LSZ Reduction Formula

A Derivation

Sine of Psi

February 9, 2016

0. References
The arguments presented in these notes are based on the LSZ Reduction Formula Wikipedia article,
Srednickis Quantum Field Theory, Peskin & Schroeders An Introduction to Quantum Field Theory,
and Schwartzs Quantum Field Theory and the Standard Model.

1. Single Particle and Multiparticle States

Previously1 we showed that if we choose our Lorentz-invariant Lagrangian2 to be,

L0 =

1
1
m20 2 ,
2
2

then we get the Klein-Gordon equation (KG eq) from the Euler-Lagrange equations (EL eq):

( 2 + m20 )(x) = 0

Note: Im using subscript zeroes to denote certain quantities as corresponding to the free theorythat is,
the theory as it appears without interactions. We found a Lorentz-invariant solution to the KG eq that we
called a field,
Z
0 (x, t) =

where k0 = wk

d3 k 1
ak eikx + ak e+ikx
3
(2) 2k

m20 + ~k 2 and the Fourier coefficients ak , ak are independent of time. In particular,

because 0 transforms as a scalar under Lorentz transformations, we call this field the free massive real
scalar field and call its Lagrangian the free massive real scalar Lagrangian. We subsequently quantized
the field, yielding the notationally-similar expression,

0 (x, t) =

1I


d3 k 1  ~ ikx
~ +ikx
a

(
k)e
+
a

(
k)e
0
0
(2)3 2k

will be using the mostly-minus metric, following the typical high-energy convention.
is technically a Lagrangian density; however, its conventional in QFT to call L a Lagrangian because we rarely care
about the Lagrangian L but frequently talk about the Lagrangian density L.
2 This

where our Fourier coefficients are now creation/annihilation operators on a Hilbert space. We say that
0 (x, t) is a quantum field and, in particular, is the free massive real scalar quantum field. The Fourier
coefficients are creation/annihilation operators in the sense that they satisfy the commutation relations of
(the continuum-equivalent of) a simple harmonic oscillator,

[
a0 (~k1 ), a
0 (~k2 )] = 0
[
a0 (~k1 ), a
0 (~k2 )] = 0
[
a0 (~k1 ), a
0 (~k2 )] = (2)3 (2k1 ) 3 (~k1 ~k2 )

We have yet to tie this to a physical interpretation. Lets do so now. We first postulate the existence of
the free vacuum. A theorys vacuum is what the universe looks like according to that theory when no
particles are present.3 In the free theory, its usually denoted by |0i, and it corresponds to an eigenstate of
the 4-momentum operator P with eigenvalue p = 0. Well normalize |0i so that h0|0i = 1. Furthermore,
we assume that the vacuum is Lorentz-invariant. That is, you cant make particles appear by boosting
yourself into a new reference frame.4
Lets define a ket |ki to describe the universe with a single particle of momentum ~k and demand it satisfy,
|ki a
0 (~k) |0i
With this definition, we have the following normalization for single-particle states5 :

hk1 |k2 i = (2)3 2k1 (~k1 ~k2 )

In perfect analogy with the simple harmonic oscillator, because we have defined that a
0 (~k) act as a creation operator that creates a single-particle with momentum ~k, it must be the case that a
0 (~k) acts as an
annihilation operator that destroys a single particle with momentum ~k. If there is no such particle to
annihilate, then the annihilation operator spits out zero. This means that the vacuum is also defined to be
3 QFT calls any excitation of the vacuum a particle, even if that excitation is not a particle in the colloquial sense. Additionally, a vacuum state might be empty of particles only in the sense that whatever stuff the vacuum is made of lacks any
excitations. For example, in many-body physics, the vacuum could be the ground state of a multi-molecular compound and
the particles could be complicated vibrational modes of the compound. In high-energy physics, the vacuum is often spacetime
itself.
4 Theres a non-trivial technicality here: a theory can sometimes have more than one vacuum. Not only is this possible, but
the Standard Model of particle physics requires a degenerate vacuum in order for particles to gain mass.
5 Exercise: Prove this equation using the definition of |ki and the properties of a
0 (~k).

Figure 1: Each curve represents values of (|~

p|, E) for invariant masses available in the free theory. The
lowest curve corresponds to single-particle states, each of which has a mass m0 . The next highest curve
corresponds to two-particle states with mass 2m0 . Above the two-particle curve lies a continuumP
of states.
This is because, for instance,Pthere exist many multiparticle states with total momentum P~ = i p~i but
differing total
p energies E = i Ei (due to relative motions of particles) thereby yielding every invariant
mass M E 2 P~ 2 larger than 2m0 .
the state that is annihilated by every annihilation operator: a
0 (~k) |0i = 0 for all ~k.
A multiparticle state is constructed by applying multiple creation operators. Unlike introductory quantum
mechanics, we need not worry about symmetrizing or antisymmetrizing any states. The commutation relations of the operators ensure the appropriate statistics are obeyed. Figure 1 summarizes the mass spectrum
of a free theory.

2. S-Matrix Elements and the Far Past, Far Future

Experiments often involve setting up an initial state, letting it evolve over time, and then extracting some
kind of physical information from the final state. QFT is often played with scattering experiments, with the
following progression of events:

1. Initial State: Fire known particles at one-another at some given energy.6

6 The particles might be fundamental particles like electrons or effective particles such as nuclei and atoms. So long as
whatever were throwing doesnt break into a new realm of physics during the experiment (e.g. throwing protons at low enough
energies as to not care about their quark structure), its good enough to be a particle.

2. ???: Particles near each other. Interactions between different particles cause quantum-weirdness.

3. Final State: New particles fly out and we measure their 4-momentum, charge, etc.

As we might expect from quantum mechanics, the probabilty of getting a certain final state from a certain
initial state is related to the overlap of their corresponding wavefunctions. If |ii describes our initial incoming
particles, then the probability that we end up with a final state |f i of final outgoing particles is related to
the quantum amplitude,

Sf i hf |iiH

which we call an S-matrix element. The subscript H indicates that this expression holds in the Heisenberg
picture, where each state is time-independent and operators evolve in time. We could alternatively use the
Schr
odinger picture, where each operator is time-independent and states evolve in time. Switching from one
picture to the other requires we recast our formally time-independent states into time-dependent forms. In
this vein, suppose we define our initial state |ii at some initial time ti . Similarly, lets define our final state
|f i at some final time tf . To complete the conversion, we must compare these states at some equal time tc .
In this language, the equal-time kets become,
h
i
c ti ) |ii U(t
c , ti ) |ii
|i(tc )i = exp iH(t
h
i
f tc ) |f i U(t
f , tc ) |f i
|f (tc )i = exp iH(t

such that,

f , tc )U(t
c , ti ) |ii = hf | U(t
f , ti ) |ii
hf |iiH = hf (tc )|i(tc )iS = hf | U(t
S
S

where the subscript S indicates when were using the Shrodinger picture. A natural follow-up question
becomes: when should our initial time and final time occur? This is tricky because such a choice is inherently
frame-dependent. QFT dodges this conundrum entirely by taking the initial and final times to t = and
t = + respectively. In this context, we call t the far past and t + the far future.7 This is,
7 Too bad we also care about unstable particles which (by virtue of their instability) cannot survive to the far future. Handling
decay phenomena requires additional care, but can be done. Its irrelevant to our current purposes however.

of course, an approximation, but its a very good one in our region of the universe. In this approximation,

Sf i =

lim

ti ,tf

f , ti ) |ii
hf | U(t
S

Much of QFT centers around calculating S-matrix elements for various processes. However, the formulae
weve developed thus far are not useful for such calculations. In particular, our discussion has been limited to
Hamiltonian-based time-evolution which muddies up special relativity because of the Hamiltonians Lorentzvariance. It would be nice to, instead, relate the calculation of S-matrix elements to the Lorentz-invariant
fields weve developed over these last few weeks. Such a relation does exist, and that relation is the LSZ
reduction formula.8 I will be deriving the LSZ reduction formula for interacting massive real scalar fields.
The generalization to massive complex scalar fields poses no complications.9

3. Interactions and the Mass Shift

Of course, if we want to discuss (nontrivial) scattering experiments, we need to include interaction terms
in our Lagrangian. But turning on interactions comes at a conceptual cost. In the free theory, it wasnt
game-breaking to leave the time at which we defined our creation/annihilation operators unspecified. This
is because the creation/annihilation operators only ever gain a phase as time evolves.10 When interactions
are present, an operator that creates a single-particle at one time may create a multiparticle state at another
time. This can be traced back to time-evolution of our fields:

+iHt

(x)
eiHt (x)e
,

where

e+iHt

X (iHt)
n
n

n!

Interaction terms in our Lagrangian will combine numerous quantum fields. By introducing such interaction
terms into our Lagrangian, we cause new equally-complicated terms in our Hamiltonian as well. Then, as
indicated, we time-evolve our fields by multiplying both sides with power series of those terms. This is how
the creation of single-particle states at any one time becomes the creation of multiparticle states at other
times.
8 Sometimes

this is simply called the reduction formula.

general forms of the LSZ reduction formula hold for massive particles with nonzero spin, but will involve polarization elements and additional renormalization parameters. Massless particles have additional symmetries that complicate the
relationship between fields and S-matrix elements.
10 Exercise: Show this using the Heisenberg picture. Additionally, show this is equivalent to |ki remaining the same up to a
phase as time evolves in the Schr
odinger picture.
9 More

Figure 2: A qualitative illustration of the typical mass spectrum of an interacting theory. Like the free
theory, we have a single-particle curve and a continuum of states with invariant mass M 2m. Note that
these now occur at multiples of the interacting mass m which is generally not equal to the free mass m0 .
There exist additional curves between the single-particle and two-particle states corresponding to 2-particle
bound states. These generically possess invariant masses M . 2m.
Additionally, the introduction of interactions complicates our vacuum. Even the free vacuum is a background
static of quantum fluctuations reminiscent of particles popping in and out of existence even when the total
particle number of the universe is zero. Interactions mean more static. This new, fuzzier vacuum is called the
interacting vacuum and is denoted |i. Like the free vacuum, it is annihilated by annihilation operators;
however, our annihilation operators are also complicated by interactions. It is a nontrivial mess to sort out
and its something were going to have to deal with if we want to prove the LSZ reduction formula.
Lastly, I want to clarify how interactions affect the masses of our particles. A free particle propagates
forever, but an interacting particle will radiate as it propagates, such thateven if its the only particle
in the universeit will end up interacting with itself. This is an important effect that well analyze more
in-depth later in the class. For now, we will only focus on a qualitative consequence of this self-interaction:
namely, the mass of an interacting particle is different than the free mass that shows up in the Lagrangian of
the theory. This is because interacting particles are constantly throwing out and subsequently reabsorbing
particles. This perpetual juggling alters the long-term propagation effects of our particles (compared to the
free theory) and thus influences each particles mass. We say that there is a mass shift between the free
mass m0 and the interacting mass m.11 We must proceed with caution.
11 You can think of the interacting mass as an effective mass. For instance, a photon in free space is massless (m = 0) so it
0
travels at the speed of light. We can fire that same photon into a superconductor, where itll slow down and act like it possesses

4. Inverting the Inhomogeneous KG Eq

Lets get quantitative. If we add new terms (generically called Lint ) to our free Lagrangian L0 , we can make
our scalar field interact with itself:

L=

1
1
m20 2 + Lint
2
2

Applying the EL eq to this gives us an inhomogeneous version of the KG eq,

( 2 + m20 )(x) = j0 (x)

Lint

We begin our derivation of the LSZ reduction formula by inverting this equation. Suppose we know a Greens
function of the KG eq. That is, suppose we have some function f (x) such that ( 2 + m20 )f (x) = 4 (x).
Conceptually, the Greens function solves the inhomogeneous KG eq for a point current. By integrating
over point currents 4 (x y) at each point y, we can reconstruct any current j0 (x) we want. This would
solve any (non-quantized) scalar theory with interactions.
It turns out the Greens functions of the KG eq are related to 2-point correlation functions such as
1 )(x
2 ) |i. We will not go into details about 2-point correlation functions right now. Instead, I
h| (x
will simply state (to be proved another day) that there are (among others) two Greens functions called the
retarded propagator ret (x) and the advanced propagator adv (x). The retarded propagator allows
us to express an interaction current in terms of its construction via fields at earlier times. Similarly, the
advanced propagator allows us to express an interaction current in terms of its influence on fields at later
times.
Based on my earlier comments, we shouldnt expect the masses of the free particles (m0 ) to equal the masses
of the interacting particles (m). Lets reorganize the inhomogeneous KG eq with that in mind:

( 2 + m20 )(x) = j0 (x)

( 2 + m20 )(x) + (m2 m20 )(x) = j0 (x) + (m2 m20 )(x)
( 2 + m2 )(x) = j0 (x) + (m2 m20 )(x)

a mass m 6= 0. This effective mass phenomena is responsible for the finite skin-depth of magnetic fields into superconductors.

Hence,

( 2 + m2 )(x) = j(x),

where

j(x) j0 (x) + (m2 m20 )(x)

This is the equation we wish to invert. Using the retarded propagator, we can build forward in time so that
our solution to the inhomogeneous KG eq will be a sum of homogeneous KG solutions in the far past and
the specific solution corresponding to the current.

(x) =

Z
Z in (x)|t= + d4 y ret (x y)j(y)
{z
} |
|
{z
}
homogeneous solution

inhomogeneous solution

The in-field in (x)|t= is a free field with mass m.12 The constant Z is called the wavefunction renormalization. We can decompose the quantized form of Equation into a sum of plane waves multiplying
operators:


in (~x)

Z
=
t=


1
1  ikx
+ikx ~
~
d k
e
a
in (k) + e
a
in (k) 
(2)3 2k
t=
3

Like we discussed earlier, we can only define one specific moment worth of creation/annihilation opera
tors a
in (~k), a
in (~k) corresponding to the full interacting field (x).
Time-evolving will mixing those creation/annihilation operators in such a way that they create/annihilate multiparticle states as well. Thus, we
shouldnt expect a
in (~k), a
in (~k) will behave like creation/annihilation operators, especially because theres a

whole inhomogeneous solution worth of complications separating (x)

and in (x). We will have to check if
a
in (~k), a
in (~k) behave like single-particle creation/annihilation operators, which is what well do in the next
section.
Theres also some ambiguity in the above formula because of an infinite phase eik . This phase is
unimportant for proving the LSZ reduction formula. To sidestep the issue, lets time-evolve the in-field
to a finite time, but do so using the free Hamiltonian H0 instead of the full Hamiltonian H. Choosing
to evolve our operators with the free Hamiltonian while an interacting Hamiltonian is present is called
the interaction picture. Its nice because we know that the free Hamiltonian doesnt mix single-particle
operators with multiparticle operators. Therefore, it wont hurt us when we examine how a
in (~k), a
in (~k)
12 To really drive this point home, the mass of the free field (x) for this same theory is some number that we label m .
0
0
Depending on the content of Lint , different mass shifts will occur, and thus different interacting theories with common m0 will
generate different asymptotic/physical masses m. Alternatively, we could demand several theories (each with their own Lint )
generate the same asymptotic mass m, in which case theyd each have to possess different values of m0 .

relate to single-particle operators in the next section. We define, for any time,

in (x) =

d3 k


1  ikx
1
~
~k, t) + e+ikx a
(
k,
t)
e
a

in
in
(2)3 2k

This implies,
a
in (~k, t) = i

d3 x eikx 0 in (x)

We recover the in-field when t .

We must do one more thing. If we proceed as we are, well butt heads with fundamental quantum mechanics.
Namely, we will be at odds with the uncertainty principle. We want our particles to be spatially localized
at the far past and far future but that can only happen if there is some uncertainty in their momenta, and
right now were only discussing exact momentum eigenstates. What we should use instead are wavepackets
of momentum eigenstates localized about some 3-momentum, say ~k:
a
in (~k, t)

d3 ` w~k (~`)
ain (~`, t) = i

d3 ` w~k (~`)

d3 x ei`x 0 in (x)

where w~k (~`) is a function of momenta that is sharply peaked around ~k. The smearing performed by this
function is vital to avoiding infinities in our proof.

5. Do a
in (~k) create single-particle states?
If we want a
in (~k), a
in (~k) to behave like creation/annihilation operators in the interacting theory, they need
to overlap only with single-particle states. That is, ideally,
h| a
in |i = 0
hk| a
in |i =
6 0
hp, n| a
in |i = 0

where |p, ni labels a multiparticle state with total 4-momentum p and other quantum numbers n.
If we can manage h| a
in (~k) |i = 0 for all ~k, then well have h| in (0) |i = 0. Suppose we instead have
10

v h| in (0) |i =
6 0. By replacing every (x) in the Lagrangian with (x)+v, we will shift this expectation
value and well have, by construction, h| in (x) |i = 0. Our creation/annihilation operators will always
appear in fields, so this is good enough.
Consider next the multiparticle states |p, ni. Each state will have some invariant mass Mn,p p p . To
start things off, note that,

hp, n| in (x) |i = hp, n| e+iP x in (0)eiP x |i ,

= e+ipx hp, n| in (0) |i ,

e+ipx An (~
p)

where An (~
p) is a Lorentz-invariant product of various 4-momenta that specify the state, but which cannot
have position dependence.
We hope that hp, n| a
in (~k) |i equals zero. For mathematical consistency (and to avoid the same issues that
made us tie the incoming particles together into wave packets), we should focus on normalizable combinations
of |p, ni:

|iq~

XZ

d3 p n,~q (~
p) |p, ni

where the n,~q (~

p) are wave packets sharply peaked around the total three-momentum ~q. We perform the
calculations at finite time and then take the limit as we approach the far future. Using Equation ,
in (~k, t) |i
q
~ h| a

= i

XZ

d3 p n,~
p)
q (~

d3 ` w~k (~`)

|i ,
d3 x ei`x 0 hp, n| (x)

d ` w~k (~`)

d3 x ei`x 0 e+ipx An (~
p),

= i

XZ

d p

n,~
p)
q (~

XZ

d3 p n,~
p)
q (~

d3 ` w~k (~`)

d3 x (p0 + `0 )ei(p`)x An (~
p)

But

d3 x ei(~p`)~x = (2)3 3 (~
p ~`), so,
in (~k, t) |i =
q
~ h| a

XZ

d3 p (2)3 (p0 + `0 )n,~

p)w~k (~
p)An (~
p)ei(p
q (~

`0 )t

where p0

q
p
2 and `0
p~2 + Mn,p
p~2 + m2 . Recall Figure 2, wherein we established that Mn,p & 2m > m.
11

Given this, p0 > `0 so that p0 `0 6= 0. By the Riemann-Lebesgue Lemma,13 the RHS goes to zero as
|t| +, so
in (~k) |i
q
~ h| a

in (~k, t) |i
q
~ h| a

= lim

=0

Hence, our a
in operator might create multiparticle states at arbitrary times but as we go into the far past
those contributions vanish. A similar argument holds for q~ h| a
in (~k) |i. This implies q~ h| in (x) |i 0
in the far past.14
If we attempt to repeat the multiparticle argument for one-particle states, we would find there exists an
integration region where p0 `0 , and hence the Riemann-Lebesgue Lemma is inapplicable. (This is good
because otherwise wed be forced to conclude in (x) = 0 in the infinite past, which would make in (x)
useless to us.) Because in (x) 6=
0 by assumption, yet h| in (x) |i = 0 and q~ h| in (~k) |0i 0, it must be
the case that in (~k) is a one-particle operator in the infinite past.
With this result in hand, lets return to the quantized form of Equation .

(x)
=

Z in (x) +

d4 y ret (x y)j(y)

In 1955, the German physicists H. Lehmann, K. Symanzik, and W. Zimmermann (the L, S, and Z of LSZ)


demonstrated15 that the single-particle behavior of in (x)
combined with the above equation imply
t=

the following weak asymptotic relation:

Z
lim
0

|i =
d3 x eipx h| 0 (x)

Z 3 ipx
Z d xe
h| 0 in (x) |i

where |i , |i are normalizable states. Note that (thanks to the limit) the left-hand side is independent of
time, while the right-hand side still carries an explicit time variable. This is only consistent if the timedependence of eipx exactly counteracts the time-dependence of the rest of the expression, which says the
operator on the RHS looks like a single-particle creation operator in the interaction picture. We will use an

13 The Riemann-Lebesgue Lemme states that the integral of a smooth function against a rapidly oscillating wave goes to
zero as the oscillations become infinitely rapid. In the limit, the wave oscillates between positive and negative so rapidly that
adjacent points cancel exactly.
14 Exercise: Show that h|
in (x) |i 0
q
~
15 Lehmann, H.; Symanzik, K.; Zimmermann, W.; Zur Formulierung quantisierter Feldtheorien Nuovo Cim. 1 (1955) 205

12

equivalent version of this expression. Specifically, via Equation ,

Z
lim
0

|i = i Z h| a
d3 x eipx h| 0 (x)
in (~k) |i ,

which is, upon reorganizing,

i
h| a
in (~k) |i =
Z

Z
lim
0

|i
d3 x eipx h| 0 (x)

( )1

We can repeat this analysis (solving the inhomogeneous KG eq) using the advanced propagator to work with
states in the future called out-states,

(x) =

Z
Zout (x) +

d4 y adv (x y)j(y)

which subsequently leads to,

h| a
out (~k) |i

i
=
Z

Z
lim

x0 +

|i
d3 x eipx h| 0 (x)

( )2

6. The Proof
Consider the matrix element,

1 ) (y
n )} |k; ii
M hf | T {(y

The symbol T { } denotes the time-ordering operator which takes in some number of fields and rearranges them such that they are ordered with the latest times to the left and the earliest times to the right.
This ensures causality when creating/annihilating particles. Well see it many times in the future. As a
quick example, suppose t1 < t2 < t3 . Then,
y1 , t1 )(~
y3 , t3 )(~
y2 , t2 )} = (~
y3 , t3 )(~
y2 , t2 )(~
y1 , t1 )
T {(~

The time-ordering ignores the commutation relations of (x)

because its really just receiving a list of what
fields are inside of it and then spitting them out in reverse-chronological order. Things get complicated when
13

we ask it to time-order multiple fields that are at the same time (then you have to start worrying about
picking a convention). We can dodge that bullet by saying no two fields occur at the same time. If we later
want two fields occurring at the same time, then we can just take the limit of our result as those two times
become identical. In this way, the subtleties of time-ordering need not hinder our proof.
Note that the matrix element M is notably more general than an S-matrix element. This is for the sake of
creating an inductive proof. Continuing to be careful about wavepackets transforms M into,
Z
M=

1 ) (y
n )}
d3 p w~k (~
p) hf | T {(y
ain (~
p) |ii

Assume there are no other particles with momentum p in the out-state; that is, ignore forward scattering.
This assumption restricts what questions the LSZ formula can answer. However, just like the case of identical
time in the time-ordering operator, the forward scattering case can be recovered after-the-fact by taking the
limit of our result as an out-state momentum approaches an in-state momentum. With this assumption,
Z

1 ) (y
n )} |ii = 0
d3 p w~k (~
p) hf | a
out (~
p)T {(y

Subtracting these and applying Equations yields,

Z

h
i
1 ) (y
n )}
1 ) (y
n )} |ii
d3 p w~k (~
p) hf | T {(y
ain (~
p) a
out (~
p)T {(y

Z
Z
i
3
1 ) (y
n )}(x)
|ii
=
d p w~k (~
p) 0 lim
d3 x eipx 0 hf | T {(y
x
Z

Z
3
ipx

0 lim
d xe
0 hf | (x)T {(y1 ) (yn )} |ii
x +

Z
Z
i
3
(y
1 ) (y
n )} |ii
=
d p w~k (~
p) 0 lim 0 lim
d3 x eipx 0 hf | T {(x)
|
{z
}
x
x +
Z

M=

(x)

Z
Z
Z
i
=
d3 p w~k (~
p) d(x0 ) 0 d3 x eipx 0 (x)
Z
Z
Z


i
3
=
d p w~k (~
p) d4 x eipx 02 (x) (x)02 eipx
Z
~ 2 m2 )eipx . Using this and integrating by parts
Note that eipx solves the KG equation, so 02 eipx = (

14

yields (noting the surface terms are eliminated by the Riemann-Lebesgue Theorem),
Z
Z
i
~ 2 + m2 )(x)
d3 p w~k (~
p) d4 x eikx (02
M=
Z
Z
i
~ 2 + m2 )(x)

d4 x eikx (02
Z
Therefore,
1 ) (y
n )} |k; ii = i
hf | T {(y
Z

(y
1 ) (y
n )} |ii
d4 x eikx ( 2 + m2 ) hf | T {(x)

On the RHS we have a time-ordered correlation function of the form we started with. We can follow the
same procedure to pull another particle from the in-state. After we remove all of the in-state particles, we
can begin removing out-state particles by a related procedure.
By induction, we obtain the LSZ reduction formula for interacting massive real scalar fields:

hp1 pn |q1 qm i =

Z Y
n 
i=1

Y

m 
i iqi xi 2
i ipj yj 2
4
2
2
1 ) (x
n )(y
1 ) (y
m )} |i

d xi
e
d yj
e
(xi + m )
(yj + m ) h| T {(x
Z
Z
j=1
4

This result is strange for a few reasons. For instance, note that we have to multiply by the KG equation
for each initial or final particle. However, we associate the initial and final particles with essentially-free
particles (up to the wavefunction renormalization), so we could reasonably expect the KG equation to vanish
for each of these particles. It turns out thats exactly what happens. The only time that the LSZ reduction
formula doesnt return zero is when there are poles in the RHS time-ordered correlation function for each
and every asymptotic particle. If the time-ordered correlation functions lacks these poles, then the LSZ
reduction formula asserts |ii never becomes |f i.
Also, it turns out the result is valid beyond single-particle excitations of the vacuum. This includes some
of the bound states illustrated in Figure 2. Despite lacking a fundamental field, there are ways to create
appropriate field combination thatwhen plugged into the LSZ reduction formulayield information about
those bound states. The theory knows about all of its asymptotic states, we just have to ask it the appropriate
questions.

15

7. Calculating the Wavefunction Renormalization

Lets return to Equation one more time. It implies,

|pi =
h| (x)



Z h| in (x) |pi

We now know that both (x)

and in (x)

Z
+
t=

d4 y ret (x y) h| j(y) |pi

create single-particle states. Hence, when we apply the KG

t=

equation to both sides, we get

Z
0=

d4 y 4 (x y) h| j(y) |pi = h| j(x) |pi

for all x. Therefore,

|pi =
h| (x)


Conceptually, in (x)

Z h| in (x) |pi

acts like a free field so it only operates on single-particle states. Meanwhile, (x)
t=

generally operates on multiparticle states as well. We can view the wavefunction renormalization as a sort of

re-weighting to take this difference into account. After all, the interacting field (x)
is in some sense losing
some of its potency on single-particle states to its interactions with multiparticle states.
When we originally wrote L (and L0 for that matter), we implicitly chose a normalization for our kinetic term:
Lkinetic = 21 . While every other term in a Lagrangian L comes with a generic variable multiplying it,
we will always choose a kinetic term with coefficient 21 . Unsurprisingly, kinetic terms dictate how particles
move in our theory such that altering that coefficient will have implications on particle propagation.16
Therefore, the wavefunction renormalization Z is extremely important in ensuring the physicality of our
theory.
The final thing I want to do in these notes is derive an expression for the wavefunction renormalization Z

16 Although, it affects propagation in a different sense than the mass shift. The details are technical and best saved until
we have better machinery. Roughly speaking, theres an object called the propagator that measures the likelihood a particle
propagates from one momentum to another. The scalar propagator is proportional to Z/(p2 + m2 ). Therefore, the mass shift
dictates the location of the propagators pole while the wavefunction renormalization dictates the intensity of the pole.

16

that we can use in the future. Note that,

h| in (x) |pi =

d3 q
eiqx h| a
in (~q) |pi
(2)3 (2q )

d3 q
eiqx h| a
p) |i
in (~q)
ain (~
(2)3 (2q )

d3 q
eiqx h| [
ain (~q), a
in (~
p)] |i
(2)3 (2q )

=
=

= eipx

Thus,

Z h| in (x) |pi =

Zeipx

Meanwhile,
x

|pi = h| eiP
h| (x)

iP x

(0)e
|pi

|pi eipx
= h| (0)

Combining these elements into Equation gives us,

|pi =
h| (0)

|pi. Well develop the means to do so in the

Therefore, Z is calculable if we know how to calculate h| (0)
coming weeks. I reiterate that both Z and the mass shift m2 m2 m20 are important in ensuring the
quantum field theory that we are using is physically sensible. They are elements of an important procedure
called renormalization. At the lowest order of perturbation theory, Z = 1 and m2 = 0. This is because
it takes two interactions in order to self-interact: a particle must first radiate, then subsequently reabsorb
that radiation. As a result, Z and m2 become nontrivial beyond leading order.

17