HAL Id: jpa-00212352
https://hal.archives-ouvertes.fr/jpa-00212352
Submitted on 1 Jan 1990
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. I. The harmonic
oscillator
R.M. More
To cite this version:
Semiclassical matrix mechanics
I. The harmonic oscillator
R. M. More
Lawrence Livermore National
Laboratory,
P.O. Box808, Livermore, California, 94550,
U.S.A.(Reçu
le Iljuillet
1989,accepté
le 25septembre 1989)
Résumé. 2014 Les matrices des variables
dynamiques
d’un oscillateurharmonique
à 1 dimensionsont calculées par une nouvelle méthode
semi-classique
formulée par More et Warren. Les éléments de matricesemi-classiques
obtenusXn, m, Pn, m :
i)
obéissent exactement auxrègles
de sélections pour les éléments de matrice nuls ;ii)
sont trèsprécis
(environ
1%)
pour les éléments de matrice non nuls ;iii)
donnent un Hamiltonien sous formediagonale ;
iv)
satisfont exactementles relations de commutation de
Heisenberg
et leséquations
d’évolution. Cespropriétés
sontvérifiées pour les éléments de matrice
Fn, m
avec m ~ n. Les résultats montrentqu’une grande
partie
de laPhysique Quantique
est contenue dans cette théoriesemi-classique
étendue. Abstract. 2014 Matrices ofdynamical
variables of the one-dimensional harmonic oscillator arecalculated
by
a new semiclassical method formulatedby
More and Warren. The results aresemiclassical matrices
Xn, m, Pn, m,
etc., whichi)
exactly obey
selection rules for zero matrix-elements,ii)
are very accurate, circa 1 %, for nonzero matrix-elements,iii)
whichdiagonalize
the Hamiltonian H, andiv)
whichexactly
satisfy
theHeisenberg
commutation relations andequations
of motion. These statements are true for matrix-elementsFn, m
with m ~ n. The resultsshow that most of the quantum
physics
is contained in the extended semiclassicaltheory.
ClassificationPhysics
Abstracts 03.65S1. Introduction.
Although
quantum
mechanics is acomprehensive
self-containedtheory
for atomic events,many
physicists
will agree there remainssomething mysterious
about thecorrespondence
between classical andquantum
theories : how does the classical behavior oflarge objects
emerge out of thedynamics
of aquantum
microworld ? Interest in this fundamentalquestion
combined with a search for bettercomputational techniques
forapplied
atomicphysics
has drawn our attention to thepossibility
ofimproving
the semiclassicaltheory.
To this end we have
explored
the calculation ofquantities
which arise in theHeisenberg
andFeynman
formulations ofquantum
mechanicsusing
the WKBapproximation
to theSchroedinger
wave-functions. In this paper we consider asimple specific
system,
theone-dimensional harmonic
oscillator,
and form the semiclassical matrix-elementsfollowing
a newcalculational scheme based on a
contour-integral
innerproduct
of WKB wave-functions[1-4].
The results are
surprising.
The matrix-elements we calculate with WKB wave-functions are in excellent
( ~
1%)
orexact
agreement
withquantum
mechanics forhalf
of the matricesrepresenting dynamical
variables such as
X, P,
X2, P 2,
and the Hamiltonian. Theagreement
isespecially
visiblebecause many matrix-elements are
exactly
zero in both WKB and exact theories.Matrix-elements on the
opposite
side of thediagonal
can bebadly
incorrect,
and so the matrices arenot Hermitian. It is as if the semiclassical
theory
ofabsorption
is very close toquantum
mechanics,
including
all rates and selectionrules,
but the results for emission are incorrect. Ofcourse, it is easy to correct the situation with the
principle
of detailed balance.We find that the semiclassical matrices of X and P
obey algebraic
relations similar to those of theHeisenberg
quantum mechanics ;
they
have the correct commutator[P, X],
they
exactly diagonalize
the Hamiltonian andobey
theHeisenberg equations
of motion for thegood
half of thesematrices,
associated withabsorption
transitions.After
seeing
theseresults,
which show that most of thequantum
physics
isalready
present
in the semiclassicaltheory,
the reader may wonder where is theboundary
betweenclassical,
quantum
and semiclassicaldescriptions
of the world[4, 5].
We do not propose any
rigid
definition orterminology ;
our paper dealsmainly
inequations
which will
speak
for themselves. Nevertheless we canidentify
three hallmarks of asemiclassical
theory :
i)
semiclassicalequations
should contain classicalingredients,
such asposition, velocity,
action, etc. ;
ii)
however,
theseingredients
are combined with the use ofquantum
ideas ofwavelength,
interference,
superposition,
etc. Forexample,
thedeBroglie wavelength
and Einstein-Bohrfrequency
relation arepart
of the semiclassicaltheory ;
iii)
theequations
are at leastpartially
derived fromquantum
mechanicsby taking
the formal limit in which Planck’s constant becomes small.To obtain a coherent structure at the semiclassical level we add another
ingredient,
which will be madeprecise
in sections 2 and 3 :iv)
the semiclassical wave-functions areanalytic
functions of the coordinates and everysingularity
has aphysical interpretation.
The matrix-elements are contourintegrals
of thesefunctions.
The form of these contour
integrals
isgiven by
aspecific
semiclassical rule ofcalculation,
which we call the restricted
interference approximation,
derived from thequantum
formulasby
systematically neglecting
interference contributions which oscillaterapidly
in the limit oflarge
quantum
numbers.(See
Eqs.
(17), (18) below).
For electrons in a Coulomb
potential,
More and Warren[1]
findreasonably
accuratematrix-elements for
dipole
andquadrupole
transitions,
including
photoelectric
processes andeven
including
line transitions between states with lowquantum-numbers,
such as ls -->2p,
using
WKB wave-functions and the restricted interferenceapproximation.
The same methodalso succeeds for other
potentials including
theDebye-screened
Coulombpotential
and the self-consistent fieldsproduced
by
bound-electronscreening
inmultiply-charged
ions[2, 3].
In references[1-4]
the matrix-elements are evaluatedby saddle-point integration
and this is asimple,
convenientapproximation
with apleasing
intuitiveinterpretation.
However for theharmonic oscillator the matrix-elements can be calculated
by
contour deformation and the results are obtained in closed form with noambiguity
from numericalapproximations.
Theseclosed-form results reveal that the semiclassical matrices are not
merely
good
approximations
but in fact
exactly satisfy
theexpected
commutation rules(for
half thematrix,
always).
The results obtained have abearing
on thequestion
ofcorrespondence
identities,
thesystems.
For theoscillator,
the semiclassical average(X2)
is exact, while(X4)
is not ; we will see this fact emerge from the structure of the semiclassical matrix for x itself.These identities are limited to the oscillator which has many
unique properties
not to be found in moregeneral
mechanicalsystems,
and the fundamentalquestion
is to determinewhich statements are valid for more
general potentials.
Thisquestion
is best addressedby
detailedstudy
of furtherspecific
cases.What is the
practical
value of this sort oftheory ?
We do not have animpressive application
for the harmonic oscillator because the
quantum
theory
itself is sostraightforward.
However,
insofar as the method works for more
complicated
quantum systems,
it is clear that one candevelop
a newcomputational technique
which issubstantially simpler
and more convenientthan the full
quantum
theory,
at theprice
of a minor(10 %)
loss of accuracy. Therefore weexpect
to see a wide range ofapplications
inatomic, molecular,
plasma
and solid-statephysics.
2. WKB wave-functions.
In this section we
develop
theanalytic properties
of the WKB wave-functions. It may be assumed that the reader is familiar with thequantum
solution asgiven by
Schiff[6]
or Landau and Lifshitz[7].
Thephysical
system
is taken to be an electronmoving
in a one-dimensionalpotential
with oscillation
frequency w
=Vk/m,
zero-point scale-length a
=Vh/mw
and energyeigenvalue En =
(n
+1/2 )
hw. The classicalturning-points
are at the solutions ofEn =
The WKB wave-vector is
m/h
times the classicalvelocity,
This is considered as a function of the
complex
coordinate z = x +i y.
We define the squareroot to have a branch-cut
from - xn
to + xn withpositive
real valuesjust
above the cut. Such abranch-cut is
required
by
thephysical possibility
of + or - velocities at anypoint x
in theclassically
allowed range. With thisdefinition,
and
where the
expression given
isinterpreted
as a contourintegral
which defines0,,(z)
for allcomplex
z.Analytic properties
ofcf>n(z)
can be establisheddirectly
fromequation
(6) (see
theAppendix),
but also theintegral
can be evaluated in closedform,
The
phase integral
is1/1t
times the action function of classical mechanics. The WKBwave-functions will be normalized with a coefficient
Nn,
The WKB
travelling
wave isIn our work we use these functions without modification
despite
their weaksingularity
(~
[x + xn ]-1l4 )
at theturning points.
Inparticular
we do not introduce anysmoothing
orconnection formula across the
turning-points.
In the calculation of matrix-elements we willsimply displace
the contour ofintegration
to avoid thesingularity.
The role of a connection formula isplayed by
equation (15)
below.The WKB functions
11, n (+ )
areanalytic
functions of zexcept
for the branch-cut on the range(- Xn, xn)
and for thepoint
atinfinity.
Theanalytic properties
arereadily developed
either from thedefining integral, equation
(6),
or from theexplicit
evaluation inequation
(7).
Onemight
be concernedwhether cP n (z)
and/orP n (± )(z)
aresingle-valued
because the contour ofintegration
inequation
(6)
can reach agiven point
(e.g.,
real x >xn)
in two ways.However,
by
virtue of thequantization
condition and theproperties
of theprefactor
1/ V q (z ),
thefunctions
’’n± )(z)
aresingle-valued
in thez-plane
cut from - Xn to xn. The derivation of thisimportant
technical result is sketched in theappendix.
Just as for
q (x ),
the branch-cut on the allowed range(-
Xn,xn )
resultsphysically
from thetwo
signs
of thevelocity.
Thepoint
atinfinity
is asingular point
of thepotential.
P n (± )
has no othersingularity.
The
asymptotic
form of1/1’ J:t )
isAlong
the real axiswm(+)
is
agrowing exponential,
which would make for anunsatisfactory
wave-function,
but its coefficientDm is purely imaginary.
Infact,
Re[ IP n (+
)(x)]
is zero for all real xhaving
[x
1
> xm. On the otherhand,
W)n(- )
is real and decreases forlarge
(real)
z.Because of this the WKB
approximation
to thequantum
wave-function may be taken to beThis is well-behaved at
large x
and turns out to be a verygood approximation
to thequantum
wave-function
everywhere
away from theturning points.
A
symmetry
formulacan be verified
by combining
properties
ofllq-(Z-)
and0,, (z)
given
in theappendix.
Equation (16)
is validonly
in the allowed rangelx
1
x,,.3. Semiclassical matrix elements.
If the wave-function of
equation (15)
isdirectly
used to calculate matrix-elementsby
thequantum
formulas the results aredisappointing, probably
because the matrix-elements areintegrals
ofoscillatory
functions so that even small errors(near
thetuming-points)
have alarge
effect on the result.The matrix-elements calculated
by
a modified method are much better. In this paper wediscuss
only
the harmonic oscillator and in this case it is theextraordinary
accuracy of theresulting
matrices whichgives
evidence for ourcomputational
scheme. However the modifiedsemiclassical rules also succeed for many other
systems
[1-3].
We use a modified inner
product
of wave-functions definedby
the contourintegral
and we calculate the matrix-element of the
quantum operator
f(P,x)
by
theanalogous
expression
For the
momentum p
we substitute the usual differentialoperator
(see
Eq.
(30) below).
Theintegration
contour C is taken to encircle the allowedregion
(for
both states n,m)
once in the clockwise sense. Because the wave functions areanalytic
functions of z the contour can bedeformed if needed.
In this paper we have included a factor 1/2 in the wave-function and for this reason
equations
(17),
(18)
do not need the factor 1/4 used in references[1-4].
The normalizationNn
ofequation
(8)
has the consequence thatUnn
= 1 for all n.Equations
(17)
and(18)
are semiclassical in the sense that the WKB functionsIF n (, )
areexplicitly
constructed from the classicalvelocity
and action functions without eversolving
aSchroedinger equation.
It is
possible
togive
arough
derivation ofequations
(17), (18)
from thequantum
theory
by
inserting
the wave function ofequation
(15)
into the usualquantum
orthogonality
and matrix-elementintegrals
and thenarguing
that terms’w’n(+) Wm(+),
Wn(-) Wm(- )
can beneglected,
atleast for
high
quantum
numbers,
becausethey
oscillaterapidly,
while the termswn(+) wm(-) ,
Similar
reasoning
can be found in theliterature,
without(evidently)
having
been testedcritically
(e.g.,
Ref.[7]).
The flaw in the derivation is that the method will be
applied
for lowquantum numbers,
and theintegrals involving
P («± )
P (± )
are
not small in that case. We therefore take thepoint
of view thatequations
(17), (18)
constitute basicassumptions
of a new semiclassicaltheory,
andwe will examine the consequences of these
assumptions.
Thephysical interpretation
ofequations
(17), (18)
isdeveloped
ingreater
detail in references[1-4].
4. Matrix-elements for the harmonic oscillator.The
surprises begin
when we evaluate the innerproduct
of WKB functions definedby
equation
(17).
The matrix U isSuch matrix-elements can be
explicitly
calculatedby deforming
theintegration
contour tolarge
z where the WKB wave-functions take theirasymptotic
forms :where C’ is a
large
circle at oo. Theintegral
isobviously
zero for all m> n. For m = n,Unn
isunity
as the result of asimple
contourintegration.
For m = n -1,
n - 3 and n -5,
Unm
is also zero. For m = n -2,
it is zero becauseBn
+Ln _ 2
= 0. Thus wehave,
inagreement
withequation
(19).
The WKB wave-functions are
exactly
orthonormal in the sense of the restricted interferenceapproximation,
astriking
result.Equation
(21)
is asurprise
in mathematical terms because the real-axis version of theintegral
inequation
(17)
has foursingularities
attuming points ;
butdespite
thesingular
integrand,
the result issimple.
Equation (21)
is remarkable inphysical
terms because afundamental
quantum property
hasemerged
fromintegrals
over classical velocities and actionfunctions.
For m = n -
4,
Un.
is not zero, and somewhatsimplifies
due to theequation :
Thus the matrix
Unm
is in exactagreement
with thequantum
theory
for all m > n -4,
but incorrect for m n 2013 4. Inparticular, Unm
is not Hermitian. We have LimUn, n -
4 =0,
soevidently
atlarge n
theregion
near thediagonal
becomes more and moren->oo
correct.
If U is
symmetrized by forming
Ùnm
=Ùmn
=Unm
Umn,
the result agreesexactly
with thequantum
theory,
Le.,
Ùnm
=8 nm.
The matrix-element of the coordinate x is
We evaluate this matrix and factor out the scale
length
Mathematically equation
(24)
is the same asequation
(20)
except
that thelarge-circle
integral
has an extra power of z.Xnm
isexactly
zero for allm> n + 1,’
zero for m = n, andXn, n -1
1 is nonzero ; we advance the indexby
one unit to writeAlso,
we haveXn,n - 2
= 0. HoweverXn,n - 3
is not zero ; the formulasimplifies
due towhich
gives
We observe that the main values
Xn, n + 1, Xn + 1 n
are very close to thequantum
values(= a [(n +"’1 )/2]112).
The fractionalerror
is== O.’02/n2.
The semiclassical matrix is notHermitian. The
symmetrized
matrixXn, m
=.Km, n
=V Xn, m Xm, n
agreesexactly
with thethe
diagonal,
butagain
decrease as n - oo while thenearby
«good
» values areincreasing.
The extreme lower-left corner of the matrix is
badly
incorrect.For the matrix
of p
= - i Ild /dx
weadopt
the semiclassical formulaThis
equation
remains semiclassical because we still have not solved aSchroedinger
equation ;
P n (--’: )
are semiclassical wave-functions.Equation
(30)
isdirectly
evaluated togive :
In this case
everything
above m = n - 3 isgood,
either exact or within 2 % of exact.Again
the matrix is non-Hermitian with
growing
incorrect values in the lower-left corner. For thegood
matrix-element we havejust
as in thequantum
theory.
Thesubsequent
rowsobey
The coefficient 1.267 = 57/45 can be verified
by
a littlealgebra.
Next we examine the commutator C =
[P, X]/ili = (PX -
XP )li h.
The result obtainedby matrix-multiplication
isThe most evident
properties
of C arei)
all matrix-elementsCn,m are
exactly
correct for allm > n - 2 and even for all m > n - 4 for n >
0 ;
ii)
except
for n =0,
theoff-diagonal
valuesThis means that C = -
U2
for all matrix elements n > 0. If the matrixCn, m
issymmetrized
by
forming
C n, m
=C m, n
=N/èn, m Cm, n
this matrix isagain
exact,C n, m = - ’6 n,
m-A most
interesting
aspect
ofequation (36)
is that the incorrect valuesXn, n - 3
andP n, n - 3
participate
inobtaining
Cn, n - 2
= 0 for n > 0.A matrix for
X2
can be constructed in two ways, firstby
matrix-multiplication
from theresult in
equation
(25),
.and
second,
by
contourintegration
ofequation
(18)
withf
=x2.
This second methodgives
Evidently
these matrices arei)
exactalong
the maindiagonal
and very accurate for allm > n -
2,
ii)
exactly
equal
for m > n -2,
so the matrixmultiplication
isvalid,
andiii)
asymptotically
correct atlarge n
for thediagonal
line with m = n - 2.Next we form a Hamiltonian matrix
Again
the matrix elements are exact for m > n -2,
and the incorrectsubdiagonal
termsHn, n - 2
approach
zero as n - oo.Again
thesymmetrized
matrix is exact.Finally,
we examine theHeisenberg equations
ofmotion, Le.,
the commutators[H, X ]
and[H, P ].
Thesequantities
shouldbe - i hP /m
andihmw 2X,
respectively.
To compare withequation
(25)
we formand to compare with
equation
(31),
we formIn this case all matrix-elements
having m > n -
1 are in exactagreement
with those inequations
(25), (31).
Evidently
asimple
rule is followed : the incorrect matrix-elementsbegin
at m =n - 4 for
U,
m = n - 3for X and P, at m
= n - 2 forX2,
C andH,
and then(evidently)
will occur at m = n - 1 forX3
and on thediagonal
forX4.
The rule is a consequence ofmatrix-multiplication
for matriceshaving
zeroes in theupper-right
corner. The result is alsoexactly
inagreement
with what onealready
knew about thecorrespondence
identities for theoscillator ;
(X2)
is exact but(x4)
is not. However as the incorrect values rise toward thediagonal,
the errors becomecomparatively
small., 5.
Summary
discussion.In this paper we have
explored
the semiclassical matrices formedby
the restricted interferenceapproximation,
equations (17), (18),
in order to see which matrix-elements arecorrect and what
algebraic properties
the semiclassical matrices have. We find that matrix-elements calculatedaccording
toequations
(17), (18)
areremarkably
accurate for all transitions with m > n for the harmonic oscillator. This is the mostimportant
result because it appears to remain valid for otherpotentials.
We find the
especially
provocative
result that theproduct
or commutator of semiclassicalmatrices is
exactly
asexpected,
i.e.,
exactly obeys
theHeisenberg matrix-algebra,
even whileThis means
that,
subject
to the restriction to m > n -N,
where Ndepends
on the matrixunder
consideration,
the semiclassical matrices form a secondinequivalent
solution
of theHeisenberg
axioms for matrixmechanics
except
for therequirement
that observables berepresented by
Hermitian matrices.The semiclassical matrices are accurate on one side of the
diagonal,
but we must ask whether it isreally
correct toidentify
thegood
matrix-elements withabsorption
(as
distinct fromemission) ;
does that identification read too much intoequations (17), (18)
which could be written with theopposite
convention ? Do the resultsimply
a basic difference betweenquantum
mechanics and semiclassical mechanics withrespect
to time-reversalsymmetry ?
Why
is theHermiticity
condition the rule which must be relaxed to obtain the semiclassicaltheory ?
Where does Planck’s constant enter as anexpansion parameter ?
Does thehigher
order WKB
theory
notonly
get
more accuratematrix-elements,
but also morediagonal lines
correct ? How many of these remarkable results are due to the
algebraic pecularities
of the harmonic oscillator ? Thesequestions
must be addressed in future work.While this paper has concentrated on
developing
exactproperties
of the semiclassicalmatrices,
we want to conclude with a reminder of the mostimportant practical
consideration. The WKB functions aresimple explicit
functions of thepotential
and thereforeequations
(17),
(18)
give
the matrix-elements of anyseparable
system
by
aquadrature.
As far as the results ofthis paper are a
guide,
it can beexpected
that matrix-elements obtainedby
thisquadrature
will be accurate.
Acknowledgment.
A
part
of this work wasperformed
while the author visited the LULI(GRECO-ILM)
Laboratory
of the EcolePolytechnique,
and the author is verygrateful
for thehospitality
of thatLaboratory.
Numerical calculations related to the results in this paper wereperformed
with the
help
of K. H. Warren.Work
performed
under theauspices
of the U.S.Department
ofEnergy by
the LawrenceLivermore National
Laboratory
under contract number W-7405-ENG-48.Appendix.
, In this
appendix
wesupply
a few more detailsconcerning
theanalytic
properties
of the WKBwave-functions. The
prefactor
in the WKBwave-function, 1 q (z ),
has an extra branch-cutwhich we take to run
from xn
to oo. With thischoice,
B/?(z)
isi)
positive
real for z = x + i ewith
1 x
[
Xnii)
positive
real for z= i y,
y > 0iii)
positive
imaginary
for z = x - i ewith
1 x 1
Xniv) positive
imaginary
for z = -i y,
y > 0Comparing (vi)
and(vii),
we find asign change
across the extra branch-cut. -ZAnalytic properties
of çb,,(z)
=
q (z’ )
dz’ can be determined from theintegral
and/or-
xn
is
negative
real for)
isnegative imaginary
In
(vi),
(vii),
theimaginary
part
isnegative ;
in fact as z - oo ,The subdominant term has a branch cut from x = xn to 0o across
which 0,, jumps
by
The
exponential
can
readily
beanalyzed by
means of these results. We arespecially
interested in the limit as z - 00. Astraightforward expansion gives
Now a small additional effort will convince the reader that
P n (± )
isanalytic everywhere
except
at z = oo
and along
the branch-cutlinking -
Xn to xn. In
particular,
the extra branch-cut from xn to 0o in0
and 0 n (z)
exactly
cancels in bothP n (+ )
and
1/1’ J- ).
Equation
(16)
is verifiedfrom
properties
(vi, vii)
of.J q (z)
and
On(Z)-References