4
4
Z4 [J] = 1
d z 6iF (0)
d x F (z x)J(x)
4!
)
Z
4 #
4
+
d x F (z x)J(x)
+ Z0 [J],
(128)
(129)
Now, lets do the same thing for 3 theory, where the interacting Lagrangian is
Lint () = 3 .
3!
The numerator of the generating functional is
"
3 #
i
1
Z0 [J].
num Z3 [J] = exp
d4 z
3!
i J(z)
Expanding to order , we get
Z
Z
i
4
d z 3iF (0) d4 x F (z x)J(x)
num Z3 [J] = 1
3!
)
3 #
Z
d4 x F (z x)J(x)
+ Z0 [J].
(130)
(131)
(132)
Z
Z
i
Z3 [J] = 1
d4 z 3iF (0) d4 x F (z x)J(x)
3!
)
Z
3 #
d4 x F (z x)J(x)
+ Z0 [J].
(133)
6
The Feynman propagator evaluated at zero is, of course, infinite. Its inclusion in the generating
functional represents the infinite self-energy of particles in the theory and must be regulated by
renormalization, which we will not consider here.
29
x1
Z4[J] = 1 +
l
4 x1
x2
il
24
x2
+
z
x3
x4
x3 +
x1
Z3[J] = 1 - l x1
2
+ il
6
x2
Figure 9: Pictorial representation of the generating functional for 4 and 3 theory
respectively
We can represent the two generating functionals Z4 [J] and Z3 [J] diagrammatically with the following Feynman rules:
1. The spacetime point zi is associated with internal points, all other variables
go with external points (recall that zi is the coordinate that occurred in the
interacting Lagrangian).
2. A line between x and z comes with a propagator F (x z).
3. Each internal point comes with a factor of i/4! for 4 theory, i/3! for 3
theory.
4. External points x come with a factor J(x).
5. Terms of the form F (0) represent closed loops, or propagators who begin
and end at the same point, joined to internal points.
6. All spacetime points are integrated over.
7. Each term in the series is multiplied by the free particle propagator Z0 [J].
These rules give the pictures in figure 9 for Z4 [J] and Z3 [J]. These diagrams stress
the general structure of the two theories. Because of the power of 4 in the interacting Lagrangian in 4 theory, the vertices in the associated Feynman diagrams are
attached to 4 legs. Similarly, the vertices in 3 theory are attached to 3 legs. A
peculiarity of 3 theory is the diagram with only one external point. As we will see
below, this gives rise to a one point correlation function which means that particles
in 3 can spontaneously self-destruct.
30
i x1
x2 - l
2 x1
il x1
4
x2
Figure 10: Pictorial representation of the two-point function h0|T [(x1 )(x2 )]|0i4 for
4 theory (top) and of the one-point function h0|T [(x1 )]|0i3 for 3 theory (bottom)
Lets calculate the 2-point function for 4 theory:
2 Z4 [J]
.
h0|T [(x1 )(x2 )]|0i4 =
J(x2 )J(x1 ) J=0
(134)
The functional differentiation is easy to do from (129) using the our previously
derived formulae for derivatives of the free field propagator (87) (90). The result
is
Z
2 Z3 [J]
h0|T [(x1 )(x2 )]|0i3 =
= iF (x1 x2 ) + O(2 ).
(136)
J(x2 )J(x1 ) J=0
This isnt terribly interesting, we see that there are no corrections to the free field
result to this level in perturbation theory. We shouldnt be surprised, there is no
way to connect to external lines to a single vertex if the vertex must be attached
to three lines. However, there is a way to connect a vertex to one external line and
one internal line, as shown in figure 10. This is borne out by the calculation of the
one-point function:
2 Z3 [J]
h0|T [(x1 )]|0i3 = i
J(x1 ) J=0
Z
i
=
F (0) d4 z F (z x1 ) + O(2 ).
(137)
4
31
x1
z
-l
x2
x3 + l
4
x1
x2
x2
+
x3
x3
+
x1
x3
x1
x2
Figure 11: Pictorial representation of h0|T [(x1 )(x2 )(x3 )]|0i3 , the three point
function, for 3 theory
Certainly a strange beast, this represents the amplitude of a particle being spontaneously created out of the vacuum, all by itself7 . We can also calculate the three
point function
2 Z3 [J]
+ F (0) d4 z [F (z x1 )F (x2 x3 )
4
+ F (z x2 )F (x3 x1 )
+ F (z x3 )F (x1 x2 )].
(138)
The first integral involves all the external legs attached to the same point, i.e. two
particles merging into one or one particle splitting in two. The second integral
is the product of free 2-point functions and the 1-point function we have already
calculated. The diagram is in figure 11.
We have hence shown how the generating functional and n-point functions can
be found from simple functional differentiation of equations (122) and (123) and
expressed in terms of Feynman diagrams. These diagrams can be converted into
scattering amplitudes via the LSZ reduction formula8 :
hp1 , . . . , pn , +|q1 , . . . , qm , i = disconnected terms
Z
n+m
+ (i)
dy1 dyn dx1 dxm
ei(p1 y1 ++pn yn q1 x1 qm xm )
(y1 + m2 ) (yn + m2 )
(x1 + m2 ) (xm + m2 )
h0|T [(y1 ) (yn )(x1 ) . . . (xm )]|0i.
7
32
In practice, these scattering amplitudes are the only meaningful quantities in quantum field theory since they are the only things that can be directly measured. So,
having arrived at a point where we can calculate hp1 , . . . , pn , +|q1 , . . . , qm , i
using the generating functional, we have completed our formulation of self-interacting
field theories in terms of path integrals.
More complicated theories, such as QED, can be quantized in terms of pathintegrals, but there are several issues that need to be addressed when writing down
Z[J]
for gauge fields. One finds that Z[J] is infinite for gauge fields A , because the
R
[dA ] integration includes an infinite number of contributions from fields related
by a simple gauge transformation. The resolution is the addition of gauge fixing
terms to the Lagrangian and the appearance of non-physical ghost fields. Such
things are beyond the scope of this paper.
Conclusions
33
References
[1] Lewis H. Ryder. Quantum Field Theory, 2nd ed.. Cambridge: 1996.
[2] Gordon Baym. Lecture Notes on Quantum Mechanics. Benjamin/Cummings:
1969.
[3] Lowell S. Brown. Quantum Field Theory. Cambridge: 1992.
[4] Viktor N. Poppv. Functional Integrals in Quantum Field Theory and Statistical
Physics. Reidel: 1983.
34