D. L.

Rubin
September, 2011
These notes are based on Sakurai, 2.4, Gottfried and Yan, 2.7, Shankar 8 &
21, and Richard MacKenzies Vietnam School of Physics lecture notes (arXiv:quant-
h/0004090v1)
1 Path Integral
Suppose we have the propagator
K(x
f
, t
f
, x
0
, t
0
) = x
f
, t
f
| x
0
, t
0
=

x
f
| e
iH(t
f
t
0
)/
| x
0
_
We can just as easily take two steps
K(x
f
, t
f
, x
0
, t
0
) = x
f
, t
f
| x
0
, t
0
=

x
f
| e
iH(t
f
t
1
)/
e
iH(t
1
t
0
)/
| x
0
_
or we could divide the total time T into N steps, with = T/N. Then
K(x
f
, t
f
, x
0
, t
0
) = x
f
, t
f
| x
0
, t
0
=

x
f
| e
iH/
e
iH/
... | x
0
_
and then we could insert the identity everywhere along the path.
K(x
f
, t
f
, x
0
, t
0
) = x
f
|e
iH/
_
dx
N1
| x
N1
x
N1
|e
iH/
_
dx
N2
| x
N2
x
N2
| . . .
. . .
_
dx
1
| x
1
x
1
|e
iH/
| x
0

=
_
dx
N1
K(x
f
, t
f
, x
N1
, t
N1
)
_
dx
N2
K(x
N1
, t
N1
, x
N2
, t
N2
) . . .
. . .
_
dx
1
K(x
1
, t
1
, x
0
, t
0
)
The amplitude is the sum of all N-legged paths.
A =

paths
A
paths

paths
=
_
dx
1
dx
2
...dx
N1
, A
path
= K
x
N
,x
N1
K
x
N1
,x
N2
...
1
Lets consider the j
th
term.
K(x
j+1
, x
j
) =

x
j+1
| e
iH/
| x
j
_
Since is small we can expand the exponential and we have
K(x
j+1
, x
j
)

x
j+1
| 1 iH/ + O(
2
) | x
j
_
Then we can insert the identity
_
dp
j
| p
j
p
j
| and
K(x
j+1
, x
j
)
_
dp
j
_
x
j+1
| 1 iH

| p
j
_
p
j
| x
j

_
dp
j
x
j+1
| p
j
p
j
| x
j
i
_
x
j+1
| H

| p
j
_
p
j
| x
j

_
dp
j
x
j+1
| p
j
p
j
| x
j
i

(
p
2
j
2m
+ V (x
j+1
))x
j+1
| p
j
p
j
| x
j

_
dp
j
2
_
e
i(x
j+1
x
j
)p
j
/
i

(
p
2
j
2m
+ V (x
j+1
))e
i(x
j+1
x
j
)p
j
/
_

_
dp
j
2
e
i(x
j+1
x
j
)p
j
/
exp
_
i

(
p
2
j
2m
+ V (x
j+1
))
_

_
dp
j
2
e
i(x
j+1
x
j
)p
j
/
exp
_
i

H
_
Now we write (x
j+1
x
j
)/ = x
j
and we have
K(x
j+1
, x
j
)
_
dp
j
2
e
i x
j
p
j
/
e
i

H
There are N such factors in the amplitude so
A
path
=
_
N1

j=0
dp
j
2
exp
_
i

N1

j=0
( x
j
p
j
H(x
j
, p
j
))
_
Thats the amplitude for one path. Now integrate over all paths
K(x
N
, x
0
) =
_
N1

j=1
dx
j
_
N1

j=0
dp
j
2
exp i

N1

j=0
( x
j
p
j
H(x
j
, p
j
))
2
As N the sum becomes an integral over all time and we write
K(x
N
, x
0
) =
_
Dx(t)
_
Dp(t) exp i

_
T
0
dt( xp H(x, p))
This is the phase space path integral. If the Hamiltonian has the standard form
H =
p
2
2m
+ V (x) then we can integrate each of the terms in the sum
K(x
N
, x
0
) =
_
N1

j=1
dx
j
exp(i

N1

j=0
V (x
j
))
_
N1

j=0
dp
j
2
exp i

N1

j=0
( x
j
p
j

p
2
j
2m
)
We use
_

e
x
2
x
dx =
_

2
/4
where the above holds for pure imaginary if it is regarded as a limit, namely if
= a + ib, a > 0 it is the limit as a 0. This is what it looks like
_
dp
2
exp i

( xp
p
2
2m
) =
_
m
2i

e
i

m x
2
/2
Putting it all together
K =
_
N1

j=1
dx
j
exp
_
i

N1

j=0
V (x
j
)
_
N1

j=0
__
m
2i
exp i

m x
2
j
2
_
=
_
m
2i
_
N/2
_
N1

j=1
dx
j
exp i

N1

j=0
_
m x
2
j
2
V (x
j
)
_
The sum is an approximation of the action of a path passing through the points
x
0
, x
1
, x
2
, ...
K =
_
Dx(t)e
iS[x(t)]
is the conguration space path integral.
1.1 Free particle path integral
The conguration space path integral for a free particle is
K =
_
m
2i
_
N/2
_
N1

j=1
dx
j
exp
_
i

N1

j=0
_
m x
2
j
2
_
_
3
K =
_
m
2i
_
N/2
_
N1

j=1
dx
j
exp
_
i

N1

j=0
m
2
_
(x
j+1
x
j
)

_
2
_
K =
_
m
2i
_
N/2
_
N1

j=1
dx
j
exp
_
im
2
N1

j=0
((x
j+1
x
j
))
2
_
K =
_
m
2i
_
N/2
_
N1

j=1
dx
j
exp
_
im
2
_
(x
N
x
N1
)
2
+ (x
N1
x
N2
)
2
+ ... + (x
1
x
0
)
2
_
_
where x
0
and x
N
are initial and nal points. The integrals are Gaussian and can be
evaluated exactly but since they are coupled it aint pretty. Lets see if we can gure
it out. First lets dene y
i
= (
m
2
)
1
2
x
i
. Then
K =
_
m
2i
_
N/2
_
N1

j=1
_
2
m
_1
2
dy
j
exp
_
i
_
(y
N
y
N1
)
2
+ (y
N1
y
N2
)
2
+ ... + (y
1
y
0
)
2
_
K =
_
m
2i
_
N/2
_
2
m
_
(N1)/2
_
N1

j=1
dy
j
exp
_
i
_
(y
N
y
N1
)
2
+ (y
N1
y
N2
)
2
+ ... + (y
1
y
0
)
2
_
Lets do the y
1
integration rst.
_
dy
1
exp i
_
(y
2
y
1
)
2
+ (y
1
y
0
)
2
_
=
_
dy
1
exp i(y
2
2
+ y
2
0
+ 2y
2
1
2y
1
(y
2
+ y
0
))
=
_
dy
1
exp i(

2y
1
(y
2
+ y
0
)/

2)
2
exp i(y
2
2
+ y
2
0
) exp
i
2
(y
2
+ y
0
)
2
=
_
dz

2
exp
1
i
(z)
2
exp
i
2
(y
2
y
0
)
2
=
_
i
2
exp
i
2
(y
2
y
0
)
2
Next we do the y
2
integration.
_
i
2
_
dy
2
exp
1
i
_
(y
3
y
2
)
2
+
1
2
(y
2
y
0
)
2
_
=
_
i
2
_
dy
2
exp
1
2i
_
3y
2
2
2y
2
(2y
3
+ y
0
)
_
exp
1
i
(y
2
3
+
1
2
y
0
)
2
4
=
_
i
2
_
dz

3
exp
1
2i
_
z
2
(2y
3
+ y
0
)
2
/3
_
exp
1
i
(y
2
3
+
1
2
y
0
)
2
=
_
i
2
_
2i
3
exp
1
2i
_
(2y
3
+ y
0
)
2
/3
_
exp
1
i
(y
2
3
+
1
2
y
0
)
2
=
_
(i)
2
3
exp
1
3i
(y
3
y
0
)
2
It looks like it goes to
_
(i)
N1
N
exp
_

1
Ni
(y
N
y
0
)
2
_
Putting it all together we have
K = lim
N
_
m
2i
_
N/2
_
2
m
_
(N1)/2
(i)
N1
2
N
1
2
exp
_

1
Ni
(y
n
y
0
)
2
_
The answer is
K = lim
N
_
m
2i
_
N/2
1

N
_
2i
m
_
(N1)/2
e
im(x

x)
2
/2N
K = lim
N
_
m
2iN
_
1/2
e
im(x

x)
2
/2N
Since N = T we have
K =
_
m
2iT
_
1/2
e
im(x

x)
2
/2T
which is of course the same as we calculated directly. Now on further investigation
we see that
K =
_
m
2iT
_1
2
exp i
m
2
_
x
N
x
0
T
_
2
T =
_
m
2iT
_1
2
exp(
i

_
T
0
L
cl
dt)
Cute huh. The coordinate space path integral for the free particle, the sum of the
action through every possible point in space, reduces to simply the classical action.
The propagator reduces to two factors, one being the phase exp
i

S
cl
5
1.2 Harmonic oscillator path integral
The coordinate space path integral for the harmonic oscillator is
K =
_
m
2i
_
N/2
_
N1

j=
dx
j
exp i
N1

j=0
_
m x
2
j
2

1
2
m
2
x
2
j
_
Now lets write
x(t) = x
cl
(t) + y(t), dx = dy, x = x
cl
+ y
Then

_
m x
2
2

1
2
m
2
x
2
j
_
=
_
m x
2
cl
2

1
2
m
2
x
2
cl
_
+

_
L
x
|
x
cl
y +
L
x
|
x
cl
y
_
+

_
m y
2
2

1
2
m
2
y
2
_
Lets look at the middle term and convert the sum to an integral
_
dt
_
L
x
y +
d
dt
_
L
x
y
_

d
dt
L
x
y
_
=
L
x
y|
t
N
t
0
+
_
dt
_
L
x

d
dt
L
x
_
y = 0
The rst term is zero because y(
0
) = y(t
N
) = 0. So
K = lim
N
_
m
2i
_
N/2
exp
i

S
cl
_
N1

j=1
dy
j
exp i
N1

j=0
_
m y
2
2

1
2
m
2
y
2
j
_
The PI over y is independent of the endpoints. It is zero at each end. It will depend
only on the total time T
K = exp
i

S
cl
Y (T), Y (T) =
_
m
2i sin T
_
1/2
and if
x(t) = Acos(t) + Bsin t, x
N
= Acos T + Bsin T, x
0
= A
Then B = (x
N
x
0
cos T)/ sin T
S
cl
=
_ _
1
2
m x
2

1
2
mx
2
_
dt
6
=
1
2
_
dt
_
m(Asin t + Bcos t)
2
m
2
(Acos t + Bsin t)
2
_
=
1
2
_
dt m
2
(A
2
sin
2
t + B
2
cos
2
t 2ABsin t cos t
(A
2
cos
2
t + B
2
sin
2
t + 2ABsin t cos t))
=
1
2
_
dt m
2
((B
2
A
2
) cos 2t 2ABsin 2t)
=
m
4
((B
2
A
2
) sin 2t + 2ABcos 2t) |
T
0
=
m
4
((B
2
A
2
) sin 2T + 2AB(cos 2T 1))
=
m
4
((x
N
x
0
cos T)
2
x
2
0
sin
2
T)
sin 2T
sin
2
T
+2x
0
(x
n
x
0
cos T)(
cos 2T 1
sin T
))
=
m
4
((x
2
N
+ x
2
0
cos 2T 2x
N
x
0
cos T)
2 sin T cos T
sin
2
T
+2x
0
(x
n
x
0
cos T)(
cos 2T 1
sin T
))
=
m
4
((x
2
N
+ x
2
0
cos 2T 2x
N
x
0
cos T)
2 cos T
sin T
+2x
0
(x
n
x
0
cos T)(
cos 2T 1
sin T
))
=
m
4
((2x
2
N
cos T + 2x
2
0
cos T 4x
N
x
0
)
1
sin T
=
m
2 sin T
((x
2
N
+ x
2
0
) cos T 2x
N
x
0
)
1.3 Principle of Least Action
Consider the conguration space path integral
K =
_
Dx(t)e
iS[x(t)]/
.
It says that a particle going from initial to nal position and time takes all possible
paths. The classical path is included but it gets no special mention. Every path
has precisely unit magnitude. The contributions from the classical path and the
totally wild path are the same. It turns out that the amplitudes interfere with each
7
other in a very special way. Consider two neighboring paths x(t) and x

(t) and let

x

(t) = x(t) + (t), with (t) small. Then we can write the action
S[x

] = S[x + ] = S[x] +
_
dt(t)
S[x]
x(t)
+ O(
2
)
The contribution of the two paths to the PI is
A e
iS[x]/
_
1 + exp
i

_
dt(t)
S[x]
x(t)
_
The phase dierence between the two paths is
1

_
dt(t)
S[x]
x(t)
. Smaller larger phase
dierence. Even paths that are very close together will have large phase dierence
for small and on average they will interfere destructively.
This is true except for one exceptional path, that which extremizes the action,
namely the classical path x
c
(t). For this path
S[x
c
+ ] = S[x
c
] +O(
2
).
The classical path and a close neighbor will have actions which dier by much less
than two randomly chosen but equally close paths.
If the problem is classical (action ), paths near the classical path will on
average interfere constructively (small phase dierence) whereas for random paths
the interference will be on average destructive. Classically, the particles motion is
governed by the principle that the action is stationary.
8