Notes on Path Intergral

Based on a course given by Ananda Dasgupta

written by Krishanu Sengupta
March 2023

Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Free Particle in 1 D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Little Bit Generalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4 Why Feynman is Not an Ordinary Person . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5 Putting The Thoughts into Math. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
6 Generalisation of Gaussian Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
7 Path Integral Involving an External Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
8 Possible Way Outs to Make Things Simple. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
9 Harmionic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
9.1 Derivation of Mehler’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
10 Alternative Calculation for the S.H.O. propagator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
11 Masslov Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
12 General Quadratic Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

1
1 Introduction
This is going to be very basic stuff, mostly that I learned during my 1st year of M.Sc. So it is better not to
speak jargon let’s jump into the stuff.
Consider the propagator for the 1D Non relativistic Quantum Mechanics
The Schroedinger equation is given as
∂
H |Ψ(t)⟩ = iℏ |Ψ(t)⟩ (1)
∂t
Here symbols have their usual meaning. And in a position basis, it becomes

ℏ2 ∂ 2 ψ(x, t) ∂ψ(x, t)
− + V (x)ψ(x, t) = iℏ
2m ∂x2 ∂t
Where ψ(x, t) is given as ⟨x|Ψ(t)⟩
We can write the above expression in this way
 
∂
iℏ − H ψ(x, t) = 0 (2)
∂t

It’s a 1st order linear differential equation of time. If we know ψ(x, ti ) then we can easily find ψ(x, t) where
t > ti . And it is given as
Z ∞
ψ(x, t) = K(x, t; x′ , ti )ψ(x′ , ti )dx′ (3)
−∞

This is very important, here K is known as Kernel or Green’s function for the Time-Dependent Schroedinger
equation and K should satisfy,
 
∂
iℏ − H K(x, t; x′ , ti ) = iℏδ(x − x′ )δ(t − ti ) (4)
∂t

This is well known Green’s function trick to solve a differential equation. One Important property of the
propagator K is
lim+ K(x, ti + ϵ; x′ , ti ) = δ(x − x′ )
ϵ→ 0

Let us verify whether the solution stated at equation 3 is satisfied by equation 2 or not
  Z ∞ 
∂ ∂
iℏ − H ψ(x, t) = iℏ − H K(x, t; x′ , ti )ψ(x′ , ti )dx′
∂t −∞ ∂t
Z ∞
= iℏδ(x − x′ )δ(t − ti )dx′
−∞

= iℏδ(t − ti )
So it is quite evident that the left-hand side of the above expression is zero when t ̸= ti , (remember t > ti )
and it satisfies.

If H has no time dependence then, we can do much more.

H |ϕn (x)⟩ = En |ϕn (x)⟩ n = 0, 1, 2, 3... (5)

Here |ϕn ⟩ are the stationary eigenstates. And they are orthogonal to each other and they also from a
complete basis.

2
 ⟨ϕn |ϕm ⟩ = δnm
X
|ϕn ⟩ ⟨ϕn | = 1
n

We can write the completeness property in position basis as

X
ϕn (x)ϕ∗n (x′ ) = δ(x − x′ ) (6)
n

Equation 6 is going to be very useful later on.

We can expand the ψ(x, ti ) in the energy eigenbasis. (at a given instant of time say ti )
X
ψ(x, ti ) = an ϕn (x)
n

And the coefficients an are given as Z ∞

an = ϕ∗n (x)ψ(x, ti )dx
−∞

And ψ(x, t) is given as X

ψ(x, t) = an ϕn (x)e−iEn (t−ti )/ℏ (7)
n
X Z ∞ 
ψ(x, t) = ϕ∗n (x′ )ψ(x′ , ti )dx′ ϕn (x)e−iEn (t−ti )/ℏ
n −∞
!
Z ∞ X
ψ(x, t) = ϕ∗n (x′ )ϕn (x)e−iEn (t−ti )/ℏ ψ(x′ , ti )dx′ (8)
−∞ n

Now comparing equation 8 with equation 3 we can write

X
K(x, t; x′ , ti ) = ϕ∗n (x′ )ϕn (x)e−iEn (t−ti )/ℏ (9)
n

We need to keep in mind that propagators only work in the forward direction of time. So we need to insert
a step function in time which ensures that.
X
K(x, t; x′ , ti ) = Θ(t − ti ) ϕ∗n (x′ )ϕn (x)e−iEn (t−ti )/ℏ (10)
n

Where Θ(t − ti ) is given as 

1 t > ti
Θ(t − ti ) =
0 t < ti
∂
− H this to K(x, t; x′ , ti )


Now apply iℏ ∂t

∂
Then we will get the following result, first consider the term iℏ ∂t acting on K

∂ X X
iℏ K(x, t; x′ , ti ) = iℏδ(t − ti ) ϕ∗n (x′ )ϕn (x)e−iEn (t−ti )/ℏ + Θ(t − ti ) ϕ∗n (x′ )En ϕn (x)e−iEn (t−ti )/ℏ
∂t n n

and H operating on K gives us

3
 X
HK(x, t; x′ , ti ) = Θ(t − ti ) ϕ∗n (x′ )En ϕn (x)e−iEn (t−ti )/ℏ
n

So we get  
∂ X
iℏ − H K(x, t; x′ , ti ) = iℏδ(t − ti ) ϕ∗n (x′ )ϕn (x)e−iEn (t−ti )/ℏ
∂t n

Now use the result that we obtain in equation (6), and we obtain
 
∂
iℏ − H K(x, t; x′ , ti ) = iℏδ(x − x′ )δ(t − ti )
∂t
So, it do satisfy the 1D differential equation(equation 4) that we obtain earlier.

The point which is very important to realise that we are considering t > ti

2 Free Particle in 1 D
2 2 2
P̂ ℏ ∂
The Free particle Hamiltonian is given as H = 2m , and in position basis it is H = − 2m ∂x2 Now, we are
about to calculate its propagator. While working with a free particle it’s always better to move on to mo-
mentum space as Ĥ and p̂ commute for a free particle.

ψ(x, t) is given as
Z ∞
1
ψ(x, t) = √ ϕ(p, t)eipx/ℏ dp (11)
2πℏ −∞

Here ϕ(p) is the momentum eigenfunction. And if we do the inverse Fourier Transform then we get
Z ∞
1
ϕ(p) = √ ψ(x, t)e−ipx/ℏ dx (12)
2πℏ −∞

Now put the expression of 12 into equation 11, then

Z ∞ Z ∞ 
1 1 ′
ψ(x, t) = √ √ ψ(x′ , t)e−ipx /ℏ dx′ eipx/ℏ dp
2πℏ −∞ 2πℏ −∞
Z ∞ Z ∞ 
1 ′
= ψ(x′ , ti )e−ipx /ℏ dx′ e(−iH(t−ti )/ℏ) eipx/ℏ dp
2πℏ −∞ −∞
Z ∞ Z ∞ 
1 ′ −ipx′ /ℏ ′ p2
= ψ(x , ti )e dx e(−i 2m (t−ti )/ℏ) eipx/ℏ dp
2πℏ −∞ −∞

Z ∞ Z ∞ −ip2 (t−ti )

′
ψ(x, t) = e 2mℏ eip(x−x )/ℏ dp ψ(x′ , ti )dx′ (13)
−∞ −∞

Now compare equation 13 with our previous result then we say K(x, t; x′ , ti ) will be

4
 ∞
−i(t − ti )p2 i(x − x′ )p
Z  
′ 1
K(x, t; x , ti ) = exp + dp (14)
2πℏ −∞ 2mℏ ℏ

Now to calculate K we need to evaluate the integral in equation 14. The thing that is know to us is the
Gaussian integral. Using that we can compute this particular integral. I will just write the result of the
Gaussian integral and compute it.
Z ∞ r  2 
π b
exp(−ax2 + bx + c)dx = exp +c ; Re(a) > 0
−∞ a 4a

So we get s
−(x − x′ )2 2mℏ
 
′ 1 π2mℏ
K(x, t; x , ti ) = exp
2πℏ i(t − ti ) 4ℏ2 i(t − ti )

im(x − x′ )2
r  
′ m
K(x, t; x , ti ) = exp (15)
i2πℏ(t − ti ) 2ℏ(t − ti )

Or we can write  
r
′ m i
K(x, t; x , ti ) = exp Acc (16)
i2πℏ(t − ti ) ℏ

Where Acc is the action of the free particle.

3 Little Bit Generalisation

Now we consider equation 10, we can modify it look much more simpler

ϕn (x) = ⟨x|ϕn ⟩ ϕ∗n (x′ ) = ⟨ϕn |x′ ⟩

So we can write
 
X −iEn (t − ti )
K(x, t; x′ , ti ) = ⟨ϕn |x′ ⟩ ⟨x|ϕn ⟩ exp
n
ℏ
!
X
= ⟨x| exp(iĤti /ℏ) |ϕn ⟩ ⟨ϕn | exp(−iĤt/ℏ) |x′ ⟩
n
!
−iĤ(t − ti )
= ⟨x| exp |x′ ⟩
ℏ

So in general we can write

K(x, t; x′ , ti ) = ⟨x| Û (t, ti ) |x′ ⟩ (17)

Where Û (t, ti ) is the time evolution operator. Equation 17 is a very useful result we are going to use a
lot while solving problems in Quantum Field theory.

5
4 Why Feynman is Not an Ordinary Person
Suppose we are doing a double-slit experiment. We have a source S and a screen that collects the particles
after they go through the double slit. It is quite well known to all of us that an interference pattern will
appear at the screen.

Figure 1: Double Slit Experiment

It’s a quite well-known experiment for all of us. But now we are going to increase the number of slits.
Instead of 2 we going to make an infinite number of slits. Once again an interference will take place. Now
we go further instead of one screen of slits let us introduce an infinite number of such slit screens in between
the source S and the screen.

Now think about it there is nothing that exists between the source and the screen right! or one can say
every point is a slit. Each of these slits screen works as an identity operator. It is an incredible idea now let
us move on and do the math which supports this idea. We denote every infinite slit screen as B1 ,B2 ,...,BM
and M → ∞. The electron which is emitted from S can choose any of the infinite possible paths. So it is
nothing but a free particle going from source S to screen.

Figure 2: Multiple Slit Interference

5 Putting The Thoughts into Math

Now I am going to change the notation a little bit for simplicity and easier for calculation.

6
 Figure 3: Motion of the particle through different slits

Initially we are writing K(x, t; x′ , ti ) instead of that we will write K(xf , tf ; xi , ti ) where xf and xi are
final and initial position respectively, tf and ti are final and initial time respectively. But the mathematical
structure remains the same.

!
−iĤ(tf − ti )
K(xf , tf ; xi , ti ) = Θ(tf − ti ) ⟨xf | exp |xi ⟩ (18)
ℏ

And in Heisenberg’s picture, it will be

K(xf , tf ; xi , ti ) = Θ(tf − ti ) ⟨xf , tf |xi , ti ⟩ (19)

3
Particle’s classical path

t
1 2 3

Figure 4: Free particle motion in space

Now at every time instant the completeness property should hold and it is given as
Z ∞
dx |x, t⟩ ⟨x, t| = 1
−∞

7
 X

Path 21
3
Path 3

t
1 2 3

Figure 5: Curved paths of a free particle in space

we can plug this identity in equation 18, it will not change the result.

K(xf , tf ; xi , ti ) = Θ(tf − ti ) ⟨xf , tf |1|xi , ti ⟩

Z ∞
= Θ(tf − ti ) dx ⟨xf , tf |x, t⟩ ⟨x, t|xi , ti ⟩
−∞
Z ∞
= Θ(tf − t)Θ(t − ti ) dx ⟨xf , tf |x, t⟩ ⟨x, t|xi , ti ⟩
−∞

Here we have use the fact that,

Θ(tf − ti ) = Θ(tf − t)Θ(t − ti ) if ti < t < tf

Then,
Z ∞
K(xf , tf ; xi , ti ) = dxK(xf , tf ; x, t)K(x, t; xi , ti ) (20)
−∞

Now we can introduce as much as identity operator we want

Z ∞ Z ∞
K(xf , tf ; xi , ti ) = ⟨xf , tf |x2 , t2 ⟩ ⟨x2 , t2 |x1 , t1 ⟩ ⟨x1 , t1 |xi , ti ⟩ dx1 dx2 (21)
−∞ −∞

Where ti < t1 < t2 < tf

Z Z
K(f |i) = dx1 dx2 K(f |2)K(2|1)K(1|i)

Suppose we put n, 1 operators b/w them,

Z Z
K(f |i) = ... dx1 dx2 ...dxn K(f |n)K(n|n − 1)K(n − 1|n − 2)...K(3|2)K(2|1)K(1|i)

K(i|i − 1) = K(xi , ti ; xi−1 , ti−1 ) (22)

8
 Figure 6: Time slicing

We can use another convention where the initial i is set to zero and f is set to n + 1, for obvious reasons
Z Yn
K(f |i) = dxi {K(i|i − 1)}K(n + 1|n) (23)
i=1

n
Z Y
K(f |i) = {K(i + 1|i)}dxi K(1|0) (24)
i=1

Let us observe the term K(i + 1|i)

K(i + 1|i) = ⟨xi+1 , ti+1 |xi , ti ⟩

!
iĤ(ti+1 − ti )
K(i + 1|i) = ⟨xi+1 | exp − |xi ⟩
ℏ
And now we do the same, we introduce an infinite number of identity operators. Here we will consider the
Hamiltonian as

p̂2
Ĥ = + V (x̂) (25)
2m
And we know that
e eB̂ ̸= eÂ+B̂
So, let us exapand the exponential

!  2
−iϵĤ iϵĤ 1 −iϵ
exp =1− + Ĥ 2 + ...
ℏ ℏ 2 ℏ
iϵ p̂2
 
=1− + V (x̂) + O(ϵ2 )
ℏ 2m
iϵp̂2 iϵ
=1− − V (x̂) + O(ϵ2 )
 2m 2 ℏ 
iϵp̂ iϵV (x̂)
= 1− 1− + O(ϵ2 )
2mℏ ℏ
−iϵp̂2
   
−iϵV (x̂)
= exp exp + O(ϵ2 )
2mℏ ℏ
So, it turns out that it is true only up to the first order in epsilon.

9
  
−iϵĤ
Now we will compute the quantity ⟨xi+1 | exp ℏ |xi ⟩ only upto first order in ϵ
!
−iϵp̂2
   
−iϵĤ −iϵV (x̂)
⟨xi+1 | exp |xi ⟩ = ⟨xi+1 | exp exp |xi ⟩
ℏ 2mℏ ℏ
Z ∞
−iϵp̂2
   
−iϵV (x̂)
= dpi ⟨xi+1 | exp |pi ⟩ ⟨pi | exp |xi ⟩
−∞ 2mℏ ℏ
Z ∞
−iϵpi 2
   
−iϵV (xi )
= dpi ⟨xi+1 |pi ⟩ exp ⟨pi |xi ⟩ exp
−∞ 2mℏ ℏ
Z ∞    2
    
1 ixi+1 pi −iϵpi −ixi pi −iϵV (xi )
= dpi exp exp exp exp
2πℏ −∞ ℏ 2mℏ ℏ ℏ
 
−iϵV (xi ) Z
exp ℏ
∞ 
i(xi+1 − xi )pi
 
−iϵpi 2

= dpi exp exp
2πℏ −∞ ℏ 2mℏ

And according to our notation, the quantity K(i + 1|i) will be the same as the above expression.
 
exp −iϵVℏ(xi ) Z ∞ 
i(xi+1 − xi )pi
 
−iϵpi 2

K(i + 1|i) = dpi exp exp (26)
2πℏ −∞ ℏ 2mℏ

And using the result

∞
b2
Z r  
2 π
exp(−ax + bx + c)dx = exp +c ; Re(a) > 0
−∞ a 4a

We get  
−iϵV (xi )
exp ℏ
r
π

−(xi+1 − xi )2

K(i + 1|i) = exp
2πℏ iϵ 2mℏ 4ℏ2 iϵ/2mℏ

m(xi+1 − xi )2
r  
m iϵ
K(i + 1|i) = exp − V (xi ) (27)
iϵ2πℏ ℏ 2ϵ2
Therefore we can write !
n
Z Y n
Y
K(n + 1|0) = dxi K(i + 1|i)
i=1 i=0
 "  #
n  m  n+1 n 2
−
Z Y
2 iϵ X m x j+1 x j
K(n + 1|0) = dxi exp  − V (xj ) 
i=1
iϵ2πℏ ℏ j=0
2 ϵ

Now we define a quantity in this manner

n  n+1
Y m  2
lim = Dx (28)
n→∞
i=1
2iϵπℏ

and we also take the limit ϵ → 0

Z  Z tf 
i
K(f inal|initial) = Dx exp L(x, ẋ) (29)
ℏ ti

10
Or we can write
 Z 
i
K(f inal|initial) = Dx exp Ai→f
ℏ
n  m  n+1 i tf
Z Y  Z 
2
K(F |I) = dxj exp dtL(x, ẋ)
j=1
2iϵℏπ ℏ ti

Where ϵ is given as ϵ = (tf − ti )/n + 1 and as n → ∞, ϵ → 0

Now our next task is to solve it for a free particle using the above formula we derived

 
n
Z Y  α  n+1 n
2
X
K(F |I) = lim dxj exp −α (xj+1 − xj )2  (30)
n→∞
j=1
π j=0

m
Where α is given as α = 2ℏiϵ

Now consider the nth term in this expression

 
Z ∞
r Z n−1  α n/2 n−1
α Y X
Kn = dxn dxj exp[−α(xn+1 − xn )2 ] exp −α (xj+1 − xj )2 
−∞ π j=1
π j=0

∞
Z r
α
Kn = dxn exp[−α(xn+1 − xn )2 ]Kn−1 (31)
−∞ π
Where Kn−1 is given as
 
Z n−1
Y  α n/2 n−1
X
Kn−1 = dxj exp −α (xj+1 − xj )2  (32)
j=1
π j=0

Clearly equation 31 is a Iterative equation and K1 is given as

Z ∞ r
α
K1 = dx1 exp[−α(x2 − x1 )2 ]K0
−∞ π
Z ∞ α
K1 = dx1 exp[−α(x2 − x1 )2 − α(x1 − x0 )2 ]
−∞ π
α Z ∞
K1 = exp[−α(x2 − x1 )2 ] dx1 exp[−α(2x21 − 2(x2 − x0 )x1 )]
π −∞

Now we set a = −2α and b = 2α(x2 − x0 ) , remember α is a imaginary number. Once we do this we can
use the result of Gaussian integral as describe in equation

α(x2 − x0 )2
α r  
π 2
K1 = exp[−α(x2 − x1 ) ] exp
π 2α 2

11
 r
α α
K1 = exp[− (x2 − x0 )2 ] (33)
2π 2

Exactly the same we can compute the value of K2

∞
r Z r
α α  α 
dx2 exp −α(x3 − x2 )2 exp − (x1 − x0 )2
 
K2 =
π−∞ 2π 2
r
∞
α2
Z   
 2 2
 3 2
= exp −α(x 3 + x 0 /2) dx2 exp −α x − 2(x 3 + x 0 /2)x 2
2π 2 ∞ 2 2

And we set a = 32 α ; b = 2(x3 + x0 /2)

Then the above equation looks like

r
α h α i
K2 = exp − (3x23 + 3/2x20 − 2(x3 + x0 /2)2 )
3π 3

r  
α −α
K2 = exp (x3 − x0 )2 (34)
3π 3

By comparing equation 33 and 34 we can write the general expression for Kn−1 by guessing
r
α h α i
Kn−1 = exp − (xn − x0 )2 (35)
nπ n
So, Kn will be

α ∞
r Z
exp −α(xn+1 − xn )2 Kn−1
 
Kn =
π
r Z∞∞ r
α 2 α h α i
exp − (xn − x0 )2
 
= dxn exp −α(xn+1 − xn )
π −∞ nπ n
r  
α α 2 2 2


= exp − (n + 1)(xn+1 + x0 /n) − n(xn+1 + x0 /n)
π(n + 1) n+1
r  
α α 2
= exp − (xn+1 − x0 )
(n + 1)π n+1

Therefore,
r  
α α
Kn = exp − (xn+1 − x0 )2 (36)
(n + 1)π n+1

As we have calculated Kn then we can write the final expression of K(F |I) by putting the value of α

12
 r  
m m 2
K(F |I) = lim exp − (xf − xi )
n→∞ 2(n + 1)ϵπiℏ 2(n + 1)ϵℏi

And we replace the (n + 1)ϵ by tf − ti

−m(xf − xi )2
 
m
r
K(F |I) = exp
2πℏi(tf − ti ) 2iℏ(tf − ti )

−m(xf − xi )2
 
m
r
K(F |I) = exp (37)
2πℏi(tf − ti ) 2iℏ(tf − ti )

And we get back our previous result.

6 Generalisation of Gaussian Integral

As we doing Gaussian integral let us study it.
Z ∞ r  2 
π b
exp −ax2 + bx + c dx =
 
exp − + c (38)
−∞ a 4a

Consider the result in more than one dimension.

exp[−ax2 + bx + c] → exp[−(...)] (39)

Where (...) is given as

(...) = a11 x21 +a22 x22 +a33 x23 +...+ann x2n +2a12 x1 x2 +2a13 x1 x3 +...+2a1n x1 xn +2an−1,n xn−1 xn +b1 x1 +b2 x2 +...+bn xn +c

We can write this fashion(using matrix)

 
a11 a12 . . . a1n
 a21 a22 . . . a2n 
 
 . . . . 
A=
 .
 (40)
 . . . 
 . . . . 
an1 an2 . . . ann

A is symmetric matrix.
And
   
b1 x1
 b2   x2 
   
.  . 
b=
.
 X=
 . 
 (41)
   
.  . 
bn xn
Then the above equation can be written as

13
 (...) = X T AX + bT X + c
So, the integral now can be written as,
Z ∞Z ∞ Z ∞
... dx1 dx2 ...dxn exp[−X T AX + bT X + c]
−∞ −∞ −∞

Now we are writing X as X = Y + t, where Y and t both are column vectors.

Then the expression inside the exponential becomes

−X T AX + bT X + c = −(Y + t)T A(Y + t) + bT (Y + t) + c

= −Y T AY − tT AY − Y T At − tT At + bT y + bT t + c

And we know that

(Y T At)T = tT AT Y
(Y T At)T = tT AY
As we know that A is a symmetric matrix.

The condition that we impose,

2tT Ay = bT y
2tT A = bT

Now take transpose on both side

2At = b
1
t = A−1 b (42)
2
So, we finally get
1
= −Y T AY + bT A−1 b + c
4

And the integral becomes

Z Z
1
I= ... dY1 dY2 ...dYn exp[−Y T AY + bT A−1 b + c]
4
Z Z
1
I= ... dY1 dY2 ...dYn exp[−Y T AY ] exp[ bT A−1 b + c]
4

We know that a symmetric matrix can be always orthogonally diagonalisable using orthogonal transfor-
mation.

14
 Y = SZ

Where S is the orthogonal matrix. Then the term exp[−Y T AY ] becomes

det(S) exp[−z T S T ASz]dz1 dz2 ...dzn

Where det(S) is the Jacobian part of the transformation. As S is orthogonal transformation det(S) = 1. So
we left with
Z Z
exp[−Z DZ]dz1 dz2 ...dzn = ... dz1 dz2 ...dzn exp[−(λ1 z12 + λ2 z22 + ... + λn zn2 )]
T

r r r
π π π
= ...
λ1 λ2 λn
π n/2
=p
det(D)
π n/2
=p
det(A)

Finally we get, Z Z Z
I= ... dx1 dx2 ...dx2 exp[−X T AX + bT X + c]

π n/2 1
I=p exp[ bT A−1 b + c] (43)
det(A) 4

And Kn is given as
   
Z n  α  n+1 n
2
Y X
Kn =  dxj  exp −α (xj+1 − xj )2 
j=1
π j=0

It is important remember xo and xn+1 are not integrated over, so we need to keep then.

 
 α n+1/2 Z n
Y
Kn = exp(−α(x2n+1 + x2o ))  dxj  exp[2α(x1 xo + xo xn+1 )]
π j=1

exp[−α(2x21 + 2x22 + ... + 2x2n − 2x1 x2 − 2x2 x3 − ... − 2xn−1 x3 )]

If we write this above expression in terms of matrix equation then we get

exp[−(X T AX + bT X)]

where

15
  
  x1
x1  0 
 x2   
   . 
 .   
X=  . 
b = 2α  (44)
 
 . 
   . 
 .   
 0 
xn
xn+1
And
 
2 −1 0 . . 0
−1 2 −1 . . 0
 
0 −1 2 −1 . .
A = α
 .
 (45)
 . . . .
 . . . .
0 0 . . . 2 n×n

Clearly, its a tri-diagonal matrix and we are to compute the determinant of this matrix and it’s inverse A−1 .
If we observe carefully then it’s quite evident we only need 4 matrix elements of A−1 they are A−1 −1 −1
11 ,A1n ,An1
−1
and Ann , as b has only two elements  
x1
 0 
 
 . 
 
 . 
b = 2α  
 . 
 
 0 
xn+1

Now to compute the determinant we gone use the following trick(for the moment let us set α = 1). Using
recursion we are going to compute this. Dn is define as the determinant of the matrix A. If we expand the
determinant then
Dn = 2 × (cof actor of 2) + (−1) × (cof actor of − 1)
Now to compute the co factor of −2 we need to observe the matrix in equation 45 carefully. If I simply
eleminate the row and column corresponding to −2 it is quite obvious we are left with the matrix say A1 of
order (n − 1) × (n − 1) and it is of the same from as A. So it’s determinant will be Dn−1 .
And for the matrix element −1 the co-factor will be (−1)(−1) × Dn−2

Dn = 2Dn−1 + (−1)(−1)1+2 (−1)Dn−2 (46)

Dn = 2Dn−1 − Dn−2 (47)

The (−1)1+2 comes due to the property of determinant.

Form the matrix in equation 45 one can easily observe that D1 = 2 (first element of the matrix0)and
D2 = 3 (determinant of the 2 × 2 matrix). And following the recursiuon relation in equation 47 we get

Dn = n + 1 (48)

16
 Now the next task is to compute the inverse elements as specified above(restore the α). A−1 −1
11 = Ann and
A−1
1n
−1
= An1 as A is a symmetric matrix. So, we need to compute only two inverse elements and they are

1 Dn−1 1 (−1)(1+n) (−1)(n−1)

A−1
11 = A−1
1n =
α Dn α Dn
n 1
A−1
11 = A−1
1n = (49)
α(n + 1) α(n + 1)

So, finally we get Kn as

π n/2
 α (n+1)/2  
2 n n xo xn+1
Kn = exp[−α(x2n+1 + x2o )] √ exp α(xo 2
+ xn+1 +2
π αn/2 n + 1 n+1 n+1 n+1
 
α α 2
Kn = √
exp − (xn+1 − xo ) (50)
π n+1 n+1

This is exactly the same expression of Kn as we got in equation 36.

7 Path Integral Involving an External Source

Till now what we have discussed it’s a free particle motion. Now let us move on to a problem where a
constant external force exists. Before jumping into the problem let us consider the propagator

Kab = K(xb , tb ; xa , ta )
Z  Z tb 
i
Kab = DxDp exp L(x, p)dt (51)
ℏ ta

where L(x, p) is the Lagrangian of the system. We can also write it using Legendre transformation. We
know that

H(x, p) = pẋ − L(x, p) (52)

L(x, p) = pẋ − H(x, p) (53)

One Thing to notice is that we are wrtting the Lagrangian as a function of (x,p) not in terms of (x,ẋ),
the reason is quite obvious the integration is over x and p. So instead of writing L(x, p) in equation 51, we
will write the R.H.S. expression of equation 53.

Then we get Z  Z tb 
i
Kab = DxDp exp (pẋ − H(x, p)) dt (54)
ℏ ta

Now the Hamiltonian of a system is given as

p2
H(x, p) = + V (x) (55)
2m

17
And the corresponding Lagrangian is given as
1
L(x, ẋ) = mẋ2 − V (x) (56)
2
Then the equation 51 becomes
 
N   N    
−
Z
dpo Y dpj dxj i X x j+1 x j
Kab = lim exp  ϵ pj − H(pj , xj )  (57)
N →∞ 2πℏ j=1 2πℏ ℏ j=0 ϵ

Suppose we have a time-dependent external force, only the time dependence.

V (x) → −J(t)x + V (x)

p2
L(x, p) = ẋp − − V (x) + xJ(t) (58)
2m

For simplicity, we set V (x) = 0. so, the equation 58 becomes as

p2
L(x, p) = ẋp − + xJ(t) (59)
2m
Then the expression of Kab becomes
 
N  N
ϵp2j
Z 
(o) dpo Y dpj dxj i X
Kab = lim exp  (pj (xj+1 − xj ) − + ϵJ(tj )xj ) (60)
N →∞ 2πℏ j=1 2πℏ ℏ j=0 2m

Now consider the term and let us expand it

N
X N
X N
X
pj (xj+1 − xj ) = pj xj+1 − p j xj
j=0 j=0 j=0
N
X +1 N
X
= pj−1 xj − p j xj
j=1 j=0
N
X
= (pj−1 − pj )xj + pN xN +1 − po xo
j=1
N
X
= (pj−1 − pj )xj + pN xb − po xa
j=1

Therefore the expression of the exponential becomes

  
N N N
i X X ϵp2j X
exp   (pj−1 − pj )xj + pN xb − po xa − + ϵJ(tj )xj 
ℏ j=1 j=0
2m j=0

  
N N 2
i X X ϵp j 
exp   (pj−1 − pj + ϵJ(tj )xj + pN xb − po xa + ϵJ(to )xa − )
ℏ j=1 j=0
2m

18
 So finally, we get
 
Z N
dpo dp1 dp2 ...dpN i iϵ X
lim exp  (pN xb − po xa + ϵxa J(to )) − p2 
N →∞ (2πℏ)N +1 ℏ 2mℏ j=0 j
N Z ∞
Y i
dxj exp[ (pj−1 + pj + ϵJ(tj ))]
j=1 −∞ ℏ

And we know that, Z ∞

exp(ipx/ℏ) = 2πℏδ(p)
−∞

Then the integral corresponding to the dxj will become

N Z ∞ N
Y i Y
dxj exp[ (pj−1 + pj + ϵJ(tj ))] = [2πℏδ(pj−1 − pj + ϵJ(tj ))]
j=1 −∞ ℏ j=1

= (2πℏ)N δ(p0 − p1 + ϵJ(t1 ))δ(p1 − p2 + ϵJ(t2 ))...δ(pN −1 − pN + ϵJ(tN ))

So finally the integral will become,

 
Z N
dpo dp1 dp2 ...dpN i iϵ X
lim exp  (pN xb − po xa + ϵxa J(to )) − p2  (2πℏ)N δ(p0 − p1 + ϵJ(t1 ))
N →∞ (2πℏ)N +1 ℏ 2mℏ j=0 j

δ(p1 − p2 + ϵJ(t2 ))...δ(pN −1 − pN + ϵJ(tN ))

 
Z N
dpo dp1 dp2 ...dpN i iϵ X
= lim exp  (pN xb − po xa + ϵxa J(to )) − p2 
N →∞ (2πℏ) ℏ 2mℏ j=0 j

δ(p0 − p1 + ϵJ(t1 ))δ(p1 − p2 + ϵJ(t2 ))...δ(pN −1 − pN + ϵJ(tN ))

Observe very carefully the first integration is over p0 , that will be replaced by p1 , and p1 will be replaced
by p2 and so on...
 
XN
p0 → p1 − ϵJ(t1 ) → p2 − ϵ(J(t1 ) + J(t2 )) → ... → pN − ϵ  J(tj )
j=1

Similarly every pk will become

N
X
pk → pN − ϵ J(tj ) (61)
j=k+1

19
Finally, we get,
   
Z N
(0) 1 i X
Kab = lim dpN exp  (pN xb − pN − ϵ J(tj ) xa )
2πℏ N →∞ ℏ j=1
  2 
N N
 ϵ X X
exp − pN − J(tk ) 

2m j=0
k=j+1

And we define two integrals as Z tb

I(t) = J(τ )dτ
ta
Z tb
J(τ )dτ = Θ(τ − t)J(τ )dτ (62)
ta

0
So, the final expression of Kab [J] will become,
Z ∞   Z tb 
(0) 1 i 1
Kab [J] = dp exp p(xb − xa ) + I(ta )xa − (p − I(t))2 dt
2πℏ −∞ ℏ 2m ta
Z tb  Z ∞
ip2 1 tb
    Z 
1 i 1 2 ip
= exp xa I(ta ) − I (t)dt dp exp − (tb − ta ) + xb − xa + I(t)dt
2πℏ ℏ 2m ta −∞ 2mℏ ℏ m ta
1/2 " 2 !#
x b − x a tb
Z tb
i m(xb − xa )2
 Z
m 1
= exp + I(t)dt + I(t)dt
2πiℏ(tb − ta ) ℏ (tb − ta ) t b − t a ta 2m(tb − ta ) ta
 Z tb 
1 2
exp xa I(ta ) − I (t)dt
2m ta

For the special case J(t) = f , we end up getting,

1/2
i m(xa − xb )3 (tb − ta )3 f 2
   
(0) m 1
Kab [J(t) = f ] = exp + (xb + xa )(tb − ta ) − (63)
2πiℏ(tb − ta ) ℏ (tb − ta ) 2 24m

If we observe the term inside the exponential it is once again the classical action.

Now consider the general case,

 
N
ϵp2j
Z
(0) dp0 dp1 dpN i X
Kab [J] = lim dx1 dx2 ...dxn ... exp  pj (xj+1 − xj ) − + ϵJ(tj )xj 
N →∞ 2πℏ 2πℏ 2πℏ ℏ j=0 2m

(0)
Kab [J] = ⟨xb , tb |xa , ta ⟩J (64)

0
∂Kab
If we now differentiate ∂J(tk ) then we get

0
∂Kab (J(t0 , J(t1 ), ..., J(tN ))) iϵ (0)
= xj Kab [J]
∂J(tk ) ℏ

20
 0
∂Kab iϵ
= ⟨xb , tb |x̂j |xa , ta ⟩J
∂J(tk ) ℏ
0
∂Kab iϵ
= ⟨xb , tb |x̂(tj )|xa , ta ⟩J (65)
∂J(tk ) ℏ

And if we perform a second derivative then we get,

 2
∂ 2 Kab
0
(J) iϵ
= ⟨xb , tb |T (x̂(tj )x̂(tl ))|xa , ta ⟩J (66)
∂J(tk )∂J(tl ) ℏ

Which gives the time-ordered product of position.

The repeated functional derivative gives a matrix element of the time-ordered product of operators.
It is a very good and efficient trick to evaluate the time-ordered product of operators. After differentiating
we simply put (J = 0). We used this kind of trick in Ising Model.

8 Possible Way Outs to Make Things Simple

Consider fluctuations around the Classical Path,

x(t) = xcl (t) + X(t) (67)

X(t) is the fluctuations around xcl (t). Taking the boundary condition into account (X(ta ) = 0; X(tb ) = 0)
we can write the xcl (t) as
xb − xa
xcl (t) = (t − ta ) + xa (68)
tb − ta
Now the action corresponding to the free particle is

Z tb
1
A[x] = mẋ2 dt
ta 2
Z tb
1
= m (ẋcl + X)˙2 dt
2 ta
Z tb Z tb Z tb
1 1
= m ẋ2cl dt + m ẋcl Ẋdt + m Ẋ 2 dt
2 ta ta 2 ta
Z tb Z tb
m tb 2
Z
d
= Scl + m (ẋcl X)dt − m ẍcl Xdt + Ẋ dt
ta dt ta 2 ta

And 2nd term X(ta ) = 0 = X(tb ) and in the 3rd term ẍcl = 0 (as it is the equation of motion)
Z tb
1
A[x] = Scl + m Ẋdt (69)
2 ta

21
 So, the propagator will be
 
n+1 Z ∞   N  2
−
Z 
m  2 i iϵ X m Xj+1 Xj
K(xb , tb ; xa , ta ) = lim dx1 dx2 ...dxn exp Scl exp   (70)
n→∞ 2πℏiϵ −∞ ℏ ℏ j=0 2 ϵ

 
i
K(xb , tb ; xa , ta ) = F (tb , ta ) exp Scl (71)
ℏ

The prefactor F (ta , tb ) can be calculated more simply. Let us try this method to solve the Harmonic
Oscillator.

9 Harmionic Oscillator
Now for a harmonic oscillator the lagrangian is given as
1 1
L(x, ẋ) = mẋ2 − mω 2 x2 (72)
2 2
Now we substitute
x(t) = xcl (t) + X(t)
and the corresponding equation of motion is given as

ẍcl + ω 2 x = 0

and the boundary condition are X(ta ) = 0 = X(tb ) and we have to substitute in the path integral as
X0 = 0 = Xn+1

So, the action becomes

Z tb h
m i
A[x] = (ẋcl + Ẋ)2 − ω 2 (xcl + X)2 dt (73)
ta 2
Z tb Z tb
m tb 2
Z
m 2
= (ẋcl − ω 2 x2cl )dt + m (ẋcl Ẋ − ω 2 xcl X)dt + (Ẋ − ω 2 X 2 )dt (74)
ta 2 ta 2 ta

Now focus on the 2nd of the equation 70

Z tb
m (ẋcl Ẋ − ω 2 xcl X)dt
ta
Z tb Z tb Z tb
d
=m (ẋcl X)dt − m ẍcl Xdt − m ω 2 xcl Xdt
ta dt ta ta
Z tb Z tb
=0+m ω 2 xcl Xdt − m ω 2 xcl Xdt
ta ta
=0

Using the equation of motion we get that the cross term goes to zero. So, we left with

m tb 2
Z
A[x] = Scl + (Ẋ − ω 2 X 2 )dt (75)
2 ta

22
 Now we are going to use the above method as discussed in equation 67
 
i
K(xb , tb ; xa , ta ) = F (tb , ta ) exp Scl
ℏ

So, we are to calculate F (tb , ta ) and for this case, it is given as

 
 m  n+1 Z ∞ n
2 im X
F (tb , ta ) = lim dx1 dx2 ...dxn exp  ((Xj+1 − Xj )2 − ω 2 ϵ2 Xj2 )
n→∞ 2πℏiϵ −∞ 2ℏϵ j=0

And the boundary condition is given as X0 = 0 = Xn+1 . Therefore the integral we obtain

π n/2
Z
dx1 dx2 ...dxn exp(−X T AX) = p (76)
det(A)

And here the matrix A is given as

2 − ω 2 ϵ2
 
−1 0 . . 0
 −1
 2 − ω 2 ϵ2 −1 . . 0 

m  0 −1 2 − ω 2 ϵ2 −1 . . 
A=   (77)
2iℏϵ 
 . . . . . 

 . . . . 
0 0 . . . 2 − ω 2 ϵ2 n×n

Clearly, the matrix A is a tri-diagonal matrix. If we observe carefully it will be very obvious to us this
matrix is very similar to the one that appear for the free particle case.So, the determinant of the matrix is
given as  m n
det(A) = Dn
2iℏϵ
Using the same idea as free particle the expression for Dn is given as

Dn = (2 − ω 2 ϵ2 )Dn−1 − Dn−2 (78)

Now we have to the solve the recursion relation in equation 74 as Dn is the determinant of the matrix A ,
we can evaluate some of initial values easily D0 ,D1 and D2 .

D−1 = 0; D0 = 1; D1 = 2 − ω 2 ϵ2 ; D2 = (2 − ω 2 ϵ2 )2 − 1

But unlike the previous case for free particle we cannot find Dn we easily for that we have to use the Ansatz
Dn ∼ rn

rn = (2 − ω 2 ϵ2 )rn−1 − rn−2
r2 = (2 − ω 2 ϵ2 )r − 1

r2 − (2 − ω 2 ϵ2 )r + 1 = 0 (79)
And the roots of this equation is given as
r
ω 2 ϵ2 ω 2 ϵ2
 
r± = 1 − ± iωϵ 1 − (80)
2 4

23
Now, if we take n → ∞ then ϵ → 0. Then we can write 0 in place of ϵ2 . Therefore we get

r± = 1 ± iωϵ (81)

if x → 0 we can write sin x ∼ x and cos x ∼ 1. So, finally we can write equation 78 as

r± = exp(±iϕ) (82)

where ϕ = ωϵ Now from here we can write the expression of Dn

n n
Dn = Ar+ + Br− (83)

Consider the initial values of Dn for n = 0 and n = −1

A B
A+B =1 + =0
r+ r−

Now solving these two above equation we get

eiϕ −eiϕ
A= B= (84)
2i sin ϕ 2i sin ϕ
So, finally the Dn becomes
exp[i(n + 1)ϕ] − exp[−i(n + 1)ϕ]
Dn =
2i sin ϕ
sin[(n + 1)ϕ]
Dn =
sin ϕ
 m n sin[(n + 1)ϕ]
det(A) = (85)
2iℏϵ sin ϕ
Therefore F (ta , tb ) is given as
s
 m  n+1 2
 m − n2 sin ϕ
F (ta , tb ) = lim π n/2
n→∞ 2πℏiϵ 2πℏiϵ sin (n + 1)ϕ
 1/2
m sin ϕ 1
= lim p
n→∞ 2πiℏϵ sin[(n + 1)ϕ]
r
mωϵ
= lim
n→∞ 2πiϵℏ sin[ω(n + 1)ϵ]
r
mω
=
2πiℏ sin[ω(tb − ta )]

And the expression for K(xb , tb ; xa , ta ) is

r  
mω i
K(xb , tb ; xa , ta ) = exp Scl
2πiℏ sin[ω(tb − ta )] ℏ

We know that
xcl = a cos ωt + b sin ωt (86)

24
 ẋcl = −ωa sin ωt + ωb cos ωt

The Lagrangian corresponding to these two above equation will become

m 2
Lcl = (ẋ − ω 2 x2cl )
2 cl
1
Lcl = mω 2 [(b2 − a2 ) cos 2ωt − 2ab sin 2ωt]
2
Corresponding the action Scl becomes

1 (b2 − a2 )
 
ab
Scl = (sin 2ωtb − sin 2ωta ) + (cos 2ωtb − cos 2ωta )
2 2ω ω
mω  2 2

= (b − a ) sin ω(tb − ta ) cos ω(tb + ta ) − 2ab sin ω(tb − ta ) sin ω(tb + ta )
2
mω
= sin ω(tb − ta )[(b2 − a2 ) cos ω(tb + ta ) − 2ab sin ω(tb + ta )]
2

And if substitute the initial condition

xa = a cos ωta + b sin ωtb

xb = a cos ωta + b sin ωtb

Now if have to solve for a and b then we get
xa sin ωtb − xb sin ωta
a= (87)
sin ω(tb − ta )

−xa cos ωta − xb sin ωta

b= (88)
sin ω(tb − ta )

And the expression for (a2 − b2 ) and 2ab is given as

1
a2 − b2 = 2 [x2a cos 2ωtb + x2b cos 2ωta − 2xa xb cos ω(tb + ta )] (89)
sin ω(tb − ta )
1
2ab = [−x2a sin 2ωtb − x2b sin 2ωta + 2xa xb sin ω(tb + ta )] (90)
sin2 ω(tb − ta )

Now if we substitute the expression for (a2 − b2 ) and 2ab into Scl will be
mω
Scl = [(x2 + x2b ) cos ω(tb − ta ) − 2xa xb ] (91)
2 sin ω(tb − ta ) a

So, the propagator finally becomes

imω [(x2a + x2b ) cos ω(tb − ta ) − 2xa xb ]

r
mω
K(xb , tb ; xa , ta ) = exp (92)
2iπℏ sin ω(tb − ta ) 2ℏ sin ω(tb − ta )

25
 One thing to remember which is very important in the entire derivation we have assumed ω(tb − ta ) << π
Another way to write it down will be
K(xb , tb ; xa , ta ) = ⟨xb | exp [−iĤ(tb − ta )/ℏ] |xa ⟩ (93)

Where Ĥ is given as
p̂2 1
+ mω 2 x̂2
Ĥ =
2m 2
Now we expand equation 90 in the energy eigenfunction basis
K(xb , tb ; xa , ta ) = ⟨xb | exp [−iĤ(tb − ta )/ℏ] |xa ⟩
X
= ⟨xb | |ψn ⟩ ⟨ψn | exp [−iĤ(tb − ta )/ℏ] |xa ⟩
n
X
= ⟨xb |ψn ⟩ exp [−iEn (tb − ta )/ℏ] ⟨ψn |xa ⟩
n
X
= ψn∗ (xa )ψn (xb ) exp [−iEn T /ℏ]
n

Where T = (tb − ta )
Now in the expression of equation 89 we have to replace the cos ωT and sin ωT with exponential.
eiωT + e−iωT
cos ωT =
2
eiωT (1 + e−2iωT )
cos ωT =
2
And
eiωT − e−iωT
sin ωT =
2i
eiωT (1 − e−2iωT )
sin ωT =
2i
Now focus on the expression 89

r  
mω imω 2 2 2xa xb
K(xb , tb ; xa , ta ) = exp (xa + xb ) cot ωT −
2iπℏ sin ω(tb − ta ) 2ℏ sin ωT
−2iωT
e−iωT
r    
mω −iωT /2 −2iωT −1/2 mω 2 2 1+e 2ωT
= e (1 − e ) exp (x + xb ) exp xa xb
πℏ 2ℏ a 1 − e−2iωT ℏ 1 − e−2iωT
r  
mω −iωT /2 1 1
= e 1 + e−2iωT − e−4iωT + · · ·
πℏ 2 8
h mω  

i 2mω
(x2a + x2b ) 1 + 2e−2iωT + 2e−4iωT + · · · exp xa xb e−iωT 1 + e−2iωT · · ·


exp −
2ℏ ℏ
r  
mω h mω 2 i 1 1
= exp − (xa + x2b ) e−iωT /2 1 + e−2iωT − e−4iωT + · · ·
πℏ 2ℏ 2 8
  mω   
2mω
1+ − (x2a + x2b )e−2iωT + · · · 1+ xa xb e−iωT + · · ·
2ℏ ℏ
r r  
mω h mω 2 2
i
−iωT /2 mω 2mω −3iωT h mω i
= exp − (xa + xb ) e + xa xb exp exp − (x2a + x2b )
πℏ 2ℏ πℏ ℏ 2 2ℏ

26
Compare  
X T
K(xb , tb ; xa , ta ) = ψn∗ (xa )ψn (xb ) exp −iEn (94)
n
ℏ
1
r
ℏω mω 2  mω 
E0 = ψ0 (x) = exp − x2
2 πℏ 2ℏ
r 1 r
3ℏω mω 2 2mω  mω 
E1 = ψ1 (x) = x exp − x2
2 πℏ πℏ 2ℏ
The general case: Mehler’s formula

∞
2ξηz − z 2 (ξ 2 + η 2 )
 
X Hn (ξ)Hn (η) n 2 − 21
z = (1 − z ) exp (95)
n=0
2n n! 1 − z2

The propagator is given as

imω [(x2a + x2b ) cos ωT − 2xa xb ]

r  
mω
K= exp
2πℏ sin ωT 2ℏ sin ωT

We will change to dimensionless quantities,

r r
mω mω
ξ= xa η= xb
ℏ ℏ
Then the above expression becomes
r   
mω i cos ωT i
K= exp 2ξη (ξ 2 + η 2 ) − (96)
2πℏ sin ωT 2 sin ωT 2 sin ωT

Now we choose z such that

z −i
2
=
1−z 2 sin ωT
 
i 2i
=−
2 eiωT − e−iωT
1
= iωT
e − e−iωT
e−iωT
=
1 − e−2iωT

Then we can write z = e−iωT

Similarly, we can write

i2 eiωT + e−iωT
 
i
cot ωT =
2 2 eiωT − e−iωT
1 + z2
=−
2(1 − z 2 )
1 z2
=− −
2 1 − z2

27
And we can also write
1 − z2
2i sin ωT =
z
Now we plug in these above expressions into the equation 92 and we get

z2
r  
mωz 2ξηz 2 2 1
K= exp + (ξ + η )(− − ) (97)
πℏ(1 − z 2 ) (1 − z 2 ) 2 1 − z2

2ξηz − (ξ 2 + η 2 )z 2 (ξ 2 + η 2 )
r    
mωz
K= exp exp −
πℏ(1 − z 2 ) 1 − z2 2
∞
!
ξ2 + η2
r  
mωz X Hn (ξ)Hn (η) n
K= n
z exp −
πℏ n=0
2 n! 2
∞  2
ξ + η2
r 
mω −iωT /2 −inωT X Hn (ξ)Hn (η)
K= e e exp −
πℏ n=0
2n n! 2

∞  2
ξ + η2
r     
mω X Hn (ξ)Hn (η) 1
K= exp − exp − n + ωT (98)
πℏ n=0 2n n! 2 2

9.1 Derivation of Mehler’s formula

This derivation will be very brief I did this one just for the completion.

We know that the generating function for Hermit polynomials is given as

2 dn
Hn (x) = (−1)n ex exp (−x2 ) n = 0, 1, 2, 3... (99)
dxn
Now consider this integral
Z ∞ Z ∞
2 −x2
exp (−t + 2xit) = e exp (−[t − ix]2 )dt
−∞ −∞
Z ∞ √
2
dt exp (−t2 + 2xit) = e−x π
−∞
∞ Z
2 1
e−x = √ dt exp (−t2 + 2xit)
π −∞
Z ∞
dn −x2 1
(e )= √ dt(2it)n exp (−t2 + 2xit)
dxn π −∞

Now put this expression into the above equation 96

2 ∞
2n ex (−i)n
Z
Hn (x) = √ tn exp (−t2 + 2itx)dt
π −∞

∞ 2 ∞
X Hn (x)z n ex X
=√ exp (−t2 + 2itx) exp (−2itz)dt
n=0
n! π −∞
∞ 2 Z ∞
X Hn (x)z n ex
=√ exp [−t2 + 2i(x − z)t]dt
n=0
n! π −∞

28
 ∞
X Hn (x)z n 2
= ex exp [−(x − z)2 ]
n=0
n!
∞
X Hn (x)z n
= exp[2zx − z 2 ] (100)
n=0
n!

Then, we can write

∞ 2 2 ∞
Hn(x)Hn (y)z n ex +y X (−1)n (2z)n ∞ ∞
X Z Z
n n!
= (uv)n exp [−(u2 + v 2 ) + 2i(ux + vy)]dudv
n=0
2 π n=0
n! −∞ −∞
2 2 Z ∞ Z ∞
ex +y
= exp [−u2 − v 2 + 2iux + 2ivy − 2uvz]dudv
π −∞ −∞
2 2 Z ∞
ex +y √ −u2 +2iux (iy−uz)2
= πe e du
π −∞
2 2 Z ∞
ex +y 2 2 2
= √ e−y e−u (1−z )+2i(x−yz)u du
π −∞
2
ex (x − yz)2
 
=√ exp −
1 − z2 1 − z2
x2 − 2xyz + y 2 z 2
 
1 2
=√ exp x −
1 − z2 1 − z2
∞
Hn(x)Hn (y)z n 2xyz − (x2 + y 2 )z 2
 
X 1
=√ exp (101)
n=0
2n n! 1 − z2 1 − z2

10 Alternative Calculation for the S.H.O. propagator

Now we will take an alternative route to derive the propagator for the Simple Harmonic Oscillator. As we
have written before the path x(t) can be written as

x(t) = xcl (t) + X(t)

And the corresponding action is given as

Z tb
m
A[x] = Scl + (Ẋ − ω 2 X 2 )dt
2 ta

With the boundary condition

X(ta ) = 0 = X(tb )
And the corresponding form of the propagator is given as
 
i
K(xb , tb ; xa , ta ) = F (tb , ta ) exp Scl
ℏ
Since the X(t) is defined only in the interval [ta , tb ] and vanishes at both ends, we can write it in terms
of a (half period) sine expansion.
∞  
X kπ(t − ta )
X(t) = bk sin (102)
T
k=1

29
 Where T is given as T = tb − ta
∞  
X bk πk kπ(t − ta )
Ẋ(t) = cos (103)
T T
k=1

And correspondingly

tb ∞ tb
πk ′ (t − ta )
   
πk(t − ta )
Z X Z
X 2 dt = bk bk′ sin sin
ta ta T T
k,k′ =1
∞
X T
= bk bk′ δkk′
2
k,k′ =1
∞
T X
= b2k
2
k=1

Similarly,
tb ∞ 
T X π 2 k 2 b2k
Z 
Ẋ 2 dt =
ta 2 T2
k=1

Now we can write the expression Z tb

m
(Ẋ 2 − ω 2 X 2 )dt
2 ta

as
Z tb ∞  2 2
 X 
m 2 2 m 2 T π k
(Ẋ − ω X )dt = − ω b2k
2
2 ta 2 2 T2
k=1

So, the expression for F (tb , ta ) will be given as

∞ Z ∞
π2 k2
   
Y i mT
F (tb , ta ) = C1 dbk exp − ω b2k
2
−∞ ℏ 4 T2
k=1
∞
!
Y 1
F (tb , ta ) = C1 p (104)
k=1
1− ω 2 T 2 /π 2 k 2

Using Euler’s formula we can write

r
ωT
F (tb , ta ) = C1 (105)
sin ωT
Where C1 is the constant. The get value of this constant can be evaluated quite easily. If we take the
limit ω → ∞ we will get back the expression for free particle.
r
m
lim F (tb , ta ) =
ω→0 2πℏiT

30
 r
m
C1 = (106)
2πℏiT

So, for the harmonic oscillator F (tb , ta ) will be

r
mω
F (tb , ta ) = (107)
2iπℏ sin ωT

Now the obvious question is which Euler’s Formula is used?

If a Polynomial f (z) has zeros at z = a,z = b,z = c ... etc. Then f (z) can be written as

f (z) = k(z − a)(z − b)(z − c)... (108)

 z z z
f (z) = k ′ 1 − 1− 1− ...
a b c

Here we are dealing with a finite-order polynomial.

Consider the function sinz z , it is an infinite order polynomial with zero’s at ±π,±2π, ±3π...

So, we can write

sin z  z z z  z 
= 1− 1+ 1− 1+ ...
z π  π 2π 2π
z2 z2 z2
   
= 1− 2 1− 2 1 − 2 ...
π 4π 9π
∞
z2 π2
Y  
= 1− 2
k
k=1

Therefore,
∞ 
z2 π2

sin z Y
= 1− 2 (109)
z k
k=1

11 Masslov Correction
Our expression for the S.H.O. propagator is valid as long as ωT < π.

For π < ωT < 2π the first factor in the expansion will be negative. As

ω2 T 2
 
1− <0
π2

This contributes a e−iπ/2 to the propagator.

31
 Similarly for 2π < ωT < 3π we have to factors of (−1). In general, for nπ < ωT < (n + 1)π we have
e−inπ/2 . This e−inπ/2 term is known as the Masslov factor.

In general, the propagator will be

r  
mω −iπ/4 −iπn/2 imω 2 2
K(xb , T ; xa , 0) = e e exp ((xa + xb ) cos ωT − 2xa xb ) (110)
2πℏ| sin ωT | 2ℏ sin ωT

This derivation is not that rigorous. Let us do a careful determination of the Maslov factor,
We will use the spectral representation to calculate the Maslov factor

The propagator is given as

∞
X
K(xb , tb ; xa , ta ) = exp [−iEn T /ℏ]ψn (xb )ψn∗ (xa )
n=0

For Harmonic Oscillator the En is given as

 
1 1  mω 1/4 −ξ2 /2
En = n + ℏω ψn = √ e Hn (ξ)
2 2n n! πℏ

p mω
where ξ = ℏ x
∞ 2 2
mω X −i(n+ 21 )ωT e−(ξa +ξb )/2
r
K(xb , tb ; xa , ta ) = e Hn (ξa )Hn (ξb )
πℏ n=0 2n .n!

ωT
Consider π ∈
/ N and let ωT = N π + ωτ where 0 < ωτ < π. Then,

e−i(n+ 2 )ωT = e−i(n+ 2 )N π e−i(n+ 2 )ωτ

1 1 1

= e−iN π/2 (−1)nN e−iN π/2

∞ 2 2
mω X −i(n+ 21 )ωT e−(ξa +ξb )/2
r
K(xb , tb ; xa , ta ) = e Hn (ξa )Hn (ξb )
πℏ n=0 2n .n!
r ∞
mω −iN π/2 −iN π/2 −(ξa2 +ξb2 )/2 X −iωτ n
= e e e (e ) (−1)nN Hn (ξa )Hn (ξb )
πℏ n=0

Now we apply the parity property of ψn ;

ψn = (−ξ) = (−1)n ψn (ξ)

Then the propagator will become

∞
e−iωτ n
r
mω −iN π/2 −iωτ /2 X
K(xb , tb ; xa , ta ) = e e exp [−(ξa2 + ξb2 )2 /2] n n!
Hn (ξa )Hn ((−1)N ξb ) (111)
πℏ n=0
2

32
 And if we recall Meher’s formula

∞
X Hn (ξa )Hn (ξb ) n
G(ξa , ξb ; z) = z
n=0
2n n!
2ξa ξb z − (ξa2 + ξb2 )z 2
 
1
= exp √
(1 − z 2 ) 1 − z2

Where the function G is almost analytical in the entire complex plane, with cuts from −∞ to −1 and
+1 to +∞.
Now if we write the propagator in terms of this G function then we get
r
mω iπ/2N −iωτ /2
K(xb , tb ; xa , ta ) = e e G(ξa , (−1)N ξb , e−iωτ ) exp [−(ξa2 + ξb2 )2 /2] (112)
πℏ

And
√ √ 2ξa ξb z − (ξa2 + ξb2 )z 2
 
1
zG(ξa , ξb ; z) = z exp √
(1 − z 2 ) 1 − z2
√ √ 2ξa ξb z − (ξa2 + ξb2 )z 2
 
2 2 1 2 2
zG(ξa , ξb ; z)e−(ξa +ξb )/2 = z exp √ e−(ξa +ξb )/2
(1 − z 2 ) 1 − z2
√ √ 2ξa ξb z − (ξa2 + ξb2 )z 2 (ξa2 + ξb2 )
 
2 2 1
zG(ξa , ξb ; z)e−(ξa +ξb )/2 = z exp − √
(1 − z 2 ) 2 1 − z2
√ √ (ξa2 + ξb2 )(1 + z 2 )
 
2 2 2ξa ξb z 1
zG(ξa , ξb ; z)e−(ξa +ξb )/2 = z exp 2
− 2
√ (113)
(1 − z ) 2(1 − z ) 1 − z2

For z = e−iωτ

z 1 1 + z2
= = −i cot ωτ
1 − z2 2i sin ωτ 1 − z2

Then the L.H.S. of the equation becomes

(ξa2 + ξb2 ) cot ωτ

 
1 ξa ξb
=√ exp −
2i sin ωτ i sin ωτ 2i

Now we will use this above expression to equation 108

mω e−iπN/2 ξa (−1)N ξb (ξ 2 + ξb2 ) cot ωτ

r  
K(xb , tb ; xa , ta ) = √ exp − a (114)
πℏ 2i sin ωτ i sin ωτ 2i

33
 Here ωτ = (ωT − N π) and sin ωτ = (−1)N sin ωT cos ωτ = (−1)N cos ωT

The term e−iπN/2 is the Maslov Factor.

The next obvious question is to ask what happens when ωT = N π

∞    
X 1
K= exp −i n + ωT ψn (xb )ψn∗ (xa )
n=0
2
∞    
X 1
= exp −i n + N π ψn (xb )ψn∗ (xa )
n=0
2
∞
X
= e−iN π/2 (−1)N n ψn (xb )ψn∗ (xa )
n=0
X∞
= e−iN π/2 ψn (xb )ψn∗ ((−1)N xa )
n=0
X∞
= e−iN π/2 ⟨xb |n⟩ ⟨n|(−1)N xa ⟩
n=0
−iN π/2
K=e ⟨xb |(−1)N xa ⟩

K = e−iN π/2 δ(xb − (−1)N xa ) (115)

So, K vanishes unless xb = (−1)N xa

Remember Z tb
m
S[x] = Scl + (Ẋ 2 − ω 2 X 2 )dt
2 ta
∞ 
mT X π 2 k 2

S[x] = Scl + − ω b2k
2
4 T2
k=1

T < π/ω for each term infinite sum, the coefficient is +ve. And no matter how we vary the path with
the endpoint fixed.

For all such variations, the additional term we get over and above Scl has to be positive.

We all know that the classical path is the least-action path. But classical mechanics doesn’t really say
that the classical path has to be the least action. It says the classical path has to be a path that is stationary.

12 General Quadratic Lagrangian

Consider the general quadratic Lagrangian of the form
1 1
L= mẋ2 + b(t)ẋx − c(t)x2 + f (t)ẋ − e(t)x (116)
2 2

34
 And the corresponding action is given as
Z tb  
1 2 1 2
S= mẋ + b(t)ẋx − c(t)x + f (t)ẋ − e(t)x dt (117)
ta 2 2

Now consider the 2 nd term of the action S.

Z tb Z tb  
2 tb d
b(t)ẋxdt = b(t)x |ta − b(t)x xdt
ta ta dt
Z tb Z tb Z tb
db 2
b(t)ẋxdt = b(t)x2 |ttba
− x dt − b(t)ẋxdt
ta ta dt ta
Z tb Z tb
db 2
2 b(t)ẋxdt = b(t)x2 |ttba − x dt
ta ta dt

So, apart from the boundary term this can be absorbed in the − 21 c(t)x2 term.

Consider the 4th term Z tb Z tb  

df
f (t)ẋdt = f (t)x|ttab − xdt
ta ta dt

Similarly, here we get a boundary term and the next term can be absorbed in the last term of the action.
Without any loss of generality, we can write general Lagrangian as
Z tb  
1 2 1 2
S[x] = dt mẋ − c(t)x − e(t)x (118)
ta 2 2

And the corresponding equation of motion is

mẍcl + c(t)xcl + e(t) = 0 (119)

Now we will do the same trick as we did before for the Harmonic oscillator and the free particle case.

x(t) = xcl (t) + X(t)

Where X(t) is the fluctuations about the classical path. With the boundary condition that is

X(ta ) = 0 = X(tb )

Therefore the action becomes

Z tb  
1 2 1 2
S[x] = dt m(ẋcl + Ẋ) − c(t)(xcl + X) − e(t)(xcl + X) (120)
tb 2 2
Z tb   Z tb   Z tb
1 2 1 2 m 2 1 2
S[x] = dt mẋcl − c(t)xcl − e(t)xcl + Ẋ − c(t)X − e(t)X dt + (mẋcl Ẋ − c(t)xcl X)dt
ta 2 2 ta 2 2 ta

35
Consider the last term of the above expression
Z tb
(mẋcl Ẋ − c(t)Xxcl )dt
ta
Z tb Z tb
= mẋcl X|ttba − mẍcl Xdt − c(t)xcl Xdt
ta ta

Using the boundary condition and the equation of motion we get

Z tb
= e(t)Xdt
ta

So, we left with

Z tb  
1 2 2
S[x] = Scl + mẊ − c(t)X dt (121)
ta 2

The propagator corresponding this action will be

Z X(tb )=0  Z tb  
i m 2 1
F (tb , ta ) = Dx exp Ẋ − c(t)X 2
(122)
X(ta )=0 ℏ ta 2 2

The propagator is given as

 
i
K(xb , tb ; xa , ta ) = F (tb , ta ) exp Scl (123)
ℏ

Then F (tb , ta ) is given as

 
 m  N2+1 Z ∞ YN N  2
iϵ X m Xj+1 − Xj cj
F (tb , ta ) = lim dXi exp  − Xj2  (124)
N →∞ 2πiℏϵ −∞ i=1 ℏ j=0 2 ϵ 2

Where Xj = X(tj ) and cj = c(tj )

And we can write the equation 120 in this form

 m  N2+1 Z N YN
F (tb , ta ) = lim dXi exp [iX T BN X] (125)
N →∞ 2πiℏϵ i=1 i=1

36
 Where BN and X is given as
   
2 −1 0 . . 0 c1 0 . . . 0
−1 2 −1 . . 0 0 c2 . . . 0
   
m 0 −1 2 −1 . . ϵ 
. . . .
A=  −  (126)
 .
2ϵℏ  . . . .
 .
2ℏ  . . .

 . . . . . . . .
0 0 . . . 2 N ×N 0 0 . . . cn N ×N

And we also know that,

∞
π N/2
Z
dx1 dx2 ...dxN exp [−X T AX] = p
−∞ det(A)

Therefore, " #
 m  N2+1 π N/2
F (tb , ta ) = lim p
N →∞ 2πiℏϵ det(−iBN )
r
m
F (tb , ta ) =
2πiℏf (tb , ta )

Where f (tb , ta ) is given as

 N
2ℏϵ
f (tb , ta ) = lim ϵ det(BN ) (127)
N →∞ m

And one of the properties of the determinant of a matrix is given as

det(λA) = λN det(A)

And I denote  N
2ℏϵ
pn = det(BN )
m
 
2ℏϵ
pn = det BN
m

Now we expand equation 122 along a particular row just like we did before to get a recursion relationship

ϵ2
 
pj+1 − 2 − cj+1 pj + pj−1 = 0 j = 2, 3, 4, ... (128)
m

with p0 = 1
pj+1 − 2pj + pj−1 cj+1
=− pj
ϵ2 m

37
 Clearly, the L.H.S looks like a second derivative term (just like we do in numerical analysis) To proceed
further we will adapt Gelfand and Yaglon process.

We define a function
ϵpj = f (tj , ta ) (129)
So, as ϵ → 0 we can write above recursion relation as

∂ 2 f (t, ta ) c(t)
=− f (t, ta ) (130)
∂t2 m

And the initial condition is given as

f (ta , ta ) = lim (ϵp0 )
ϵ→0

f (ta , ta ) = lim ϵ1 = 0
ϵ→0

And
∂f (t1 , ta ) f (t1 ) − f (t0 )
|t=ta = lim
∂t ϵ→0 ϵ
∂f (t1 , ta )
|t=ta = lim (p1 − p0 )
∂t ϵ→0

ϵ2
 
∂f (t1 , ta )
|t=ta = lim 2 − c1 − 1
∂t ϵ→0 m
∂f (t1 , ta )
|t=ta = 1
∂t

We can test the above result. If we set c = mω 2 then

∂2f
= −ω 2 f
∂t2
And the corresponding solution is

f (t, ta ) = A cos [ω(t − ta )] + B sin[ω(t − ta )]

1
The initial condition is given as A = 0 and B = ω
r
mω
F (tb , ta ) =
2πℏ sin ω(tb − ta )

The forced harmonic oscillator c = mω 2 , e(t) is arbitrary.

Since the Gelfand-Yaglon equation is independent of e(t) and f (ta , t0 ) and hence, F (ta , tb ) is the same
as above. The difference is Scl .

38
 Consider a family of classical trajectories staring from x = xa at t = ta , parameterized by p : x(p, t)
Each trajectory obeys the Euler-Lagrange E.O.M.
 
d ∂L ∂L
− =0
dt ∂ Ẋ ∂X

Differentiating both sides w.r.t p

!
d ∂ 2 L ∂X ∂2L ∂2L ∂2L
   
∂ Ẋ ∂X
+ − − =0
dt ∂ Ẋ 2 ∂p ∂X∂ Ẋ ∂ Ẋ∂X ∂p ∂x2 ∂p

Now we define
∂X ∂ Ẋ
=J = J˙
∂p ∂p

Then,
d ∂2L ˙
 2 
∂2L ∂2L ∂2L
 
d ∂ L
J + J+ J− J− J =0
dt ∂ Ẋ 2 dt ∂X∂ Ẋ ∂ Ẋ∂X ∂X∂ Ẋ ∂X 2
 2    2 
∂2L

d ∂ L ˙ d ∂ L
J + − J =0
dt ∂⊥X 2 dt ∂X∂ Ẋ ∂X 2

Since,
p
x(p, t) = xa + (t − ta ) + ...
m
And
1
J= (t − ta ) + O((t − ta )2 )
m
at
t = ta J =0
and
∂J 1
=
∂t m

For
1 1
L= mẊ 2 − c(t)X 2 − c(t)X (131)
2 2

∂2L ∂L ∂2L
=m = c(t) =0
∂ Ẋ 2 ∂X 2 ∂X∂ Ẋ

The equation for J becomes

mJ¨ + c(t)J = 0 (132)

39
 mJ satisfies the same second order ODE as fay , and the same initial conditions. So, for our quadratic
Hamiltonian.
fay = mJ
∂x(p, t)
fay =
∂p

So, our function f (tb , ta ) = fay (tb , ta )

∂xb
fay (tb , ta ) = m
∂p

And the action integral is given as Z tb

S= L(X, Ẋ)dt
ta

And on variation we get

Z tb   
∂L ∂L d ∂L
δS = δX|ttba + −
∂ Ẋ ta ∂X dt ∂ Ẋ

For Classical paths

δScl = pb δXb − pa δXa

Where  
∂Scl
pa = −
∂Xa
So, now we can write the function f as
∂Xb
f =m
∂pa
m
f= ∂pa
∂Xb

∂ 2 Scl
 
1 1
=− (133)
f m ∂Xa ∂Xb

Then the propagator for a quadratic Hamiltonian is completely determines by the classical action.
s
∂ 2 Scl
  
1 Scl nπ
K(xb , tb ; xa , ta ) = exp i − (134)
2iπℏ ∂Xb ∂Xa ℏ 2

40