Solutions to Feynman-Hibbs classical action problems

Peter Haggstrom
www.gotohaggstrom.com
mathsatbondibeach@gmail.com

December 10, 2021

1 Introduction
Feyman developed his path integral approach to quantum mechanics in his PhD thesis and later
he and Albert Hibbs produced a textbook on path integrals [1]. In this book, Feynman and
Hibbs pose a series of fundamental problems which relate to the principle of least action. To set
the scene for these problems they give a short derivation of the conditions for an extremum. I
will simply reproduce that derivation here ( see pages 26-27 of [1] ). More detailed expositions
of the calculus of variations concepts can be found in [2], [3].
The idea is to find the particular path x(t) out of all possible paths where a particle starts
at point xa at time ta and ends at point xb at time tb . In classical theory there is only once
such path but it quantum theory there are multiple trajectories each with a certain amplitude
(represented by a kernel in the relevant integral representation ) and the aim is to sum over all
these trajectories between the start and end points.
In classical theory there is a quantity S which is associated with each path and the classical path
x(t) is the one for which S is a minimum (more precisely an extremum ). Thus if we perturb the
path slightly the value of S is unchanged to first order.

1
The definition of S is as follows:

Z tb
S= L(ẋ, x, t) dt (1)
ta

where L is the Lagrangian for the system. For a particle of mass m subject to potential energy
V (x, t) (ie a function of position and time) the Lagrangian is:

m 2
S= ẋ –V (x, t) (2)
2

We suppose that we vary the path x(t) a little by δx(t) with condition that the end points be
fixed ie:

δx(ta ) = δx(tb ) = 0 (3)

For x(t) to be an extremum we must have:

δS = S[x + δx] − S[x] = 0 (4)

to the first order of δx(t). Thus we have:

2
 Z tb
S[x + δx] = L(ẋ + δ ẋ, x + δx, t) dt
ta
Z tb " #
∂L ∂L
= L(ẋ, x, t) + δx + δ ẋ dt
ta ∂x ∂ ẋ
Z tb " #
∂L ∂L
=S[x] + δx + δ ẋ dt
ta ∂x ∂ ẋ
Z tb " #tb Z
tb
∂L ∂L d  ∂L 
=S[x] + δx dt + δx − δx dt using integration by parts
ta ∂x ∂ ẋ ta dt ∂ ẋ
ta
" #tb Z
tb
∂L h d  ∂L  ∂L i
=S[x] + δx − δx − dt
∂ ẋ ta dt ∂ ẋ ∂x
ta
" #tb Z
tb
∂L h d  ∂L  ∂L i
∴ δS = δx − δx − dt
∂ ẋ ta dt ∂ ẋ ∂x
ta
(5)

" #tb
But from (3) we must have that δx ∂L
∂ ẋ = 0 so that the extremum is the curve along which
ta
the following condition is satisfied:

d  ∂L  ∂L
− =0 (6)
dt ∂ ẋ ∂x

More detailed verifications of this condition can be found at pagesR 9-17 of [2]. Feynman and
Hibbs denote the extreme value of the classical action integral S = L dt by Scl .

2 Problem 2-1
m 2
For a free particle L = 2 ẋ . Show that the action Scl corresponding to the classical motion of
a free particle is:
m (xb − xa )2
Scl = (7)
2 tb − ta

2.1 Solution to Problem 2-1

Using condition (6) we see that:

d  ∂L  ∂L
− = mẍ = 0 (8)
dt ∂ ẋ ∂x

So ẍ = 0 which implies that x(t) = vt + d. With x(ta ) = xa and x(tb ) = xb we have that:

3
 !
xb − xa
x(t) = t + xa tb − xb ta (9)
tb − ta

xb −xa
On this path the velocity v(t) is constant and is simply tb −ta .

The classical action is given by:

Z tb
m 2
Scl = ẋ dt
ta 2
Z tb
m
= v(t)v(t) dt
2 ta
" #
m h itb Z tb
= v(t)x(t) − x(t)dv(t)
2 ta ta
" #
m h itb Z tb (10)
= v(t)x(t) − x(t)ẍ(t) dt
2 ta ta
mh i
= v(tb )x(tb ) − v(ta )x(ta ) since ẍ(t) = 0
2
m h xb − xa  x − x 
b a
i
= x(tb ) − x(ta )
2 tb − ta tb − ta
m (xb − xa )2
=
2 tb − ta

3 Problem 2-2
m 2
For a harmonic oscillator L = 2 (ẋ − ω 2 x2 ). With T equal to tb − ta show that the classical
action is:

mω  2
(xb + x2a ) cos ωT − 2xb xa

Scl = (11)
2 sin ωT

3.1 Solution to Problem 2-2

Using condition (6) we see that:

d
(mẋ) + mω 2 x = mẍ + mω 2 x = 0 =⇒ ẍ + ω 2 x = 0 (12)
dt

Hence a solution to (12) is given by:

x(t) = A sin ωt + B cos ωt (13)

0 0
Note that t = t − ta is also a solution to (12). With t = ta , t = 0 and x(0) = xa and with
0
t = tb , t = tb − ta and x(T ) = xb . Therefore

x(0) = xa = B (14)

4
Hence:

xb = X(T ) =A sin ωT + B cos ωT

=A sin ωT + xa cos ωT (15)
xb − xa cos ωT
∴A=
sin ωT

With t 7→ t − ta and T = tb − ta we have:

Z T
m
Scl = (ẋ2 − ω 2 x2 ) dt
2 0
T
m T 2 2
Z Z
m 2
= v (t) dt − ω x dt
2 2 0
"0 #
m h iT Z T m T 2 2
Z
= v(t)x(t) − x(t)ẍ(t) dt − ω x dt using (10) for the first integral
2 0 0 2 0
" #
m h iT Z T m T 2 2
Z
2
= v(t)x(t) − x(t)(−ω x) dt − ω x dt using (12) for the first integral
2 0 0 2 0
m 
= v(T )x(T ) − v(0)x(0)
2
m 
= ẋ(T )x(T ) − ẋ(0)x(0)
2
(16)

But from (13)-(15):

!
xb − xa cos ωT
x(t) = sin ωt + xa cos ωt (17)
sin ωT

Hence:
!
xb − xa cos ωT
ẋ(T ) = ω cos ωT − xa ω sin ωT (18)
sin ωT

and
!
xb − xa cos ωT
ẋ(0) = ω (19)
sin ωT

Putting it all together in (16) we have (noting that x(T ) = xb and x(0) = xa ):

5
 m 
Scl = ẋ(T )x(T ) − ẋ(0)x(0)
2" ! !#
m xb − xa cos ωT xb − xa cos ωT
= ωxb cos ωT − xa xb ω sin ωT − xa ω
2 sin ωT sin ωT
mω h i
(20)
= xb (xb − xa cos ωT ) cos ωT − xa xb sin2 ωT − xa (xb − xa cos ωT )
2 sin ωT h
mω i
= x2b cos ωT − xa xb cos2 ωT − xa xb sin2 ωT − xa xb + x2a cos ωT
2 sin ωT h
mω i
= (x2b + x2a ) cos ωT − 2xa xb
2 sin ωT

4 Problem 2-3
m 2
Find Scl for a particle under a constant force f , that is, L = 2 ẋ + f x.

4.1 Solution to Problem 2-3

Using condition (6) we see that:

mẍ = f (21)

Thus:

f 2
x(t) = t + ct + d (22)
2m

is a solution to (21) and as in Problem 2-2 we note that x(t − ta ) is also a solution to (21).
We have from (22):

x(ta − ta ) = x(0) = xa = d (23)

and
f
x(tb − ta ) =xb = (tb − ta )2 + c(tb − ta ) + xa
2m
(24)
xb − xa f
∴c= − (tb − ta )
tb − ta 2m

Hence since x(t − ta ) is a solution to (21) we have with τ = t − ta :

f 2 h xb − xa f i
x(τ ) = τ + − (tb − ta ) τ + xa (25)
2m tb − ta 2m

Therefore:

fτ xb − xa f
ẋ(τ ) = + − (tb − ta ) (26)
m tb − ta 2m

6
Hence we have that:

m 2
L(τ ) = ẋ (τ ) + f x(τ )
2
mhfτ xb − xa f i2 f 2 τ 2 hx − x f i
b a
= + − (tb − ta ) + + fτ − (tb − ta ) + xa f
2 m tb − ta 2m 2m tb − ta 2m
m h f 2 τ 2  xb − xa f 2 2f τ  x − x
b a f i
= + − (t b − ta ) + − (t b − t a ) + (27)
2 m2 tb − ta 2m m tb − ta 2m
f 2τ 2 hx − x
b a f i
+ fτ − (tb − ta ) + xa f
2m tb − ta 2m
f 2τ 2 m h xb − xa f i2 hx − x
b a f i
= + − (tb − ta ) + 2f τ − (tb − ta ) + f xa
m 2 tb − ta 2m tb − ta 2m

The classical action is thus given by:

( )
tb −ta
f 2τ 2 m h xb − xa hx − x
Z i2
f b a f i
Scl = + − (tb − ta ) + 2f τ − (tb − ta ) + f xa dτ
0 m 2 tb − ta 2m tb − ta 2m
" #tb −ta
f 2τ 3 m h xb − xa f i2 hx − x
b a f i
2
= + − (tb − ta ) τ + f − (tb − ta ) τ + f xa τ
3m 2 tb − ta 2m tb − ta 2m
0
f 2 (tb − ta )3 m h xb − xa f i2 hx − x
b a f i
= + − (tb − ta ) (tb − ta ) + f − (tb − ta ) (tb − ta )2 + f xa (tb − ta )
3m 2 tb − ta 2m tb − ta 2m
f 2 (tb − ta )3 m (xb − xa )2 f f2
= + − (xb − xa )(tb − ta ) + (tb − ta )3 + f (xb − xa )(tb − ta )
3m 2 (tb − ta ) 2 8m
f2
− (tb − ta )3 + f xa (tb − ta )
2m
−f 2 m (xb − xa )2 f
= (tb − ta )3 + + (xb − xa )(tb − ta ) + f xa (tb − ta )
24m 2 (tb − ta ) 2
2 2
−f m (xb − xa ) h xb − xa i
= (tb − ta )3 + + f (tb − ta ) xa +
24m 2 (tb − ta ) 2
2 2
−f m (xb − xa ) f (xa + xb )(tb − ta )
= (tb − ta )3 + +
24m 2 (tb − ta ) 2
2 2
−f 3 m (xb − xa ) f (xa + xb )T
= T + + with T = tb − ta
24m 2 T 2
(28)

5 Problem 2-4
Classically, the momentum is defined as:

∂L
p= (29)
∂ ẋ
Show that the momentum at a final point is:

7
  ∂L  ∂Scl
=+ (30)
∂ ẋ x=xb ∂xb

while the momentum at an initial point is:

 ∂L  ∂Scl
=− (31)
∂ ẋ x=xa ∂xa

Hint: Consider the effect on (5) of a change in the end points.

5.1 Solution to Problem 2-4

6 0 since we are varying xb . This means
In equation (5), ta is fixed so δx(ta ) = 0 but δx(tb ) = δxb =
that while the value of the quantity in (32) vanishes at the initial point there is a contribution
from the variation in the final point:

" # tb
∂L ∂L
δScl = δx = δxb × −0 (32)
∂ ẋ ∂ ẋ tb
ta

Hence we can write in the limit:

∂Scl  ∂L 
+ = (33)
∂xb ∂ ẋ x=xb

When we keep the final point fixed with δx(tb ) = 0 and vary the initial point we get:

" #tb
∂L ∂L
δScl = δx = −δxa × (34)
∂ ẋ ∂ ẋ ta
ta

Hence we can write in the limit:

∂Scl  ∂L 
− = (35)
∂xb ∂ ẋ x=xa

We can check the logic of these results by applying them to Problems 2-1 and 2-2. For example,
in Problem 2-1 we have that:

∂Scl (xb − xa )
=m (36)
∂xb tb − ta
xb −xa
which is clearly a momentum since on the classical path the velocity is a constant tb −ta . Fur-
thermore from (9) we have that:
!
xb − xa
x(t) = t + xa tb − xb ta (37)
tb − ta

8
so that

xb − xa
ẋ(t) = (38)
tb − ta

Thus:
∂L (xb − xa )
= mẋ = m (39)
∂ ẋ tb − ta
∂L
noting that there is no explicit dependence on x for the evaluation of ∂ ẋ at x = xb .
With Problem 2-2 we have that:

∂Scl mω
mω

= 2xb cos ωT − 2xa = xb cos ωT − xa (40)
∂xb 2 sin ωT sin ωT

∂L
= mẋ (41)
∂ ẋ

From (17) we have that:

 x − x cos ωT 
b a
ẋ(t) = ω cos ωT − xa ω sin ωT
sin ωT (42)
 x cos ωT − x 
b a
=ω
sin ωT

Hence:

 x cos ωT − x  ∂S
b a cl
mẋ(t) = mω = (43)
sin ωT ∂xb
∂L
Again there is no explicit dependence on x for evaluation of ∂ ẋ at x = xb .

6 Problem 2-5
Classically the energy is defined as:

E = ẋp − L (44)

Show that the energy at a final point is:

 ∂L  ∂Scl
ẋb − L(xb ) = − (45)
∂ ẋ x=xb ∂tb

while the energy at an initial point is:

∂Scl
+ (46)
∂ta

9
Hint: A change in the time of an end point requires a change in path, since all paths must be
classical paths.

6.1 Solution to Problem 2-5

R tb
A naive approach to this problem might run this way. Since Scl = ta
L(ẋ, x, t) dt one might
argue that:

Z tb
∂Sc ∂
= L(ẋ, x, t) dt
∂t ∂t ta
Z tb
∂L(ẋ, x, t) (47)
= dt
ta ∂t
=L(ẋ, x, tb ) − L(ẋ, x, ta )

which is not −E ie negative energy.

A diagram helps:

There is a change in the classical path caused by the change in time of the path while holding
the initial endpoints xa and xb fixed. Note that η(ta ) = 0 but η(tb ) 6= 0. The total change in the

10
action involves two components: first, the bit of the classical path between ta and tb and then,
secondly, the bit from tb to tb + ∆t. The new path can be represented by:

x(t) = xcl (t) + η(t) (48)

subject to η(ta ) = 0.
We now go through the process of working out δScl where η(t) is the deviation function. In
terms of the action integral we have:

Z tb +∆t Z tb Z tb +∆t
Scl = L dt = L dt + L dt = S1 + S2 (49)
ta ta tb

So that:

δScl = δS1 + δS2 (50)

In working out δS will deal with the two integrals on the RHS of (49) separately:

Z tb

Z tb

δS1 = L ẋcl (t) + η̇(t), xcl (t) + η(t), t dt − L ẋcl (t), xcl (t), t dt
ta ta
Z tb Z tb

∂L ∂L 

= L ẋcl (t), xcl (t), t + η(t) + η̇(t) dt − L ẋcl (t), xcl (t), t dt
ta ∂x ∂ ẋ ta
Z tb 
∂L ∂L 
= η(t) + η̇(t) dt
ta ∂x ∂ ẋ cl
Z tb Z tb
∂L h ∂L itb d  ∂L  (51)
= η(t) dt + η(t) − η(t) dt
ta ∂x ∂ ẋ ta ta dt ∂ ẋ
Z tb !
∂L d  ∂L  h ∂L itb
= − η(t) dt + η(t)
ta ∂ ẋ dt ∂ ẋ ∂ ẋ ta
| {z }
=0 since this is a classical path
h ∂L i
= η(t) since η(ta ) = 0
∂ ẋ tb

Now for δS2 , since the end points are fixed, the Lagrangian can be taken as fixed at time tb and
the variation is due to the extra travel between tb and tb + ∆t. If we had any continuous function
R a+∆t
f (x) we could estimate the integral a f (t) dt by ∆t × f (a) where ∆t is very small. In essence
this is a continuity argument since for t very close to a, f (t) is very close to f (a). Thus:

Z tb +∆t
δS2 = L(ẋcl (tb ), xcl (tb ), tb ) dt = ∆t L(ẋcl (tb ), xcl (tb ), tb ) = ∆t L(tb ) (52)
tb

Hence:

11
 h ∂L i
δScl = η(t) + ∆t L(tb ) (53)
∂ ẋ tb

Now we can get an expression for η(tb ) as follows with the help of the diagram above, noting
that the variation η(t) is negative and the gradient at xcl (ta ) is positive. Thus:

xcl (tb + ∆t) − xcl (tb ) = −η(tb ) (54)

Therefore:

η(tb )
ẋcl (tb ) = − (55)
∆t
or

η(tb ) = −ẋcl (tb ) ∆t (56)

Plugging (56) into (53) we have:

h ∂L i
δScl = − ẋcl (tb ) ∆t + ∆t L(tb )
∂ ẋ tb
(57)
h ∂L i
= L − ẋcl ∆t
∂ ẋ tb

Finally we get:

∂Scl h ∂L i
= L − ẋcl (58)
∂tb ∂ ẋ tb

whch is the same as (45) on multiplying both sides of (58) by −1. Equation (58) can be expressed
in the Hamiltonian formulation as:

∂Scl
= −H(tb ) (59)
∂tb

To calculate the second part of the question we consider this diagram:

12
This time (49) becomes:

Z tb Z tb Z ta +∆t
Scl = L dt = L dt − L dt = S1 − S2 (60)
ta +∆t ta ta

We approximate δS2 as before (ie see (52) ):

δS2 = ∆t L(ta ) (61)

For completeness the derivation of δS1 is set out below although one could go straight to it
because it follows the same lines as before:

13
 Z tb

Z tb

δS1 = L ẋcl (t) + η̇(t), xcl (t) + η(t), t dt − L ẋcl (t), xcl (t), t dt
ta ta
Ztb Z tb

∂L ∂L 

= L ẋcl (t), xcl (t), t + η(t) + η̇(t) dt − L ẋcl (t), xcl (t), t dt
ta ∂x ∂ ẋ ta
Z tb 
∂L ∂L 
= η(t) + η̇(t) dt
ta ∂x ∂ ẋ cl
Z tb Z tb
∂L h ∂L itb d  ∂L  (62)
= η(t) dt + η(t) − η(t) dt
ta ∂x ∂ ẋ ta ta dt ∂ ẋ
Z tb !
∂L d  ∂L  h ∂L itb
= − η(t) dt + η(t)
ta ∂ ẋ dt ∂ ẋ ∂ ẋ ta
| {z }
=0 since this is a classical path
∂L i
= − η(ta ) since η(tb ) = 0
∂ ẋ ta

Thus:
h ∂L i
δScl = − η(ta ) − ∆t L(ta ) (63)
∂ ẋ ta

We have from the diagram (remembering that the variation η(t) is negative and we actually have
a positive derivative at ta ):

xcl (ta + ∆t) − xcl (ta ) = −η(ta ) (64)

Therefore:

η(ta )
ẋcl (ta ) = − (65)
∆t
or

− η(ta ) = ∆t ẋcl (ta ) (66)

Hence:

h ∂L i
δScl = ∆t ẋta − ∆t L(ta )
∂ ẋ ta
h ∂L i (67)
= ẋta − L(ta ) ∆t
∂ ẋ ta

Finally we have:

∂Scl ∂L
= ẋta − L(ta ) = H(ta ) (68)
∂ta ∂ ẋ ta

14
7 References
[1] Richard P. Feynman and Albert R. Hibbs , “Quantum Mechanics and Path Integrals”, Emended
by Daniel F. Styer, Dover, 2010

[2] I.M. Gelfand and S.V. Fomin, “Calculus of Variations”, translated and Edited by Richard
A. Silverman, Dover, 2000.
[3] Cornelius Lanczos, ”The Variational Principles of Mechanics”, Fourth Edition , Dover, 1970

8 History
Created
31 December 2018

10 December 2021: Typo corrected at the bottom of page 3. Originally it was x(t) = xt + d and
should have been x(t) = vt + d.

15