Anda di halaman 1dari 30

Wave function confinement via transfer matrix methods

Jeffrey D. Olson and Jonathan Lee Mace Citation: Journal of Mathematical Physics 44, 1596 (2003); doi: 10.1063/1.1554763 View online: http://dx.doi.org/10.1063/1.1554763 View Table of Contents: http://scitation.aip.org/content/aip/journal/jmp/44/4?ver=pdfcov Published by the AIP Publishing

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

JOURNAL OF MATHEMATICAL PHYSICS

VOLUME 44, NUMBER 4

APRIL 2003

Wave function connement via transfer matrix methods


Jeffrey D. Olsona)
Department of Physics, University of Texas, Austin, Texas 78712

Jonathan Lee Maceb)


DX-2, Los Alamos National Laboratory, MS-C920, Los Alamos, New Mexico 87545

Received 21 August 2002; accepted 11 December 2002 The exact transfer matrix approach used in studying sectionally constant potentials in one dimension is generalized to cylindrical and spherical geometries, where the potential depends only on radius. In each geometry two transfer matrices sufce to completely describe the wave function: one for handling a discontinuity in potential and one for handling a delta-function potential barrier. This method is then applied to the problem of conning a wave function in a cylindrical conguration using only a series of carefully placed delta function potential barriers. It is found that connement can be made to increase nearly exponentially with the number of barriers if placed correctly, but that this arrangement has an exponentially sharp dependence on both barrier position and energy. 2003 American Institute of Physics. DOI: 10.1063/1.1554763

I. INTRODUCTION

It is a well known result in one-dimensional quantum mechanics that a lattice of evenly spaced potential barriers, such as delta functions, can give rise to bands of perfect transmission by particles with energies in allowable ranges, even when that energy is signicantly less than the potentials involved in the lattice. This is due to how the boundary conditions imposed on the particles wave function by the laws of quantum mechanics give rise to resonances. It seems like it should be possible to exploit this feature of wave mechanics in geometries other than that of planar lattices. For example, can an arrangement of cylindrically symmetric concentric barriers be found that amplies the probability for a particle of a particular energy to be found in the interior region of the barriers? The same question could be asked of an arrangement of spherically symmetric barriers. In seeking to answer this question we found that it was necessary to extend a particularly useful method used in studying one-dimensional planar quantum system to cylindrical and spherical geometries: namely, the method of transfer matrices. Transfer matrices have long been used in studying one-dimensional scattering problems not only in quantum mechanics,15 but in electrodynamics and optics as well. The primary advantage of using transfer matrices is the systematic manner in which wave functions are matched at potential boundaries. This allows for systematic calculation of scattering parameters such as transmittance and reectance. The ease with which the approach can be implemented on a computer allows calculations with large numbers of barriers that would be otherwise intractable. Walker and Gathrights article5 provides a thorough survey of the principals and methods involved. It turns out that the basic idea can be extended to cylindrical and spherical systems. This has recently been done for the electromagnetic problem6,7 while the corresponding problem in quantum mechanics has only been briey covered. One study8 develops the transfer matrices for the spherical case, while another9 makes some use of cylindrical transfer matrices in investigating a double barrier problem. Neither study presents the general transfer matrix for the planar/cylindrical/spherical problem, nor do either of them derive the transfer matrix for a delta function potential. Both
a

Electronic mail: jdolson@physics.utexas.edu Author to whom correspondence should be addressed. Electronic mail: jonathan@lanl.gov 1596 2003 American Institute of Physics

0022-2488/2003/44(4)/1596/29/$20.00

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1597

problems are addressed here. It is found that, independent of geometry, two matrices sufce to completely describe the propagation of the wave function: one for transferring over a discontinuity in the potential and one for transferring over a delta function barrier. In the general formulation it is no longer possible to separate out the propagation matrix from the other two matrices as is done for the planar case in Ref. 5. Each matrix is dependent on where the discontinuity or delta function is located. Hence we lose some of the power of planar transfer matrices. Nevertheless, the matrices developed are fully rigorous and applicable to any quantum system possessing the necessary symmetries. In the rst section of this article we present the derivation of the generalized transfer matrices and highlight some of their important properties. We then consider the question of connement in a cylindrically symmetric system composed of an array of delta function potentials. We make use of the transfer matrices developed to nd a conguration of barriers that gives near optimal connement. We also develop the analogous problem in a planar system for purposes of comparison. We then provide some numerical results for the connement problem obtained from our solution. Finally, in the appendixes we highlight some properties of the algebraic group that arises when studying generalized transfer matrices.
II. MATRIX DEVELOPMENT

dinger equation Consider the dimensionless version of the time independent Schro 2 r 2 e v r 0. Here the constant is given by 1

2 mV 0

a0 ,

where V 0 and a 0 are the units of energy and length, respectively. The quantities r, e, and v in 1 are given by taking their dimensional counterparts and dividing by a 0 or V 0 as appropriate. We phrase the problem in dimensionless terms in order to ease computer implementation for which the transfer matrix method is particularly well suited. In a region where v is constant, Eq. 1 is simply Helmholtzs equation with k e v :
2 k 2 r 0.

Using separation of variables10 gives the radial equations

d2 k 2 r 0, dr 2

n 1, n 2, n 3,

3a 3b 3c

d2 2 1 d 2 k r 0, dr 2 r dr r2

2 d 1 d2 r 0, k 2 dr 2 r dr r2

where and are the azimuthal quantum numbers and n indicates whether we are working in planar ( n 1), cylindrical ( n 2), or spherical ( n 3) coordinates. The general solution to 3 can be written as

r A r B r ,

where ( r ) and ( r ) are linearly independent solutions. Of the possible choices for ( r ) and ( r ) we will use forms that represent outgoing and incoming waves, respectively:

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1598

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

r e ikr ,
1 r H kr , 1 r h kr ,

r e ikr ,

n 1, n 2, n 3,

5a 5b 5c

2 r H kr , 2 r h kr ,

where H and h are the cylindrical and spherical Hankel functions. We make this choice simply because scattering problems are best phrased in terms of incoming and outgoing waves. However, the transfer matrices we will derive are independent of the choice made for ( r ) and ( r ).

A. Discontinuity

Suppose we have a discontinuity in the potential at r a . Say that it rises from v 1 to v 2 . We demand that the radial wave function and its rst derivative be continuous at r a :

1 a 2 a , 1 a 2 a .
These conditions can be written nicely in a matrix form:

2 A1 B1 2

A2 , B2

where everything is evaluated at r a . Note that the subscripts on the functions refer to the constant k. Thus i ( a ) has k k i e v i . To simplify we introduce the following notation: Mi a Thus 6 can be rewritten as M1 a 1 M2 a 2 . Solving for 2 , we get
1 2 M 2 a M1 a 1 .

i a i a

i a i a

Ai . Bi

This gives us the desired transfer matrix for going from region 1 to region 2:
1 n M 2 a M1 a

2 2 22 2
1

2 2 2 1 2 2 2 1

1 1

2 1 21

1 2 1 2 2 1 2 1

Again, everything is evaluated at r a . We can simplify the notation somewhat if we introduce a generalized Wronskian:

j . m i , j i j i

When i j , this generalized Wronskian is just the normal Wronskian. For the particular forms of ( r ) and ( r ) we have chosen, the normal Wronskian is given by

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1599

m i , i

2 ik i , 4i , a

n 1, n 2, n 3. 10

2i 2 , a ki

Note that none of these are ever zero since we deliberately choose ( r ) and ( r ) to be linearly independent solutions. With this notation we can rewrite 8 as n m 1 , 2 1 m 2 , 2 m 2 , 1

m 1 , 2 m 2 , 1

11

Any solution to 4 can be used in the above expression for the discontinuity transfer matrix. For reference, the explicit components of n for each n and for the ( r ) and ( r ) that we have chosen are listed in Appendix B.
B. Delta function barrier

Now suppose we have two shells separated by a delta function barrier at r a :


v r

ra , c nr n1

where the constant c n is given by

c n

0,
2

1, 2, 4,

n 1, n 2, n 3,

and is the strength of the barrier. We still demand that the radial wave function be continuous at the boundary, but the rst derivative will suffer a discontinuity due to the delta function. We can calculate the discontinuity by integrating 1 from r a to r a : 0cn where n is given by

dr r n 1

d2 n n1 d 2 2 e v r r , 2 dr r dr r

n , n 2, 1 , n 3.
First, note that the integral over the terms involving e and n will vanish due to the continuity of . If we rewrite the remaining terms slightly, we get

n 1,

d n1 d r r dr dr dr

2 v r r r n 1 dr

2 cn

r r a dr .

Integrating both sides gives us r n1 d r dr

2 a , cn

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1600

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

a 2 1 a a 2 1 a

2 a , c na n1

2 a 2 a , 2 c na n1 1

the last line holding since is continuous at r a . We rewrite this as

a n 1 a 2 a n 2 a , 1
where

12

2 . 2 c na n1

13

Again we can write the boundary conditions on at r a in a convenient matrix form,

A1 B1 n

A2 , B2

14

where everything is evaluated at r a . We introduce the following notation: N a We can then rewrite 14 as N a 1 N a 2 . Solving for 2 we have
1 2 N a N a 1 .

a a n a

a n a

14

15

We now have the desired transfer matrix for going from region 1 to region 2 over a delta function barrier:
1 n N a N a

n 1 m , n

2n m ,

16

where again everything is evaluated at r a . As we explained above the Wronskian is never zero, so this matrix always exists. For convenience we dene the constant

n i
and write the delta function transfer matrix as n 1 i n

2n , m ,

17

18

For the ( r ) and ( r ) listed in 5 the constant n is given by

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1601

2 , 2k 2 , 8 2 k , 8

n 1, n 2, n 3. 19

Note that when k is real, n is also real. Recall that k will be real and positive when the potential v is less than the particle energy e. When this is true ( r ) and ( r ) will be complex conjugates, and 18 can be put into a more convenient form: n
C. Properties

1 i n 2 i n 2

i n * 2 1 i n 2

20

By exploiting a few simple symmetry and conservation properties, we can make some general statements about the transfer matrices linking various regions without resorting to explicit calculations.1,1113 Recall that the probability current density in the radial direction is given by j r d d* i * . 2m dr dr

21

When is of the form 4 with ( r ) ( r ) * ( e v ) we can express this current as a sum of incident and reected current densities without any interference terms: j r d d* i . A2B2 * 2m dr dr

22

Probability ux conservation requires that the quantity A 2 B 2 be invariant in regions of the same potential so that ( r ) is the same. This, in turn, imposes the restriction M zM z 23

on the transfer matrices linking two such regions. In addition, the principle of symmetry under time reversal requires that the transfer matrix M satisfy M * xM x . 24

The matrices z and x in these equations are the usual Pauli spin matrices though, of course, no spin is involved here. Together these two conditions require that the transfer matrix linking any two regions of the same potential less than e must belong to the pseudo-unitary group SU1,1. Every matrix in this group can be written in the form


a b* b a*

with

a 2 b 2 1.

25

Pseudo-unitary matrices are less common, and hence less familiar to physicists, than unitary or Hermitian matrices. Some of the relevant properties of this group are reviewed in Appendix A as well as in Ref. 11. It happens to be the case that the determinant of a matrix connecting two regions of the same potential is equal to unity regardless of the value of the potential v . An explicit calculation will show that the determinants of the transfer matrices given in 11 and 18 are

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1602

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

det n

m 1 , 1 , m 2 , 2

26a 26b

det n 1.

According to Eq. 10 the determinant of n depends only on the ratio k 1 / k 2 . By taking a product of discontinuity and delta function transfer matrices one can see that the determinant of an arbitrary transfer matrix linking a region at potential v 1 to a region at potential v 2 depends only on the ratio k 1 / k n . As a check on the logical consistency of the arguments leading to the expressed forms for the transfer matrices one should verify that they give reasonable answers in certain limits. First off, if one transfers over a discontinuity of zero height or over a delta function barrier of zero strength one would expect to get the identity for a transfer matrix. Inspection of 11 and 18 shows that this is the case: n v , v , a 1, n v ,0,a 1. 27a 27b

Physical intuition also suggests expected inverses for the various transfer matrices. Namely jumping from v 2 to v 1 should cancel a transfer from v 1 to v 2 at the same location, and a transfer over a strength delta function should cancel a transfer over an strength delta function at the same location. A slightly longer inspection will reveal this to be the case:
1 n v 1 , v 2 , a n v 2 , v 1 , a , 1 n v , , a n v , , a .

28a 28b

III. WAVE FUNCTION CONFINEMENT

Consider a cylindrically symmetric potential in two dimensions that is zero everywhere except in some annular region centered at the origin. One can pose the question of how to congure the potential in that region to maximally amplify an incoming matter wave so as to effectively trap it in the center region. Given the innite number of ways in which one might congure a potential, this could be a very difcult problem. We will consider only a very restricted subcase. Namely we will allow our potential to consist of only delta function potential barriers of all the same strength, so that the only free parameters are the number of barriers and their locations. Furthermore, we will assume that the incoming matter wave is cylindrically symmetric: 0. In order to attack this problem we must rst make precise what we mean by maximally conning a wave function in the internal region. Let us suppose that our potential has m delta function barriers. There are then m 1 regions: the internal region containing the origin, the external region containing innity and m 1 interbarrier regions see Fig. 1. In each region we will assume that ( r ) is written in the form 4 with and as given in 5b. We label the coefcients in the internal region A 0 and B 0 , and successively label the coefcients in each region by A j and B j . The coefcients in the external region are then A m and B m . Since the imaginary parts of the Hankel functions go to innity at the origin we must have A 0 B 0 29

to insure that the wave function be physical. According to 22 the total probability current in the internal region must then be zero. That is, any probability that enters that region must eventually be reected back out. By conservation of probability the total current in each successive region must also be zero. This in turn implies that

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1603

FIG. 1. A cylindrical setup with m barriers at radii r i and corresponding wave functions i .

A j B j for each j in the range 0,m . Specically, we must have A m B m . Any incoming matter wave will eventually be reected back out. We can then dene the connement coefcient as the ratio of internal to external probability amplitudes:

A 0 2 B 0 2 . A m 2 B m 2

30

It is this quantity that we wish to maximize by choosing m positions at which to place the barriers. We will label these optimal positions r 0 through r m 1 . At each of these positions there will be a corresponding delta function transfer matrix connecting the coefcients in the adjoining regions: j 1 j j . Connecting the internal and external regions will be the total transfer matrix Tm m 1 1 0 with m Tm 0 . 31

Of course, we are free to choose the phase of the wave function at will. Equation 31 together with 29 provides a relation between the phase of A m and that of B m . Using this relation we can * we already know that their magnitudes are the same. always choose the phase so that A m B m As each of the delta function transfer matrices j are of the form given in 25, Eq. A12 in Appendix A guarantees that the coefcients in each region will satisfy A jB* j . Thus, from here onward we will denote j in each region by j The connement coefcient then becomes C
c 0 2 . c m 2

cj . c* j

32

33

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1604

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

Although we have only been talking about a cylindrical conguration, we can pose a similar problem in planar or spherical congurations. In the planar case we need only put an innitely high barrier at the origin, so that any wave coming in from innity is reected back no matter what the energy. The constraint on the internal coefcients in this situation is that the wave function be exactly zero at the origin. This requires that A 0 B 0 . Thus all of our comments on zero total probability current hold in the planar case as well, and we can again choose the phase so that A j B * j . The spherical case is identical to the cylindrical case in form, the only real difference being that the transfer matrices have a different dependence on r. We can then summarize the necessary constraints on the internal coefcients in planar, cylindrical and spherical congurations: c* 0 c 0 pure imaginary pure real pure real n 1,

* c 0 c0 * c 0 c0

n 2, n 3.

Before looking specically at any one particular geometry we will rst outline the general solution, and only then examine the special cases of the planar and cylindrical congurations.

A. General solution

For m barriers we have the relation m Tm r 0 , r 1 ,..., r m 1 0 , where 0

c0 , c* 0

cm * . cm

What we want to do is choose the r j s so that the quantity C


c 0 2 c m 2

is maximized. One can either think of this as xing the magnitude of c m and maximizing the internal amplitude c 0 or, equivalently, as xing the magnitude of c 0 and minimizing that of c m . Our point of view will typically be the latter. Strictly speaking, this is an optimization problem in m variables. One obvious approach would be to write down the connement coefcient as a function of the r j s, take m partial derivatives and set them all equal to zero simultaneously. Not only does this approach become quickly intractable for more than a small number of barriers but, in general, it does not give the global minimum, only a local one. A much simpler approach is to independently attack each successive barrier starting with the rst one. That is, choose r 0 to minimize the ratio c 1 / c 0 . Having xed r 0 , choose r 1 to minimize the ratio c 2 / c 1 and so on out to r m 1 , the rationale being that the product of these ratios will give a near minimal total ratio:
c m c 1 c 2 c m . c 0 c 0 c 1 c m1

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1605

This approach fails when choosing a suboptimal ratio c j / c j 1 that affects the phase of c j in such a way as to make another ratio small enough to more than compensate for the increase in c j / c j 1 . In the planar problem, this turns out not to be possible, so that the above approach does indeed give the optimal solution. The cylindrical and spherical cases are not so well behaved. However, trial and error suggests that while this approach is not optimal it gives fairly good results. So now the question becomes how to nd r j r j 1 that minimizes the ratio c j 1 / c j assuming that c j and r j 1 are known. We will call this ratio the amplication ratio at the j th barrier. It turns out that we can place some immediate bounds on the extent to which we can minimize or maximize such a ratio. Since j is of the form j

cj , c* j

Eq. A25 of Appendix A tells us that c j is proportional to the vector norm j , so that the amplication ratio can be written
c j 1 j 1 r j . j j c j

34

Furthermore, from Eq. A26 we know that this quantity will always lie between the singular values of ( r ):

2 r

r 0 1 r . 0

35

Thus, no matter what r j we pick we will always have


c j 1 2 r j . c j

So before attempting to minimize the amplication ratio, we will take a look at the singular values of the delta function transfer matrix. In the long run, the solution of this problem is facilitated by rst casting the delta function transfer matrix in a slightly different form which may appear more complicated at rst. Since we are assuming here that e v 0 so that the wave function may propagate freely we start with Eq. 20 ( 0 is assumed to be given: r and make the following denition,

1 i r 2 i 2 r

i 2 r * 1 i r 2

r tan 1 r 2 .

36

Since ( r ) 2 is always real and in the range 0,, ( r ) will be real and in the range 0,/2. Furthermore, ( r ) is either constant or a monotonically decreasing function of r for the s given in 5. From here on out we will assume that the r dependence of and is understood, and we will not indicate it expressly. Using 36 we proceed to write the elements of ( r ) in polar form: 1 i 2 1 2 4 e i tan and
1 2

1 tan2 e i sec e i

37

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1606

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

22
so that

2 2 * *

i 2 i 2

i tan . * *

38

Equations 37 and 38 together with their complex conjugates give us the following form for ( r ):

sec e i i tan

i tan

sec e i

39

In this form it is easy to see that the determinant of ( r ) is unity: det r sec2 tan2 1 Utilizing Eq. A21 we can easily write down the singular values for ( r ):

1 r sec tan , 2 r sec tan .

40a 40b

Moreover, since the determinant of ( r ) is 1, these singular values are inverses of each other:

1 r 2 r 1.
As we remarked earlier our amplication ratio will always lie between these singular values. We can show this explicitly by nding an expression for the amplication ratio. This is facilitated by rst calculating the polar decomposition of ( r ),

0 e i

sec

i tan

* i e

i i tan e *

sec

41

Note that the rst matrix in this decomposition is unitary and that the second is Hermitian and has the same singular values as ( r ). This is the nature of the polar decomposition. Unitary matrices only rotate vectors, leaving the Hermitian matrix to do the stretching. Hence, we will focus on it. T If we multiply this matrix on the right by ( c j , c * j ) , we get

d j1 d* j1

sec i tan

i tan

* i e

i e *

sec

c j sec ic * j tan

* i e

i c* j sec ic j tan * e


cj c* j

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1607

FIG. 2. Bounded amplication ratio.

We have used d j 1 here instead of c j 1 because we have not yet multiplied by the unitary matrix in 41. However, we still have d j 1 c j 1 . Taking the ratio of d j 1 and c j we get

*c * d j1 j i sec i tan e . cj c j
Now dene

42

r i
Note that is of the form

*c * j i e . c j

43

r e i r ,
We can then write 42 as

1.

d j1 sec tan . cj The absolute value of this ratio is equal to our amplication ratio,
d j 1 c j 1 sec tan . c j c j

44

As 1 this quantity clearly lies between the singular values 2 and 1 :

2 sec tan 1 .

45

When 1 the amplication ratio equals 2 and when 1 the ratio equals 1 . Figure 2 shows a typical plot of Eq. 45. The solid line is the amplication ratio, while the dashed lines are the bounding singular values 2 and 1 . The point indicated in the gure is the rst point at which the amplication ratio is equal to 2 . A more explicit form for the amplication ratio is

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1608

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

FIG. 3. The tangent point versus the true minimum.

c j 1 sec2 tan2 2 sec tan cos 1/2. c j

46

Note that choosing 1 ( ) does not necessarily give the minimum value for the amplication ratio as is, in general, a function of r. However, it should be clear from Fig. 2 that the true minimum will always occur very near 1. Figure 3 shows a blowup of the region around this point showing the true minimum versus the tangent point. Instead of choosing the true minimum of the amplication ratio for r j we will use the point at which the amplication ratio is equal to 2 , the reason being that the equation determining the singular values is simpler in form than the equation determining the amplication ratio. This allows us to write down both the transfer matrix at r j and the overall connement coefcient in a particularly simple form. Going back to 43 and setting 1 gives us the following condition on and at r j : c* j i i e . * cj 47

This is the critical equation that must be solved for r j . Note that it is recursive in nature. One must rst solve it for r 0 in order to nd c 1 , and then solve it for r 1 in order to nd c 2 , and so on. As the equation usually permits multiple solutions, it is the rst solution larger than r j 1 that is chosen, as this will give the smallest ratio. Applying this to Eq. 41 gives us the form of the transfer matrix at r j :

0 e
i

sec tan c* j cj

tan sec

cj c* j

48

Knowing j , r j and c j we can nd c j 1 and repeat the process: j

e i cj c* 0 j

0 e i

e i c j sec c j tan c* 0 j sec c * j tan

0 e i

2c j 2e ic j c j1 . i * * 2c * e c c 2 j j j1 49

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1609

We now have a recursive function for determining the c j s in each region assuming all of the r j s are known: c 1 2 r 0 e i r0 c 0 , c 2 2 r 1 e i r 1 c 1 2 r 0 2 r 1 e i r 0 r 1 c 0 ,

c m giving us the ratio

m1 j0

2 r j e i r j c 0 ,

50

c m r c 0 j 0 2 j

m1

51

and the connement coefcient C


c 0 2 c m 2

m1 j0

1 r j

52

recalling that 1 ( r ) 2 ( r ) 1 . By 40 we know that 1 ( r ) is always greater than 1. If happens to be constant as in the planar problem, then 1 will be constant as well and 52 reduces to
m C 2 1 .

Here the connement coefcient rises exponentially with the number of barriers. If is a decreasing function of r as is the case in cylindrical and spherical geometries, 1 ( r ) will fall and eventually approach unity as r . Therefore the dependence on the number of barriers is high at rst but tapers as more barriers are added. Note that Eq. 50 also gives us an expression for the c j / c * j ratio that occurs in the j th transfer matrix. This will be useful for nding an explicit form for the total transfer matrix: cj c* j
B. Planar case

e 2ik0 rk

j1

c0 c* 0

53

As we mentioned earlier, the planar case turns out to be exactly solvable by the method of the previous section. This is largely a result of the simple fact that here is a constant:

r tan 1 const.
This in turn implies that singular values of the transfer matrix are also both constant:

1 r sec tan const, 2 r sec tan const.


At each stage, therefore, we are free to choose an r j that achieves the minimum possible amplication ratio no matter what the positions of the other barriers. The only problem remaining is whether or not we can nd an exact expression for each of these r j s. Due to the simplicity of ( r ) this, too, is possible.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1610

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

Starting with 47, c* j i i e * cj and, using the form for from 5a, we get e 2 ikr j i c* j cj e i.

We can then use Eq. 53, recalling that is constant, to write the ratio of c * j and c j as c* j cj leaving us with e 2 ikr j ie i 1 2 j . Taking the logarithm of both sides and solving we nd that r j 1 3 1 2 j tan 1 . 2k 2 e 2i j c* 0 c0 e 2 i j , 54

55

As one might expect this gives us a constant spacing between barriers: 1 r j r j 1 tan 1 . k Note, however, that the distance between the origin and the rst barrier at r 0 ) is different than the interbarrier spacing: r 0 1 3 tan 1 . 4k 2k

We can write out an explicit form for the transfer matrix at the j th barrier by applying 54 to the general form 48: j

e i 0

0 e i

sec tan e 2 i j

tan e 2 i j sec

UH j .

Note here that the unitary matrix U is independent of j by virtue of the constant nature of . Having an expression for the individual transfer matrices we would also like to write out an explicit form for the total transfer matrix over m barriers: Tm m 1 m 2 1 0 . Though this may appear complicated at rst, a few simple matrix manipulations will make it possible. The rst step is to notice that H j U UH j 1 . Multiplying both sides by U 1 on the right gives H j UH j 1 U 1 U j H0 U j .

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1611

Applying this result to j gives j U j 1 H0 U j . 56

In this form the product of successive transfer matrices is made easier since the internal Us all cancel: Tm m 1 m 2 1 0 Um H0 U m 1 Um 1 H0 U m 2 U2 H0 U 1 UH0 Um Hm 0 . 57 Due to their simple forms the m th powers of U and H0 are not hard to calculate: Tm

e i 0

0 e i

sec tan

tan sec

1 e im 2 0

0 e im

m m 1 2 m m 1 2

m m 1 2 m m 1 2

58

We have inadvertently given the polar decomposition of the total transfer matrix as Um is unitary and Hm 0 is Hermitian actually symmetric and positive-denite. Taking this matrix and operating on 0 we expect to recover the connement coefcient: Tm and as expected C In terms of this is C 1 2 2 m .
C. Cylindrical case

im im i 2e im i im 2e


m 2

2 m 2 1 .

The cylindrical case is necessarily more complicated than the planar one as is no longer constant, but depends on r. More specically it is a monotonically decreasing function of r:

r tan 1 J kr 2 Y kr 2 .
In order to simplify notation we will denote and at the j th barrier by

j r j tan 1 r j 2 ,
j r j .

* is 1 for the cylindrical case we can write 53 as As the ratio c 0 / c 0


c* j cj e 2ik0k
j1

c* 0 c0

e 2ik0k.

j1

59

Our equation for determining r j is then c* j j1 j i i e j ie i j 2 k 0 k . cj * j 60

Unfortunately this equation is transcendental in form and must be solved numerically for each of the r j .

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1612

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

Calculating the explicit forms for the transfer matrices in the cylindrical case is more complicated as well, but here the approach used in the planar case is at least generalizable. The specic polar decomposed form for the j th transfer matrix is readily found from 59 and 48: j

e i j 0

0 e i j

sec j tan j e 2 i k 0 k
j1

tan j e 2 i k 0 k sec j

j1

Uj Hj .

Now the unitary part depends on j as expected. We would like to perform some analogous matrix manipulations to those used in the planar case in order to arrive at the total transfer matrix. More care must be taken here, though, as U now depends on j. We rst dene the matrix Sj

sec j tan j

tan j sec j

61

Some algebra will show that H j can be written as H j U j 1 U j 2 U0 S j U j 1 U j 2 U0 1 with H0 S0 . To further simplify notation we dene the product matrix
U j U j 1 U j 2 U1 U0 , U 0 1,

so that
1 H j U j S j U j .

62

As in the planar case, if we multiply two consecutive transfer matrices together the internal unitary matrices cancel:
1 1 1 j j 1 U j H j U j 1 H j 1 U j 1 S j U j U j S j 1 U j 1 U j 1 S j S j 1 U j 1 .

Doing this for the whole sequence of transfer matrices gives us the total transfer matrix: Tm m U m S m , 63

It turns out that it does not matter in which order the U j matrices or S j matrices are multiplied since they are self-commuting:
Ui , U j 0, Si , S j 0.

64a 64b

Again we have inadvertently arrived at a polar decomposition for the total transfer matrix. This time, however, instead of two matrices both raised to the m th power we have two products of m different matrices. Fortunately the products are not hard to compute, even if they look a bit complicated:
U m

e i 0m1 0 e

0
i 0m1

65

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1613

FIG. 4. A wave function in a planar conguration. Note the even spacing of the barriers. e 1, 3.898.

1 S m 2

1 j 2 j 2 j 1 j 2 j 1 j 1 j 2 j

66

1 where is shorthand for m j 0 . Applying Tm to 0 one obtains the connement coefcient given in 52 as we would hope.

D. Results

Now that we have calculated explicit forms for the barrier positions and the transfer matrices for both the planar and cylindrical congurations, we can run some numerical trials and verify that we indeed have the predicted connement. Figures 4a and 4b show probability densities in a planar conguration. Figures 5a, 5b, and 6 show corresponding probability densities in a cylindrical conguration. Note here the strong dependence on . Figure 7 shows how the connement coefcient varies with the number of barriers in both planar and cylindrical congurations. One, perhaps unexpected, result found in these numerical trials was how sensitive the connement coefcient was to both the energy of the particle and the positions of the barriers. Figures

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1614

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

FIG. 5. A wave function in a cylindrical conguration. Note how the barrier spacing decreases with increasing r. e 1, 3.898.

8a and 8b show the extremely sharp energy dependence. In general it is found for both the planar and cylindrical congurations that the higher the spike the sharper it is, much as for a sequence of functions approaching a delta function. Figures 9a and 9b show the dependence on the barrier positions for a three barrier system. As one might expect, the connement coefcient is much more sensitive to the position of the rst barrier than it is to the last. This is because each of the r j s depend only the positions preceding it. If the rst barrier position is perturbed, everything is destroyed, but if the last barrier position is perturbed, we are still guaranteed connement at least as good as in the case of m 1 barriers.
IV. CONCLUSIONS

We have found that we can write down generalized transfer matrices for wave functions propagating in sectionally constant potentials in planar, cylindrical, and spherical geometries. Two matrices sufce to describe how the wave function propagates across potential boundaries: one for moving over a discontinuity in potential and one for moving past a delta function potential barrier. The form of these matrices is independent of the geometry. One simply needs to choose the appropriate linear independent solutions to the radial equation and insert these into the transfer

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1615

FIG. 6. A cylindrical conguration with seven barriers. By doubling we have increased the connement coefcient by three orders of magnitude. e 1, 7.797 and C 6.12 105 .

matrix expressions. By approximating a real potential with sectionally constant pieces the transfer matrices can be used to nd the wave function in any quantum system possessing the necessary symmetries. In considering the connement problem we nd that it is indeed possible to achieve an extremely large connement ratio with an arrangement of only a few carefully placed delta function barriers. In the planar problem an analytic, closed form expression was found for not only the transfer matrix and connement coefcient, but the optimal barrier positions as well. The exact solution gives a truly exponential dependence for the connement coefcient on the number of barriers. In the cylindrical arrangement an analytic form was found for the transfer matrix and connement coefcient when specic near-optimal barrier positions were chosen. However, these barrier positions need to be solved for numerically. It was found that this approximate solution gives a decreasing exponential dependence on the number of barriers, while still allowing a connement ratio several orders of magnitude larger than unity. The analysis and arguments for

FIG. 7. Connement coefcient as a function of the number of barriers in a cylindrical and planar congurations. The solid line is for a planar conguration with 3.898. The dotted and dashed lines are for a cylindrical conguration with 3.898 and 7.797 respectively. e 1 for all.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1616

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

FIG. 8. Dependence of the connement coefcient on energy in a planar conguration with 3.898.

the cylindrical problem apply equally well to the spherical problem. One simply needs to replace the cylindrical Hankel or Bessel functions with the corresponding spherical ones. Independent of geometry, however, it is found that the connement coefcient is extremely sensitive to both the barrier positions and the energy of the particle. Without an extreme ne tuning of both of these parameters the connement disappears.
APPENDIX A: THE TRANSFER MATRIX GROUP SU1,1

Flux conservation FC together with symmetry under time reversal T requires that the transfer matrix linking any two regions of the same potential in planar, cylindrical, or spherical geometries belong to the special pseudo-unitary group SU1,1. We review some of the properties of this group here. The pseudo-unitary group U1,1 is dened as group of linear transformations on C2 that preserves the indenite sesquilinear form

*v 2 u,vu* 1 v 1 v 2

A1

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1617

FIG. 9. Dependence of the connement coefcient on barrier positions in a planar conguration with three barriers. m 3, e 1 and 3.898

or equivalently as the set of nonsingular matrices that preserves the indenite metric z the usual Pauli spin matrix M zM z . Transformations of this form leave the pseudo-norm A2

u , u u 1 2 u 2 2

A3

invariant and so are ux conserving. The addition of T symmetry requires that the matrices belong to the subgroup of U1,1 possessing unit determinant, denoted SU1,1. This group can be written as the following set of 2 2 complex matrices: SU 1,1


a b* b a*

a2b2 1 .

A4

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1618

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

Unlike unitary and Hermitian matrices, pseudo-unitary matrices are not normal they do not commute with their adjoint but rather pseudo-normal i.e., normal with respect to the given indenite metric. In particular, if A U(1,1), then A 1 zA z . For an arbitrary element of SU1,1 this is a b b* a* A5

a* b

b* a

A6

1. Isomorphisms

As noted in Ref. 11 there are other realizations of the group SU1,1 that are sometimes convenient. In particular, SU1,1 is isomorphic to both SL2,R, the group of 2 2 real matrices with unit determinant, and Sp2,R, the real symplectic group. The isomorphism between SU1,1 and SL2,R is suggested by the fact that both the trace and determinant of elements in SU1,1 are real. If A SU(1,1) is given by A then tr A 2 Re a , det A 1, A7a A7b


a b b* a*

from which it immediately follows that the eigenvalues of A are either purely real and inverses or each other or complex conjugates of modulus one. That the same is true of matrices in SL2,R suggests that elements in SU1,1 are related to real matrices by a similarity transformation. In fact, for any A SU(1,1), SAS 1 SL 2,R , where S is the unitary matrix S Likewise, for any B SL(2,R), S 1 BS SU 1,1 . A9 1 A8a

& i

1 i

& 1

i i

A8b

Thus, SU1,1 and SL2,R are conjugate subgroups of GL2,C and so isomorphic. The real symplectic group Sp2,R is dened as the group of linear transformations that preserve an antisymmetric bilinear form on R2 , or, equivalently, as the group of real matrices that satisfy M TJM J , where J A10a

0 1

1 0

A10b

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1619

One can show that a matrix M satises this equation if and only if M SL(2,R) so that Sp(2,R) SL(2,R). Hence, we have the isomorphisms SU 1,1 SL 2,R Sp 2,R .
2. Action

A11

As a subgroup of GL2,C, the group SU1,1 acts on C2 . Due to its relationship with SL2,R, however, it leaves some important subsets of C2 invariant. In addition, via conjugation by S, SU1,1 acts directly on R2 . * . That is, consists of all vectors of the form Dene v C2 v 1 v 2


a b b* a*

c , c*

c C.

It is important to note that is not a subspace of C2 , as it is not closed under scalar multiplication. In fact, the subspace spanned by is all of C2 . However, one may note that is a subspace of C2 when considered as a four-dimensional vector space over R. In either case, it is easy to see that SU1,1 leaves invariant: SU(1,1) ,


S R2 ,

c ac b * c * d . c* d* a * c * bc

A12

Another way to see this is to rst observe that S 1 R2 . A13

Indeed, for x , y R we have S

x iy x & . x iy y

A14

It is then easy to see that the action of SU1,1 on is entirely equivalent to the action of SL2,R on R2 : SU 1,1 S 1 SL 2,R S S 1 SL 2,R S S 1 SL 2,R R2 S 1 R2 . Going back to A12 we note that d ac b * c * c a b *

c* c a b * sgn c * 2 , c

A15

so that the ratio d / c is a complex number depending only on a, b, and c (sgn(c*)sgn(c)1 eic). In particular,
d a 2 b 2 2 a b cos a b 2 c 1/2, c

d c a b * sgn c * 2 . We noted that is not closed under multiplication by C. However, it should be noted that the set so obtained is also invariant under SU1,1: z v z C and v . If we write A16

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1620

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

z re i ,

vs

e i , e i

then we can write an arbitrary member of as u rs

e i . e i

A17

In this form it should be clear that consists of all those vectors in C2 whose components have equal magnitude: u C2 u 1 u 2 . A18

It is easy to see that SU1,1 leaves invariant: Let A SU(1,1), z C, v and u z v . Then Au A z v z A v z v u , so that SU 1,1 .
3. Singular value decomposition

A19

The singular value decomposition SVD of a matrix A is given by A U V , where U and V are unitary matrices and is a diagonal matrix with decreasing but non-negative values on the main diagonal. If A is square and nonsingular, then will be positive denite. The diagonal elements of are called the singular values of A. They are given by the positive square roots of the eigenvalues of A A . The SVD of a matrix is important because it reveals clearly how much the matrix can stretch or shrink an arbitrary vector. The singular values of A are precisely the 2-norm lengths of the semi-major axes of the hyperellipsoid dened by A v v 1 . See Ref. 14 for more information. For A SU(1,1) the SVD is straightforward to calculate. The eigenvalues of A A are given by A A AA a b 2 , so that the singular values of A are

a2b2

2 a *b *
a2b2

2 ab

A20

1,2 a b .

A21

We have taken the positive square root in both cases as det(A)1 and so a b . It is then straightforward to calculate the SVD of A: U 1

& sgn b

sgn a

sgn a sgn b

A22a

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1621

V and

& sgn ab

1 sgn ab

A22b

ab

0
ab

A22c

Note that this decomposition takes us out of the group SU1,1 as U and V are unitary not pseudo-unitary.
4. Polar decomposition

The polar decomposition of a matrix A is the equivalent of the complex number decomposition z re i . It gives a unitary matrix times a positive-denite Hermitian matrix R or vice versa: A R R . This decomposition is useful for separating the actions of rotation and stretching. The matrix R or R ) will be seen to contain the singular values of A and is thus responsible for stretching, while U, being unitary, only rotates. The polar decomposition of a matrix A is easily calculated from the SVD. For A SU(1,1) we have A R UV V V or A R U U UV

sgn a 0

0 sgn a *

b * sgn a *
a

b sgn a

A23a

b * sgn a
a

b sgn a *

sgn a 0

0 sgn a *

A23b

Unlike the SVD, the polar decomposition of A leaves us in SU1,1 as the matrix is both unitary and pseudo-unitary.

5. Norm

An interesting property of SU1,1 is that p-norm of any matrix in SU1,1 appears to be independent of p. In particular, let A SU(1,1). Then
A 1 A 2 A ab.

A24

Recall that the p-norm of a matrix A is dened as


A p sup
v

Avp , 0 vp

where the vector p-norm is given by


v p v 1 p v 2 p v n p 1/p .

Thus the p-norm of A is the maximal amount that A will stretch a vector. Note that if v dened in Appendix A, Sec. 2, then

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1622

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

v p 2 1/p v i .

A25

Since A leaves invariant, A v p 2 1/p v i and the quantity


A v p v i vp v i

is independent of p. However, such a statement is not generally true for v . The 2-norm, or spectral norm, of a matrix is the most important. One can show that it is given by its largest singular value,
A A 2 1 .

Moreover, one can show that for any A GL(2,C) and any v C2 that

Av 1 . v

A26

For A SU(1,1) and v ( c , c * ) T we can write


Avp a 2 b 2 2 a b cos a b 2 c 1/2, vp

which, by varying the cosine, obviously lies between


ab
Avp ab vp

27

for a b .
APPENDIX B: TRANSFER MATRIX EXPRESSIONS

For reference, we list here the components of the discontinuity and delta function transfer matrices for the choices for ( r ) and ( r ) made in 5.
1. Discontinuity

Equation 11 for the discontinuity transfer matrix is n v 1 , v 2 , a m 1 , 2 1 m 2 , 2 m 2 , 1

m 1 , 2 m 2 , 1

Using 9 and 10 for the generalized Wronskian gives us 1 k 1 k 2 e ia k 1 k 2 1 2 k k 2 k 1 e ia k 1 k 2 2

k 2 k 1 e ia k 1 k 2 k 1 k 2 e ia k 1 k 2

ia m 11 m 12 , 8 m 21 m 22

3 where

n 11 n 12 ia 2 k 2 , 2 2 1 n 21 n 22

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

J. Math. Phys., Vol. 44, No. 4, April 2003

Wave function connement/transfer matrix methods

1623

1 2 2 2 1 1 m 11 k 2 H ak 1 H 1 ak 2 H 1 ak 2 k 1 H ak 2 H 1 ak 1 H 1 ak 1 , 2 2 2 2 2 2 m 12 k 2 H ak 1 H 1 ak 2 H 1 ak 2 k 1 H ak 2 H 1 ak 1 H 1 ak 1 , 1 1 1 1 1 1 m 21 k 2 H ak 1 H 1 ak 2 H 1 ak 2 k 1 H ak 2 H 1 ak 1 H 1 ak 1 , 2 1 1 1 2 2 m 22 k 2 H ak 1 H 1 ak 2 H 1 ak 2 k 1 H ak 2 H 1 ak 1 H 1 ak 1 ,

and
1 2 2 2 1 k2 1 h n 11 k 2 h ak 1 h 1 a 1 ak 2 k 1 h ak 2 h 1 ak 1 1 1 h 1 ak 1 , 2 2 2 2 2 n 12 k 2 h ak 1 h 1 ak 2 1 h 1 ak 2 k 1 h ak 2 h 1 ak 1 2 1 h 1 ak 1 , 1 1 1 1 1 n 21 k 2 h ak 1 h 1 ak 2 1 h 1 ak 2 k 1 h ak 2 h 1 ak 1 1 1 h 1 ak 1 , 2 1 1 1 2 n 22 k 2 h ak 1 h 1 ak 2 1 h 1 ak 2 k 1 h ak 2 h 1 ak 1 2 1 h 1 ak 1 .

2. Delta function

Equation 18 for the delta function transfer matrix is n 1 i n Using 19 for n we get 1 2 1 1 i 2 k e 2 iak
1 2 2 H ak H ak 1 2 8 H ak

e 2 iak 1

2 1 i

2 2 H ak

1 2 H ak H ak 2 2 h ak 1 2 h ak h ak

1 2 2 k h ak h ak 3 1 i 1 2 8 h ak
1 2

E. Merzbacher, Quantum Mechanics, 3rd ed. Wiley, New York, 1998. D. Kiang, Am. J. Phys. 42, 785 1974. 3 T. M. Kalotas and A. R. Lee, Am. J. Phys. 593, 48 1991. 4 D. W. L. Sprung, H. Wu, and J. Martorell, Am. J. Phys. 6112, 1118 1993. 5 J. S. Walker and J. Gathright, Am. J. Phys. 625, 408 1994. 6 E. X. Ping, J. Appl. Phys. 7611, 7188 1994. 7 M. A. Kaliteevskii, R. A. Abram, V. V. Nikolaev, and G. S. Sokolovski, J. Mod. Opt. 465, 875 1999. 8 T. M. Kalotas, A. R. Lee, and V. E. Howard, Am. J. Phys. 593, 225 1991. 9 E. X. Ping, J. Quant. Elec. 317, 1210 1995.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

1624
10 11

J. Math. Phys., Vol. 44, No. 4, April 2003

J. D. Olson and J. L. Mace

S. Hassani, Mathematical Physics: A Modern Introduction to its Foundations Springer-Verlag, New York, 1999. A. Peres, J. Math. Phys. 245, 1110 1983. 12 P. A. Mello and J.-L. Pichard, J. Phys. Paris I 14, 493 1991. 13 P. Pereyra, J. Math. Phys. 363, 1166 1995. 14 G. H. Golub and C. F. Van Loan, Matrix Computations, 3rd ed. John Hopkins University Press, Baltimore, 1996.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 169.226.11.193 On: Fri, 17 Jan 2014 17:07:06

Anda mungkin juga menyukai