Professional Documents
Culture Documents
A THESIS
Submitted to the University of Madras
in Partial fulfilment of the requirement
of the Degree of
DOCTOR OF PHILOSOPHY
By
KUMAR ESWARAN
1
January 1973
1
The support of the University Grants Commission is gratefully acknowledged
PREFACE
This thesis embodies the results of the investigations carried out by the au-
thor during the period 1969-1972 in the Department of Theoretical Physics of
the University of Madras, on the phase and coherence aspects of quantum sys-
tems.
The thesis consists of four chapters and an appendix. Chapter II embodies the
results of our studies on generalized phase operators for the quantum harmonic
oscillator. In Chapter III we present an approximate solution for the energy
levels of the anharmonic oscillator, which we obtain through the use of action
and phase variables. Chapter IV concerns itself with the behaviour of small
wavepackets in spherically symmetric potentials and, in particular, the question
whether there exist any wave packets which remain coherent for all time.
The work reported in this thesis has been done in collaboration with and
under the guidance of Professor P.M. Mathews. It forms the subject matter of
five papers, two of which have already been published:
and three others (see footnote)1 which are in the course of publication.
To the knowledge of the author the results of the thesis have not formed any
part of any earlier thesis.
ACKNOWLEDGEMENTS
Finally thanks are due to Mr. Mohamed Jawad for his excellent typing of
the thesis.
1 NOTE ADDED: (by Author) These three papers have subsequently been published
1
CONTENTS
Chapter I : Page 3
Introduction
Chapter II : Page 9
The Action & Phase (Angle) Variables & the Energy Levels
of the Anharmonic Oscillator
Chapter IV : Page 26
Chapter V : Page 36
Appendix and References
2
Chapter 1
Introduction
This thesis deals with some of the problems of the quantum theory which de-
spite the varied successes of the theory since it first originated [1-4], have yet
remained unresolved. Broadly speaking the problems we consider have to do
with the role of the phase variable in relation to specific quantum systems (the
harmonic and the anharmonic oscillators) and the question of coherence in three
dimensional systems.
The problem of the phase variable for the quantum harmonic oscillator arises
when one tries to find a well behaved quantum mechanical opeator correspond-
ing to the phase or angle variable φ (canonically conjugate to the action variable
J) of the classical oscillator. The existence of a operator which is conjugate to
the number operator was postulated many years ago by Dirac [5]. Dirac pro-
posed to introduce a phase operator φop through the definition
1
a = exp(iφop )N 2 , (1.1)
where N is the number operator:
1
N = ~(N + ), N = a† a (1.2)
2
a and a† being the familiar annhilation and creation operators obeying the
commutation rule
[a, a† ] = 1 (1.3)
The equation (1.1) was meant to be the analogue of the classical expression
mω 1 1
( ) 2 (x + ip/mω) = eiφ J 2 , (1.4)
2
which follows from the definition of the classical phase through
2J 1 1
x=( ) 2 cosφ, p = (2Jmω) 2 sinφ, (1.5)
mω
If exp(iφop ) of (1.1) is taken to be an unitary operator, as one would expect,
then the sustitution of (1.1) in (1.3) would immediately yield
exp(iφop ) N exp(−iφop ) = N + 1, (1.6)
3
But this cannot be: a similarity transformation of N cannot change it to N+1
since the spectrum of eigenvalues of the two operators cannot coincide. In fact
the impossibility of defining an unitary operator exp(iφop ) through (1.1) was
first pointed out by Susskind and Glogower [6]. Indeed if one assumed the
existence of such an operator satisfying the commutation relation
or equivalently, with
U = C + iS , (1.10)
[U, N ] = U ; (1.11)
from (1.11) it is easy to see that U is a step down operator for the number
eigenstates.
Taking,
U |ni = (1 − δn0 )|n − 1i (1.12)
Susskind and Glogower [6], have solved the eigenvalue problem for the C and S
operators, and shown that their eigenvalue spectra are continuous and fill the
range (−1, 1) and that the eigenstates form a complete set.
1 Varga and Aks, Ref. [7], claim to have obtained a representation of equation (1.7) in a
basis related to the positive momentum eigenfunctions of a particle in a one dimensional box
(−1/2, 1/2), see also Refs. [8] and [9]. Their argument rests on the identity of the eigenvalue
spectra of VP (P,Q being the momentum and position operators of the particle in the box
and V the projection operator to the states of non-negative momentum) and of the number
operator N, from which they infer that these positive momentum states and the number
eigenstates are related by an unitary transformation. They propose then that φop be defined
through the unitary transformation V QV → φop . However, their arguments seem untenable:
while the zero eigenvalue of N is nondegenerate, that of VP is infinitely degenerate(since all
negative momentum eigenstates of P become eigenstates belonging to the eigenvalue zero of
VP). Thus the assumed unitary transformation does not exist.
2 A similar approach as has been employed by Louisell [10 ] is useful in dealing with the
problem of the azimuthal angle conjugate to Lz in angular momentum theory, eg. see Ref.
[11], for the discussion of the phase problem in angular momentum. Carruthers and Nieto
Ref. [12], provide an excellent review on these subjects.
4
The cosine and sine operators defined through (1.10) to (1.12) are not the
only possible operators that can be obtained. It was pointed out by Lerner [13]
that equation (1.12) is not really mandatory, one could have more generally
where f (n) is arbitrary except for certain restrictions that are needed to ensure
that the C and S have eigenvalue spectra extending from -1 to +1. Lerner has
given examples for various forms for f (n) , following from the use of various
kinds of symmetrizations in going from the classical to the quantum variables.
The eigenstates for the C and S operators for a particular choice of f (n) have
been worked out by J. Zak [11]. Since U is not an Hermetian operator the eigen-
values of U will not be necessarily real. Lerner, Huang and Walters [14] have
proved the interesting fact that all complex numbers z with |z| < 1 are eigen-
values of U. The corresponding eigenstates form an overcomplete set. Recently,
some of the spectral properties of the phase operators have been obtained also
by Ifantis [15],[16] using the observation that C and S belong to a special class
of tridiagonal operators.
The problem of the anharmonic oscillator has been the subject of a great
deal, [18] -[20], of recent work which has been motivated largely by the hope
of exploiting the analogy of the anharmonic oscillator and the nonlinear fields
to gain some insight into the latter. The results are however of considerable
interest. Bender and Wu [19] showed that if the energy eigenvalues En of the
5
hamiltonian
p2
H= + αx2 + βx4 (1.14)
2m
are calculated by a perturbation method in powers of β the series fails to
converge for any β . They have traced the reason for this divergence to the
peculiar analytical properties of En (β) in the complex β plane. Despite the
singularities of En (β), it turns out that the values of En for physical β can be
clearly reproduced by P adé approximants [21]-[22]. Other methods of comput-
ing E (eg. in terms of the roots of a Hill determinant) have also been explored
recently[23]-[25]. However throughout this work one does not find a simple for-
mula for quick and practical computation of the energy levels. We show in
Chapter III of this thesis that by a semiclassical method based on the use of
the action and angle variables an approximate formula which serves this pur-
pose can be obtained. It not only reduces to the harmonic oscillator spectrum
as β → 0 but also reproduces the known asymptotic (large n) expression for
En (β) . We also derive a scaling relation (for the exact eigenvalues) which is
more general than any previously known in the literature [20] and show that
our approximate formula for En (β) does obey this scaling relation.
a statistical ensemble around some point in phase space) see M.Born and D.J. Hooton, and
M.Born Refs. [29]-[31].
4 Such solutions have been called ‘The Double solutions’, for a discussion of this, see Ref.
[28].
6
already referred to in Sec.1, and they are characterised by a minimum value for
the product of the extensions in coordinate and momentum spaces. But even in
the case of such an important system as a particle in a coulomb-type potential,
nothing seems to be known about the behaviour of a small three dimensional
wavepacket as time progresses, and in particular, as to whether there exist co-
herent (in the sense of non-spreading) wave packets in such a potential.
7
packet) involve various cross correlations eg. h(~x − h~xi)R (~
p − h~
pi)T i. It turns
out that one can get a closed system of first order differential equations with
constant coefficients for a set of six such correlated functions together with the
four quantities (1.15). Taking these ten quantities as the elements of a ten
component column vector u, we can write the equations in the matrix form:
du
= ωM u, (1.17)
dt
where M is a 10×10 matrix in which the particular central potential enters, only
through a single parameter, and ω is the frequency of the orbit. The question
of coherence of the wave packet then boils down to whether (1.17) has solutions
such that dudt = 0. We find that formally such solutions do infact exist, since
M happens to have zero eigenvalues (four of them); but it turns out that none
of these solutions is compatible with the essential positivity of the quantities
(1.15) . This difficulty manifests itself through the non-diagonalizability of the
matrix M . Positivity can be ensured only by allowing some of the uncertainties
defined through (1.15) to increase indefinitely. This is the case in particular for
the uncertainty in the tangential components of position: There is no choice of
initial conditions by which spreading of the wavepacket in this direction can be
prevented. The case of πR (which measures the uncertainty in the radial com-
ponent of momentum) is similar. On the other hand the tangential component
of momentum and the radial component of position as well the components per-
pendicular to the plane of motion remain well defined, their uncertainties being
constant or varying harmonically at worst. The details of these calculations are
carried out for the case of particular interest, namely the coulomb potential.
But we observe that the results hold quite generally, for central potentials, the
only exception (where true coherent states exist) being the three dimensional
harmonic oscillator. In this case the M matrix becomes diagonalizable.
8
Chapter 2
1. Introduction
We have noted in the last chapter, though an unique well-defined phase op-
erator conjugate to the number operator N of the quantum harmonic oscillator
does not exist, the notion of phase can be introduced through cosine and sine
operators C and S defined through the commutation relations:
[C, N ] = i S and [S, N ] = −i C (2.1)
or equivalently, with
[U, N ] = U where U = C + iS . (2.2)
Equation (2.2) implies that
U |ni = f (n) |n − 1i , with f (0) = 0 , (2.3)
where the sequence of numbers f (n) has to be so chosen as to ensure that the
hermetian operators C and S have the right kind of eigenvalue spectrum, namely
a continuous one extending from -1 to +1 , consistent with their interpretation
as cosine and sine operators. It was Lerner’s observation that the necessary and
sufficient conditions for this purpose5 are that
f (n) → 1 , as n→∞ (2.4)
and that it be possible to express the f (n) in terms of another sequence
gn (0 ≤ gn ≤ 1) (2.5)
1 2
f (n) = gn (1 − gn−1 ) . (2.6)
4
5 See Lerner,Huang and Walters Ref. [14] for a detailed proof of the ‘chain sequence condi-
9
Subject to these conditions one can make any of a vast variety of choices for f (n)
and hence for C and S. This freedom of choice raises interesting possibilities.
One may ask, for instance, whether one can define C, S in such a way that
within some class of coherent states. Such a question is prompted by the fact
that coherent states are wave packets with a classical motion and classically,
the cosine and sine functions have a vanishing Poisson bracket, the phase angle
being uniquely defined. We will consider this in section 5, but first we solve the
the eigenvalue problem for C and S in all generality. We show in section 2 that
if
C |λi = λ |λi , (2.8)
and |λi is expanded in terms of the number eigenstates |ni as
n=∞
X
|λi = Cn (λ) |ni , (2.9)
n=0
then the Cn (λ) form a set of orthogonal polynomials in λ . We also verify that
the states |λi form a complete set, with the eigenvalue parameter λ ranging
from -1 to +1 (provided the chain sequence condition holds). It is remarkable
that there are such elegant properties associated with the phase operators and
that they are demonstrable even when the f (n) is quite arbitrary. For the famil-
iar special case [6],[12] when f (n) = (1 − δn0 ) , the coefficients of transformation
Cn (λ) reduce to the Gegenbauer polynomials. Similar properties can be de-
duced with regard to eigenstates of the sine operator.
6 Actually there is one normalizable state satisfying his extrememum condition, namely | 0i,
the ground state . However this is a state in which (4C)2 (4S)2 has its maximum value.
10
2. The Eigenstates of the Generalized Cosine and Sine Operators
Consider now the eigenvalue problem for C, eq. (2.8), with C given in terms
of U defined by (2.3), as
1
C = ( U + U† ) (2.10)
2
On sustituting (2.9) in (2.8) and using (2.10) and (2.3) we obtain the following
recursion relations for the coefficients Cn (λ) :
2 λ C0 (λ) = C1 (λ) f (1),
2 λ C1 (λ) = C2 (λ) f (2) + C0 (λ) f (1),
− − − − − − − − − − − − − − −,
2 λ Cn (λ) = Cn+1 (λ) f (n + 1) + Cn−1 (λ) f (n), (2.11)
To solve eqs. (2.11) it is necessary to rewrite them in the form:
f 2 (n)
Fn+1 (λ) = 2 λ − , (2.12)
Fn (λ)
where
Cn (λ)
Fn (λ) = f (n), (2.13)
Cn−1 (λ)
By virtue of ( 2.13) we can write
Qn
Fr (λ)
Cn (λ) = Qr=1 n C0 (λ), (2.14)
r=1 f (r)
Qn
Further it is a consequence of (2.12) thet the product r=1 Fr (λ) is a polynomial
of degree n in λ. It follows from (2.14) that Ĉn (λ) ≡ Cn (λ)/C0 (λ) is also such
a polynomial. It may be verified that its explicit form is
Ĉn (λ) ≡ Cn (λ)/C0 (λ)
" n #−1 { n }
Y X2
11
(n)
values of r such that r ≤ { n+1 2
2 }. For other values of r , Dr (f ) could be
defined to be zero; in any case they do not occur within the range of summation
in the expansion (2.15). We observe further that the definitions (2.16) lead to
the identity
(n−2)
Dr(n−1) (f 2 ) + Dr−1 (f 2 ) f 2 (n) = Dr(n) (f 2 ) , (2.17)
which is very useful in verifying explicitly that (2.15) for Cn (λ) is indeed the
solution of the recursion relations (2.11).
We will now show that the polynomials Ĉn (λ) form an orthogonal set, that
is, a weight function ρ(λ) can be found such that
Z +1
ρ(λ) Ĉm (λ) Ĉn (λ) dλ = δmn (2.18)
−1
(where An , Bn and Ln are constants, with An > 0 and Ln > 0 ) and if the
coefficient of xn in Pn (x) is kn then the necessary and sufficient condition for
Pn (x) to be orthonormal is that
kn An kn kn−2
An = ; Ln = = 2 (2.20)
kn−1 An−1 kn−1
2 f (n − 1) 2n
An = ; Bn = 0 ; Ln = and kn = Qn (2.21)
f (n) f (n) r=1 f (r)
It is evident that they satisfy (2.20). Hence the polynomials Ĉn (λ) form an
orthonormal set, as asserted earlier. Since the spectrum of C is known [13],[14]
to extend from -1 to +1, this is the domain over which the orthogonality relation
holds, and this fact has been used in writing the limits of integration in (2.18).
We have not been able to determine the weight function ρ(λ), however. If
we choose thep coefficient C0 (λ) (which is not determined by the recurrence
relations) as ρ(λ), we can write (2.18) in view of (2.15) as
Z +1
Cm (λ) Cn (λ) dλ = δmn (2.22)
−1
It is expected that the infinite set of orthonormal functions Cm (λ) form a com-
plete set for normalizable functions of λ over the interval (−1, 1). In fact this can
be inferred from the orthonormality of the states |λi which are the eigenstates
of the hermitian operator C :
X
Cm (λ) Cm (λ0 ) = hλ|λ0 i = δ(λ − λ0 ) (2.23)
m
12
whence for arbitrary f (λ),
Z
f (λ0 ) δ(λ − λ0 ) dλ0
f (λ) =
X Z
= Cm (λ) f (λ0 )Cm (λ0 ) dλ0 (2.24)
m
Thus any f (λ) can be expanded in terms of the set Cm (λ) which is therefore
complete.
While the polynomials Ĉn (λ) cannot be identified with any known set when
the f (n) are arbitrary, in the special case when f (n) = (1 − δn0 ) they reduce to
the Gegenbaur polynomials. This can be verifiedp by inspection of (2.15). The
weight function ρ(λ) in this case is ρ(λ) = (1 − λ2 ). Similar properties can
be proved for the coefficients of expansion (in terms of number eigenstates) of
the eigenstates of the S-operator. In fact it turns out that the coefficients are
just in Cn (λ).
Since the commutator of the number operator N and the cosine operator
C is the operator i S (and not a c-number) the equation which determines the
state for which (∆N )(∆C) is a minimum is quite complicated as pointed out
by Jackiw. But if the uncetainty product is defined to be
(∆N )2 (∆C)2
,
hSi2
the condition for this to be a minimum is relatively simple,8 and following
Jackiw, we will adopt this definition for our purposes. The relevant equation is
then
(N + i γ C) |ψi = λ |ψi, (2.25)
where
1
λ = hN i + i γ hCi and γ = [(∆N )2 /(∆C)2 ] 2 , (2.250 )
Jackiw has obtained the solution of this equation for the special case f (n) =
(1 − δn0 ), and we will generalize it here for arbitrary f (n). For a state |ψi
n=∞
X
|ψi = an (λ) |ni , (2.26)
n=0
which satisfies the minimum condition (2.25), we obtain the following recurrence
relations with the aid of equations (2.2) and (2.3):
iγ
a1 (λ) f (1) = λ a0 (λ) (2.27)
2
− − − − − − − − − − − − − − −,
2(λ − n)
an (λ) = an+1 (λ) f (n + 1) + an−1 (λ) f (n), (2.28)
iγ
8 See Jackiw, Ref [17 ]; and Carruthers and Nieto, Ref.[12], for a further discussion of this
question.
13
Observing that φn (λ) , defined by
an (λ)
φn (λ) ≡ f (n),
an−1 (λ)
satisfies
2 (λ − n) f 2 (n)
φn+1 (λ) = − , (2.29)
iγ φn (λ)
one finds, using the same procedure as in the solution of (2.11), that the ratio
Qn
an (λ) φr (λ)
= Qr=1n , (2.30)
a0 (λ) r=1 f (r)
th
is a poly nomial of
the 2n degree
in λ. Eq. (2.26) with (2.30) gives the state
(∆N ) (∆C)2
which minimises hSi2 .
Jackiw has pointed out the relevance of this kind of uncertainty product when
dealing with a pair of operators whose commutator is not a c-number. From his
treatment it follows that the states which minimise Q must satisfy the equation:
! !
Ĉ Ŝ Ĉ Ŝ
−i +i |ψi = 0 ; (2.31)
∆C ∆S ∆C ∆S
where
Ĉ = C − hCi and Ŝ = S − hSi. (2.32)
This would be ensured of course if
!
Ĉ Ŝ
+i |ψi = 0 , (2.33)
∆C ∆S
and we now proceed to solve this more restrictive equation. We first rewrite it
as
(1 + γ)U + (1 − γ)U † |ψi = 2 λ |ψi ,
(2.34)
where
12
(∆C)2
γ = , (2.35)
(∆S)2
and
λ = hCi + iγ hSi. (2.36)
If we now assume an expansion
n=∞
X
|ψi = an |ni , (2.37)
n=0
14
for |ψi in terms of the number eigenstates, we obtain from (2.34) the following
recurrence relations for the expansion coefficients:
eq.(2.38) simplifies to
where
λ
x= p . (2.41)
1 − γ2
Eq. (2.40) is identical to (2.11), so that we can write down the solution imme-
diately:
#−1 {n
2}
" n
Y X
bn (x) = b0 (x) f (r) (2x)n−2r (−1)r Dr(n−1) (f 2 ), (2.42)
r=1 r=0
Hence
#−1 {n
2}
" n
Y X
n/2
an (x) = a0 (x) q f (r) (2x)n−2r (−1)r Dr(n−1) (f 2 ), (2.43)
r=1 r=0
or
2
1
|α| < (2.46)
q
where
bn+1 (x)
α = lim (2.47)
n→∞ bn (x)
The value of α may be found from eq.(2.40) reexpressed as
bn+1 bn
f (n + 1) + f (n) = 2 x . (2.48)
bn−1 bn−1
15
On letting n → ∞, remembering that f (n) → 1, we get
α2 − 2 x α + 1 = 0. (2.49)
This equation describes an ellipse in the λ-plane. we now proceed to prove this
assertion. There are two cases to be considered : γ 2 > 1 and γ 2 < 1. If γ 2 > 1,
we write p
λ = γ 2 − 1 sinh θ, (2.54)
where θ is in general complex,
θ = u + iv.
1 + γ iv γ 2 − 1 −iv
λ= e − e ≡ µ + iν (say) (2.56)
2 2(1 + γ)
16
eigenstates of U found by Lerner at. al. [14]. It has also been shown by Lerner
et. al. that the eigenstates of U form a complete (or rather an over- complete)
set. By using a similar argument as that used by Cahill [42], we can assert that
for general values of γ , the set of all states |ψi characterized by λ lying within
the ellipse of convergence form an over complete set.
we define
1
α = N 2 eiφ
10 We have defined these functions using the same notation as in Ref. [12], except that here
17
for coherent states the following expressions can be easily obtained :
1
hα|C|αi = N 2 exp(−N ) cos φ ψ1 (N ), (2.60a)
1
hα|S|αi = N exp(−N ) sin φ ψ1 (N ),
2 (2.60b)
1 1 N
hα|C 2 |αi = exp(−N ) [ f 2 (1) + ξ1 (N ) + cos 2φ ψ2 (N )], (2.60c)
4 4 2
1 1 N
hα|S 2 |αi = exp(−N ) [ f 2 (1) + ξ1 (N ) − cos 2φ ψ2 (N )], (2.60d)
4 4 2
1 1
hα|C 2 + S 2 |αi = exp(−N ) [ f 2 (1) + ξ1 (N )], (2.60e)
2 2
hα|C 2 − S 2 |αi = N exp(−N ) cos 2φ ψ2 (N ), (2.60f)
hα|CS + SC|αi = N exp(−N ) sin 2φ ψ2 (N ), (2.60g)
i
hα|CS − SC|αi = exp(−N ) [ξ2 (N ) + f 2 (1)]. (2.60h)
2
The quantities:
which measure the degree of incompatibility of the number phase operators can
be evaluated with the aid of eqs (2.60). Note that for the coherent states
(∆N )2 = hN 2 i − hN i2 = N
N
f 2 (0) = 0, f 2 (1) =
N −1
f 2 (r) = 1 (f or r > 1), (2.62)
We must verify that f (n) defined in (2.62) do follow the chain sequence condi-
18
11
tion:
1 2
f (n) = gn (1 − gn−1 ), (2.64a)
4
with g0 = 0 and 0 ≤ gn ≤ 1. (2.64b)
Now since
1 2
f (1) = g1 (1 − g0 ), (2.65)
4
from eq.(2.62) we have
1 N
g1 = (2.66)
4 N −1
Since
1 2
g2 (1 − g1 ) = f (2)
4
1
=
4
we have
1
g2 = /(1 − g1 ).
4
1/4
gn = .
1 − gn−1
Hence,
1/4
gn = (2.67)
1/4
1−
1/4
..... −
1 − g1
Now using the method of continuous fractions [44] and eq(2.66) it turns out
that one can sum up eq. (2.67) , to obtain
1 nN − 2(n − 1)
gn = (2.68)
2 (n + 1)N − 2n
sequence conditions (2.64) they do not have the property Limit: f (n) = 1 as n → ∞. It
is for this reason, the spectra of C and S will not span the entire region (-1,1). We are
grateful to Professor Lerner for pointing this out (Private Communication). The choice of
(2.62) overcomes this difficulty.
19
i.e.
1 N (n + 1) − 2n
≤1 .
2 N (n + 2) − 2(n + 1)
i.e.
N n − N − 2n ≤ 2nN + 4N − 4n − 4,
whence
2(n + 2)
≤ N. (2.72)
n+3
The maximum value of the l.h.s. of (2.72) is attained when n → ∞, which is 2.
Therefore if
N ≥ 2, (2.73)
equation (2.71) is satisfied. Since we must also have
gn+1 ≥ 0, (2.74)
i.e.,
n
N ≥2 (2.75)
n+1
The maximum value of r.h.s of (2.75) is 2. Therefore we see that (2.73) ensures
that (2.70) holds. We have shown that there exist phase operators such that
hα|[C, S]|αi = 0 for all coherent states having |α| = N with N ≥ 2. Since
the chain sequence conditions with the requirement limn→∞ f (n) = 1 are sat-
isfied the spectra of these operators will span the entire region (-1,1). It is
apparant from eq. (2.60h) that there could be other phase operators which are
more complicated than the simple choice given in (2.62) which will satisfy these
conditions.
20
Chapter 3
1. Introduction
p2
H= + αx2 + βx4 (3.1)
2m
If the nth energy level of (3.1) is denoted by En (m, α, β) , then a general scaling
law can easily be established by writing m = µm0 and x = ξ y (ξ = constant)
in the eigenvalue equation
~2 ∂ 2 ψ
− + (αx2 + βx4 ) ψ = En (m, α, β) ψ,
2m ∂x2
21
to obtain
~2 ∂ 2 ψ
− + (α µ ξ 4 y 2 + β µ ξ 6 y 4 ) ψ = ξ 2 µ En (m, α, β) ψ .
2m0 ∂y 2
Comparing these two equations, we discover that
En (m, α, β) = f En (m0 , α0 , β0 )
(3.2a)
where f is given by
1/2 1/3
a b a2
f= = = ; (3.2b)
µ µ2 b
in terms of the ratios
m α β
µ= , a= and b = ; (3.2c)
m0 α0 β0
which have to be so chosen that
µ a3 = b2 . (3.2d)
Our scaling law is a generalization of that found in Symanzik [20]. The latter
corresponds to setting µ = 1 in our equation (3.2).
A familiar special case, ( Simon Ref. [20]) is obtained when µ = 1,
b = β and α = 1,
2
En (m0 , 1, β) = β 1/3 En (m0 , β − 3 , 1) (3.3)
which exhibits the β 1/3 type of behaviour of the energy levels and the cube root
type of singularity at β = 0.
x = A cn(φ, k) , (3.4a)
1
2βA2 2
1
p = mẋ = − (2mα) A 1 +
2
sn(φ, k) dn(φ, k) , (3.4b)
α
d
where the fact that du cn(φ, k) = −sn(u, k) dn(u, k) and φ̇ has been written as a
constant since φ is an angle variable. The value of the constant and the relation
between A and the modulus k of the elliptic functions are so chosen that H is
free of φ. We have
βA2 αk 2
k2 = or A2 = (3.5)
α + 2βA2 β(1 − 2k 2 )
12 For definitions of the elliptic functions used inthis Chapter and some of their elementary
22
On using (3.4) and (3.5) H reduces to
H = A2 (α + βA2 ), (3.6)
2βA2
p ∂p 3 ∂x
+ 2αx + 4βx = 2αA 1 + , (3.11a)
m ∂A ∂A α
p ∂p ∂x
+ 2αx + 4βx3 = 0, (3.11b)
m ∂φ ∂φ
∂x ∂x
and multiplying (3.11a) by ∂φ and (3.11b) by ∂A , subtracting and using eqs.
(3.4) we have
12
2βA2
1
{x, p}A,φ = (2mα) A 1 +
2
. (3.12)
α
Hence from (3.8) and (3.9) we obtain
∂J 1 1
= (2m) 2 A α + 2βA2 2
∂A
Z A
1 1
J = (2m) 2 A α + 2βA2 2 dA
0
12 " 32 #
2mα3 2βA2
= 1+ −1 . (3.13)
6β α
Thus: (
12 ) 43
2
α 2
H = 1+ 3βJ − 1 . (3.14)
4β mα3
23
4. The Energy Levels
On substituting this in eq. (3.14) and noting that the second term in curly
brackets now dominates we get:
4
18π 3 12 n + 1 3 β~4 13
2
En ≈ 1 2 (3.20)
Γ 4m2
4
24
This expression coincides 13 with the asymptotic formula given in Titch-
marsh [46] for the eigenvalues in the case of the pure x4 potential. It may be
noted that the energy increases as n4/3 β 1/3 for large n. Finally it is a trivial
matter to verify that our approximate formula obeys the scaling law eqs. (3.2)
satisfied by the exact energy levels. Actual numerical computation shows that
even for low lying energy levels the formula reproduces the energy values within
a few percent, the error being of the same order as the W.K.B. evaluation [47]
which is more complicated.
13 The apparent missing of a factor of 2 within the curly brackets is because the range of x
25
Chapter 4
Coherence in Quantum
Mechanical Systems
1. Introduction
most unlikely that the case of other types of orbits would be very different.
26
and calculating the time derivatives using the quantum equation of motion:
dA 1 ∂A
= [A, H] + , (4.2)
dt i~ ∂t
with
p2 1
H = + mω 2 x2 , (4.3)
2m 2
one finds from the expressions χ̇ and π̇ involve also hxp + p xi, but
27
2
h a† (0)i = ha† (0)i2 , (4.10b)
These are necessary and sufficient conditions for a harmonic oscillator wavepacket
to maintain a constant size or extension. They are of course trivially satisfied
by the coherent states |αi which are eigenstates of a. At the other extreme we
have the energy eigenstates which being stationary states have constant values
of χ and π. In between there is a whole class of semicoherent states for which
eqs (4.10) are valid. An interesting example is provided by states of the type
where |αi and |βi are coherent states and N is a normalization factor, N =
− 1
1 − |hβ|αi|2 2 . It may be easily verified that the equations (4.10) do hold good
in these states. Note that (4.11) defines |ψi as the projection of a coherent state
|αi orthogonal to another coherent state |βi. Therefore the ordinary coherent
states |αi are not obtainable as the limit of |ψi when the two superposed states
|αi and |βi are made identical : in the limit α → β, |ψi does not exist. We will
refer to the states (4.11) as semicoherent states. it is a curious fact that in the
semicoherent state (4.11), the mean position or (momentum) of the oscillator
is exactly the coherent state |αi i.e. hψ|x(t)|ψi = x̄α (t) ≡ hα|x(t)|αi. The
position probability density |ψ(x, t)|2 in the state (4.11) is describable in terms
of two gaussian packets of constant amplitude and width (oscillating simple
harmonically with phases characterised by arg α and arg β ) together with an
interference term which ensures that χ and π remain time independent. Their
values are
2 |hα|βi|2 |α − β|2
~
χ = 1+ , (4.12a)
2mω (1 − |hα|βi|2 )
2 |hα|βi|2 |α − β|2
mω~
π = 1+ . (4.12b)
2 (1 − |hα|βi|2 )
Clearly the product of these exceed the minimum value ~ 2 /4 unlike the case of
true coherent states.
This example illustrates the wide variety of states which can satisfy the con-
dition of constant values for the uncertainties ∆x and ∆p. Of these the true
coherent states have the property of being small wavepackets around the mean
position. In the search for coherent states in three dimensions, in the remaining
sections of this chapter, we use the constancy of uncertainties of various quanti-
ties as a criterion. But we also demand small size, so that semi-coherent states
(if at all they exist) are excluded from consideration.
28
and momentum in the radial and tangential directions as well as the directions
perpendicular to the plane of the orbit. We define the quantities
* 2 +
h~xi
χR = · (~x − h~xi) , (4.13a)
R
* 2 +
1 h~pi
χT = 2 2 · (~x − h~xi) , (4.13b)
m ω R
D E
2
χz = {z − hzi} , (4.13c)
* 2 +
1 h~xi
πR = 2 2 · (~p − h~pi) , (4.13d)
m ω R
* 2 +
1 h~
pi
πT = 4 4 · (~p − h~pi) , (4.13e)
m ω R
1 D 2
E
πz = 2 2 {pz − hpz i} , (4.13f)
m ω
The subscript R indicates the radial component, T the component tangential
to the orbit (and hence parallel to h~ pi ) and z the component normal to the
plane of the orbit which is taken to be the x − y plane. The factors of mω
are included to reduce the dimensions of all the quantities to the dimension of
(length2 ). Here m is the mass of the particle and ω the angular frequency of
its motion in the circular orbit. Of course by the radius of the orbit we mean
|h~xi|. It is evident that
dV
mω 2 R = ≡ V 0 (R) . (4.14)
dr r=R
29
1
π̇z = − V33 χ̇z , (4.16b)
m2 ω 2
where V33 is defined by
V0
2 2
∂ V 2
00 z 0z
V33 = = V −V 3 +
∂z 2 0 r r r 0
V 0 (R)
= = m ω2 . (4.17)
R
The subscript 0 indicates that the quantity concerned is to be evaluated at the
position of the orbit; the fact that the orbit has been chosen to be in the x-y
plane (hzi = 0) has been used in the above reduction. Equations of exactly
the same form16 as (4.16), with V33 = m ω 2 , hold also for a one dimensional
oscillator.On solving these equations we obtain
1 1
χz (t) = (χz (0) + πz (0)) + (χz (0) − πz (0)) cos2ωt
2 2
1
+ χ̇z (0) sin2ωt (4.18a)
2ω
πz (t) = (χz (0) + πz (0)) − χz (t) . (4.18b)
Clearly the extension of the wavepacket perpendicular to the orbital plane re-
mains bounded; infact, initial conditions may be so chosen as to make χz and
πz time independent.
potential in one dimension. While the equations are exact for the harmonic oscillator, the
derivation of these equations for other potentials involves the neglect of third and higher order
derivatives of V, in a Taylor’s expansion around the point h~ xi, the ‘centre’ of the wavepacket,
a procedure which is valid for ‘small’ wavepackets (i.e. χz << R2 ).
30
These are the correlations 17 along various compnents of (~x − h~xi) and
p − h~
(~ pi). The µ’s contain one component of (~x − h~xi) and one of (~p − h~pi) and
the convention regarding the subscripts in this case is that the first subscript
indicates which component of (~x − h~xi) is involved while the second subscript
p −h~
refers to the component of (~ pi). The equations for the ten coupled quantities
written in matrix form are as follows:
χ̇R 0 0 2 0 0 0 1 0 0 0 χR
χ̇T
0
0 −2 0 0 0 0 0 0 1 χT
χ̇RT −1 1 0 0 0 0 0 1 1 0 χRT
π̇R 0 0 0 0 0 2 η 0 0 0 πR
π̇T
= ω
0 0 0 0 0 −2 0 0 0 −1 πT
π̇RT
0
0 0 −1 1 0 0 η −1 0 πRT
µ̇RR 2η 0 0 2 0 0 0 2 2 0 µRR
µ̇RT
0
0 −1 0 0 1 −1/2 0 0 1/2 µRT
µ̇T R
0
0 η 0 0 1 −1/2 0 0 1/2 µT R
µ̇T T 0 −2 0 0 2 0 0 −2 −2 0 µT T
(4.20)
or
du
= ω M u, (4.21)
dt
Where M and u stand for the square matrix and the column vector which
appear in the r.h.s. of eq.(4.20). The parameter η is defined as
r ∂ω 2
η = − 1+ 2
ω ∂r r=R
for equations of the above type, where λi and u(i) are respectively the eigenvalues
and column eigenvectors of M . In the present case however, the solutions are
not of this kind, because it so happens that M is not a diagonalizable matrix (it
does not have ten independent eigenvectors) except for the special case η = −1.
The characteristic equation for M can be verified to be
31
that M is diagonalizable if η = −1 but not for any other value of η. However
one has for any η,
which shows that in general M has only two (instead of four) independent eigen-
vectors belonging to the eigenvalue 0 and only one eigenvector each (instead of
1
two) corresponding to the eigenvalues ±(η − 3) 2 . This means that the Jordan
canonical form [50] of M is
0 1 . . . . . . . .
. 0 1 . . . . . . .
. . 0 . . . . . . .
. . . 0 . . . . . .
. . . . β 1 . . . .
Mc = SM S −1 =
(4.26)
. . . . . β . . . .
. . . . . . −β 1 . .
. . . . . .
. −β . .
. . . . . . . . 2β .
. . . . . . . . . −2β
When
α
V (r) = (4.27)
r
η = 2, (4.28)
the eigenvalues of M are
0, ±i , ±2i . (4.29)
Rewriting eq. (4.21) as
dv
= ω Mc v , v = Su ; (4.30)
dt
we can readily obtain the general solution for this set of coupled equations for
32
the components of
With this definition, it is easy to see from the form of M , eq.(4.26), with η = 2,
that
u = S −1 v
= c1 + c2 ωt + c3 (ωt)2 u(1) + (c2 + 2 c3 (ωt)) u(2)
where u(i) = S −1 v (i) . The u(i) evidently satisfy equations (4.33) with Mc
replaced by M . It can easily be verified that since M is a real matrix
and hence for the reality of the solution (which is required in view of the nature
of the components of u ) we must choose
18 The u(i) in (4.37) are not normalized and u(1) , u(4) , u(5) , and u(9) are arbitrary up to
a factor. However such a factor can be absorbed into the integation constants c1 , c2 , .... etc.,
and one can easily check by differentiation that (4.34) with (4.37) is the most general solution
of (4.21), with η = 2.
33
u(1) u(2) u(3) u(4) u(5) u(6) u(9)
0 0 2/9 1 0 2i/3 1
1 0 0 3 2 2i/3 −4
0 −1/3 0 0 −i/2 −1/2 2i
1 0 0 0 1 i/3 −1
0 0 1/18 1 0 i/3 1
0
−1/6
0
0
−i/2
0
i
(4.37)
0 2/3 0 0 i 1/3 −2i
0 0 −1/9 −1 0 −i/2 −1
−1 0 0 −1 −3/2 −i/2 2
0 1/3 0 0 i 1/3 −4i
Since the quantity in curly brackets is evidently a Hermetian operator the above
expectation value can vanish only if
h~xi h~
pi
· (~x − h~xi) + 2 2 · (~p − h~pi) |ψi = 0 (4.39)
R m ω R t=0
Now let us suppose that the x-axis is chosen so as to pass through the mean
position at t = 0, so that hxi = R, hyi = hzi = 0 at t = 0 and hpx i = hpy i = 0,
and hpz i = mωR. Then eq.(4.39) reduces to
~ ∂
(x − R) + −R ψ(x, y, z) = 0 , (4.40)
imω ∂y
34
ie.
1 ∂ψ mω
= i (2R − x) . (4.41)
ψ ∂y ~
Hence h mω i
ψ = exp i (2R − x) y f (x, z). (4.42)
~
where f (x, z) is any arbitrary function of x and z. It is evident that the wave-
function merely oscillates and does not fall off in the y direction. Thus there
does not exist any compact wavepacket which does not spread indefinitely.
It is to be noted however that χR (the uncertainty in the radial component of
position) and πT ( the uncertainty in the tangential component of momentum)
either remain constant or vary utmost periodically. This is because, as may be
seen from the explicit expressions (4.37) for the u(i) , the columns u(1) , u(2) and
u(5) do not contribute to χR and πT and these are the columns which appear
with factors of t or t2 in (4.34). On the other hand u(1) and u(5) contribute to
χT and this quantity increases indefinitely i.e. the wavepacket (even if it is well
defined initially) spreads gradually in the tangential direction.
19 This statement is of course for ‘small’ wavepackets, subject to the approximaton of ne-
glecting third and higher derivatives of V in a Taylor’s series expansion around hxi. In the case
of the harmonic oscillator of course all such derivatives vanish identically and the nonspreading
is a rigorous property.
35
Chapter 5
Appendix
We will now derive the equations of spread (4.20), for a wavepacket moving in
a circular orbit in an arbitrary central potential V (r).
The Hamiltonian for such a system is given by
3
X pi2
H = + V (r) (5.1)
i=1
2m
For our derivation we will have occasion to use the equation of motion (4.2) and
the commutation relations
∂f (x)
[pi , f (x)] = −i~ (5.2)
∂xi
together with the Taylor’s expansion of V (~x), about the ‘centre’ h~xi of the
wavepacket:
3
X ∂V
V (~x) = V (h~xi) + . (xi − hxi i)
i=1
∂xi ~x=h~xi
3 3
∂2V
1 XX
+ . (xi − hxi i) (xj − hxj i)
2! i=1 j=1 ∂xi ∂xj ~x=h~xi
+ ... (5.3)
∂2V
1 ∂V h xi i h xj i ∂ 1 ∂V
Vij ≡ = δij + (5.4b)
∂xi ∂xj ~ xi
x=h~ r ∂r r ∂r r ∂r r=R
Equations (4.20) have been derived for a bound wavepacket moving in a circular
orbit. For such an orbit we have
2
h p~ i = m2 ω 2 R2 (5.5a)
h p~ i . h ~x i = 0 , (5.5b)
36
and
1 ∂V
m ω2 = . (5.5c)
R ∂R
When these results are used, some of the expressions (4.13) and (4.19) simplify
somewhat. For example
1
χR = hxi i hxj i hxi xj i − R2 (5.6a)
R2
1 1
χRT = hp i i hxj i hx x
i j i , (5.6b)
mω R2
1 1
πR = 2 2 hx i i hx j i hp p
i j i , (5.6c)
m ω R2
1 1
πRT = 3 3 hxi i hpj i hpi pj i , (5.6d)
m ω R2
1 1
µRR = hxi i hxj i hpi xj + xi pj i . (5.6e)
mω R2
Here and in the sequel, we use the summation convention with respect to re-
peated indices, which range from 1 to 3.
To derive the equation corresponding to the first row in the matrix equation
(4.20), we note from (5.6a) that
dχR 1 1 d
i~ = 2 hxi i hxj i h[xi xj , H]i + i~ 2 hxi xj i (hxi i hxj i) , (5.7)
dt R R dt
d
where we have used the quantum equation of motion to evaluate dt hxi xj i . On
d
evaluating the commutator and substituting dt hxi i = hp i i/m, we get
1 dχR
= µRR + 2 χRT , (5.8)
ω dt
which corrsponds to the first row in the matrix equation (4.20). This equation
is exact.
We will now derive the equation for πR corresponding to the 4th row in
(4.20), this however involves an approximation depending on the small size of
the wave packet. From (5.6c) we have
dπR 1 hxi i hxj i d hpi pj i d
= 2 2 hpi pj i + {hxi i hxj i}
dt m ω R2 dt R2 dt
1 hxi i hxj i ∂V ∂V
= 2 2 − pj + p i
m ω R2 ∂xi ∂xj
1 hpi pj i
+ 2 2 {hxi i hpj i + hpi i hxj i} , (5.9)
m ω mR2
37
d
where the quantum equation of motion has been used to evaluate dt hpi pj i. To
proceed further we expand ∂V /∂xi in a Taylor’s series about the position h~xi :
∂V
pj = Vi hpj i + Vit h( xt − hxt i) pj i
∂xi
1 ∂V
= hxi i hpj i + {hxt pj i − hxt i hpj i} Vit . (5.10)
R ∂R
Vi and Vit are given in eqs. (5.4). We have dropped the third and higher order
derivatives in the Taylor’s expansion. This is a good approximation for most
potentials of interest provided the size of the wave packet is much less than the
radius of the orbit. From (5.10) we have
hxi i hxj i ∂V hxi i hxj i
− p j = − {hxt pj i − hxt i hpj i} Vit . (5.11)
R2 ∂xi R2
Note that the contribution to (5.11) from the first term in (5.10) is zero because
of (5.5b). Similarly the second term in (5.11) is zero, hence
hxi i hxj i ∂V hxi i hxj i
− pj = − hxt pj i Vit .
R2 ∂xi R2
hxi i hxj i 1 ∂V hxi i hxt i ∂ 1 ∂V
= − hxt pj i δit +
R2 R ∂R R ∂R R ∂R
hxt i hxj i 1 ∂V ∂ 1 ∂V
= − hxt pj i +R
R2 R ∂R ∂R R ∂R
R ∂ ω2
hxt i hxj i 2
= − hxt pj i mω 1 + 2
R2 ω ∂R
mω 2
= hxt i hxj i hxt pj i η (5.12)
R2
where we define
R ∂ ω2
η = − 1+ 2
ω ∂R
r V 00
= − (5.13)
V 0 r=R
Putting (5.12) into (5.9) and using (5.6d) and (5.6e) we have
1 dπR
= 2 πRT + η µRR , (5.14)
ω dt
which is the desired equation. Similarly all the other equations in (4.20) can be
derived.
Before we close the appendix we make a final remark, namely that all the
quantities which are defined in (4.19) and which do not appear to be properly
symmetized are symmetric for circular orbits. Consider for example µT R , eq.
(4.19c) :
1 1
µT R = 2 2 {h~p i · (~
x − h~
x i)} {h~
x i · (~
p − h~
p i)} ,
m ω R2
1 hpi i hxj i
= 2 2 {hxi pj i − hxi i hpj i} (5.15)
m ω R2
38
In the case of circular orbits this reduces to
1 hpi i hxj i
µT R = hxi pj i
m2 ω 2 R2
1 hpi i hxj i
= 2 2 hpj xi + i ~ δij i
m ω R2
1 hpi i hxj i
= 2 2 hpj xi i (5.16)
m ω R2
confirming that the order of factors in the product whose expectation value is
taken is irrelevant.
*********************************************
REFERENCES
39
27. E.H. Kennard : Z. Physik 44, 326 (1927)
28. Louis de Broglie : “Nonlinear Wave Mechanics”, Elsever Publ.,
New York (1960)
29. M. Born and D.J. Hooton : Z. Physik 142, 201 (1955)
30. M. Born and D.J. Hooton : Proc. Camb. Phil. Soc. 52, 287 (1956)
31. M. Born : Z. Physik 153, 372 (1958)
32. G. Petiau : Comptes Rendus 238, 998, 1568 (1954)
33. G. Petiau : Comptes Rendus 239, 344, 792, 1158 (1954)
34. G. Petiau : Comptes Rendus 240, 2491 (1955)
35. F. Karolhazy : Nuovo Cimento 42A, 390 (1966)
36. E. Schrodinger : Z. Physik 14, 664 (1926)
37. R.J. Glauber : Phys. Rev. 131, 2766 (1963)
38. P. Carruthers and M.M. Nieto : Amer. Journ. of Phys. 33, 537 (1965)
39. E.C.G. Sudarshan : Phys. Rev. Lett. 10, 277 (1963)
40. G. Szego : “ Orthonormal Polynomials”, American Mathematical Society
Colloquium Publ. Providence R.I. Vol. 23 (1965)
41. P. Carruthers and M.M. Nieto : Phys. Rev. Lett. 14, 387 (1965)
42. K.E. Cahill : Phys. Rev. 138, B 1566 (1965)
43. K. Eswaran : Nuovo Cimento 70B, 1 (1970)
44. H. Barnard and J.M. Child : “ Higher Algebra” , Macmillan and Co. (1959)
45. E.T. Whittaker and G.N. Watson : “ A Course of Modern Analysis”,
Cambridge Univ. Press, Cambridge (1952) Chapter XXXI.
46. E.C. Titchmarsh : “ Eigenfunction Expansion”, Oxford Univ. Press,
Oxford (1962) Chapter VII, p 144.
47. Pao Lu : Journ. of Chem. Phys. 47, 815 (1967)
48. Yu. E. Murakhver : Soviet Phys. Doklady 10, 1079 (1966).
(Translated from Doklady Akad. Nauk. SSSR 165, 5261 (1965))
49. A. Messiah : “Quantum Mechanics”, North Holland Publ.
Amsterdam (1966) Vol I, Chapter VI.
50. F.R. Gantmacher : “ The Theory of Matrices ”, Chelsea Publ. New York
(1960) Vol I, Chapter VII.
**********************************************
40