Sei sulla pagina 1di 20

Chapter 2

Quantum Mechanical
Path Integrals
In this chapter the s t a n d a r d version of nonrelativistic q u a n t u m mechanics
will be reformulated in terms of a p a t h integral. T h e p a t h integral will be
derived as a limiting expression from the fundamental structure of nonrela­
tivistic q u a n t u m mechanics, and will be valid for a wide variety of systems.
In this sense the p a t h integral is a representation of q u a n t u m mechanical
amplitudes equivalent to the usual wave mechanical or m a t r i x formulations
of these same amplitudes. This is extremely useful since there is no ambi­
guity in the p a t h integral's definition. For this reason the form of the p a t h
integral derived in this case serves as a motivating form for a generalization
of the p a t h integral to all q u a n t u m processes.
In Sec. 2.1 the relevant aspects of basic q u a n t u m mechanics are re­
viewed. In Sec. 2.2 these are used to find the p a t h integral form for a
q u a n t u m mechanical amplitude. In S e c 2.3 the idea of the p a t h integral
as a "sum over histories" is presented as a conceptual generalization of
the results of Sec. 2.2. Some of the formidable mathematical difficulties
associated with this generalization are sketched.

2.1 Quantum Mechanics


T h e Hamiltonian formulation of classical mechanics, discussed in Sec. 1.3,
serves as the starting point for the operator form of q u a n t u m mechanics
[1, 2, 3]. In t h e Hamiltonian formalism the generalized coordinates, qj, and
their canonically conjugate momenta, pj, are the fundamental mechanical

31
32 Q u a n t u m Mechanical P a t h Integrals

"observables" of a particle or system. In q u a n t u m mechanics these become


operators defined on an abstract Hubert space Ti t h a t represents all possible
q u a n t u m mechanical configurations in which the system may be observed.
These operators, denoted Qj and P j , are assumed to obey the commutation
relation
[Q P ]
h k= QjPk - PkQj = ihSjk , (2.1)
where h is Planck's constant. T h e elements of the Hubert space Ti are
called states and are written IV*)· T h e operators Q and Ρ are assumed to
have a complete set of eigenstates | q) and | ρ) such t h a t

QSU) = *\9)*PJ\P)=PJ\P)> (2-2)

so t h a t qj and pj are the eigenvalues of the respective operators.


T h e Hilbert space Ti is equipped with an inner product. If | φ) and | φ)
are any two states in Ti, then their inner product is denoted ( φ \φ), and
obeys
(φ\φ) = (φ\φ)·, (2.3)

so t h a t its form is analogous to the inner product for the functions of Sec.
1.3 or the vectors of Sec. 1.4. T h e inner products of the \ p) and | q) states
are assumed to obey

<*l?'> = * " ( ? - ? ' ) > (Ρ\ρ') = δ (ρ-Ρ'),


η
(2-4)

so t h a t these states are orthonormal in the continuum sense. T h e Dirac


delta appearing in (2.4) is understood to have the test function space L , 2

the space of all square integrable functions. It is also assumed t h a t these


states are complete, so t h a t the states \p) and \q) b o t h span the Hilbert
space Ti. In the continuum normalization of (2.4) this means t h a t

Jd p\p)( \
n
P = l, | < ί ? | ? } < ? 1 = 1>
η
(2.5)

where the limits on the integrals of (2.5) must run over the entirety of the
phase space available to the system. Coupling the algebra of (2.1) with t h e
inner product (2.4) gives the coordinate representation of the m o m e n t u m
operator

( \ \ >)
q Pj q = -ih-£-6 ( - ').
n
q q (2.6)

This, in turn, gives the inner product


Sec. 2.1 Quantum Mechanics 33

T h e factors present in (2.7) are necessary in order t h a t

(q\q') = Γ d p(q\p)(p\ ')


n
q

J — OO
dp
•I.
n

_ v - ( f - i ' ) / » = S (q
e
n
- q') . (2.8)
-oo (2*ft)

T h e formal similarity of (2.1) t o (1.50) indicates t h a t the classical me­


chanical Poisson brackets of two observables, A(p q) and J3(p, g), must b e
}

replaced in q u a n t u m mechanics by commutators in the manner

{A(Pl q), B(p, q)} p>q - i [ A ( P , Q ) , B(P Q)] .


t (2.9)

If Ο is a q u a n t u m mechanical observable, i.e., some function of Q, P , and


possibly i, then (1.49) and the formal identity (2.9) gives

i h ^ = ih?£ + [0,H), (2.10)

where Η = H(P,Q>i) is the Hamiltonian of the system under analysis.


T h u s , the time development of a q u a n t u m mechanical observable is driven
by the Hamiltonian of the system, just as it is in classical mechanics.
T h e physical system is represented by a s t a t e | φ) in t h e Hubert space
7i and t h e state is assumed to b e of unit length, so t h a t (φ \φ) = 1, It
y

follows from (2.10) t h a t the time development of the state | φ) is given by


the Schrodinger equation

H\rP,t) s = ih^\x!>,t) . s (2.11)

T h e s t a n d a r d differential version of the Schrodinger equation used in wave


mechanics may be obtained from (2.11) by forming t h e inner product with
the position eigenstate |<j), and identifying t h e wave-function φ^,ί) =
( 5 I Vs * )s* This results in the differential equation

H{-ih^q)^{q,t) = ih^rP(q,t) . (2.12)

W h e n the system is in the state | φ), the expectation value of a measure­


ment of the observable 0 ( P , Q) is given by

(O) = (φ 10(P,<2)\Φ) = Jdqφ·( ,t)0(-ih£- 2 ,q)φ( ,t) ,


ς (2.13)
34 Q u a n t u m Mechanical P a t h Integrals

leading t o the interpretation of 1^(9)| as a probability density. T h e nor­


2

malization of the state,

(φ\φ)= j dq\t!>(q,t)\ 2
= l, (2.14)

is simply a reflection of the fact t h a t t h e total probability of observing the


particle must be unity.
In the Schrodinger picture of q u a n t u m mechanics the states, \ψ) are 3}

manifestly time dependent, while the observables are not, so t h a t , if Η has


no explicit time dependence, the s t a t e at time t evolves from t h e s t a t e at
t = 0 according to

\φ^) =αρ^~Η^\φ) .
3 3 (2.15)

If i f is time dependent, then (2.15) must b e replaced with the time-ordered


integral of H. This is denoted by

\φ,t)s = Τ { e x p [ - 1 j f dr ( r ) ] } \φ ) s . (2.16)

Time ordering means expanding t h e exponential in (2.16) so t h a t the oper­


ators are ordered sequentially according t o their time argument, with t h e
latest at the left and earliest at t h e right. This is most easily expressed by
introducing the step function 0 ( r ) , which is defined as

1 if r > 0,
θ(τ) = { (2.17)
0 if r < 0.

E x e r c i s e 2 . 1 : Show t h a t θ(τ) has t h e integral representation

- F t -^—^ , (2.18)
.oo 2πί ω-ic '
v

and t h a t

p(t) = 6{t). (2.19)

Using t h e step function, the time-ordered product of two time-dependent


operators can be written

T { A ( < ! ) B ( i ) } = 9(h - t )A(t )B(t )


2 3 1 2 + θ(ί 2 - *ι)£(ί )Λ(ίι) .
2 (2.20)
S e c . 2.2 T h e P a t h Integral Derived 35

T h e generalization of (2.20) to the time-ordered product of many operators


is obvious.
In the Heisenberg picture of q u a n t u m mechanics the observables, O , h

are time dependent, and from (2.10), they satisfy

ihj O„
t = ih^O H + [O ,H)
h . (2.21)

T h e equality of the matrix elements in the respective pictures gives

H(Φ \0„(t)\φ) = (Φ^Η 5 \O \φ t)


s } s , (2.22)

so t h a t differentiation of (2.22) gives

i h ^ = [0„,H]. (2.23)

If Η has no explicit time dependence, then the Heisenberg picture operators


of (2.23) at time t can written in terms of the operators at t = 0 in t h e
form
0 (t) = e l O e- l
H .
iHi n
H
iHi h
(2.24)
If Η is explicitly time dependent, then it is necessary to use time ordering,
as in (2.16). T h e Heisenberg picture states, | φ ) , are time independent. It H

is clear t h a t the Schrodinger picture and Heisenberg picture can b e chosen


to coincide at some specific time. In what follows it will be assumed t h a t
this time is t = 0, and any state with no explicit time dependence displayed
can be considered to be either a Heisenberg picture s t a t e or a Schrodinger
picture state at t = 0.
In t h e p a t h integral formulation of q u a n t u m mechanics the object of ut­
most interest is the transition amplitude. If the system is in the Schrodinger
picture s t a t e \ <j>,t ) at the time ί , then the transition amplitude to the
a α

state I φ,ίι,) at the time ti is defined as

Ζ(ψ,φ) = (ψ,η\φ,ί ) α . (2.25)

T h e transition amplitude then gives the probability of the system transiting


between the respective states as Ρ = \Z\ , so t h a t Ζ describes a quantum
2

process. In the next section the transition amplitude is given a p a t h integral


representation.

2.2 The Path Integral Derived


In this section the transition amplitude will be given a p a t h integral rep­
resentation by applying the results of the previous section. T h e pivotal
36 Q u a n t u m Mechanical P a t h Integrals

idea, first noted by Dirac [4], is t h a t the infinitesimal q u a n t u m mechanical


transition amplitude is governed by t h e value of the classical action. Us­
ing fairly intuitive arguments, this idea was developed by Feynman [5, 6]
into the p a t h integral representation of the finite transition amplitude, and
later presented in detail in the book by Feynman and Hibbs [7]. T h e p a t h
integral derived in this section will be identical in form to t h a t first devel­
oped by Feynman. However, a derivation based on the canonical structure
of q u a n t u m mechanics has the advantage of demonstrating t h a t the p a t h
integral must yield results identical to those obtained by the s t a n d a r d meth­
ods of wave mechanics or m a t r i x manipulation. For t h a t reason it is an
immensely valuable heuristic device for testing generalizations of the p a t h
integral form to systems where a rigorous derivation is impossible.
It is therefore very important to list the assumptions t h a t go into this
derivation, so t h a t they may be kept in mind during any generalization.
First, it will be assumed t h a t the limits on the statement of completeness
(2.5) are dboo. Second, it will be assumed t h a t the Hamiltonian Η has no
explicit time dependence, although this is not necessary. Third, and this
is for simplicity of notation, the system will be considered to be t h a t of a
one-dimensional single particle moving in a potential.
T h e first step in the derivation of the p a t h integral [3, 8,9] is to construct
the "instantaneous" eigenstates, \q,t) and of the Heisenberg picture
operators Q(t) and P(t). These are defined by

|«,0 = «"""Ί«>, \p,t)=e i \p).


iHt H
(2.26)

It is to be noted t h a t these states are not Schrodinger picture states. How­


ever, these states are complete, since it follows from (2.5) t h a t
oo
dq\q,t){q,t\
/ •OO

= exp(iHt/h) (^j dq\q)(q\ j


S
exp(-iHt/h) = 1. (2.27)

T h e state | q,t) is an eigenstate of the Heisenberg picture operator Q(t) in


the sense t h a t
Q(t)\q,t) = q\q,t) . (2.28)
These states also have the valuable property t h a t , for \ ψ,ί) a Schrodinger
picture state,
{q,-t\* t) = (q\1>)=1>(q).
i (2.29)
A similar pair of statements holds for the instantaneous m o m e n t u m eigen­
states.
Sec. 2.2 T h e P a t h Integral Derived 37

These states can be used to derive the p a t h integral form for the transi­
tion element between Schrodinger picture states. T h e object of interest, Z,
is defined as the inner product of the instantaneous eigenstates at different
times, so t h a t
Z(q ,t ,q ,t )
a a b b = (q ,t \q ,ta) b b a , (2.30)

where it is assumed t h a t t > t . Knowledge of the form for Ζ allows


b a

the calculation of the more general form (2.25), since it follows t h a t t h e


transition element between Schrodinger picture states is given by

oo
dq dq (\l>,-t
a b b \q >t )(q ,t
b b b b \ q ,t
a a )(q ,t
a a | φ, —t )
a

/ -oo
CO
dqadq ip*(q mq )Z(q ,t ,q ,t )
b b a a a b b , (2.31)
/
•oo
where properties (2.27) and (2.29) have been used. Once the initial and
final states of the system are specified by the normalized forms for $(q)
and <^(g), the system propagates in time from φ to φ through t h e function
Z . For this reason the transition element (2.30) is sometimes referred to
as the propagator, since it contains all the information regarding the time
development of the system.
T h e time interval t — t is first partitioned into Ν infinitesimal steps of
b a

duration e = (t — t )/N, where the limit Ν —• oo is understood in every­


b a

thing t h a t follows. Next, JV—1 complete sets of intermediate instantaneous


I q) eigenstates are inserted into the transition element sequentially at each
of the respective times t = t + ne. This gives n a

oo
dq - - r f g ^ . i ( q , t
x b b Ι^-υ^-ι)
/ •oo
χ(Αν-ι,*λγ-ι|··' Ιϊβι'β) r ( -32)
2

so t h a t the whole transition element has been reduced to the product of


Ν transition elements, which are infinitesimal in the sense t h a t their time
difference approaches zero. Each one of the infinitesimal transition elements
may now be analyzed. Since the time difference is e between the two states,
it follows from the definition (2.26) t h a t the j t h element is given by

{q , j+1 t j + 1 I qj ,<;) = < q j+1 \ -**(™>/»


β | qj ) . (2.33)

In order to proceed further it is necessary to select a convention t h a t will


be used consistently to reduce all the infinitesimal elements. To evaluate
the m a t r i x element of the exponentiated Hamiltonian, the exponential must
38 Q u a n t u m Mechanical P a t h Integrals

be expanded in a power series. After the expansion all the Ρ operators will
be moved to the left and all the Q operators will be moved to the right.
For want of a better n a m e this will be referred to as coordinate ordering.
T h e final result of coordinate ordering a product of P ' s and Q s will be y

t h a t all the Ρ operators lie on the left side of the expression and all t h e
Q operators will lie on t h e right side of the expression. T h e coordinate
ordering operation will be denoted by C{· · · } , so t h a t , for example,

C{PQP } 2
= PQ 3
. (2.34)

Obviously, t h e opposite convention can be selected, and it is n a t u r a l


to consider the possibility t h a t t h e existence of different ordering schemes
might induce some ambiguity in the resulting p a t h integral [10]. T h a t t h e
same form results is given as Exercise 2.4. T h e advantage of coordinate
ordering is t h a t an arbitrary function of Q and P , denoted f(Q,P), has
the m a t r i x element

< ΡI C{ f(Q, P)} I q > = f(q, p) ( ρ I q) , (2.35)

so t h a t a coordinate-ordered function of t h e operators may be reduced t o a


c-number (classical number) function by taking its m a t r i x element. Result
(2.35) can be demonstrated by using a power series representation of t h e
c-number function / and by noting t h a t coordinate ordering suppresses t h e
presence of any commutators between Q and P.
T h e next step in evaluating (2.33) is t o observe t h a t

e-ieH(P,Q)/n = c { e-i<H(P,Q),n J + ^ ( 2 3^

T h e proof of this statement for a general Hamiltonian is given as Exercise


2.2. However, a demonstration for the particularly simple b u t physically
meaningless form Η = aP + bQ is instructive because of its use of the
Baker-Campbell-Hausdorff theorem. T h e operator version of this theorem
is identical to the form proved in S e c 1.4 for matrices. Using this theorem
shows t h a t
e-ie(aP+bQ)/h = ^ieaP/n^iebQ/h ie ab/2h

2
^ (2.37)

so t h a t the effects of the commutator (2.1) are 0(c ) and are therefore irrel­ 2

evant in the e —» 0 limit. Since the commutator represents q u a n t u m effects,


or quantum corrections, this verifies the intuitive notion t h a t , for infinites­
imal time periods, t h e time development of t h e system is overwhelmingly
dominated by classical dynamics. T h e generalization of this result, showing
t h a t t h e 0(c ) t e r m in (2.36) is ignorable, can be accomplished by using
2
S e c . 2.2 T h e P a t h Integral Derived 39

the Trotter product formula [11, 12]. If A and Β are any two b o u n d e d
operators, t h e Trotter product formula gives

exp[t(A + B)] = ^lim^ exp (^^j e x


P • ( 2 3 8
)

E x e r c i s e 2.2: Prove (2.38) and a d a p t the theorem to demonstrate t h a t ,


in t h e e —» 0 limit, the commutators generated by a Hamiltonian of t h e
form Η = Ρ /2m + V(Q) are irrelevant t o the evaluation of (2.33).
2

These results show t h a t the infinitesimal transition element (2.33) be­


comes, for e « 0,

CO
dPi{*i+i\Pi){Pi\e- ' \<li) UH{P Q),n

/ -CO

« Γ d P j e- i i H
^f (q
h
j + 1 IPJUPJUJ). (2.39)
J — OO
This infinitesimal element can be simplified further by using (2.7) t o give

(qj+i IPJ )(PJ\ <u > = 2 ^ e ,


' P i ( i i + l
" i i ) M
· ( - °)
2 4

Using t h e fact t h a t qj is the coordinate value associated with the s t a t e at


time tj allows the formal identification

(2.41)
H?,7<«+>-«) = $ » «
Using this identification, t h e infinitesimal m a t r i x element can be written

(qj+i,tj+i\qj,tj) * J ^ exp j-^c ftfl-#(ρ^·)]}

= «Φ · ·
(2 42)

where
£(Pi IQj)= Pj Qj ~ (Pj>9j) H
· (2.43)
First discussed in Sec. 1.3, £(pj, qj) is the Lagrangian density in Hamilton's
formulation of the classical mechanical system.
40 Quantum Mechanical P a t h Integrals

T h e finite transition element Ζ can finally b e written as t h e product of


t h e Ν infinitesimal elements, giving

{qb,tb 19a,t ) a

N-l
dpo dpN-i , ,
2πΛ d q 1 " d q N _ 1 exp . ( · )2 4 4

where t h e identifications q = q a n d q = gj are implicit in (2.44). T h e


Q a N

argument of t h e exponential in (2.44) h a s t h e form of a Riemann s u m ,


enabling t h e identification

lim2>APi,?i)= / dtC(p,q) = S[p(t),q(t),t ,t ]


a b , (2.45)
i _ >
j=0 J t
*

T h u s , the classical action h a s appeared in t h e q u a n t u m mechanical t r a n ­


sition element. T h e path integral measure appearing in (2.44) is written
formally as

l i m ^ . . . ΪΕϋζλ d q i ... = VpVq . (2.46)

Form (2.46), when combined with t h e exponential of t h e action, techni­


cally does not meet t h e mathematical criteria required for a probability
measure [13]; nevertheless, it will be referred t o as t h e p a t h integral mea­
sure throughout what follows. This problem is discussed further in Sec.
2.3.
T h e final form of t h e transition element is then given by

(ft, *» I i « , *« ) = j " Vp Vq exp j i jf * dt C(p q) j , y (2.47)

where t h e limits on t h e q integrals are present t o remind t h e user t h a t q a n d 0

q„ are identified as q a n d Expression (2.47) is t h e fundamental form


a

for t h e p a t h integral version of t h e propagator. T h e measure appearing


in (2.47) ranges over t h e entire phase space available t o t h e particle as it
propagates from q t o g&. a

E x e r c i s e 2 . 3 : Consider a Hamiltonian Η t h a t is explicitly time depen­


dent in t h e sense t h a t time-dependent c-number functions (not velocity-
dependent potentials) may appear in t h e Lagrangian density. Use (2.16)
to show t h a t t h e resulting p a t h integral representation of t h e transition
element is unchanged in form.
Sec. 2.2 T h e P a t h Integral Derived 41

Exercise 2 . 3 gives t h e important result t h a t Lagrangian densities describing


systems with time-dependent parameters can immediately b e given a p a t h
integral representation with no difficulty.

E x e r c i s e 2.4: Consider a definition of coordinate ordering where t h e Ρ


operator is moved to the right and the Q operator is moved to the left.
Derive the form of the p a t h integral for this definition.

E x e r c i s e 2.5: Extend t h e p a t h integral formalism to a system with η


degrees of freedom.

Some applications of t h e p a t h integral found in the literature do not


have Vp appearing in the measure. This is not necessarily incorrect since
it is possible to integrate all the pj appearing in (2.42) for a large class of
Lagrangians. A n example of this occurs when the Lagrangian density of
the system has the form

C(p,q) = p q - ^ - V ( q ) , (2.48)

Since this gives rise to an integral Gaussian in p , it is possible t o evaluate


exactly t h e ρ integration in all the infinitesimal transition elements (2.42).
It follows from (1.107) t h a t

m
2wihe β Χ
^ h 6

m
exp (2.49)

where the identification (2.41) has again been m a d e . In this case the p a t h
integral has become

<9^ί6|9α,<α> = ^ ^ β χ ρ | ^ 6
Λ £ ( 9 , 9 ) | , (2.50)

where
C{q\q) = \mq -V{q).
2
(2.51)
42 Q u a n t u m Mechanical P a t h Integrals

T h e factors resulting from the dp integrations have been absorbed into the
measure, giving the definition

(2.52)

T h e form (1.38) of the classical mechanical action has emerged in this case.
It is form (2.50) which will be used t o generalize t h e p a t h integral in the
next section, although (2.47) is a more general and useful form t h a n (2.50).
T h e p a t h integral form (2,50) has been derived assuming t h a t t h e po­
tential V(q) appearing in the Lagrangian density is not velocity dependent.
Some physical systems are characterized by a velocity-dependent potential.
It is still possible to derive a p a t h integral for the propagator; however,
there may be ambiguities in the final form. T h e demonstration of this is
left as the following exercise.

E x e r c i s e 2.6: Consider a one-dimensional system with a point mass


moving in the velocity-dependent potential

Derive the form of the p a t h integral equivalent to (2.50) and discuss any
ambiguities t h a t are present in the coordinate ordering problem for this
potential.

2.3 The Sum over Histories


In the previous section the p a t h integral expression for the transition am­
plitude was derived from the s t a n d a r d operator formulation of q u a n t u m
mechanics. Although the measure in the second form (2.50) of the p a t h
integral is singular in the e —• 0 limit, t h e p a t h integral, as defined in t h e
previous section, must yield results t h a t are identical to any other valid
approach t o t h e same problem. T h a t this is true will be seen in detail in
the next chapter. However, for t h e moment the explicit evaluation of t h e
p a t h integral will b e deferred and attention will be focussed on generalizing
the concepts contained in, and indicated by, the form of the p a t h integral
in hand. It will be argued t h a t such a generalization is not rigorously de­
fensible at this time, and t h a t it is well to remember this when applying
p a t h integrals to new problems. In generalizing the concepts of the p a t h
Sec. 2.3 T h e S u m over Histories 43

integral it is also necessary t o remember the assumptions t h a t went into


t h e derivation of t h e previous section.
It is t h e outstanding feature of the p a t h integral t h a t the classical action
of t h e system has appeared in a q u a n t u m mechanical expression, a n d it is
this feature t h a t is considered central t o any extension of t h e p a t h integral
formalism. It is the p a t h integral of (2.50), obtained by integrating t h e
m o m e n t a pj> t h a t will serve as t h e form for motivating t h e conceptual
generalization. In the first step, t h e integrations over the qj will be replaced
by a Riemann s u m of the form
oo oo

/
dqj -> Σ C
*> (· )
2 53

•°° n i = -oo
where e is understood to be t h e infinitesimal measure element on all t h e
q

qj integrations. T h e p a t h integral (2.50) then becomes


N/2 0 0 0 0

Πχ~ — 00 η λγ—ι - 00=

. N-l
χ exp , (2.54)
3=0

where η and η _ are defined by q = n e and qb = n e .


0 Ν χ a Q q N q

T h i s form of the p a t h integral then represents the s u m over all possible


sets of values for t h e Ν— 1 integer variables {rij}. Each set of values
is weighted by the exponential of t h e value of the action for t h a t set of
values. Specifying a set of {rij} is equivalent t o specifying a set of {fy},
and these values represent the intermediate values of the particle's position
as it moves from q to qb. In effect, specifying a set of values {rij} defines a
a

discrete path from q to i.e., a set of Ν—1 intermediate positions for t h e


a

particle. In t h e e —• 0 limit any piecewise continuous p a t h from q t o qb


q a

can b e represented in this manner. T h u s , in t h e limit, t h e p a t h integral can


be viewed as a s u m over all piecewise continuous p a t h s from q t o qb in t h e a

time interval tb — < , with each p a t h receiving a weighting factor given by


a

the exponential of the classical action along t h a t p a t h . In this way every


piecewise continuous p a t h represents a possible "history" of t h e particle's
motion, and t h e p a t h integral representation of the transition element is a
weighted s u m over all possible histories.
It is this observation t h a t is t h e starting point for the generalization of
the p a t h integral, derived in the last section, to systems whose quantization
by other techniques may not b e well understood. In this approach t o
44 Quantum Mechanical P a t h Integrals

describing a q u a n t u m process, one begins by writing the classical action S


for t h e system under consideration. This specifies the dynamical variables
of the system, e.g., q(t) in the case of q u a n t u m mechanics. T h e transition
amplitude Ζ of t h e quantized system is then assumed to be given by

(2.55)
paths

where the p a t h s must be chosen t o go from some initial configuration of


the dynamical variables to some final configuration. In (2.55) t h e formal
derivatives of (2.41) are taken to be the exact statements, and in t h a t
sense (2.55) is a continuum version of the p a t h integral derived in t h e
previous section. T h e advantage of such a picture for q u a n t u m processes
is the intuitive power t h a t it brings. In the p a t h integral formulation t h e
classical p a t h s of Newton have reappeared in the quantized system, allowing
a graphical interpretation of q u a n t u m processes. However, the deceptive
simplicity of statement (2.55) cloaks many subtle, and as yet unresolved,
mathematical difficulties, and at least a cursory discussion of some of these
must be m a d e .
First, there are numerous classical systems for which no sensible quan­
t u m theory exists. Cases of this are very easy to find, even in simple
one-dimensional q u a n t u m mechanical systems. A simple example is t h e
potential V = otq — /?g . If β is a positive constant the potential is un­
2 4

bounded from below. Even though it is possible, at least formally, to write


down classical solutions to the equation of motion, the q u a n t u m theory
cannot be stable since there is no normalizable ground state. Another prob­
lem associated with converting a classical mechanical system t o a q u a n t u m
mechanical p a t h integral is t h a t it is not obvious, at first glance, how to
implement classical constraints on motion in a q u a n t u m mechanically con­
sistent manner. At an even more subtle level, it will be seen t h a t some field
theories cannot be consistently quantized because of q u a n t u m mechanical
anomalies in their conservation laws. T h e p a t h integral approach cannot
possibly improve such situations, and it may even obscure some of t h e
problems if it is indiscriminately applied.
Second, t h e motivating form (2.54) indicates t h a t the only difference in
weighting each p a t h receives is t h e exponential of the value of the action
associated with the p a t h , and t h a t the prefactors, i.e., t h e e factors in (2.54),
are the same for all p a t h s . T h a t this is true for all q u a n t u m systems, or
rather might b e an artifact of the assumptions made in deriving the form
(2.50), is not α priori obvious. For example, the assumption was m a d e
t h a t the limits on the intermediate qj integrations were ± o o . If, instead,
the system under consideration is a particle in a one-dimensional box of
Sec. 2.3 T h e S u m over Histories 45

width L, t h e n clearly the integrations over the qj should have t h e range 0


to L, and t h a t in itself will give a different result for (2.47). Obtaining a
form similar t o (2.50) raises additional difficulties. Later in this section a
similar system, a point mass on a circle, will be analyzed, and these points
will become manifestly clear. There it will not be apparent t h a t the p a t h
integral derived from t h e q u a n t u m mechanical solution to the problem h a s
any reference to t h e classical action for the system.
T h i r d , t o say t h a t there are "many" p a t h s from q t o q would b e
a b

something of an understatement. However, there are many pathological


p a t h s t h a t cannot contribute. For example, if t is a rational number and
a

t is an irrational number, then a possible p a t h would be given by t h e


b

Dirichlet function,

{ q

q
a

b
if t is rational,

if t is irrational.
(2-56)

Such a p a t h is not piece wise continuous and by t h a t criterion should b e ex­


cluded from the sum. In addition, the set of all piecewise continuous p a t h s
contains b o t h discontinous and nondifferentiable p a t h s , and these are, in
some sense, "far away" from the typically smooth and completely contin­
uous classical trajectories t h a t extremize the action. Intuitively, p a t h s far
from the classical trajectories would be expected to contribute little to t h e
p a t h integral, since a gross deviation from classical behavior violates t h e
idea t h a t q u a n t u m corrections t o most physical processes are small. Of
course, q u a n t u m mechanics counters this intuition with the presence of a
discrete set of b o u n d states. This is far from the classical behavior of t h e
system, which allows a continuum of b o u n d states t o occur in attractive
potentials. In the p a t h integral it is unclear how t h e b o u n d s t a t e s t r u c t u r e
is manifested in a s u m over histories, since a discrete set of b o u n d states
corresponds classically to a discrete set of bounded trajectories.
T h a t p a t h s associated with large values for the action do not con­
t r i b u t e significantly to the p a t h integral can be inferred from the R i e m a n n -
Lebesgue lemma [14]. This states t h a t , if f(a) is an integrable function,

Hm J daf(a)sin{at) = 0 . (2.57)

T h e Riemann-Lebesgue lemma can be proved by treating lim*—^ s i n ( a / )


as a distribution. It then follows t h a t

lim s i n ( a i ) = / dt — sin(a<) = | a / dte i a t


= πα δ(α) , (2.58)
Jo d t
J-co
46 Quantum Mechanical P a t h Integrals

which agrees with result (2.57).

E x e r c i s e 2.7: Show t h a t linrij cos(otf) = 0.

These results show t h a t , as a distribution,

lim e t a t
= π
π ιιαα<6(α) , (2.59)
t—oo

which vanishes when integrated against a well-behaved function. In this


sense, p a t h s with infinite values for the action will not contribute to the
p a t h integral.
These remarks serve to draw attention to the major difficulty in gener­
alizing the p a t h integral to a sum over histories, and t h a t is the question of
precisely how this s u m is to be performed or defined in a sensible way. T h e
idea of summing or integrating over p a t h s can be made mathematically
rigorous only by the formal definition of a path measure on the space of all
p a t h s [15, 16], This measure must be defined consistently with the weight­
ing factor of the exponentiated action, since t h a t is the functional being
integrated. To motivate this with a familiar example, it is recalled t h a t
the dt appearing in the Riemann sum of t h e exponential is an example of a
type of measure called the J o r d a n measure and has t h e form tj+i — tj. This
is by no means the most general form of measure, even for the real line.
To demonstrate the relation between the function being integrated and the
measure used t o integrate it, one need only point out t h a t the function
q(t) defined by (2.56) is not an integrable function using J o r d a n measure.
If such a pathological function is to be rendered integrable, then a more
general measure, referred to as the Lebesgue measure, must be defined.
T h e discussion of formal measure theory and, in particular, the problem
of defining a measure over the space of p a t h s is very difficult and requires
a level of m a t h e m a t i c s well beyond the scope of this book [17]. However,
some aspects of summing over p a t h s can be m a d e clearer with a modicum
of effort.
One of t h e chief difficulties in constructing a rigorous measure for t h e
continuum version of the p a t h integral is the presence of the i in its def­
inition. T h e oscillatory n a t u r e of the exponential creates difficulties in
including "unruly" p a t h s , i.e., nondifferentiable and discontinuous p a t h s ,
since the exponentiated action tends toward a distribution. A simple ex­
clusion of the unruly p a t h s would be incorrect. This is so because, while
the differentiate and continuous p a t h s are everywhere dense in the set of
all p a t h s , it can be shown t h a t , taken alone, they are inadequate t o give a
correct evaluation of the p a t h integral. In point of fact, a careful analysis
S e c . 2.3 T h e S u m over Histories

of t h e situation shows t h a t these p a t h s form a set of measure zero [18].


T h e situation is much like t h a t of the real number line, where the set of all
rational numbers can be shown to be a set of measure zero, even though
they are everywhere dense in the real line.
Unfortunately, because the exponentiated action tends toward a dis­
tribution, the p a t h integral defined by (2.55) prevents the definition of a
well-behaved measure, and therefore an alternate scheme must be found
if a m a t h e m a t i c a l definition is to be made. It is the oscillatory n a t u r e of
the p a t h integral t h a t gives rise t o the distributions. If the oscillations
were suppressed, then it might b e possible to define a sensible measure for
p a t h s . It is with this hope t h a t much of the rigorous work done on p a t h
integrals defines t h e m for Euclidean time. This means t h a t t h e integrand
of t h e action undergoes the so-called Wick rotation, which amounts t o re­
placing t with — ir. In effect, the p a t h integral is analytically continued t o
imaginary time, evaluated, and analytically continued back by t h e inverse
Wick rotation, τ —• it, to yield the final result. T h e necessity of this an­
alytic continuation arises from the need t o evaluate the generalization of
the oscillatory Gaussian integrals first encountered in Sec. 1.5. T h e r e , t h e
integration was defined by an analytic continuation identical to the Wick
rotation. Under the Wick rotation the form (2.50) for the p a t h integral
becomes

where, for the simple case (2.48), the Euclidean action SE is given by

(2.61)

T h e factors in the measure of the p a t h integral become real under the Wick
rotation, so t h a t

(2.62)

where e is effectively dr. T h e Euclidean action (2.61) is identical in form


T

to the integral of the Hamiltonian, written in terms of q instead of p . It


will be shown in C h a p t e r 4 t h a t this is no coincidence. However, for now
it is necessary only to notice t h a t , unless the potential is u n b o u n d e d from
below, large deviations from the classical trajectory, as well as unruly p a t h s ,
are exponentially suppressed. T h e Wick rotation makes a mathematically
meaningful measure possible.
48 Q u a n t u m Mechanical P a t h Integrals

It is worth noting t h a t , under a Wick rotation, the differential form of


t h e Schrodinger equation, for V = 0, becomes

£ ^ = φ , (2-63)

which is the diffusion equation t h a t governs Brownian motion. Wiener [19]


was the first t o construct a p a t h integral representation for Brownian mo­
tion and t o show t h a t a well-behaved measure could b e defined. It has been
argued, using sophisticated techniques, t h a t t h e p a t h integral h a s a well-
defined measure in t h e Euclidean region, and this argument will b e taken
over in the remainder of this book. T h e interested reader is recommended
to the books by Rivers [13] and by Glimm and Jaffe [20].
At best, these remarks may convey the difficulty in making t h e contin­
u u m p a t h integral of (2.55) a well-defined mathematical object. In practice
it is common t o approximate the s u m over p a t h s by using the subset of
piecewise continuous p a t h s t h a t are differentiable and continuous and t h a t
yield a finite action. In particular the classical trajectory is believed to
represent t h e m a x i m u m weight p a t h . It is hoped, b u t unsubstantiated in
most cases, t h a t this approximation gives t h e most i m p o r t a n t properties
of the q u a n t u m transition amplitude. In C h a p t e r 3 the role of the classical
trajectory in the q u a n t u m transition amplitude will be discussed for several
extremely simple cases.
This section will close with the derivation of the p a t h integral represen­
tation for the transition amplitude in a simple one-dimensional q u a n t u m
system with properties significantly different t h a n those originally assumed.
T h e system is t h a t of a point mass constrained t o move on a circle of radius
R. T h e classical action for this system, in terms of the angular variable 0,
is given by
£(θ, θ) = \rnR e2 2
- V(9) . (2.64)
T h e potential V is assumed t o b e periodic, so t h a t V(9 + 2 π η ) = V(0),
where η is an arbitrary integer. T h e Hamiltonian has t h e s t a n d a r d form

Η(ρ θ)
$) = ± ρ θ
2
+ ν(θ). (2-65)

T h e position eigenstates of the system are | θ ), and these are complete in


the sense t h a t

άθ\θ)(θ\ = 1. (2.66)
JO

T h e configuration space representation of t h e m o m e n t u m operator is

{β\Ρ\6') = -ί~6(β-β')- (2.67)


Sec. 2.3 T h e S u m over Histories 49

Because t h e system is on a circle, t h e eigenfunctions of all observables


must b e periodic. It is straightforward t o show t h a t t h e eigenstates of t h e
observable p$ are no longer continuous, b u t are indexed by an integer n,
and are therefore denoted \p ) . It follows from (2.67) t h a t t h e normalized
n

m o m e n t u m eigenfunctions for t h e system are given by

(0\ ) Pn = - ^ e i n e
. (2.68)

These eigenfunctions are orthonormal

(Pn\p )= m Γ ^e , ( m
- n )
* = S mn , (2.69)
Jo 1*
and complete in t h e sense t h a t
OO . 0 0

Σ ( \Ρη)(Ρη\θ')
θ
= ± Σ e i n { 0
- 9 , )
= W - n , (2.70)
n = —oo n = —oo
where t h e Dirac delta is clearly periodic. These results can be used t o
derive t h e p a t h integral form for t h e transition element (θ(, ίι\θ ,ί ). } α α

E x e r c i s e 2.8: Use the previous results to show t h a t the p a t h integral


representation of t h e transition amplitude is given by

(Obitb | o ,t )
a a

^ oo oo
lim / d9 ·
Je a
l

Σ··· Σ
• • ^ - ' ( 2 7 Γ ) λ γ
x
' n = —oo nAT-i= —oo
( . N-l 0

[«,Μ»/..-·,)-^-^·,)] I . (2.71)

where t h e integrations over the 9j are from 0 to 2π.

T h e r e are formal similarities between (2.71) and t h e general form (2.47)


if t h e classical m o m e n t u m is identified as ρ = rihjR. However, this forces
the classical m o m e n t u m to be discrete, and such a constraint is never
present in classical mechanics. Furthermore, it is not clear a t this stage
t h a t a form similar to (2.50) can b e obtained or, if it can b e , what t h e lim­
its on t h e θ integrations should b e . T h u s , passing over to a p a t h integral
of t h e form (2.55) for this system seems, at this point, a questionable step.
T h i s problem will be resolved in the next chapter when (2.71) is analyzed
to reveal its underlying structure.
50 Quantum Mechanical Path Integral

References
[1] A. Messiah, Quantum Mechanics, Wiley, New York, 1966.
[2] L. Schiff, Quantum Mechanics, Third Edition, McGraw-Hill, New
York, 1968.
[3] R. Shankar, Principles of Quantum Mechanics, Plenum Press, New
York, 1980.
[4] P.A.M. Dirac, Physikalische Z. der Sowjetunion 3 , 64 (1933).
[5] R.P. Feynman, Rev. Mod. Phys. 2 0 , 367 (1948).
[6] T h e interaction between Dirac and Feynman is described in S. Schwe-
ber, Rev. Mod. P h y s . 5 8 , 449 (1986).
[7] R.P. Feynman and A.R. Hibbs, Quantum Mechanics and Path Inte­
grals, McGraw-Hill, New York, 1965.
[8] W . Tobocman, Nuovo Cimento 3 , 1213 (1956).
[9] E.S. Abers and B.W. Lee, Phys. Rep. 9, 1 (1973).
[10] Various aspects of ordering problems are discussed in J.S. Dowker, J.
M a t h . P h y s . 17, 1873 (1976); B . Gaveau and L.S. Schulman, J . M a t h .
P h y s . 3 0 , 3019 (1989).
[11] Applications of the Trotter product formula to p a t h integral structures
is found in E. Nelson, J . M a t h . Phys. 5, 332 (1964).
[12] An alternative derivation of the q u a n t u m mechanical p a t h integral
from the Trotter product formula can be found in L.S. Schulman,
Techniques and Applications of Path Integration, Wiley, New York,
1981.
[13] See, for example, R . J . Rivers, Path Integral Methods in Quantum Field
Theory, Cambridge University Press, New York, 1987.
[14] See, for example, E.C. Titchmarsh, Introduction to the Theory of
Fourier Integrals, Oxford University Press, Oxford, 1948.
[15] L.L. Lee, J . M a t h . P h y s . 17, 1988 (1976).
[16] A. TVuman, J . M a t h . Phys. 1 9 , 1742 (1978).
[17] P.R. Halmos, Measure Theory, Van Nostrand, New York, 1950.
[18] J . R . Klauder, Acta. Phys. Aust., Suppl. X I , 341 (1973).
[19] N. Wiener, J. M a t h , and Phys. 2, 131 (1923); N. Wiener, A c t a M a t h .
5 5 , 117 (1930); see also M. Kac, Probability and Related Topics in the
Physical Sciences, Interscience, New York, 1959.
[20] A. Glimm and A. Jaffe, Quantum Mechanics—A Functional Integral
Point of View, Springer-Verlag, Berlin, 1981.

Potrebbero piacerti anche