HAL Id: jpa-00247513
https://hal.archives-ouvertes.fr/jpa-00247513
Submitted on 1 Jan 1991
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Semiclassical matrix-mechanics. II. Angular momentum operators
Richard More
To cite this version:
Richard More. Semiclassical matrix-mechanics. II. Angular momentum operators. Journal de
Physique II, EDP Sciences, 1991, 1 (2), pp.97-121. �10.1051/jp2:1991150�. �jpa-00247513�
J
Phys.
II1 (1991) 97-121 FtVRIER 1991, PAGE 97Classification
Physics
Abstracts 03 655Semiclassical matrix-mechanics.
II. Angular momentum operators
RJchard M More
Lawrence Liverrnore National
Laboratory
L-321, Livermore, CA 94550, US A(Received
27 June J990,accepted
mfinal
form 29 October J990)Abstract. Semiclassical
angular
momentum matrices are calculated using a new contour-integral
formula for matrix-elements WKBsphencal
harrnonic functions are found to beexactly
orthonormal with thecontour-integral inner-product
Matnx-elements obtained from these wave-functions are accurate to about 29b and the matrices
obey
the cornrnutation relationsexpected
of quantumangular
momentum operators The semiclassical wave-functions are related to asuperposition of allowed classical orbits and this illustrates the connection to the
Feynman path-
summation formulation of quantum mechanics for electrons m a
spherically
syrnrnetncpotential.
1.
Innoducdon.
A
key step
in thedevelopment
Ofquantum
mechanics was the introduction Of matrices torepresent dynamical
Variables such as position x Or momentum p[I]
Matricesappeared
asgeneralizations
Of Founer componentsx(w), p(w)
used m the classical calculation Ofradiation from atoms From a modern point of mew the matrices
X,
P are givenby quantum mechanics,
and the connection with Fouriercomponents
is anapproximate
orlimiting correspondence
with quantities defined m the classicaltheory [2]
Thecorrespondence
is Validin the limit of
large quantum
numbers for transitions with smallchanges
inquantum number,
i e
,
for a small subset of matrix-elements near the
diagonal
One can ask how tointerpret
the rest of thematrix,
Or moreexactly,
how tO relate thegeneral
matnx-element tO a semiclassicalquantity
Recently,
More and Warrenproposed
an answer tO th~s question, a semiclassical formula in which the matnx-element is determinedby
a contourintegral
Over WKB wave-functions[3].
This formula has a
Simple physical interpretation
for radiative transitions Of atomic electrons the transition is associated with a second-Order Orbit intersection which Occurs at auniquely
,defined radius for transitions
having
substantialmatrix-elements,
Such asls-2p, 2s-3p, 2p-3d.
The matrix-element calculated in this way is accurate to a few percent,
quite generally,
forlarge
Or Smallchanges
m thequantum
numbers The method also works for radialintegrals
associated with
quadrupole
transitions and for thephotoelectric
effectWhen
applied
tO the harmonicoscillator,
thecontour-integral
formula g1Ves a surprising result for the first time a semiclassicaltheory
agrees in almost every respect with the98 JOURNAL DE
PHYSIQUE
II M 2quantum treatment
[4].
The semiclassical matrices have the correct commutator[P, X]
=
ihl, diagonahze
the Hamlltoman andexactly obey
the matrixequations
of motion(e
g,[H,
P ocli~.
However the matrices are not Hermitian and the excellent results described holdonly
for matrix-elementsF~,~
with mm n.In this paper the contour
integral
method isapplied
to theangular
momentum of anonrelativistic electron in a
Spherically Symmetric potential.
The mostinteresting
questions will be : How accurate are the semiclassical results ? Do the semiclassical matncesobey
thealgebraic
relations of theHeisenberg
matrix mechanics ? What is thephysical
interpretation of semiclassicalSpherical
harnlonic functions ?Section 2 outlines the
contour-integral
method Semiclassicalsphencal harmonii
functionsare discussed in Section 3
Angular
momentum matncesL~, L~, L~
are calculated in Section 4They
are within 2-3 fb of thequantum
matnces andexactly satisfy
theimportant algebraic
relations :~~X,
~y] ~~=
jj)
and CyChC permUtabons
The semiclassical matrices
L~
andL~
are
exactly
correct howeverL~
andL~
areSlightly
non-Hermitian. Thus the semiclassical
theory
agrees with quantumtheory
for all but onealgebraic property
of theangular
momentum matrices.These
unexpected
results areprobably
related to the fact that the WKBangular
functionsare
exactly
orthonormal when their innerproduct
isflefined by
thecontour-integral
formula(Sect. 4)
Because the semiclassical matrix elements are
analytic
functions of the quantum numbersI,
m it is
possible
to evaluate them withhalf-integer
Values such asI
=
1/2 (Sect. 5). Angular
momentum matnces obtained th~s way are
exactly
the Pauh matricesSemiclassical
angular
functionsYi~(@, 4
can be combined with WKB radial functionsR~i(r)
to make a three-dimensional wave-functionp~i~(r).
This function has a direct geometncinterpretation
related toFeynman's
quantum mechanics : for an electron moving ina
sphencally
symmetncpotential
the semiclassical wavefunction is obtainedby
superposition of allowedclassjcal irbits
of afamily
Such that one orbit passesthrou)h
eachpoint
m Space(Sect 6)."
Orbits of th~s
family
aregenerated by
rotating aSingle
orbitthrough
Variousankles
thephase assigned
to each orbit is determinedby
the rotationangle
Th~s constructio6 g1Ves a detailed connection between classicalpaths,
rotational symmetry and the WKB wave-function.
lihe
overallgoal
ofthis
workis to
develoj
acomplete
semiclassicaltheory
which will describe theentire'range
ofphenomena
encountered in quantummechanics,,as
far aspossible.
Of courseiriany
aspects of the desiredtheory
arealready
known[5, 6].
Insearching
for the rules of
thq,semiclassical theory
we are assistedby
comparisons with thequantum theory
At thesaine
time, the semiclassicaltheory
alsogives
a better intuitiveundersfianding
of
qujntum theory.
2.
Contour-integral
fornlulafion.The usual semiclassical
theory
is based upon thd Bohr-Sommerfeldquantization
rule for one- dimensional motion,ip~ ix)
dx=
2 ark
(n
+ '(2)
c~ 2
M 2 SEMICLASSICAL MATRIX-M.ECHANICS II 99
where p~
ix)
= momentum oc
[E~ U(x)
"~ withE~
= energy
eigenvalue
,
x is a
coordinate,
allowed to be acomplex number,
andU(x)
is thepotential,
assumed -to be ananalytic
function of x Thepositive
andnegative
signs of p~ix) correspond
to the two directions of motion in the ± x-directionmathematically
p~ix)
has a branch-cutalong
the real x-ax~s in theclassically
allowed range whereE~
mU(x) Assuming
this allowed range is asingle interval,
x~~~ w x wx~~~, p~ix)
will bepositive
and realjust
above the branch-cut and the contourC~
encircles the allowed range once in the clockwise sense. With theseconventions,
the Bohr-Sommerfeld rule is a contourintegral
The WKB
travelhng
wave>
~ ± p~(x') dx'/h ± iw/4
~~
~~~fi
~ '~ ~~~is
typically
ananalytic
function ofx'ha~ing
a branch-cut
along
the allowed rangex~~~~ w x w x~~~~. Both p~
ix), pi ix)
may besipgular
atthe.singularitips
ofU(x)
Theprefacior
I/ fi
and(he
action pix')
dx' have additionalbranch-cuts,
but these cancel in thewave-function because of
equktion (2) Specific analytic properties
, of the WKBspherical
harmonic functions are summanzed m
appendix
B.One can construct a real wave-function
p~ ix)
defined for real xby
the equation~niX)
=
lllll Re
~t IX
+ tEl
+~i IX
+ IE)1 14)
This function oscillates in the allowed range x~~~~ w x wx~~~ and has the correct bounded behavior in the forbidden ranges It g1Ves a
good approximation
to the quantum wave-function aside from
(weak) singularities
at theturning
points x~~~~, x~~~~Equation (4)
can be used mplace
of the usual connection formulas.The matrix-element of a
dynamical
Variablefix, p)
will be calculatedby
the contour-integral for~nula [3]
Fnm
"j
~i IX) / IX, > ~i IX)
dX15)
where C encloses the allowed range for both states n, m in the clockwise sense Here
fix, fl)
is thequantum
operator expression, but equation(5)
is semiclassical because the WKB wave-functions areexplicit
combinations of classical momentum and action functions.Although equation (5)
is s1mllar to thequantum
formula for a matr~x element there are alsospecific
differences The quantum formula is not a contourintegral
and contains additionaltennis associated with the
integrals
ofp( fp(, pi fpj.
The additional tennis causeoscillatory (interference)
modulation of theprobability density
and their omission can be called a restrictedinterference
approximation InAppendix
A it is shown thatequation (5)
converges with the quantum expression in the semiclassical limit where the
oscillatory integrals
become small The bestagreement
with thequantum theory
is obtained when thep~
wave is taken for the statehaving
the smaller allowed range of classical motion.Equation (5)
g1Vessatisfactory
results for matrix-elements of several systems[3, 4]
and in section 4 we will see that it succeeds again for thespherical
harnlomc functions. An extension to three-dimensional systems is indicated m section 6(see especially Eq (78) below).
Temporarily adopting
equation(5),
One can ask whether it fOrn1s the basis for a consistent100 JOURNAL DE
PHYSIQUE
II M 2semiclassical
theory
Forexample,
for thespecial
casef
=
I,
equation(5) implies
a contour-integral inner-product
of semiclassicalwave-functions,
Un,
m "
j ~i IX) ~i IX)
dX16)
In part I of th~s paper, equation
(6)
was evaluated with WKB wave-functions for the harnlonic oscillator with theunexpected
result U~~>= 3~~> for all states n'm n For radial motion in the Coulomb
potential
and also in a screened Coulombpotential,
it was found thatintegrals U()
were 3~~. + O
(10~
~[3].
With the usual inner
product,
WKB wave-functions are not evenapproximately
orthonor- mal and therefore these results are evidence that equations(5), (6)
cangive
a moresatisfactory
semiclassicaltheory
For the WKB
spherical
harnlonicfunctions,
theorthonornlahty
relation is exact,U(j~
=
Sit.
for alli'm I (see Eq (30) below)
U~~ =
corresponds
tonornlalizing
the wave-function in tennis of the classicalround-tnp travel-time,
this is a well-known normalization of WKB functions[6]
However thisnornlahzation agrees with the quantum
normalization, p~
~ dx =I, only
in the semiclassi- cal limitAnother
special
case of equation(5)
is n= m, the semiclassical expectation Value
In
"
j ~i /~t
dX17)
When the
quantity f
is a functionfix),
this issimply
~ j2 j /lx)
~
j8)
n " ~~
w
~'The
right-hand
side is the classical time-averageoff,
which isusually
agood
approximation to the quantumexpectation-Value
even for lowquantum-numbers.
Forexample,
results from equation(8)
areexactly
correct for the expectation Values(I/r), (1/r~)
for electrons in aCoulomb
potential [7]
We now compare equation
(5)
to theHeisenberg correspondence principle
In the limit oflarge
andnearly-equ41
quantumnumbers,
equation(5)
gives for the casef
=
fix)
d~ '(En E~)
$
dX'IAFnm
" ~ ~j / IX)
eP X
=
la1~ j $ fix)
e- ~"~~f~ cc
flxlt))
e~ ~"~ dt19)
Here
quantities
without asubscnpt
are evaluated on an average orbit(e.g., p(x)m
p~
ix)
m p~
ix)
forlarge
andnearly-equal
n,m).
This is a well-known calculation[6]
In order to denve equation(9)
it is necessary to omit theoscillatory
termsp( fp(, pi fpj
and theseare
already
omitted from equation(5)
soF~~ automatically
reduces to the Founer componentoff
for the casein
m « n. Ingeneral
whenin
m is notsmall,
one cannotspeak
of Founercomponents
because the two orbits are too different to beapproximated by
asingle
space-time trajectory.
M 2 SEMICLASSICAL MATRIX-MECHANICS II 101
Th~s is
especially
evident m calculations of thephotoelectric
matrix-element where one orbit is bound and the other extends out to infinite distance so there is nomeaningful
averagetrajectory Equation (5)
still givesgood
results for these matrix-elements[3]
Thus equation
(5)
does notdisagree
with theHeisenberg correspondence relation,
but rathergeneralizes
itby
giving a semiclassicalapproximation
to the entire matnx off.
Equation (5)
for the matrix-elements is a step toward acomplete
semiclassicaltheory,
but manyquestions'remain
concerning the interpretation and thegeneralization
to three-dimensional systems
and/or time-dependent
atomicdynamics
With someeffort,
one canemploy
equations(5)-(8)
to denve solutions for certainspecific
three-dimensional and time-dependent systems.
A freshapproach
to the moregeneral
case is givenby
the orbitsuperposition
method discussed in section 6.3.
Spherical
harmonic functions.Semiclassical
sphencal
harmonic functions areeasily
constructedby
the WKB method[6, 8, 9]
We sketch th~s calculation in order to draw attention to a fewimportant
technical points.The associated
Legendre
equation may be written~
ill -x~)~~j
+~A- '~~
&= 0.(10)
dx dx
1-x~
In
quantum mechanics,
this equation arisesby
separation of variables and theeigenvalue
A is shared with the radial equation. The quantumeigenvalue
is A=
iii +1),
whereI
is the
(integer) angular
momentum However in the semiclassicaltheory
it is normal to assumeA
=
(I
+ ~ill)
2
where
I
is an integer. With this
choice, tie
WKB wave-functions&i~
turn out to besingle-
valued
analytic functions,
a stnctrequirement
for themethod
of this paper The same separation constant, equation(11),
isnormally
used in the WKB solution of the radialequation [3, 6-9]. Surprinsingly,
the semiclassical matrixL~
has thequantum eigenvalues iii
+1) despite
the use ofequation ill)
m the wave functions(Sect 4)
For now we considerpositive integer
m(0
< m wI).
For a solution & of the form &= exp
Ii
aix) ],
equation(10)
becomes
~~
~~~~[i j~~ ~~~~ j~~1~~2
~~~ ~~ ~~~i ~'
~~~~For
large
quantum numbers theright-hand
side isrelatively
small when it is omitted one has the zero-order equation for «:~~ ~~~
~~ ~"
(~+~ )~_ m2
~ l
x2
The
right-hand
side of equation(12)
isequated
to the first-order term on the left to give«~i~ =
~ log
2ill x~) ~~°
dX(14)
Thls is the usual WKB
technique.
Planck's constant does not appear in equations
(10)-(14).
Inphysical
terms the expansionparameter
is Ilit
+1/2)
and therefore it is necessary tocarefully inspect
the results for small102 JOURNAL DE
PHYSIQUE
II N 2I Mathematically
the expansion is notuniformly,valid
because ofsingular
behavior atx = ± I ai~d at the zeroes of
d«o/dx.
The zeroes of
d«o/dx
areturning
points, located at x = ± J~,,
~2
~ i 2' ~~~~
~
~
The classical
motiin
is
confinid
to theregion xi
=
(cos ii
w xo, that is, excluded
from~a
cone around the
iz-axis.
Thephysical significance
of th~s restriction becomes deaf in section 6.It is useful to wnte
q(x)
=
/~ (16)
With this notation equation
(13)
becomes~°
=
(i
+~~~~~. (17)
X I x
Equation (17)
isexplicitly
solved m tennis of functionsfix),
~ix)
definedby
sin
f
=
~
cos
f
=
~~~~
(18)
J~ J~
Sin ~ =
~
@
CDS ~ =
jh II 9)
xo i-x J~
i-x~
The definitions are appropriate because the trigonometric relations
sin~ @'+ cos~
=1 areobdyed
fora(I
x. Theboundary
values off,
~ areii± xo)
= ~
i± xo)
= ±
]. 120)
The derivatives are
~~ q/~)
~~) [
ji
~) ~X)
~~~~
For either sign of m it follows that
«oix)
=
(f
+fix) im1
~ix) 122)
At this point, equation
(22)
appears to be a mathematical relationhaving
little intuitive content. In section 6 we will see that it has asimple
geometncinterpretation
in effect it is thetransformation of a
simple
waveexp(i (I
+1/2)
y defined in a rotated coordinate system.We combine equations
(14)-(22)
to define the semiclassicaltravelling-wave
for allm(- I
w mmI by
~+ ~~~ aim
qix)
+ ix±
l~
+2q
ix)
ix/fi
±'~'~~
fi
~~~~,' xo
~
fi2
M 2 SEMICLASSICAL MATRIX-MECHANICS II 103
The
nght-hand
sidedepends
on the sign of monly through
the coefficient ci~, which is taken to be~fm "
) e~~~~+m)
w/2~ "
(24) Equations (23), (24) imply
At
m
lx)
=)~ et
m
lx) 125)
The
magnitude
of ci~ is chosen to normalize the functions with the innerproduct
ofequation (6).
Thephase
agrees with a well-known definition of thespherical
harmonic functions[10, 11].
The formulas
simplify
for m=
0 The appropriate
special
case ofequation (23)
may be written~Xp ±1 + ~ ~
~~
~~~~ ~~ ~~~~~~ ~~
j2
i'+~
sin
~~~~
2
These are
travelhng
waves to compare to theLegendre polynomial
we fornl.Pf
=
Re
[Pi
+Pi ]. (27)
Comparisons
are given infigure
,
the
agreement
isgood
away from thetu£rung
points whichoccur at x = ± I
Equation (27)
is a knownasymptotic
fornlula for theLegendre
function[12].
..
The WKB approximation to the
spherical
harnlonic function is obtained froml'lLl@, 4
= Re18im
+8iml ~Pml4
~im4=
Re
18im
+8iml /~
2128)
gr
Numerical compansons show that
equation (28)
is agood approximation
apart from thesingularity
at theturning points
± J~ Asample
comparison is given in table1Table I
Comparison of
WKB and exactSpherical
Harmonicsfor
cos@=005,
4
=0f
m WKB Exact Difference0 0 0.3184 0.2821 12 8 fb
0 0 0239 0.0244 2 5 fb
0.3685 0.3451 6 8 fb
2 0 0.3160 0.3130 0 fb
2 0 0381 0.0386 7 fb
2 2 0 4105 0.3853 6 5 fb
4. Matrix-elements and
matrix-algebra.
In this section the
contour-integral
formula is used to calculate the innerproduct
Of WKB.wave-functions and matrix-elements of x=cos@
Angular
momentum matricesL~,
L~, L~
are constructed and theiralgebraic properties
examined104 JOURNAL DE PHYSIQUE II, M 2
s
=
i o
Legendre function
Exaci Legendre function Po o exact
WKB
>,
os os
a H
W W
8 0 8 0
d <
.0 s -o s
.i .i o
'~
.08 -06 -04 .02_ 0 02 04 06 0810
'~~10
.08 -06 .04--02 0 02 04 06 0810cos8 cos0
a)
b)s S
=2
y
= 3
Legendre function
Legendre function
i o exact
exact
WKB WKB
os
~
m #
8 0 8 0
$ $
~ ~
.os .os
-i o -1
.15 .1
-10 -08 -06 -04 .02 0 02 04 06 08 10 -10 .08 -06 -04,02 0 02 04 06 08 10
cos0 cos0
c) d)
Fig I Companson of quantum and semiclassical Legendre functions for several angular momenta
The semiclassical functions
~'J$
areobviously
orthonormal withrespect
to the index m because the#-dependence
is asimple
factorexp(im 4 )
as m quantum mechanics Thus it isonly
necessary to calculate the innerproduct
of functionshaving
the same mUi[~
=
Aim lx) elm lx)
dx129)
From
equation (23)
we see that the factor(I -x~)
appears with powers(m(,
+
[m
in8+,
fJ~respectively
and therefore cancels inequation (29)
This means that themtegrand
isanalytic except
for branch-cutsalong
the allowed ranges(-xo, xJ)
andN 2 SEMICLASSICAL MATRIX-MECHANICS II 105
( go, go),
where xJ,go
are turningpoints
for quantum numbersif, m)
andif',
m).
Inparticular
theintegrand
ofequation (29)
has nobranch-points
at x = ±1 Therefore the contour ofintegration
may bedisplaced
tolarge ix
and theintegral
iseasily
evaluatedby residues,
giving :U)j~
=
an,
for allf'
m
I (30)
The
diagonal
valueiU)Q)
isunity
because of the normalization chosen inequation (24)
The surpnse is theorthogonality
for all pairsf, I'
withf
<
I'
The calculation can berepeated
for thespecial
case m=
0 where the
simpler
functional fornl ofequation (26)
gives anelementary
tngonometncintegral
lpi lx) PI lx)
dx= ~
~ 3fi 131)
This is
exactly
theorthonornlallty
relation forLegendre polynomials
It would be difficult to obtainanything
like this result for WKB functions without use of the contourintegral
innerproduct.
Next we examine the matrix-element of x
= cos between states
ii, m)
andii', m).
Contour
integration
shows this matnx-element is zero forI=I'
and for alli'm
I+1
Thls means the selection rules agreeexactly
with quantum mechanics. ForI'= I
+ I we find~ ~
i
go
~ ~' +/fi
~'~~~~
~~~~
~ ~~
'~ l +
+/I Xl
~~~~Table II shows that this agrees
closely
with thequantum qatrix-element
even for samllquantum
numbersNext we fornlmatnx-elements of the
angular-momentum operators L~ L~, L~, using
the usualrepresentations
i~
= -1
~
d4
I+
=
i~
±i~
=
e~~~ (± I
+ i cotI (33)
°@
°4
Table II Test
of
equation(32) for
the matrix-elementsof
x= cos
QM
and SC denotequantum and semiclassical tiatrix-elements
I,
mI
+ 1, m
QM
SC difference1,0 2,0
0.5164 0.5000 3 2 fb1,1 2,1
0.4472 0.4269 4.6 fb2,0 3,0
0 5071 0.5000 4 fb2,1 3,1
0 4781 0 4705 6 fb2,2 3,2
0.3780 0 3657 3 2 fb3,0 4,0
0.5040 0.5000 0 8 fb3,1 4,1
0 4880 0 4839 0 8 fb3,2 4,2
0 4364 0.4316 fb3,3 4,3
0 3333 0 3240 2 8 fb106 JOURNAL DE PHYSIQUE II N 2
Substitutihg
x,= cos andperforming
the integration over theangle 4,
we -have(I"~~
+l~(iS i''~~)
~j
8im+1~~ ~)~ fi~ Aim
d~(34)
l -X
For m m
0,
we use the wave8j~
~ i which has the smaller allowed range in
agreement
with the ruledeveloped
inAppendix
A For m<
0, et,
~
has the smaller allowed range, so we take
8j~
with8(~
~i
To calculate matrix-elements of L_ we fornl the Hermltian
adjoint
ofL~
In
equation (35)
the waves chosen are8j~
and8(~~i
for mm 0 and vice versa form'<
0 In this case thei-
wave is not associated with the shorterbranch-cut,
this agreeswith the
procedure
followedin'part
I of this paper and leads tointeresting
but non-Hertnitian matrices.Calculation of these matnx-elements is a
straightforward
contourintegration.
The contour may bedisplaced
tolarge xi
because the factors'il x~)
cancel and one mustsiniply
extract the rdsidue of the
integrand
atxi
- oJ.
The results are
simplified by defining
~(*(~'~' ji
~~
fi)jm<
~
~f+I-<m+i<
11 +/fi)'~+~'
~~~~Then we have
,
~
IL
+
)'. z
+ '=
~~ji '~~1~[l~ i i
>.
137)
j I
~i, mIf
m)/r
m » °j38)
~'~
±' if
+ m + I r m
< 0 where x,
go
are defined with respect to the pairsii,
m
)
andii,
m + I
)
It is useful to write out these matrices for
I
= 1, 2
(Tabs III, IV)
Table III Semiclassical
angular
momentum matricesfor
L=
0.000 0 719 0 000
L~
= 0.695 0 000 0 719
0.000 0 695 0 000
0.000 0 7191 0.000
L~
=0 6951 0 000 0.7191
0.000 0 6951 0.000
000 0 000 0.000
L~ = 0.000 0 000 0 000
o coo o.o00 1.000
N 2 SEMICLASSICAL MATRIX-MECHANICS II 107
Table IV. Semiclassical
angular
momentum matricesfor
L=
2
o coo 1.022 o coo o coo o coo
0 979 0 000 .1 228 o 000 o.oo0
L~
= 0 coo 1.221 0.000 228 o 0000 000
0.00i.
221 o 000 0220 000 0.000 o 000 0 979 o coo
.000
0221 0.000 o.000 0.000o.9791 o 000 1.2281 o.000 0.000
L~
=0.000 221t o.000 228t o.000
0.o00 o coo 1.221 0 oo0 1.0221
0 000 coo 0 oo0 o 979t 0 000
-
2 000 o coo o coo o coo o cooo coo o00 o coo o coo o coo
L= = o 000 0.000 o 000 0 000 o coo
0.000 0 000 0.000 1.000 0.000
0.000 0 000 0.000 0 000 2.000
The most important feature of these matnces is their overall accuracy each entry is within 2 fb of the
corresponding quantum
value However the matricesL~ L~
arenon-Hermttianj
atypical
result of calculations based onequation 15).
The matnces for
L~, L~
have theproperty
that if theproduct
is fornled between matrix-elements on
opposite
sides of thediagonal,
the result is the square of thequantum
matrix- element. This symmetryproperty
holds also for theharnloniq
oscillator[4]
Given these sen~iclassical matrices
L~, L~, L~
an immediate question is whetherthey obey algebraic
conditions whichparallel
the laws of theHeisenberg
matrix mechanicsThe matrices given in tables
III,
IVexactly satisfy
thefollowing algebraic
relations :[L~, L~]
=tL~, [L~, L=]
=iL~, [L=,L~]
=
tL~ (39)
L~
= ma ~~.140)
L~
=
Lj
+L(
+Lj
=
iii
+1) 3~~> 141)
These are the
algebraic
relationsobeyed by
thequantum angular
momentum matrices.Equation (41)
isespecially
surpnsing in view of equation(11)
Thus the
contour-integral
semiclassicaltheory reproduces
the main results of the quantumtreatment. The matrix-elements of x=cos@
obey
the quantum selection-rules. Thesemiclassical matrices L have the correct
eigenvalues, exactly obey
the commutation relations but are non-Hernlitian and inexactby
a few percent.Since the situation
closely parallels
that found for the harmonic oscillator onemight
suspectthere is a
general proof
of thesealgebraic relations, perhaps
based on the classical Poisson brackets We have not found such aproof
and m fact there is a sort of counterexample provided by
the radial matrix-elements for the Coulombpotential,
whichonly approximately satisfy
the desiredalgebraic
relationsPerhaps
the mostgeneral
conclusion is that semiclassicalresults are
good
approximations containing most features of the quantum picture.108 JOURNAL DE
PHYSIQUE
II N 25.
HaW-integer spin.
The wave-function defined
by
equation(23)
is asingle-valued
function of x whenI
andm are
integers
Thls may be testedby following
thechanges
m8f~
when x varies around a contour that just surrounds the allowed rangei
xo,xJ)
Surprisingly, fJf~
is alsosingle-valued when'i =1/2,
m=
±1/2
Thls means that thefunctions
fJf~
can be used to calculate matrix-elements for aspin-1/2
systemFor this calculation we substitute xJ
=
go
=@
in equations
(36)-(38).
The results are~ l
jo
I~~ 2 0
L~
= ~(42)
~ l
j-
I o~~2
oDespite
this nice result the calculation does not give an intuitive picture or mechanical model for spinphenomena
That is morelikely
to be foundby
the method of section 6.6.
Superposition
of classicalpaths.
This ~section examines the meaning of WKB wave-functions and the connection between classical
paths
and quantum eigenstates The discussion assembles a number ofelementary
orknown results
[9, 13-15]
andinterprets-
them in terms of thesuperposition
of classicalpaths [16]
The
emphasis
here is on the construction of semiclassical wave-functions rather thaneigenvalue spectra
orspace-time
Green's functionsII 7, 18, 19]
The constructionincorporates
the EBK quantization rule[14, 15]
There are three ways to obtain the three-dimensional semiclassical wave-function
first,
it is aproduct
of one-dimensional WKB functions for theseparate
coordinatesir,
@,4 )
second,
it is the result ofrotating
asingle
classical orbit togenerate
aspace-filling family
of copies of theorbit,
with each copyassigned
aphase
deternlinedby
the rotationangles
and third it is constructedby
invariant vector-integration of the momentum fieldpir
definedby
the
family
of orbits.For an electron moving m a
sphencally symmetric
atomicpotential Vir)
with energyE~i,
theproduct waye-function
is~nfm i~ )
"
I R>
i~) 8tm iC°S
~ 1°m
i ) 143)
The radial wave-function
R~i
is~
a~f ±i p~1(r') dr'/h ±iw/4
~~
~~~fi
~ ~~~~The radial momentum
p~iir)
= r
pi (r
isp~iir)
=
fi jE~i
+evir)
2~~~~ (45)
mr
N 2 SEMICLASSICAL MATRIX-MECHANICS II 109
The
angular
wave-function is taken fromequation (28)
of section 3 Theangular
momentum quantum numbers areL
=
(I+ (46)
2
L~
= m(47)
The
phase
of theproduct
wave-function in equation(43)
is(S"(lpnfl~)d~+ lLf~Lz~)+Lz4 148)
The two tennis in
parenthesis
come fromequation (22)
for thephase
of8(~. Analysis
of theclassical orbits will now make clear the meaning of equation
(48)
Until the work of
Feynman [16]
it waswidely
believed that one could not understand quantum states in tennis of classicalorbits, ispecially
states of smallquantum
numbers such as theIs, 2s, 2p
statesIndeed,
there are several important differences between classical motionalong
an orbit and thequantum
statehaving
the same energy andangular
momentum. These differences include some of the more difficult or unique features of the quantumtheory
I)
The classical motion is localizedalong
a trajectory, r = roit ), leading
to atime-averaged probability density
of the fornlp
jr
oc 3jr
roit)
dt149)
Thls function is nonzero
along~a
curve or at most aplanar
surface If pir)
wereregarded
asthe square of some
wave-function,
m an effort to reconcile classical and quantumdescnptions,
that wave-function would havelarge (or infinite)
derivativesperpendicular
to the classicaltrajectory,
and therefore alarge (or infinite)
kinetic energycomputed by
the quantum formula This is quite unlike the quantum state, which gives asmoothly
varyingprobability density
and a kinetic energy close to the classical valueii)
The classical electron isaccelerated, especially
when it passes close to the nucleus, and this leads to afluctuating
ortime-dependent
electric current However the current vectorcalculated from a quantum energy
eigenstate
istime-independent
iii)
While the classical Orbit hassharply
defined values for all threecomponents
of theangular
momentum vectorL,
thequantum
state does not,typically only L~
andL~
havesharply
defined valuesThese
important
differences are reconciled if thequantum
state iscompared
to asuperposition of classical
particles
on afamily
of classicalpaths
Thefamily
F consists of allowed orbits such that one orbit traverses each point r m the regionpermitted by
the conservationlaws,
and thesuperposition
includes copies of the electron distributedalong
each orbit
(see Fig 4).
i)
Because orbits of thefamily
passthrough
each point m thepermitted volume,
there is aprobability
for a copy of the electron to be at any position and hence a smoothprobab1llty density
As will be seenbelow,
the quantum and classical kinetic energies arenearly equal.
ii)
The current vector associated with thesuperposition
of classicalparticles
is time-independent
because thecopies
of the electron flow m a smooth and laminar fashion around their orbits. As one copy moves away from thenucleus,
anotherapproaches
and there isonly
a
steady
average currentjir).
l10 JOURNAL DE PHYSIQUE II M 2
'
, I
~
Fig 2 An allowed classical orbit located m the x-y
plane
The point of largest'radius is located by theangle
a Theangle
yit)
locates the current point on the orbit The orbit is taken to be anellipse (corresponding
to the Coulomb potential) forsimplicity
but the results apply to any central-forcemotion
,
z
'i
Fig
angle p
M 2 SEMICLASSICAL MATRIX-MECHANICS II ii1
z
X
Fig
4 A one-parameterfamily
of orbits defined by varying a for constant go and fl Thecomplete
two-parameterfamily
is constructed by allowing fl to vary Orbits of the family passthrough
each pointm the
spatial
regJon boundedby
twospheres
and a cone around the z-axis Thlsfamily
of orbits generates the senudassicaleigenfunction
of equation (43)iii)
Thefamily F(E,
L ~,L~)
fills space, i-e-, one orbit of thefamily
traverses eachpoint
r inthe
permitted
region. However one cannot not construct aspace-filling
superposition ofpa~hs
without
allowing
twocomponents
of L to vary(see
below nearEq. (76))
The
possibility
ofconstructing
a precise picture of thistype
is best seenby working through
some
examples
Wethereby
obtainsimple
semiclassical pictures for various knownsolution>
of
Schroedinger's
equation,together
with intuitive denvations of many formulas of quantumtheory
In this paper we consider the bound states of an electron m a
sphencally syrnmetnc potential.
First we construct afamily F(E, L~, L~)
of classical orbits allhaving
the sameenergy
E~i(< o),
totalangular
momentum L=
IL (,
and z-componentL~.
The
family
is notuniquely
determinedby
the classical mechanics in this case, because of thedegeneracy
of thesphencally
symrnetnc systemThe discussion and
figures
refer toelliptic
orbits in a Coulombpotential (Figs 2, 3)
Thlscase is chosen for
simplicity
of presentation but the method is not limitedto'that casel However,
the Coulomb system has additionaldegeneracy (or symmetry)
whichplays
noesiential
rolem the
following
discussionl12 JOURNAL DE
PHYSIQUE
II N 2We
begin
with an orbit m the x-yplane,
withangular
momentum vectoralong
z, and rotated in the x-yplane
so that the point oflargest
radius r~~~ occurs at anangle
a from the x-ax1s The
angle
yit)
locates the current point on theorbit,
relative to the x-ax~s The orbit is givenby (Fig. 2)
x = r y a cos y
y = r
iy
a sin yz =
o
(50)
The function
riy
a is obtainedby
solution of the classical equations of motion, as is thetime-dependence
of yit)
For the Coulombpotential,
r =
i
l~llijl~~
n
isi)
(~
=~~~ (52)
t mr
Here r~ and
e are constant for orbits of the
family
FEquation (52) applies
for any central- forcepotential.
We rotate the orbit
by
anangle
Ho around the x-axis, lo that theangular
momentum vector becomesL
=
10,
L sino, L cos
o
(53)
and then rotate
by
anangle p
around the z-axis, giving the finalangular
momentum vectorL
=
iL
sin Ho sinp,
L sin o cosp,
L coso) (54)
The
angle
Ho is constant,r =
rojE,
L ~,Lz,
a,p,
yit)) j56)
The set of orbits defined
by
varyinga and
p
define a two parameterfamily
of orbits,
together
with the
time-dependence
ofyit)
thisfamily
covers thepermitted
region of three- dimensional space. Tl~is is venfiedby forming
the Jacobianderivative,
~ii~iii~~
=Ii t
x
II
=
r2t
cos y sin~o 157)
This Jacobian is nonzero
except
at isolatedpoints Equation (56)
can be written inspherical polar
coordinatesir,
@,4
with the resultr =
rjy
aj58)
cos = sin y sin
Ho (59)
sin
e*~~
=
icos
y ± i sin y cosHo) e*'~ (60)
The momentum p is
~
dro PIE,
L,
Lz,
a,fl,
YIt))
= m