You are on page 1of 18

ANNALS

OF PHYSICS

1%

123-140 (1990)

Wigners
N.L.
Institute .for

Function
BALAZS*

and Tunneling
AND A. VOROS+

Theoretical Physics, Unioersily of California, Santa Barbara, California 93106 Received September 13, 1989

We construct Wigners functions of a system with the potential-iw2q2 to exhibit and resolve some paradoxical features present in this description. Among others, we show that tunneling at negative energies arises through real trajectories associated with positive energies, in contradistinction of the usual WKB picture where this effect comes about by using complex trajectories of the correct energy. We resolve this puzzle by showing (a) that the initial data of this quanta1 description already contain wrong energy regions, (b) the interference phenomenon of WKB wave functions which provides the main contribution to Wigners function in the semiclassical domain becomes strongly non-local in the presence of a separatrix, which not only allows, but necessitates the presence of these wrong classical trajectories in the present description. c: 1990 Academic Press, Inc.

Special examples often illuminate general problems. In this note we will study the quantum mechanical scattering on a concave parabolic potential, often called the inverted harmonic oscillator. This problem has a long history, and has been completely solved, as far as the wave mechanical problem is concerned [ 11. However, we will concentrate on studying the problem from the point of view of Wigners function. We do this to augment our understanding how the quanta1 phenomena of tunneling and scattering are described within this formalism. As we shall see, this description yields some unexpected, and seemingly paradoxical features due in part to (a) the scattering condition expressed in this formalism; (b) the role of Plancks constant in the selection of initial data for Wigners function, leading to a utilization of real orbits in tunneling; (c) the presence of separatrices in the classical description. Since one can obtain an exact analytical expression for Wigners function (Eq. (lo)), these points can be explicitly studied.
I. THE SETTING OF THE PROBLEM

Let the classical Hamiltonian be H = P2/2rn - mo*Q*/2, where w is not the frequency, but the Liapunov exponent of the classical motion. We rescale this classical
* Permanent address: 11794. Permanent address: CEDEX, France-Member Department of Physics, State University Theorique, of New York, Saclay. Stony Brook, New York

Service de Physique of CNRS.

CEA-CEN

F-91 191 Gif-sur-Yvette

123
0003-4916/90 $7.50
Copyright ci 1990 by Academic Press. Inc. All rights of reproduction m any form reserved

124 Hamiltonian to quantum q = Q/e,

BALAZS

AND

VOROS

so as to make it dimensionless, in a manner immediately adaptable theory, by introducing the dimensionless variables p = P/+, and h = H/iiw. This yields
h = (pq2)/2. (1)

The classical solution curves h = E are hyperbola branches in the p, q spaces as shown in Fig. 1. The diagonals are the separatrices, and are the phase space trajectories for E= 0, corresponding to paths which connect infinity with the top of the barrier. The E > 0, E< 0 families of hyperbola are disjoint, and arranged as shown in the figure. The negative E trajectories correspond to classical paths which rebound from the potential hill, without being able to pass through it; the positive E solutions correspond to classical paths which go over the barrier without being reflected by it. The arrows on the solution curves show the direction of motion on a phase trajectory. The scattering convention usually introduced requires that for q > 0 only transmitted particles should exist, thus no solution curve reaching q + +oo should be used which has an arrow pointing to the left. In other words, all solution curves below the p = q diagonal should be omitted. In order to incorporate conveniently this condition into the quantum description we introduce classically a coordinatization of the phase space in which this diagonal is a coordinate curve. Put

u=(P-4)/J% u=(P+4)/$.
Then the Hamiltonian is
h = uv.

(2)

(3)

In this parametrization the scattering condition implies that no solution should be utilized for which u is negative. If we now quantize the problem u and v become operators. Since the commutator of ti and 6 is equal to the commutator of 8 and 4, we can consider u a momentum

FIG.

1.

Classical

phase plane

with

representative

orbits.

WIGNERS

FUNCTION

AND

TUNNELING

125 we consider the operator (4)


(5)

and u a coordinate. association

Hence choosing a p representation

zi+ u,

13 + id/du

fi -+ i(zZ + fiti) + (uid/du + i(d/du)u)/2,

this being the symmetrized dimensionless Hamiltonian operator, The eigenvalue equation for the eigenfunction f(u) reads as

with the singular solution


f= U-w+ 112) (u I=-01,

(7)

and a similar solution for negative u values. The two solutions are separated by the singularity. The scattering condition requires that we select (7) as the solution for u > 0, and zero otherwise, f =u-E+112)@(U). (8)

(In the rest of the article E will always denote the energy of the eigenstate, and h or E will refer to the energy parameter in Wigners function.) In Appendix A we show that this solution is indeed the correct one in the q representation. We would like to stress that the wave functions obtained using the u representation are as good as the ones using the q representation, since the 45 rotation described by (2) is a classical canonical transformation to which an exact unitary transformation of the wave function corresponds, with the transformation function (U 1q),
(u 1 q) = lcm/&4 aq1 12 eiS-yh

with S(u, q)/h = q2/2 + J&q + u2/2 using our dimensionless variables. (While this association is true in general only in the semiclassical limit, for linear canonical transformations it is an exact relation. Moreover, it is a consequence of the detinition of Wigners function that it transforms as a scalar under linear unimodular transformations [2], and thus Wigners function Wk(u, o) gives immediately WE(P, 4) = K(U(P? 4)? 4P, q)).)
II. hZZLES

Our aim is to construct Wigners function associated with a stationary quantum mechanical state of energy E satisfying the scattering condition. What are our expectations? Consider E<O. Figure 2a shows the classical situation. The solution curve

126

BALAZS

AND classical

VOROS

quanta1 expectations

FIG. 2. Top part: On the left, potential right, the associated phase portrait. Bottom transmitted beam of particles; on the right,

hill with incoming and reflected beam of particles; on the part: On the left, potential hill with incoming, reflected, and expected (but incorrect) Wigners function.

corresponding to the incoming and reflected particle is the left hand branch of the hyperbola h = E, the classical distribution function will be proportional to a Dirac delta function, 6(h - E). The right hand branch cannot be utilized since it violates the scattering condition. In the quanta1 case, using a semiclassical approximation, we expect something similar to the sketch shown in Fig. 2. On the left, 6(h - E) should be replaced by an Airy function. (The immediate vicinity of its peak is represented by the shaded ribbon.) Similarly, we might expect that the particles, after being transmitted by tunneling should be described by an Airy function exponentially damped in E; its peak is represented by the shading over the upper half of the hyperbola on the right. (The lower half of the hyperbola cannot be used because it describes particles coming from the right.) The two structures are presumably connected in some manner, depicting tunneling in this description. However, this picture must be wrong. Wigners function gives a full description of the quanta1 situation, including tunneling and scattering. At the same time it is a wellknown fact that for Hamiltonians at most quadratic in the variables the integral-differential equation describing the time evolution of Wigners function is identical to the classical Liouville equation, a first order partial differential equation. Its characteristics are the classical phase space trajectories, h = constant, and on each characteristic the solution has the same value, the initial value. Consequently, if Wigners function does not vanish on the p > 0 part of the right branch of the hyperbola, it cannot vanish on the p < 0 part either. Since the scattering condition requires this vanishing (for p < 0) the function must vanish on the whole

WIGNERS

FUNCTION

AND

TUNNELING

127

right hand branch obliterating (in the quanta1 case) the transmission altogether. Thus, the right hand branch of the E < 0 hyperbola cannot be used at all to construct Wigners function for the transmitted particles. How do they appear then in this picture? This is puzzle 1. The only way to circumvent this predicament is to use the h > 0 hyperbolae as well, even though the energy of our quanta1 state is negative. These, however, come into play only if Wigners function for E < 0 is compelled to spill over the h = 0 separatrix. Why should this spillover be compulsory? This is puzzle 2. Thus, puzzles 1 and 2 arise from the fact that the equivalence of Wigners integral differential equation with Liouvilles equation and the scattering condition imposed precludes tunneling, unless characteristics of the wrong energy must also be utilized in the time evolution of a quantum state. This fact leads to subsidiary questions. (1) How can transmitted particles be described by Wigners function in the classically allowed region if not by an Airy type approximation over the classically allowed phase space where h - E? (2) How can a negative energy eigenfunction give rise to a description utilizing classical trajectories with h positive? In particular: (a) will this not generate an inconsistency if we use negative energy, WKB wave functions to describe the state; (b) how can the particles transmitted by positive h trajectories still possess negative energies, as they must?

III. THE CONSTRUCTION OF WIGNERS FUNCTION

Given a wave function f(u) (U being a momentum function w(u, u) is obtained as w(u, 11)=(1/2x) j dsf(u+s/2)f*(u-s/2)ePi,

like coordinate),

Wigners (9)

with !I = 1. According to Eq. (8) f(u) = u-(/~)-~, u>O; _ f(u)=O, eigenstate with energy E, satisfying the scattering condition. w,(u, u)= (l/271) jy2,, ds(u+s/2)-12--iE = 0, Putting s = 2ux we immediately w,(u, u) = W,(h; u) =(lIn) = 0,
595/199/l-9

U-CO, in the energy Then u > 0, u < 0.

(a-s/2)--(12+iEe,

find

j,

dx(1+x)~/2~(1-x)-/2)+~e2i~~,

u>O 24< 0.

(lOa)

128 Putting

BALAZS

AND

VOROS

x = tanh y we obtain the alternative

form a>0 (lob)

w,(h; u)= (I/X) I_, dy(l/cosh y)ep2ihtahY+iEy,

(or its complex conjugate, W being real). Since the wave functions are not normalizable the integral of W over p and q diverges. We fixed the multiplicative factor by requiring that j W,(h; u) dh = 1, or W,(O) = l/cash nE. The real integral in Eq. (1Oa) can also be expressed as a complex contour integral which identifies it with a Laguerre function W,(h; u) = (cash nE)-l (l/274 $ dze4jhz(z+ 1/2))1/2)-iE W, (z - 1/2))(12)fiE (11)

= (cash TIE)- e2ihL- (,,*,-iE(-4ih)/r(1/2-

the last line being obtained through the integral representation of the Laguerre functions. (See Appendix B.) We note that the parameters u, u appear in the combination UZ) = h, as expected. It is less expected that the expression also depends explicitly on the sign of U, incorporating the scattering condition. In the present case the energy constant surface consists of several disjoint pieces, and the signs of u and u are additional constants of the motion labeling the pieces. In the future we will omit the u label in W,(h; u) and agree that it describes Wigners function for u positive (it being zero otherwise). With Eq. (11) our task is in principle solved. However, the Laguerre functions for complex arguments and order are not tabulated and their asymptotic expansions are not studied. For this reason, in the next section we study this function directly.

IV. THE

ANALYSIS

OF WIGNERS

FUNCTION

Three energy parameters appear in the problem, the energy of the eigenstate &BE; the classical energy fiwh(p, q) associated with the parameters p and q appearing in Wigners function; and ho. They are present in the two ratios E and h. We now analyze W,(h). We note that if both E and h change sign nothing is altered in (lOa) (since w is real). Thus we can confine our interest to one sign of E. Let us take E < 0, the analysis of tunneling, and study W,(h) as a function of h, starting with h < E. According to Eq. (10) for a given E (and u positive), W depends on h only. Classically we would find 6(h - E), thus W would vanish everywhere save on the hyperbola branch h = E, intersecting the negative q axis if E is negative (or the positive p axis if E is positive). In the semiclassical region we expect this delta function to be broadened, thus we consider h values near E. Thus E and h have the same sign. This broadening can be easily understood if we evaluate (lob) by the method of

WIGNER'S FUNCTION AND TUNNELING

129

steepest descent. For large h and E and h near E the saddle points are at small values of y; hence we can expand the integrand in Eq. (lob) and find (l/n) Srn he m
WI

~ E).v--eihY3/3 =

(2/(2h)u3)

~i( _

2(h

~)/(2h)u3),

c (2/(2E)3) Ai( -2(h - E)/(2E)3)

(12)

the familiar Airy type approximation. As the different values of h sweep around E, W exhibits ripples on the convex side of the h = E curve, (h < E), and an exponential damping on the other side, in the region bounded by the h = E( ~0) curve, the separatrix p = -4, and the boundary p = q corresponding to the u = 0 curve which separates the W= 0 region from the rest. The Airy approximation ceases to exist outside this wedge. This can be understood easily, if we reinterpret the argument of the Airy function geometrically as follows [3]. Take a point yE(pE, qE) on the h = E curve and draw a straight line through this point along the afIine normal N, to the curve at this point. All points x : (p, q) on this normal curve are given by x(d) = yENEd, where d is a scalar parameter and measures in this sense the separation of the two points along the affine normal. Then the Airy function dependence at x(d) is simply Ai( -2,4/fi213), (where we restored all the physical dimensions of p and q). Consequently, only at those points x is this approximation possible which lies on an affine normal direction of some point of the h = E curve. In the present case N, at the point yE is along the radius vector yE (see Appendix C), thus one can only apply this approximation at those points x which lie on a straight half line connecting the h = E curve to the origin, i.e., at the points in the quadrant between the asymptotes of the h = E curve. Near the separatrix Jhl is small, and a different approximation is to be used, (Appendix C). This gives, for 0 > h B E 1 WE(h) = (cash nE) &,(4 m) a monotonic function of h. If h is small but positive and E large negative we find 1 WE(h) = (cash nE) Jcl(4 $-El) an oscillatory function of h. (A saddle point analysis indicates that the oscillatory behavior starts at h= -1/16E.) A general view of W,(h) is given in Fig. 3 for several different energies. V. THE RESOLUTION OF THE PUZZLES In Section II we raised several puzzles implied in the present problem. Wigners integral-different equation is equivalent to Liouvilles equation; hence all solutions of the initial value problem are obtained by displacing the initial data

130

BALAZS

AND

VOROS

0.6-

0.4-

0.2-

B.B-

-9.2-

-8.4-

FIG. 3. Graphs of Wigners function, tional to J,(h). For E >O the oscillations graph. )

W,(h), for different values of E. (For E=O, W,(h) is proporof W,(h) for h >O are no longer visible on the scale of the

WIGNERS

FUNCTION

AND

TUNNELING

131

c I.0-

8.8-

0.6-

0.44 0.2-

0.0-

-B.Z-

-0.61
-9.5

I
-7.5

,
-5.5

,
-3.5

,
-1.5

.I,
8.5

,
2.5

, h
4.5

0.8 1

-8.6 I
-13.0

-Il.0

I
-9.0

I
-7.0

I
-5.0

I
-3.0

I
-1.0

I I

I.
I.0

h
3.6

FIGURE

3-Continued

59WYY/I-IO

132

BALAZS

AND

VOROS

along the characteristics, the h = constant curves. However, the characteristics for which E < 0, and which originate at q = -co, p > 0 do not lead to the q > 0 region, hence, seemingly, no tunneling is possible! This puzzle is resolved by observing that a Wigner function for a stationary state of negative energy, E < 0 cannot be confined to the negative h(p, q) region allowed by the scattering conditions. This region is the wedge around the negative q axis bounded by the q = +p separatrices. The scattering condition absolutely confines W to the u > 0 region; the additional requirement of a vanishing W for h > 0 would imply the further condition W = 0 for u > 0. We now express these conditions on the wave function in the o and u representation. Accordingly we seek a wave function $(u) which vanishes for u > 0, and whose Fourier transform d(u) vanishes for u < 0. However, such a function cannot exist by Painlevts theorem [4]. Indeed, Ii/(u) = 1; du d(u)e, thus it is analytical in U. Hence, if it vanishes on part of the real axis it will vanish everywhere, contrary to the requirement that it be finite for negative u values. Consequently, the wave functions of physically realizable states must produce such associated Wigner functions which must spill across the u = 0 separatrix (since the spillover across the u = 0 separatrix is forbidden by the scattering condition). Thus, the correct choice of the initial form of a general W function is influenced by quanta1 conditions and will in general contain fi. (It must correspond to the projector of a physically realizable state for pure states; or it must correspond to a positive density operator in general.) Hence the transfer of the initial data corresponding to tunneling is accomplished by the real characteristics for which h(q, p) > 0 although the quantum state has E<O. It is a well known fact that Wigners function associated with an energy eigenstate is not sharp in the classical energy. (The average of powers of the classical Hamiltonian h, averaged over Wigners function, are not powers of the average of h.) In the usual case this leads to a trivial quanta1 broadening of the classical picture. If, however, a separatrix is present, this, usually innocuous broadening leads to unexpected results. Now the broadening must cover the separatrix, and, as time passes the broadening will slide along the separated trajectories to regions far apart. These are, then, the explanations to puzzles 1 and 2. We turn now to the subsidiary questions which are linked. (a) What is the reason that the transmitted particles in the classically allowed region are not described by an Airy type approximation to Wigners function? (b) How can the use of positive h trajectories be reconciled with the fact that the quantum state has negative energies? Consider the elegant interference argument which leads to the Airy approximation of Wigners function [S]. We insert in the defining expression the WKB wave functions and find that the oscillatory integrand has two stationary points which contribute significantly to the value of W,(p, q) in the classically allowed region.

WIGNERS

FUNCTION

AND

TUNNELING

133

These are found by a chord construction: the two stationary points s1, s2 contributing to W&J, q) are the endpoints of that chord of the h = E curve for which the p, q point is the bisector. At the same time the h = E curve is that curve which was utilized (a la Maslov) to construct the WKB wave function used. As the argument of W,, the (p, q) point, approaches the h = E curve the s,, s2 points conflue with the (p, q) points, leading to the Airy function behavior. (see Fig. 4.) In the present case the WKB wave functions utilized are constructed over the two branches of the $J -q2) = E < 0 hyperbola with the p negative segment of the right branch omitted (since there are no incoming particles from q = +co). Now a constructive interference can occur in two different ways. As before, a self-interference of one branch. For example, at the point (p, q) the standard chord construction results in the stationary points S, , s2 and the possibility of the s,, s2 confluence yields the Airy approximation. However, a point p, q can now be the bisector of a chord ending on two different branches of the h = E < 0 hyperbola in the points s;, s; which are again stationary points. Given E, these (p, q) points lie on a h = E > 0 hyperbola. Thus, the mutual interference of WKB associated waves with the different branches of the h = E < 0 hyperbola leads to an interference at points with h positive. These saddle points cannot coalesce with each other and with (p, q); thus they will not lead to the Airy form. We stress that the effect is due to the separatrix since it produced the two branches of the hyperbola, and it prohibited the confluence of the saddle points on the different branches. This also answers point(b). We see that the mutual interference from different branches will yield contributions where h has the opposite sign to E. At the same time the resulting Wigner function is oscillatory which allows the average value (h ) to be negative in regions where the h > 0 trajectories hold sway. The average energy per particle associated with the region for which h > 0 and p > 0 is given by
ce> =.f dpdqhW&) .f d 4 W,(h)

a q q \ \ >

FIG. 4. (a) Conventional chord construction where the h(p, q) = E curve has a single branch. The points sl, s2 are the locations of the saddle points in the integrand of kV; the point (p, q) is the bisector of the s, , s2 chord. (b) The two different chord constructions if the h(p, q) = E curve has two branches. The chord s,. s2 has (p, q) as its bisector; sir s2 lie on the same branch. The chord s; , 3; has (p, q) as its bisector; s;, s; lie on different branches.

134

BALAZS

AND

VOROS

integrated over the wedge shaped re ion between the top part of the two diagonals. Putting h = E, p = fi cash a, q = t 2s sinh a, this becomes
(e) = jam d& C-W&)/$ d& WE(&).

According to Eq. (lob) WE(s) satisfies a differential equation derivable from the one defining L,(z). This gives
f (E dW,/d&) +4(&-E) W,= 0;

integrating

this differential

equation from E= 0 to E= cc we immediately


(e) = E.

find that

Thus, for E < 0 the oscillations of WE(&) indeed yield a negative (E). These explanations resolve the original puzzle; nevertheless a certain unease remains, since the WKB description in a classically permitted region has the same form irrespective of whether this region is actually reached by trajectories in the classical limit or not. Yet this property does not carry over to the Wigner description, even though we use WKB wave functions. This comes about since the interference mechanism which forms the basis of the semiclassical evaluation of Wigners function yields profoundly different results depending on whether a separatrix is present or not. VI.
THE REFLECTION AND TRANSMISSION COEFFICIENTS

We compute now the reflection and transmission coefficients using Wigners function. At a point q the current density is given by j dpp W,(h), where the range of integration depends on the particular current we seek. Consider Fig. 5 and q = q < 0; the current on the left is defined as

the first term (which will be negative) is the current j, due to reflection. The second term is that part of the incoming current, ji;, for which h < 0; the last term, j& is the incoming current for which h > 0. The sum of the last two terms gives the incoming current jinc. On the right hand side q= q >O and the current is
cc Iright = I 4 pWdp=&,

where j, is the current in the transmitted beam. Current conservation requires that jleft = j+&f. This is satisfied because the second term of jleft is equal and opposite to

WIGNER'S FUNCTION AND TUNNELING

135

FIG. 5. Visualization
particles currents with positive, are proportional

of currents in scattering. The incoming currents j&, j,;, are produced by esp. negative energies; j, and j, are the transmitted and reflected currents. The to the number of streamlines intercepted by the 9 = 9 and the 9 = 9 lines.

the first, hence cancels the first, and the last term of, jleft, in turn, is equal to jright. These observations follow immediately from the fact that W is constant along the h = constant trajectories, and the same number of trajectories go through the corresponding p segments of integration. The transmission and reflection coefficients C, and CR are defined as
CT = j,(q + + cc )/i,c(q + -a 1= CR=-cl i,:;i,, =*+a

1 1 +!X
W,(h) dh thus dh.

with CI= lj,l/lj,l.


As q + - 00, Ij,l + j:. ,mW,(h) dh, j, + j;

a=jam W,(h) dh; j~ ,n W,(h)


A straightforward calculation in Appendix E yields a = ,+ZEn, and
2nE CT=

l+e

+2nE

the standard answer. VI CONCLUSIONS The present study shows that the presence of separatrices in the classical phase space alters our intuitive understanding of the picture conjured .up by Wigners

136

BALAZS

AND

VOROS

functions. The usually innocuous broadening of the classical distribution function by quantum effects will be sent by the separatrix into widely separated regions, leading to a spreading of Wigners function into regions which are associated with the wrong sign of the classical energy. This, then, allows tunneling to be produced by classical orbits of the wrong sign. Nevertheless, this does not conflict with the usual semiclassical explanation of Wigners function as being the result of an interference by WKB waves because the presence of the separatrix produces an interference which has strong nonlocal consequences in the (p, q) space. These results, then, suggest that while in the usual WKB description tunneling can be conceived as mediated classically by complex classical orbits, in the Wigner picture the real classical orbits can also play a significant role. The role of the complex classical orbits in the Wigner description requires further clarification.

APPENDIX

A: THE WAVE FUNCTION IN THE q REPRESENTATION of Abramowitz,


y+(ax*-a)y=O,

According to the notation

Stegun [6] we seek that solution of


(AlI

(their

Eq. (19.21.1)),

which has only a transmitted

wave for x$ Ial. This is

y = E(a, x)x

U(ia, xe- in4), (their Eq. (19.17.9)) where U(a, 2) N ez214 ers-s=/2Sns1
l/2 ds ,

(AZ)

with c1 being a counterclockwise contour with 0. Since the Schrodinger equation is

around a branch cut connecting

-co

d2$/$q2 + q2t+b + 2EIl/ = 0

(A3)

we must put a = -E and x = $q. Inserting this in (A2) and changing the integration we obtain
i(q2/2 + fiqu

variable s to u by s = ei3n4u,

+ ~12 +, - ( 112 ) - iE du

(A4)

where we have chosen the branch cut to lie on the negative real axis. Since the generating function of the canonical transformation (p, q) + (u, u) is q2/2 + ,,bqu + u*/2, the unitary transformation (q 1 u) is proportional to ei(q2/2 + J&u + U2/2), th us the wave function U- (1/2)-iE does indeed correspond to an eigenfunction with the asymptotic properties required by the scattering condition.

WIGNER'S FUNCTION AND TUNNELING APPENDIX

137

B: LAGLJERRE FUNCTIONS

The Laguerre polynomials

are defined as
L,(x) = e $ (xe-~-)

if they are normalized so that the coefficient of .Y be unity. Then by Cauchys theorem 1 111L,(x)e- = -!- 4
2ni

dy(ye-?)
(y-xy+

where the closed contour of integration surrounds the points y = 0, x. This definition remains valid if II ceases to be an integer, and thus the points y = 0, x become branch points. Putting y = ~(4 + Z) we obtain the symmetric expression

where the contour surrounds the branch points z = f 4 and the branch cut connecting them.

APPENDIX

C: THE AFFINE INTERPRETATION

OF W,(h)

Let x: (p, q) be the argument of W(p, q) and let yE(A) : (pE(;l), q,(l)) be any point on the curve H(p, q) - E = 0, where A is an arbitrary parameter along the curve. Then

defines the affine arclength along the H= E curve. The afline normal N(s), normal to the H = E curve at the point y(s) is given by
N(s) = d2y/ds2,

((3)

thus
x = y + A d 2y/ds2 (C3)

defines A. In the present case the curve is $(p - q2) = E, E < 0, and the relevant branch is
4E(P) = - &=z (C4)

138

BALAZSAND

VOROS

choosing p as the parameter. That is,


Y,(P) : (P, 4E(P)), dy,l@ : (4 p/q&

d2y,Mp2 : (0, - Wk:), ds/dp =

1
0
: (

Pl4E -2EJq;
1

13 = ( -2E)13
qE

NE

d2y,lds2

_ 2EJ2,3(P, qE) = ( _ &3

YET

and
x=y,+N,A=y, A
( -2q2:3 *

Squaring and subtracting each component

we find

or

from which
h-E n=(2E)Ll?

if

h near E.

Hence Ai( -24) = Ai( -2(h - E)/(2E)j3). The same as the Airy-factor in Eq. (12).

APPENDIX

D: ASYMPTOTES

OF W,(h) FOR SMALL h, LARGE E

According to (lob)
e+2ih

W(h) = ___

cash nE

Lp,,,,,wiE(-4ih)/f

(Dl)

Put c2jhW(h) into Laguerres differential differential equation


hW+

equation. Then W satisfies the following


W=O. P2)

W+4(h-E)

WIGNER'S FUNCTION AND TUNNELING

139

If h is small and E is large, neglect h next to E in the last term, giving (D3) The solution finite at h = 0 is J( 4i Jz) = Zo(4 Jz,,

giving W up to an E dependent factor. Since W,(O) can be easily calculated from the integral representation (or from the normalization of L,(z) at z = 0) we obtain (04)

APPENDIX

E: THE COMPUTATION

OF THE TRANSMISSION

COEFFICIENT

C, = LX/( 1 + a), with CI= A/B. Here

A= lam W,(h) & = i

j dy

e+2i,

2ni+sinh

y
ezfEJ. .

B= so W,(h) dh = &. j & -?c 4-A

The integrals are to be taken on the contours which avoid y = 0 as indicated, (to make the integration over h convergent at the limits + co, -co). Now
A=& B= - &

Principal Principal

value integral + k Residue( y = 0) value integral + t Residue(y = 0).

Hence
A + B = Residue( y = 0) = 1 A - B = k.

Principal

value integral
)d/r, dy =&

&B,l/m

ni

- dy (e 2iEt -e-2 X sinh v

tanh nE

140 whence

BALAZSAND

VOROS

A = enE/(enE + eCnE) B=e-E/(&=E+e-E) M= e2nE.

ACKNOWLEDGMENTS
The authors owe their thanks to many: the National Science Foundation for sponsoring their visit to the Institute for Theoretical Physics at Santa Barbara (NSF Grant No. PHY82-17853, supplemented by funds from NASA) and the Institute itself for their hospitality. NLB also acknowledges the support of (NSF Grant No. PHY88-16234) and the repeated and kind hospitality of the Service de Physique Theorique, CEN-Saclay (France), where much of the work has been initiated. Finally, NLB thanks Professor E. Wigner for an illuminating discussion on these matters some years ago.

REFERENCES
1. E. C. KEMBLE, 2. N. L. BALAZS 3. N. L. BALAZS, Phys.

Rev. 48
London,

(1935),

560;

L. D.

LANDAU

AND

E. M.

LIFSHITS,

Quantum

Mechanics,

p. 177, Pergamon,

1958.

AND B. K. JENNINGS, Phys. Physica A 102 (1980), 236.

Rep. 104 (1984),

347.

4. Oeuvres de Paul Painleve, Ed., CNRS 1974, Vol. II, p. 55; see R. F. STREATEZR AND A. S. WIGHTMAN, PCT, Spin & Statistics, and All That, p. 74, Benjamin, New York, 1964. 5. M. V. BERRY, Philos. Trans. R. Sot. London A 281 (1977), 237. 6. M. ABRAMOWITZ AND I. A. STEGUN, Handbook of Mathematical Functions, pp. 686720, U.S. National Bureau of Standards, Washington, D.C., 1964.

You might also like