Solution to Non-Relativistic Path Integral Problem

This is a problem regarding non-relativistic single particle quantum mechanics and

path integrals.
2
Pop
Imagine a Hamiltonian for a particle moving in one dimension given by Hop = 2M
+µ
(µ can be thought of as a chemical potential). For this Hamiltonian, the Schroedinger
equation for the one-particle wave function φ(x) takes the form Oop φ(x) = 0, where
∂ 1 ∂2
Oop = i ∂t + 2M ∂x2
− µ.

1. Compute hxf , tf |xi , ti i and show that

h i
p2
dp ip(xf −xi )−i
Z
2M
+µ (tf −ti )
hxf , tf |xi , ti i = e . (1)
2π

We actually more or less did this in the first or second lecture of the year. The
idea is to recognize that the states above are in the Hamiltonian basis and that
they evolve according to |x, ti = eiHop t |xi, where |xi is the Schroedinger basis state.
Thus,

hxf , tf |xi , ti i = hxf |e−iHop (tf −ti ) |xi i

dp
Z
= hxf |pihp|e−iHop (tf −ti ) |xi i
2π h 2 i
dp p
Z
−i 2M +µ (tf −ti )
= hxf |pihp|e |xi i
2π h i
p2
Z
dp ipxf −i 2M +µ (tf −ti ) −ipx
= e e e i
2π h 2 i
dp ip(xf −xi )−i 2M p
Z
+µ (tf −ti )
= e (2)
2π

2. Check that P (xf − xi , tf − ti ) = −ihxf , tf |xi , ti iθ(tf − ti ) is a Greens function of the

Schroedinger equation operator: i.e.

Oop P (x − x0 , t − t0 ) = δ(x − x0 )δ(t − t0 ) . (3)

This is shown by first noting that

h i h i
p2 p2
p2 p2
(" # )
dp ip(x−x0 )−i dp
Z Z
2M
+µ (t−t0 ) ip(x−x0 )−i 2M
+µ (t−t0 )
Oop e = +µ − −µ e
2π 2π 2M 2M
= 0. (4)
 Thus,
h 2
i
dp ip(x−x0 )−i 2M p
+µ (t−t0 ) d
Z
0 0
Oop P (x − x , t − t ) = −i e i θ(t − t0 )
2π h 2 i dt
dp 0 )−i p +µ (t−t0 )
Z
ip(x−x 2M
= −i e iδ(t − t0 )
2π
dp ip(x−x0 )
Z
= e δ(t − t0 )
2π
= δ(x − x0 )δ(t − t0 ) . (5)

R
3. Evaluate hxf , tf |xi , ti i explicitly by performing the dp above.
h 2
i
dp ip(xf −xI )−i 2M
p
Z
+µ (tf −ti )
e
2π
dp ip(xf −xi )−i p2 (tf −ti )
Z
−iµ(tf −ti )
= e e 2M
2π  
i(tf −ti ) 2p(xf −xi )M (xf −xi )2 M 2
p2 − + i(xf −xi )2 M
dp −
Z
−iµ(tf −ti ) 2M (tf −ti ) (tf −ti )2 2(tf −ti )
= e e e
2π
2
f −xi ) M
−iµ(tf −ti )
Z
dp0 − i(tf −ti ) p0 2 i(x2(t )
= e e 2M e f −t i
2π
i(xf −xi )2 M
s
−iµ(tf −ti )+ 2(t −ti )
M
= e f , (6)
2πi(tf − ti )
2
R +∞ q
π
where we used the standard result −∞ dte−at = a
to perform the p0 integral.

We will now obtain the same result following a path integral approach.

4. Consider first the general case where we have a 1-dimensional L of form

L(ẋ(t), x(t)) = c1 ẋ2 (t) + c2 x2 (t) , (7)

with action
Z
S[x(t)] = dt00 L(ẋ(t00 ), x(t00 )) , (8)

where the time derivative is in the dummy integration variable t00 . Expand S[x(t)]
about an arbitrary trajectory x0 (t) using

S[x(t)] = S[x0 (t) + y(t)]

 
Z
δS  1
Z
0 δ2S 
= S[x0 (t)] + dt y(t) + 2 dt dt y(t) y(t0 ) + . . .

δx(t) x0 (t) δx(t)δx(t0 ) x0 (t)
 

(9)
 δx(t) δẋ(t) d
Hint: Use rules like δx(t0 )
= δ(t − t0 ) and δx(t0 )
= dt
δ(t − t0 ) — sometimes it will be
useful to partial integrate the latter form under some integral.

We need,
" #
δS Z
00 00 d 00 00 00
= dt 2c1 ẋ(t ) 00 δ(t − t) + 2c2 x(t )δ(t − t)
δx(t) dt
Z
= dt00 2 [−c1 ẍ(t00 )δ(t00 − t) + c2 x(t00 )δ(t00 − t)]
= 2 [−c1 ẍ(t) + c2 x(t)] . (10)

δ2S h
0 0
i
= 2 −c 1 δ̈(t − t ) + c 2 δ(t − t ) . (11)
δx(t)δx(t0 )

Higher functional derivatives give zero since the double derivative already has no
dependence upon the functional form. (This is a special feature of the quadratic
form for L assumed.) Putting these two results into Eq. (9), we have

S[x(t)] = S[x0 (t) + y(t)]

Z
= S[x0 (t)] + dt 2 [−c1 ẍ0 (t) + c2 x0 (t)] y(t)
Z h i
+ 21 dt dt0 y(t) 2 −c1 δ̈(t − t0 ) + c2 δ(t − t0 ) y(t0 )
Z
= S[x0 (t)] + 2 dt [−c1 ẍ0 (t) + c2 x0 (t)] y(t)
Z
+ dt y(t) [−c1 ÿ(t) + c2 y(t)]
(12)

5. Derive the Euler-Lagrange equation for the general L(ẋ, x) from the action principle
δS
δx(t)
= 0.

Referring to the above first functional derivative result we find that

δS
= 2 [−c1 ẍ(t) + c2 x(t)] = 0 (13)
δx(t)

implies that the solution x0 (t) obeys

c2
ẍ0 (t) = x0 (t) . (14)
c1
6. Now compute hxf , tf |xi , ti i using path integral techniques. You must look back in
our notes and you will find the expression
 
Z x(tf )=xf dx(t)  eiS[x(t)]
hxf , tf |xi , ti i = q (15)
x(ti )=xi 2πi/M

where the  stuff is to remind you of what is actually present when you go to the
discrete limit. Note that we have made M explicit here, whereas we had set M = 1
in the notes to simplify our expressions obtained after performing the dpi integrals.
In the discrete limit, the fully correct expression was (assuming the simple potential
V = µ and using the notation adopted for this problem):
n n+1 2
1 dxk M xj − xj−1
Z Y n Xh  io
hxf , tf |xi , ti i = q lim q exp i −µ
2πi/M n→∞
k=1 2πi/M j=1 2 
(16)

where the j indices label the intermediate points separated by  in time and xn+1 =
xf and x0 = xi are fixed. In the above, I have assumed the same H and associated
L as in part 1). We will evaluate this by using the expansion techniques above,
δS
expanding around the classical path x0 (t) which solves δx(t)
= 0. You wish to show
that
hxf , tf |xi , ti i = eiS[x0 (t)] h0, tf |0, ti i , (17)

with
 
y(tf )=0 dy(t)
Z R
 ei dt c1 ẏ 2 (t)
h0, tf |0, ti i = q
y(ti )=0 2πi/M
 n+1  
n n+1
"  2 #
1 M yj − yj−1
Z Y  X 
≡ q  dyk exp i , (18)
2πi/M k=1

j=1 2  

M
where for this case we took c1 = 2
and c2 = 0 and we have yn+1 = y0 = 0.

To see that this is the correct procedure, note that expanding about extremum
function x0 (t) implies that the linear term in the expansion vanishes and we obtain
for the case in hand where c2 = 0
Z Z
S[x(t)] = S[x0 (t)] + dt y(t) [−c1 ÿ(t)] = S[x0 (t)] + c1 dt ẏ(t)ẏ(t) . (19)
 Now, it is obvious by writing x(t) = x0 (t) + y(t) and then inserting the above result
that
Z x(tf )=xf Z y(tf )=0
iS[x(t)]
d[x(t)]e = d[y(t)]eiS[x0 (t)+y(t)]
x(ti )=xi y(ti )=0
Z y(tf )=0 R
iS[x0 (t)] dtẏ 2 (t)
= e d[y(t)]eic1 . (20)
y(ti )=0

7. Now evaluate h0, tf |0, ti i explicitly. For this purpose, go back to the discrete version
of Eq. (18) which we obtain by noting that
 
n+1 2
yj − yj−1
Z 
dt c1 ẏ 2 (t) =
X
lim  c1  (21)
→0,n→∞
j=1 
and write the sum as a bilinear form

κy T Ay , (22)

where y(tf ) = yn+1 = 0 and y(ti ) = y0 = 0 and A is a symmetric matrix containing

only numbers. Then use the result derived in class that says
(2π)n/2
Z n o n o
dx1 . . . dxn exp − 21 xi Aij xj + xi Ji = √ exp 21 Ji A−1
ij J j , (23)
det A
setting Ji = 0. You will obviously need to compute det A — it should be fairly easy.

Having done this, take the continuum limit of your result, n → ∞,  → 0.

The key step here is to get the correct matrix A. To do so, we consider a couple of
examples. For n = 2, we would have (dropping c1 and  for the moment)

(y3 − y2 )2 + (y2 − y1 )2 + (y1 − y0 )2 = 2y22 + 2y12 − 2y2 y1 (24)

where we used the boundary conditions y3 = y0 = 0. The right hand side can be
written as   
2 −1   y1 
(y1 , y2 )  , (25)


−1 2 y2
where we call the matrix displayed A2 . If you do the same thing for n = 3 you will
have   
 2 −1 0   y1 
  
(y1 , y2 , y3 ) 
 −1   y2  ,
2 −1    (26)
  
0 −1 2 y3
 where the displayed matrix is denoted by A3 . It is now a simple matter the check
the det A2 = 3 and det A3 = 4, which generalizes to det An = n+1. The appropriate
result is then κy T Ay with An as exampled above and κ = c1 /. We thus want the
following integral:
 n+1
1
Z
q  dy1 . . . dyn exp {iκyi Aij yj }
2πi/M
 n+1
1 (2π)n/2
= q  √ n √
2πi/M −2iκ n+1
 n+1
1 (2π)n/2
= q  q n √
2πi/M −2ic1 / n+1
 
1 (2π)n/2
= q  q n √
4πc1
2πi/M M
n+1
 
1 (2π)n/2
= q  √ n √
2πi/M 2π n+1
" #1/2
M
=
2πi(tf − ti )

where we used  = (tf − ti )/(n + 1).

8. Finally, solve the Euler Lagrange equation for the boundary conditions

x(tf ) = xf , x(ti ) = xi (27)

and call the solution x0 (t). Compute S[x0 (t)]. Then, plug this and your result for
h0, tf |0, ti i into Eq. (17) and verify that this answer agrees with the answer you
obtained in part 3.

For c2 = 0, we already know that the Euler-Lagrange equation simply says that
ẍ0 (t) = 0, implying that ẋ0 (t) = constant ≡ v0 . Thus, we have (recall c1 = M/2)
M 2
L= v
2 0
− µ and
tf M 2
Z  
S[x0 (t)] = dt v −µ
ti 2 0
M 2
= v (tf − ti ) − µ(tf − ti )
2 0
 M (xf − xi )2
= (tf − ti ) − µ(tf − ti )
2 (tf − ti )2
M (xf − xi )2
= − µ(tf − ti ) . (28)
(tf − ti )

From which we obtain

iM (xf −xi )2
iS[x0 (t)] (tf −ti )
−iµ(tf −ti )
e =e , (29)

which duplicates the main factor of Eq. (6).

Putting all this together as in Eq. (17), we get

iM (xf −xi )2
" #1/2
(tf −ti )
−iµ(tf −ti ) M
hxf , tf |xi , ti i = e (30)
2πi(tf − ti )

which agrees precisely with Eq. (6). Now, I could have used a less general machinery
to get this answer for the c2 = 0 case, but the notation and machinery employed
make it clear how one might approach a c2 6= 0 problem in the most efficient manner.