Vous êtes sur la page 1sur 55

1

Path Integrals in Quantum Mechanics

M3R Project in Mathematical Physics


Nicolas Martinez Robles

Supervisor: Prof Y Chen

June 2005

Imperial College
London
2

Path Integrals in Quantum Mechanics

Index Page

Introduction 3
Outline 4

1) Historical Motivation 5
2) Further Tools from Quantum Mechanics 7
a) Review from the Calculus of Variations and Classical Mechanics 7
b) The Time Translation Operator 8
b) Completeness Relations 9
3) Path Integral Formulation of Quantum Mechanics 10
4) Recovering Limits 17
a) The Principle of Least Action from Path Integrals 17
b) The Method of Steepest Descent 19
c) The Schrödinger Equation from Path Integrals 19
5) Solving Problems in Quantum Mechanics using Path Integrals 22
a) The Free Particle 22
b) The Harmonic Oscillator 24
c) A Counter Example and Effective Action 27
d) Perturbation Methods and the Born Series 28
e) The Scattering Matrix 29
f) Introduction to Feynman Rules 34
g) Coulomb Scattering and Rutherford’s Formula 36
h) Euclidean Formalism 39
i) An Electron in an Electromagnetic Field 41
j) The Aharonov-Bohm Effect 43
6) The Need of a Quantum Theory of Fields and Conclusion 47

Appendix 49

References 51
3

Introduction

Path integrals represent a third different way of doing quantum mechanics.


As we will point out several times in the project, the path integral approach of
quantum mechanics is equivalent to wave mechanics.
However, many courses in quantum mechanics do not include some topics in path
integrals because they represent a more difficult way to solve easy problems (e.g. free
particle, harmonic oscillator…) and are thus left for a course in quantum field theory.

One may then ask: what is the purpose of including path integrals in a course of non-
relativistic quantum mechanics? The answer is that path integrals offer a considerably
much more intuitive way of approaching the subject.


quantisation
Lagrangian Mechanics Path Integral Formulation

classical limit

 


quantisation
Hamiltonian Mechanics Schrödinger Wave Mechanics

classical limit

Normally, quantum mechanics is presented from the wave mechanics approach which
is a quantisised version of Hamiltonian mechanics through the operators of position
and momentum. Then the classical limit is seen through the Hamilton-Jacobi equation
and the WKB method.

What if we now decide to bridge quantum mechanics from the Lagrangian


formulation of dynamics instead of the Hamiltonian version? This would lead to the
path integral formulation. However, Hamiltonian mechanics and Lagrangian
mechanics are equivalent, so it would be sensible to expect path integrals to be
equivalent to the wave mechanics, and this is precisely the case.
The classical limit can also be recovered from path integrals.
4

Outline

In this project, we will prove these claims but we will not adopt such a restrictive use
of classical mechanics, in order to fully build path integrals we will make use of the
concept of a Hamiltonian in classical physics. However, we will emphasize the use of
the Lagrangian in quantum mechanics instead of the use of the Hamiltonian.
The equivalence between these two formulations of mechanics allows us to do this.

Later we will solve some standard problems from quantum mechanics using path
integrals to show the parallelism between these two methods.

It has been an objective to try to put results from different sources in a new light, by
clarifying proofs and creating a coherent set of examples and applications which are
related to each other and of increasing complexity. To be self contained we have
included all the necessary proofs from analysis in the Appendix and a review of
several tools from classical and quantum mechanics

First we set up path integrals and show the classical limit in two different ways, a
heuristic argument is given in (4), whereas a more mathematical argument is given in
(5) by proving the following

Path Integrals ⇒ Schrödinger's Equation ⇒ Hamilton-Jacobi ( → 0).

Then the first application is the free particle (6a), which will come up in more
sophisticated problems such as the harmonic oscillator (6b), effective action (6c) and
for the perturbation and scattering processes of chapter 6 in general.
The choice of applications has arisen naturally from the rich variety and difficulty of
examples that can be solved using path integrals.

Finally, the last part is concerned with the breaking down of the quantum mechanics
in a relativistic framework, and brief hints of how this problem can be solved by the
use of path integrals.
5

1 Historical Motivation
Back in 1933, the founding days of quantum mechanics, Dirac observed that the
action plays a dominant role in classical mechanics (he regarded the Lagrangian
formulation to be much more fundamental that the Hamiltonian one) but that it was
completely ignored in the emerging quantum theory. In the language that we will use
in this paper, he considered that the propagator “corresponds” to exp(iS / ) where S is
the classical action. Indeed this had been a fruitful approach in the classical limit since
it connects the Schrödinger, Hamilton-Jacobi (see (2)) and conservation of probability
equations.
To see this link, set
ψ = ψ (r, t ) = a (r, t )eiS ( r , t ) /  = aeiS / 
in the time dependent Schrödinger equation:
∂  2 2 iS / 
i (aeiS /  ) = − ∇ (ae ) + V (r, t )aeiS / 
∂t 2m
which gives
∂S  2 ∇ 2 a 1 2  1 ∂a  2  ∇a 
− + − ∇S − V + i   + ∇ S+ ⋅ ∇S  = 0
∂t 2m a 2m  a ∂t 2m m a 
both, the real part and the imaginary part have to be zero hence
∂S 1 2 ∂S 2 ∇2a
+ ∇S + V = +H = ≠0
∂t 2m ∂t 2m a
the quantum version of the Hamilton-Jacobi equation, note that it is not zero, and
1 ∂a 1 2 1 ∇a ∂a 2  ∇S 
+ ∇ S+ ⋅ ∇S = + ∇ ⋅  a2  = 0,
a ∂t 2m m a ∂t  m 
the law of conservation of probability. From here, the WKB method follows.
Certainly this indicated that the approach of using exp(iS / ), or the action S for that
matter, was on a promising track.

Following Dirac’s suggestion, in 1948 Richard Feynman successfully derived a third


formulation (the first ones being of course the matrix and wave formulations) of
quantum mechanics. The main idea behind his approach was that the propagator could
be written as the sum of all possible paths, not just the classical one, that connect the
initial and final points. Each of the paths are now equally important and they
contribute exp(iS / ) to the propagator.
6

This idea went beyond Dirac’s first speculation because it said that a quantum particle
takes all the paths, and the amplitudes are added according to quantum mechanical
rules that were known.

The concept of path integration was not properly understood at the time and this
resulted in Feynman’s paper being denied publication in the Physical Review!

As it is often the case in physics, path integrals do not really add something new to the
known results of quantum mechanics, in fact, they are normally very difficult to
compute. Their real power resides in quantum field theory and in the fact that they
provide an intuitive and more natural way to look at quantum mechanics.

Finally, it is interesting to note two applications.


A quantum theory of finance [see Baaquie] can be derived from path integration and a
certain kind of path integral known as the Polyakov path integral (along with
conformal fields) leads to string theory [see Polchinski], a candidate explanation of
quantum gravity.
7

2 Further Tools from Classical and Quantum Mechanics


It is not the place to include a very dense explanation, only the main results are stated
and the reader is referred to Abers, Feynman, Goldstein, Landau and Merzbacher.

a) Brief Review from the Calculus of Variations and Classical Mechanics


Finding the stationary values (either maximum or a minimum) of an integral like
tb

I := ∫ dtf ( y, y )
ta

(with appropriate boundary conditions) leads to δI = 0 and the Euler-Lagrange


equations
∂f d ∂f
− = 0.
∂y dt ∂y
If
1
f = L(qi , qi ) = mqi2 − V (qi )
2
where L is the Lagrangian and set I := S (termed the action) then δS = 0 (principle of
least action)
tb
1 
δS = δ ∫ dt  mqi2 − V (qi )  = 0
ta  2 

∂L d ∂L
0= − = −V '(qi ) + mqi ,
∂qi dt ∂qi
implies Newton’s equations. A system with such a Lagrangian is called conservative
because it conserves the energy (the only case we treat here) which is defined as
E = ∑ pi qi − L

and E is also the Hamiltonian H which can be written


pi2
H= + V (qi )
2m
where pi = mqi is the momentum.

H satisfies Hamilton’s equations


∂H ∂H
= qi = − p i .
∂pi ∂qi
The equations of motion can be recovered from the Hamilton-Jacobi equation
8

∂S
+ H = 0.
∂t

b) The Time Translation Operator


Define the time translation operator U (t , t0 ) as the operator that relates the state of a
system at two different times
| ψ (t )〉 = U (t , t0 ) | ψ (t0 )〉.
The time translation operator has the following composition rule
U (t2 , t1 )U (t1 , t0 ) = U (t2 , t0 ),

and we must have U (t , t ) = I therefore U (t , t0 )† = U (t , t0 ) −1 = U (t0 , t ), i.e. U is unitary.

Differentiate U (t , t0 ) with respect to t and let t → t0 , define

lim d
t →t0 U (t , t0 ) = −iK (t0 )
dt
K represents what would happen if we make a small time translation.
Since U is unitary
d d
0= I =tlim
→t0
U (t , t0 )†U (t , t0 )  = −i  K (t0 )† − K (t0 ) 
dt dt
hence K is self-adjoint.

The purpose of introducing U (t , t0 ) is because the dynamics of a quantum system is

contained within it. To see this we must first relate U (t , t0 ) to the Hamiltonian.
The transition to quantum from classical mechanics demands that
 qi , p j  = iδij
PB

giving the Schrödinger equation


d
i | ψ〉 = H | ψ〉
dt
write this as
d
i U (t , t0 ) | ψ〉 = H (t )U (t , t0 ) | ψ(t0 )〉.
dt
However we must have
d
i U (t , t0 ) = H (t )U (t , t0 ),
dt
9

since ψ (t0 ) is an arbitrary state. But as we have shown U (t0 , t0 ) = I so our previous

equation can be written in integral form


t
i
U (t , t0 ) = I − ∫ dt '( H (t ')U (t ', t0 )).
 t0

General dynamical interest gives us that the Hamiltonian should be a constant of the
motion, hence
U (t , t0 ) = e− iH ( t −t0 ) /  .
c) Completeness Relations
In Dirac bracket notation the completeness relations can be written as
〈 p | p '〉 = δ( p − p ') 〈 q | q '〉 = δ(q − q ')

eipq /  e − ipq / 
〈 q | p〉 = 〈 p | q〉 = .
2π 2π
For instance, the first equation of the second line says that the momentum eigenstate
is a plane wave in the coordinate representation.

By normalization, the following hold


∞ ∞

∫ dq | q〉〈 q | =
−∞
∫ dp | p〉〈 p | = 1.
−∞

These will later prove to be vital to path integrals.


10

3 Path Integral Formulation of Quantum Mechanics

Consider a particle moving in one dimension (for simplicity, but we can generalize to
any number of dimensions in a fairly straight forward way), and we take the
Hamiltonian to be of the usual form
p2
H = H ( p, q ) = + V (q ), (0)
2m
i.e. a constant of the motion, which gives dynamical interest. We will not set  = 1 yet.
Suppose we have a particle in a position q at time t = 0 then the fundamental question
to ask is: what is the probability amplitude that it will be at some other position q ' say
at a later time t = T ?
From our discussion above we know that we can solve this using the time-translation
operator in the following way. Suppose the initial state is
| ψ(0)〉 = | q〉 ,
denote by A the amplitude that the state evolves in time and going to the state
| ψ(T )〉 = | q '〉.
Then A can be written as
A = 〈 q ' | ψ (T )〉 ≡ K (q ', T ; q,0) = 〈 q ' | e −iHT /  | q〉.
K is known as the propagator from the initial state to the final state, and it is
independent of the origin of time, i.e.
K (q ', T + t ; q, t ) = K (q ', T ; q,0).
The idea now is to derive an expression for this amplitude in the form of a summation
(or integral in the case of the continuum) over all possible paths between the initial
and final points. By using this process what we are doing actually is deriving path
integrals from quantum mechanics, but what Feynman did was to derive path integrals
differently and then he proved that they were equivalent to the standard formulation
of quantum mechanics.
By the composition rule we know that the time evolution in the above amplitude can
be separated into two smaller time evolutions, i.e.
A = 〈 q ' | e − iHT /  | q〉 = 〈 q ' | e −iH (T − t1 ) /  e− iHt1 /  | q〉. (1)
The key idea now comes from quantum mechanics, and it is to bring a well known
identity operator as follows
11

∞ ∞

∫ dq | q〉〈q | =
−∞
∫ dp | p〉〈 p | = 1
−∞

obviously we are interested in the first integral at the moment, so we introduce it in


the amplitude
∞ ∞
A = 〈 q ' | e − iH (T − t1 ) /  ∫−∞ dq1 | q1〉〈q1 | e 1 | q〉 = −∞∫ dq1K (q ',T ; q1, t1 ) K (q1, t1, q,0). (2)
− iHt / 

  
=1

This says that if a process can occur a number of ways, the amplitudes for each of
these ways add, it is the rule for combining amplitudes. When a particle goes from q
to q ', then the particle must be somewhere during the journey at an intermediate time
t1 say; so labelling that corresponding point q1 , we compute the amplitude for

propagation via the point q1 [this is the product of the propagators in (2)] and integrate

over all possible intermediate positions.

For N intermediate points, divide the time interval T in equal spaces as ε = T / N .

Figure 2.1: Discretization of time

Then the propagator becomes


A = 〈 q ' | (e − iH ε /  ) N | q〉
∞ ∞
= 〈 q ' | e −iH ε /  ∫ dqN −1 | qN −1 〉〈 qN −1 | e −iH ε /  ∫ dqN − 2 | qN − 2 〉
−∞ −∞

∞ ∞
×〈 qN − 2 |  ∫ dq2 | q2 〉〈 q2 | e −iH ε /  ∫ dq1 | q1 〉〈 q1 | e −iH ε /  | q〉
−∞ −∞


= ∫ dq  dq
−∞
1 N 〈 q ' | e − iH ε /  | qN −1 〉〈 qN −1 | e− iH ε /  | qN − 2 〉  〈 q2 | e −iH ε /  | q1 〉〈 q1 | e − iH ε /  | q〉


 N −1  N
= ∫ ∏ j ∏

−∞  j =1
dq 
j =0
K q j +1 , q j (3)
12

where we have defined q0 = q and qN = q '. Note however that the initial and final
positions are not integrated over. Hence the amplitude is the integral of the amplitude
of all N-legged paths, as illustrated below

Figure 2.2: Sum over paths

If we leave aside the mathematical justifications, we can see that when N becomes
large, this clearly tends to a sum over all possible paths of the amplitude for each path.
A= ∑A
paths
path ,

where

 N −1  N −1

∑ = ∫−∞  ∏ dq j  Apath = ∏ K qk +1 , qk .
paths j =1  k =0

Let us now evaluate the propagator K in the expression. We have defined it to be


K q j +1 , q j = 〈 q j +1 | e− iH ε /  | q j 〉

= 〈 q j +1 |1 − iH ε /  + O(ε 2 ) | q j 〉

= 〈 q j +1 | q j 〉 − iε / 〈 q j +1 | H | q j 〉 + 〈 q j +1 | O(ε 2 ) | q j 〉. (4)

The first term is just a delta function



dp j
∫ 2π e
ip j ( q j +1 − q j ) / 
〈 q j +1 | q j 〉 = δ ( q j +1 − q j ) = . (5)
−∞

The second term in (4) requires a little bit more of caution to be evaluated.
Assume as we have said before that the Hamiltonian is of the form (0) then evaluate
each of the two terms
pˆ 2 ∞ ∞
p 2j
〈 q j +1 | | q j 〉 = ∫ ∫ dpdp '〈 q j +1 | p '〉〈 p ' | | p〉〈 p | q j 〉
2m −∞ −∞ 2m
13

ip ' q j +1 /  − ipq j / 
∞ ∞
e p 2j e
= ∫ ∫ dpdp '
−∞ −∞ 2π 2m
〈 p ' | p〉
2π
∞ ∞
dpdp ' p 2j
= ∫∫
−∞ −∞ 2π
exp 
 i /  ( p ' q j +1 − pq )
j 

2m
δ( p − p ')


dp j p 2j
= ∫ 2π
−∞
exp ip j / (q j +1 − q j ) 
2m
. (6)

In the first line, p̂ operates to the right. Similarly

 q + qj 
〈 q j +1 | V (qˆ ) | q j 〉 = V  j +1  〈 q j +1 | q j 〉 , (7)
 2 
where V (qˆ ) operates to the left. This last expression is asymmetric between q j and q j +1.

Define q j = (q j +1 + q j ) / 2 then

〈 q j +1 | V (qˆ ) | q j 〉 = V (q j )δ(q j +1 − q j )

dp j
= ∫ 2π exp ip
−∞
j / (q j +1 − q j )  V (q j ).

The exact choice of q j +1 or q j does not matter in the continuum limit, which we shall

take later.
The middle term of (4) is then

 p j 
∞ 2
dp j
〈 q j +1 | H | q j 〉 = ∫−∞ 2π  j j +1 j   2m + V (q j ).
exp  ip /  ( q − q )   (8)
 
An alternative way to explain this can be done in two lines

 pˆ 2 ∞ ∞
 p 2j 
〈 q j +1 |  + V (qˆ )  ∫ dp j | p j 〉〈 p j | q j 〉 = ∫ dp j  + V (q j +1 )  〈 q j +1 | p j 〉〈 p j | q j 〉
 
 2m  −∞
  −∞  2m 
=1

dp j  p 2j

 ip ( q − q ) / 
= ∫  + V (q j +1 )  e j j +1 j

−∞ 2π  2m 
if we remember that p̂ operates to the left and V (qˆ ) to the left.
Finally combining (4) and (8) we have

 p j 
∞ 2
dp j iε ∞ dp j
K q j +1 , q j = ∫ exp ip j / (q j +1 − q j )  − ∫ exp ip j / (q j +1 − q j )   + V (q j ) 
−∞ 2π  −∞ 2π  2m 
+ 〈 q j +1 | O ( ε 2 ) | q j 〉
14

Now we are going to neglect the terms O(ε 2 ) this is called the rule of discarding the

O(ε 2 ) terms, and it is one of the main topics in putting Feynman path integrals in a
more rigorous framework.

dp j  iε  p 2j  
K q j +1 , q j = ∫  exp ip j / (q j +1 − q j )  − exp ip j / (q j +1 − q j )   + V (q j )  
−∞ 2π 
   2m  

dp j  iε  p 2j  
= ∫−∞ 2π  j j +1 j  1 −   2m + V (q j ) 
exp  ip /  ( q − q ) 
  

dp j
= ∫ 2π exp ip
−∞
j / (q j +1 − q j )  exp  −iε / H ( p j , q j ) 


dp j
= ∫ 2π exp iε / ( p (q
−∞
j j +1 − q j ) / ε − H ( p j , q j ))  (9)

where in the third line we have approximated the integrand to an exponential function.
In the amplitude there are N such factors, so by suggestively setting
q j := (q j +1 − q j ) / ε we can write
N −1 ∞
 N −1 dp j   iε N −1 
Apath = ∏ K q j +1 , q j = ∫−∞  ∏  exp  ∑ ( p j q j − H ( p j , q j )) . (10)
j =0 j = 0 2π    j =0 
The propagator becomes
∞ N −1
K= ∫ ∏ dq A
−∞ j =1
j path


 N −1  ∞  N −1 dp j   iε N −1 
= ∫−∞  ∏
 dq j  ∫  ∏  exp  ∑ ( p j q j − H ( p j , q j )) . (11)
j =1  −∞  j = 0 2π    j =0 
There are N momentum integrals, one for each interval, and N − 1 positions integrals,
one for each intermediate position. Remember that q j has no special meaning in the

computations that follow.


Another key step here towards the continuum limit is that as N → ∞, ε → 0 in such a
way that εN → T − 0, then the sum inside the exponential approximates to an integral
over all functions p = p(t ) and q = q(t ). Define

 N −1 dp j   N −1 
Dp (t ) :=  ∏  and Dq(t ) :=  ∏ dq j 
 j = 0 2π   j =1 
then
15


iT 
K = ∫ Dp (t ) Dq (t )exp  ∫ dt ( pq − H ( p, q )) 
−∞ 0 

iT 
= ∫−∞ Dp (t ) Dq (t )exp  ∫ dtL(q, q ) 
0 

i 
= ∫ Dp(t ) Dq(t )exp   S[q(t )] 
−∞
(12)

this result is known as the phase-space path integral or Hamiltonian Path Integral.
It is better to think of (12) just as a notation for the more precise (11).
We can carry out the integration with respect to p in (11), write it as
∞ N −1
 N −1  ∞ N −1 dp j  iε N −1  p 2j  
K = ∫ ∏ dq j exp  −iε ∑ V (q j )  ∫ ∏ exp  ∑  p j q j − 
 2m  
−∞ j =1  j =0 −∞ j = 0 2π   j = 0 
the p integrals are uncoupled Gaussians, i.e.

dp j  iε  p 2j   m
∫−∞ 2π    j j 2m   = 2πiε exp(iεmq j /(2))
2
exp  p 
q −
  
we have been completely careless about the convergence of the integral, but it is not
the aim here to establish such details.
Applying this procedure to the propagator gives
∞ N −1
 iε N −1  N −1 m
K= ∫−∞ ∏ dq j exp  −

∑ V ( q j ) ∏ 2 π iε
exp(iεmq 2j /(2))
j =1  j =0  j =0

 m 
N / 2 ∞ N −1
 iε N −1  mq 2j 
= 
 2πiε 
∫∏ j  ∑
dq exp  

− V ( q j  .
)  (13)
−∞ j =1  j =0  2  
In a more compact notation of the continuum we have
∞ N / 2 ∞ N −1
 m 
∫−∞ Dq(t ) = N →∞  2πiε  ∫ ∏ dq
lim
j
−∞ j =1


i 
K= ∫ Dq(t )exp   ∫ dtL(q, q ) 
−∞

∫ Dq(t )e
iS [ q ( t )] / 
= , (14)
−∞

where S is again the classical action, but (14) should be viewed as a notation for the
more precise expression (13) as N → ∞. This integral is known as the configuration
space path integral or Lagrangian Path Integral and is due to Feynman.
16

The integration measure Dq(t ) is defined as the indicated limit. The propagator is
given by the integral over all possible paths, weighted by a quantity whose phase is
the value of the action along each path.
In order to link wave functions with propagators we have to realize that if we are
given a wave function ψ (qa , ta ) at an initial time ta , then using the propagator the

corresponding wave function at final time tb is



ψ (qb , t b ) = ∫ dq
−∞
a K (qb , t b , qa , ta )ψ (qa , t a ).

We have considered only the case when the Hamiltonian is of the form (0), however
(14) tends to be correct even when the Hamiltonian has another form (example later).

What Feynman and Hibbs tried to do, was to postulate path integrals as a complete
substitute to the Schrödinger equation and non-relativistic quantum theory in their
book Quantum Mechanics and Path Integrals, all this had been hinted in Feynman’s
PhD thesis The Principle of Least Action in Quantum Mechanics.

However there are cases (rare, yet important) where the path integral approach is just
not right (example later). This is the reason why we have derived path integrals from
the Hamiltonian.

Before proceeding to the computation of path integrals in quantum problems we will


find the classical limit of path integrals as  → 0.
17

4 Recovering Limits
a) The Principle of Least Action from Path Integrals
According to (14) the propagator is

∫ Dq(t )e
iS [ q ( t )] / 
K= ,
−∞

a superficial look at this equation tells us two surprising facts.


The first fact is that when a particle goes from one position to another, it does take all
possible paths between these two positions. And the second one, not so easy to
reconcile with daily experience is that all possible paths contribute a little, that is if we
have a classical path S [qc ] and a non-classical path S [qnc ], we find that the first and

the second paths are both complex numbers of unit magnitude exp iS [qc ]/  and

exp iS [qnc ]/  respectively.


Why should the classical path be more important than any other path?
Consider two neighbouring paths q1 and q2 , as shown in Figure 3.1 and they both

contribute to the path integral. Let q2 (t ) = q1 (t ) + η(t ), where ηis small.


Expand in terms of Taylor series

δS [q ]
∫ dt η(t ) δq(t ) + O(η ).
2
S [q2 (t )] = S [q1 (t ) + η(t )] = S [q1 ] +
−∞

The functional derivative means in the notation of the calculus of variations


δq (t )
= δ(t − t ')
δq '(t )
where the last delta is a Dirac delta function.
As we have said above, both paths contribute exp iS[q1 ]/  and exp iS [q2 ]/  to the path
integral, so their contribution is

i  i  ∞
δS [q1 ] 
A = exp  S [q1 ] + exp   S [q1 ] + ∫ dt η(t ) + O ( η2 )  
     −∞ δq (t )  

i  i ∞ δS [q1 ]  
= exp  S [q1 ]  1 + exp  ∫ dt η(t ) 
     −∞ δq (t )  

where we have omitted as usual the O(η2 ) terms.


18

Figure 3.1: Neighbouring paths

This means that the difference in phase between the two paths, which determines the
interference between the two contributions, is

1 δS [q1 ]

 −∞
dt η(t )
δq (t )
.

Hence the smaller the value of Planck’s constant (it is very small indeed),
 = 1.0545887... × 10−27 erg − sec

= 6.582173... × 10−16 eV − sec


the larger the phase difference between two given paths.
If the paths are very close together, so that the difference in actions is extremely small,
for sufficiently small  the phase difference will still be large, and on average
destructive interference occurs.
There is one exceptional path for which this argument takes a different view, it is the
path that extremizes the action, and we know from classical mechanics that this has to
be the classical path qc (t ). This has the following property S [qc + η] = S [qc ] + O(η2 ).

Figure 3.2: Interference of paths


19

Thus the classical path and a very close neighbour will have actions which differ by
much less than any two other but equally close paths, see Figure 3.2.

Hence for fixed closeness of two paths, paths near the classical path will on average
interfere constructively (small phase difference) whereas for other paths, the
interference will be on average destructive (large phase difference).
Therefore, as expected, when the action is much larger than Planck’s constant, the
important contribution to the path integral comes from the region where the path that
extremizes the path integral is, i.e. the motion is dominated by the principle of least
action that is stationary (i.e. the classical principle of least action), the Euler-Lagrange
equations follow and all the equations of motion of classical dynamics follow from
this point.

b) The Method of Steepest Descent


As  → 0 in
∞ ∞
i 
∫−∞ Dq(t )exp   ∫ dtL(q, q )  = −∞∫ Dq(t )e
iS [ q ( t )] / 
K=

the integral is dominated by the paths of stationary phase, that is


δS
=0
δq (t )
which gives the classical equations of motion (Euler-Lagrange), c.f. (2.a).

c) The Schrödinger Equation from Path Integrals


So far we have not used Schrödinger’s Equation in the process of building a
formalism for path integrals. We are now going to try to recover the time dependent
version of the Schrödinger’s Equation.
Schrödinger’s Equation is a differential equation hence it determines infinitesimal
changes in the wave function. Consequently, in order to derive it, we only have to
examine the infinitesimal form of the transition amplitude. Consider the explicit form
of (13) but in the case of infinitesimal ε, then

 m 
1/ 2  iε  m  q − q 2  q j + q j −1   
K (qb , tb = ε, qa , ta = 0) =   exp    j j −1
 − V    
 2πiε     2  ε   ε   
20

and we have said before the transition amplitude is the propagator which gives the
propagation of the wave function in the following way

ψ ( q, ε) = ∫ dq 'U (q, ε, q ',0)ψ(q ',0).
−∞

Substituting the transition amplitude in the wave function one has


1/ 2 ∞
 m   im iε  q + q '  
∫ dq 'exp  2ε (q − q ')
2
ψ ( q, ε) =   − V  ψ (q ',0),
 2πiε  −∞   2  
perform the change of variable z = q '− q, which gives
1/ 2 ∞
 m   im iε  z 
∫ dz exp  2ε z
2
ψ ( q, ε) =   − V  q +   ψ (q + z ,0).
 2πiε  −∞   2 
We have to make some qualitative reasoning to proceed with this integral. We have
said that ε is an infinitesimal, hence if z is large then the term
im 2
z
2 ε
would lead to rapid oscillations and all such contributions will average out to zero.
Hence the dominant contribution will come from the region of integration
1/ 2
 2πε 
0≤ z ≤  ,
 m 
where the change in the first exponent is of order of unity. The interest in that now we
can power expand in Taylor series, and since we are interested in the infinitesimal
behaviour, we can keep terms up to order ε.
ψ ( q, ε) =
1/ 2 ∞
 m   im   iε  z 2 
∫ dz exp  2ε z
2
=   1 − V  q +  + O(ε )  ψ (q + z ,0)
 2πiε  −∞    2 
1/ 2 ∞
 m   im   iε  z 2  1 2 3 
∫ dz exp  2ε z
2
=   1 − V  q +  + O(ε )   ψ(q,0) + zψ '(q,0) + z ψ ''(q,0) + O ( z ) 
 2πiε  −∞    2  2! 
1/ 2 ∞
 m   im  iε z2 2 
∫ dz exp  2ε z
2 3
=    ψ (q,0) − V (q )ψ (q,0) + zψ '(q,0) + ψ ''(q,0) + O( z , ε ) .
 2πiε  −∞   2 
Let us now evaluate each term separately, using Gaussian integrals, we have
∞ 1/ 2
 im 2   2πiε 
∫−∞ dz exp  2ε z  =  m 

 im 
∫ dz exp  2ε z
2
z = 0
−∞ 
21

∞ 1/ 2
 im 2  2 iε  2πiε 
∫−∞ dz exp  2ε z  z = m  m  .

As before, we have not been concerned about the convergence of such integrals,
although in order to compute them we would need to use regularization tools
1/ 2

 im 2  lim

 im 2  lim  π  1/ 2
 2πiε 
∫−∞ dz exp  2ε z  =δ→0+ ∫−∞ dz exp  2ε z − δ  =δ→0+  im 
 δ − 2 ε 
=
 m 
 ,

and similarly for the other integrals, but again, it is not the purpose of this project to
establish such details, and we will just assume them.

To end this computation, we need to plug these results back in the wave function

 m 
1/ 2
 2πiε 1/ 2  iε iε  2πiε 
1/ 2

ψ ( q, ε) =       ψ ( q ,0) − V ( q ) ψ ( q ,0) +   ψ ''(q,0) + O(ε )  
2

 2πiε   m    2m  m   

iε iε
= ψ (q,0) + ψ ''(q,0) − V (q )ψ(q,0) + O (ε 2 )
2m 
write this as
iε   2 ∂ 2 
ψ (q, ε) − ψ (q,0) = − − 2
+ V (q )  ψ (q,0) + O(ε 2 ).
  2m ∂q 
Finally take the limit as ε → 0 and we have

∂  2 ∂ 2 
i ψ ( q, t ) =  − 2
+ V ( q )  ψ ( q, t )
∂t  2m ∂q 
which is the result we wanted. Thus, Schrödinger’s Equation is contained within the
path integral formulation. This result points at the fact that path integrals are
equivalent to the wave formulation of quantum mechanics.

Classical mechanics also follows from the path integral formulation.

Path Integrals ⇒ Schrödinger's Equation ⇒ Quantum version of Hamilton-Jacobi

∂S 1 2 ∂S 2 ∇2a
+ ∇S + V = +H =
∂t 2m ∂t 2m a
now set  → 0 to obtain the limit we wanted
∂S
+ H = 0,
∂t
the Hamilton-Jacobi equation.
22

5 Solving Problems in Quantum Mechanics using Path Integrals


Since we have already shown the connection between the quantum case and the
limiting classical case, we set  = 1 and we call the starting and final times of the
motion ta and tb .

a) The Free Particle


The simplest problem to solve is that of a free particle, that is when V (q ) = 0.
The action in this case is just
t b
1
S [q ] = ∫ dt mq 2
ta 2

where q is any path that goes from qa to qb . Expand q (t ) around some particular path

q0 (t ) which of course we can see that it will have to be the classical path.
For a free particle the Lagrangian depends only on q (t ) so
tb t
∂L(q0 ) 1 b ∂ 2 L(q0 )
S [q ] = S [q0 ] + ∫ dt δq (t ) + ∫ dt
 2
δq (t ) 2 .
ta ∂q 2 ta ∂q

There are no higher order terms in this expansion. Choose q0 (t ) to be the classical path

and by Hamilton’s principle δS = 0, the linear term vanishes, so we are left with
t
1 b ∂ 2 L(q0 )
S [q ] = S [q0 ] + ∫ dt δq (t ) 2 .
2 ta ∂q 2

The classical path in this case in a straight line, i.e.


qb (t − ta ) + qa (t − tb )
q0 (t ) =
tb − ta
so L has to be constant and
m (qb − qa ) 2
S [q0 ] = .
2 tb − t a
The expression for the propagator is then

∫ Dqe
iS [ q ]
K (qb , tb , qa , ta ) =
−∞

∞  t
i b ∂ 2 L(q0 ) 
= ∫ Dδq exp iS[q0 ] + ∫ dt 2
δq (t ) 2 
−∞  2 ta ∂q 
23

∞  1 tb ∂ 2 L(q ) 
= exp(iS [q0 ]) ∫ Dδq exp  ∫ dt 2
0
δq (t ) 2 
−∞  2 ta ∂q 

where δq (t ) is any path that starts from 0 at ta and returns to 0 at tb . Since the classical
action vanishes for such paths, the second factor is the propagator for paths with
qa = qb = 0 :

K (qb , tb , qa , ta ) = K (0, tb ,0, ta ) exp(iS [q0 ]).


We are left with the evaluation of the propagator on the RHS, which is not easy.
Let us now plug these results into the path integral (13) we have
N / 2 ∞ N −1 2
 m  mε N −1  q j +1 − q j 
K (0, tb ,0, ta ) = lim
N →∞  
 2πiε 
∫ ∏ dq j exp i ∑ ε 
2 j =0 
−∞ j =1 
N / 2 ∞ N −1
 m  mε
∫ ∏ dq (qN − qN −1 ) 2 + (qN −1 − qN − 2 ) +  + (q1 − q0 ) 
lim
= N →∞   j exp i
 2πiε  −∞ j =1 2

where we set q ' = qN = qb and q = q0 = qa , as the starting and final points. The integrals
are coupled Gaussian, which makes their evaluation much harder (not like before).
The result is (see Appendix)
N /2 ( N −1) / 2
lim  m  1  2πiε  im(qb − qa ) 2
K (0, tb ,0, ta ) = N →∞     exp
 2πiε  N m  2N ε
and the final propagator is
N /2 ( N −1) / 2
 m  1  2πiε  im(qb − qa ) 2
K (qb , tb , qa , ta ) =lim
N →∞     exp exp(iS [q0 ])
 2πiε  N  m  2Nε

m im (qb − qa ) 2
= exp . (15)
2πi (tb − ta ) 2 tb − ta
lim
Note that tb → ta K (qb , tb , qa , ta ) = δ(qb − qa ). It would be convenient to double check this

result using methods of standard quantum mechanics, which is done as follows:


K = 〈 q ' | e− iH ( tb − ta ) | q〉


∫ dp | p〉〈 p | q〉
− i ( tb − ta ) p 2 /(2 m )
= 〈q ' | e
−∞


2
∫ dpe b a 〈 q ' | p〉〈 p | q〉
− i (t − t ) p /(2 m )
=
−∞


dp
∫ 2π e
− i ( tb − ta ) p 2 /(2 m ) + i ( q ' − q ) p
=
−∞
24

m im (qb − qa ) 2
= exp ,
2πi (tb − ta ) 2 tb − ta

since the last integral is Gaussian. This result agrees with our previous path integral
evaluation (15) but it has been considerably easier to compute.

b) The Harmonic Oscillator


The next example will be the harmonic oscillator, where the Lagrangian is
1 2 1
L = L(q, q ) = mq − mω2 q 2 .
2 2
For simplicity here, set the initial time to be ta = 0 and like before, expand about the
classical path
t
1 b  ∂ 2 L(q0 ) ∂ 2 L(q0 ) 

2
S [q ] = S [q0 ] + dt  2
[δ 
q (t )] + 2
[δ q (t )]2 
2 0  ∂q ∂q 
t
mb
= S [q0 ] + ∫ dt ([δq (t )]2 − ω2 [δq (t )]2 )
20

since the second term is independent of the end points we can write
K (qb , tb , qa , 0) = K (0, tb , 0, 0) exp(iS [q0 ]). (16)
Hence we are left with the evaluation of the classical action, this is a standard problem
in classical dynamics

S [q0 ] = [(qa2 + qb2 ) cos ωtb − 2qa qb ].
2sin(ωtb )
Before proving this, let us finish the expression for the propagator K. Let y (t ) be any
path such that at the end points we have ya = yb = 0, then

 tb  my 2 mωy 2   lim  m ( N +1) / 2
K (0, tb ,0,0) = ∫
−∞
Dy exp i ∫ dt 
 0  2
−   = N →∞ 
2  

 2πiε 
IN .

We now need to compute the integral term I N as follows


∞ ∞
 m ∞ 2 2 2 
IN → ∫−∞ ∏ dy n exp iε 2 ∑ ( y n − ω yn ) 
n =1  n =1 


 m ∞ 
= − ∫ ∏ dyn exp iε ∑ ( yyn2 + ω2 yn2 ) 
−∞ n =1  2 n =1 
∞ ∞
∞ ∞
 m ∞ ∞ 
= ∫ ∏ dyn ∫−∞ ∏ dy k exp iε 2 ∑∑ ( yk M kn yn ) 
−∞ n =1 k =1  n =1 k =1 
25

where M is the operator


 d2 
M := −  2 + ω2  ,
 dt 
this will allow us to use the matrix simplectic approach to solve the problem. It is
important to note at this point that since M contains derivatives, it mixes up the
different functions ym (t ). For full rigour, the derivatives ought to be replaced by finite

difference, but we are interested in the case when N → ∞ so the answer will be correct.
Here M is a real symmetric matrix, this means that it can be diagonalized with an
orthogonal (real and unitary) matrix. The integral can be solved as follows. Consider
multiple integrals of the form
 
I = ∫ ∫ ∏ dqn exp  −∑ Aij qi q j 
n  ij 
if A is a real symmetric matrix it can be diagonalized. The new variables are yi and

∑A y
j
ij i = λ i yi

the transformation is effected by an orthogonal matrix, so its Jacobian is the unity, and
  π πn / 2
I = ∫ ∫ ∏ dyn exp  −∑ λ i yi2  = ∏ = .
n  ij  n λn det A
Applying this to the problem we have that as N → ∞
K0
K (0, tb ,0,0) =
det M
where K 0 is a constant independent of ω.
Hence we are left with evaluation of the determinant of the matrix. First we need to
find the eigenfunctions
 d2 
−  2 + ω2  un (t ) = λ nun (t )
 dt 
by setting λ n + ω2 = ωn we have

un (t ) = A sin(ωnt ) + B cos(ωnt )

using the boundary condition un (0) = un (tb ) = 0 we can find the value of ωn
2
 nπ 
ωntb = nπ ⇒ λ n =   − ω2
 tb 
hence the determinant of M is
26

∞ 
∞    ωt  
2
 nπ  sin ωtb
det M = ∏   − ω2  = K1 ∏ 1 −  b   = K1
  tb 
n =1     nπ  
n =1  ωtb

where again K1 is a constant that is independent of ω and the sin representation as an


infinite product is a well known theorem of complex analysis.
Combining both constants we have

ωtb
K (0, tb ,0,0) = K 2
sin ωtb

where once more K 2 is independent of ω, so (16) becomes

ωtb  imω 
K (qb , tb , qa ,0) = K 2 exp  [(qa2 + qb2 ) cos ωtb − 2qa qb ] .
sin ωtb  2sin(ωtb ) 
In order to determine this constant we have to let ω → 0 and this will give us the limit
of the free particle, which has propagator (15)
ω→ 0  im 2 2  m im (qb − qa ) 2
K (qb , tb , qa ,0) = K 2 exp  [(qa + qb ) − 2qa qb ] = exp
 2tb  2πi (tb − ta ) 2 tb − t a

hence, once we solve for the constant we arrive at


mω  imω 
K (qb , tb , qa ,0) = exp  [(qa2 + qb2 ) cos ωtb − 2qa qb ] . (17)
2πi sin ωtb  2sin(ωtb ) 
In order to fully prove (17), we need to derive the classical action and the identity for
the sine function using complex variables. The first problem is solved as follows.
The motion of a harmonic oscillator is
q (t ) = Q sin(ωt + z )
where Q and z are constants. The boundary conditions for this problem are
qb = Q sin(ωtb + z ) and qa = Q sin z
which give us
qb − qa cos ωtb
Q cos z = ,
sin ωtb
now compute the action
t t
mb 1 b

S [q0 ] = ∫ dt (q(t ) 2 − ω2 q (t ) 2 ) = mω2Q 2 ∫ dt cos(2ωt + 2 z )


20 2 0

1
= mω2Q 2 (sin(2ωt + 2 z ) − sin 2 z ).
4
27

In order to proceed we need to use the boundary conditions and the relation that
follows from the boundary conditions that we have shown above
2qb
Q 2 sin(2ωtb + 2 z ) = (qb cos(ωtb ) − qa )
sin(ωtb )
and finally we have
2qb
Q 2 sin(2 z ) = (qb − qa cos z )
sin(ωtb )
which is the result we wanted.

c) A Counter Example and Effective Action


Before proceeding into more applications of path integrals and the Euclidean
formalism, it is convenient to give a counter example where this method is wrong,
which was discovered by Lee and Yang in 1962.
Suppose we have the following Lagrangian
1 2
L= q f (q )
2
this describes a system with a velocity-dependent potential, using Hamilton’s
equation we can find the momentum
∂L
p=  (q )
= qf
∂q
and the Hamiltonian
1 2 1 p2
H = pq − L = q f (q ) =
2 2 f (q)
which is clearly not of the form (0). Substitute this in (14)

i 
K (qb , tb , qa , ta ) = ∫ Dq exp   S
−∞
eff 

where
 i 
Seff = ∫ dt  L(q, q ) − δ(0) ln f (q ) 
 2 
this is not the action of the path integrals, it is the effective action. So before
proceeding to compute path integrals, we have to make sure the Hamiltonian is of the
form (0), but we knew this to be true for the cases of the free particle and of the
Harmonic oscillator. In general, all physical situations have conservation of energy, so
(0) holds, but we have to be careful that this is always the case.
28

d) Perturbation Methods and the Born Series


Sometimes it is useful to expand the potential in (14), i.e. perturbation, set  = 1,

i 1 
∫ Dq(t )exp   ∫ dt  2 mq
2
K (qb , tb ; qa , ta ) = − V (q)  
−∞ 

 1   
∫ Dq(t )exp i ∫ dt 2 mq  exp  −i ∫ dtV (q ) 
2
=
−∞ 
and now expand as usual the exponential

 1 2 1 2 
= ∫−∞ Dq(t )exp i ∫ dt 2 mq  1 − i ∫ dtV (q) − 2 ( ∫ dtV (q) ) + 

= K 0 (qb , tb ; qa ,0) + K1 (qb , tb ; qa ,0) + K 2 (qb , tb ; qa ,0) +  (20)
This would only be possible if the variables inside the path integrals are ordinary
numbers, so that they all commute, and this is precisely the case. Later we will
establish matters of convergence in the Euclidean formalism. Let us now evaluate
each of the terms in the expansion of (20)

 1 2 m  im (qb − qa ) 2 
K 0 (qb , tb ; qa ,0) = ∫
−∞
Dq (t )exp ∫ 2

i dt 
mq 

=
2πitb
exp 
 2 tb


we know this from the free particle problem.
The second term needs to be evaluated explicitly, in (13) make the following
substitution (just for convenience) z = ( N + 1) / 2 so that we have shifted the integral

 m 
( N +1) / 2 ∞ N  iε N +1  mqn2 
K (qb , tb ; qa , ta ) =lim
N →∞   ∫−∞ ∏ dqn exp  ∑ − V (qn )   (21)
 2πiε  n =1   n =1  2 
note that the indices have changed. Discretize the integral, which gives us

 1 
K1 (qb , tb ; qa ,0) = −i ∫ Dq (t )exp i ∫ dt mq 2  ∫ dtV (q )
−∞  2 
( N +1) / 2 ∞
 m  N
 N +1 mqn2  N
= −i lim
N →∞  
 2πiε 
∫ ∏ n iε∑
−∞ n =1
dq exp  ∑ εV (qk ).
n =1 2  k =1

Now divide the sum inside the exponential and the product
k / 2 ∞ k −1
N
 m   k mqn2 
= −i lim
N →∞ ∑ ε 
k =1  2πiε 
∫ ∏ j iε∑
dq exp
n =1 2 

−∞ j =1

( N +1 − k ) / 2 ∞

 m  N
 k mqn2 
× ∫ V (qk )dqk   ∫ ∏ dq j exp iε∑ .
−∞  2πiε  −∞ j = k +1  n =1 2 
29

Here we have two propagators, separated by the integral of the potential with respect
to the coordinates qk .

Now take the limit as N → ∞ and identify both propagators


tb ∞
m  im (qb − q1 ) 2  m  im (q1 − qa ) 2 
K1 (qb , tb ; qa ,0) → ∫ dt1 ∫ dq1 exp   V ( q1 ) exp  
0 −∞ 2πi (tb − t1 )  2 tb − t1  2πit1  2 t1 
tb ∞
= ∫ dt1 ∫ dq1 K 0 (qb , tb , q1 , t1 )V (q1 ) K 0 (q1 , t1 , qa , t1 )
0 −∞

∞ ∞
= ∫ dt1 ∫ dq1K0 (qb , tb , q1, t1 )V (q1 ) K0 (q1, t1, qa , t1 ),
−∞ −∞

using the fact that K 0 (qb , tb , q1 , t1 ) = 0 for tb < ta .


We have to develop a pattern to evaluate the remaining terms of the potential.
Let us do a similar computation for the second term using the same way of breaking
up the factors
K 2 (qb , tb , qa , ta )

1  1  2
=− ∫
2 −∞
Dq (t ) exp i ∫ dt  mq 2   ∫ V (q )dt
 2 
( )
t t ∞ ∞
1b b
2 ∫0 ∫0 −∞
=− dt1 dt 2 ∫ dq1 ∫ dq2 K 0 ( qb , tb , q2 , t 2 )V ( q2 ) K 0 ( q2 , t 2 , q1 , t1 )V ( q1 ) K 0 ( q1 , t1 , qa ,0),
−∞

and for general terms we have


K n (qb , tb , qa , ta ) = K 0 (qb , tb , qa , ta )
∞ ∞
−i ∫ dq1 ∫ dt1 K 0 (qb , tb , q1 , t1 )V (q1 ) K 0 (q1 , t1 , qa , ta )
−∞ −∞

∞ ∞ ∞ ∞
1
− ∫ dq1 −∞∫ dt1 −∞∫ dq2 −∞∫ dt2 K0 (qb , tb , q1, t1 )V (q1 ) K0 (q1, t1, q2 , t2 )
2 −∞

×V (q2 ) K 0 (q2 , t2 , qa , ta ) +  (22)


At this moment, the pattern should become apparent. The nth order term has n factors
of the potential V (q ) always in the middle of n + 1 free propagators with the right

arguments. In front of each integral term, there is a coefficient of (−i ) n but we do not
have the factor of 1/ n! from the Taylor expansion because there are n! different ways
you can order the factors V (qk ). The same in other words, consider the second case
30

∞ ∞ ∞ ∞
1 1
∫ dt ' ∫ dt '' (V (t ')V (t '') + V (t ')V (t '') ) = ∫ dt ' ∫ dt '' ( θ(t '− t '')V (t ')V (t '') + θ(t ''− t ')V (t ')V (t '') )
2 −∞ −∞ 2 −∞ −∞
∞ ∞
= ∫ dt1 ∫ dt2 θ(t1 − t2 )V (t1 )V (t2 )
−∞ −∞

and so on for higher order terms. An easy way to visualize this process is to think of
the propagator as a different kind of sum over paths, i.e. the particle may travel freely
from qb to qa in time tb to 0, or it can travel from qb to q1 in time tb to t1 , interact in those

locations, then travel again from q1 to q2 in time t1 to t2 , interact again and move on any
number of times and so on. These are known as the Born series.

Figure 6.1: The Born series

Summarizing, write down


1. the free propagators
2. the potential −iV [q(t )] in between then
3. integrate over all the coordinates and time
4. final step is to add up all the amplitudes with different number of interactions

e) The Scattering Matrix


This method of perturbations can be applied to the S matrix. We know that if we are
given a wave function ψ(qa , ta ) at an initial time ta , then using the corresponding wave

function at final time tb is



ψ(qb , t b ) = ∫ dq
−∞
a K (qb , t b , qa , ta )ψ (qa , t a ).

Substitute the Born series in this expression and let the potential depend on time, i.e.
31


ψ(qb , t b ) = ∫ dq
−∞
a K 0 (qb , t b , qa , ta )ψ (qb , t b )

∞ ∞ ∞
−i ∫ dt ∫ dq ∫ dqa K 0 (qb , t b , q, t )V (q, t ) K 0 (q, t , qa , ta )ψ(qb , t b ) +  (23)
−∞ −∞ −∞

Under the assumption that the series converges, the terms that have been omitted
modify the last K 0 to the full propagator K, so that we can leave them out having an

exact integral equation for the wave function ψ.


Assume that at the limit ta → −∞ ψ is free (i.e. plane wave φ with potential zero).
Then in that case the term in the first line of (23) is also a plane wave because of the
free propagation of ψ(qa , t a ), hence we may write
∞ ∞
ψ(qb , t b ) = φ(qb , t b ) − i ∫ dt ∫ dq K 0 (qb , t b , q, t )V (q, t ) ψ(q, t ).
−∞ −∞

Let us now plug the final wave function into the Schrödinger equation
1 2 ∂
∇ ψ(qb , t b ) + i ψ(qb , t b ) = V (qb , t b ) ψ(qb , t b )
2m ∂tb
remember that we have dropped Planck's constant, and that the Laplacian is taken
with respect to the coordinates. The potential can be dropped when we insert the free
wave, so that this can be written in terms of K 0 as

1 2 ∂
∇ K 0 (qb , t b , q, t ) + i K 0 (qb , t b , q, t ) = iδ(qb − q )δ(tb − t )
2m ∂tb

where the term δ(tb − t ) has been included, but it can be dropped according to the way
we have built path integrals in this paper. This is a Green's equation for the
Schrödinger equation above, where K 0 is the Green function.
When we set up the experimental conditions before a scattering process, we assume
that the particle is free at t = −∞, which is what we have been doing, then the particle
scatters, and is free again at t = ∞, which is what we need to consider now. This must
seem contradictory, because a free particle (has a definite energy and momentum) is
governed by a plane wave, then this plane wave spreads over all space and time, the
issue is that this spreading also includes the centre of the interaction potential V, so
that the particle can never be free (have zero potential again).
Essentially, this is translated into a problem of switching the potential on and off
slowly so that we can have a free particle again in the distant future.
32

In order to be able to solve this problem, we need to use the adiabatic hypothesis, and
the reason why the potential should not be turned on and off too quickly is because it
would imply that the time dependence of V results in the scattering centre emitting or
absorbing energy, and this must not happen. This implication is seen through Fourier
transformations.
There are two possible ways to proceed now.
We could either consider the "initial" condition that ψ is a plane wave and then
assume that V → 0 for large negative t but keeping ta even further in the past, or we

could consider the "final" condition that ψ becomes free for large positive t and
keeping tb even further in the future.

Suppose we take the first option, and we denote by ψ + (qb , tb ) a wave which was free at

t = −∞ and this means that we use the 'retarded' propagator K 0 (q, t , q ', t ') which is zero

for t ' > t. The plane wave is denoted ψ −∞ (qa , ta ).

If we had decided to do this the other way around, we would have used ψ − (qa , ta )

(check this) consisting of a wave which becomes free at t = ∞, and an 'advanced'


propagator K 0 (q, t , q ', t ') which is zero for t ' < t. The corresponding plane wave is now

denoted by ψ ∞ .

The plane wave ψ ∞ has definite energy and momentum, and we are interested in the
amplitude of detecting a final particles with these qualities, i.e. the amplitude of
detecting the plane wave ψ ∞ .
Let S denote this amplitude, which we call the scattering amplitude. S is the overlap of
the wave functions

∫ dq ψ (qb , tb )ψ + (qb , tb )
*
S= b ∞
−∞

∞ ∞

∫ dq ∫ dq ψ
*
= b a ∞ (qb , tb ) K 0 (qb , tb , qa , ta )ψ −∞ (qa , ta )
−∞ −∞

∞ ∞ ∞ ∞
−i ∫ dqb ∫ dq ∫ dqa ∫ dt ψ*∞ (qb , tb ) K 0 (qb , tb , q, t )V (q, t ) K 0 (q, t , qa , ta )ψ −∞ (qa , ta )
−∞ −∞ −∞ −∞

∫ dq ψ
*
= b ∞ (qb , tb )φ(qb , tb )
−∞

−i (same integral), (24)


33

where φ is of course a plane wave. If we use box normalisation then


1
ψ −∞ (q, t ) = exp(i ( pa q − Eat ))
τ
1
ψ ∞ ( q, t ) = exp(i ( pb q − Ebt ))
τ
where pa and pb are the initial and final momenta, τ is the volume of the box.

The initial and final energy is related to the momenta by E = p 2 /(2m). Since the

volume of the box is arbitrary we can set it to be (2π)3 for convenient simplifications
later. Using these formulas in the first integral of (24) we arrive at
∞ ∞
1 i ( pb qb − Eb tb )
∫ dqb ψ ∞ (qb , tb )φ(qb , tb ) = ∫ dq
*
b e φ(qb , tb )
−∞ −∞ τ

1 1
= ∫ dq
−∞
b
(2π) 3/ 2
ei ( pb qb − Eb tb )
(2π)3 / 2
eiEb tb

since Et is equivalent to qp hence


∞ ∞
1
∫−∞ dqb ψ (qb , tb )φ(qb , tb ) = (2π)3 −∞∫ dqb e b b = δ( pb ).
* ip q

The scattering matrix or S matrix hence becomes


Sba = δ( pa − pb ) − i (integral) (25)

One of the elements of the matrix Sba is S, which we have termed the scattering

amplitude. The term δ( pa − pb ) corresponds to no interaction giving momentum


conservation and a unit S matrix. However the integral in (25) which can be seen in
the third line of (24) represents the interactions. Let us call it A
∞ ∞ ∞ ∞
A = −i ∫ dqb ∫ dq ∫ dqa ∫ dt ψ
*
∞ (qb , tb ) K 0 (qb , tb , q, t )V (q, t ) K 0 (q, t , qa , ta )ψ −∞ (qa , ta )
−∞ −∞ −∞ −∞

the amplitude that a given state ψ ∞ comes from a particular state ψ −∞ is represented by

A. This is an expression for the scattering amplitude in terms of the propagator and the
potential. Using Feynman rules, this can be translated into an easier language, but
this would take us beyond the scope of this paper. Instead a brief introduction will be
given here.

f) Introduction to Feynman Rules


A may be represented by the following diagram
34

Figure 6.2: Feynman Rule for Amplitude A.

It is of interest now to do the reverse process, how can we translate a diagram into an
expression for the scattering amplitude. This can be done by making the
correspondence

Figure 6.3: Feynman Rule for Scattering Amplitude.

also it is necessary to multiply by ψ −∞ and by ψ*∞ at the ends of the diagram and to
integrate over the two relevant spatial variables. The second term of the amplitude
then has the following Feynman rule

Figure 6.4: Feynman Rule for the Second Order Amplitude

with corresponding expression


∞ ∞ ∞ ∞ ∞ ∞

∫ dqa ∫ dq ∫ dt ∫ dq ' ∫ dt ' ∫ dqb ψ ∞ K0 (qb , tb , q ', t ')V (q ', t ') K 0 (q ', t ', q, t )V (q, t ) K 0 (q, t , qa , ta ).
2
A(2) = ( −i ) *

−∞ −∞ −∞ −∞ −∞ −∞
35

It is very clear that the Feynman rules are not really of much use in this case because
we are dealing with a non-relativistic theory, however in quantum field theory they
are essential.
These rules have been considered only in terms of coordinates, but for convenience,
they can be translated in terms of momenta. Let us make this translation.
Denote by Ω( p, t , p0 , t0 ) the amplitude that a particle with momentum p0 at time t0 be

later observed to have momentum p1 at time t1. Then


∞ ∞
Ω( p1 , t1 , p0 , t0 ) = ∫ dq0 ∫ dq1 exp(−ip1q1 ) K (q1, t1, q0 , t0 )exp(−ip0q0 ).
−∞ −∞
(35)

From here onwards, a vector notation is adopted. In a) the free particle propagator was
shown to be
3/ 2
 m   im (q 0 − q1 ) 2 
K 0 (q1 , t1 , q 0 , t0 ) =   exp  
 i (t1 − t0 )   2 t1 − t0 
Hence in this case, Ω0 can be written as
3/ 2 ∞
 m  ∞
 im (q 0 − q1 ) 2 
Ω0 (p1 , t1 , p 0 , t0 ) =  
 i (t1 − t0 ) 
∫ 0 −∞∫ 1
−∞
dq dq exp [ i (p 0 ⋅ q 0 − p1 ⋅ q1 ) ] exp 
 2 t1 − t0 
.

In order to evaluate this integral, we need to make the following change of variables
q = q 0 − q1 Q = q 0 + q1 p = p 0 − p1 P = p 0 + p1 ,
this simplifies the following expression
2(p 0 ⋅ q 0 − p1 ⋅ q1 ) = P ⋅ q + p ⋅ Q.
3
This transformation has Jacobian ( 12 ) = 81 , so that
3/ 2 ∞ ∞
1 α 
∫ dQ exp(ip ⋅ Q / 2) ∫ dq exp(iP ⋅ q / 2)e ,
2
i αq
Ω0 (p1 , t1 , p 0 , t0 ) =  
8  iπ  −∞ −∞

where α is defined as α := m /(2(t1 − t0 )). One has


∫ dQ exp(ip ⋅ Q / 2) = 8(2π) δp = 8(2π) δ(p


3 3
0 − p1 ),
−∞

hence the amplitude becomes


3/ 2 ∞
α
∫ dq exp(iP ⋅ q / 2 + iαq ).
3 2
Ω0 (p1 , t1 , p 0 , t0 ) = (2π) δ(p 0 − p1 )  
 iπ  −∞

Using the formula (see Appendix)


36


 b2  π
∫−∞
2
dx exp( − ax + bx + c ) = exp  + c
 4a  a
the integral is fully evaluated as
 iP 2 (t1 − t0 ) 
Ω0 (p1 , t1 , p 0 , t0 ) = (2π)3 δ(p 0 − p1 )exp  − .
 8m 
By the change of variable formulas we have P 2 = 4p 02 , so that

 ip 2 (t − t ) 
Ω0 (p1 , t1 , p 0 , t0 ) = (2π)3 δ(p 0 − p1 )exp  − 0 1 0  . (36)
 2m 
As it has been said, this propagator gives the amplitude for observing a particle with
momentum p1 at time t1 , given that is has been observed with momentum p 0 at time t0 .

This quantity has Fourier transform K 0 (q1 , t1 , q 0 , t0 ) given by the inverse of (35), which
gives when substituted in (36)
∞ ∞
1
K 0 (q1 , t1 , q 0 , t0 ) = ∫ dp1 −∞∫ dp0 exp(ip1 ⋅ q1 )Ω0 (p1 , t1 , p0 , t0 ) exp(−ip0 ⋅ q0 )
(2π)6 −∞

1 ∞  z2 
3 ∫
= dz exp 
i z ⋅ (q1 − q 0 ) − (t1 − t0 )  , (37)
(2π) −∞  2m 
this equation will be crucial in the discussion that follows.

g) Coulomb Scattering
In this paper, the Hydrogen atom has not been treated because it requires a lot of
transformations and techniques from path integrals that would take us too far from the
applications of quantum mechanics. The interested reader should check Kleinert.
However, even if Coulomb potentials are not treated here (in exact solution, that is),
we can study Coulomb scattering. The idea of introducing vector notation is because
the Coulomb potential is in 3D, denote it by V (q, t ).
The scattering amplitude in the first Born approximation was seen to be
∞ ∞ ∞ ∞
A = −i ∫ dq1 ∫ dq ∫ dq 0 ∫ dt ψ*∞ (q1 , t1 )(q b , tb ) K 0 (qb , tb , q, t )V (q, t ) K 0 (q, t , q a , ta )ψ −∞ (q 0 , t0 )(q a , ta ).
−∞ −∞ −∞ −∞

Let us now substitute K 0 from (37) and use the box normalisation wave functions for

ψ ∞ and for ψ −∞ discussed above, then


∞ ∞ ∞ ∞ ∞ ∞
i 1
A=− ∫ dq1 −∞∫ dq−∞∫ dq0 −∞∫ dt −∞∫ dz −∞∫ dz '
τ (2π)6 −∞
37

  p2   q2 
× exp  −i  pb ⋅ q1 − b t1   exp i  z ⋅ (q1 − q) − (t1 − t )  
  2m    2m 

 q '2   p 2 
×V (q, t )exp i  z '⋅ (q − q 0 ) − (t − t0 )   exp i  p a ⋅ q 0 − a t0   .
 2m   2m  

- Integration over the coordinates q1 and q 0 gives the delta functions

(2π)3 δ(pb − z ) and (2π)3 δ(z '− p1 ).

- Integration over the variables z and z ' eliminates the time terms t0 and t1 ,

i
A=− ∫ dqdtV (q, t )exp[i((pa − pb ) ⋅ q − ( Ea − Eb )t )]
τ −∞

where Ei = pi2 /(2m). Let us now introduce the Coulomb potential

Ze 2
V = V (q, t ) = ,
4πε0 r
and integrate over t, then

i Ze2 dq
A = − 2πδ( Ea − Eb )
τ ∫ r exp[i(pa − pb ) ⋅ q].
4πε0 −∞

Because of the singularity, this integral will not converge, hence a factor of e− γr needs
to be introduced and then we need to let γ → 0. Doing this process gives us

dq 4π 4π

−∞ r
exp[i (p a − pb ) ⋅ q] =
(p a − pb ) 2
= 2,
p
so that in the end
i Ze 2
A = − 2π δ( Ea − Eb ).
τ ε0 q 2
This is an amplitude (scattering amplitude which is used to compute the scattering
cross section σ) , what we want is the probability that a particle emerges with
2
momentum pb and this is given by A .

Once again, assuming box normalisation with volume τ we have that the probability
that a particle emerges with momentum between pb and pb + dpb is given by

2 τdpb
probability = A .
(2π)3
Suppose that the interaction lasts time T, then the number of particles # per second
emerging in this momentum range is given by
38

2
A dpb
#= τ .
T (2π) 2
How do we obtain the cross section σ ? Divide by the incident flux and integrate over
the final momentum. There are 1/ τ incident particles per unit volume each with
speed pa / m, so that the flux is given by pa /(τm) particle per second per unit area. The
cross section is obtained when we integrate
∞ 2
A τm τ
σ = ∫ dpb .
−∞ T pa (2π)3
2
The probability A involves the square of a delta function. This can be computed as

follows, proceed using the definition of the delta function


T /2 2
1
2π −T∫/ 2
2 lim
δ( Ea − Eb ) = T →∞ dt exp[i( Ea − Eb )t ]

2
lim sin[( Ea − Eb )T / 2] T
= T →∞ = δ( Ea − Eb )
π( Ea − Eb ) 2π

because we have lim 2 2


n →∞ sin ( nx ) /( nx ) = πδ( x ).

Summarizing, the cross section is


mZ 2e4 d 3 pb mZ 2e 4 1
σ=
4π2ε02 ∫∫∫ pa q 4 δ( Eb − Ea ) = 4π2ε02 ∫∫∫ p qa
4
pb2 dpb δ( Eb − Ea )d Ω

where Ω is not to be confused with momentum propagator.


Finally pa = (2mEa )1/ 2 and pb2 dpb = (2m3 Eb )1/ 2 dEb so that integrating over Eb gives

m 2 Z 2e 4 dΩ
σ=
4π2ε 02 ∫∫ q 4

where, because of the delta function pa = pb = p. Let θ be the angle between p a and pb

then q 2 = 4 p 2 sin 2 (θ / 2). Knowing that p = mw gives the differential cross section
2
d σ  Ze2  1
= 2  4
d Ω  8πε0 mv  sin (θ / 2)

an equation known as Rutherford’s formula.

h) Euclidean Formalism
So far we have computed integrals of the sort
39

∞  tb 
∫−∞ Dq exp i ∫ dtL(q, q ) 
 ta 
(26)

for (13) using the Gaussian integral



π
∫ dxe
− ax 2
=
−∞ a
assuming they were convergent. Let us manipulate these integrals, set
t = −iw (27)
we are rotating the time-integration contour by π / 2 so that it runs along the imaginary
axis. Hence (26) becomes
∞  tb  dq  
∞  itb  dq  
∫−∞ Dq exp ∫
 ta
dwL  q , i  ∫
 dw   −∞
= Dq exp  − ∫ dwLE  q,
 ita

 dw  
2
 dq  1  dq 
LE = − L  q, i =   + V (q)
 dw  2  dw 
where LE stands for the Euclidean Lagrangian. Note that we have

tb − ta = −iT

and that the “Euclidean” propagator would be (T > 0)

K E (qb , T , qa ,0) = 〈 qb | e− HT | qa 〉
essentially we have just decided to put the imaginary unit somewhere else, so our
problem is shifted to the following: if we have a particle in potential that does not
depend on velocity then

 T  dq  
K E (qb , T , qa ,0) = ∫ Dq exp  − ∫ dwLE  q,  (28)
−∞  0  dw  
where
2
1  dq 
LE = − L(iw) = m  + V (q ).
2  dw 
If we go back to our example of the free particle, we can now write its Euclidean
propagator as

m  m (qb − qa ) 2 
K E (qb , T , qa ,0) = exp  − .
2πT  2 T 
We can always find the real time propagator by analytic continuation back to the
original variables.
40

So far it seems that we have translated a problem into another without actually solving
it, or getting further knowledge of it. How could such a simple transformation like (27)
help us? The integral in (28) is real and has no convergence problem, also since the
integrand has a sharp peak we can normally get a good approximation by expanding
the integrand about its maximum.
Let us look at an application.

Let ψ n (q ) be the energy eigenfunctions, with eigenvalues En . The Euclidean


propagator can be written as
K E (qb , T , qa ,0) = 〈 qb | e− HT | qa 〉

= ∑ 〈 qb | ψ n 〉 e − EnT 〈ψ n | qa 〉 = ∑ ψ n (qa )ψ*n (qb )e− EnT . (29)


n n

It is known that all physical spectrums have a lower bound, hence all the energies can
be chosen positive by choosing the zero point of the energy suitably. This implies that
all the terms go to zero as T increases but the term with the ground state energy
survives the longest.
This fact will provide us with a way to extract the energy of the lowest state. Let us do
this for the harmonic oscillator, where we know that the spectrum is
 1
En =  n +  ω.
 2
Recall the real time propagator

mω  imω 
K (qb , tb , qa ,0) = exp  ( qa2 + qb2 ) cos(ωtb ) − 2qa qb  
2πi sin(ωtb )  
 2sin(ωtb ) 
in our new language, the Euclidean form is

mω  mω 
K E (qb , T , qa ,0) = exp  − ( qa2 + qb2 ) cosh(ωT ) − 2qa qb   .
2π sinh(ωT )  
 2sinh(ωT ) 
The leading term in the propagator is

lim mω − mωqa2 / 2 − mωqb2 / 2 −ωT / 2


T →∞ K E (qb , T , qa ,0) = e e e (30)
π
since the hyperbolic functions have the following property
1
lim
T →∞ sinh(ωT ) =Tlim→∞ cosh(ωT ) = eωT .
2
If we now compare (29) and (30) we have
41

mω − mωqa2 / 2 − mωqb2 / 2 −ωT / 2


∑ψ
n
n (qa )ψ*n (qb )e − EnT =
π
e e e as T → ∞

which gives eigenfunctions


1/ 4
 mω  2
ψ 0 (q) =   e− mωqa / 2
 π 
with corresponding eigenvalues
ω
E0 = .
2
This is a known result in quantum mechanics.
What can we deduce now from the exponential in the Euclidean propagator? First
expand it in Taylor series and then multiply this expansion by

1 1
lim
T →∞ = e −ωT / 2
sinh(ωT ) 2

which is in the square root. Each of the terms goes like


  1 
exp  −  n +  ωT 
  2 
which means that the spectrum is
 1
En =  n +  ω,
 2
for integer n, in agreement with what we knew.

i) An Electron in an Electromagnetic Field


We have mentioned in paragraph 2 that the path integral approach tends to be correct
even when the Hamiltonian is not of the form (0). Let us now give an example of such
a case. The spin interaction in the following example will be neglected.
Consider an electron in a general external electromagnetic field with magnetic
potential A(r, t ) and electrostatic potential eφ(r, t ), note that this quantity is a scalar.
It is a known fact from classical electromagnetism that the Hamiltonian is given by
1
H= (p + eA(r, t )) 2 − eφ(r, t ) (31)
2m
clearly, this Hamiltonian is not of the form (0) and it is not a trivial matter to see how
this problem can be solved by using path integrals. Can we use (14)? The answer is
yes, let us prove it.
42

The general phase space path integral is


∞ ∞  tb  1 
K (rb , tb , ra , ta ) = ∫ Dr ∫ Dp exp i ∫ dt  p ⋅ r − (p + eA(r, t )) 2 + eφ(r, t )   .
−∞ −∞  ta  2m  

It is now convenient to introduce a change of variables that simplifies the integrand


but that leaves the measures of path integration unchanged, that would be
p n ' = p n + eA(r, t ) then d 3 pn = d 3 pn '. The phase space becomes
∞ ∞  tb  p '2 
K (rb , tb , ra , ta ) = ∫−∞ −∞∫
Dr Dp 'exp  i ∫ dt
 ta 
 (p '− eA (r , t )) ⋅ 
r −
2m
+ eφ(r, t )  
 
∞  tb 

p2 
= ∫ Dr ∫ Dp exp i ∫ dt  p ⋅ r − − eA(r, t ) ⋅ r + eφ(r, t )  
−∞ −∞  ta  2m  
it is necessary now to complete the square and then integrate over the momentum
2
1 2 1  r  r 2 1 2 1 1
p ⋅ r − p =−  p + 

− =− ( p − mr ) + mr 2 = mr 2
2m 2m  2 ( − 2m )  4 ( − 2m )
1 1
2m 2 2

∞  tb 
K (rb , tb , ra , ta ) = ∫−∞ Dr exp i ∫ dtL(r, r ) 
 ta 
where L is the Lagrangian for a charge −e in a magnetic field
1 2
L(r, r ) = mr − eA(r, t ) ⋅ r + eφ(r, t ).
2
Even if (0) is not general, it does hold for an external electromagnetic field. Let us
now finish the computation of the propagator
∞  tb 
K (rb , tb , ra , ta ) = ∫−∞ Dr exp i ∫ dt ( L0 (r, r ) − eA(r, t ) ⋅ r + eφ(r, t )) 
 ta 
∞  tb   tb 
= ∫ Dr exp  i ∫ dtL0 (r, r )  exp  −ie ∫ dt ( A(r, t ) ⋅ r − φ(r, t )) 
 t   ta 
−∞  a 
∞  tb 
= ∫−∞ Dr exp(iS 0 (r ))exp  −ie ∫ ( A(r, t ) ⋅ dr − φ(r, t )dt ) 
 ta 
Let us take the following gauge transformation
A (r, t ) → A '(r, t ) = A(r, t ) + ∇Λ (r, t )

φ(r, t ) → φ '(r, t ) = φ(r, t ) − Λ (r, t )
∂t
using this new gauge, the propagator becomes
43

∞  rb  dr ∂ 
K '(rb , tb , ra , ta ) = ∫ Dr exp(iS0 (r )) exp  −ie ∫  A(r, t ) ⋅ dr − φ(r, t )dt − ∇Λ (r, t ) ⋅ − Λ (r, t )  
−∞  ra  dt ∂t  

∞  rb  d 
= ∫ Dr exp(iS0 (r ))exp  −ie ∫  A (r, t ) ⋅ dr − φ(r, t )dt − Λ (r, t )dt  
−∞  ra  dt  

= K (rb , tb , ra , ta )exp[ie(Λ (ra , ta ) − Λ (rb , tb ))].

At time ta the wave function of the electron is ψ (r, ta ) and at later time tb the wave

function is
ψ (r, tb ) = ∫∫∫ d 3ra K (rb , tb , ra , ta )ψ (r, ta ).

3

The most important fact is that if the wave function ψ (r, t ) satisfies Schrödinger’s
equation in the original gauge, then
ψ '(r, t ) = exp(−ieΛ (r, t ))ψ (r, t )
also satisfies Schrödinger’s equation in the new gauge. It might seem like the wave
function is gauge dependent, but there are no observable gauge dependent quantities.
A very important result has just been proved. Gauge transformations are allowed by
classical electromagnetism as long as magnetic and electrostatic potentials are
transformed with a factor that depends on t and r. The new wave function must also
be transformed from the old by a factor that depends on t and r, that is Λ (r, t ).

j) The Aharonov-Bohm Effect


With this new tool, we can now attack a problem that has no classical analogue.
It is called the Aharanov-Bohm effect.
Let us set up the following experiment in 2D. Suppose that a beam of electrons with
momentum k is moving in the y direction and hits an impenetrable plane at y = 0. The
beam can pass through two holes in the plane, on the z axis at x = ± d / 2.
Moreover, the electrons are detected on a screen located some distance y = L d
behind the plane.
44

The probability that the electrons strike the screen depends on the distance from the
centre- this is the two slit pattern.
Interference pattern cannot be produced if the splits are wide because the beam is not
precise (localized) enough. Yet, the approximation that the beam has definite energy
is not correct if the splits are narrow. Aharanov and Bohm computed the appropriate
length of the splits in their paper that gives the single slit diffraction pattern
modulating the double slit interference pattern.
When we have the optimal length it is safe to say that the wave function to the right of
the barrier is the sum of two spherical wavelets, each coming from one of the slits.
1 i ( kr1 −ωt ) 1 i ( kr2 −ωt )
ψ (r, t ) ∼ e + e , (32)
r1 r2

where r1 and r2 are the distances from the observation point to slits 1 and 2 respectively.

Denote by ra some point located on the plane at y = 0, and by rb some point on the

detecting screen at y = L.
Thus the wave function on the screen is determined by the that of the plane, that is

ψ (rb , tb ) = ∫∫∫ d ra
3
∫ Drexp(iS ) K (r , t , r , t )ψ(r ).
0 b b a a a (33)

3 −∞

The stationary points of the path integral are the classical paths given by (32), the
integrands are in phase only in some region close to the classical paths.
It was worked in their paper, that the two terms add up to the double slit interference
pattern (we assume this result)
45

2 4 πd sin θ
ψ (r, t ) = 2
cos 2
r λ
where as we have said, d is the distance between the slits, r the distance to the point
half way between them, λ the be Broglie wavelength and finally sin θ = z / L.

In order to use our results from the magnetic field we have to modify the experiment.
Change the interference so that before the electrons hit the screen they go through a
region where there is no magnetic field but where the potential A is not zero.
Experimentally speaking, this can be achieved by adding a very long cylindrical
solenoid, whose axis is parallel to the z axis, passing through the imaginary triangle
we get by connecting the two slits to the centre of the interference pattern on the
detecting screen.
For convenience, take the origin of the coordinates where the axis of the solenoid
intersects with the z = 0 plane. The magnetic field is constant inside the magnet but
zero outside. Adopting cylindrical coordinates we have
 B nˆ ρ ≤ R0
B= 0 z
0 ρ ≥ R0

 12 B0nˆ z × r ρ ≤ R0
A (r ) =  1 2 . (34)
 2ρ R0 B0nˆ φ ρ ≥ R0

Let us now consider the paths neighbouring one of classical paths (which is a straight
line from one of the slits to the detector, i.e. the observation point). We have shown in
the previous discussion that the contribution of that classical path to the propagator is
the contribution of the same path the magnetic field vanishes B = 0,

 rb 
contribution = exp(iS0 (r ))exp  −ie ∫ dr ⋅ A(r )  .
 r 
 a 
The key fact is now that this integral is the same for all the neighbouring paths of one
of the classical paths because B = ∇ × A = 0.
However the extra phase is not the same in the vicinity of the other classical path
since a path cannot be deformed into another path without intersecting the region
where the magnetic is not zero, B ≠ 0.
46

Before activating the magnetic field, consider the case when we have B = A = 0,
denote by ψ1 (r ) and ψ 2 (r ) the contributions to ψ(r ) from all the paths near each of the
two classical paths. We can use (32) to approximate these.
Let us now activate the magnetic field in the solenoid. The change is given by
   
ψ (r ) = ψ1 (r )exp  −ie ∫ dr ⋅ A(r )  + ψ 2 (r )exp  −ie ∫ dr ⋅ A(r ) 
 Γ   
 1   Γ2 
where Γ1 and Γ 2 are the classical straight lines from the source through one of the slits
to the detector in the screen (observation point). This equation can be re-written in a
more appealing way by factoring out one the phases:
 
ψ (r ) = exp  −ie ∫ dr ⋅ A(r )  [ ψ1 (r ) + ψ 2 (r ) exp(−ieϑ)]
 Γ 
 1 

ϑ := ∫ dr ⋅ A(r )
C

where the contour C is defined as C = Γ1 − Γ 2 . Because C is a closed contour that cuts


through the solenoid, the difference between the integrals is not zero. We are now in
position to apply Stokes theorem using the potential of (34) to compute the phase
difference ∆φ

∆φ = φ+ − φ− = eϑ = eπR02 B0 = eΦ
here Φ represents the magnetic flux through the contour.
The Aharanov-Bohm effect states that the interference pattern shifts as B is changed.
Normally ∆φ will not be zero, nor a multiply of 2π. This effect has been observed
experimentally.
We would expect from classical physics the orbit of a classical charged particle to be
unaffected by the magnetic (vector) potential in a region where the magnetic field is
zero. Yet in quantum theory, the potential has effects on the phase of a wave function
even when the classical paths are only through a region where the magnetic field
vanishes.
The Aharanov-Bohm effects needs a potential A in a region where the magnetic field
is non existent, however the gauge has remained untouched. It depends on the
magnetic flux through the contour, leaving the gauge quantity invariant.
A is measurable in quantum mechanics only up to a gauge transformation.
47

6 The Need of a Quantum Theory of Fields and Conclusion


It is a good moment to show the need of a quantum theory of fields, in non-relativistic
quantum mechanics we have E = p 2 /(2m) , adopting a 3D vector notation we have
2
U (t ) = 〈q | e− i ( p /(2 m )) t / 
| q0 〉

| q 0 〉 ∫∫∫ d 3 p | p〉〈p |
2
= 〈q | e − i ( p /(2 m )) t / 

= ∫∫∫ d 3 p 〈q | e −i ( p
2
/(2 m )) t / 
| p〉〈p | q 0 〉

3

= ∫∫∫ d 3 p e −i (p
2
/(2 m )) t / 
〈q | p〉〈p | q 0 〉
3

1
3 ∫∫∫
2
= d 3 p e −i (p /(2 m ))t /  eip ⋅q /  e − ip ⋅q0 / 
(2π)
3

1
3 ∫∫∫
2
= d 3 p e −i (p /(2 m ))t /  eip ⋅( q − q0 ) / 
(2π)
3
3/ 2
 m  2
=  eim (q − q0 ) /(2 t )
. (18)
 2πit 
Unfortunately, this expression is non-zero for all q and t, this means that a particle can
travel between any points in an arbitrarily short time. This clearly violates the
principle of causality, hence it could never be a relativistic theory. Unfortunately
again, this is problem is not solved by using a relativistic expression like

E= p 2 + m 2 , following the same procedure we have

p 2 + m2 / 
U (t ) = 〈q | e− it | q0 〉

1
(2π)3 ∫∫∫
p 2 + m2 /  ip ⋅ ( q − q0 ) / 
= d 3 p e − it e

3


1
∫ dp p sin( p q − q
p2 + m2 / 
= 2 0 )e −it
2π q − q 0 0

the evaluation of this integral is much more difficult, and the interested reader should
check Peskin and Schroeder (1995), who would then be referred to Gradshteyn and
Ryzhik (1980).
It suffices to say that it can be evaluated explicitly in terms of Bessel functions, and
that for q 2 t 2 (fortunately the case of physical interest because it is outside the
light-cone), we have the behaviour
48

q2 −t 2
U (t ) ∼ e − m (19)
which means that the propagation amplitude is small yet non-zero outside the light-
cone and again causality is violated.

One way to solve this is using quantum field theory, and a very fruitful approach to
building a quantum field theory (such as quantum electrodynamics) is using path
integrals which is how it was done originally by Feynman et al.

A main goal of this paper was to end in such a way that a chapter in path integrals in
quantum field theory could continue. Hopefully, the main tools of path integrals have
been studied, and from here onward a quantisised version of field theory can be
constructed.
49

Appendix
We need the following formula for the evaluation of the propagator of the free particle.

Claim:

∫ dxe
− x2
I := = π
−∞

Proof
Consider
∞ ∞

∫ dxe ∫ dye = ∫∫ dxdye− ( x


2 − x2 − y2 2
+ y2 )
I = .
−∞ −∞
2

Transform to polar coordinates x = r cos θ, y = r sin θ and dxdy = rdrd θ


2π ∞ 2π 2π
1 1
∫ ∫ d θdr (re ) = ∫ d θ[−e − r ]∞0 = ∫ d θ = π,
2 2
−r
I2 =
0 0 20 20

the result follows. 

Using the appropriate substitutions we have



π
∫ dxe
− ax 2
= ,
−∞ a

 b2  π
∫−∞
2
dx exp( − ax + bx + c ) = exp  + c .
 4a  a

Claim:
∞ ∞
i n πn  iλ 
∫ dx1  ∫ dxn exp iλ[( x1 − a) + ( x2 − x1 ) +  + (b − xn ) ] =
2 2 2
exp  (b − a) 2 
−∞ −∞ (n + 1)λ n  n + 1 
Proof (by induction).
Assume it is true for n, and show that it is true for n + 1.
∞ ∞

∫ dx1  ∫ dxn +1 exp iλ[( x1 − a) + ( x2 − x1 ) +  + (b − xn ) ]


2 2 2

−∞ −∞


i n πn  iλ 
n ∫
= dxn +1 exp  ( xn +1 − a ) 2  exp iλ (b − xn +1 ) 2 
(n + 1)λ −∞ n +1 
2 ∞
 i n πn   1 
= n 
 (n + 1)λ 
∫ dx
−∞
n +1 exp iλ 
n +1
( xn +1 − a ) 2 + (b − xn +1 ) 2 .

50

1 n+2 2
( xn +1 − a ) 2 + (b − xn +1 ) 2 = y − 2 y (b − a ) + (b − a ) 2
n +1 xn +1 − a = y n + 1

2
n+2 n +1  1
=  y− (b − a )  + (b − a ) 2 .
n +1  n+2  n+2
Let λ − ((n + 1) /(n + 2))(b − a ) = z so that the integral becomes

i n πn  n+2 2 iλ 2 i n +1πn +1  iλ 
n ∫
dz exp iλ z + (b − a )  = n +1
exp  (b − a ) 2  .
(n + 1)λ −∞  n +1 n+2  (n + 1 + 1)λ n + 2 

51

References

Classical Mechanics:
Classical Mechanics, Third Edition – Goldstein, Poole and Safko
Addison Wesley, 2002

Mechanics, Third Edition – L.D. Landau and E.M. Lifshitz


Pergamon Press, 1978

Mathematical Methods:
Complex Variables: Introduction and Applications, Mark J. Ablowitz and Athanassios
S. Fokas
Cambridge University Press, 1997

Path Integrals:
Handbook of Feynman Path Integrals – C. Grosche and F. Steiner
Springer Tracts in Modern Physics, 1998

Feynman Integral and Random Dynamics in Quantum Physics: A Probabilistic


Approach to Quantum Dynamics – Zbigniew Haba
Kluwer Academic Publishers, 1999

Path Integral Methods and Applications – Richard MacKenzie


Lectures given at Rencontres du Vietnam: VIth Vietnam School of Physics, 1999-
2000

Path Integral Methods – T. Kashiwa, Y. Ohnuki and M. Suzuki


Oxford Science Publications, 1997

Path Integrals in Quantum Mechanics, Statistics, and Polymer Physics, and Financial
Markets, Third Edition – Hagen Kleinert
World Scientific Publishing Company; 3rd edition (2004)
52

Path Integrals in Physics Volume 1: Stochastic Process & Quantum Mechanics – M.


Chaichian and A. Demichev
Institute of Physics Publishing; 1st edition (July 15, 2001)

Path Integrals in Physics Volume 2: Quantum Field Theory, Statistical Physics &
Other Modern Applications – M. Chaichian and A. Demichev
Institute of Physics Publishing; 1st edition (July 15, 2001)

Quantum Field Theory:


Field Theory: A Path Integral Approach – Ashok Das
World Scientific Lecture Notes in Physics

Quantum Field Theory – Lewis H. Ryder


Cambridge University Press, 1989

The Quantum Theory of Fields: Volume I Foundations – Steven Weinberg


Cambridge University Press, 1995

An Introduction to Quantum Field Theory – Michael Peskin and Daniel V. Schroeder


Addison Wesley, 1995

Path Integral Methods in Quantum Field Theory – R.J. Rivers


Cambridge Monographs on Mathematical Physics

Quantum Mechanics:
Quantum Mechanics – Ernest S. Abers
Addison Wesley, 2004

Quantum mechanics and path integrals – R.P. Feynman and A.R. Hibbs
McGraw Hill, 1965

Quantum Mechanics, Third Edition – Eugen Merzbacher


John Wiley & Sons, Inc., 1998
53

Quantum Mechanics: Nonrelativistic Theory – L.D. Landau and E.M. Lifshitz


Pergamon Press, 1977

Scientific Papers and Others:


Physikalisches Z. der Sowjetunion 3 (1933), 64 – P.A.M. Dirac

Rev. Modern Phys. 20 (1948), 267 – R.P. Feynman

Phys. Rev. 80 (1950), 440 R.P. Feynman

Phys. Rev. 115 (1959), 485 – Y. Aharonov and D. Bohm

http://scienceworld.wolfram.com/physics/Aharonov-BohmEffect.html

String Theory, Vol. 1 : An Introduction to the Bosonic String – Joseph Polchinski


Cambridge Monographs on Mathematical Physics, 1998

Quantum Finance: Path Integrals and Hamiltonians for Options and Interest Rates –
Belal E. Baaquie – Cambridge University Press, 2004
54
55

Vous aimerez peut-être aussi