November 1992

NTZ 29/92

AN INTRODUCTION INTO
THE FEYNMAN PATH INTEGRAL

arXiv:hep-th/9302097v1 20 Feb 1993

CHRISTIAN GROSCHE
International School for Advanced Studies
Via Beirut 4
34014 Trieste, Miramare, Italy

Lecture given at the graduate college Quantenfeldtheorie und deren Anwendung in der Elementarteilchen- und Festk
orperphysik, Universit
at Leipzig,
16-26 November 1992.

Abstract. In this lecture a short introduction is given into the theory of

the Feynman path integral in quantum mechanics. The general formulation
in Riemann spaces will be given based on the Weyl- ordering prescription,
respectively product ordering prescription, in the quantum Hamiltonian. Also,
the theory of space-time transformations and separation of variables will be
outlined. As elementary examples I discuss the usual harmonic oscillator, the
radial harmonic oscillator, and the Coulomb potential.

Contents

Contents
I
II

Introduction
General Theory
II.1 The Feynman Path Integral
II.2 Weyl-Ordering
II.3 Product-Ordering
II.4 Space-Time Transformations
II.5 Separation of Variables
III Important Examples
III.1 The Free Particle
III.2 The Harmonic Oscillator
III.3 The Radial Path Integral
III.3.1 The General Radial Path Integral
III.3.2 The Radial Harmonic Oscillator
III.4 Other Elementary Path Integrals
III.5 The Coulomb Potential
III.5.1 The 1/r-Potential in R2
III.5.2 The 1/r-Potential in R3 - The Hydrogen Atom
III.5.3 Coulomb Potential and 1/r-Potential in D Dimensions
III.5.4 Axially Symmetric Coulomb-Like Potentials
Bibliography

Page
2
5
5
10
17
21
30
33
33
34
40
40
50
57
59
61
68
74
81
89

Introduction

I INTRODUCTION
It was Feynmans (and Diracs [20]) genius [32, 33] to realize that the integral kernel
(propgator) of the time-evolution operator can be expressed as a sum over all possible
paths connecting the points q and q with weight factor exp iS(q , q ; T )/h , where
S is the action, i.e.
X

K(q , q ; T ) =
A eiS(q ,q ;T )/h
(1.1)
all paths

with some appropriate normalization A.

Surprisingly enough, the same calculus (same in the sense of a nave analytical
continuation) was already know to mathematicians due to Wiener in the study of
stochastic processes. This calculus in functional space (Wiener measure) attracted
several mathematicians, including Kac (who mentioned being influenced by Feynmans
work!), and was further developed by several authors, where best known is the work
of Cameron and Martin. The standard reference concerning these achievements is the
review paper of Gelfand and Yaglom [37], where all these early work was first critically
discussed.
Unfortunately, the discussion between physicists and mathematicians remains
near to nothing for quite a long time, except for [37]; the situation changed with
Nelson [80], and nowadays there are many attempts to understand the path integral
mathematically despite its pathalogies of infinite measure, infinite sums of phases
with unit absolute values etc.
In particular, the work of Morette-DeWitt starting with her early paper [76] gave
rise to a beautiful theory of the semiclassical expansion in powers of h
[18, 19, 77-79].
As it is known for quite a long time, the propagator can be expressed semi-classically as
ei SCl /h , with SCl the classical action, times a prefactor. This prefactor is remarkably
simple, namely one has

41

1
2 i h

D/2

KW KB (x , x ; t , t ) = [g(x )g(x )]
s




i
2 SCl [x , x ]

det
exp
SCl [x , x ] . (1.2)
xa xb
h
g = det(gab ) of some possible metric structure of a Riemannian space, and the
determinant


2 SCl [x , x ]
M := det
(1.3)
xa xb
is known as the Pauli-van Vleck-Morette determinant. The semiclassical (WKB-) solution of the Feynman kernel (we use the notions semiclassical and WKB simultaneously)
is based on the fact that the harmonic oscillator, respectively, the general quadratic
Lagrangian, is exactly solvable and its solution is only determined by the classical
path and not on the summation over all paths. As it turns out, an arbitrary kernel
can be expanded in terms of the classical paths as an expansion in powers of h
. The
semiclassical kernel (at least its short time representation) is known since van Vleck,
derived by the correspondence principle. Later on Pauli [83] has given a detailed discussion in his well-known lecture notes. More rigorously the short time propagator for
the one-dimensional case was discussed by Morette [76] in 1951 and a few years later
2

Introduction

by DeWitt [17] extending the previous work to D-dimensional curved spaces. Concerning the semi-classical expansion, every path integral with a Hamiltonian which
is quadratic in its momenta can be expanded about the semiclassical approximation
(1.2) giving a consistent and converging theory, compare DeWitt-Morette [18, 19, 74,
75, 77].
Supersymmetric quantum mechanics provides a very convenient way of classifying
exactly solvable models in usual quantum mechanics, and a systematic way of
addressing the problem of finding all exactly solvable potentials [56].
By e.g. Dutt et al. [16, 29] it was shown that there are a total of twelve different potentials. A glance on these potentials shows that their corresponding Schrodinger equation leads either to the differential equation of the confluent hypergeometric differential
equation with eigen-functions proportional to Laguerre polynomials (bound states)
and Whittaker functions (continuous states), respectively to the differential equation
of the hypergeometric differential equation with eigen-functions proportional to Jacobi polynomials (bound states) and hypergeometric functions (continuous states).
In Reference [16] this topic was nicely addressed and it was shown that in principle
the radial harmonic oscillator and the (modified) Poschl-Teller-potential solutions are
sufficient to give the solutions of all remaining ones, together with the technique of
space-time transformations as introduced by Duru and Kleinert [27, 28, 62] (see also
reference [49], Ho and Inomata [55], Steiner [90], and Pak and Sokmen [82]) enables
one to give the explicit path integral solution of the Coulomb V (C) (r) = e2 /r (r > 0)
and the Morse potential V (M ) (x) = (
h2 A2 /2m)(e2x 2ex ) (x R) (compare e.g.
reference [46] for a review of some recent results). The same line of reasoning is true for
the path integral solution of the (modified) Poschl-Teller potential [5-35] which give in
turn the path integral solutions for the Rosen-Morse V (RM ) (x) = A tanh xB/ cosh2 x
(x R), the hyperbolic Manning-Rosen V (M Ra) (r) = A coth r + B/ sinh2 r (r > 0),
and the trigonometric Manning-Rosen-like potential V (M Rb) (x) = A cot x+B/ sin2 x
(0 < x < ), respectively [44].
These lecture notes are far from being a comprehensive introduction into the
whole topic of path integrals, in particular if field theory is concerned. As old as
they be, the books of Feynman and Hibbs [34] and Schulman [86] as still a must for
becoming familiar with the subject. A more recent contribution is due to Kleinert
[64]. Myself and F. Steiner are presently preparing extended lecture notes Feynman
Path Integrals and a Table of Feynman Path Integrals [50, 51], which will appear
next year.
Several reviews have been written about path integrals, let me note Gelfand and
Jaglom [37], Albeverio et al. [1-3], DeWitt-Morette et al. [19, 79], Marinov [73], and
e.g. for the topic of path integrals for Coulomb potentials [46].
The contents of the lecture is as follows. In the next Chapter I outline the basic
theory of the Feynman path integral, i.e. the lattice definition according to the Weylordering prescription in the Hamiltonian and a related prescription which is of use
in several applications in my own work. Furthermore, the technique of canonical
coordinate-, time- and space-time transformations will be presented. The Chapter
closes with the discussion of separation of variables in path integrals.
In the third chapter I present some important examples of exact path integral
evaluations. This includes, of course, the harmonic oscillator in its simplest form. I also
discuss the radial path integral with its application of the exact treatment of the radial
(time-dependent) harmonic oscillator. The chapter concludes with a comprehensive
discussion of the Coulomb potential, including the genuine Coulomb problem, its D3

Introduction

dimensional generalization, and a generalization with additional axially-symmetric

terms.

Acknowledgement
Finally I want to thank the organizers of this graduate college for their kind invitation
to give this lecture. I also want to thank F. Steiner which whom part of the presented
material was compiled.
4

II.1

The Feynman Path Integral

II GENERAL THEORY
1. The Feynman Path Integral
In order to set up the requirements of the path integral formalism we start with the
generic case, where the time dependent Schrodinger equation in some D-dimensional
Riemannian manifold M with metric gab and line element ds2 = gab dq a dq b is given by


h
h2
LB + V (q) (q; t) =
(q; t).
(1.1)

2m
i t
is some state function, defined in the Hilbert space L2 - theR space
of all square
integrable functions in the sense of the scalar product (f1 , f2 ) = M g f1 (q)f2 (q)dq
[g := det(gab ), f1 , f2 L2 ] and LB is the Laplace-Beltrami operator
1
1

LB := g 2 a g 2 g ab b = g ab a b + g ab (a ln g )b + g ab,a b

(1.2)

(implicit sums over repeated indices are understood).

h
2
LB + V (q) is usually defined in some dense subset
The Hamiltonian H := 2m
2
D(H) L , so that H is selfadjoint. In contrast to the time independent Schrodinger
equation, H = E, which is an eigenvalue problem, and equation (1.1) which are
both defined on D(H), the unitary operator U (T ) := e i T H/h describes the time
evolution of arbitrary states L2 (time-evolution operator): H is the infinitessimal
generator of U . The time evolution for some state reads: (t ) = U (t , t )(t )
Rewriting the time evolution with U (T ) as an integral operator we get
Z p

(q ; t ) =
g(q )K(q , q ; t , t )(q ; t )dq ,
(1.3)
where K(T ) is the celebrated Feynman kernel. Equations (1.1) and (1.3) are
connected. Having an explicit expression for K(T ) in (1.3) one can derive in the
limit T = 0 equation (1.1). This, on the other hand proves that K(T ) is indeed
the correct integral kernel corresponding to U (T ). A rigorous proof includes, of course,
the check of the selfadjointness of H, i.e. H = H . Due to the semigroup property
U (t2 )U (t1 ) = U (t1 + t2 )

(1.4)

(H time-independent) we have

K(q , q ; t + t ) =

K(q , q; t + t)K(q, q ; t + t )dq

(1.5)

In Rthe
of an Euclidean
space, where gab = ab , S is just the classical action,
R
 mcase
2
q

V
(q)
dt
=
L
(q,
q)dt,

and we get explicitly [q (j) := (q (j) q (j1) ),

SCl =
cl
2
q (j) = q(tj ), tj = t + j, = (t t )/N , N , D = dimension of the Euclidean
space] :


 N2D N1

Z
N 

X
Y
i
m 2 (j)
m
dq (j) exp
K(q , q ; T ) = lim
q V (q (j) )
.
N 2 i h
h

T
2

j=1

j=1

General Theory

q(tZ
)=q

q(t )=q

( Z 
 )
i t m 2
q V (q) dt
Dq(t) exp
h t

(1.6)

For an arbitrary metric gab things are unfortunately not so easy. The first formulation
for this case is due to DeWitt [17]. His result reads:
K(q , q ; T )
=

q(tZ
)=q

q(t )=q

:= lim

( Z 
 )
R
i t m
g DDeW q(t) exp
dt
gab (q)qa qb V (q) + h
2
h t
2
6m

N1 Z q
m ND/2 Y
g(q (j) ) dq (j)
2 i h
j=1


N 
i X
2
m
h
gab (q (j1) )q a,(j) q b,(j) V (q (j1) ) +
R(q (j1) )
exp

h
2
6m
j=1

(1.7)

(R = g ab (cab,c ccb,a + dab ccd dcb cad ) - scalar curvature; abc = g ad (gbd,c + gdc,b
gbc,d ) - Christoffel symbols). Of course, we identify q = q (0) and q = q (N) in the
limit N . Two comments are in order:
R
1) Equation (1.7) has the form of equation (1.6) but the corresponding S = Ldt
is not the classical action, respectively the Lagrangian L is not the classical
a b
Lagrangian L(q, q)
=m
2 gab q q V (q), but rather an effective one:
Z
Z
Sef f = Lef f dt (Lcl VDeW )dt.
(1.8)
2

The quantum correction VDeW = 6m

R is indispensible in order to derive from
the time evolution equation (1.3) the Schrodinger equation (1.1) (e.g. [24]). The
appearance of a quantum correction V is a very general feature for path integrals
defined on curved manifolds; but, of course, V h2 depends on the lattice
definition.
2) A specific lattice definition has been chosen. The metric terms in the action are
evaluated at the prepoint q (j1) . Changing the lattice definition, i.e. evaluation
of the metric terms at other points, e.g. the postpoint q (j) or the mid-point
q(j) := 12 (q (j) + q (j1) ) changes V , because in a Taylor expansion of the relevant
terms, all terms of O() contribute to the path integral. This fact is particularly
important in the expansion of the kinetic term in the Lagrangian, where we have
4 q (j) / O().
One of the basics requirements of the construction of the path integral is Trotters
[93] product formula. Discussions and proofs can be found in many textbooks on
functional integration and functional analysis (e.g. Reed and Simon [85], Simon [87]).
Let us shortly notice a simple proof [87] :
Theorem: Let A and B be selfadjoint operators on a separable Hilbert space so that
A + B, defined on D(A) . . . D(B), is selfadjoint. Then
N

.
(1.9)
ei T (A+B) = s-lim ei tA/N ei tB/N
N

II.1

The Feynman Path Integral

If furthermore, A and B are bounded from below, then

et(A+B) = s-lim

etA/N etB/N

N

(1.10)

Proof ([80,87]): Let ST = ei T (A+B) , VT = ei T A , WT = ei T B , UT = VT WT , and

let T = ST for some H, with underlying Hilbert space H. Then

N1

X

j
Nj1
N

UT /N (ST /N UT /N )
k(ST UT /N )k =


j=0

N sup k(ST UT /N )k.

(1.11)

0sT

Let D(A) D(B). Then s1 (Ss 1) i(A + B) for s 0 and



Us 1
Vs 1
Ws 1
= Vs (i B) + Vs
iB +

s
s
s
i B + i A + 0

(1.12)

hence

lim

h
i
N k(ST /N UT /N )k 0,

for each D(A) D(B).

(1.13)

Now let D denote D(A) D(B) with the norm k(A + B)k + kk kkA+B .
By hypothesis, D is a Banach space. With the calculations of equations (1.111.13), {N (St/N Ut/N )} is a family of bounded operators from D to H with
supN {N k(ST /N UT /N )k} < for each . As a result, the uniform boundedness
principle implies that

N k(ST /N UT /N )k CkkA+B

(1.14)

for some positive C. Therefore the limit (1.14) is uniform over compact subsets of
D. Now let D. Then s s is a continuous map from [0, T ] into D, so that
{s |0 s T } is compact in D. Thus the right-hand side of equation (1.14) goes to
zero as N The proof of equation (1.10) is similar.
7

General Theory

The conditions on the operators A and B can be weakened with the requirement
that they are fulfill some positiveness and possesses a particular self-adjoint extension.
Thus we see that ordering prescriptions in the quantum Hamiltonian and corresponding lattice formulations in the path integral are closely related (compare also
[22, 23]). This can be formulated in a systematic way (e.g. [68] ):
We consider a monomial in coordinates and momenta (classical quantities)
M (n, m) = q 1 . . . q n p1 . . . pm

(1.15)

We want to have a correspondence rule according to





i
i
exp
(uq + vp) D (u, v; q, p) (u, v) exp (uq + vp)
h
h

to generate operators q, p from coordinates q, p. This produces the mapping




n+m D (u, v; q, p)
1 

.
M (n, n) n+m 

u1 . . . un v1 . . . vm u=v=0
i

(1.16)

(1.17)

Some better known examples are displayed in the following table

Correspondence Rule

(u, v)

Weyl

Symmetric
Standard
Anti-Standard
Born-Jordan

cos

uv
2


uv
exp i
2


uv
exp i
2

uv uv
sin
2
2


Ordering Rule
n  
1 X n nl m l
q p q
l
2n
l=0
1 n m
(q p + pm qn )
2

qn pm

pm qn

n
1 X ml n l
p
q p
m+1

l=0

We can make this correspondence explicit. Let us consider the Fourier integral


Z
i
d
(uq + ivp) dudv
A(p, q) = A(u, v) exp
h


(1.18)
Z
i
1
d
A(p, q) exp (uq + ivp) dpdq
A(u, v) =
(2h)D
h

We define the operator A (p, q) via



Z
i

dv)(u, v) exp (uq + ivp) dudv,

A (p, q) = A(u,
h

(1.19)

II.2

Weyl-Ordering

which gives



Z
i
i
1
A(p, q)(u, v) exp u(q q) v(p p) dudvdpdq. (1.20)
A (p, q) =

(2h)D
h
h

The inverse transformation is denoted by A (p, q) and has the form



Z
1
i
i

1
A (p, q) =
Tr A(p, q)[(u, v)] exp
u(q q) + i v(p p) dudv. (1.21)

(2h)D
h
h

In particular, the -function has the correspondence operator



Z
1
i
i

(p p, q q) =
(u, v) exp u(q q) v(p p) dudv. (1.22)
(2h)D
h
h

Let us turn to the Hamiltonian. Given a classical Hamiltonian Hcl (p, q) the
quantum Hamiltonian is calculated as


Z
i
i
H(p, q) = exp
up + vq (u, v)Hcld
(u, v)dudv

h h


(1.23)
Z
i
i
1
d
exp up uq Hcl (p, q)dpdq.
Hcl (u, v) =
(2h)2D
h
h

Obviously, by proposing a particular classical Hamiltonian depending on variables q

and q , respectively, produces a particular Hamiltonian function,
Z
1

(q q , y) eivy/h Hcl [p, 12 (q + q ) v]dudv
H(p, q , q ) =
(1.24)
(2h)D
For = 1 and = cos uv
2 , respectively, we obtain Hamiltonian functions according to
H(p, q , q ) = H(p, 12 (q + q ))

H(p, q , q ) = 21 [Hcl (p, q ) + Hcl (p, q )]

(1.25)

which are the matrix elements

< q |H(p, q)|q >

(1.26)

of the quantum Hamiltonians which are Weyl- and symmetrically ordered, respectively.
Choosing (u, v) = exp[i(1 2u)uv] [54] yields

q ab
ih

h2 1
g (q)pa pb + ( 21 u)pa g ab ,b
( u)2 g ab ,ab + VW eyl . (1.27)
2m
m
2m 2
with the well-defined quantum potential
i
h2 ab d c
h2 h ab
ab
ab
g a b + 2(g a ),b + g ,ab
(1.28)
VW eyl =
(g ac bd R) =
8m
8m
u = 12 corresponds to the Weyl prescription and is clearly emphazised.
In the next two Sections we discuss two particular ordering prescriptions and the
corresponding matrix elements which will appear as the relevant quantities to be used
in path integrals on curved spaces, it will be the Weyl-ordering prescription (leading
to the midpoint rule in the path integral) and a product ordering (leading to a product
rule in the path integral).
Another approach due to Kleinert [63, 64] I will not discuss here.
H(p, q; u) =

General Theory

2. Weyl-Ordering
A very convenient lattice prescription is the mid-point definition, which is connected to
the Weyl-ordering prescription in the Hamiltonian H. Let us discuss this prescription
in some detail. First we have to construct momentum operators [83]:



ln g
h

a
,
a =
pa =
+
(2.1)
i q a
2
q a
R

which are hermitean with respect to the scalar product (f1 , f2 ) = f1 f2 g dq. In
terms of the momentum operators (2.1) we rewrite the quantum Hamiltonian H by
using the Weyl-ordering prescription [49,69,72]
H(p, q) =

1 ab
(g pa pb + 2pa g ab pb + pa pb g ab ) + VW eyl (q) + V (q).
8m

(2.2)

Here a well-defined quantum correction appears which is given by [49,72,81]:

VW eyl

i
2 ab d c
h
2 h ab
h
ab
ab
=
g a b + 2(g a ),b + g ,ab
(g ac bd R) =
8m
8m

(2.3)

The Weyl-ordering prescription is the most discussed ordering prescription in the

literature. Let us start by defining it for powers of position- and momentum- operators
q and p, respectively [69]:

 m X
m
1
m!
m r
(2.4)
(q p )W eyl =
qml pr ql ,
2
l!(m

l)!

l=0
with the matrix elements (one-dimensional case)
 m X
m
1
m!
< q |(q p )W eyl |q > =
q ml < q |pr |q > q l
2
l!(m l)!

l=0
m

Z
dp
q + q
r i p(q q )/
h
=
p e
(2h)D
2

m r

(2.5)

and all coordinate dependent quantities turn out to be evaluated at mid-points. Here
was used that the matrix elements of the position |q > and momentum eigen-states
|p > have the property
< q |q >= (g g )1/4 (q q ),

< q|p >= (2h)D/2 ei pq/h .

(2.6)

This power rule is nothing but a special case of a more general prescription. Of course,
well behaved operator valued functions are included.
We have the Weyl-transform of an operator A(p, q) as [72]:

Z
A(p, q) dv ei pv < q v2 |A(p, q)|q + v2 >,
(2.7)

10

II.2

Weyl-Ordering

with the inverse transformation

Z
1
A(p, q) =
dpdqA(p, q)(p, q)

(2h)D
Z
(p, q) = du ei pu |p u2 >< p + u2 |.

Let us consider some simple examples. First of all for some operator f (p)

Z
dv ei pv/h < q v2 |f (p)|q + v2 >

Z
= dvdp ei pv/h < q v2 |f (q)|p >< p |q + v2 >

1
=
dvdp ei(pp )v/h f (p ) = f (p).
D
(2h)

(2.8)

(2.9)

Hence (and similarly)

f (p) f (p),
f (q) f (q).

Straightforwardly one shows the following correspondence

(2.10)

F (q)pi g ij (q)pj F (q)



1 1 2
ij
ij
ij
pi pj F (q)g +
F (q)g ,ij (q) + F,i (q)F,j (q)g (q) F,ij (q)F (q)g (q)
2 2
(2.11)
2

ij

Special cases are


pi pj F (q) pi


pi F (q)pj pi


pi F (q)pj pi +



ih

ih

pj
F (q)
2 q i
2 q j


ih

ih

pj +
F (q)
2 q i
2 q j


ih

ih

pj +
F (q).
2 q i
2 q j

(2.12)

For the case that F (q) is some symmetric N N matrix we have



1
ij
ij
ij
pi pj F (q) + 2pi F (q)pj + F (q)pi pj pi pj F ij (q).
4

(2.13)

There is, of course, a one-to-one correspondence between the function A(p, q) and the
operator A(p, q), called Weyl-correspondence. It gives a prescription how Weyl
ordered operators
can be constructed by the classical counterpart. Starting now
with an operator A, the Weyl-correspondence gives an unique prescription for the
11

General Theory

construction of the path integral. We have for the Feynman kernel for an arbitrary
N N [which is due to the semi-group property of U (T ), i.e. U (t1 +t2 ) = U (t1 )U (t2 )]:
 

 


i


K(q , q ; T ) = q  exp H(p, q) q
h





N 
N1
Y
YZ p

 (j1)
i
T
(j)
(j)
(j)
q  exp
. (2.14)
H(p, q) q
g dq
=

h

N
j=1
j=1
Observing

du ei uq < q |p u2 >< p + u2 |q >




Z
i
q + q
i
1

du exp
p(q q ) + u q
=
(2h)D

h
h
2



q +q
(2.15)
= ei p(q q )/h q
2

< q |(p, q)|q > =

and making use of the Trotter formula e i t(A+B) = slimN e i tA/N e i tB/N
and the short-time approximation for the matrix element < q | e i H |q >:

i
< q | exp i H(p, q)/h |q (j1) >
Z

Z


 (j1)
1
i h(p,q)/
h
(j) 
=
dudv exp i(q q)u + i(p p)v q
>
< q  dpdq e
(2h)D



Z
i
i (j)
1
(j1)
(j)
dp exp
p(q q
) H(p, q ) ,
(2.16)
=
(2h)D
h
h
(j)

N

where q(j) = 12 (q (j) + q (j1) ) is the mid-point coordinate.

Let us consider the classical Hamilton-function
Hcl (p, q) =



1 ab 
g (q) pa Aa (q) pb Ab (q) + V (q).
2m

(2.17)

The corresponding quantum mechanical operator is not clearly defined due to the
factor ordering ambiguity. Let us try the manifestly hermitean operator
 1
 1

1 1 
H(p, q) =
g 4 pa Aa (q) g 2 (q)g ab (q) pb Aa (q) g 4 (q) + V (q).
2m

The Weyl-transformed of H reads


2 1 1 1 ab

h
g 2 g 2 g ,ab
H(p, q) = Hcl (p, q) +
2m 4

1 1

1
1
1

1 1

ab
ab
+ g 4 ,a g 4 ,b g 2 g g 4 ,ab g 4 g
2
2
12

(2.18)

II.2

Weyl-Ordering


2  m l ab
h
la mb g R
= Hcl (p, q) +
8m
i
h2 h ab
g a b + 2(g ab a ),b + g ab,ab
= Hcl (p, q) +
8m

(2.19)

and, of course, VW eyl appears again. For the path integral all quantities have to
be evaluated at q(j) . Note: The Weyl-transformed of the the Weyl-ordered operator
(2.16) is just the classical Hamiltonian (2.17). Inserting all quantities in equation
(2.13) we obtain the Hamiltonian path integral:

K(q , q ; T ) = [g(q )g(q )]

14

lim

N1
YZ

dq

(j)

j=1

N Z
Y

j=1

dp(j)
(2h)D

N h
i X
i
q (j) p(j) Hef f (p(j) , q(j) )
.
exp

h
j=1

(2.20)

The effective Hamiltonian to be used in the path integral (2.20) reads,

1 ab (j) (j) (j)
g (
q )pa pb + V (
q (j) ) + VW eyl (
q (j) ).
2m

H(p, q) = Hef f (p(j) , q(j) ) =

(2.21)

With the help of the famous Gaussian integral

Z

dx =
exp
p

p2 x2 qx

q2
4p2

(2.22)

and, respectively, by its D-dimensional generalization

Z

iqp 21 g ab pa pb

Rn

dp e

= (2)

D/2

det(gab ) exp


1
a b
gab q q ,
2

(2.23)

we get by integrating out the momenta the Lagrangian path integral which reads
(M P = Mid-Point):

K(q , q ; T ) = [g(q )g(q )]

41

q(tZ
)=q

q(t )=q
1

:= [g(q )g(q )] 4 lim

m
2 i h

 N2D


 Z t
i
Lef f (q, q)dt

g DM P q(t) exp
h t

N1
YZ

j=1

dq (j)


 
N q
Y
m
i
(j)
a,(j)
b,(j)
(j)
(j)
gab (
q )q
q
V (
q ) VW eyl (
q ) .

g(
q (j) ) exp
h

2
j=1

(2.24)

13

General Theory

Equation (2.24) is, of course, equivalent with equation (1.7). The mid-point prescription arises here in a very natural way, as a consequence of the Weyl-ordering
prescription. This is a general feature that ordering prescriptions lead to specific
lattices and that different lattices define different V .
To proof that the path integral (2.24) is indeed the correct one, one has to show
that with the corresponding short time kernel
q
 m D/2
1
K(q (j) , q (j1) ; ) =
[g(q (j1) )g(q (j) )] 4 g(
q (j) )
2
i

h

 
i m
(j)
a,(j)
b,(j)
(j)
(j)
gab (
q )q
q
V (
q ) VW eyl (
q ) .
exp
h 2

(2.25)

and the time evolution equation (1.3) the Schrodinger equation (1.1) follows. For this
purpose a Taylor expansion has to performed in equation (1.3) yielding
(q , t) +

2 (q ; t)
(q ; t)
(q ; t)
+
B
a
b
+ . . . , (2.26)
= B0 (q ; t) + Bq b
q q
t
q b
q b q a

where the coefficients in the expansion are given by

B0 =

Bq a q b

dq

g(q )K(q , q ; )

m D/2 1 i [V (q )+V (q )]/h

g 4 (q ) e
2 Zi h


1
im

a
b
21
4
q )g (q ) exp
(2.27)
q a gb q
dq g (
2h
Z
p
= dq g(q )K(q , q ; )q b


Z
 m D/2 1
1
im
12

4
a
b
q )g 4 (q ) exp
g (q ) dq g (
q a gb q q b

2 i h
2h
(2.28)
Z
p
= dq g(q )K(q , q ; )q a q b


Z
 m D/2 1
1
im
21

4
a
b
4
q )g (q ) exp
g (q ) dq g (
q a gb g q a q b

2 i
2h
(2.29)

Bq b

From these representations it is clear that we need equivalence relations (in the sense
of path integrals) for q a q b etc. They are given by
i h ab
g
m
2


i h  ab cd
a
b
c
d
g g + g ac g bd + g ad g bc .
q q q q =

m
q a q b =

14

(2.30)
(2.31)

II.2

Weyl-Ordering

q a q b q c q d q e q f

3
i h h ab cd cd
=

g g g + g ac g bd g ef + g ad g bc g ef
m

+ g ab g ce g df + g ab g cf g de + g cd g ae g bf + g cd g af g be + g ac g be g df + g ac g bf g de

i
+ g bd g ae g cf + g bd g af g ce + g ad g be g cf + g ad g bf g ce + g bc g ae g df + g bc g af g de .

(2.32)

We just show the identity (2.30), the proof of the remaining ones is similarly. Let us
consider the integral
I(gcd ) =

dq exp

im
q c gcd q d
2h

m
2 i hg

(2.33)

with the gcd as free parameters. Differentiation with respect to one of the parameters
gives on the one hand:
dq exp

m
1/2
1
g
=
2 i h gab
2

im
I(gcd ) =
gab
2h


im
c
d
q gcd q q a q b
2h

(2.34)

and on the other

I(gcd ) =
gab

m
1
g 1/2 g ab = g ab I(gcd ).
2 i h
2

(2.35)

Here we have used the formula for the differentiation of determinants: g/gab = g g ab .
Combining the last two equations yield (2.30).
Let us denote by = q q and q = q . The various Taylor expanded contributions
yield:
#
1
g 1/4 (q )g 1/2(q ) g 3/4 (q) 1 a a + (4a b + 3a,b ) a b .
8
(2.36)
#
#


im
im
gab (q ) a b exp
gab (q) a b
exp
2h
2h


m 
m
gav vbc,d + gau avd vbc + guv uad vbc a b c d
gab bcd a c d
1+
2 i h
8 i h

2

1
m
v u a b c d e f
+
(2.37)
gav gdu bc ef .
2 2 i h
"

Here the various derivatives of the metric tensor gab have been expressed by the
15

General Theory

Christoffel symbols. Thus combining the last two equations yield:

#
#


i
m
i
m
g 1/4 (q )g 1/2(q ) exp
gab (q ) a b g 3/4 (q) exp
gab (q) a b
2h
2h

2



m
1
m
d
a b c
a
gav gdu vbc uef a b c d e f
ga dbc +
1 a +
2 i h
2 2 i h


1
m 
v
a
v
u
v
a b c d
a b
gav bc,d + gau vd bc + guv ad bc + (4a b + 3a,b ) .

8 i h
8
(2.38)
Let us start with the Bab -terms. We get immediately by equation (2.30):
i h ab
g .
2m

(2.39)

i h
[a g ab + (a g ab )]
2m

(2.40)

Bab =

Similarly:
Ba =

For the 2 - and 4 -terms in B0 we get:

i h ab
1
(4a b + 3a,b ) a b =

g (4a b + 3a,b ),
8
8m

m 

gav vbc,d + gau avd vbc + guv uad vbc a b c d

8 i h
i h ab h
=

g 8a b + cab,c + 2a,b
8m

i
+ guv g cd (2uad vbc + uab vcd ) + 5cab cbd + 2dac cbd .

(2.41)

(2.42)

For the 6 -terms equation (2.32) yields


2
m
1
gav gdu vbc uef a b c d e f
2 2 i h
i
i h ab h
=

g 4a b + 4cab c + 4dac cbd + guv g cd (2uac vbd + uab vcd ) .

8m
Therefore combining the relevant terms yields finally:


Z
im
1
a b
1/4

2
gab (q ) d
g (q )g (q ) exp
2h
 m D/2 

i h ab
c
d c
c
1+
=

g a,b ab c + 2ac bd ab,c

2 i h
8m


 m D/2
i
=

exp
VW eyl .
2 i h
h
16

(2.43)

(2.44)

II.3

Product-Ordering

Inserting all the contributions into equation (2.67) yields the Schrodinger equation
(1.1).
Let us emphasize that the above procedure is nothing but a formal proof of the
path integral. A rigorous proof must include at least two more ingredients
1) One must show that in fact
lim |[e i T H/h K(T )]| 0

(2.45)

for all H (H: relevant Hilbert space).

2) One must show that the domain D of the infinitessimal generator of the kernel
K(T ) is in fact identical with the domain of the Hamiltonian corresponding to
the Schr
odinger equation (1.1), i.e. the infinitessimal generator is the (selfadjoint)
Hamiltonian.
It is quite obvious that these strong mathematical requirements hold only under certain
assumptions on the potential involved in the Hamiltonian. Here Nelson [80] has shown
the validity of the path integral for one dimensional path integrals. Some instructive
proof can be furthermore found in the books of Simon [87] and Reed and Simon [85].
Also due to Albeverio et al. [1-3] is a wide range of discussion to formulate the Feynman
path integral without the delay of go back to the definition of Wiener integrals.

3. Product-Ordering
In order to develop another useful lattice formulation for path integrals we consider
again the generic case [43]. We assume that the metric tensor gab is real and symmetric
and has rank(gab ) = D, i.e. we have no constraints on the coordinates. Thus one
a b
can always find a linear transformation C : qa = Cab yb such that LCl = m
2 ab y y
(b)
with ab = Cac gcd Cdb and where is diagonal. C has the form Cab = ua where
the ~u(b) (b {1, . . . , d}) are the eigenvectors of gab and ab = fc2 ac bc where fa2 6= 0
(a {1, . . . , d}) are the eigenvalues of gab . Without loss of generality we assume fa2 > 0
for all a {1, . . . , d}. (For a time like coordinate qa one might have e.g. fa2 < 0, but
cases like this we want to exclude). Thus one can always find a representation for gab
which reads,
gab (q) = hac (q)hbc (q).
(3.1)
(a)

(b)

Here the hab = Cac fc Ccb = uc fc uc are real symmetric D D matrices and satisfy
hab hbc = ac . Because there exists the orthogonal transformation C equation (3.1)
yields for the y-coordinate system (denoted by My ):
ab (y) = fc2 (y)ac bc .

(3.2)

equation (3.2) includes the special case gab = ab . The square-root of the determinant

of gab , g and the Christoffels a read in the q-coordinate system (denoted by Mq ):



h,a
h

h,a

g = det(hab ) =: h, a =
.
(3.3)
, pa =
+
h
i qa
2h
17

General Theory

The Laplace-Beltrami-operator expressed in the hab reads on Mq :

M
LBq


= hac hbc



 ac
bc
2
h,a ac bc
h
bc
ac h
+
h +h
+
h h
q a q b
q a
q a
h
q b

and on My :
M
LBy





fb,a

1 2
+
2fa,a
.
= 2
fa ya2
fb
ya

(3.4)

(3.5)

With the help of the momentum operators (3.3) we rewrite the Hamiltonian in the
product-ordering form
H=

2 Mq
h
1 ac
LB + V (q) =
h (q)pa pb hbc (q) + V (q) + VP rod (q),
2m
2m

(3.6)

with the well-defined quantum correction





2
h
ac bc
ac bc h,ab
ac bc h,a h,b
bc h,a
ac
bc h,b
4h h ,ab + 2h h
h h
h ,b
+ 2h
+ h ,a
.
VP rod =
8m
h
h
h
h2
(3.7)
On My the corresponding VP rod (y) is given by
2 1
h
VP rod (y) =
8m fa2



fb,a
fb

2




 
fa,a
fa,aa
fb,a
fa,a
fb,a
2
+2
. (3.8)
4
+4

fa
fa
fa
fb
fb ,a

The expressions (3.7) and (3.8) look somewhat circumstantial, so we display a special
case and the connection to the quantum correction VW eyl which corresponds to the
Weyl-ordering prescription.
1) Let us assume that ab is proportional to the unit tensor, i.e. ab = f 2 ab . Then
VP rod (y) simplifies into
2
+ 2f f,aa
2 (4 D)f,a
.
4
8m
f

2D

VP rod (y) = h

(3.9)

This implies: Assume that the metric has or can be transformed into the special
form ab = f 2 ab . If the dimension of the space is D = 2, then the quantum
correction VP rod vanishes.
2) A comparison between (3.7) and (2.3) gives the connection with the quantum
correction corresponding to the Weyl-ordering prescription:
VP rod

2
h
2hac hbc,ab hac,a hbc,b hac,b hbc,a .
= VW eyl +
8m

(3.10)

In the case of equation (3.2) this yields:

2
fa fa,aa
2 fa,a
h
VP rod (y) = VW eyl +
4m
fa4

18

(3.11)

II.3

Product-Ordering

These equations often simplify practical applications.

Next we have to consider the short-time matrix element < q | e i T H |q > in order to derive the path integral formulation corresponding to our ordering prescription
(3.6).
We proceed similar as in the previous section. We consider the short-time approximation to the matrix element [ = T /N, g (j) = g(q (j) )]:
< q (j) | e i H/h |q (j1) >< q (j) |1 i H/h|q (j1) >
1 Z
[g (j) g (j1) ] 4
ipq (j)
=
e
dp
(2h)D
i h
i

< q (j) |hac pa pb hbc |q (j1) > < q (j) |V + VP rod |q (j1) > .
2m
h

(3.12)

Therefore we get for the short-time matrix element ( 1):

1 Z
[g (j) g (j1) ] 4
< q |e
|q
>
dp
(2h)D


i ac (j) bc (j1)
i
i
i
(j)
(j)
(j)
pq
h (q )h (q
)pa pb V (q ) VP rod (q ) .
exp
h
2mh
h
h
(3.13)

(j)

i H/
h

(j1)

The choice of the post point q (j) in the potential terms is not unique. A prepoint, mid-point or a product-form-expansion is also legitimate. However,
changing from one to another formulation does not alter the path integral, because
differences in the potential terms are of O(), i.e. of O(2 ) in the short-time Feynman
kernel and therefore do not contribute. The Trotter formula e i T (A+B) := s
limN (e i T A/N e i T B/N )N states that all approximations in equations (3.12) to
(3.13) are valid in the limit N and we get for the Hamiltonian path integral
(j)
in the product form-definition [hab = hab (q (j) )]:

1
K(q , q ; T ) = [g(q )g(q )] 4 lim

exp

N 
i X
h

j=1

pq (j)

N1
YZ
j=1

N
(j)
Y
dp

dq (j)
D
(2
h
)
j=1

ac,(j) bc,(j1) (j) (j)

h
h
pa pb V (q (j) ) VP rod (q (j) )
.

2m

(3.14)

Performing the momentum integrations we get for the Lagrangian path integral
19

General Theory

in the product form-definition: (P F =Product-Form)

K(q , q ; T )
=

q(tZ
)=q

q(t )=q

:= lim

 
 Z t 
m
i
a b
hac hbc q q V (q) VP rod (q) dt
g DP F q(t) exp
h t
2

N1 Z q
m  N2D Y
g(q (j) ) dq (j)
2 i h
j=1


N 
i X
m (j) (j1) a,(j) b,(j)
hac hbc q
q
V (q (j) ) VP rod (q (j) )
exp
.

h
2
j=1

(3.15)

In the last step we have to check that the Schrodinger equation (1.1) can be deduced
from the short-time kernel of equation (3.15). Because one can always transform from
the q-coordinates to the y-coordinates, which is a linear orthogonal transformation
and thus does not produce any quantum correction in the path integral (3.14) defined
on My , we shall use in the following the representation of equation (3.2). Looking
at the Hamiltonian formulation of the path integral (3.14) we see that the unitary
transformation y = C 1 q changes the metric term pa hac hcb pb pa ab pb , i.e. in the
correct feature, the measure dq (j) dp(j) dy (j) dp(j) due to the Jacobean J = 1 so
that the Feynman kernel is transformed just in the right manner.
We restrict ourselves to the proof that the short-time kernels of equations (2.25)
and (3.15) are equivalent, i.e. we have to show (
y = (y + y )/2):


p
im
i

41
a
b
g(
y) exp
[g(y )g(y )]
ab (
y )y y V (
y ) i hVW (
y)
2h
h



i
im

2 a

fa (y )fa (y ) y V (y ) i hVP rod (y ) . (3.16)

=
exp
2h
h

Clearly, e i V (y)/h =
e i V (y )/h for the potential term. It suffices to show that a
Taylor expansion of the g and the kinetic energy terms on the left-hand side of equation
(3.15) yield an additional potential V given by
V (y) = VP rod (y) VW eyl (y) =

2
(y) fa (y)fa,aa (y)
2 fa,a
h
.
4m
fa4 (y)

(3.17)

We consider the g-terms on the left-hand side of equation (3.16) and expand them in
a Taylor-series around y . This gives (a = (ya ya ), fa (y ) fa ):


p
1 fc fc,ab fc,a fc,b a b

14
(3.18)
.
[g(y )g(y )]
g(
y) 1
8
fc2
Exploiting the path integral identity (2.30) we get by exponentiating the O()-terms,

2 
p
i h fa fa,bb fa,b

41
[g(y )g(y )]
g(
y ) exp
.
(3.19)
8m
fa2 fb2
20

II.4

Space-Time Transformation

Repeating the same procedure for the exponential term gives:





im
im
a b

a a
exp
ab (
y ) exp
fa (y )fa (y )
2
h
2
h
#
"
i
(fc fc,ab fc,a fc,b ) a b c c .
1
8m
We use the identity (2.30) to get




im
im
a b

a a
ab (
y ) exp
fa (y )fa (y )
exp
2
h
2
h
"
#
2
2
i h fa fa,bb fa,b
i h fa fa,aa fa,a
exp
+
.
8m
fa2 fb2
4m
fa4

(3.20)

(3.21)

Combining equations (3.20) and (3.21) yields the additional potential V and
equation (3.16) is proven. Thus we conclude that the path integral (3.15) is welldefined and is the correct path integral corresponding to the Schrodinger equation
(1.1).
A combination of lattice prescriptions were in the metric gab in the kinetic term in
the exponential and in the determinant g different lattices are used exist also in the
literature. Let us note the result of McLaughlin and Schulman [70]:

K(q , q ; T ) = lim

N1 Z q
m ND/2 Y
g(q (j) dq (j)
2 i h
j=1


N 
i X
m
exp
gab (
q (j) )q (j) a q (j) b U (
q (j) )
h

(3.22)

j=1

with the effective potential

e
U (q) = V (q) A(q) q F (q),
c
A(q) an additional vector potential term and F (q) given by


F (q) = h2 Fabcd g ab g cd + g ac g bd + g ad g bc

1 
Fabcd =
gab,cd 2g ef abe cdf
48m

(3.23)

(3.24)

[abc = 21 (gab,c + gac,b gbc,a )]. According to DOlivio and Torres [21] the effective
potential can be rewritten as
i
e
2 h
h
ab
c
d
c
R + g (ad bc + b ac ) .
U (q) = V (q) A(q) q +
c
8m

which is up the derivative term the result of Mizrahi [72].

21

(3.25)

General Theory

4. Space-Time Transformation
Let us consider a one-dimensional path integral
K(x , x ; T ) =

x(tZ
)=x

x(t )=x

" Z 
 #
m 2
i t
x V (x) dt .
Dx(t) exp
h t
2

(4.1)

It is now assumed that the potential V is so complicated that a direct evaluation of

the path integral is not possible. We want to describe a method how a transformed
path integral can be achieved to calculate K(T ) or G(E), respectively, nevertheless.
This method is called space-time transformation. This technique was originally
developed by Duru and Kleinert [27,28]. It was further evolved by Steiner [90], Pak
and Sokmen [82], Inomata [59], Kleinert [62, 64] and Grosche and Steiner [49]. In the
rigorous lattice derivation of the following formul we follow ourselves [49] and Pak
and Sokmen [82].
Let us start by considering the Legendre-transformed of the general one-dimensional
Hamiltonian:


h2 d2
d
HE =
+ V (x) E
(4.2)
+ (x)
2m dx2
dx
which is hermitean with respect to the scalar product
Z
Rx

(f1 , f2 ) = f1 (x)f2 (x)J(x)dx,

J(x) = e (x )dx .
Introducing the momentum operator


1
h d
+ (x) ,
px =
i dx 2

(4.3)

d ln J(x)
dx

(4.4)

p2x
2 2
h
+ V (x) +
[ (x) + 2 (x)] E
2m
8m

(4.5)

(x) =

HE can be rewritten as
HE =

with the corresponding path integral

KE (x , x ; T ) = ei T E/h K(x , x ; T ),

(4.6)

where K(T ) denotes the path integral of equation (4.1).

Let us consider the space-time transformation
x = F (q),

dt = f (x)ds

(4.7)

with new coordinate q and new time s. Let G(q) = [F (q)], then
 2
 

2

d
d
1
F
(q)

E =
+ V [F (q)] E.
+
G(q)F
(q)

H
2
2

2m F (q) dq
F (q) dq
22

(4.8)

II.4

Space-Time Transformation

2
= fH
E:
With the constraint f [F (q)] = F (q) we get for the new Hamiltonian H
 

2  2

d
d
F
(q)

=
+ f [F (q)][V (F (q)) E]
H
+ G(q)F (q)
2
2m dq
F (q) dq
(4.9)


h2 d2
d

=
+ f [F (q)][V (F (q)) E],
+ (q)
2m dq 2
dq

where (q)
= G(q)F (q) F (q)/F (q). The corresponding measure in the scalar
product and the hermitean momentum are


Rq
p
1
h d

J(q) = g(q) = e (q )dq .

+ (q) ,
(4.10)
pq =
i dq
2

expressed in the position- q and momentum operator pq is

The Hamiltonian H
= pq + f [F (q)][V (F (q))E ] + V (q)
(4.11)
H
2m
with the well-defined quantum potential
" 
#
2


h2
F (q)
F (q)
2
V (q) =
3
2
+ G(q)F (q) + 2G (q)F (q) .
(4.12)
8m
F (q)
F (q)
Note that for G 0, V is proportional to the Schwartz derivative of the transfor is
mation F . The path integral corresponding to the Hamiltonian H
, q ; s ) =
K(q

q(tZ
)=q

Dq(s)

q(t )=q

( Z 
 )
i s m 2
exp
q f [F (q)][V (F (q)) E] V (q) ds .
h 0
2

(4.13)

As it is easily checked it is possible to derive from the short time kernel of equation
(4.13) via the time evolution equation
Z

, q ; s )(q
; s)dq
(q ; s) = K(q
(4.14)

the time dependent Schrodinger equation

s) = i h
H (q;

(q; s).
s

(4.15)

) and the
The crucial point is now, of course, the rigorous lattice derivation of K(s

is given in terms of K(T ) by

relation between K(s
) and K(T ). It turns out that K
the equations
Z
1

e i T E/h G(x , x ; E)dE

K(x , x ; T ) =
2 i h

(4.16)
Z

41

G(x , x ; E) = i[f (x )f (x )]
K(q
, q ; s )ds .
0

23

General Theory

This we want to justify!

Let us consider the path integral K(T ) in its lattice definition

K(x , x ; T ) = lim

N1 Z
m ND/2 Y
dx(j)
2 i h
j=1


N 
i X
m (j)
exp
(x x(j1) )2 V (
x(j) )
.
h

2
j=1

(4.17)

We consider a D-dimensional path integral. To transform the coordinates x into

the coordinates q we use the midpoint prescription expansion method. It reads that
one has to expand any dynamical quantity in question F (q) which is defined on the
points q (j) and q (j1) of the j th -interval in the lattice version about the midpoints
q(j) = 21 (q (j) + q (j1) ) maintaining terms up to order (q (j) q (j1) )3 . This gives:
F (q

(j)





q (j)
q (j)
(j)
(j)
) F (q ) F (q
) = F q +
F q
2
2


3

F (q) 
1
(j)
(j)
(j) F (q) 
qn(j) qk
+ qm
+ ....
= qm

qm q=q (j) 24
qm qn qk q=q (j)
(4.18)
(j)

(j1)

Introducing the abbreviations

(j)
F,m

we get therefore
2

(j)


F (q) 

,
qm q=q (j)

(j)
F,mnk


3 F (q) 

qm qn qk q=q (j)

2

1
(j)
(j) (j) (j)
(j) (j)
= qm F,m + qm qn k F,mnk
24
1
(j)
(j) (j) (j)
(j)
(j) (j)
(j)
qm
qn(j) F,m
F,n + qm
qn(j) qk ql F,m
F,nkl
12

2
1 i h
1
(j)
(j) (j) (j)
(j) (j)
=q

F,m
F,nkl Fmnkl
(q (j) )
m qn F,m F,n +
12 m

(4.19)

(4.20)

according to equation (2.30) with the abbreviation

(j)

(j)

1
(j) (j) 1
Fmnkl
(q (j) ) = (F,m
F,n ) (F,k F,l )1
(j)

(j)

(j)

(j)

(j)
(j) 1
(j)
(j) 1
+ (F,m
F,k )1 (F,l F,n
) + (F,m
F,l )1 (F,k F,n
) .

(4.21)

Furthermore we have to transform the measure. Because

N1
Y
j=1

dx

(j)


Y
 F (q (j) )  (j) N1
dq

F;q (q (j) )dq (j)
 q (j) 


N1
Y
j=1

j=1

24

(4.22)

II.4

Space-Time Transformation

does not have a symmetric look, we must expand about q(j) . We get:
N1
Y
j=1

N
Y
Y

1/2 N1

1
dx(j) = F;q (q )F;q (q )
dq (j)
F;q (q (j) )F;q (q (j1 ) 2
j=1

j=1

N
Y
Y

1/2 N1
F;q (
q (j) )
dq (j)
F;q (q )F;q (q )

j=1

j=1



1 F;q,m (
q )F;q,n(
q (j) ) F;q,mn (
q (j) )
(j)
(j)
1

qm qn
8
F;q2 (
q (j) )
F;q (
q (j) )
(j)

N
Y
Y

 1 N1
F;q (
q (j) )
dq (j)
=
F;q (q )F;q (q ) 2
j=1

j=1



i h (j) (j) 1 F;q,m (
q (j) )F;q,n(
q (j) ) F;q,mn(
q (j) )
exp
(F F )

8m ,m ,n
F;q2 (
q (j) )
F;q (
q (j) )

(4.23)

Thus we have the coordinate transformed path integral

, q ; T )
K(q

= [F;q (q )F;q (q )]

1/2

lim

N1 Z
m ND/2 Y
dq (j)
2 i h
j=1

 
i m (j) (j)
(j)
q qn F,m (
q (j) )F,n (
q (j) ) V (
q (j) )
F;q (
q ) exp

h 2 m
j=1


q (j) )F;q,n(
q (j) ) F;q,mn(
q (j) )
h2 (j) (j) 1 F;q,m (
(F F )

8m ,m ,n
F;q2 (
q (j) )
F;q (
q (j) )

h2 (j) (j) 1
(j)

F F F
(q )
8m ,m ,nkl mnkl
N
Y

(4.24)

This path integral has the canonical form (2.24). In order to see this note that
the factor F,m (
q (j) )F,n (
q (j) ) in the dynamical term can be interpreted as a metric
gnm appearing in an effective Lagrangian. Thus we can rewrite the transformed
Lagrangian
p metric (Gervais and Jevicki [38]): Firstly, we have
p in terms of this
F;q (q) = det(gnm (q)) g(q). In the expansion of the determinant about the
midpoints we then get
1/2
[F;q (q (j) )F;q (q (j1 )




q
1 mn (j)
(j)
(j)
mn
(j)
(j)
(j)
g (
q )gmn,kl + g ,k (
q )gmn,l (
q ) qk qk . (4.25)
q(
q ) 1+
16
Using now successively the identities
g,l (q)
= g mn (q)gmn,l (q),
g(q)

m
F,kl
(q)

m
m
kl (q)F,l (q),

25

m
ml,k (q)



1 g,l (q)
=
2 g(q) ,k
(4.26)

General Theory

we can cast the quantum potential into the form

V (q) =

2 mn
h
g (q)klm (q)lkm (q)
8m

(4.27)

which is just the Weyl-ordering quantum potential without curvature term. In the
present case the curvature vanishes since we started in flat Euclidean space which
remains flat during the transformation.
In particular equation (4.24) takes in the one-dimensional case the form [F
dF/dq]:
, q ; T ) = [F (q )F (q )] 21 lim
K(q

N1 Z
m N/2 Y
dq (j)
2 i h
j=1

( "
#)
N
2
2 (j)
Y
i m
h F (
q )
2 (j)
(j)
F (
q (j) ) exp

qm
qn(j) F (
q ) V (
q (j) )
4

h 2h
8m F (
q (j) )
j=1

(4.28)
2

which is the same result as taking for D = 1 in VW eyl the metric tensor gab = F .
E . To
The infinitesimal generator of the corresponding time-evolution operator is H
proceed further we perform a time transformation according to
s = s(t ) = 0,

dt = f (s)ds,

s = s(t )

(4.29)

and we make the choice f = F . Translating this transformation into the discrete
notation requires special care. In accordance with the midpoint prescription we must
first symmetrize over the interval (j, j 1) to prefer neither of the endpoints over the
other, i.e.
t(j) = = s(j) F (q (j) )F (q (j1) ),
Expanding about midpoints yields
(

(j)

2 q (j)
F (
q ) 1+
4
2

(j)

s(j) = s(j) s(j1) (j) .

"

F (
q (j) )

F (
q (j) )

F (
q (j) )
F (
q (j) )

2 #)

(4.30)

(4.31)

Insertion yields for each jth -term [identify V (x(j) ) V [F (q (j) )] V (q (j) )]


 

N
N N1
i m 2 (j)
m  2 Y (j) Y
(j)
exp
dx
x V (x ) + E
2 i h
h 2
j=1
j=1

N1
Y
j=1

F (q

(j)

)dq

(j)

N 
Y

j=1

m  12
2 i h



  
1 (j) (j) 4 (j)
i m 2 (j) 2 (j)
(j)
F (
q ) q + F (
V (x ) + E
q )F (
q ) q
exp
h 2
12
26

II.4

Space-Time Transformation

N 
h
i 21 N1
Y
Y
(j)


dq
= F (q )F (q )
j=1

j=1

m
2 i (j) h

 21

"" (

 (j) 2 1

q (j) )
m
F (
q )
i
2 q (j) F (
exp

1+

(j)

(j)

h 2
4
F (
q )
F (
q (j) )
j=1
)##


h
i
(j)
4 (j)
F
(
q
)

q
2
2 q (j) +
(j) F (
q (j) ) V (q (j) ) E
.
12 F (
q (j) )
h
i 21  m  N2 N1
Y


=
F (q )F (q )
dq (j)
2 i h
j=1
"" (
h
i
i m 2 (j)
2
exp
q (j) F (q (j) ) V (q (j) ) E
h 2
" 
#)##
2
F (q (j) )
F (q (j) )
ih
3
2 (j)
.

8m
F (q (j) )
F (q )
N
Y

Let us assume that the constraint

Z s

dsf [F (q(s))] = T

(4.32)

(4.33)

has for all admissible path a unique solution s > 0. Of course, since T is fixed, the
time s will be path dependent. To incorporate the constraint let us consider the
Green function G(E):
Z

dT ei T E/h K(x , x ; T )
G(x , x ; E) = i
0
Z
(4.34)

1
i T E/
h

dE e
G(x , x ; E).
K(x , x ; T ) =
2 i h

This incorporation is far from being trivial, in particular the fact that the new timeslicing (j) is treated in the same way as the old time-slicing , i.e. (j) is considered
as being fixed, but can however be justified in a more comprehensive way, see e.g. [6,
35, 59]. We now observe that G(E) can be written e.g. in the following ways [62]:
Z

< x | e i s f (x)(HE)/h |x > ds

(4.35a)
G(x , x ; E) = f (x )
0
Z

= f (x )
< x | e i s (HE)f (x)/h |x > ds
(4.35b)
0
Z

p
h
i s f (x)(HE) f (x)/

= f (x )f (x )
|x > ds .
< x |e
0
(4.35c)
The last line give for the short-time matrix element in the usual manner
<x(j) | e i

(j)

f 2 (x)(HE)f

1
2

(x)/
h

|x(j1) >
27

General Theory

< x(j) |1 i (j) f 2 (x)(H E)f 2 (x)/h|x(j1) >

q
i (j)
(j) (j1)
=< x |x
>
f (x(j) )f (x(j) ) < x(j) |H E|x(j1) >
h
q
i (j)
(j)
(j1)
E |F (q (j1) ) >
f [F (q (j) )]f [F (q (j))] < F (q (j) )|H
=< F (q )|F (q
)>
h
i

(j1) >
=< q (j) |q (j1) > < q (j) |H|q
h



Z
1
i
i 2 i 2 (j)
i
(j)
(j)
(j)
=
dpq exp
pq q
p F (
q )(V (
q ) E) V (
q )
2
h
2mh q
h
h
 21



i m 2 (j) 2 (j)

m
(j)
(j)
exp
q F (
q )(V (
q ) E) V (
q ) ,
=
2 i h
2h
h
h
(4.36)
where the well-defined quantum potential V is given by
"
 2 #
h2
F (q)
F (q)
V (q) =
3
.
2
8m
F (q)
F (q)

(4.37)

Note that this quantum potential has the form of a Schwartz-derivative. This
procedure justifies our symmetrization prescription of equations (4.30). Thus we arrive
finally at the transformation formul:
Z
1

dE e i T E/h G[F (q ), F (q ); E]
K(x , x ; T ) =
2 i h

(4.38)
Z

12

G[F (q ), F (q ); E] = i[F (q )F (q )]
ds K(q , q ; s )
0

) is given by
where the path integral K(s
, q ; s ) = lim
K(q

m
2 i h

 21 N1
YZ

dq (j)

j=1


N 
i X
m 2 (j)
2 (j)
exp
q F (
q )(V (
q (j) ) E) V (
q (j) )
.
h

2
j=1

(4.39)

This is exactly the path integral (4.13) and thus we have proven it. Note that only
for D = 1 we have the simple transformation property
i m 2 (j)
i m 2 (j)
x
q .
2
2

(4.40)

For higher dimensional problem this may only possible for one coordinate kinetic term.
Anyway, the requirement for higher dimensional problems are
F;q = F,m F,n
28

(4.41)

II.5

Separation of Variables

for at least one pair (m, n) (1, . . . , D).

Finally we consider a pure time transformation in a path integral. We consider
equation (4.35c), i.e.
Z

(4.42)
ds < q | e i s f (HE) f /h |q >,
G(x , x ; E) = f (x )f (x )
0

where we assume that the Hamiltonian H is product ordered. Then

G(x , x ; E)
=

f (x )f (x )

ds

f (x )f (x )

N1
YZ

dq

(j)

j=1

j=1

N
Y

N1
YZ

< q (j) | e i

f (HE)

f /
h

|q (j1) >

dq (j)
ds lim
1
N
(g g ) 4
0
j=1
 
N Z
q
Y
dp(j)
i
(j)
(j)

p q f (q (j) )f (q (j1) )
exp
D
(2
h
)
h

j=1
 ac (j1) bc (j)

h (q
)h (q ) (j) (j)
(j)
(j)

pa pb + V (q ) + VP rod (q ) E
2m
Z s
N1


Y
ND/2
1
m
g(q (j) )
dq (j)
= (f f ) 2 (1D/2) lim
D
(j)
N 2 i
h
f (q )
j=1

"
N
i X
m hac (q (j1) )hbc (q (j) ) a,(j) b,(j)
p
q
q
exp
(j1) )f (q (j) )
h
2
f
(q
j=1
#)


i f (q (j) ) V (q (j) ) + VP rod (q (j) ) E
=

= (f f )

1
2 (1D/2)

, q ; s )ds
K(q

(4.43)

with the path integral

, q ; s ) =
K(q

q(tZ
)=q

q(t )=q

p
g DP F q(t)

)
( Z 


i s m a b
hac hcb q q f V (q) + VP rod (q) E ds .
exp
h 0
2

(4.44)

ac ) and equation (4.44) is of the canonical

ac = hac /f and g = det(h
Here denote h
product form. Note that for D = 2 the prefactor is one.

29

General Theory

5. Separation of Variables
Let us assume that the potential problem V (x) has an exact solution according to
" Z 
 # Z
m 2
i t
Dx(t) exp
x V (x) dt = dE e i E T /h (x ) (x ).
h t
2

x(tZ
)=x

x(t )=x

(5.1)
Here dE denotes a Lebesque-Stieltjes integral to include discrete as well as
continuous states. Now we consider the path integral
R

K(z , z , x , x ; T )
=

z(tZ
)=z

z(t )=z

f d (z)

i Z

d
Y

gi (z)Dzi (t)

i=1

x(tZ
)=x

d
Y

k=1

x(t )=x

Dxk (t)



d
d

t
X
X
V (x)
m
dt
exp
x 2k
gi (z)zi2 + f 2 (z)
+
W
(z)
+
W
(z)
h

t
2
f 2 (z)

i=1

k=1

Z Y
d
d
N1 Z
Y
m  N2D Y
(j)
(j)
d (j)
(j)
dxk
f (z )
gi (z )dzi
= lim
N 2 i
h
i=1
j=1
k=1

N
d
d
i X
X
X
(j)
(j)
m
exp
gi (z (j1) )gi (z (j) )2 zi + f (z (j1) )f (z (j) )
2 x k
h
2
j=1
i=1
k=1



(j)
V (x )
2 (j) + W (z (j) ) + W (z (j) ) . (5.2)

f (z )


Here (z, x) (zi , xk ) (i = 1, . . . , d ; k = 1, . . . , d, d + d = D) denote a D-dimensional

coordinate system, gi and f the corresponding metric terms, and W the quantum
potential. For simplicity I assume that the metric tensor gab involved has only
diagonal elements, i.e. gab = diag(g12 (z), g22 (z), . . . , gd2 (z), f 2 (z), . . . , f 2 (z)]. Of course,
Qd
det(gab ) = f 2d i=1 gi2 f 2d G(z). The indices i and k will be omitted in the following.
We perform the time transformation
s=

d
f 2 [z()]

s = s(t ),

s(t ) = 0,

(5.3)

where the lattice interpretation reads /[f (z (j1) )f (z (j) )] = (j) . Of course, we
identify z(t) z[s(t)] and x(t) x[s(t)]. The transformation formul for a pure time
transformation are now
Z
1

dE e i ET /h G(z , z , x , x ; E)
(5.4)
K(z , z , x , x ; T ) =
2 i h

30

II.5

Separation of Variables

G(z , z , x , x ; E) = i[f (z )f (z )]

1D/2

, z , x , x ; s ),
ds K(z

(5.5)

) is given by
where the transformed path integral K(s
, z , x , x ; s )
K(z

z(sZ
)=z p
G(z)
=
Dz(s)
2d
f
(z)
z(0)=z

x(sZ
)=x

Dx(s)

x(0)=x

( Z  

 )
s


2
m g (z) 2
i
z + x 2 V (x) f 2 (z) W (z) + W (z) + f 2 (z)E ds
exp
2
h 0
2 f (z)
Z

, z ; s ),
= dE e i E s /h (x ) (x )K(z
(5.6)

with the remaining path integration

, z ; s ) =
K(z

z(sZ
)=z

G(z)

f 2d (z)

z(0)=z

Dz(s)

( Z 
 )


i s m g 2 (z) 2
2
2
exp
z f (z) W (z) + W (z) + f (z)E ds . (5.7)
h 0
2 f 2 (z)
Of course, in the path integrals (5.6,5.7) the same lattice formulation is assumed as in
the path integral (5.2). Note the difference in comparison with a combined space-time
transformation where a factor [f (z )f (z )]1/4 would instead appear. We also see that
for D = 2 the prefactor is identically one. We perform a second time transformation
) effectively reversing the first:
in K(s
Z s
=
f 2 [z()]d,
= s
(5.8)
0

with the transformation on the lattice interpreted as (j) = (j) f (z (j1) )f (z (j) ).
Therefore we obtain the transformation formul
Z

1

, z ; E )
K(z , z ; s ) =
dE e i E s /h G(z
(5.9)
2 i h

Z

Dd
1

2
G(z , z ; E ) = i[f (z )f (z )]
d ei E /h K(z
,z ; )
0
(5.10)

with the transformed path integral given by

K(z
,z ; ) =

z(Z
)=z

z(0)=z

G(z) Dz()

( Z 
 )
E
i m 2
2
d .
g (z)z W (z) W (z) + 2
exp
h 0
2
f (z)
31

(5.11)

General Theory

Plugging all the relevant formul into equation (5.5) yields

Z

dE (x ) (x )
K(z , z , x , x ; T ) = [f (z )f (z )]
Z
Z
Z
Z

1
i E( T )/
h 1

dE e
ds e i s (E +E )/h

d
dE
2h 0
2h

( Z 
z(Z )=z
 )
p
E
i m 2
2

d .
g (z)z W (z) W (z) + 2
G(z) Dz() exp
h 0
2
f (z)

D/2

z(0)=z

(5.12)
The d dE-integration produces just T , whereas the dE ds -integration can be
evaluated by giving E +E a small negative imaginary part and applying the residuum
theorem yielding E E . Therefore we arrive finally at the identity [45]

K(z , z , x , x ; T ) = [f (z )f (z )]

z(tZ
)=z

z(t )=z

p
G(z) Dz(t) exp

D/2

dE (x ) (x )

( Z 
 )
E
i t m 2

dt .
g (z)z 2 W (z) W (z) 2
h t

2
f (z)
(5.13)

Note that this result can be short handed interpreted by inserting



m
2 i (j) h

D/2

 

m
i
2 (j)
(j)
(j)
exp
x V (x )
h 2 (j)
Z
(j)
= dE(j) e i E(j) /h (j) (x(j1) )(j) (x(j) )

(5.14)

with (j) = /[f (z (j1) )f (z (j) )] for all j th and applying the orthonormality of the
in each j th -path integration. Equation (5.14) describes therefore a short-cut to
establish equation (5.12) instead of performing a time transformation back and forth.

32

III.1

The Free Particle

III Important Examples

In this Chapter I present some of the most important path integral solutions which
are
1) The free particle as a simple example.
2) The harmonic oscillator in its basic form, where we allow in addition a time
dependent frequency.
3) Path integration in polar coordinates. We shall discuss the various features of
properly defined path integrals including the Besselian functional measure.
In particular the (time-dependent) radial harmonic oscillator will be exactly
evaluated.
4) The Coulomb potential.

1. The Free Particle

For warming up we calculate the path integral for the free particle in an D-dimensional
Euclidean space. From the representation
K(x , x ; T )
=

x(tZ
)=x

Dx(t) exp

im
2
h

x(t )=x

= lim

m
2 i h

x 2 dt
t

N
X

2
im
x(j) x(j1) .
dx(j) exp
2h j=1

 N2D N1
YZ
j=1

(1.1)

it is obvious that the various integrations separate into a D-dimensional product. All
integrals are Gaussian. Starting with ( = 1, . . . , D):
(0)
K N=2 (x(2)
, x ; 2)


 m Z
m
m
(1)
(2)
(1) 2
(1)
(0) 2
=
dx exp
(x x )
(x x )
2 i h
2 i h
2 i h
r


m
m
(2)
(0) 2
exp
(x x ) ,
=
4 i h

4 i h

(1.2)

it is easy to show that

K

(0)
(x(N)
, x ; N )



m
m
(N)
(0) 2
exp
(x
x ) .
2 i h
N
2ih
N
33

(1.3)

Important Examples

Thus we get in the limit N (note N = T ):

D/2


m
i m
2
K(x , x ; T ) =
exp
(x x )
2 i h
T
2
hT


Z
hp2
1

exp i p(x x ) i T
dp.
=
(2)D
2m

(1.4)

From this representation the normalized wave-functions and the energy spectrum can
be read off (p RD ):
p2 h2
Ep =
2m

ei px
(x) =
,
(2)D/2

(x RD )

(1.5)

which is the known result.

The corresponding energy-dependent Green function is given by (D = 1):
(1)

(x , x ; E) = i

i ET /
h

K(x , x ; T ) e

r


|x x |
m
exp i
2mE .
dT =
2E
h
(1.6)

For the D-dimensional case we obtain

G(D) (x , x ; E)

D/2 
 12 (1D/2)


m
|x x |2
m
2
|x x |
= 2i
2mE
K1D/2 i
2 i h

2E
h
(1.7)
1
 D/2 



2 (1D/2)
m
m 2
|x x |
(1)
=i
|x x |2
2mE ,
(1.8)
H1D/2
2
h
2E
h
where use has been made of the integral representation [40, p.340]
Z

dt t

exp

and K (z) =

i
2

(1)

pt
4t

ei /2 H (i z), K 12 (z) =

a
=2
4p

 2

K ( ap ),

(1.9)

/2z ez .

2. The Harmonic Oscillator

After this easy task we proceed to the path integral calculation of the harmonic
oscillator. We consider the simple one-dimensional case with the Lagrangian
L[x, x]
=

m 2 c(t) 2
x
x + b(t)xx e(t)x
2
2
34

(2.1)

III.2

The Harmonic Oscillator

Here we assume that the various coefficients may be time-dependent. The path integral
has the form

K(x , x ; t , t ) =

x(tZ
)=x

Dx(t) e

i
h
S[x]

x(t )=x

x(tZ
)=x

Dx(t) e

x(t )=x

i
h

R t
t

L[x,x]dt

(2.2)

In this quadratic Lagrangian, of course, an ordering problem appears. This is due to

the stochastic nature of the path integral, the time-integral in the exponent is not
a Riemann-integral but a so called Ito-integral. This we have already discussed in
chapter II. The ordering ambiguity appears in the b(t)-term, where we have in the
corresponding Hamiltonian a px-term. The consequence is that we have to use the
mid-point formulation, i.e. (
x(j) = 12 (x(j) + x(j1) ), and c(j) = c(t(j) ), etc., see also
Schulmans book [86]):
K(x , x ; t , t )
=

( Z 
 )
i t m 2 c(t) 2
x
x + b(t)xx e(t)x dt
Dx(t) exp
h t
2
2

x(tZ
)=x

x(t )=x

= lim

N1 Z
m  N2 Y (j)
dx
2 i h

j=1


N 
i X




(j)
2
m (j)
c
exp
x x(j1)
(j)
.
(
x(j) )2 + b(j) x
(j) x(j) x(j1) e(j) x
h

2
2
j=1
(2.3)

In the following we use the notations in equations (2.2) and (2.3) as synonymous. Let
us expand the path x(t) about the classical path xCl (t), i.e.:
x(t) = xCl (t) + y(t),
where y(t) denotes a fluctuating path about the classical one. The classical path
obeys, of course, the Euler Lagrange equations:
L[xCl , x Cl ]
d L[xCl , x Cl ] L[xCl , x Cl ]
=

= 0,
xCl
dt
x Cl
xCl

xCl (t ) = x ,

xCl (t ) = xCl .
(2.4)

Expanding we obtain for the action



Z t 
1 2 S[x] 
m 2 c 2
2

S[x] = S[xCl ] +
y = S[xCl (t ), xCl (t )] +
y y + by y dt.
2 x2 x=xCl
2
2
t
(2.5)
35

Important Examples

The linear functional variation vanishes due to the Euler-Lagrange equations (2.4).
For the path integral we now find


i

S[xCl (t ), xCl (t )] F (t , t ),
K(x x ; t , t ) = exp
h

"
#
y(tZ )=0
(2.6)
Z


im t

2
2
2
Dy(t) exp
F (t , t ) =
y (t)y dt .
2
h t
y(t )=0

Here we have used the abbreviations m 2 (t) = c(t) + b(t),

y = y(t ) and y = y(t ).
Now
F (t , t ) = lim FN
N

Z
Z
m  N2 (1)
= lim
dy (N1)
dy
N 2 i
h

N h

i
X
m
(j)
(j1) 2
2 (j) 2 (j) 2
(y y
)
y
exp

2 i h
j=1


(2.7)

Let us introduce a N 1-dimensional vector z = (x1 , . . . , xN1 )T and the (N 1)

(N 1) matrix B:

2 2 (1) 2

..

B=
.

1
2 2 (2) 2
..
.

...
...
..
.

0
0

...
...

0
0
..
.
2

0
0
..
.

(N2) 2

2
1

1
2 (N1) 2
2

(2.8)

Thus we get
FN

m 2
=
2 i h

N1

z exp

m T

z Bz
2 i h

m
2 i h det B

 12

(2.9)

Therefore

F (t , t ) =

m
2 i h
f (t , t )

 21

where

f (t , t ) = lim det B.
N

Our final task is to determine det B. Let us consider the j j matrix

2 2 (1) 2
1
...
0
0

1
2 2 (2) 2 . . .
0
0

.
.
.
.
.
(j)

..
..
..
..
..
B =

0
0
. . . 2 2 (j1) 2
1
0
0
...
1
2 2 (j) 2
36

(2.10)

(2.11)

III.2

The Harmonic Oscillator

One can show that the following recursion relations holds:

det B (j+1) = (2 2 (j+1) 2 ) det B (j) det B (j1)
with det B (1) = 2 2 (1) 2 and det B (0) = 1. Let us define g (j) = det B (j) , then we
have
g (j+1) 2g (j) + g (j1) = 2 (j+1) 2 g (j) .

Turning to a continuous notation we find for the function g(t) = f (t, t ) a differential
equation:
g(t) + 2 (t)g(t) = 0, with g(t ) = 0, g(t
) = 1.
(2.12)
The
follow from g(t ) = g0 = lim0 det B (0) and g(t
) = lim0
 two last equation

g(t + ) g(t ) / = lim0 (det B (1) det B (0) ) = 1. Finally we have to insert
g(t ) = f (t , t ) into equation (2.10).
At once we recover the free particle with g = t t = T .
The case of the usual harmonic oscillator with (t) = (time-independent) is given
by
1
g(t) = sin T.

To get the path integral solution for the harmonic oscillator we must calculate its
classical action. It is a straightforward calculation to show that it is given by:
i
m h 2

S[xCl (t ), xCl (t )] =
(x Cl + x Cl ) cos T 2xCl xCl .
(2.13)
2 sin T
and we have for the Feynman kernel:

K(x , x ; T ) =

m
2 i h
sin T

 21

exp



x x
m
2
2
(x +x ) cot T 2
. (2.14)

2ih

sin T

By the use of the Mehler-formula [31, Vol.III, p.272]:


 n


X
4xyz (x2 + y 2 )(1 + z 2 )
1 z
1
exp
e
,
Hn (x)Hn (y) =
2)
2
n!
2
2(1

z
1

z
n=0
(2.15)
where Hn denote the Hermite-polynomials,
we
can
expand
the
Feynman
kernel
p
p
according to (identify x m/h x , y m/h x and z = e i T ):
(x2 +y 2 )/2

K(x , x ; T ) =

e i T En /h n (x )n (x )

(2.16)

n=0

with energy-spectrum and wave-functions:

En = h
(n + 12 ),
r

 41



m
m 2
m
Hn
n (x) =
x exp
x .
22n hn!2
h
2
h
37

(2.17)
(2.18)

Important Examples

Equation (2.14) is as it stands only valid for 0 < T < . Let us investigate it for
larger times. Let
T =

n
+ ,

with

n N0 ;

0 < < /

(2.19)

then we have sin T = ei n sin , cos T = ei n cos and equation (2.14) becomes
 21
m
K(x , x ; T ) =
2h sin



i
i 1
i m h 2
2

exp
(x + x ) cos T 2x x
+n +
2 2
2
h sin T

(2.20)

which is the formula for the propagator with the Maslov correction (note sin =
| sin T |). Now let 0, i.e. we consider the propagator at caustics. We obtain


 12



i
i m h 2
m
i n
n
2
i n

= lim
(x + x ) 2 e
xx
+
exp
K x ,x ;
0 2 i h

2
2
h


1
1
n
n 2
= lim
exp 2 (x (1) x ) i
a0
a
2
a


i
(2.21)
= exp n (x (1)n x ).
2

Let us finally determine the energy dependent Green function. We use the integral
representation [40, p.729], a1 > a2 , ( 21 + ) > 0):


Z

a1 + a2
2 x
exp
t cosh x I2 (t a1 a2 sinh x)dx
coth
2
2
0
( 1 + )
= 2
W, (a1 t)M, (a2 t).
t a1 a2 (1 + 2)

(2.22)

p
Here W, (z) and M, (z) denote Whittaker-functions. We reexpress ex = x/2
[I 21 (x) + I 12 (x)] apply equation (2.22) for = E/2
h and use some relations between
Whittaker- and parabolic cylinder-functions to obtain
r


m
E
1
1

G(x , x ; E) =
2 h
2 h
#
#
"r
" r
2m
2m
(x + x + |x x |) D 1 + E
(x + x |x x |) .
D 1 + E
2
h

2
h

h
h
(2.23)
From this representation we can recover by means of the expansion for the -function

(1)n
(z) =
+
n!(z + n)
n=0
38

et tz1 dt
1

III.2

The Harmonic Oscillator

and some relations of the parabolic cylinder-functions and the Hermite polynomials,
the wave-functions of equation (2.18).
Let us note that we can use equations (2.14, 2.23) to obtain recursion relations for
the Feynman kernel and the Green function, respectively, of the harmonic oscillator
2
2
in D dimensions. Let ~x Rn and define = x~ + x~ , = ~x ~x , then
1 (D2)
K
(~x , ~x ; T )
2


1 i T m

=
K (D2) (~x , ~x ; T ).
e
2
2
h

K (D) (~x , ~x ; T ) =

(2.24)

For the Green functions this gives

(D) (|; E)
G(D) (~x , ~x ; E) G
(D2) (|; E)
(D) (|; E) = 1 G
G
2 


1 m
(D2)
(D)

G (|; E) =
G
(|; E h)
2
2 h

(2.25)
(2.26)
(2.27)

Introducing furthermore = 12 (|~x + ~x | + |~x ~x |), = 12 (|~x + ~x | |~x ~x |) we

obtain for D = 1, 3, 5 . . . .
r
  D1

2
m
1
E
1
1

G (~x , ~x ; E) =
2 h
2 h
2
!
r
 D1


2
2m

D 1 + E
2
D 1 + E
2
h

2
h

h
(D)

!
2m
,
h
(2.28)

respectively,
r
  D1

2
1
m
1
D
E
G(D) (~x , ~x ; E) =

2 h
2
h
2
!
r

 D1


2m
m 2

1
+

D D + E
2
D D + E
2
h

2
h

h
h

!
2m
.
h
(2.29)

Let us note that the most general solution for the general quadratic Lagrangian is due
to Grosjean and Goovaerts [52, 53].

39

Important Examples

3. The Radial Path Integral

3.1. The General Radial Path Integral

Radial path integrals have been first discussed by Edwards and Gulyaev [30] and
Arthurs [4]. Edwards and Gulyaev discussed the two- and three-dimensional cases,
Arthurs concentrated on D = 2. Further progress have been made by Peak and
Inomata [84] who calculated the path integral for the radial harmonic oscillator
including some simple applications. See also [49] for the ordering problematics
which give rise to the various quantum potentials in the formulation of the radial
path integral. In this section we derive the path integral for D-dimensional polar
coordinates. We follow in our line of reasoning references [49] and [92], where these
features have been first discussed in their full detail. Similar topics can be also found
in Bohm and Junker [5] from a group theoretic approach. We will get an expansion
in the angular momentum l, where the angle dependent part can be integrated out
and a radial dependent part is left over: the radial path integral. We discuss some
properties of the radial path integral and show that it possible to get from the short
time kernel the radial Schrodinger equation.
The path integral in D-dimensions has the form:
K (D) (x , x ; T )
=

( Z 
 )
i t m 2
x V (x) dt
Dx(t) exp
h t
2

x(tZ
)=x

x(t )=x

Z
Z
m ND/2
(1)
= lim
dx dx(N1)
N 2 i
h


N 
i X
m (j)
exp
(x x(j1) )2 V (x(j) )
.
h

2h
j=1


(3.1)

Now let V (x) = V (|x|) and introduce D-dimensional polar coordinates [31, Vol.II,
Chapter IX]:
x1 = r cos 1
x2 = r sin 1 cos 2
x3 = r sin 1 sin 2 cos 3
...
xD1 = r sin 1 sin 2 . . . sin D2 cos
xD = r sin 1 sin 2 . . . sin D2 sin
40

(3.2)

III.3.1

The Radial Path Integral

where 0 ( = 1, . . . , D 2), 0 2, r =
V (x) = V (r). We have to use the addition theorem:
(1)

PD

2
=1 x

1/2

. Therefore

(2)

cos (1,2) = cos 1 cos 1

+

D2
X

(1)
cos m+1

(2)
cos m+1

m=1

m
Y

sin n(1)

sin n(2)

n=1

D1
Y

sin n(1) sin n(2) ,

(3.3)

n=1

where (1,2) is the angle between two D-dimensional vectors x(1) and x(2) so that
x(1) x(2) = r (1) r (2) cos (1,2). The metric tensor in polar coordinates is
(gab ) = diag(1, r 2 , r 2 sin2 1 , . . . , r2 sin2 1 . . . sin2 D2 ).

(3.4)

If D = 3 equation (3.3) reduces to:

cos (1,2) = cos (1) cos (2) + sin (1) sin (2) cos((1) (2) ).
The D-dimensional measure dx expressed in polar coordinates is
dx = r D1 drd = r D1

D1
Y

(sin k )D1k drdk

k=1

d =

D1
Y

(sin k )D1k dk .

k=1

(3.5)

d denotes the (D 1)-dimensional surface element on the unit sphere S D1 and

(D) = 2 D/2 /(D/2) is the volume of the D-dimensional unit (S D1 -) sphere. The
determinant of the metric tensor is given by
2

D1
Y
D1k
D1
(sin k )
g = det(gab ) = r

(3.6)

k=1

With all this the path integral equation (3.1) yields:

K (D) (r , { }, r , { }; T ) =
Z
Z
Z
 m ND/2 Z
D1
D1
r(N1) dr(N1) d(N1)
r(1) dr(1) d(1)
= lim
N 2 i
h
0
0


N
Y
 i
im 2
2
(j,j1)
r(j) + r(j1) 2r(j) r(j1) cos
V (r(j) ) .

exp
2
h
h
j=1
(3.7)
{} denotes the set of the angular variables. For further calculations we need the
formula [40, p.980]:
 

X
z
z cos
e
=
()
(l + )Il+ (z)Cl (cos ),
(3.8)
2
l=0

41

Important Examples

(for some 6= 0, 1, 2, . . . ), where Cl are Gegenbauer polynomials and I modified

Bessel functions. Equation (3.8) is a generalization of the well known expansion in
three dimension where =
e

z cos

1
2

(remember Cl2 = Pl , [40, p.980]):

r

X
(2l + 1)Il+ 21 (z)Pl (cos )
2z

(3.9)

l=0

Note: It is in some sense possible to include the case D = 2, i.e. = 0 if one uses
l lim0 ()Cl = 2 cos l [40, p.1030], yielding finally [40, p.970]:
z cos

Ik (z) ei k .

(3.10)

k=

The addition theorem for the real surface (or hyperspherical) harmonics Sl on the
S D1 -sphere has the form [31, Vol.II, Chapter IX]:
M
X

Sl ((1) )Sl ((2) ) =

=1

1 2l + D 2 D2
Cl 2 (cos (1,2) )
(D) D 2

(3.11)

with = x/r unit vector in Rd and the M linearly independent Sl of degree l with
M = (2l + D 2)(l + D 3)!/l!(D 3)!. The orthonormality relation is
Z

dSl ()Sl () = ll .
(3.12)
Combining equations (3.8) and (3.11) we get the expansion formula
z((1) (2) )

=e

z cos (1,2)

2
= 2
z

 D2
X
M
2
X

Sl ((1) )Sl ((2) )Il+ D2 (z). (3.13)

l=0 =1

Let = (D 2)/2 > 0 in equation (3.8), then the jth term in equation (3.7) becomes




i
m
im 2
(j,j1)
2
(r + r(j1) ) V (r(j) ) exp
r(j) r(j1) cos
Rj = exp
2h (j)
h
2 i h




 D2

2
im 2
2 i h
i
D2
2
exp
=
(r + r(j1) ) V (r(j) )

mr(j) r(j1)
2
2h (j)
h





X
D2
D
m

lj +
1 Ilj + D2
r(j) r(j1) Clj 2 (cos (j,j1) )
2
2
i h
lj =0


i
im 2
2
(r + r(j1) ) V (r(j) )
exp
2h (j)
h

X
M

X
m

Sljj ((j) )Sljj ((j1) ).

r(j) r(j1)

Ilj + D2
2
i h
=1

2 i h
= 2
mr(j) r(j1)

lj =0

 D2
2

42

(3.14)

III.3.1

The Radial Path Integral

With the help of equations (3.11) and (3.14) now (3.7) becomes:
K (D) (x , x ; T )
D2
2

= (r r )

X
M
X

Sl ( )Sl ( )

l=0 =1

N Z
Z
m
r(N1) dr(N1)
r(1) dr(1) . . .
lim
N i
h
0
0




N
Y
im 2
m
i
2
exp

(r + r(j1) ) V (r(j) ) Il+ D2

r(j) r(j1) .
2
2h (j)
h
i h


(3.15)

j=1

Therefore we can separate the radial part of the path integral:

K

(D)

(r , { }, r , { }; T ) =

(D)

X
l+D2
l=0

D2

D2
2

Cl

(cos ( , ) )Kl (r , r ; T )
(3.16)

with
Z
Z
m N/2
= (r r )
dr(1)
dr(N1)
lim
N 2 i
h
0
0



N
N

Y
i X m

l [r(j) r(j1) ] exp

(r(j) r(j1) )2 V (r(j) )
h

2
j=1
j=1

(D)
Kl (r , r ; T )

D2
2

(3.17)

and the functional measure

(D)

(z(j) ) =

2z(j) ez(j) Il+ D2 (z(j) ),

(3.18)

where z(j) = (m/ i h)r(j) r(j1) .

In the literature often use is been made of the asymptotic form of the modified
Bessel functions
1

I (z) (2z) 2 ez(

41 )/2z

(|z| 1,

(z) > 0).

Then the functional measure becomes (we ignore (z) = 0):

(
#)
"
2
i h
1
D2
(D)
l (z(j) ) exp

l+
2mr(j) r(j1)
2
4
(D)

and Kl

(3.19)

(3.20)

reads:

(D)
Kl (r , r ; T )

D1
2

= (r r )

lim

m N/2
2 i h

dr(1)

"
N
i X
1
D2 2
m
2 (l + 2 ) 4
2
(r(j) r(j1) ) h
exp
h
2
2mr(j) r(j1)
j=1
43

dr(N1)
#

V (r(j) )
.

(3.21)

Important Examples

This last equation seems to suggests a Lagrangian formulation of the radial path
integral:
(D)

kl

(r , r ; T )=

( Z "
2
(l + D2
i t m 2
2 )
r h2
Dr(t) exp
h t
2
2mr 2

r(tZ
)=r

r(t )=r

1
4

V (r)

#)

(3.22)
with
= (r r )
But nevertheless, equation (3.17)
can be written in a similar manner with a nontrivial functional measure:
(D)
kl (r , r ; T )

(D)

kl

(D1)/2

(r , r ; T ) =

r(tZ
)=r

r(t )=r

(D)
Kl (r , r ; T ).

( Z 
 )
t
i
m
(D)
l [r 2 ]Dr(t) exp
r 2 V (r) dt
h t
2

(3.23)

and the whole l-dependence is in l , as defined in equation (3.18) in the lattice

formulation. The asymptotic expansion of I in equation (3.19) in problematic because
it is only valid for (z) > 0, but we have (z) = 0. The complete expansions in this
case reads [40, p.962]

ez X (1)k ( + k + 21 )
I (z)
2z k=0 (2z)k k!( k + 12 )

( + k + 21 )
exp[z ( + 12 )i] X 1

+
(2z)k k!( k + 12 )
2z
k=0






2 41
2 14
1
1
+ exp z +
.
i +

exp z
2z
2z
2
2z

(3.24)

(+ for /2 < arg(z) < 3/2, - for 3/2 < arg(z) < /2). Langguth and
Inomata [67] argued that the inserting of the expansion (3.20) in the path integral
(3.17) can be justified (adopting an argument due to Nelson [80]), but things are not
such as easy. Our arguments are as follows:
1) The Schr
odinger equation





2 d2
h
D1 d

2 l(l + D 2)

+h

+
+ V (r) (r; t) = i h
(r; t) (3.25)
2
2
2m dr
2r dr
2mr
t

can be derived from the short time kernel of the path integral (3.17).
2) For r , r 0 equation (3.24) has the right boundary conditions at the origin. We
have for z 0:


1
2
z + 2 1 + O(z 2 )
2z ez I (z)
(2)!!
and it follows:

(D)
l (z(j) )

r l+(D1)/2
r l+(D1)/2
44

(r 0, r > 0)

(r 0, r > 0)

(3.26)

III.3.1

The Radial Path Integral

(D)

(D)

for all j. If Rn,l and en,l denote the (normalized) radial state functions and
energy levels, respectively, of the D-dimensional problem, we have
(D)

Kl

(r , r ; T ) =

(D)

(D)/
h

(D)

Rn,l (r )Rn,l (r ) e i T en,l

/
h

(3.27)

n=0

and thus for any but s-states - including the factor r (D1)/2 in equation (3.27) we have:
(D)
Rn,l (r) r l+(D1)/2 r (D1)/2 = r l
(3.28)
(D)

which is the correct boundary condition for Rn,l .

Both points are not true for the path integral (3.22), because
1) we did not find any obvious method for deriving the full Schroding er equation
(3.25), the angular dependence comes out wrong.
2) In the Euclidean region T i ( > 0) the functional measure is vanishing like
(
"
#)
1
1
D2 2
2
)

(l
+
(l + D2
2
4
2
4
exp h
+

N
2mr1 r
2mr(N1) r

N1

1
D2
2
X (l + 2 ) 4
(3.29)
exp h
N
2mr(j) r(j1)
j=1

which is also vanishing for r , r 0, but not in the right manner. The points
r = r = 0 are essential singularities, where as the correct vanishing is powerlike.
3) Equation (3.22) seems very suggestive, but it is quit useless in explicit calculations.
Even in the simplest case D = 3, l = 0, V = 0, i.e.
 m N/2
D1
(D)
Kl (r , r ; T ) = (r r ) 2 lim
N 2 i
h

Z
Z
N
im X

dr(1)
dr(N1) exp
(3.30)
(r(j) r(j1) )2
2h

0
0
j=1

cannot be calculated. The integrals turn out to be repeated integrals over errorfunctions which are not tractable. Also the method of Arthurs [4] fails. In this method
one assumes that the integrations in the limit 0 are effectively from to +.
The path integral (3.30) then becomes


im
(D)
D1
2
2
exp
Kl (r , r ; T ) = (r r )
(r r )
2h
(D)

which is wrong. The mirror term Kl

(r , r ; T ) is missing.

Let us discuss some properties of the path integral (3.22)

(D)
kl (r , r ; T )

r(tZ
)=r

(D)
l [r 2 ]Dr(t) exp

r(t )=r

45

( Z 
 )
i t m 2
r V (r) dt .
h t
2

(3.31)

Important Examples

(D)

1) The whole l- and D-dependence is contained in l

(D)

[z(j) ] =

Therefore we can conclude:

(D)

, as defined in (3.18):

2z(j) ez(j) Il+ D2 (z(j) ).

2

(3)

[r 2 ] = l+ D3 [r 2 ]

(3.32)

(3.33)

and all dimensional dependence of the path integral (3.31) can be deduced from
the three dimensional case:
(D)

kl

(3)

(r , r ; T ) = kl+ D3 (r , r , T ).

(3.34)

(3)

From now on we will denote kl (T ) = kl (T ).

2) The boundary conditions we have already discussed; we have for the radial wave
(D)
(D)
functions ul and Rl respectively:
(D)

ul

(r) r l+

D1
2

(D)

Rl

(r) r l

(r 0).

(3.35)

We therefore state: The radial path integral kl (T ) can be written as a superposition

of two one dimensional path integrals:
(1)

(1)

kl (r , r ; T ) = kl (r , r ; T ) (1)l kl (r , r ; T )

(3.36)

with
(1)
kl (r , r ; T )

= lim

m  N2
2 i h

dr(1)

dr(N1)


N 
i X
m 2
(D)
l [r(j) r(j1) ] exp

r(j) V (|r(j) |)
.
h

2
h
j=1
j=1
N
Y

(3.37)

4) From equation (3.36) for the case l = V = 0 we have two free particle path integrals
the solution being:



i m
i m
2
2
exp
(r r ) exp
(r + r )
=
2T h
2T h
 21





i m
i m
m
2
1 exp
,
(r r )
rr
exp
=
2 i h
T
2T h
hT
(3.38)
the second term is the so called mirror term. It is not allowed to drop it, even for
very small T 0, since this violates the boundary condition which demands that
k0V =0 vanishes for r , r 0. These two parts of the Bessel functions express nothing
k0V =0 (r , r ; T )

m
2 i h
T

 21 

46

III.3.1

The Radial Path Integral

else than the famous mirror principle. Actually, in the above Euclidean form k0V =0 is
nothing else but the heat kernel on the half line r 0, which has the representation:
K (3) (r , r ; T ) = K (1) (r , r ; T ) K (1) (r , r ; T )

(3.39)

where K (1) denotes the one-dimensional heat kernel in R. In the language of diffusion
processes: r = 0 is an absorbing boundary and the job is done by the functional
(D)
measure l .
Let us discuss the radial path integral in the language of the Weyl-ordering and
product-ordering rule [49]. We consider the Schrodinger equation in D-dimensions
with a potential V (|x|) = V (r):


h2

LB + V (r) (r, {}; t)

(3.40)
ih
(r, {}; t) =
t
2m
with LB the Laplace Beltrami operator. Introducing the D-dimensional Legendre
operator:


 2
 2
1

2
+
+...
+ (D 2) cot 1
+ (D 3) cot 2
L(D) =
12
1
2
sin2 1 22


1
2
2

1
+
+
cot

+
.
D2
2
D2
sin2 1 . . . sin2 D3 D2
sin2 1 . . . sin2 D2 2
(3.41)
the Laplace operator is:
LB =

2
D1
1
+
+ 2 L2(D) .
2
r
r r r

(3.42)

LB yields:
Rewriting the Hamiltonian H = 2m

p2
H(pr , r, {p , }) = r
2m


1
1
1
2
2
2
p +
+
p +... +
p + VW eyl ({}) (3.43)
2mr 2 1 sin2 1 2
sin2 1 . . . sin2 D2
with



1
1
2
h
1+
+... +
.
VW eyl (r, {}) =
8mr 2
sin2 1
sin2 1 . . . sin2 D2

and the hermitean momenta



D1
h
+
pr =
i r
2r


h
D1
p =
+
cot
i
2
h
p =
.
i
47

(3.44)

(3.45)

Important Examples

Because of the special nature of gab , the product ordering gives the same quantum
potential as the Weyl-ordering, i.e. we have VW eyl = Vprod . In particular we have
the equivalence
N1
Yp
j=1

g (j) exp

N
i X
h

j=1

(j)
LN
, { (j) }, r (j1) , { (j1) })
Cl (r

14

=(g
g )

N p
Y

j=1

g(j)

exp

i N (j) (j)
L (
r , { })
h Cl

(3.46)

where
(j)
LN
, { (j) }, r (j1) , { (j1)})
Cl (r


m
(j)
(j1) 2
d2 (j) ( (j) (j1) )2 + . . .
(j)
(j1) 2
d
(j)
2
) +r
= 2 (r r
(1 1
) + sin
1
2
2
2

d
d
2 (j)
2 (j1)
(j)
(j1) 2
(3.47)
. . . + (sin 1 . . . sin D2 )(
)

denotes a classical Lagrangian on the lattice. LN

r (j) , {(j) }) is similar defined as
Cl (
(3.47), except that one has to take all trigonometrics at mid-points. The features are,
of course, direct consequences of the product form definition as described in section
II.3. With the correct Hamiltonian (3.43) we can consider the Hamiltonian path
integral for the Feynman kernel K:
K (D) (r , { }, r , { }; T )
1

= (g g ) 4

r(t )=r

{(tZ
)= }

Dr(t)Dpr (t)D{}(t)D{p }(t)

r(t )=r
{(t )= }

# )
( Z "
D1
X
i t
pr r +
p Hef f (pr , r, {p }, {}) dt
exp
h t
=1
Z
N Z
N1
Y
Y
dpr(j) d{p(j) }
41
= (g g )
dr(j) d{(j) }
lim
N
(2h)D
j=1
j=1

i X

pr(j) r(j) + {p(j) }{(j) } Hef f (pr(j) , r(j) , {p(j) }, {(j) })

exp

h
j=1

(3.48)

48

III.3.1

The Radial Path Integral

with the effective Hamiltonian

p2r(j)

Hef f (pr(j) , r(j) , {p(j) , (j) }) =

2m

1
1
1 2
p22 + +
p2(j) + VP rod ({(j) })
p(j) +
+
(j)
(j)
(j)
d
d
d
d
2
2
2
2
1
2mr(j)
sin 1
sin 1 . . . sin D2

(3.49)

After the integration over all momenta we get (V VW eyl = Vprod ):

K (D) (r , r , { }, { }; T )
=

r(t )=r

{(tZ
)= }

r(t )=r
{(t )= }

)
( Z

i t  N
V (r, {}) dt ,
LCl (r, r,
{, })
Dr(t) D(t) exp
h t

Z
N D N1 Z
m  2 Y D1
dr(j) dr(j) d(j)
= lim
N 2 i
h
j=1 0

N
i X

exp
LN
(r
,
{
},
r
,
{
})

V
(r
,
{
})
(j)
(j1)
(j1)
(j)
(j)
ef f (j)
h

j=1


(3.50)

with the effective Lagrangian

Lef f (r, r,
{, })
m
= [r 2 + r 2 12 + r 2 sin2 1 22 + + r 2 (sin2 1 . . . sin2 D2 ) 2 ] V (r, {})
2
(3.51)
defined in the same way as LN
Cl in equation (3.47).
The path integral (3.50) with the Lagrangian given by equation (3.47) is too
complicated for explicit calculations. We therefore try to replace equation (3.47)
under the path integral (3.50) by the following expression:
LCl (r, {}, r,
:= m R2
2 Vc ({})
LCl (r, {}, r,
{})
{})
2

(3.52)

where Vc has to be determined and denotes the D-dimensional unit vector on the
S D1 -sphere. Thus we try to replace LCl by a simpler expression and hope that
Vc + VW eyl is simple enough. We have
((1) (2) )2 = 2(1 cos (1,2))
49

(3.53)

Important Examples

with the addition theorem (3.3). We shall use equation (3.53) to justify the replacement (3.52) and thereby derive an expression for Vc . We start with the kinetic term
(x(j) x(j1) )2 expressed in the polar coordinates (3.2), R = r (not fixed), and expand
it in terms of r and . After some tedious calculations we obtain the following
identity


i N (j)
(j)
(j1)
(j1)
L (r , { }, r
, {
})
exp
h Cl




i m (j) 2
(j1) 2
(j) (j1)
(j,j1)
(j)
(j)
r
+r
2r r
cos
i Vc (r , { })
(3.54)
=
exp
h

with

Vc (r

(j)

, {

(j)



2
h
1
1
}) =
++
1+
(j)
(j)
(j)
8mr (j) 2
sin2 1
sin2 1 . . . sin2 D2

(3.55)

(Vc is the same whether or not r (j) = 0). The result is that the potential Vc generated
by these steps cancels exactly VW eyl (r, {})! Therefore we get:

K (D) (r , r , { }, { }; T ) =

r(t )=r

{(tZ
)= }

r(t )=r
{(t )=}

( Z 
 )
i t m 2
x V (r) dt ,
Dr(t) D(t) exp
h t
2

(3.56)
where x 2 has to be expressed in polar coordinates. In the lattice formulation x 2 reads
2
2
x 2 [r(j)
+ r(j1)
2r(j) r(j1) cos (j,j1) ]/2 .

(3.57)

Therefore we arrive at equation (3.52) and both approaches are equivalent as it should
be.
Let us note that equation (3.54) as derived in [49] is effectively a regularization
of the highly singular terms in V . This corresponds to the 1/r 2 -terms in the radial
path integral which have been regularized by the functional measure l [r 2 ]. However,
the functional measure method gives a quite practical tool to regularize such singular
terms.

3.2. The Radial Harmonic Oscillator

Let us now discuss the most important application of equation (3.17), namely the
harmonic oscillator with V (r) = 12 m 2 r 2 . The calculation has first been performed
by Peak and Inomata [84]. However, we present the more general case with timedependent coefficients following Goovaerts [39]. This example will be of great virtue
in the solution of various path integral problems.
50

III.3.2

The Radial Harmonic Oscillator

We have to study
Kl (r , r ; )

Z
Z
m N/2
dr(1)
dr(N1)
lim
= (r r )
N 2 i
h
0
0


N 
Y
im
i
(D)
2
2 (j) 2
l [r(j) r(j1) ] exp

(r(j) r(j1) )
m (t )r(j)
2h
2
h
j=1
N/2 Z

Z
m
2D
r(1) r(1)
r(N1) dr(N1)
= (r r ) 2 lim
N i
h
0
0




N 
Y
im 2
m
i
2
2 (j) 2
exp

(r + r(j1) )
m (t )r(j) Il+ D2
r(j) r(j1)
2
2h (j)
2
h
i h
j=1
 N/2
Z
Z
2
2

i
(r
+r
)/2
2D
r(1) dr(1)
r(N1) dr(N1)
e
= (r r ) 2 lim
N
i
0
0
h
i
2
2
2
exp i((1) r(1)
+ (2) r(2)
+ + (N1) r(N1)
)


Il+ D2 ( i r(0) r(1) ) . . . Il+ D2 ( i r(N1) r(N) )

1D
2

= (r r )

2D
2

lim KlN (T )

(3.58)

where KlN (T ) is defined by the iterated integrals. Furthermore we have set = m/

h
2
2 (j)
and (j) = [1 m (t )/2]. We are now using [40, p.718]
Z

x2

xe



1 (2 + 2 )/4

J (x)J (x)dx =
e
I
2
2

which is valid for () > 1, | arg

[84] one can show that
Z

i r2

re

(3.59)

| < /4 and , > 0. With analytic continuation



ab
i (a2 +b2 )/4 i
e
I
I ( i ar)I ( i br)dr =
2
2 i

(3.60)

is valid for > 1 and () > 0. By means of equation (3.60) we obtain for KlN (T ):
KlN (T )

  N2
Z

Z

i 2
2
=
r(1) dr(1)
r(N1) dr(N1)
(r + r )
exp
i
2
0
0
h
i
2
2
2
exp i((1) r(1) + (2) r(2) + + (N1) r(N1) )


Il+ D2 ( i r(0) r(1) ) . . . Il+ D2 ( i r(N1) r(N) )
2

N
2
2

=
exp i pN r + i qN r Il+ D2 ( i N r r )
2
i
51

(3.61)

Important Examples

where the coefficients N , pN and qN are given by

N =

N1
Y
k=1

pN

2k

N1
X 2k
=
2
4k
k=1

qN =

2
4N1

1 = ,

k+1

1 = 1 ,

k
Y

=
2k
j=1

k+1 = k+1

(k 1)

2
.
4k

(3.62)

We must now determine these quantities. Let us start with the evaluation of k .
Putting
yk+1
2k
=
(3.63)

yk
we obtain
k+1 = k+1

2
4k

yk+1 2yk + yk1

+ k2 yk = 0.
2

(3.64)

In the limit N this gives a differential equation for y

y + 2 (t)y = 0
with solution y = (t + t) (t). Since on one hand side
y1 y0 + y 0 y0
2
y1 y0 1 2y0 ,

( 0),
( 0),

and on the other



y1 y0
1 2
y 0 =
=
1 1 y0 1,

( 1),

we have the boundary conditions

(0) = 0,

(0)

= 1.

Now let us expand yk+1 [(k + 1)] + O(3 ), we observe that

yk+1 2yk + yk1 + 2 2 (t + k)yk = 0
52

(3.65)

III.3.2

The Radial Harmonic Oscillator

is satisfied up to second order in . Therefore expanding (3.64)

k+1 =

[(k + 2)] + O(3 )

yk+2
=
.
2 yk+1
2 [(k + 1)] + O(3 )

Consequently
lim N

N1
(j) + O(3 )
m Y
= lim
N
h j=1 [(j + 1)] + O(3 )

m (0)
m
=
.
N
h (N )
h(T )

= lim

(3.66)

Similarly
lim qN



m
[(N 1)] + O(3 )
= lim
1
N 2
h
[N ] + O(3 )
m(T
)
m [N ] [t + (N 1)]
=
.
= lim
N 2
h
[N ]
2
h(T )

(3.67)

Finally we must calculates pN . We have

lim pN


N1
X 2k

= lim
N
2
4k
k=1


N1
m
1 X

=
lim

2
h N
2 [(k + 1)]
k=1


Z T
dt
1
m
lim

=
2
2
h N
(t)


Z T
m
m
dt
(T )
=

lim
(T )
(T ).
2
2
h(T ) N

2
h(T )
(t)


(3.68)

Furthermore we find
Z
()
dt
lim () = lim
()
2
N
N
(t)
(0) + (0)

lim
= (0)

=1
0



()

1
= lim
lim ()

0
0

()
(0)[(0)

+ (0)]

= 0,
= lim
0
[(0) + (0)]

and therefore (t) is satisfying the boundary conditions

(0)
= 0.

(0) = 1,
53

(3.69)

Important Examples

(t) is found to satisfy the differential equation

+ 2 (t) = 0.
Therefore we have
P (T ) = lim pN =
N

(3.70)

m (T )
2
h (T )

(3.71)

and we have finally for the path integral of the radial harmonic oscillator with timedependent frequency:
Kl (r , r ; T )

= (r r )

2D
2






mr r
m
i m (T ) 2 (T
) 2
.
exp
r +
r
Il+ D2
2
ih
(T )
2
h (T )
(T )
ih
(T )

(3.72)

In particular for (t) = = const.:

1
sin (t t ),

(t) = cos (t t )

(t) =

(t)
= cos (t t )

which yields the radial path integral solution for the radial harmonic oscillator with
time-independent frequency
Kl (r , r ; T )

= (r r )

2D
2





mr r
m
i m 2
2
.
exp
(r + r ) cot T Il+ D2
2
ih
sin T
2
h
ih
sin T

(3.73)

It is possible to get the same result, if one starts with the D-dimensional path integral
in Cartesian coordinates:
K(x , x ; ) =

x(tZ
)=x

Dx(t) exp

x(t )=x

im
2
h


2

x 2 2 x dt

insert for every dimension the solution of the one-dimensional oscillator

r



2xk xk
m
i m
2
2

(xk + xk ) cot T
exp
K(xk , xk ; T ) =
2ih sin T
2
h
sin T

(3.74)

(3.75)

and uses equations (3.8) and (3.11).

The next step is to calculate with the help of equation (3.73) the energy-levels and
state-functions. For this purpose we use the Hille-Hardy-formula [40, p.1038]
 
 X

1
n
t/2
t n! e 2 (x+y)
2 xyt
1
1+t
I
=
exp (x + y)
(xy)/2 Ln() (x)Ln() (y).
1t
2
1t
1t
(n
+

+
1)
n=0
(3.76)
54

III.3.2

The Radial Harmonic Oscillator

With the substitution t = e2 i T , x = mr /h and y = mr /h in equation (3.73)

we get finally:
Kl (r , r ; T ) =

l
l
e i T EN /h RN
(r )RN
(r )

(3.77)

N=0



D
(3.78)
EN = h N +
2
s


l+ D2



2
m 2
m 2
2m ( Nl
m 2 (l+ D2
l
2 + 1)
2 )
RN (r) =

r
r L N l
r .
exp
hr D2 ( N+l+D
h

2
)
2
(3.79)
The path integral for the harmonic oscillator suggests a generalization in the index
l. This will be very important in further applications. For this purpose we consider
(D)
equation (3.17) with the nontrivial functional measure l (r 2 ) l [r 2 ]:
1
Kl (r , r ; T ) =
rr

r(tZ
)=r

r(t )=r

( Z 
 )
t
m
i
Dr(t)l [r 2 ] exp
r 2 V (r) dt
h t
2

(3.80)

in the Schrodinger
The functional measure corresponds to a potential Vl = h
2 l(l+1)
2mr2
equation. Assuming that we can analytically continue in l with () > 1, then
2 1

we get for an arbitrary potential V (r) = h

2 2mr24 navely inserted into the radial path
integral
( Z 
 )
1
2

m
i t m 2
4
r h2
2 r 2 dt .
Dr(t) exp
h t
2
2mr 2
2

r(tZ
)=r

r(t )=r

(3.81)

This path integral must now be interpreted in terms of the functional measure in
equation (3.18). Define (z mr(j1) r(j) / i h):
2

[r ] = lim

N
Y
p

2z(j) ez(j) I (z(j) ).

(3.82)

j=1

Then equation (3.81) must be interpreted with the help of equation (3.82) as
( Z 
 )
1
2

i t m 2
m
4
Dr(t) exp
r h2
2 r 2 dt
2
h t
2
2mr
2

r(tZ
)=r

r(t )=r

:=

r(tZ
)=r

"

im
Dr(t) [r 2 ] exp
2
h

r(t )=r

2

r 2 2 r dt


 



r r m
mr r
i m 2
2
.
r + r cot T I
exp
=
ih
sin T
2
h
ih
sin T
55

(3.83)

Important Examples

This important equation has been derived by Peak and Inomata [84] with the
help of the three-dimensional radial harmonic oscillator, and by Duru [26] with the
corresponding two-dimensional case. Duru considered the radial potential problem
a
+ br 2
2
r

V (r) =

(3.84)

with some numbers a and b. Identifying a = (

h2 /2m)(2 14 ) and b = 21 m 2 it is
simple calculation to express equation (3.83) in terms of a and b, which is omitted
here. However, these authors did not discuss this path integral identity in terms of the
functional measure language. That such a procedure is actually legitimate is beyond
the scope of these notes. It was justified by Fischer, Leschke and M
uller [35] for the
radial path integral and for the Poschl-Teller path integral as well (see below), where
one also has to be careful with the appropriate Besselian functional measure. In the
functional measure interpretation equation (3.83) can be found in references [49] and
[92]. Equation (3.83) is very important in numerous applications.
Let us note the free particle case. In the limit 0 we obtain in equation (3.73)
(=0)

Kl

(r , r ; T )
 


mr r
m
i m 2
2

2D
exp
(r + r ) Il+ D2
= (r r ) 2
2
ih
T
2
hT
ih
T


Z
2
2D
ih
T p
dp p exp
Jl+ D2 (pr )Jl+ D2 (pr )
= (r r ) 2
2
2
2m
0

(3.85)

with wave-functions and energy spectrum

p (r) = r

2D
2

p Jl+ D2 (pr),
2

2 p2
h
Ep =
.
2m

(3.86)

The one-dimensional case gives (i.e. motion on the half-line)

p (r) = r p J 12 (pr) =

2
sin pr,
r

Ep =

2 p2
h
.
2m

(3.87)

However, there is an ambiguity in the boundary condition for r = 0 for D = 1. The

present case here, i.e. p (0) = 0, corresponds to a specific self adjoint extension of the
Hamiltonian H = h2 d2 /dx2 for functions L2 ([0, )) on the half-line.
Finally we calculate the energy dependent Green function for the radial harmonic
oscillator. Making use of the integral representation of the previous section we obtain
Gl (r , r ; E)
Z
ei ET /h Kl (r , r ; T )dt
=i
0





E
[ 12 (l + D
m 2
m 2
2 h
)]
W E , 1 (l+ D2 )
=
r M E , 1 (l+ D2 )
r .
2
h 2
2
2
h 2
2
h >
h <
(r r )D/2 (l + D
2)
(3.88)
56

III.4

Other Elementary Path Integrals

From this representation we can recover by means of the expansion for the -function
the wave-functions of equation (3.79).
The corresponding Green function for the free particle has the form
(=0)

Gl

(r , r ; E)
2m

h(r r )

D2
2

h(r r )

D2
2



 r
 r
1 mE
1 mE

Il+ D2
(r + r |r r |) Kl+ D2
(r + r + |r r |)
2
2
i 2
i 2
h2
h2
(3.89)
i m

Jl+ D2
2

r



r
mE
mE
(1)

(r + r |r r |) Hl+ D2
(r + r + |r r |) ,
2
2
h2
2
h2
(3.90)

where use has been made of the integral representation

 


Z
h

i h

i
dx
ab
a+b
I
= 2I p a b K p a + b .
exp px
x
2x
2x
0
(3.91)
4. Other Elementary Path Integrals
There are two further path integral solutions based on the SU(2) [5, 25, 60] and
SU(1, 1) [5, 61] group path integration, respectively. The first yields the path integral
identity for the solution of the Poschl-Teller potential according to
K (P T ) (x , x ; T )
=

( Z 

 )
2 14
i t m 2
2 2 14
h
Dx(t) exp
+
dt
x
h t
2
2m sin2 x
cos2 x

x(tZ
)=x

x(t )=x


ih
T
(,)
(,)
2
(x )l
(x )
( + + 2l + 1) l
=
exp
2m
l=0



ih
T
2

= sin 2x sin 2x
exp
(2J + 1)
2m
1

J=0, 2

(cos 2x )DJ+ , (cos 2x ) (4.1)

(2J + 1)DJ+
,
2

with the wave-functions given by

p
(,)
(x) = (2J + 1) sin 2x DJ+ , (cos 2x)
n
2

57

(4.2a)

Important Examples

l!( + + l + 1)
= 2( + + 2l + 1)
( + l + 1)( + l + 1)
1

 12

(sin x)+ 2 (cos x)+ 2 Pn(,) (cos 2x).

(4.2b)

 21
l!( + + l + 1)
+ + 2l + 1
=
2++1
( + l + 1)( + l + 1)

2 sin 2x (1 cos 2x) 2 (1 + cos 2x) 2 Pn(,) (cos 2x).



(4.2c)

Here, of course we can analytically continue from integer values of m and n to, say,
real numbers and , respectively .
Similarly we can state a path integral identity for the modified Posch-Teller
potential which is defined as
V

(,)



2 14
2 2 14
h
.

(r) =
2m sinh2 r cosh2 r

(4.3)

This can be achieved by means of the path integration of the SU(1, 1) group manifold.
One gets
K

(M P T )

i2 
ih
T h
(r , r ; T ) =

2(k1 k2 n) 1
2m
n=0


Z
ih
T 2
(,)
(,)
(4.4)
p .
dp p
(r )p (r ) exp
+
2m
0

NM
X

(r ) exp
(,)
(r )(,)
n
n

Introduce the numbers k1 , k2 defined by: k1 = 12 (1 ), k2 = 21 (1 ), where the

correct sign depends on the boundary conditions for r 0 and r , respectively.
In particular for 2 = 14 , i.e. k2 = 14 , 43 , we obtain wave-functions with even and odd
parity, respectively. The number NM denotes the maximal number of states with
0, 1, . . . , NM < k1 k2 21 . The bound state wave-functions read as:
1

n(k1 ,k2 ) (r) = Nn(k1 ,k2 ) (sinh r)2k2 2 (cosh r)2k1 + 2

2 F1 (k1 + k2 + , k1 + k2 + 1; 2k2 ; sinh2 r)

 12
2(2

1)(k
+
k

)(k
+
k
+

1)
1
1
2
1
2
Nn(k1 ,k2 ) =
(2k2 )
(k1 k2 + )(k1 k2 + 1)

(4.5)

( = k1 k2 n), (there is a factor 2 missing in reference [36]). Note the equivalent

formulation
(,) (r)

n!(n + + + 1)
= +
2
(n + + 1)(n + + 1)
58

 12

(1 x) 2 (1 + x)

1
2

Pn(,) (x), (4.6)

III.5

The Coulomb Potential

with the substitutions = 2k2 1, = 2(k1 k2 n)1 = 21, x = 2/ cosh2 r1 with

i1
h
p
dx.
the incorporation of the appropriate measure term, i.e. dr = (1 + x) 2(1 x)
The scattering states are given by:

1
1

p(k1 ,k2 ) (r) = Np(k1 ,k2 ) (cosh r)2k1 2 (sinh r)2k2 2

2 F1 (k1 + k2 , k1 + k2 + 1; 2k2 ; sinh r)

r
h
p
sinh
p
1
(4.7)
Np(k1 ,k2 ) =
(k1 + k2 )(k1 + k2 + )

(2k2 )
2 2

i 21

(k1 + k2 + 1)(k1 + k2 + 1) ,

[ = 12 (1 + ip)].
It is possible to state closed expressions for the (energy dependent) Green functions
for the Poschl-Teller and modified Poschl-Teller potential, respectively. For the PoschlTeller potential it has the form (Kleinert and Mustapic [65])
m
(m1 LE )(LE + m1 + 1)
sin 2x sin 2x
ih

(m1 + m2 + 1)(m1 m2 + 1)



(m1 +m2 )/2
(m1 m2 )/2
1 cos 2x
1 + cos 2x

2
2



(m1 +m2 )/2
(m1 m2 )/2
1 cos 2x
1 + cos 2x

2
2


1 cos 2x
2 F1 LE + m1 ; LE + m1 + 1; m1 m2 + 1;
2


1 + cos 2x
2 F1 LE + m1 ; LE + m1 + 1; m1 + m2 + 1;
2

G(x , x ; E) =

(4.8)

with m1/2 = 21 ( ), LE = 12 + 12 2mE /h and x x . A similar expression

is valid for the modified Poschl-Teller-potential Green function (see [65]) which is,
however, omitted here.
5. The Coulomb Potential
The hydrogen atom is, of course, one of the most interesting subjects in quantum
mechanics. In the beginnings of quantum mechanics and with the atom model of
Rutherford it was a riddle how to tract a system which is from the classical physics
point of view unstable and doomed to vanish into pure radiation. Its was Bohrs
genius, postulating the famous rules
H that only a countable number of orbits are allowed
satisfying the quantum condition pdq = nh (n N). However, this old quantum
mechanics was not sufficient, because it fails e.g. in the case of the helium-atom, and
it was Pauli who solved the hydrogen problem in the terms of the new quantum
mechanics developed by Schrodinger and Heisenberg. It is surprising, that Pauli did
59

Important Examples

not use a differential equation and solves the corresponding Eigenvalue problem as
Schrodinger did later, instead he exploited in fact the hidden SO(4) symmetry of
the Kepler-Coulomb-problem. This symmetry gives classically rise to an conserved
quantity, called Lenz-Runge vector. This additional symmetry allows also a separation
of the Coulomb problem in parabolic coordinates. This vector points from the focal
point of the orbit the perihel.
Ever since the success of quantum mechanics the hydrogen atom was the model
to test the theory, may it be Diracs relativistic quantum mechanics, where first the
fine-structure constant = e2 /hc arises.
It was for a long time a really nuisance that this important physical system
could not be treated by path integrals. Calculating wave functions and energy
levels remains more or less a simple task in the operator language, but even the
construction of the Green function (resolvent kernel) was impossible for a long time.
It takes as long as 1979 as Duru and Kleinert [27, 28] finally applied a long-known
transformation in astronomy (Kustaanheimo-Stiefel transformation [66]) in the path
integral of the Coulomb problem and were successful. Even more, their idea of
transforming simultaneously coordinates and the time-slicing opens new possibilities in
solving the huge amount of unsolved path integral problems. In particular, Coulombrelated potentials, in the sense, that all these problems can be reformulated in terms
of confluent hypergeoemtric functions, like the Morse-potential could be successfully
treated. However, the original attempt of Duru and Kleinert was done in a more or
less formal manner, and it does not takes a long time when it was refined by Inomata
[58] and Duru and Kleinert [28].
Closely related to the original Coulomb problem is, of course, the 1/r-potential
discussion in RD . The D = 2 case in discussed in this subsection, whereas the original
Coulomb problem in the next.
As it turns out its Schrodinger equation for the Coulomb potential is separable in
four coordinate systems:
1) Spherical coordinates:
x = r sin cos
y = r sin sin
z = r cos

(r 0, 0 , 0 2).

(5.1)

Separation of variables in this coordinate system is, of course, not a specific feature
of the Coulomb problem, but a common property of all radial potentials.
2) Parabolic coordinates:
x = cos
y = sin
1
z = ( 2 2 )
2

(, 0, 0 2).

(5.2)

Here electric and magnetic fields can be introduced without spoiling separability.
60

The 1/r-potential in R2

III.5.1

3) Spheroidal coordinates:
x = sinh sin sin
y = sinh sin cos
z = cosh cos + 1

( 0, 0 , 0 2).

(5.3)

Here a further charge can be introduced and therefore these coordinates are
suitable for the study of the hydrogen ion H2+ .
4) Spheroconical coordinates:
x = r sn(, k) dn(, k )

r 0, || K, || 2K ,

y = r cn(, k) cn(, k )
z = r dn(, k) sn(, k )

0 k, k 1,

k 2 + k = 1.

(5.4)

sn(, k), cn(, k) and dn(, k) denote Jacobi elliptic functions (e.g. [40, pp.910])
of modulus k and with real and imaginary periods 4K and 4iK , respectively.
Here the corresponding wave functions in and can be identified with the wave
functions of a quantum mechanically asymmetric top.
We will discuss the path integral for the Coulomb system in the first two of these
coordinate systems. For the usual polar- and parabolic coordinate coordinate system
the path integration can be exactly performed. In the remaining two, however, the
theory of special functions of these coordinate is poorly developed and no solution
seems up to now available. Nevertheless it is possible to formulate the Coulombproblem in these coordinate systems and point out some relations to other problems
connected to these coordinates [46].
It is surprising that the Coulomb path integral was first solved in Cartesian
coordinates (where the Kustaanheimo-Stiefel transformation works) and not in polar
coordinates. Looking at the appropriate formul we see that even in the onedimensional case (in polar coordinates) we need the one-dimensional realization of
the Kustaanheimo-Stiefel transformation and a space-time transformation.
5.1. The 1/r-Potential in R2 [28, 57]
We consider the Euclidean two-dimensional space with the singular potential V (r) =
Z e2 /r (r = |x|, x R2 ). Here, e2 denotes the square of an electric charged Z
its multiplicity (including sign). However, as already noted, this potential is not the
potential of a point charge in R2 . The classical Lagrangian now has the form
L(x, x)
=

m 2 Z e2
x +
.
2
r

(5.5)

Of course, bound and continuous states can exist, depending on the sign of Z. The
path integral is given by
K(x , x ; T ) =

" Z 
 #
m 2 Z e2
i t
dt .
Dx(t) exp
x +
h t
2
r

x(tZ
)=x

x(t )=x

61

(5.6)

Important Examples

Discussion of this path integral are due to Duru and Kleinert [27,28] and Inomata
[58]. However, the lattice-formulation is not trivial for the 1/r-term. In fact, it is too
singular for a path integral, respectively, a stochastic process and some regularization
must be found. This is known for some time, and it turns out that the KustaanheimoStiefel transformation does the job.
As it turns out one must perform a space-time transformation in order to solve
the path integral (5.6). We perform first the transformation
 

2
2
x1 = ,
x2 = 2,
u
R2 ,
(5.7)

which casts the original Lagrangian into the form

Z e2
L(u, u)
= 2m( 2 + 2 )(2 + 2 ) + 2
.
+ 2
The metric tensor (gab ) and its inverse (g ab ) are given by



1
1
1 0
2
2
ab
(gab ) = 4( + )
,
(g ) =
2
2
0 1
4( + ) 0

(5.8)

0
1

(5.9)

with determinant g = det(gab ) = 16( 2 + 2 )2 and dV = dx1 dx2 = 4( 2 + 2 )dd.

The hermitean momenta corresponding to the scalar product
Z Z
(1 , 2 ) = 4
dd ( 2 + 2 )1 (, )2 (, )
(5.10)

are given by



h

p =
+
,
i 2 + 2




h

p =
+
.
i 2 + 2

(5.11)

Following our general theory of Chapter II.5 we start by considering the Legendre
transformed Hamiltonian


2
Z e2
h2 2
+

E.
(5.12)
HE =
2m x21
x22
r
which gives

 2

h2

1
2
Z e2

HE =
+

E.
2m 4( 2 + 2 ) 2 2
2 + 2

(5.13)

Therefore the time-transformation is given by = f (, ) with f (, ) = 4( 2 + 2 )

and the space-time transformed Hamiltonian has the form

2  2
2

=
4Z e2 4E( 2 + 2 )
+
H
2m 2 2
(5.14)
1 2
2
2
2
2
=
(p + p ) 4Z e 4E( + )
2m
62

The 1/r-potential in R2

III.5.1

with the momentum operators and vanishing quantum potential V :

p =

h
,
i

p =

h
,
i

V = 0.

(5.15)

This two-dimensional transformation can be interpreted as a two-dimensional Kustaanheimo-Stiefel transformation [66]. It has the general feature that it performs a change
of variables to square-root coordinates, note r = 2 + 2 . The general properties of
transformations like this states the following problem (see e.g. [28, 46] for some review
of the relevant literature): For which values of D does exist a formula
2
2
(x21 + + x2D )(y12 + + yD
) = z12 + + zD
,

(5.16)

where the zi are homogeneous bilinear forms in x and y. Following a theorem of

Hurwitz and Lam this type of transformation can only be realized in the spacedimensions D = 1, 2, 4, 8. The assumption about z now implies that
z = B(x) y
X
X
zi =
(B k )li xk yl =
B(x)li yi .
k,l

(5.17)
(5.18)

For the special case x = y one has

X

zi2

X

x2j

2

(5.19)

and the matrix B satisfies the condition

t

B(x)B(x) = |x|2 .

(5.20)

The one-dimensional case is trivial, but we shall use it in the calculation of the
path integral for the 1/r-potential in polar coordinates. The D = 2 case we have just
encountered. The four-dimensional variation of this transformation has been used a
long time ago in astronomy for the purpose of regularizing the Kepler problem. The
D = 8 case is degenerate. Here the matrix B has the form:

x1
x2

x3

x4
B(x) =
x5

x6

x7
x8

x2
x1
x4
x3
x6
x5
x8
x7

x3
x4
x1
x2
x7
x8
x5
x6

x4
x3
x2
x1
x8
x7
x6
x5

x5
x6
x7
x8
x1
x2
x3
x4

and maps R8 R and is thus of no further use.

63

x6
x5
x8
x7
x2
x1
x4
x3

x7
x8
x5
x6
x3
x4
x1
x2

x8
x7

x6

x5

x4

x3

x2
x1

(5.21)

Important Examples

In the next Section (hydrogen-atom) we need the D = 4 case which is more involved
and not one-to-one.
To incorporate the time transformation
Z t
d
,
s = s(t )
(5.22)
s(t) =
4r()

t
and its lattice definition 4
r (j) s(j) = 4
u(j) 2 (j) into the path integral (5.6) we
now use from the set of equations (II.4.35), namely equation (II.4.35a). Observing
that in the path integral (5.6) the measure changes according to
N 
Y

j=1

m
2 i h
t(j)

N1
Y

(j)
(j)
dx1 dx2

 N1
N 
Y
m
1 Y

d (j) d (j)
=
r
2 i h
(j)
j=1

j=1

(5.23)

j=1

we arrive at the path integral transformations equations (u R2 ):

Z
1
dE e i T E/h G(x , x ; E)
K(x , x ; T ) =
2 i h



Z

ds K(u , u ; s ) + K(u , u ; s )
G(x , x ; E) = i

(5.24)
(5.25)

is given by
where the space-time transformed path integral K
Z
Z

4 i Z e2 s /
h

D(s) D(s)
K(u , u ; s ) = e
( Z 
 )
i s m 2
( + 2 ) + 4E( 2 + 2 ) ds .
exp
h 0
2

(5.26)

Note that the factor 1/r has been exactly canceled by means of equation (II.4.35a).
This is a specific feature of this potential problem. Furthermore we have taken into
account that our mapping is of the square-root type which gives rise to a sign
ambiguity. Thus, if one considers all paths in the complex x = x1 + ix2 -plane from
x to x , they will be mapped into two different classes of paths in the u-plane: Those
which go from u to
u and those going from u to u . In the cut complex x-plane
for the function u = x2 these are the paths passing an even or odd number of times
through the square root from x = 0 and x = . We may choose the u corresponding
to the initial x to lie on the first sheet (i.e. in the right half u-plane). The final u
can be in the right as well as the left half-plane and all paths on the x-plane go over
into paths from u to u and those from u to u [28]. Thus the two contributions
arise in equation (5.25). Equation (5.26) is interpreted as the path
p integral of a twodimensional isotropic harmonic oscillator with frequency = 8E/m. Therefore
we get



m
u u
4 i Z e2 s
m

2
2

K(u , u ; s ) =
(u + u ) cot s 2
exp

.
2ih
sin s
h
2 i h

sin s
(5.27)
64

The 1/r-potential in R2

III.5.1

Introducing now two-dimensional polar-coordinates

=
so that u u =

r cos

,
2

r sin

,
2

(r > 0, [0, 2])

r r cos( )/2. Thus

Z
m ds
G(x , x ; E) =
h 0 sin s





4 i Z e2 m
m r r

exp
s
(r + r ) cosh
cos
h
2ih

ih
sin s
2

(5.28)

Using the expansion




=
ei l I2l (z)
cos z cos
2

(5.29)

(5.30)

l=

we get for the kernel G(x , x ; E):

1 X i l( )
G(x , x ; E) =
e
Gl (r , r ; E),
2

(5.31)

l=

where the radial kernel Gl (E) is given by

Gl (r , r ; E)

 


Z
m
m r r
4 i Z e2 s
2m ds

exp

(r + r ) cot s I2l
=
h

sin s
h
2ih

ih
sin s
0
2 i p
h
m , substitution u

Z

2p
hs
m

and Wick-rotation)
  
2m
du
2i
2p r r

=
exp
u + i p(r + r ) coth u I2l
h 0 sinh u
ap
i sinh u
(Substitution sinh u = 1/ sinh v)
2
Z 





v ap
2m
coth
exp i p(r + r ) cosh v I2l 2 i p r r sinh v dv
=
h 0
2

(Reinsert hp = 2mE)
r
r


1
Z e2
m 1
m
1

+l

=
2E (2l)!
2
h
2E
r r
r
r


8mE
8mE

W Z e2 m ,l
2 r> M Z e2 m ,l
2 r< . (5.32)
h

2E
h

2E
h
h
(Set =

In the last step we have used (2.22) and r> , r< denotes the larger (smaller) of r , r ,
respectively. The representation (5.32) shows that Gl (E) has in the complex energyplane poles which are given at the negative integers nl = 0, 1, 2, . . . at the argument
65

Important Examples

of the -function and a cut on the real axis with a branch point at E = 0. Thus we
get a discrete and continuous spectrum, respectively:
EN =
Ep =

mZ 2 e4
,
2
h2 (N 12 )2

2 p2
h
,
2m

(N = 1, 2, . . . )

(p R).

(5.33)
(5.34)

Here we have introduced the principle quantum number N := nl + l + 1. This is the

well-known result.
To determine the discrete spectrum we consider the first line in equation (5.32),
identify in the Hille-Hardy formula t = e2u , x = mr and y = mr and thus get

m X (2 i p r r )2l
Gl (r , r ; E) =
i
h n=0 n + l + 12 ap

n!

(2l)

exp[ i p(r + r )]L(2l)

n (2 i pr )Ln (2 i pr ).
(2l + n)!

(5.35)

Taking equation (5.35) and the nth residuum gives the energy levels and the wave
functions. Inserting into equation (5.29) thus yields the Green functions for the
discrete levels for the two-dimensional 1/r-potential

X
X
N,l (r , )N,l (r , )
G(x , x ; E) = h

EN E

(5.36)

N=1 l=

and the wave functions of the discrete spectrum are given by:
 12 
l
2r
(N l 1)!
N,l (r, ) =
a2 (N 12 )3 (N + l 1)!
a(N 12 )




r
2r
(2l)
exp
.
i l LNl1
a(N 12 )
a(N 21 )


(5.37)

Furthermore we have used the abbreviation a = h

2 /(mZ e2 ) (the first Bohr-radius).
The correct normalization to unity can be seen from the relation
Z

h
i2
2(N + 2 )(N + l + )!
(2l++1)
,
x2l++2 ex LNl1 (x) dx =
(N l 1)!

(5.38a)

respectively
Z

+a x

h
i2
(2n + + 1)(n + + 1)
()
Ln (x) dx =
.
n!
66

(5.38b)

The 1/r-potential in R2

III.5.1

In order to determine the continuous wave functions we use the dispersion relation
[40, p.987]
Z




dx e2ix 21 + + ix 12 + ix Mix, ()Mix, ()

 


2
2
=
. (5.39)
exp ( + ) tanh J2
cosh
cosh

and get (
hp = 2mE )
Gl (r , r ; E)
Z
= 2 i p

  

ds
2
p rr
4 i Z e2

exp
s p(r + r ) cot s I2l

sin s
h
sin s
0


Z
4 i Z e2
i

ds exp
s
=
h
r r 0


Z
|( 21 + l + ip)|2
4 i p
phs

exp
+p Mi p,l (2 i pr )M i p,l (2 i pr )dp
2
(2l + 1)
m
0
 
Z

dp
exp
=h

2 2
ap
h p /2m E
0




i
i
12 + l + ap
21 + l ap

(5.40)
M i ,l (2 i pr )M i ,l (2 i pr ).

ap
ap

2 r r (2l)!

Here again the residuum at E = h

2 p2 /2m has been taken in the
integral. Thus the
wave functions of the continuous spectrum are given by (
hp = 2mE):
r




1

i
1
1
p,l (r, ) =
exp

+l+
i l M i ,l (2 i pr). (5.41)
ap
4 2 r (2l)!
2
ap
2ap
Note that
 
 


 1


 1
i
i




 2 + l + ap M api ,l (2 i pr) =  2 + l + ap M api ,l (2 i pr)

(5.42)

and the wave-functions are therefore basically real. These results of the 1/r potential
for the discrete and continuous spectrum, respectively, are equivalent with the results
of the operator approach. The complete Feynman kernel therefore reads
K(x , x ; T ) =

e i T EN /h N,l (r , )N,l (r , )

l=0 N=1

Z
X
l=0

dp e i T Ep /h p,l (r , )p,l (r , )

(5.43)

with wave functions as given in equations (5.37) and (5.41) and energy spectrum
equations (5.33, 5.34).
67

Important Examples

5.2. The 1/r-Potential in R3 - The Hydrogen Atom [27, 28, 41, 55, 57, 58]
We consider the Euclidean three-dimensional space with the singular potential V (r) =
Z e2 /r (r = |x|, x R3 ) with Z and e2 as in the previous Section. This potential
corresponds in the space R3 , of course, to the potential of a point charge and is thus
the kernel of the Poisson equation in R3 . The classical Lagrangian has the form
L(x, x)
=

m 2 Z e2
x +
.
2
r

(5.44)

The path integral is given by

K(x , x ; T ) =

" Z 
 #
m 2 Z e2
i t
dt .
x +
Dx(t) exp
h t
2
r

x(tZ
)=x

x(t )=x

(5.45)

This path integral was first successfully solved (actually the Green function or resolvent
kernel) by Duru and Kleinert [27], followed by further contributions of Inomata [58]
and Duru and Kleinert [28], Ho and Inomata [55] Grinberg, Mara
non and Vucetich
[41,42] and Kleinert [62].
The transformation in the two-dimensional case corresponds to the two-dimensional
realization of the Kustaanheimo-Stiefel transformation and we must look for a generalization. However, we have a the relevant space R3 whereas the relevant KustaanheimoStiefel transformation maps R4 R3 which means that the transformation in question is not one-to-one. We have namely [28]
u3
x1
u2
x2 =

u1
x3
u4

u4
u1
u2
u3

u1
u4
u3
u2

u1
u2
u3 u2
,
u3
u4
u4
u1

(5.46)

or in matrix notation x = A u with x R3 and u R4 , respectively. Several

modifications of the matrix A are used in the literature, e.g. [10,11]

u3
u

4
A=
u1
u2

u4
u3
u2
u1

u1
u2
u3
u4

u2
u1
,
u4
u3

(5.46b)

and see [66] as well. The transformation has the property At A = u21 + u22 + u23 + u24 =
r = |x| and no Jacobean exist. However, a fourth coordinate can be described by
x4 (s) = 2

(u4 u1 u3 u2 + u2 u3 u1 u4 )d.

(5.47)

Therefore we must circumvent this problem. The idea [27,58] goes at follows: We
consider the lattice formulation of equation (5.45) and introduce a factor one by
68

The 1/r-Potential in R3 - The Hydrogen Atom

III.5.2

means of
r
1=

m
2 i h
T


m

2
d exp
( )
2ih
T



N Z
 m  N2 Y
i m 2 (j)
(j)
= lim
d exp

N 2 i
h
2
h

j=1

(5.48)

which gives for equation (5.45)

Z
N1 Z
m 2N Y
3 (j)
d (j)
d x
d lim
K(x , x ; T ) =
N
2
i

j=1


N 
i X
2
m 2 (j)
Ze
exp
( x + 2 (j) ) + (j)
h

2
r
j=1

(5.49)

with the to-be-defined quantity r(j) . We now realize the transformation A(u) on
(j)
(j1)
(j)
) (we use in the following the abbreviation x4 ):
midpoints [58] u
a = 12 (ua +ua
xa(j)

=2

4
X

(j)

Aab (
u(j) )ub

(5.50)

b=1

We have

(j) 2

At (
u(j) ) A(
u(j) ) = u
1

(j) 2

+u
2

(j) 2

+u
3

(j) 2

+u
4

=: r(j)

(5.51)

which defines r(j) . Furthermore

(j)

(j)

(j)

(j)

2 x1 + 2 x2 + 2 x3 + 2 x4

(j)
(j)
(j)
(j) 
= 4
r (j) 2 u1 + 2 u2 + 2 u3 + 2 u4 =: 4
r (j) 2 u(j) .
(j)

(5.52)

(j)

Infinitesimal this has the form dxa = 2Aab (

u(j) )dua (sums over repeated indices
understood) and the Jacobean of this transformation exists and is given by
(x1 , x2 , x3 , )
= 24 r(j) 2 .
(u1 , u2 , u3 , u4 )

(5.53)

Thus we arrive at the transformed classical Lagrangian

L(u, u)
= 2mu2 u 2 +

Z e2
.
u2

(5.54)

We now repeat the reasoning of the previous Section by observing that under the
time-transformation equation (5.22)
s(t) =

d
,
4r()
69

s = s(t )

Important Examples

and its lattice definition 4

r (j) s(j) = 4
u(j) 2 (j) the measure in the path integral
transforms according to
N 
Y

j=1

2 N1
N1
N 
Y
m 2 Y
1 Y
m
(j) 4 4 (j)
d4 u(j) . (5.55)

(2
u ) d u =

2
(j)
2 i h
(4r ) j=1 2 i h

j=1
j=1

Respecting again equation (II.4.35a) which produces a factor 4r we get the path
integral transformation
Z
1
dE e i T E/h G(x , x ; E)
K(x , x ; T ) =
2 i h

Z
2

, u ; s )
ds e4 i Z e s /h K(u
G(x , x ; E) = i

(5.56)

is given by
where the space-time transformed path integral K
, u ; s )
K(u
2N

Z
1
m

=
d lim
N 2 i h
4r


N 
N1
X
YZ
m 2 (j)
i
u + 4E u
(j) 2
d4 u(j) exp

h
2
j=1

j=1

m
2 i h
sin s

2 Z

d
exp
4r



u u
m
2
2

(u + u ) cot s 2

.
2ih

sin s

(5.57)

Here we have used once again the known solution for the harmonic oscillator, where
we have
harmonic oscillator with frequency
p actually a four-dimensional isotropic

=
8E/m. Note that the factor 1/r has not been canceled as in the twodimensional case.
Up to now we have not discussed the problem of an eventually quantum correction
appearing in the transformation procedure. As usual we consider the Legendre
transformed Hamiltonian
Z e2
2
h
(3)
E.
HE =
2m
r

(5.58)

Let us write HE in the coordinates:

u1 = u cos cos
u2 = u cos sin
u3 = u sin cos
u4 = u sin sin

(u = |u| =

r, = 2 , = )

(0 , 0 2)
70

(5.59)

The 1/r-Potential in R3 - The Hydrogen Atom

III.5.2

Then the Schr

odinger equation in polar coordinates




Z e2
2
1 2
h2 2
+
+ L
E (r, , ) = 0

2 r 2
r r r 2
r
is transformed into


 2

h2 1
Z e2

3
2

+
4K 2 E (u) = 0
2 4s2 s2
s s
s
with
K2 =

1
2
2
2
2

(tan

cot
)

cos

2
sin2 2
2

(5.60)

(5.61)

(5.62)

is the Casimir-operator of the group SO(4). Writing back into u1 , . . . , u4 we get for
HE in the coordinates q R4 :
= 4u2 H
E = 1 4 4Z e2 4Eu2
H
2m
and with puk = i h
/uk :
4

1 X 2
puk 4Z e2 4Eu2 .
Hef f (pu , q) =
2m

(5.63)

k=1

and no quantum correction appears. Therefore equation (5.57) is correct.

To perform the -integration in equation (5.57) we introduce polar coordinates

cos
2
2

u2 = r sin sin
2
2

u3 = r cos cos
2
2

u4 = r cos sin
2
2
u1 =

r sin

0 2 .
0 4

(5.64)

We have with equation (5.47)

(s) =

( cos )r()d.

(5.65)

The polar coordinates give the expansion

u u = r r






+

+

+ cos cos
cos
cos
sin sin
2
2
2
2
2
2
(5.66)
71

Important Examples

Using the expansion

z cos

ei I (z)

(5.67)

twice, respecting due to the explicit form of (s) from equation (5.65) above
Z

d(s )
=
r

(5.68)

and that the integration over yields 41 2 we get for the resolvent kernel:

2


Z
m
m
4 i Z e2 s
i

ds
exp

(r + r ) cot s
G(x , x ; E) =
0
2ih
sin s
h
2ih

 



r
X

r
r r
m

m
i ( )
I
. (5.69)
sin sin
cos cos

e
I
ih
sin s
2
2
ih
sin s
2
2
=

The last sum can be exactly performed with the help of



z1 z2 e i
z1 z2 ei

 2

q

z12

z22

2z1 z2 cos

(1)n In (z2 )I+n (z1 ) ei n ,

n=

(5.70)

in particular


q
2
2

z + z + 2zz cos ,
I (z)I (z ) = I0

(5.71)

and the trigonometric expressions sin2 2 = 21 (1 cos ) and cos2 2 = 21 (1 + cos ).

Thus

2
Z
i
m

G(x , x ; E) =
ds
0
2ih
sin s

 


m r r
m

4 i Z e2 s

, (5.72)

(r + r ) cot s I0
cos
exp
h
2ih

ih
sin s
2
where cos = cos cos + sin sin cos( ). The I0 -Bessel function can be
expanded in terms of Legendre-polynomials:



2X
I0 z cos
=
(2l + 1)Pl (cos )I2l+1 (z).
2
z

(5.73)

l=0

This relation can be derived from the general expansion



kz
2

X
( + l)
I (kz) = k
(1)l (2l + )2 F1 (l, l + ; 1 + ; k 2 )I2l+ (z).
l!(1 + )
l=0
(5.74)

72

The 1/r-Potential in R3 - The Hydrogen Atom

III.5.2

Equation (5.73) is recovered with the identifications = 1, = 0, k = cos 2 and

Thus the resolvent kernel can be expanded
Pl (z) = (1)l 2 F1 (l, l + 1; 1; z+1
2 ).
according to
G(x , x ; E) = G(r , , , r , , ; E)

1 X
=
(2l + 1)Pl (cos )Gl (r , r ; E)
4

(5.75)

l=0

l
X

Yln ( , )Yln ( , )Gl (r , r ; E),

(5.76)

l=0 n=l

where the radial Green function Gl (E) is given by

Gl (r , r ; E)





Z
ds
2m
m r r
4 i Z e2 s m

=
exp

(r +r ) cot s I2l+1
h
2ih

ih
sin s
h r r 0 sin s
r
r


i Z e2 m
1
m
1
1+l+
=
rr
2E (2l + 1)!
h
2E
!
!
r
r
8mE
8mE
r> M Z e2 m ,l+ 1
r< . (5.77)
W Z e2 m ,l+ 1
h

2E
2
h

2E
2
h2
h2

From the poles of the -function at z = nr = 0, 1, 2, . . . we read off the bound

state energy levels which are given by
EN =

mZ 2 e4
,
2
h2 N 2

(N = nr + l + 1 = 1, 2, 3, . . . )

(5.78)

which are the well-known Balmer levels.

The determination of the wave functions is analogous to the previous twodimensional case and thus we only state the result for the bound state wave
functions of the hydrogen atom:
1


l



r
2r
2r
2 (N l 1)! 2
(2l+1)
exp
LNl1
Yln (, ).
N,l,n (r, , ) = 2
N
a3 (N + l)!
aN
aN
aN
(5.79)
A quite complicated attempt was made by Grinberg, Mara
non and Vucetich [41]
to determine these functions from the wave functions of the corresponding fourdimensional oscillator. The continuous state wave functions of the hydrogen
atom read as:




1
i

p,l,n (r, , ) =
1+l+
exp
M i ,l+ 1 (2 i pr)Yln (, ).
ap
2
ap
2ap
2(2l + 1)!r
(5.80)
73

Important Examples

Of course, these wave functions form a complete set. The spectrum of the continuous
states is, of course, given by
2 p2
h
Ep =
,
(p > 0).
(5.81)
2m
These results coincides with the operator approach. The complete Feynman kernel
therefore reads
X

K(x , x ; T ) =
e i T EN /h N,l (r , )N,l (r , )
l=0 N=1

Z
X
l=0

dp e i T Ep /h p,l (r , )p,l (r , )

(5.82)

with wave functions as given in equations (5.79) and (5.80) and energy spectrum
equations (5.78) and (5.81), respectively.
5.3. Coulomb Potential and 1/r-Potential in D Dimensions [15, 55, 91]
In this Section we discuss the 1/r-potential in D dimensions. For D = 3 we have, of
course, the Coulomb problem. The whole calculation is quite similar as in the previous
cases, we just have to use a Kustaanheimo-Stiefel transformation in one dimension.
The D-dimensional path integral problem was first discussed by Chetouani and
Hammann [15], whereas the radial problem by Ho and Inomata [55] and [91].
We start with the D-dimensional path integral with the singular potential V (r) =
q1 q2 /r, where r = |x| (x RD ):
K(x , x ; T ) =

" Z 
 #
m 2 q1 q2
i t
dt .
x +
Dx(t) exp
h t
2
|x|

x(tZ
)=x

x(t )=x

(5.83)

We denote the coupling by q1 q2 to emphasize the possibility that two charges q1

and q2 can interact which each other. As we have studied very explicitly in Section
III.3.1, the angular variables can be integrated out yielding
K(x , x ; T ) =

X
l=0

(D)

Sl

(D)

({ })Sl

({ })Kl (r , r ; T ),

(5.84)

where the radial kernel Kl (T ) is given by

Kl (r , r ; T )
= (r r )

1D
2

r(t )=r

(r r )

1D
2

" Z 
1)2
(l + D
m 2
i t
2
Dr(t) exp
r h2
h t
2
2mr 2

r(tZ
)=r

 #
q1 q2
dt
+
r

" Z 
 #
t
m
q
q
i
1
2
dt .
r 2 +
Dr(t)l+ D 1 [r 2 ] exp
2
h t
2
r

r(tZ
)=r

r(t )=r

1
4

74

(5.85)

III.5.3

Coulomb Potential and 1/r-Potential in D Dimensions

(D)

The Sl ({}) are the hyperspherical harmonics on the S D1 -sphere. The corresponding Hamiltonian to the path integral (5.85) has the form


h2 d2
l(l + D 2) q1 q2
D1 d
H =
+h
2
+

2
2m dr
r dr
2mr 2
r
(5.86)
1
D
2

1)

(l
+
1 2
q
q
1 2
2
4
=
p +h
2

2m r
2mr 2
r
with the momentum operators equation (3.45). We now perform the space-time
transformation
Z t
d
,
s = s(t ),
r = F (u) = u2
(5.87)
s(t) =
4r()

t
2

with f = 4u2 , thus that F = f and we get the space-time formed Hamiltonian
1
2
1 2
2 (2l + D 2) 4
p +h

4Eu2 4q1 q2
(5.88)
H=
2m u
2mu2
with the momentum operator


h d
2D 3
pu =
.
(5.89)
+
i du
2u
Here we have applied the results of space-time transformations for one-dimensional
Hamiltonians. Thus we get the transformation formul
Z
1

Kl (r , r ; T ) =
dE e i T E/h Gl (r , r ; E)
2 i h

(5.90)
Z
4 i q1 q2 s /
h

23 D
ds e
Kl (u , u ; s ),
Gl (r , r ; E) = 2 i(u u )
0

l (s ) is given by
where the kernel K
l (u , u ; s )
K

u(t )=u

" Z 
 #
s
i
m
Du(s)2l+D2 [u2 ] exp
u 2 + 4Eu2 ds
h 0
2

u(tZ
)=u

with =

u u
p





m 2
mu u
m
2
,
exp
(u + u ) cot T I2l+D2
ih
sin T
2ih

ih
sin T

8E/m. The Green function we obtain in the usual way with the result

Gl (r , r ; E)

= (r r )

(5.91)

2D
2

2m
h

ds
sin s




m
4 i q1 q2 s
m
r r

exp

(r + r ) cot s I2l+D2
h
2
hi
ih
sin s


p
r
m
q1hq2 2E
m l + D1
2
1D
2

= (r r )
2E
(2l + D 2)!
r
r


8mE
8mE
W q1 q2 m ,l+ D2
2 r> M q1 q2 m ,l+ D2
2 r< .
h

2E
2
h

2E
2
h
h


75

(5.92)

Important Examples

Thus it is easy to display the wave functions and spectrum for the discrete contribution
for the D-dimensional 1/r-problem [a = h
2 /(m|q1 q2 |), n N]:
 21 
D/2 
l
1
2
(N l 1)!
2r
N,l (r, {}) =
(N + l + D 3)!
2(N + D3
a(N + D3
a(N + D3
2 )
2 )
2 )




r
2r
(2l+D2)
(D)
exp
LNl1
Sl ({})
D3
D3
a(N + 2 )
a(N + 2 )


(5.93a)


 21



1
a(N + l + D 3)!(N L 1)!


2r
(D)
D1
r 2 WN+ D3 ,l+ D2
Sl ({})
2
2
a(N + D3
)
2

= (1)Nl1

EN =

1
N + D3
2

mq12 q22
,
2
2
)
2
h (N + D3
2

(N = 1, 2, . . . )

(5.93b)
(5.94)

For the continuous spectrum we get similarly

D1
i
+ ap
1D l +
2
p,l (r, {}) = r 2
2(2l + D 2)!


mq12 q22
(D)
exp
M i ,l+ D2 (2 i pr)Sl ({}) (5.95)
2
ap
2
2
h p
with Ep =

h
2 p2
2m .

The complete Green function thus has the form


mq12 q22
N,l (r , { })N,l (r , { })
exp i T 3
G(x , x ; E) =
D3 2
2
h (N + 2 )
N=1 l=0


Z
X
p2
p,l (r , { })p,l (r , { }). (5.96)
exp i h
T
+
2m
0

X
X

l=0

Finally we can state the following path integral identity =

Z

8E/m :

" Z 
 #
t
m

i
dt
r 2 +
Dr(t) [r 2 ] exp
h t
2
r

Z
2m r r
=
dE e i T Eh
i h

 


Z
m r r
4is
m
ds

exp

(r + r ) cot s I2

sin s
h
2ih

ih
sin s
0
p
r
Z
q q
m
)
m ( + 21 1h 2 2E
dE e i T E/h
= ih

2E
(2 + 1)

76

III.5.3

Coulomb Potential and 1/r-Potential in D Dimensions

r



r
8mE
8mE
W q1 q2 m ,
2 r> M q1 q2 m ,
2 r<
h

2E
h

2E
h
h
)
(
Z
Z

X
dp
1
n (r )n (r ) +
p (r )p (r ).
dE e i T E
= ih

E
E

E
n
p

0
n=0

(5.97)

The discrete state wave functions and energy spectrum are given by (a = h
2 /mq1 q2 )
 12 
+ 12
n!
2r
1 a(n + 2 + 1)
a(n + + 12 )
2




2r
r
(2)
L
(5.98a)
exp
a(n + + 12 ) n
a(n + + 12 )

 21


1
2r
(1)n
Wn++ 21 ,
=
n + + 12 an!(n + 2 + 1)
a(n + + 21 )
(5.98b)

1
n (r) =
n++

En =

m2
.
2
h2 (n + + 21 )2

(5.99)

For the continuous states we have similarly for the wave functions and the energy
spectrum
r


1
i

1 ( + 2 ap )
M i , (2 i pr)
exp
(5.100)
p (r) =
ap
2 (2 + 1)
2ap
with Ep =

h
2 p2
2m .

In particular this gives the identities

m ( 12 + )
W, (2 i kr> )M, (2 i kr< )
ih
k (1 + 2)
n!
1
m X
=
hk n (n + 2 + 1) n + + 21
h
i
1
exp k(r + r ) (4k 2 r r )+ 2 Ln(2) (2kr )Ln(2) (2kr )
Z
1 m dp ep
+
2 hk p i

( 12 + + ip)( 12 + i p)

Mi p, (2 i kr )M i p, (2 i kr ),
2
(1 + 2)
p
or alternatively (and explicitly, = (q1 q2 /h) m/2E)

(5.101)

r
r

r

m ( 12 + )
8mE
8mE

2 r> M,
2 r<
W,
2E (1 + 2)
h
h
Z

X
1
dp
=h

n (r )n (r ) + h

p (r )p (r ). (5.102)
E E
Ep E
0
n=1 n
77

Important Examples

Let us note that it is not difficult to analyze the Green function by taking the nth
residuum in equation (5.92). Using the appropriate expansion for the -function again
we get
r

X
m
(1)n
1
1
q

G(r , r ; E) =
q
q
1
m
n! n + + 1 2
2En (2 + 1)
n=0
2
h

2En




2r>
2r<
Wn++ 21 ,
Mn++ 21 ,
+ regular terms
a(n + + 12 )
a(n + + 12 )

(5.103)

which gives using the relations of the Whittaker-functions with the Laguerre-polynomials the wave functions (5.92) and the energy spectrum (5.94). In particular we have
for D = 1
 21
 
2r r/an (1) 2r
1
,
e
Ln
n (x) =
an3
an
an
mZ 2 e4
En = 2 2 ,
nN
2
h n


i
1
e/2ap M i , 1 (2 i pr).
p (x) = 1
ap 2
ap
2


(5.104)
(5.105)
(5.106)

Because, the domains x < 0 and x > 0 are separated the wave-functions (5.104,5.106)
are doubly degenerated.
We can study several special cases (see [15]):
i) We first consider D = 2 and get
G(x , x ; E)

iq q p m
m X |l| h1 2 2E
2E
(2|l|)!
l=
r
r



8mE
8mE
i l( )

2 r> M i q1 q2 m ,|l|
2 r<
e
W i q1 q2 m ,|l|
h

2E
h

2E
h
h
r



Z
du
2mE
q1 q2
2m
2m
exp
u
(r + r ) coth u

=
h 0 sinh u
h
E
h




2 2mE r r
1 X i l( )
e
I2l

2
h sinh u
l=



Z
m du

8mE
.
cosh
=
r r cos
h 0 sinh u
h sinh u
2
r
r


q1 q2
2m
2mE

exp
u
(r + r ) cosh u .
(5.107)
h
E
h2
1

(r r ) 2
=
2

This is equivalent with equation (5.29).

78

III.5.3

Coulomb Potential and 1/r-Potential in D Dimensions

ii) We consider D arbitrary. The Green function reads in the integral representation,
where we expand in addition the hyperspherical harmonics in terms of Gegenbauerpolynomials
r



Z
2mE
2m 2D du
q1 q2
2m

G(x , x ; E) =
(r r ) 2
exp
u
(r +r ) coth u
h
sinh u
h
E
h
0




r
X
) D2
(2l + D 2)( D2
2mE
r
2
(1,2)
2
. (5.108)

Cl 2 (cos
)I2l+D2
h sinh u
4 D/2
l=0

The l-summation can be performed by means of equation (5.74) and taking into
account that we have for the Gegenbauer-polynomials


1
(n + 2)
2

Cn (cos ) =
2 F1 2 + n, n; + ; sin
n!(2)
2
2


(5.109)
1
n (n + 2)
2
= (1)
.
2 F1 2 + n, n; + ; cos
n!(2)
2
2
Thus we get
G(x , x ; E)



Z
1D
du
2mE
q1 q2
2m
2m 2D
(r r ) 2 (4) 2
exp
u
(r +r ) coth u

=
h
sinh u
h
E
h
0

X
(l + D 2)

(1)l (2l + D 2)
l!( D1
2 )
l=0





(1,2)
D1
2 2mE r r
2
2 F1 l + D 2, l;
I2l+d2
; cos
2
2
h sinh u
3D

 1D


2
2

(1,2)
2m
4h

r r cos
=
h
2
2mE
r



Z
du
2mE
q1 q2
2m

u
(r + r ) coth u

D+1 exp
h
E
h
(sinh u) 2
0



(1,2)
2 2mE r r
(5.110)
cos
I D3
2
h sinh u
2
Alternatively this can be written as
G(x , x ; E)

 3D

 1D
Z
2
4
du
4h
r r + x x
2m

=
D+1
h
2
2mE
(sinh u) 2
0
r
r





q1 q2
2m
2 2mE r r + x x
2mE
exp
u
(r +r ) coth u I D3

2
h
E
h
h sinh u
2
1D

 2
Z
du
m
2h
2
2 3D
4

=
(x y )
D+1
h
2mE
(sinh u) 2
0
r
p






2mE
x2 y 2
2m
q1 q2
exp

2mE
, (5.111)
u
x coth u I D3
2
h
E
h
h sinh u
79

Important Examples

where x = r + r and y = |x x |. For D = 2 we recover, of course, equations

1
(5.29,5.107) with I 21 (z) = (2/z) 2 cosh z.
D = 1 yields with (1,2) 2:
G(x , x ; E)
Z
2m du
xx
=
h
sinh u
0
r



 
2mE
q1 q2
2 2mE x x
2m

exp
u
(x + x ) coth u I1

h
E
h
h sinh u
r
r


q1 q2
m
m
1
=

2E
h
2E


r
r
8mE
8mE

W q1 q2 m , 1
(5.112)
2 r> M q 1 q 2 m , 1
2 r< .
h

2E 2
h

2E 2
h
h
Furthermore one can show, by using the relation:


d
zdz

m 

I (z)
z

I+m (z)
z +m

that the D-dimensional Green function for the 1/r-potential is connected with the
(D 2)-dimensional Green function by the relation
GD (x|y; E) =

1
GD2 (x|y; E).
2y y

(5.113)

Therefore
GD (x|y; E) =

GD (x|y; E) =

2yy

 D1
2

G1 (x|y; E),

2yy

 D2
2

G2 (x|y; E),

D = 1, 3, 5, . . . ,
(5.114)
D = 2, 4, 6, . . . .
(5.115)

80

III.5.5

Axially Symmetric Coulomb-Like Potential

5.4. Axially Symmetric Coulomb-Like Potential [7-13, 46-48, 88, 89]

It is possible to consider the even more complicated potential problem, namely [46]
LCl =

q2
m 2
(x + y 2 + z 2 ) + p
2
x2 + y 2 + z 2

1
ch2
z
bh2
p

2
2
2
2
2m x + y
2m (x + y ) x2 + y 2 + z 2
2

q
bh
m 2
ch cos
(r + r 2 2 + r 2 sin2 2 ) +

2
2
r
2m r 2 sin2
2mr 2 sin
2

(5.116a)

(5.116b)

m
2q 2
bh
ch
2 2
m 2
( + 2 )(2 + 2 ) + 2 2 2 + 2

.
2
2
+ 2
2m 2 2
m 2 2 ( 2 + 2 )
(5.116c)

Here I have displayed the corresponding classical Lagrangian in cartesian, polar and
parabolic coordinates. It belongs to the class of potentials mentioned long ago by
Makarov, Smorodinsky, Valiev and Winternitz [71] which are separable in a specific
coordinate system and furthermore exactly solvable. This potential was discussed by
Carpio-Bernido, Bernido and Inomata [10] starting from cartesian coordinates and
using the four-dimensional realization of the Kustaanheimo-Stiefel transformation.
However, this potential is separable in polar as well as parabolic coordinates and we
shall give the explicit calculation for all of them.
1) Cartesian coordinates
In our line of reasoning we follow reference [10], however with some simplifications because we have already discussed the realization of the four-dimensional KustaanheimoStiefel in the discussion of the hydrogen atom. We consider the path integral representation corresponding to equation (5.116b)

K(x , x ; T ) = lim

N1
YZ
m  3N
2
dx(j) dy (j) dz (j)
2 i h
j=1

N 
i X
m 2 (j)
q2
exp
( x + 2 y (j) + 2 z (j) ) + (j)
h
2
r
j=1
2

(j)

bh
1
z
ch

.
(j)
2
(j)
2
(j)
(j)
2m r
2m r (r 2 z (j) 2 )
z

(5.117)

Similarly to equation (5.45) we insert a factor one

1 = lim



N N Z
i m 2 (j)
m  2 Y (j)
d exp
.

2 i h
2
h

j=1
81

(5.118)

Important Examples

We repeat the steps from equations (5.49) to (5.57), i.e. we realise the four-dimensional
Kustaanheimo-Stiefel transformation on midpoints and arrive together with the
transformation formul
Z
1

dE e i T E/h G(x , x ; E)
K(x , x ; T ) =
2 i h

(5.119)
Z

4 i q 2 s /
h

G(x , x ; E) = i
ds e
K(u , u ; s ),
0

with the transformed path integral

Z
1

1 (u , u , u , u ; s ) K
2 (u , u , u , u ; s ),
K(u , u ; s ) =
d K
1
1
2
2
3
3
4
4
4r

(5.120)

and the kernels K1 (s ), K2 (s ), respectively, are given by

1 (u1 , u1 , u2 , u2 ; s )
K
=

u1 (tZ )=u
1

Du1 (s)

u1 (t )=u1

u2 (tZ )=u
2

Du2 (s)

u2 (t )=u2

)
( Z 
2 
(b
+
c)
h
i s m 2
ds
(u + u 22 ) + 4E(u21 + u22 )
exp
h 0
2 1
2m(u21 + u22 )

1 (tZ )=
1

1 D1 (s)

1 (t )=1

1 (tZ )=
1

D1 (s)

1 (t )=1

( Z 
)
1
b
+
c

i s m 2
4
exp
( + 21 21 ) + 4E21 h2
ds
h 0
2 1
2m21
(5.121)

1
p
2 1 1

ei 1 (1 1 )

1 =

1 (tZ )=
1

1 (t )=1

( Z 
)
1
2
i s m 2
2b+c + 4
2
D1 (s) exp
+ 4E1 h
ds
h 0
2 1
2m21

m
i 1 (
1 1 )
e
2 i h
sin s =
1
 


m1 1
m 2
2

( + 1 ) cot s I1
.
exp
2ih
1
ih
sin s

(5.122)

Here we have introduced two-dimensional polar coordinates u1 = 1 cos 1 , u2 =

1 sin 1 , separated the 1 and the 1 path integration in the usual way, applied
82

III.5.5

Axially Symmetric Coulomb-Like Potential

the
p solution of theradial harmonic oscillator and have used the abbreviations =
8E/m, 1 = + 2 + b + c. Similarly

m
ei 2 (2 2 )

2 i h
sin s =
2
 


m2 2
m 2
2

( + 2 ) cot s I2
exp
2ih
2
ih
sin s

2 (u , u , u , u ; s ) =
K
3
3
4
4

(5.123)

with u3 = 2 cos 2 , u4 = 2 sin 2 , 2 = + 2 + b c. We identify variables [CayleyKlein parameters, c.f.equation (5.59)]

1 =

r cos

2 =

+
1 =
2

r sin

2 =
2

(5.124)

(0 , 0 2, 0 4, d = r d ). Collecting factors we thus obtain



2

Z
m
m
4 i q 2 s
i

ds
exp

(r + r ) cot s
G(x , x ; E) =
0
2ih
sin s
h
2ih


 


X
m r r

m r r

i ( )

e
I1
I2
.
sin sin
cos cos
ih
sin s
2
2
ih
sin s
2
2
=

(5.125)

(Compare with equation (5.69)). Using the addition theorem [31, Vol.II, p.99]:
z
J (z cos cos )J (z sin sin )
2
= (sin sin ) (cos cos )

X
( + + l + 1)( + l + 1)
J++2l+1 (z)

(1)l ( + + 2l + 1)
l! 2 ( + 1)(l + + 1)
l=0

2 F1 (l, + + l + 1; + 1; sin2 )2 F1 (l, + + l + 1; + 1; sin2 ).

(5.126)

we get in the usual way by performing he s integration

G(x , x ; E)

 


1 X i ( )

1
2
=
sin sin
e
cos cos
2 =
2
2
2
2

X
l=0

(1 + 2 + 2l + 1)

(1 + 2 + l + 1)(1 + l + 1)
l! 2 (1 + 1)(2 + l + 1)
83

Important Examples





2
2 F1 l, 1 + 2 + l + 1; 1 + 1; sin
2 F1 l, 1 + 2 + l + 1; 1 + 1; sin
2
2
Z

ds
m
4 i q 2 s /
h

r r h 0 sin s





m
m r r

exp
(r + r ) cot s I1 +2 +2l+1
2ih

ih
sin s

1 X i ( ) X (1 + 2 + 2l + 1)
(1 + 2 + l + 1)l!
e
=
2 =
2
(1 + l + 1)(2 + l + 1)
l=0

 


1
2 (1 ,2 )
( , )
cos cos
Pl
(cos )Pl 1 2 (cos )
sin sin
2
2
2
2
r
1
m [ 21 (1 + 2 ) + l + 1 p]

rr
2E
(1 + 2 + 2l + 2)
Wp,l+ 1+1 +2 (2 i kr> )Mp,l+ 1+1 +2 (2 i kr< ).
(5.127)


with the abbreviations p = q m/2Eh2 , k = +2mE/h. This is the result of

reference [10]. The energy-levels are determined by the poles of the -function and
are given by
mq 4
En = 2
,
n N0 .
(5.128)
2 2
2
h (n + l + 1 + 1 +
)
2
2

The corresponding wave functions will be calculated below, where we separate the
radial from the angular path integrations directly.
2) Polar coordinates
We consider the path integral corresponding the Lagrangian (5.116a):
K(x , x ; T )
=

r(tZ
)=r

Dr(t)

r(t )=r

(tZ
)=

(t )=

sin D(t)

(tZ
)=

D(t)

(t )=

( Z 
i t m 2
exp
(r + r 2 2 + r 2 sin2 2 )
h t
2


 )
1
bh2
q2
ch2 cos
h2
1+

+
dt
r
2m r 2 sin2 8m
2mr 2 sin2
sin2

1 X i ( )
e
K (r , r , , ; T )
=
2 =

with the kernel K (T ), which in turn is also separated:

K (r , r , , ; T )
84

(5.129)

III.5.5

Axially Symmetric Coulomb-Like Potential

= (r r sin sin ) 2

r(tZ
)=r

Dr(t)

r(t )=r

(tZ
)=

D(t)

(t )=

( Z 
2
i t m 2
2 + b + c cos
2 2
exp
(r + r ) h
h t
2
2mr 2 sin2

1
4

 )
q2
h2
+
+
dt
r
8mr 2

= (r r sin sin ) 2

r(tZ
)=r

rDr(t)

(tZ
)=

(t )=

r(t )=r

( Z "
 2
i t m 2 4m 2
D(t) exp
r +
r
h t
2
2
2

# )
2
1
1
2
2
2

+
b

+
b
+
c

q
h

4
4
+
+
dt
h2
h2
2
2
2
2
8mr cos
r
8mr 2
8mr sin
=

(1 ,2 )

(1 ,2 )

( )l

( )Kl (r , r ; T ),

(5.130)

l=0

the (Poschl-Teller) wave functions

( , )
l 2 1 ()

l! (1 + 2 + l + 1)
1 + 2 + 2l + 1
=
1+
+
2 1 2
(1 + l + 1)(2 + l + 1)
(1 cos )

1
2

(1 + cos )

2
2

(1 ,2 )

Pl

 21

(cos ),

(5.131)

(1,2 as before) and the radial path integral

1
Kl (r , r ; T ) =
rr

( Z 
)
1
2

i t m 2 q2
4
Dr(t) exp
r +
h2
dt
h t
2
r
2mr 2

r(tZ
)=r

r(t )=r

(5.132)

=
+ 2l + 1] . This path integral is nothing but a Coulomb potential path
integral, however with generalized angular momentum. Thus we just analytically
continue the known result in l and obtain

1
2 (1 + 2

r
m ( + 1 p)

Wp,+ 12 (2 i kr> )Mp,+ 12 (2 i kr< ).

2E (2 + 2)
(5.133)
The energy spectrum was already stated in equation (5.128) and the corresponding
1
G (r , r ; E) =
rr

85

Important Examples

wave functions are

ei (1 ,2 )

(r, , ) =

()
2 l

 21 


n!
2
2r

(n + + 12 )2 a3 (n + + 12 )(n + 2 + 2)
a(n + + 12 )






r
2r
(2+1)

exp
Ln

a(n + + 12 )
a(n + + 21 )

r



( + 12 + ap
)

M i ,+ 1 (2 i pr)
exp
ap
2
2 r(2 + 2)
2ap

(5.134)

and a denotes the Bohr radius.

3) Parabolic coordinates
The potential described in equation (5.116c) is also separable in parabolic coordinates,
in the operator approach as well in the path integral formalism. We consider the path
integral formulation corresponding to the Lagrangian (5.116c):
K(x , x ; T ) K( , , , , , ; T )
=

(tZ
)=

D(t)

(t )=

(tZ
)=

D(t)( 2 + 2 )

(tZ
)=

D(t)

(t )=

(t )=

( Z 
m
i t m 2
( + 2 )(2 + 2 ) + 2 2 2
exp
h t
2
2

 )
h2 (b 14 ) ch2
2 2
2q 2
dt

+ 2
+ 2
2m 2 2
m 2 2 ( 2 + 2 )

1 X i ( )
=
e
K ( , , , ; T )
2 =

(5.135)

with the kernel K (T )

K ( , , , ; T ) = ( )

12

(tZ
)=

D(t)

(t )=

( Z 
i t m 2
exp
( + 2 )(2 + 2 )
h t
2

(tZ
)=

D( 2 + 2 )

(t )=

 
1
2
ch2
2 2
2q 2
2b+ 4
dt
h

+ 2
+ 2
2m 2 2
m 2 2 ( 2 + 2 )

This path integral is now tracked by a time-transformation according to

Z t
d
d
2 + 2 ,
= (j)
s(t) =
,
(j)
2
2
t () + ()
86

(5.136)

(5.137)

III.5.5

Axially Symmetric Coulomb-Like Potential

therefore
Z
1
K ( , , , ; T ) =
dE e i ET /h G ( , , , ; E)
2 i h

Z
2

( , , , ; s )
G ( , , , ; E) = i
ds e2 i q s /h K

(5.138)

with (note that this is a two-dimensional time-transformation and the prefactor is

one):
( , , , ; s ) = K
1 ( , ; s ) K
2 ( , ; s ).
K
(5.139)
1,2 (s )
This gives for the kernels K
1 ( , ; s ) = 1
K

( Z 
 )
s
i
m
1 [ 2 ]D(s) exp
2 + E 2 ds
h 0
2

(tZ
)=

(t )=

 


m
m 2
m
2

=
exp
( + ) cot s I1
ih
sin s
2ih

ih
sin s
(5.140)
p

2 (s )
with = 2E/m, 1 = + 2 + b + c. Similarly for K
2 ( , ; s ) = 1
K

(tZ
)=

(t )=

( Z 
 )
s
i
m
2 [ 2 ] exp
2 + E 2 ds
h 0
2

 


m
m
m 2
2

=
exp
( + ) cot s I2
ih
sin s
2ih

ih
sin s
(5.141)

(2 = + 2 + b c). Integrating over s yields

(bound)

(x , x ; E) = h

= n1 ,n2

N ( , , )N ( , , )
EN E
=0

(5.142)

with the energy spectrum (note = q 2 /hN )

EN =

mq 4
,
2
h2 N 2

N = n1 + n2 + 1 +

1 + 2
.
2

The wave functions have the form


 21
ei
2
2n1 !n2 !
,n1 ,n2 (, , ) =

2 a2 N 3 (n1 + 1 + 1)(n2 + 2 + 1)

 2 
1 
2





2 + 2 (1 ) 2

(2 )
Ln1
Ln2
.
exp

aN
aN
2aN
aN
aN
87

(5.143)

(5.144)

Important Examples

For the continuous spectrum we obtain

(cont.)

(x , x ; E) = h

Z
X

dp

with
Ep =

p,, ( , , )p,, ( , , )
(5.145)
Ep E

2 p2
h
2m

(5.146)

and the wave functions

1+1
1+2
i
i
i
ei [ 2 + 2 ( + ap )][ 2 + 2 (
p,, (, , ) =
3
(1 + 1 )(1 + 2 )
(2) 2

M i (+
2

1
i
ap ), 2

( i p 2 )M i (
2

i
)]
ap

2
i
ap ), 2

( i p 2 ). (5.147)

Finally we state the relation of the Green function in parabolic coordinates. We have

2 X

i m
G( , , , , , ; E) =
ei ( )
2 i h

=
 


Z

m
m
ds
I2
I1

ih
sin s
ih
sin s
sin2 s
0


2 i q 2 m 2
2
2
2

exp
s
( + + + ) cot s .
h
2ih

(5.148)

Note the similarity to the u1 , u2 , u3 , u4 approach. Actually the Kustaanheimo-Stiefel

approach in cartesian coordinates produces a separation in parabolic coordinates. We
switch bach to polar coordinates by means of

2 = r z = r(1 cos ) = 2r sin2

2
2 = r + z = r(1 + cos ) = 2r cos2

(5.149)

Usingpagain the addition theorem (5.126) with a rescaling s 2s , /2 so that

= 8E/m we recover the Green function (5.125).
Let us finally note that even more complicated Coulomb-like potentials can be
exactly solved [47, 48] which are, however, only separable in parabolic coordinates.

88

Bibliography

BIBLIOGRAPHY
[1] S.Albeverio: Some Recent Developments and Applications of Path Integrals; in: M.C.Gutzwiller
et al.(eds.): Bielefeld Encounters in Physics and Mathematics VII; Path Integrals From meV to
MeV, 1985 (World Scientific, Singapore, 1986).
[2] S.Albeverio, P.Blanchard and R.Hegh-Krohn: Some Applications of Functional Integration;
in Mathematical Problems in Theoretical Physics, p.265 (Eds.: R.Schrader, R.Seider, and
D.A.Uhlenbrock) Lecture Notes in Physics 153 (Springer-Verlag, Berlin, 1982).
[3] S.Albeverio, Ph.Combe, R.Hegh-Krohn, G.Rideau, M.Sirgue-Collin, M.Sirgue and R.Stora
(Eds.): Feynman Path Integrals, Lecture Notes in Physics 106 (Springer-Verlag, Berlin, 1979).
[4] A.M.Arthurs: Path Integrals in Polar Coordinates; Proc.Roy.Soc.(London) A 313 (1969) 445;
Path Integrals in Curvilinear Coordinates; Proc.Roy.Soc.(London) A 318 (1970) 523.
[5] M.B
ohm and G.Junker: Path Integration Over Compact and Noncompact Rotation Groups;
J.Math.Phys. 28 (1987) 1978.
[6] D.P.L.Castrigiano and F.St
ark: New Aspects of the Path Integrational Treatment of the Coulomb
Potential; J.Math.Phys. 30 (1989) 2785.
[7] M.V.Carpio-Bernido: Path Integral Quantization of Certain Noncentral Systems with Dynamical
Symmetries; J.Math.Phys. 32 (1991) 1799.
[8] M.V.Carpio-Bernido: Green Function for an Axially Symmetric Potential Field: A Path Integral
Evaluation in Polar Coordinates; J.Phys.A: Math.Gen. 24 (1991) 3013.
[9] M.V.Carpio-Bernido and C.C.Bernido: An Exact Solution of a Ring-Shaped Oscillator Plus a
c sec2 /r2 Potential; Phys.Lett. A 134 (1989) 395.
[10] M.V.Carpio-Bernido, C.C.Bernido and A.Inomata: Exact Path Integral Treatment of Two
Classes of Axially Symmetric Potentials; in Third International Conference on Path Integrals
From meV to MeV, 1989, p.442; Eds.: V.Sa-yakanit et al.(World Scientific, Singapore, 1989).
[11] M.V.Carpio-Bernido and A.Inomata: Path Integral Treatment of the Hartmann Potential; in
Bielefeld Encounters in Physics and Mathematics VII; Path Integrals From meV to MeV,
1985; eds.:M.C.Gutzwiller et al.(World Scientific, Singapore, 1986).
[12] L.Chetouani, L.Guechi and T.F.Hammann: Exact Path Integral for the Ring Potential; Phys.
Lett. A 125 (1987) 277.
[13] L.Chetouani, L.Guechi and T.F.Hammann: Exact Path Integral Solution of the Coulomb Plus
Aharonov-Bohm Potential; J.Math.Phys. 30 (1989) 655.
[14] L.Chetouani, L.Guechi and T.F.Hammann: Generalized Canonical Transformations and Path
Integrals; Phys.Rev. A 40 (1989) 1157.
[15] L.Chetouani and T.F.Hammann: Coulomb Greens Function, in a n-Dimensional Euclidean
Space; J.Math.Phys. 27 (1986) 2944.
[16] R.De, R.Dutt and U.Sukhatme: Mapping of Shape Invariant Potentials Under Point Canonical
Transformations; J.Phys.A: Math.Gen. 25 (1992), L 843.
[17] B.S.DeWitt: Dynamical Theory in Curved Spaces. I. A Review of the Classical and Quantum
Action Principles; Rev.Mod.Phys. 29 (1957) 377.
[18] C.DeWitt-Morette: Feynman Path Integrals: I.Linear and Affine Technique, II.The Feynman
Green Function; Commun.Math.Phys. 37 (1974) 63.
[19] C.DeWitt-Morette, A.Maheswari and B.Nelson: Path Integration in Non-Relativistic Quantum
Mechanics; Phys.Rep. 50 (1979) 255.
[20] P.A.M.Dirac: The Lagrangian in Quantum Mechanics; Phys.Zeitschr.Sowjetunion 3 (1933) 64;
reprinted in: Quantum Electrodynamics (Ed.J.Schwinger) Dover, New York, 1958, p.312;
[21] J.C.DOlivio and M.Torres: The Canonical Formalism and Path Integrals in Curved Spaces;
J.Phys.A: Math.Gen. 21 (1988) 3355; The Weyl Ordering and Path Integrals in Curved Spaces;
Path Summation: Achievements and Goals, Trieste, 1987, p.481, eds: S.Lundquist et al. (World
Scientific, Singapore, 1988).
[22] J.S.Dowker: Covariant Feynman Derivation of Schr
odingers Equation in a Riemannian Space;
J.Phys.A: Math., Nucl.Gen. 7 (1974) 1256.
[23] J.S.Dowker: Path Integrals and Ordering Rules: J.Math.Phys. 17 (1976) 1873.
[24] J.S.Dowker and I.W.Mayes: The Canonical Quantization of Chiral Dynamics; Nucl.Phys. B 29
(1971) 259.

89

Bibliography

[25] I.H.Duru: Path Integrals Over SU(2) Manifold and Related Potentials; Phys.Rev. D 30 (1984)
2121.
[26] I.H.Duru: On the Path Integral for the Potential V = ar2 + br2 ; Phys.Lett. A 112 (1985) 421.
[27] I.H.Duru and H.Kleinert: Solution of the Path Integral for the H-Atom; Phys.Lett. B 84 (1979)
185.
[28] I.H.Duru and H.Kleinert: Quantum Mechanics of H-Atoms From Path Integrals; Fortschr.Phys.
30 (1982) 401.
[29] R.Dutt, A.Khare and U.P.Sukhatme: Supersymmetry, Shape Invariance, and Exactly Solvable
Potentials; Amer.J.Phys. 56 (1988), 163.
[30] S.F.Edwards and Y.V.Gulyaev: Path Integrals in Polar Co-ordinates; Proc.Roy.Soc.(London) A
279 (1964) 229.
[31] A.Erd
elyi, W.Magnus, F.Oberhettinger and F.G.Tricomi (Eds.): Higher Transcendental Functions, Vol.I-III (McGraw Hill, New York, 1955).
[32] R.P.Feynman: The Principle of Least Action in Quantum Mechanics; Ph.D.Thesis, Princeton
University, May 1942.
[33] R.P.Feynman: Space-Time Approach to Non-Relativistic Quantum Mechanics; Rev.Mod.Phys.
20 (1948) 367.
[34] R.P.Feynman and A.Hibbs: Quantum Mechanics and Path Integrals (McGraw Hill, New York,
1965).
[35] W.Fischer, H.Leschke and P.M
uller: Changing Dimension and Time: Two Well-Founded and
Practical Techniques for Path Integration in Quantum Physics; Universit
at Erlangen-N
urnberg
preprint, August 1991, J.Phys.A: Math.Gen. 25 (1992).
Path Integration in Quantum Physics by Changing the Drift of the Underlying Diffusion Process;
Universit
at Erlangen-N
urnberg preprint, February 1992.
[36] A.Frank and K.B.Wolf: Lie Algebras for Systems With Mixed Spectra.I. The Scattering P
oschlTeller Potential; J.Math.Phys. 25 (1985) 973.
[37] I.M.Gelfand and A.M.Jaglom: Die Integration in Funktionenr
aumen und ihre Anwendung in
der Quantentheorie; Fortschr. Phys. 5 (1957) 517; Integration in Functional Spaces and its
Applications in Quantum Physics; J.Math.Phys. 1 (1960) 48.
[38] J.-L.Gervais and A.Jevicki: Point Canonical Transformations in the Path Integral; Nucl.Phys.
B 110 (1976) 93.
[39] M.J.Goovaerts: Path-Integral Evaluation of a Nonstationary Calogero Model; J.Math.Phys. 16
(1975) 720.
[40] I.S.Gradshteyn and I.M.Ryzhik: Table of Integrals, Series, and Products (Academic Press, New
York, 1980).
[41] H.Grinberg, J.Mara
non and H.Vucetich: Atomic Orbitals of the Nonrelativistic Hydrogen Atom
in a Four-Dimensional Riemann Space Through the Path Integral Formalism; J.Chem.Phys. 78
(1983) 839;
The Hydrogen Atom as a Projection of an Homogeneous Space; Zeitschr.Phys. C 20 (1983) 147.
[42] H.Grinberg, J.Mara
non and H.Vucetich: Some Remarks on Coulomb and Oscillator Problems in
RN ; KINAM 5 (1983) 127.
[43] C.Grosche: The Product Form for Path Integrals on Curved Manifolds; Phys.Lett. A 128 (1988)
113.
[44] C.Grosche: Path Integral Solution of a Class of Potentials Related to the P
oschl-Teller Potential;
J.Phys.A: Math.Gen. 22 (1989) 5073.
[45] C.Grosche: Separation of Variables in Path Integrals and Path Integral Solution of Two Potentials
on the Poincar
e Upper Half-Plane; J.Phys.A: Math.Gen. 23 (1990) 4885.
[46] C.Grosche: Coulomb Potentials by Path Integration; Fortschr.Phys. 40 (1992), 695.
[47] C.Grosche: Path Integral Solution of a Non-Isotropic Coulomb-Like Potential; Phys.Lett. A 165
(1992), 185.
[48] C.Grosche: Path Integral Solution of Two Potentials Related to the SO(2,1) Dynamical Algebra;
Trieste preprint, SISSA/186/ 91/FM, December 1991.
[49] C.Grosche and F.Steiner: Path Integrals on Curved Manifolds; Zeitschr.Phys. C 36 (1987) 699.
[50] C.Grosche and F.Steiner: Feynman Path Integrals; to appear in Lecture Notes in Physics.

90

Bibliography

[51] C.Grosche and F.Steiner: A Table of Feynman Path Integrals; to appear in Springer Tracts in
Modern Physics.
[52] C.C.Grosjean: A General Formula for the Calculation of Gaussian Path-Integrals in Two and
Three Euclidean Dimensions; J.Comput.Appl.Math. 23 (1988) 199.
[53] C.C.Grosjean and M.J.Goovaerts: The Analytical Evaluation of One-Dimensional Gaussian
Path-Integrals; J.Comput.Appl. Math. 21 (1988) 311.
[54] A.C.Hirshfeld: Canonical and Covariant Path Integrals; Phys.Lett. A 67 (1978) 5.
[55] R.Ho and A.Inomata: Exact Path Integral Treatment of the Hydrogen Atom; Phys.Rev.Lett. 48
(1982) 231.
[56] L.Infeld and T.E.Hull: The Factorization Method; Rev.Mod.Phys. 23 (1951), 21.
[57] A.Inomata: Exact Path-Integration for the Two Dimensional Coulomb Problem; Phys.Lett. A
87 (1982) 387.
[58] A.Inomata: Alternative Exact-Path-Integral-Treatment of the Hydrogen Atom; Phys.Lett. A
101 (1984) 253.
[59] A.Inomata: Remarks on the Time Transformation Technique for Path Integration; in Bielefeld
Encounters in Physics and Mathematics VII; Path Integrals from meV to MeV, 1985, p.433;
Eds.: M.C.Gutzwiller et al. (World Scientific, Singapore, 1986);
Recent Developments of Techniques for Solving Nontrivial Path Integrals; in Path Summation:
Achievements and Goals, Trieste, 1987, p.114; Eds.:S.Lundquist et al. (World Scientific,
Singapore, 1986);
Time Transformation Techniques in Path Integration; in Path Integrals from meV to MeV,
p.112; Eds.: V.Sa-yakanit et al. (World Scientific, Singapore, 1986);
[60] A.Inomata and R.Wilson: Path Integral Realization of a Dynamical Group; Lecture Notes in
Physics 261, p.42 (Springer-Verlag, Berlin, 1985); Factorization-Algebraization-Path Integration
and Dynamical Groups; in Symmetries in Science II, eds.: B.Gruber and R.Lenczewski, p.255
(Plenum Press, New York, 1986).
[61] G.Junker and M.B
ohm: The SU(1, 1) Propagator as a Path Integral Over Noncompact Groups;
Phys.Lett. A 117 (1986) 375.
[62] H.Kleinert: How to do the Time Sliced Path Integral for the H Atom; Phys.Lett. A 120 (1987)
361.
[63] H.Kleinert: Quantum Mechanics and Path Integral in Spaces with Curvature and Torsion;
Mod.Phys.Lett. A 4 (1990) 2329.
[64] H.Kleinert: Path Integrals in Quantum Mechanics, Statistics and Polymer Physics (World
Scientific, Singapore, 1990).
[65] H.Kleinert and I.Mustapic: Summing the Spectral Representations of P
oschl-Teller and RosenMorse Fixed-Energy Amplitudes; J.Math.Phys. 33 (1992) 643.
[66] P.Kustaanheimo and E.Stiefel: Perturbation Theory of Kepler Motion Based on Spinor Regularization; J.Rein.Angew.Math. 218 (1965), 204.
[67] W.Langguth and A.Inomata: Remarks on the Hamiltonian Path Integral in Polar Coordinates;
J.Math.Phys. 20 (1979) 499.
[68] F.Langouche, D.Roekaerts and E.Tirapegui: Functional Integration and Semiclassical Expansion
(Reidel, Dordrecht, 1982).
[69] T.D.Lee: Particle Physics and Introduction to Field Theory (Harwood Academic Publishers,
Chur, 1981).
[70] D.W.McLaughlin and L.S.Schulman: Path Integrals in Curved Spaces; J.Math.Phys. 12 (1971)
2520.
[71] A.A.Makarov, J.A.Smorodinsky, Kh.Valiev and P.Winternitz: A Systematic Search for Nonrelativistic Systems With Dynamical Symmetries; Nuovo Cimento A 52 (1967), 1061.
[73] M.S.Marinov: Path Integrals in Quantum Theory: An Outlook of Basic Concepts; Phys.Rep. 60
(1980) 1.
[72] M.M.Mizrahi: The Weyl Correspondence and Path Integrals; J.Math.Phys. 16 (1975) 2201.
[74] M.M.Mizrahi: On the Semiclassical Expansion in Quantum Mechanics for Arbitrary Hamiltonians; J.Math.Phys. 18 (1977) 786.

91

Bibliography

[75] M.M.Mizrahi: Phase Space Path Integrals, Without Limiting Procedure; J.Math.Phys. 19 (1978)
298.
[76] C.Morette: On the Definition and Approximation of Feynmans Path Integrals; Phys.Rev. 81
(1951) 848.
[77] C.Morette-DeWitt: Feynmans Path Integral: Definition Without Limiting Procedure; Commun.Math.Phys. 28 (1972) 47.
[78] C.Morette-DeWitt: The Semiclassical Expansion; Ann.Phys.(N.Y.) 97 (1976) 367.
[79] C.Morette-DeWitt and K.D.Elworthy: New Stochastic Methods in Physics; Phys.Rep. 77 (1981)
122.
[80] E.Nelson: Feynman Integrals and the Schr
odinger Equation; J.Math.Phys. 5 (1964) 332.
[81] M.Omote: Point Canonical Transformations and the Path Integral; Nucl.Phys. B 120 (1977)
325.
[82] N.K.Pak and I.S
okmen: General New-Time Formalism in the Path Integral; Phys.Rev. A 30
(1984) 1629.
[83] W.Pauli: Die allgemeinen Prinzipien der Wellenmechanik, in S.Fl
ugge: Handbuch der Physik,
Band V/1 (Springer-Verlag, Berlin, 1958).
[84] D.Peak and A.Inomata: Summation Over Feynman Histories in Polar Coordinates; J.Math.Phys.
10 (1969) 1422.
[85] M.Reed and B.Simon: Methods of Modern Mathematical Physics, Vol.II (Academic Press, New
York, 1975).
[86] L.S.Schulman: Techniques and Applications of Path Integration (John Wiley & Sons, New York,
1981).
[87] B.Simon: Functional Integration and Quantum Physics (Academic Press, New York, 1979).
[88] I.S
okmen: Exact Path-Integral Solution of the Ring-Shaped Potential; Phys.Lett. A 115 (1986)
249.
[89] I.S
okmen: Exact Path-Integral Solution for a Charged Particle in a Coulomb Plus AharonovBohm Potential; Phys.Lett. A 132 (1988) 65.
[90] F.Steiner: Space-Time Transformations in Radial Path Integrals; Phys.Lett. A 106 (1984) 356.
[91] F.Steiner: Exact Path Integral Treatment of the Hydrogen Atom; Phys.Lett. A 106 (1984) 363.
[92] F.Steiner: Path Integrals in Polar Coordinates From eV to GeV; in Bielefeld Encounters
in Physics and Mathematics VII; Path Integrals From meV to MeV, 1985, p.335; Eds.:
M.C.Gutzwiller et al.(World Scientific, Singapore, 1986).
[93] H.F.Trotter: On the Product of Semi-Groups of Operators; Proc.Amer.Math.Soc. 10 (1959) 545.

92