ANNALS OF PHYSICS 103, 198-207 (1977)

A New Approach to Gaussian Path Integrals and the

Evaluation of the Semiclassical Propagator

S. LEVIT AND U. SMILANSKY

Department of Nuclear Physics, Weizmann Institute of Science, Rehovot, Israel

Received February 26, 1976

The expansion of path variations in terms of solutions of Morse’s boundary problem is

applied in order to evaluate Gaussian path integrals. Together with a recently discovered
theorem on infinite products of eigenvalues of Sturm-Liouville type operators this yields
an expression for the most general semiclassical propagator. The properties of the latter are
investigated in the light of the Morse theory. The general methods developed here are
illustrated by the example of a charged particle moving in a homogeneous magnetic field.

1. INTRODUCTION

A new type of uniform semiclassical approximation (USCA) has been developed

and applied in recent years to molecular scattering [l-3] and nuclear physics [4-61.
This approximation is based on the semiclassical limit of the S matrix, which can be
derived from Feynman’s path integral approach [7] to quantum mechanics. The
starting point for this approach is the quantum mechanical propagator, expressed
through path integrals of various forms [7-lo]. The propagator is written symbolically
as
K = e(ijn)r[pathlD[path], (1.1)
s

where K stands for the propagator, T is the classical action functional, and D [path] is
the path differential. The explicit form of T and the meaning of the path differential
depend on the specific form and representation used [7, 91. The classical paths satisfy
the variational principle
ST=O. (1.2)
One may expand formally

T = TC’ + (1/2!)(S2T)C1 + (1/3!)(PT)cl + ... , (1.3)

where (S”T)cl stands for the nth variation of the functional T from the classical path
The substitution of (1.3) into (1.1) gives

K = e(i/ll)Tcl e~i/i)l(l/2)(~*T)C1+~~~lD[path] = e(i/fi)T" .R, (1.4)

s
198
Copyright 0 1977 by Academic Press, Inc.
All rights of reproduction in any form reserved. ISSN 0003-4916
 THE SEMICLASSICAL PROPAGATOR 199

where R is the reduced path integral. It is calculated by integrating over the space of
path variations from the classical path. In the semiclassical limit the stationary
phase approximation to the integral K is used. This amounts to retaining only the first
two terms in (1.3), and one faces, therefore, the problem of evaluating the Gaussian
path integral
6 = - e(i:2rr)(s’T)C’D[path].
(1.5)
J

This problem has been discussed in the general framework of the path integral
evaluation via the limiting procedure, i.e., as a limit of multidimensional integral
[7,9, 11, 121.
Another technique for performing the same integral was proposed by Davison [13],
in which the paths are expanded in terms of a complete set of orthogonal functions.
This idea has been applied in practice to a few simple cases, where the Fourier expan-
sion has been used [14, 151. Tt leads to tedious calculations even for simple applica-
tions [16] and becomes practically intractable for more general cases.
In the present work we propose a more convenient choice of a complete orthogonal
basis for the expansion of the path variations. The basis consists of the eigenfunctions
of Morse’s boundary problem [17]. As will be shown below, this is a natural choice,
which is specific to the system and leads to great simplification in calculating the path
integrals. At this point, the propagator is expressed in terms of the infinite product of
eigenvalues of Morse’s boundary problem [17]. Finally, one applies the results of a
recently discovered theorem [18] through which the above mentioned infinite product
can be written in a closed form.
The powerful tools of the Morse theory are used to present the phase of the propa-
gator in terms of the number of focal points on the classical trajectory.
The traditional Lagrange form [7] of the path integral representation of the
propagator is treated in Section 2. An application to the problem of a charged particle
moving in a magnetic field is given in Section 3. In this case the exact propagator is
already of the Gaussian type and no further approximation is necessary.

2. THE SEMICLASSICAL PROPAGATOR

A general physical systemwith m degreesof freedom is considered. The path integral

expression for the propagator [7] is

K(q”, f’; q’, f’) = s ,(i:n)rrs(t’l~[q(t)], (2.1)

where q = {ql ,..., q,J are the generalized coordinates of the system,

TMt)l = j-;” Uq, 4 t) df. (2.2)

200 LEVIT AND SMILANSKY

The integration in (2.1) is performed over all possible paths satisfying

4(f) = q’ q(C) = q”, (2.3)

and D[q(t)] is the “path differential” [7, 121.
Throughout this work it will be assumedthat there exist at least one classicalpath
which satisfiesthe boundary conditions (2.3). If there are more than one classicalpath,
it will be assumedthat they are far enough apart from each other to allow the separate
treatment of the contribution of each path to (2.1).
One introduces the path variation from the classicalpath

7Jt> = 4,(t) - CI$f) (i = 1,..-, m> (2.4)

so that
Q’) = TjJt”) = 0 (i = I,..., m). (2.5)
The secondvariation of the action functional T is then readily found:

The summation over dummy indices is implied in this work and the superscript ( )cl
indicates evaluation along the classicalpath. It is seenthat the second variation (2.6)
is itself a functional with respect to the path varitions q(t).
The so-called secondary Lagrangian is now defined [17]

252(7)~
7j, t> = Rij(t) rli77j+ 2Qij(t) Tiqj + Pij(r> -ii?ii 9 (2.7)
where the following notations are used

h(t) = (e)” (& = Rj,), (2.8a)

Qi&> = (g&J (2.8b)

(Pij = Pj,). (2.8~)

The matrix P(t) is the masstensor of the system. We limit it to be positive definite so
that det P(t) > 0 for t’ < t < t”.
The second-variation functional (2.6) is written with the aid of (2.7) as

(my = ~[~(t)] = j-l” 252(q,7j, t) dt, (2.9)

and the reduced propagator (1.5) is

Z?(r”, t’; qcl(t)) = 1 eci121i)l[nct)lD[rl(t)l. (2.10)

 THE SEMICLASSICAL PROPAGATOR 201

The functional Z[r](t)] can be cast into a more convenient form by performing the
time integration by parts. One obtains

with

fl,j = -(d/df)[~~j(d/dj) + Qjzl + [Qdd/dt) + &I (i,j= l)..., m). (2.12)

The variations q(t) are continuous m-component vector functions on the interval
[t’, t”], satisfying the conditions (2.5). The introduction of a scalar product

(7, 7’) == [yp %(j> Q’(f) dt (2.13)

makes the path space of the integral (2.10) into a Hilbert space which is an external
product of two spaces: m-dimensional configuration space of the physical system, and
a Hilbert space of the continuous scalar functions on [t’, t”] vanishing at t’ and t”.
Thus the functional Z[q(t)] can be expressedas

@l(t)1 = (‘1,497 (2.14)

where fl was defined in (2.12).

By choosing some orthonormal basis in this Hilbert space, say {zzor(t)}, one may
expand any variation as
q(t) = c aw(r). (2.15)

The functional Z[r(t)] is then written as

MO1 = 2 4w3anao, (2.16)

*,6
where
A,, = (u”, ALP). (2.17)

A basis {u”(t)} which diagonalizes the matrix {A,,} is given by the characteristic
solutions of Morse’s boundary problem [17]

-A,,23%
.+ XUi= L aQ(u,4 j) aqu, f-2,
t) + hzli = 0, (2.18a)
dt azi, - aui
u<(t’) = u,(t”) = 0 (i = I,..., m), (2.18b)

where the function Q(u, zi, t) is the secondary Lagrangian, introduced in (2.7).
In order to evaluate the path integral (2. lo), one uses(2.15) with {u”(t)} as solutions
of (2. IS) to obtain

(2.19)
202 LEVIT AND SMILANSKY

where {ha} are the eigenvalues of (2.18). They are all real, since the operator A is self-
conjugate in the space of path variations [17, p. 811.
In (2.19) J denotes the Jacobian which connects the “elements of integration”

D[y(t)] = J n da,‘. (2.20)

The transformation (2.15) is linear, hence J is independent of 7(f) and may be taken
out of the integration (2.19). Moreover, if one uses, instead of {p(t)}, some other
orthonormal basis, say {z?(t)), the Jacobian J would be unchanged due to the uni-
tarity of the transformation {u “) -+ {P}. J is then some normalization constant for the
path integral (2.10) [7, 131.
The evaluation of (2.19) is now trivial and one gets

@t”, t’) = J n (2ix/qh”)‘~“. (2.21)

Thus, apart from the normalization constant, the most general Gaussian path integral
is completely defined by the classical action [7], and the eigenvalues of the operator
A (2.12).
The right-hand side of (2.21) can be expressed in a closed form. For this purpose it
is useful to discuss the absolute value and the phase of the reduced propagator&t”, t’)
separately.

(a) Detetwzination of 1l?(t”, t’)i

By making use of the previously discussed invariant properties of the Jacobian J
one may transform (2.21)

lqt”, t’) = Rfm@“,

t’) (JJX&ee/flxy, (2.22)
1 a
where Krree and {A,“,,,: are the reduced propagator and the corresponding eigenvalues
of the physical system with the unperturbed Lagrangian

LflY?C? = $P<j(q, t) 4i4j a (2.23)

For convenience we call this system “free.” The operator A for this system is

A:?’ = --(d/Lit) Pij(t)(d/dt), Pii = P&“‘(t), t). (2.24)

Consider now the following initial value problem

&p = 0 (2.25a)
?p(t’) = 0, ?ji(k.)(f) = 2&’ (k = I,..., I??), (2.25b)

and the corresponding problem for Afree

Ayy = 0, (2.26a)
;jj”‘(t /) = 0, 7gqty = aj, (k = I,..., m). (2.26b)
 THE SEMICLASSICAL PROPAGATOR 203

It has been shown [ 181 that the ratio of the infinite products in (2.22) is

(2.27)

The first equality on the left-hand side of (2.27) follows because k + Xg,, for large
&‘S.
Before proceeding to obtain a final expression for 1K(t”, t’)l we should examine the
physical meaning of the functions 7J”‘(t) defined in (2.25) and show their relation to
the properties of the classical path qC’(t).
As was discussed above, the classical path qCl(t) is defined by the boundary condi-
tions (2.3) at t = t’ and at t = t” contrary to the common practice of specifying the
initial conditions at t = t’. This feature is typical of the semiclassical theory and is
discussed at length by Miller [f 91. However, once the classical path is found, the
initial velocities are specified and

4% q’, 4”) = P@, 4’3 4’(s’, 4% (2.28)

where 4’ = gcl(t’, y’, 4”).
Consider the m-parameter family of classical paths qcl(t, q’, p), p = (pL1,..., p,), all
passing through q’ at t = t’, which contains the path qcl(t, q’, 4’) for p = 4’. One can
study the way by which the classical path changes due to an infinitesimal change in the
initial velocities. A set of “response functions” {aqF’(t, q’, ~)/a~j},=~, is then obatined
and it can be easily shown that they are a solution Eqs. (2.25a) with the initial condi-
tions (2.25b), so that [17, pp. 9-1 l]

?jf’(t) = [2qc1(t, 4’2 ~)/2/Lj]lu=d’ z [iiqF’(t)/2~j(t’)]. (2.29)

The sameholds for the “free” system:

fji(j)(l) = [ap~‘(t)/a~j(r’)] (2.30)

are the solutions of (2.26).
Equations (2.25a) are known as Jacobi equations in variational calculus or as
Poincare equations in stability theory.
Using the relations (2.29), (2.30) (2.27) and (2.22) the absolute value of the reduced
propagator may be written as
j l7 / = I l&r,, j 1&K&f”, t’)/o(t”, 1’)!‘/2, (2.3 1)
where
d(t, t’) = det [ai: (t)/apj(r’)], (2.32)
dfree(r, t’) =: det [&$(r)/Z@j(t’)]. (2.33)

The “free” propagator k,,, may be found by the limiting procedure [7, 121

(2.34)
204 LEVIT AND SMILANSKY

Hence

1k(t”, t’)l = 1& In”’ 1 det B /-“‘, (2.35)

,
where the relations

p:‘(t) = [a.(qC1, cj”‘, t)/a& (2.36a)

and
d(t”, t’) = det [~&t”)/~pj(t’)] . det P(P) (2.36b)
have been used.
The result (2.35) expresses the absolute value of the reduced propagator in terms
of the Van Vleck determinant [20]

det[@Sct(q”, q’)/aq; i?qi’], (2.37)

and it guarantees the unitarity of the semiclassical propagator [l].

(b) The phase Q-I@“, t’)

Let us examine the right-hand side of the expression (2.22). The requirement that
the matrix (Pij(t)) is positive definite for all t’ < t < t”, implies that all A&, > 0.
The explicit form of&&t”, t’) may be easily found from (2.34)

k&(t”, t’) = (&)“” (det $” Pij’ df))-l”, (2.38)

and the determinant which appears in (2.38) is seen to be positive. Hence appart from
the trivial factor (27A-“i2, the phase of the reduced propagator&t”, t’) is given from
(2.22) by (-~r/2), where Y is the number of negative eigenvalues of the operator A
(2.12) counted with their multiplicity. They are found by the solution of the boundary
problem (2.18), which has been extensively studied by Morse [17]. In analogy with
the theory of quadratic forms in finite dimensional spaces, v is called the index of the
classical path qcl(t, q’, q”) on which the functional I[?;r(t)] is defined.
The useful concept of focal point is then introduced. The point qCl(T) (T # t’) on the
classical path is a focal point relative to q’ = qCl(t’), if Eqs. (2.25a) have a nonzero
solution which vanishes at t = t’ and t = T. The number of such linearly independent
solutions defines the multiplicity of the focal point. Since the general solution of
(2.25a), which vanishes at t = t’, is a linear combination of the solutions qch)(t)
(k = l,..., m) of the initial value problem (2.25), it is obvious that qcl(T) is a focal
point if the determinant (2.32) vanishes at t = 7. The rank of the matrix {aq~‘(t)/a&‘}
reduces at the focal point by the number, which is equal to the multiplicity of the point.
Geometrically, a focal point is a contact point of the given classical path with the
envelope of the m-parameter family of paths, all passing through q’ at t = t’ and
containing the path qCl(t, q’, q”) [21, p. 871. The envelope surface is often referred to
as the focal surface or the caustic.
The main result of the Morse theory is that the index v of the classical path is finite
 THE SEMICLASSICAL PROPAGATOR 205

and is equal to the count of focal points on this path strictly between its end points,
provided that each focal point is counted with its multiplicity.
This result is important when the contributions from several classical paths are
considered and their relative phases affect the propagator through the interference
phenomenon.
In practice, in order to calculate the reduced propagator k(t”, t’) one has to solve
Eqs. (2.25). By following the time development of the determinant d(t, t’) (2.32) one
obtains the phaseof the propagator by counting the number of zeros along the path,
whereasthe absolute value of the propagator is given by d(t”, t’).
When one of the eigenvalues of (2.18) vanishes, the end point q” is itself a focal
point. This corresponds to the caseof several coalescingclassicalpaths, which connect
the given end points. The number of such paths is equal to the multiplicity of the zero
eigenvalue, which clearly cannot exceed m, the number of degreesof freedom.
The infinite value of the semiclassicalpropagator in the case of a vanishing eigen-
value indicates that the quadratic approximation to the action is insufficient in the
“direction” p(t) which corresponds to the vanishing k. Higher-order terms should
be included and the so-called “uniform approximation” should replace the primitive
quadratic approximation [1,4, 5, 22, 231.

3. APPLICATIONTOACHARGEDPARTICLEINACONSTANTMAGNETICFIELD

The path integral form of the propagator describing a charged particle moving in a
constant uniform magnetic field has been previously treated using either Davies’
prescription [16] or the limiting procedure [lo]. The method developed in the previous
section allows for an elegant treatment of this problem.
The motion in the direction parallel to the field H (chosen as the z-axis) is free and
is not coupled to the other degreesof freedom. The problem is thus reduced to the
two-dimensional problem for the motion from the point (x’, y’; t’ = 0) to (x”, y”;
t” = T). The Lagrangian is
L = &rn [(zP + j+) + 2w (xj - yi)], (3.1)
where m is the mass, and 4 eH/mc = w is the Larmor frequency of the charged
particle. The boundary problem (2.18) is now written as

mfjl - 2mw?j, + XQ = 0,
(3.2a)
mfj2 + 2mw?j, + hrj2 = 0

?m = rlz@) = rim = rlzm = 0. (3.2b)

Note that in Eq. (3.2) there is no reference to a specific classical trajectory. This
follows becauseL is a quadratic function in the coordinates and the velocities.
One can readily find the eigenvalues of (3.2),
h(n) = m(7r2rz2/T2)[1- (~T/TQz)~] (n = 1, 2,...). (3.3)
206 LEVIT AND SMILANSKY

Using the result (2.22) and observing that {hen)}are doubly degenerate, one obtains
for the reduced propagator

(3.4)

where the relations

A&e = m(7T2n2/T2) (n = 1, 2,...),

(3.5)
Rfree(T, 0) = (m/27ritv)
have been used.
Finally with the aid of the Euler formula (sin x/x) = fi (1 - x~/?T%~)one gets
n=1
K(T, 0) = (m/2~rihT)(w T/sin 0T). (3.6)
One can easily derive the sameresult by solving the initial value problem (2.25) and
evaluating the determinant (2.32).
Consider time intervals T < n/w. The classical path connecting the points (x’, v’)
and (x”, JI”) is an arc spanning 2wT radians. All other classical trajectories passing
through (x’, v’) are arcs spanning the sameangle at time T, but having different radii
which depend on the initial velocity at (x’, y’). As long as T < n/w this family of
paths does not have an envelope, and the index v of the classical trajectory is zero.
Once T > n/w the arcs grow into circles, all passing the point (x’, v’) several times.
Thus the caustic of the family of paths for this system is degenerateto a point (x’, y’).
The index of the classicaltrajectory can be derived either from the geometrical picture
or by studying (3.3) yielding the result that v is twice the integer value of UT/r.

ACKNOWLEDGMENT

We would like to thank Mr. D. Melnik for valuable discussions and comments.

REFERENCES

1. W. H. MILLER, J. Chem. Phys. 53 (1970), 1949, 3578.

2. R. A. MARCUS, J. Chem. Phys. 54 (1971), 3965.
3. P. PECHUKAS, Phys. Feu. 181 (1969), 166, 174.
4. S. LEVIT, U. SMILANSKY, AND D. PELTE, Phys. Lett. B53 (1974). 39.
5. H. MASSMANN AND J. 0. RASMUSSEN, NW/. Phys. A 243 (1975), 155.
6. T. KOELING AND R. A. MALFLIET, Phys. Rep. C22 (1975), 181.
7. R. FEYNMAN AND A. R. HIBBS, “Quantum Mechanics and Path Integrals,” McGraw-Hill,
New York, 1965.
8. C. GARROD, Rev. Mod. Phys. 38 (1966), 483.
9. M. C. GUTZWILLER, J. Math. Phys. 8 (1967), 1979.
10. D. PEAK AND A. INOMATA, J. Math. Phys. 10 (1969), 1422.
11. I. M. GEL’FAND AND A. M. YAGLOM, J. Math. Phys. 1 (1960), 48.
12. G. J. POPADOPOULOS, Phys. Rev. D 11 (1975), 2870.
 THE SEMICLASSICAL PROPAGATOR 207

13. B. DAVISON, Proc. Roy. Sot. Ser. A 225 (1954), 252.

14. W. K. BURTON AND A. H. DE BORDE, N~rovo Cimento 2 (1955), 197.
15. H. DAVIES, Proc. Cambridge Philos. Sot. 53 (1957), 199.
16. M. L. GLASSER, Phys. Rev. B 133 (1964), 831.
17. M. MORSE, “The Calculus of Variations in the Large,” Amer. Math. Sot., Providence, R. I.,
1934.
18. S. LEVIT AND U. SMILANSKY, Proc. Amer. Math. Sot., to appear.
19. W. H. MILLER, Classical limit quantum mechanics and the theory of molecular collisions, Adu.
Chew. Phys. 25 (1974), 69.
20. J. M. VAN VLECK, Proc. Nat. Acad. Sci. U.S.A. 14 (1928), 178.
21. G. A. BLISS, “Lectures on the Calculus of Variations,” Univ. Chicago Press, Chicago, 1946.
22. J. R. STINE AND R. A. MARCUS, J. Chem. Phys. 59 (1973), 5145.
23. J. N. L. CONNOR, Chem. Phys. Lett. 25 (1974), 611.

.59+=3/r-14