IN PHASE SPACE
An Overview with Selected Papers
R. Monaco
Forthcoming
Matter Particled - Patterns, Structure and Dynamics
Selected Research Papers of Yuval Ne'eman
edited by R. Ruffini and Y. Verbin
vol,34
OUANTUM MECHANICS
IN PHASE SPACE
I
Editors
Cosmas K. Zachos
Argonne National Laborato y , USA
David B. Fairlie
University of Durham, UK
Thomas L. Curtright
University of Miami, USA
wp World Scientific
N E W JERSEY
L O N D O N * SINGAPORE * B E l J l N G
S H A N G H A I * HONG KONG * T A I P E I
CHENNAI
Published by
The editors and the publisher would like to thank the following for their permission to reproduce the articles
found in this volume:
Academic Press, American Association of Physics Teachers, American Physical Society, Cambridge University
Press, Elsevier Science B.V., Institute of Physics Publishing, Moscow Institute of Physics and Technology,
Progress of Theoretical Physics, Springer-Verlag, Taylor & Francis Group.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center,
Inc., 222RosewoodDrive, Danvers, MA01923, USA. Inthis casepermission tophotocopy is notrequiredfrom
the publisher.
CONTENTS
Preface
vii
1 Introduction
5 Ehrenfests Theorem
10
11
7 Time Evolution
13
16
17
10 Propagators
18
11 Canonical Transformations
19
21
24
25
15 Omitted Miscellany
26
28
References
31
vi
39
Index
43
vii
PREFACE
Wigners quasi-probability distribution function in phase space is a special (WeylWigner) representation of the density matrix. It has been useful in describing transport in
quantum optics; nuclear physics; and quantum computing, decoherence, and chaos. It is
also of importance in signal processing and the mathematics of algebraic deformation. A remarkable aspect of its internal logic, pioneered by Groenewold and Moyal, has only emerged
in the last quarter century. It furnishes a third, alternative, formulation of quantum mechanics, independent of the conventional Hilbert space, or path integral formulations.
In this logically complete and self-standing formulation, one need not choose sides between coordinate and momentum space. It works in full phase-space, accommodating the
uncertainty principle, and it offers unique insights into the classical limit of quantum theory. The variables (observables) in this formulation are c-number functions in phase space
instead of operators, with the same interpretation as their classical counterparts, but are
composed together in novel algebraic ways.
This volume is a selection of 23 useful papers in the phase-space formulation, with an
introductory overview which provides a trail-map to these papers and an extensive bibliography. (Still, the bibliography makes no pretense to exhaustiveness. An up-to-date database
on the large literature of the field, with special emphasis on its mathematical and technical aspects, may be found at http://idefix.physik.uni-freiburg.de/Nstar/en/download.html)
The overview collects often-used formulas and simple illustrations, suitable for applications
to a broad range of physics problems, as well as teaching. It provides supplementary material that may be used for a beginning graduate course in quantum mechanics. D. Morrissey
is thanked for the helpful comments and Prof Curtright would also like to express his thanks
to Ms Diaz-Heimer.
Errata and other updates to the book may be found on-line at
ht tp://server.physics.miami.edu/Ncurtright/QMPS
Introduction
There are at least three logically autonomous alternative paths to quantization. The first is
the standard one utilizing operators in Hilbert space, developed by Heisenberg, Schrodinger,
Dirac, and others in the 1920s. The second one relies on path integrals, and was conceived
by Dirac [Dir33] and constructed by Feynman.
The third one (the bronze medal!) is the phase-space formulation, based on Wigner's
(1932) quasi-distribution function [Wig321 and Weyl's (1927) correspondence [Wey27] between quantum-mechanical operators in Hilbert space and ordinary c-number functions in
phase space. The crucial composition structure of these functions, which relies on the *product, was fully understood by Groenewold (1946) [Gro46], who, together with Moyal
(1949) [Moy49], pulled the entire formulation together. Still, insights into interpretation
and a full appreciation of its conceptual autonomy took some time to mature wHith the work
of, among others, Takabayasi [Tak54], Baker [Bak58], and Fairlie [Fai64].
This complete formulation is based on the Wigner function (WF), which is a quasiprobability distribution function in phase space:
(1)
It is a generating function for all spatial autocorrelation functions of a given quantummechanical wave function $ ( x ) . More important, it is a special representation of the density
matrix (in the Weyl correspondence, as detailed in Section 12). Alternatively, in a 2ndimensional phase space, it amounts to
(2)
where $ ( x ) = ( X I $ )
There are several outstanding reviews on the subject: Refs. HOS84, Tak89, Ber80, BJ84,
Lit86, deA98, Tat83, Coh95, KN91, Kub64, DeG74, KW90, Ber77, Lee95, DahOl, Sch02,
DHSOO, CZ83, Gad95, HH02, Str57, McD88, Leo97, Sny80, Ba163, BFF78.
Nevertheless, the central conceit of the present overview is that the above input wave
functions may ultimately be bypassed, since the WFs are determined, in principle, as the
solutions to suitable functional equations in phase space. Connections to the Hilbert space
operator formulation of quantum mechanics may thus be ignored, in principle-even though
they are provided in Section 12 for pedagogy and confirmation of the formulation's equivalence. One might then envision an imaginary world in which this formulation of quantum
mechanics had preceded the conventional Hilbert-space formulation, and its own techniques
and methods had arisen independently, perhaps out of generalizations of classical mechanics
and statistical mechanics.
1
It is not only wave functions that are missing in this formulation. Beyond the ubiquitous (noncommutative, associative, pseudodifferential) operation, the +product, which
encodes the entire quantum-mechanical action, there are no linear operators. Expectations
of observables and transition amplitudes are phase-space integrals of c-number functions,
weighted by the WF, as in statistical mechanics. Consequently, even though the W F is not
positive-semidefinite (it can be, and usually is negative in parts of phase space [Wig32]), the
computation of expectations and the associated concepts are evocative of classical probability theory. Still, telltale features of quantum mechanics are reflected in the noncommutative
multiplication of such c-number phase-space functions through the *-product, in systematic
analogy to operator multiplication in Hilbert space.
This formulation of quantum mechanics is useful in describing quantum transport processes in phase space. Such processes are of importance in quantum optics [SchO2, Leo97,
SMOO], nuclear and particle physics [BakGO, SP81, MM84, CC03, BJY041, condensed matter [MMP94, DBB02, KKFR89, BP96, KLO1, JBMOS], the study of semiclassical limits
of mesoscopic systems [Imr67, OR57, Sch69, Ber77, KW87, OM95, MS95, MOT98, V089,
V0781, and the transition to classical statistical mechanics [JD99, Fre87, BD98, Raj83,
CV98, SMOO, FZOl, Za1031.
It is the natural language to study quantum chaos and decoherence [JN90, ZP94, BC99,
KZZ02, KJ99, Zu91, FBA96, Kol96, GH93, CL03, OCO3] (of utility in, e.g. quantum computing [BHP02]), and provides crucial intuition in quantum mechanical interference problems [Wis97], probability flows as negative probability backflows [BM94, FMSOO], and measurements of atomic systems [Smi93, Dun95, Lei96, KPM97, LvoOl, JS02, BHS02, Ber02,
Cas911.
The intriguing mathematical structure of the formulation is of relevance to Lie Algebras
[FFZ89],martingales in turbulence [FanOS],and string field theory [BKMOS]. It has recently
been retrofitted into M-theory advances linked to noncommutative geometry [SW99] (for
reviews, see Refs. CasOO, HarOl, DNO1, HS02), and matrix models [TayOl, KS021; these apply space-time uncertainty principles [Pei33, Y089, JY98, SSTOO] reliant on the *-product.
(Transverse spatial dimensions act formally as momenta, and, analogously to quantum mechanics, their uncertainty is increased or decreased inversely to the uncertainty of a given
direction.)
As a significant aside, the W F has extensive practical applications in signal processing,
filtering, and engineering (time-frequency analysis) , since time and frequency constitute a
pair of Fourier-conjugate variables just like the and p pair of phase space.a
For simplicity, the formulation will be mostly illustrated for one coordinate and its
conjugate momentum, but generalization to arbitrary-sized phase spaces is straightforward
[DM86],including infinite-dimensional ones, namely scalar field theory [DitSO, Les84, Na97,
CZ99, CPPO1, MM941: the respective WFs are simple products of single-particle WFs.
"Thus, time-varying signals are best represented in a W F as time-varying spectrograms, analogously to a music score ,
i.e. the changing distribution of frequencies is monitored in time [BBL80, Wok97, QC96, MH97, Coh95, GroOl]: even
though the description is constrained and redundant, it gives an intuitive picture of the signal that a mere time profile
or frequency spectrogram fails to convey. Applications abound [CGBSl, Lou96, MH97] in bioengineering, acoustics,
speech analysis, vision processing, turbulence microstructure analysis, radar imaging, seismic data analysis, and the
monitoring of internal combustion engine-knocking, failing helicopter component vibrations, and so on.
f(z,p)=
27r
S d y $* (x
5.)
e-iyp$ (z
+ zy).
ti
(4)
-;
one space dimension, by virtue of nondegeneracy, $J has the same effect as $J*, and f turns out to be p-even, but
this is not a property used here.
CThisone is called the Husimi distribution [Tak89, TA991, and sometimes information scientists examine it on account
of its non-negative feature. Nevertheless, it comes with a heavy price, as it needs to be dressed back to the
W F for all practical purposes when expectation values are computed with it, i.e. it does not serve as an immediate
quasi-probability distribution with no further measure (see Section 13). The negative feature of the W F is, in
the last analysis, an asset, not a liability, and provides an efficient description of beats [BBL80, Wok97, QC96,
MH97, Coh951; cf. Fig. 1. If, instead, strictly inequivalent (improper) expectation values were taken with the Husimi
distribution without the requisite dressing of Section 13, i.e. as though it were a bonaifide probability distribution, such
expectation values would reflect loss of quantum information: they would represent classically coarsened observables
[WO87].
n
5
c-number functions. Such functions are often classical quantities but, in general, are
uniquely associated with suitably ordered operators through Weyls correspondence rule
[Wey27]. Given an operator ordered in this prescription,
1
(2d2
.>I
,
(7)
the corresponding phase-space function g(z,p ) (the Weyl kernel function of the operator)
is obtained by
PHP,
r-5.
(8)
(9)
The kernel function g(z,p ) is often the unmodified classical observable expression, such
as a conventional Hamiltonian, H = p2/2m V(z), i.e. the transition from classical mechanics is straightforward. However, it contains h corrections when there are quantummechanical ordering ambiguities, such as in the observable kernel of the square of the
angular momentum C C: This contains a term, -3h2/2, introduced by the Weyl ordering [She59, DS82, DS021, beyond the mere classical expression ( L 2 ) ,and accounts for the
nontrivial angular momentum of the ground-state Bohr orbit. In such cases (including
momentum-dependent potentials), even nontrivial O(h) quantum corrections in the kernel
functions (which characterize different operator orderings) can be produced efficiently without direct, cumbersome consideration of operators [CZ02, Hie841. More detailed discussion
of the Weyl and alternate correspondences is provided in Sections 12 and 13.
In this sense, expectation values of the physical observables specified by kernel functions
g ( z , p ) are computed through integration with the WF, in close analogy with classical
probability theory, except for the non-positive-definiteness of the distribution function. This
operation corresponds to tracing an operator with the density matrix (cf. Section 12).
Given a specification of observables, the next step is to find the relevant W F for a given
Hamiltonian. Can this be done without solving for the Schrodinger wave functions $, i.e.
not using Schrodingers equation directly? Indeed, the functional equations which f satisfies
completely determine it.
Firstly, its dynamical evolution is specified by Moyals equation. This is the extension
of Liouvilles theorem of classical mechanics, for a classical Hamiltonian H ( z ,p), namely
atf + {f,N} = 0, to quantum mechanics, in this language [Wig32, Moy491:
(10)
The right-hand side of (10) is dubbed the Moyal Bracket (MB), and the quantum
commutator is its Weyl correspondent. It is the essentially unique one-parameter (ti) associative deformation of the Poisson brackets of classical mechanics [Vey75, BFF78, FLS76,
Ar83, Fle90, deW83, BCG97, TD971. Expansion in ti around 0 reveals that it consists of
the Poisson bracket corrected by terms O(h).
The equation (10) also evokes Heisenbergs equation of motion for operators, except that
H and f here are classical functions, and it is the *-product which enforces noncommutativity. This language makes the link between quantum commutators and Poisson brackets
more transparent.
Since the *-product involves exponentials of derivative operators, it may be evaluated
in practice through translation of function arguments ( Bopp shifts),
(12)
(13)
The cyclic phase-space trace is directly seen in the representation (14) to reduce to a plain
product, if there is only one * involved:
dpdx f * g
dpdx fg =
dpdx g * f .
(16)
Moyals equation is necessary, but does not suffice to specify the W F for a system. In
the conventional formulation of quantum mechanics, systematic solution of time-dependent
equations is usually predicated on the spectrum of stationary ones. Time-independent purestate Wigner functions *-commute with H , but clearly not every function *-commuting with
H can be a bona-fide W F (e.g. any * function of H will *-commute with H ) .
6
Static WFs obey more powerful functional *-genvalue equations [Fai64] (also see
Refs. Kun67, Coh76, Dah83):
iti
=
f ( V )* H ( z , p ) = E f ( z , d
(17)
where E is the energy eigenvalue of fj$ = E$. These amount to a complete characterization
of the WFs [CFZ98].
Lemma 1 For real functions f ( z , p ) , the Wigner form (4) for pure static eigenstates is
equivalent to compliance with the *-genvalue equations (17) (8 and S parts).
Proof
(18)
.f * H
=E f(Z,P),
(19)
(20)
Hence, real solutions of (17) must be of the form f = Jdy e-iyp$*(x - ?y)$(x
;y)/2",
such that fj$ = E$.
0
Equation (17) lead to spectral properties for WFs [Fai64, CFZ981, as in the Hilbert
space formulation. For instance, projective orthogonality of the * genfunctions follows from
associativity, which allows evaluation in two alternate groupings:
f * H * g = E f f * g = E , f*g.
7
(21)
f*g=O.
f*H*f=Ef f*f=H*f*f,
and hence f
*f
(23)
f*f.(f.
(24)
Pure state fs then *-project onto their space. In general, it can -e shown [Tak54, CFZ981
that, for a pure state,
fa
*f b =
1
Ja,b f a
(25)
'
The normalization matters [Tak54]: despite linearity of the equations, it prevents superposition of solutions. (Quantum mechanical interference works differently here, in comportance
with density matrix formalism.)
By virtue of (16), for different *-genfunctions, the above dictates that
dpdx f g
= 0.
(26)
Consequently, unless there is zero overlap for all such WFs, at least one of the two must
go negative someplace to offset the positive overlap [HOS84, Coh951-an illustration of the
feature of negative values. This feature is an asset and not a liability.
Further, note that integrating (17) yields the expectation of the energy,
(27)
H(x,p)f (x,p)dxdp = E
dxdp f 2
1
=-
(28)
i.e. the overlap increases to a divergent result in the classical limit, a s the WFs grow increasingly spiky.
dThis discussion applies to proper WFs, corresponding to pure states' density matrices. E.g. a sum of two WFs
is not a pure state in general, and does not satisfy the condition (6). For such generalizations, the impurity is
[Gro46] 1 - h ( f ) = s d x d p (f - h f 2 ) 2 0, where the inequality is only saturated into a n equality for a pure state.
For instance, for w
(fa fb)/2 with fa * f b = 0, the impurity is nonvanishing, d x d p (w - hw') = 1/2. A pure
state affords a maximum of information, while the impurity is a measure of lack of information [Fan57, Tak541-it is
the dominant term in the expansion of the quantum entropy around a pure state [Bra94].
In classical (non-negative) probability distribution theory, expectation values of nonnegative functions are likewise non-negative, and thus result in standard constraint inequalities for the constituent pieces of such functions, e.g., moments of the variables. But
it was just seen that for WFs which go negative for an arbitrary function g, (1gI2)need not
be 2 0. This can be easily seen by choosing the support of g to lie mostly in those regions
of phase-space where the W F f is negative.
Still, such constraints are not lost for WFs. It turns out they are replaced by:
Lemma 2
(9**9) 1 0.
(29)
In Hilbert space operator formalism, this relation would correspond to the positivity of the
norm. This expression is non-negative because it involves a real non-negative integrand for
a pure state W F satisfying the above projective Condition",
/dpdx(g**g)f = h
dxdp(g**g)(f*f) = h
dxdp(f*g*)*(g*f) = h
dxdplg*f12. (30)
0
+ bx + cp,
(31)
for arbitrary complex coefficients a , b, c. The resulting positive semi-definite quadratic form
is then
a*a+b*b(z*x) +c*c(p*p)
+ (a*b+b*a)(x)+(a*c+c*a)(p)
for any a, b, c. The eigenvalues of the corresponding matrix are then non-negative, and thus
so must be its determinant. Given
2
x*x=x,
p*p=p2,
ih
ih
p*x=px--,
2
x*p=px+-,
2
(33)
= ((x - (xc)>2),
@PI2
= ((P- (P)l2>,
(34)
ti
AxAp>-.
2
(36)
eSimilarly, if fl and f2 are pure state WFs, the transition probability (I JdzQit; ( Z ) & ( Z ) ~ ~between
)
the respective states
is also non-negative [OWSl],manifestly by the same argument [CZOl], namely J d p d x f i f2 = ( 2 7 r F ~J) ~d x d p lfl*f2I2 2
0.
10
The ti entered into the moments constraint through the action of the * product [CZOl].
More general choices of g likewise lead to diverse expectations inequalities in phase space;
e.g. in six-dimensional phase space, the uncertainty for g = a + bL, + cL, requires 1(1+ 1) 2
m(m l),and hence 1 2 m, etc. [CZOl, CZ021. For a more extensive formal discussion of
moments, cf. Ref. N086.
Ehrenfests Theorem
(37)
at = V J , f ,l
serves to prove Ehrenfests theorem for expectation values. For any phase-space function
k ( x , p ) with no explicit time dependence,
= zJdxdp
= Jdxdp
(H* f
f *H)*k
fgk,Hl= (gk,Hl).
(38)
(39)
and Moyals equation above. The z , p equations of motion reduce to the classical ones of
Hamilton, x = a p H , p = -&H.
Moyal [Moy49] stressed that his eponymous quantum evolution equation (10) contrasts
to Liouvilles theorem for classical phase-space densities,
dfcl
afcl
(40)
- =-+;:a,.fcl+papfcl=o.
dt
at
Specifically, unlike its classical counterpart, in general, f does not flow like an incompressible
fluid in phase space.
For an arbitrary region R about a representative point in phase space,
(41)
That is, the phase-space region does not conserve in time the number of points swarming
about the representative point: points diffuse away, in general, without maintaining the
density of the quantum quasi-probability fluid, and, conversely, they are not prevented
from coming together, in contrast to deterministic flow. For infinite R encompassing the
entire phase space, both surface terms above vanish to yield a time-invariant normalization
10
11
for the WF. The O(ti2)higher momentum derivatives of the WF present in the MB (but
absent in the PB-higher space derivatives probing nonlinearity in the potential) modify
the Liouville flow into characteristic quantum configurations [KZZ02, FBA96, ZP941.
To illustrate the formalism on a simple prototype problem, one may look at the harmonic
oscillator. In the spirit of this picture, one can, in fact, eschew solving the Schrodinger
problem and plugging the wave functions into (4); instead, one may solve (17) directly for
H = (p2 x2)/2 (with m = 1, w = 1):
[(x
iti
+ ,itia,)" + ( p - -8z)2
2
- 2E] f ( ~ , p=
) 0.
(42)
For this Hamiltonian, the equation has collapsed to two simple PDEs. The first one, the
Imaginary part,
(43)
restricts f to depend on only one variable, the scalar in phase space, z = 4 H / h = 2(x2
p 2 ) / h . Thus the second one, the Real part, is a simple ODE,
(44)
Setting f ( z ) = exp(-z/2)L(z) yields Laguerre's equation,
za;
'1
+ (1 - .)az + -t i 2
L ( z ) = 0.
(45)
L, = -e"
for n = E / h - 1/2 = 0 , 1 , 2 , . . . , so the
Lo=1,
L1=l--
* gen-Wigner
4H
ti'
(46)
8H2
L2=--h2
8H
ti
+ 1,
1..
(47)
But for the Gaussian ground state, they all have zeros and go negative. These functions
become spiky in the classical limit ti + 0 ; e.g. the ground state Gaussian fo goes to a 6
function.
11
14
U,(x,p;t) = e,
=1
+ (it/h)H(x,p) + (it2!/ti)
~
(it/ h)3
H * H + 7H * H * H + ..., (54)
for arbitrary Hamiltonians. The solution to Moyals equation, given the W F a t t = 0, then,
is
f ( x , p ; t ) = U?(X,P;t) *f(x,p;O) *U*(x,p;t).
(55)
In general, just like any *-function of H , the *-exponential (54) resolves spectrally
[Bon84]:
(56)
(Of course, for t = 0, the obvious identity resolution is recovered.) In turn, any particular
*-genfunction is projected out formally by
(57)
(7)
-1
= [cos(;)]
exp [;Htan(;)]
(58)
- --
x*H - H*x
iii
= apH = p
(59)
(60)
where the concluding member of these two equations hold for the oscillator only. Thus, for
the oscillator,
x(t) = a : c o s t + p s i n t ,
p(t) = p c o s t - x s i n t .
(61)
As a consequence, for the oscillator, the functional form of the Wigner function is
preserved along classical phase-space trajectories [Gro46]:
f ( x , p; t ) = f ( x cos t
- p sin t ,p cos t
+ x sin t ;0).
(62)
(62)
fAs an application, note that the celebrated hyperbolic tangent *-composition law of Gaussians follows trivially, since
[BFF78]. That
these amount to *-exponentials with additive time intervals, exp,(tf) * exp,(Tf) = exp,[(t T)f)],
is,
14
16
In Diracs interaction representation, a more complicated interaction Hamiltonian superposed on the oscillator one leads to shape changes of the W F configurations placed on
the above turntable, and serves to generalize to scalar field theory [CZ99].
More generally, to represent all operators on phase space in a selected basis, one looks
at the Weyl-correspondents of arbitrary ) . 1 (bl, referred to as nondiugonal WFs [Gro46].
These enable investigation of interference phenomena and the transition amplitudes in the
formulation of quantum-mechanical perturbation theory [BM49, W088, CUZOl].
Both the diagonal and the non-diagonal WFs are represented in (2), by replacing p
+
I$a>($bI:
=$a(x)
* J ( P ) * $;(a:)
(63)
The representation on the last line is due to Ref. Bra94 and lends itself to a more compact
and elegant proof of Lemma 1. Just as pure-state diagonal WFs obey a projection condition,
so too the non-diagonals. For wave functions which are orthonormal for discrete state labels,
da: $:(a:)$b(z) = dab, the transition amplitude collapses to
J dxdp
fab
(x,P ) = dab
(64)
To perform spectral operations analogous to those of Hilbert space, it is useful to note that
these WFs are *-orthogonal [Fai64],
( 2 r f i ) fba
* f d c = dbcfda
(65)
(2.h)
fab
( z l , P l )fba
( z 2 , P 2 ) = 6 (z1 - x 2 )
6(P1 - P 2 )
(66)
a,b
For example, for the SHO in one dimension, non-diagonal WFs are
(67)
(cf. coherent states [CUZOl, DG80)]. Explicitly, in terms of associated Laguerre polynomials, these are [Gro46, BM49, Fai641
ei(n-k)
arctan(p/z)
(-Ik
(68)
7rfi
16
17
fkn
* H = Ek f k n
(69)
i)
i)
The energy *-genvalue conditions are (En - /ti = n, an integer, and (Ek - / h = k,
also an integer.
The general spectral theory of WFs is covered in Refs. BFF78, FM91, Lie90, BDW99,
CUZOl.
Given the spectral properties summarized, the phase-space perturbation formalism is selfcontained: it need not make reference to the Hilbert-space treatment [BM49, W088, CUZO1,
SS02, MS961.
For a perturbed Hamiltonian,
H (.,P>
= Ho(.,P)
Hl(z,P) ,
(70)
03
(71)
k=O
k=O
* f n = Enfn = f n * H .
(72)
(73)
For example, consider all polynomial perturbations of the harmonic oscillator in a unified
treatment, by choosing
H I = eYX+6P
*
*
0
0
= YX+JP = eYX
e,
e6P
e(Y6/2 =
e6P
eY"
e-iY6/2
(74)
to evaluate a generating function for all the first-order corrections to the energies [CUZOl],
00
00
(75)
n=O
n=O
hence
(76)
17
18
From the spectral resolution (56) and the explicit form of the +exponential of the
ti], it follows that
oscillator Hamiltonian (58) [with eit --t s and ELo) = ( n
+ i)
(77)
and hence
E(*)(s) =
-
dxdp eyx+6pexp
7rti(l+ s )
1-s
exp
[fi.4
(72
+ 62)
1-s
I-s
(--)
ti
x+p
lfs
(78)
(79)
and so on. All the first order corrections to the energies are even functions of the
parameters-only even functions of x and p can contribute to first-order shifts in the oscillator energies.
First-order corrections to the WFs may be similarly calculated using generating functions for nondiagonal WFs. Higher order corrections are straightforward but tedious. Degenerate perturbation theory also has an autonomous formulation in phase-space, equivalent
to Hilbert space and path-integral treatments.
10
Propagators
Time evolution of general WFs beyond the above treatment is discussed at length in
Refs. BM49, Ber75, GMSO, CUZOl, BR93, wo82, W002, FMO3. A further application
of the spectral techniques outlined is the computation of the W F time-evolution operator from the propagator for wave functions, which is given as a bilinear sum of energy
eigenfunct ions,
(80)
as it may be thought of as an exponentiated effective action. (Henceforth in this section,
take h = 1.)
This leads directly to a similar bilinear double sum for the W F time-transformation
kernel [Moy49],
f b a (5,P>e
ah
18
-i(E,-Eb)t
f a b ( X , p) .
(81)
19
Defining a big star operation as a *-product for the upper-case (initial) phase-space
variables,
(82)
it follows that
T ( X , Pi
x,p ;t)*.fdc(X,
p )=
f b c ( z , P ) e-
i (Ec-Eb) t
fdb(X,P) 7
(83)
/ d X d P T ( x , p ;X , P ;t ) f d c ( X P
, ) = fdc(x,p)e-i(Ec-Ed)t
= UL1*fdC(z,p;
O)*U, = f d c ( z , p ;t ) .
(84)
For example, for a free particle of unit mass in one dimension, H = p 2 / 2 , WFs propagate
according to
6(z-X-Pt)6(p-P),
(85)
(86)
11
Canonical Transformations
d X d P = dxdp { X ,P }
(87)
{u,v),p
{ X ,q
= -- -- ,
axap apax
x p =
1,
{X,P}XP =
(88)
1.
(89)
Yes, for linear canonical transformations [KLOl], but clearly not for general canonical
transformations. Still, things can be put right, by devising general covariant transformation
rules for the *-product [CFZ98]: the WF transforms in comportance with Diracs quantum
canonical transformation theory [Dir33].
20
K l ( G q ,
(90)
the energy eigenfunctions transform in a generalization of the "representation-changing"
Fourier transform [Dir33],
(91)
(In this expression, the generating function F may contain ti corrections [BCT82] to the
classical one, in general-but for several simple quantum-mechanical systems it manages
not to [CG92, DG021.) Hence [CFZ98], there is a transformation functional for WFs,
I ( x , p ;X, P ) , such that
f ( ~ , p=
) J d x d p T(Z,P;
x,P)*FT(X,P ) =
d x d I~( Z , P ;
x,P ) F(X,P ) ,
(92)
where
--s
-
" I2n
2
(93)
+ ~ P -Yi F * ( x - -,x
Y
- -Y
>2 + i
2
~ ( x Y-, x
2
I'
+ 2)
H ( w ) * q x ,P ;
'H(XP).
(94)
(94)
That is, if F satisfies a *-genvalue equation, then f satisfies a *-genvalue equation with the
same eigenvalue, and vice versa. This proves useful in constructing WFs for simple systems
which can be trivialized classically through canonical transformations.
A thorough discussion of MB automorphisms may start from Ref. BCW02 . (Also see
Refs. Hie82, DKM88, GR94, DV97, Hak99, KL99, DPOl.)
Time evolution is a canonical transformation [Dir33], with the generator's role played
by the effective action A of the previous section, incorporating quantum corrections to both
phases and normalizations; it connects initial wave functions to those at a final time.
For example, for the linear potential, with
H=p2+z,
(95)
(95)
(96)
T then evaluates to
Tin
P7 ; X , P ;t )
21
-
- 8n2t
it
t X.)
p%-2t
= 6(p+t-P) 6(x-2tp-t2-X)
= -2t
16
= d(x-X-(p+P)t)
+ iPY - -2 (y + Y )+ 2t (x - X)(y - y )
6 (p - - - - .-x>
d Y d y exp -iyp
2t
6(P-p-t).
(97)
The 6 functions enforce exactly the classical motion for a mass= 1 / 2 particle subject to
a negative constant force of unit magnitude (acceleration = -2). Thus the W F evolves
classically as
f ( x , p ;t ) = f(x
- 2pt
-t2,p
+ t ;0 ) .
(98)
Note that time independence follows for f ( z , p ; 0) being any function of the energy variable,
since x p2 = x - 2pt - t 2 + ( p + t ) 2 .
The evolution kernel T propagates an arbitrary W F through just
f ( x , p ;t ) =
x,
d X d P T ( x , p ; P ;t ) f(X,
P ;0 ) .
(99)
The underlying phase-space structure, however, is more evident if one of the wave-function
propagators is given in coordinate space, and the other in momentum space. Then the
path integral expressions for the two propagators can be combined into a single phase-space
path integral. For every time increment, phase space is integrated over to produce the new
Wigner function from its immediate ancestor. The result is
T (x,p ;
x,p ;t )
(100)
o>*eiz2P2e-2(x--zz)(p-Pz),
= 1 ~ ~ l d p l J d z 2 d p 2 e 2 ( z - z ~ ) ( ~ - m ) e - (~Z-lz; t~\ v
x l2 ;0) (pl; tlp2;
7 9
where ( q t lz2;O) and ( p 1 ; t (p2;O)are the path integral expressions in coordinate space,
and in momentum space. Blending these x and p path integrals gives a genuine path
integral over phase space [Ber80, DK851. For a direct connection of U, to this integral, see
Refs. Sha79, Lea68, SamOO.
12
This section summarizes the bridge and equivalence of phase-space quantization to the
conventional formulation of quantum mechanics in Hilbert space. The Weyl correspondence
merely provides a change of representation between phase space and Hilbert space. In
itself, it does not map (commutative) classical mechanics to (non-commutative) quantum
mechanics, but it makes that deformation map easier to grasp, defined within a common
representation, and thus more intuitive.
Weyl [Wey27] introduced an association rule mapping invertibly c-number phase-space
functions g(x,p ) (called phase-space kernels) to operators 8 in a given ordering prescription.
21
22
Specifically, p
H p,
x H r, and, in general,
1
- d r d o d x d p g ( z , p ) exp [ir(p - p) io(F - x)] .
(101)
(2Tl2
The eponymous ordering prescription requires that an arbitrary operator, regarded as a
power series in F and p, be first ordered in a completely symmetrized expression in F and p,
by use of Heisenberg's commutation relations, [F,p] = ih.
A term with m powers of p and n powers of F is obtained from the coefficient of T ~ O "
in the expansion of (rp +
which serves as a generating function of Weyl-ordered
polynomials [GF91]. It is evident how the map yields a Weyl-ordered operator from a
polynomial phase-space kernel. It includes every possible ordering with multiplicity one,
e.g.
@(F, p)
6p2x2
P2F2
(102)
In general [McC32],
(103)
Phase-space constants map to the identity in Hilbert space.
In this correspondence scheme, then,
l"r6 =
(104)
(104)
dxdpg.
Conversely [Gro46, Kub64, HOS841, the c-number phase-space kernels g ( x ,p) of Weylordered operators @(F, p) are specified by p H p, F H x, or, more precisely, by the "Wigner
map,"
(105)
since the above trace reduces to
dz
e i r u ~ / (zle-iuFe-irp
2
@lz)= 27r
d z ( z - hrI@jlz)eiu(rh12--z).
(106)
Thus, the density matrix inserted in this expression [Moy49] yields the hermitean generalization of the Wigner function (63) encountered,
=27r / d y e - i Y w z
ti
p)i&+ ZY)
= fb*.hP)
(107)
22
23
(108)
and thus expectation values follow for m = n, as utilized throughout in this overview.
Hence,
{0m| exp i(cr$ + Tp)\ij>m) = I dxdp fm(x,p) exp i(ax + rp),
(109)
r/xf+'^rf-r^)
(110)
7/ =
(p'-p),
rj = <7-(p'-p)
(111)
e x p i r ( p - p) + a(j - x) ( f * g ) ( x , p ) ,
(112)
The ^-product thus defines the transition from classical to quantum mechanics. In fact,
the failure of Weyl-ordered operators to close under multiplication may be stood on its
head [Bra02], to define a Weyl-symmetrizing operator product which is commutative and
constitutes the Weyl transform of fg instead of the non-commutative f * g. (For example,
1x * p = Ixp + ih i-> 2jp = pp + pp + ih. The classical piece of 2x * p maps to the Weyl
23
24
symmetrization of the operator product, 2xp H ~p p ~ . )One may then solve for the PB in
terms of the MB, and, through the Weyl correspondence, reformulate Classical Mechanics
in Hilbert space as a deformation of Quantum Mechanics, instead of the other way around
[Bra02].
Arbitrary operators 6 ( ~ , p )consisting of operators F and p, in various orderings, but
with the same classical limit, could be imagined rearranged by use of Heisenberg commutations to canonical completely symmetrized Weyl-ordered forms, in general with O(h)
terms generated in the process. Each one might then be inverse-mapped uniquely to its
Weyl-correspondent c-number kernel function g in phase space. [In practice, there is the
more direct Wigner transform formula (105), which bypasses a need for an actual rearrangement.] Thus, operators differing from each other by different orderings of their JS and ps
correspond to kernel functions g coinciding with each other at O ( h o ) but
, different at O ( h ) ,
in general. Hence, in phase-space quantization, a survey of all alternate operator orderings
in a problem with such ambiguities amounts t o a survey of the quantum correction O(h)
pieces of the respective kernel functions, i.e. the inverse Weyl transforms of those operators,
and their study is systematized and expedited. Choice-of-ordering problems then reduce to
purely *-product algebraic ones, as the resulting preferred orderings are specified through
particular deformations in the c-number kernel expressions resulting from the particular
solution in phase space [CZOa].
13
The Weyl correspondence rule (101) is not unique: there are a host of alternate equivalent
association rules which specify corresponding representations. All these representations
with equivalent formalisms are typified by characteristic quasi-distribution functions and
*-products, all inter-convertible among themselves. They have been surveyed comparatively
and organized in Refs. Lee95, BJ84, on the basis of seminal classification work by Cohen
[Coh66, Coh761, and are favored by virtue of their different characteristic properties in
varying applications.
For example, instead of the operator exp(i7p i a ~of) the Weyl correspondence, one
might posit, instead [Lee95, HOS841, antistandard ordering,
(113)
,
specifies the Kirkwood-Rihaczek prescription; or else stanwith w = e x p ( i h ~ a / 2 ) which
dard ordering, w = exp(-ih~a/2) on the right-hand side of the above, for the Mehta
prescription; or normal and antinormal orderings for the Glauber-Sudarshan prescriptions, generalizing to w = exp[?(T2 a2)]for the Husimi prescription [Hus40, Tak891;
or w = cosh[a(T2 a2)]for the Rivier prescription; or w = s i n ( h ~ a / 2 ) / ( h ~ o / 2 for
) , the
Born-Jordan prescription; and so on.
The corresponding quasi-distribution functions in each representation can be obtained
as convolution transforms of each other [Coh76, Lee95, HOS841, and likewise the kernel func-
24
25
tion observables are convolution dressings of each other, as are their *-products [Dun88,
AW70, Ber751.
Example For instance, the Husimi distribution follows from a Gaussian smoothing linear
conversion map [W087, Tak89, Lee951 of the WF,
f H =~
( f=) exp
(114)
ti
(115)
Expectation values of observables now entail equivalence conversion c;essings of tlLerespective kernel functions and a corresponding *-product [Ba79, OW81, V089, Tak89, ZacOO],
which now cannot be simply dropped inside integrals. For this reason, distributions such
as this Husimi distribution (which is positive-semidefinite [Car76, OW81, Ste801) cannot be
automatically thought of as bona-fide probability distributions. This is often dramatized as
the failure of the Husimi distribution f H to yield the correct x- or p-marginal probabilities,
upon integration by p. or . x , respectively [OW81, HOS841. Since phase-space integrals are
thus complicated by conversion dressing convolutions, they preclude direct applications of
the Schwarz inequality and the standard inequality-based moment-constraining techniques
of probability theory, as well-as routine completeness and orthonormality-based functional
analytic operations. (Ignoring the above equivalence dressings and, instead, simply treating
the Hussimi distribution as an ordinary probability distribution in evaluating expectation
values results in loss of quantum information-effectively coarse-graining to a classical
limit.)
Similar caveats also apply to more recent symplectic tomographic representations
[MMT96, MMMO1, Leo971, which are positive semi-definite too, but also do not constitute conventional probability distributions.
W f 1 g H=
1 Q(f),%7)
25
(116)
26
as utilized in Dirads heuristics. Instead, the Weyl correspondence map (101) from functions
to ordered operators,
1
m(f) (W2
+ ia(p - x)],
(117)
m(tIf791)
[Wf),
Wd]
*
(118)
It is the MB, then, instead of the PB, which maps invertibly to the quantum commutator.
That is to say, the deformation in phase-space quantization is nontrivial: the quantum
functions, in general, do not coincide with the classical ones [Gro46], and involve O(h)
corrections, as extensively illustrated in, e.g. Refs. CZ02, DS02, CH86; also see Ref. Got99.
An alternate abstract realization of the above MB algebra in phase space (as opposed
to the Hilbert space one), m(f),is [FFZ89, CFZm981
W)=f* .
(119)
Realized on a toroidal phase space, with a formal identification ti H 27r/N, it leads to the
Lie algebra of SU(N) [FFZ89],by means of Sylvesters clock-and-shift matrices [Sy182]. For
generic h, it may be thought of as a generalization of SU(N) for continuous N , allowing for
taking the limit N + 00.
Essentially (up to isomorphism), the MB aIgebra is the unique one-parameter deformation of the Poisson bracket algebra [Vey75, BFF78, FLS76, Ar83, Fle90, deW83, BCG97,
TD971, a uniqueness extending to the star product. Isomorphism allows for dressing transformations of the variables (kernel functions and WFs, as in Section 13 on alternate orderings), through linear maps f H T ( f ) ,which leads to cohomologically equivalent starproduct variants, i.e. [Ba79, Vo89, BFF781
(120)
15
Omitted Miscellany
Phase-space quantization extends in several interesting directions which are not covered in
such a summarizing introduction.
The systematic generalization of the *-product to arbitrary non-flat Poisson manifolds
[Kon97], is a culmination of extensions to general symplectic and Kahler geometries [Fed94,
KisOl], and varied symplectic contexts [Ber75, RTOO, CPPO2, BGLOl]. For further work on
curved spaces, cf. Refs. APW02, BF81, PT99. For extensive reviews of mathematical issues,
cf. Refs. Fo189, Hor79, W098, AW70. For a connection to the theory of modular forms, SCC
26
Ref. RajOZ. For WFs of discrete (finite systems), cf. Refs. w0087, ACW98, RA99, RGOO,
BHP02.
Spin is treated in Refs. Str57, VG89, AWOO; and forays into a relativistic formulation
in Ref. LSUO2 (also see Refs. CS75, Ran66).
Inclusion of Electromagnetic fields and gauge invariance is treated in Refs. Mue99,
LF94, LFO1, JVS87, ZC99, K000. Subtleties of Berrys phase in phase space are addressed
in Ref. SamOO.
28
Selected Papers
16
The decisive contributors to the development of the formulation are Hermann Weyl (18851955), Eugene Wigner (1902-1995), Hilbrand Groenewold (1910-1996), and Jose Moyal
(1910-1998). The bulk of the theory is implicit in Groenewolds and Moyals seminal papers. But this has been a slow story of emerging connections and chains of ever-sharper
reformulations. Confidence in the autonomy of the formulation accreted slowly. As a result, attribution of critical milestones cannot avoid subjectivity: it cannot automatically
highlight merely the earliest occurrence of a construct, unless that has also been conclusive
enough to yield an indefinite stay against confusion about the logical structure of the
formulation.
H Weyl (1927) [Wey27] introduces the correspondence of Weyl-ordered operators to
phase-space (c-number) kernel functions (as well as discrete QM application of Sylvesters
(1883) [Sy182] clock-and-shift matrices).
J von Neumann (1931) [Neu31], in a technical aside off a study of the uniqueness of
Schrodingers representation, includes a Fourier transform version of the *-product which
promotes Weyls correspondence rule to full isomorphism between Weyl-ordered operator
multiplication and *-convolution of kernel functions.
E Wigner (1932) [Wig321 introduces the eponymous phase-space distribution function
controlling quantum mechanical diffusive flow in phase space. It specifies the time evolution
of this function and applies it to quantum statistical mechanics. (Actually, Dirac (1930)
[DirSO] has examined a similar object for the electron density in a multielectron ThomasFermi atom, but interprets the negative values as a failure of his semiclassical approximation,
and dismisses the full quantum object.)
H Groenewold (1946) [Gro46]. (Based on Groenewolds thesis work.) A seminal but
somewhat unappreciated paper which achieves full understanding of the Weyl correspondence and produces the W F as the classical kernel of the density matrix. It reinvents
and streamlines von Neumanns construct into the standard *-product, in a systematic
exploration of the isomorphism between Weyl-ordered operator products and their kernel
function compositions. It further works out the harmonic oscillator WF.
J Moyal (1949) [Moy49] amounts to a grand synthesis: it establishes an independent
formulation of quantum mechanics in phase space. It systematically studies all expectation
values of Weyl-ordered operators, and identifies the Fourier transform of their momentgenerating function (their characteristic function) to the Wigner Function. It further interprets the subtlety of the negative probability formalism and reconciles it with the
uncertainty principle and the diffusion of the probability fluid. Not least, it recasts the time
evolution of the Wigner function through a deformation of the Poisson bracket into the
Moyal bracket (the commutator of *-products, i.e. the Weyl correspondent of the Heisenberg commutator), and thus opens up the way for a systematic study of the semiclassical
limit. Before publication, Dirac contrasts this work favorably to his own ideas on functional
28
29
30
Y Kim and E Wigner (1990) [KW90] is a classic pedagogical discussion on the spread of
wavepackets in phase space, uncertainty-preserving transformations, coherent and squeezed
states.
B Fedosov (1994) [Fed941 initiates an influential geometrical construction of the
*-product on all symplectic manifolds.
T Curtright, D Fairlie, and C Zachos (1998) [CFZ98] illustrates more directly the equivalence of the time-independent *-genvalue problem to the Hilbert space formulation, and
hence its logical autonomy; formulates Darboux isospectral systems in phase space; works
out the covariant transformation rule for general nonlinear canonical transformations (with
reliance on the classic work of P Dirac (1933) [Dir33]; and thus furnishes explicit solutions
to practical problems on first principles, without recourse to the Hilbert space formulation.
Efficient techniques for perturbation theory are based on generating functions for complete
sets of Wigner functions in T Curtright, T Uematsu, and C Zachos (2001) [CUZOl]. A
self-contained derivation of the uncertainty principle in phase space is given in T Curtright
and C Zachos (2001) [CZOl].
M Hug, C Menke, and W Schleich (1998) [HMS98]introduces and exemplifies techniques
for numerical solution of *-equations on a basis of Chebyshev polynomials.
31
REFERENCES
AW70
APW02
AWOO
Ant01
An97
Ar83
ACW98
Bak58
Bak6O
B584
Ba163
BRWK99
BBL80
Bar45
BM49
BKM03
BGLOl
BFF78
BF81
Ba79
B JY04
BC99
Ber80
Ber75
Ber77
BCG97
Ber02
BP96
BHS02
BHP02
BV94
Bon84
BCT82
BM94
BDW99
G Agarwal and E Wolf, Phys Rev D2 (1970) 2161; ibid 2187, ibid 2206
M Alonso, G Pogosyan, and K-B Wolf, J Math Phys 43 (2002) 5857 [quantph/0205041]
J-P Amiet and S Weigert, Phys Rev A63 (2000) 012102
J-P Antoine, J-P Gazeau, P Monceau, J Klauder, and K Penson, J Math Phys 42
(2001) 2349 [math-ph/0012044]
F Antonsen, [gr-qc/9710021]
W Arveson, Comm Math Phys 89 (1983) 77-102
N Atakishiyev, S Chumakov, and K B Wolf, J Math Phys 39 (1998) 6247-6261
G Baker, Phys Rev 109 (1958) 2198-2206
G Baker, I McCarthy and C Porter, Phys Rev 120 (1960) 254-264
N Balasz and B Jennings, Phys Repts 104 (1984) 347
R Balescu, Equilibrium and Nonequilibrium Statistical Mechanics (Wiley-Interscience,
New York, 1963)
K Banaszek, C Radzewicz, K Wddkiewicz, and J Krasinski, Phys Rev A60 (1999)
674-6 77
H Bartelt, K Brenner, and A Lohmann, Opt Commun 32 (1980) 32-38
M Bartlett, Proc Camb Phil SOC41 (1945) 71-73
M Bartlett and J Moyal, Proc Camb Phil SOC45 (1949) 545-553
I Bars, I Kishimoto, and Y Matsuo, Phys Rev D67 (2003) 126007
I Batalin, M Grigoriev, and S Lyakhovich, Theor Math Phys 128 (2001) 1109-1139
[hep-th/0101089]
F Bayen, M Flato, C Fronsdal, A Lichnerowicz, and D Sternheimer, Ann Phys ( N Y )
111 (1978) 61-110; ibid 111-151; Lett Math Phys 1 (1977) 521-530
F Bayen and C Fronsdal, J Math Phys 22 (1981) 1345-1349
F Bayen, in Group Theoretical Methods in Physics, W Beiglbock et all eds, Lecture
Notes in Physics 94 (Springer-Verlag, Heidelberg, 1979) pp 260-271
A Belitsky, X Ji, and F Yuan, Phys Rev D69 (2004) 074014
M Benedict and A Czirjbk, Phys Rev A60 (1999) 4034
F Berezin, Sow Phys Usp 23 (1980) 763-787
F Berezin, Comm Math Phys 40 (1975) 153-174
M Berry, Philos Trans R SOCLondon A287 (1977) 237-271
M Bertelson, M Cahen, and S Gutt, Class Quant Grav 14 (1997) A93-A107
P Bertet et al, Phys Rev Lett 89 (2002) 200402
B Biegel and J Plummer, Phys Rev B54 (1996) 8070-8082
M Bienert, F Haug, W Schleich, and M Raizen, Phys Rev Lett 89 (2002) 050403
P Bianucci et al, Phys Lett A297 (2002) 353-358
R Bishop and A Vourdas, Phys Rev A50 (1994) 4488-4501
J Gracia-Bondia, Phys Rev A30 (1984) 691-697
E Braaten, T Curtright, and C Thorn, Phys Lett B118 (1982) 115
A Bracken and G Melloy, J Phys A27 (1994) 2197-2211
A Bracken, H Doebner, and J Wood, Phys Rev Lett 83 (1999) 3758-3761;
J Wood and A Bracken, J Math Phys 46 (2005) 042103
31
32
BCW02
Bra02
BR93
Bra94
BD98
BAD96
CC03
CZ83
Car76
Cas91
CasOO
CH87
CH86
CV98
CL03
Coh95
Coh66
Coh76
cs75
CPPOl
CPPO2
CGB91
CG92
CUZOl
cz99
CFZ98
CFZm98
CZOl
cz02
DS82
DahOl
DS02
Dah83
33
DG80
DK85
DG02
deA98
DeG74
DBBO2
DV97
dew83
DPOl
Dir3O
Dir33
Dir45
DKM88
Dit9O
DM86
DNOl
DHSOO
Dun95
Dun88
EGV89
Fai64
FFZ89
FM91
Fan03
Fan57
FBA96
FZOl
Fed94
Fey87
FM03
FLS76
Fle90
Fo189
FraOO
33
34
Fre87
FMSOO
Gad95
GM80
GF91
GH93
Got99
GR94
GroOl
Gro46
Gro76
Hak99
HKN88
Han84
HarOl
HS02
HSD95
Hie82
Hie84
HOS84
HH02
Hor79
HL99
Hud74
HMS98
Hus40
Imr67
JBMO3
JS02
JVS87
JN90
JY98
JD99
KO00
KZZ02
KJ99
KL99
KLOl
34
35
KN9 1
KW90
KW87
KisOl
KKFR89
Ko196
KS02
KL94
Kon97
Kub64
Kun67
KPM97
Les84
Lea68
Lee95
Lei96
LPM98
Leo97
LFOl
LF94
LSUO2
Lie90
Lit86
Lou96
LVOOl
MS95
MM84
MMT96
MMMOl
MS96
MMP94
McD88
McC32
MOT98
MH97
Moy49
MM94
Y Kim and M Noz, Phase Space Picture of Quantum Mechanics, Lecture Notes in
Physics v 40 (World Scientific, Singapore, 1991)
Y Kim and E Wigner, Am J Phys 58 (1990) 439-448
Y Kim and E Wigner, Phys Rev A36 (1987) 1293; ibid A38 (1988) 1159
I Kishimoto, JHEP 0103 (2001) 025
N Kluksdahl, A Kriman, D Ferry, and C Ringhofer, Phys Rev B39 (1989) 7720-7735
A Kolovsky, Phys Rev Lett 76 (1996) 340-343
A Konechny and A Schwarz, Phys Repts 360 (2002) 353-465
H Konno and P Lomdahl, J Phys SOCJ p 63 (1994) 3967-3973
M Kontsevich, Lett Math Phys 66 (2003) [q-alg/9709040]; ibid 48 (1999) 35-72
[math.QA/9904055]
R Kubo, J Phys SOCJ p 19 (1964) 2127-2139
W Kundt, Z Nut Forsch a22 (1967) 1333-1336
C Kurtsiefer, T Pfau, and J Mlynek, Nature 386 (1997) 150
B Lesche, Phys Rev D29 (1984) 2270-2274
B Leaf, J Math Phys 9 (1968) 65-72; ibid 9 (1968) 769-781
H-W Lee, Phys Repts 259 (1995) 147-211
D Leibfried et all Phys Rev Lett 77 (1996) 4281
D Leibfried, T Pfau, and C Monroe, Physics Today 51 (April 1998) 22-28
U Leonhardt, Measuring the Quantum State of Light (Cambridge University Press,
Cambridge, 1997)
M Levanda and V Fleurov, Ann Phys (NY) 292 (2001) 199-231
M Levanda and V Fleurov, J Phys: Cond Matt 6 (1994) 7889-7908
B Lev, A Semenov, and C Usenko, J RUSSLaser Res 23 (2002) 347-368 [quantph/0112146]
E Lieb, J Math Phys 31 (1990) 594-599
R Littlejohn, Phys Rep 138 (1986) 193
P Loughlin, ed, Special Issue on Time Frequency Analysis: Proceedings of the IEEE
84 (2001) No 9
A Lvovsky et al, Phys Rev Lett 87 (2001) 050402
M Mallalieu and C Stroud, Phys Rev A51 (1995) 1827-1835
J Martorell and E Moya, Ann Phys (NY) 158 1-30
S Mancini, V Manko, and P Tombesi, Phys Lett A213 (1996) 1-6
0 Manko, V Manko, and G Marmo, in Quantum Theory and Symmetries: Krakow
2001 Proceedings, E Kapuscik and A Horzela, eds (World Scientific, 2002) [quantph/Oll2112]
M Marinov and B Segev, Phys Rev A54 (1996) 4752-4762
P Markowich, N Mauser, and F Poupaud, J Math Phys 35 (1995) 1066-1094
S McDonald, Phys Rep 158 (1988) 337-416
N McCoy, Proc Nut Acad Sci USA 19 (1932) 674
B McQuarrie, T Osborn, and G Tabisz, Phys Rev A58 (1998) 2944-2960; T Osborn,
M Kondrateva, G Tabisz, and B McQuarrie, J Phys A32 (1999) 4149-4169
W Mecklenbrauker and F Hlawatsch, eds, The Wigner Distribution (Elsevier, Amsterdam, 1997)
J Moyal, Proc Camb Phil SOC45 (1949) 99-124
S Mr6wczy6ski and B Muller, Phys Rev D50 (1994) 7542-7552
35
36
Mue99
Nag7
NO86
Neu3l
OW81
OC03
OR57
OM95
Pei33
PT99
QC96
Raj83
Raj02
Ran66
RTOO
Rie89
RA99
R092
Roy77
RGOO
Sam00
Sch02
Sch69
sw99
SSTOO
ss02
Sha79
She59
SP81
SMOO
Smi93
Sny80
Ste80
Str57
Sy182
Tak54
Tak89
Tat83
TayO1
36
37
TZM96
TA99
TD97
VG89
V089
V078
Vey75
vH51
W087
W088
Wey27
Wig32
Wis97
Wok97
W098
W082
woo2
Woo87
Yo89
ZacOO
zc99
Za103
ZP94
zu91
39
SELECTED PAPERS
1 H Weyl
J v Neumann
E Wigner
111
J Moyal
100
H Groenewold
91
167
T Takabayasi
202
40
G Baker
D Fairlie
10
N Cartwright
11
250
A Royer
12
244
253
13
14
41
15
16
17
369
416
R Feynman
in Essays in Honor of David Bohrn, B Hiley and F Peat, eds (Routledge and Kegan Paul,
London, 1987) pp 235-248
Negative Probability
426
18
19
B Fedosov
20
21
450
476
490
42
22
495
43
INDEX
Chebyshev polynomial, 13
Bohr ground-state orbit, 5
Born-Jordan prescription, 24
classical limit, 3, 8, 11, 24, 25
classical mechanics in Hilbert space, 24
classical trajectories, 14
Cohen classification, 24
coherent states, 13, 15, 16
Darboux transformations, 13
deformation, 6, 24, 26, 29
diffusion in phase space, 10
distribution function, 4
dressings, 25, 26
Ehrenfests theorem, 10
entropy, 3
field theory, 2
Gaussian, 3, 11, 12, 14, 25
Glauber-Sudarshan prescription, 24
Liouville potentials, 13
Liouvilles theorem, 10
marginal probability, 3, 15, 25
Morse potentials, 13
Moyal bracket, 6, 24, 26
Moyals equation, 5, 10
music score, 2
negative probability, 3
noncommutative geometry, 2
Poschl-Teller potential, 13
path integral, 21
perturbation, 17
positivity, 9
projective orthogonality, 7
spreading wavepacket, 15
*-product, 1, 2, 6, 10, 12, 19, 25, 26
*-genvalue, 7, 30
harmonic oscillator, 11
Husimi prescription, 24, 25
time-frequency analysis, 2
tomographic representation, 25
transition amplitude, 9, 16
turntable, 16
isospectral pairs, 13
uncertainty, 2, 9, 15
Kirkwood-Rihaczek prescription, 24
Laguerre, 11, 16
45
**
46
H. Weyl,
* nber eine neue Begriindung der Quantenmechanik, ZS. f. Phys. 40, 809,
1927; 44, 1, 1927. Vgl. farner P. A. M. D i r a c , Proc. Royal SOC.(A) 113, 621,
1937, und D. H i l b e r t , J. v. N e u m a n n , L. N o r d h e i m , nber die GrundIagen
der Quantenmechanik, Xath. Ann. 98, 1, 1927.
** Diese Verknupfung mit der Gruppentheorie liegt in ganz anderer Richtung
als die Untersuchungen von Herrn W i g n e r , die erkennen lassen, daB die Struktur
der Spektren nach ihrer qualjtativen Seite hin durch die bestehende Symmetricgrnppe bestimmt ist (mehrere Arbeiten in der ZS. f. Phys. 40, 492 und 883; 43,
624, 1926/1927).
47
1. Teil.
1. M a t h e m a t i s c h e G r u a d b e g r i f f e , d i e H e r m i t e s c h e n
F o r m e n betreffend. Die in der nberschrift angekundigien Grundbegriffe und -tatsachen stelle ich bier in der Nomenklatur der mehrdimensionalen analytischen G eometrie kurz zusammen. Das Abweichende
von der gewahnlichen vz-dimensionalen Beometrie liegt darin, da4 die
Komponenten der V e k t o r e n
F = (XI,
$2,
(1)
* * -1
1*
48
H. Weyl,
nicht nur reelle, sondern beliebige komplexe Zahlen sein konnen, und
daI3 als Q u a d r a t d e s B e t r a g e s eines Vektors dementsprechend die
,Hermitesche Einheitsform"
IF 12 = X I $ *
+ x 2 z 2 + ... + xnza
(2)
der Metrik zugrunde liegt (der Querstrich bedeutet den ubergang zur
konjugiert komplexen Zahl). Vektoren (1) werden in der ublichen W eise
mit Zahlen multipliziert und addiert. Sie bilden eine n-dimensionale
lineare Mannigfaltigkeit, den Vektorraum oder V e k t o r k b r p e r B,; d. h.
es lassen sich auf mancherlei Art n Vektoren e,: e,*, ..., e z so auswahlen,
daS jeder Vektor F auf eine und nur eine Weise in der Form
= x:eT
xge:
-.. x",:
sich darstellen lellt. Wird z. B. e: als der Vektor ei = (0, 0, . . ., 1, 0, ..., 0)
gewehlt (1 steht an i-ter Stelle), so fallen die ,Komponenten $' von
in bezug auf das Koordinatensystem (e:, e,: ..., et)" mit den ,,absoluten
Komponenten xi zusammen. Ein Koordinatensystem, in welchem das
Quadrat des Betrages von sich durch die Komponenten xi des willkurlichen Vektors F mittels der Formel (2) ausdriickt, heille normal.
A l l e n o r m a l e n K o o r d i n a t e n s y s t e m e s o l l e n als g l e i c h b e r e c h t i g t
g e 1t e n , das durch unseren arithmetischen Ausgangspunkt bedingte spezielle
Koordinatensystem (ei) sol1 unter ihnen seine ausgezeicbnete Stellung verlieren. In Zukunft bedeutet daher auch ei ein beliebiges normales KOordinatensystem, xi die darauf bezuglichen Komponenten des Vektors
Die Formeln fur den fibergang vom Koordinatensystem ei zu einem
anderen ei lauten allgemein:
+ +
(3)
Die Bedingungen, welche die Koeffizienten ei erfiillen mussen, damit
eine , u n i t a r e T r a n s f o r m a t i o n " vorliegt, welche zwischen zwei n o r m a l e n Koordinatensystemen vermittelt, sind leicht aus der Definition zu
ermitteln und entsprechen genau den a u s der elementaren analytischen
Geometrie gelainfigen. Wenn wir mit E die Matrix 11 eik I( bezeichnen
und der * das Transponieren einer Matrix, die Vertauschung von Zeilen
und Spalten bedeutet, 1 aber die die Identitat darstellende Einheitsmatrix,
so lauten sie:
-
BE* = E*E = 1.
Die Formeln (3) oder, wie ich jetzt lieber schreiben will:
(4)
49
haben bekanntlich noch eine zweite Bedeutung; sie stellen, unter Zugrundelegung des festen normalen Koordinatensystems der ei, eine u n i t a r e
A b b i l d u n g d e s V e k t o r r a u m e s auf s i c h s e l b e r dar, vermoge deren
dem Vektor 7 =
zi ei der Vektor 7' =
xi ei zugeordnet wird. Ich
bezeichne diese Abbildung kurz mit r' = T E . Dann druckt sich die
Zusammensetzung zweier Abbildungen
naturgemOB durch T'' = T ( E X ' ) aus - El E' folgen sich von links nach
rechts, wie wir zu lesen gewohnt sind -, und man befindet sich in
Einklang mit der ublichen Festsetzung des Matrizenkalkuls, nach
welcher aus
E = ezk 111 E' = II e;k II
II
+ + +
( ~ 9 )= ~ 1 5 i
sgY2
..*
ZnFn
einzufiihren. (97) ist das Konjugierte zu (q 9). Man wird zwei Vektoren
s e n k r e c h t aufeinander nennen, wenn ihr skalares Produkt verschwindet.
Zwei von 0 verschiedene Vektoren gehbren demselben S t r a h l an,
wenn der eine aus dem anderen durch Multiplikation mit einer (komplexen, von 0 verschiedenen) Zahl hervorgeht. Ein Strahl kann eindeutig bezeichnet werden durch einen ihm angehtirenden Vektor F vom
Betrage 1 (Einheitsvektor). Aber dieser ist seinerseits durch den Strahl
nicht eindeutig bestimmt, sondern an Stelle von F kann mit gleichem
Recht jeder Vektor E 7 treten, der aus ihm durch Multiplikation mit einer
beliebigen Zahl E vom absoluten Betrage 1 hervorgeht. Das ist wesentlich anders als im gewtihnlichen R a m , wo nur die Doppeldeutigkeit
1 ubrigbleibt. Fasse ich eine unitare Abbildung (4)
ekes Vorzeichens
auf nicht J s Abbildung des V~ktor-,sondern des Strahlenkijrpers (homogener Standpunlrt), so soll sie kurz eine D r e h u n g heiDen. E und E'
stellen dieselbe Drehung dar: ENE', wenn E' = E X ist; E bedeutet
dabei, wie im folgenden stets, einen Zahlfaktor vom Betrage 1.
50
ci
H.Weyl,
A @ )=
UiklC,.Xk,
(5)
i, k = l
=aik
oder
A" = A
(6)
(T'S) = A (T 9).
Die Grundtatsache fiir H e r m i t e sche Formen ist der S a t z v o n d e r
H a u p t a c h s e n t r ansf o r m a t i o n : Rin normales Koordinatensystem ei kann
zu A so gewghlt werden, dad i n ihm
(7)
A @ ) = a,z,Z,
a,z,Z,
f UnXnZn
wird. Die E i g e n w e r t e a l , aa, . . ., a, sind eindeutig durch die
H e r m i t e s c h e Form bestimmt (natiirlich nur bis auf die Reihenfolge).
Was die zugehirrigen H a u p t a c h s e n oder E i g e n v e k t o r e n ei betrifft,
so steht es mit ihnen i n Hinsicht der eindeutigen Bestimmtheit folgendermaoen. Seien etwa die Eigenwerte a,, a2, a3 einander gleich, = a, und
von den iibrigen verschieden. Dann gehijrt zum Eigenwert a der von
den Grundvektoren el, e,, e3 aufgespannte dreidimensionale Eigenraum 3(a),
der aus allen Vektoren T von der Gestalt x, el
xae,
x3 e, besteht; in
ihm ist (ell e,, e,) ein normales Koordinatensystem. Die zu den
- a *
51
Quantenmechanik und Gruppentheorie.
n u m e r i s c h v e r s c h i e d e n e n Eigenwerten a', a", .. . gehorigen Teilrgume W (a'), 8 (a"), .. ., die gegenseitig aufeinander senkrecht stehen,
sind durch A eindeutig determiniert; in jedem von ihnen kann aber das
normale Koordinatensystem willkurlich gewshlt werden. Das letzte
bedeutet in dem angenommenen Beispiel, daO zl, za,x3 untereinander noch
einer beliebigen unitaren Transformation unterworfen werden konnen,
o b e dal3 die Normalform (7) zerstort wird.
Zwei Hermitesche Formen A , B lassen sich dann und nur dann
s i m u l t a n auf Hauptachsen transformieren, wenn die Koeffizientenmatrizes vertauschbar sind: A B = B A. Ein entsprechender Satz gilt
fur mehr als zwei Hermitesche Formen, ja fur irgend eine endfiche oder
unendliche Gesamtheit solcher Formen.
8 2.
D e r p h y s i k a l i s c h e Begriff d e s r e i n e n F a l l e s . Ich
exemplifiziere am Beispiel des magnetischen Elektrons, weil hier sehr
einfache, aber vom klassischen Standpunkt paradoxe Verhdtnisse vorliegen. Nach der Annahme von G o u d s m i t und U h l e n b e c k , die sich
seither bestens bewahrt hat, mu0 man dem Elektron ein eigenes I m p u l s moment zuschreiben, dessen Komponente G~ in einer beliebigen Richtung,
etwa der z-Richtung, nur der beiden Werte
1 und - 1 fiihig ist, wenn
h / 4 x als Einheit zugrunde gelegt wird. Man kann sich vorstellen, daD aus
einem gegebenen Elektronenstrom, durch ein Verfahren analog dem bekannten S t e r n - G e r 1a c h schen Experiment zum Nachweis der Richtungsquantelung bei Atomen, der Schwarm derjenigen Elektronen ausgesondert
wird, fir welche 6, den Wert
1hat. Die Elektronen dieses Schwarms (-5%
mogen keine Storung erfahren, so daB fur sie alle dauernd mit Sicherheit 6, den Wert
1 besitzt. In einem solchen Elektronenschwarm
haben wir (wenn Wjr noch von Ort und Geschwindigkeit der Elektronen abstrahieren) einen , r e i n e n Fallu vor uns: er ist von einer
inneren Homogenitat, die prinzipiell nicht mehr gesteigert werden kann.
Denn alle physikalischen Fragen, welche sich sinnvoll mit Bezug auf ihn
stellen lassen, finden eine von v o r n h e r e i n a n g e b b a r e n u m e r i s c h
b e s t i m m t e Antwort. Solche Fragen sind allein die folgenden: 1st r
irgend eine Richtung, mit welcher Wahrscheinlichkeit hat fur ein Elektron des G,,-Schwarms die Gr6l3e 6,. den Wert
1 oder - l ? Die
numerisch bestimmte Antwort lautet: Wenn 6 der Winkel ist, den die
r- mit der x-Richtung bildet, so sind die beiden Wahrscheinlichkeiten bzw.
6
= cosaT
und
6
2
= sina -.
52
H. Weyl,
Die Wahrscheinlichkeit ist als Hsufigkejt im Elektronenschwarm zu verstehen; sie wiirde sich, wenn mit dem Schwarm das Aussondernngsexperiment in der r-Richtung vorgenommen wiirde, in dem Searkeverhhltnis der beiden Teilstrahlen bekunden *. Hlltten wir am Anfang statt
der x- eine andere, die x-Richtung zugrunde gelegt, so hstten wir einen
anderen reinen Fall, den Elektronenschwarm Gzt bekommen. I n ihm
4
hat <iz mit der Wahrscheinlichkeit cosa- den Wert
1, mit der Wahr2
6
scheinlichkeit sina- den Wert - 1, wenn 4 = (r,x) ist; insbesondere
2
hat 6% rnit Sicherheit den Wert
1. Dieser reine Fall ist von dem
ersten v e r s c h i e d en , weil die gleichen physikalischen Fragen hier andere
numerische Antworten finden. Es gibt so viele verschiedene reine F U e ,
wie es verschiedene Richtungen x gibt. Wir kiSnnen aus solchen reinen
StrBmen Ejz, Gzt, .. . Mischungen in irgend einem Verhaltnis herstellen.
Die Haufigkeit, rnit welcher in einem solchen Nischstrom ein 6, = 1
oder - 1 ist, hiingt von dem Xischungsverhaltnis ab. Wir sind hier
umgekehrt darauf angewiesen, aus den experimentell beobachteten Haufigkeiten Schliisse auf die Konstitution des Nischstromes zu ziehen. Der
Unterschied zwischen reinem Fall und Mischung, den ich hier aufstelle,
ist analog zu den biologischen Begriffen der ,reinen Linie (innerhalb
der reinen Linie gelten die M endelschen Vererbungsgesetze) und der
,Populationu (auf welche sich clie Gesetze von G a l t o n bezogen). Hier
wie dort ist es eine wichtige Aufgabe der Experimentierkunst, reine
Linien zu isolieren. Die Unterscheidung : Theorie der reinen Bile
einerseits, Statistik der Gemenge andererseits, scheint mir fundamental fur
die richtige Erfassiing des Sinnes der Quantenmechanik.
An dem Tatbestand, die Elektronenschwsrme betreffend, wie er
bisher beschriebeo wurde, ist nichts Paradoxes. Statt vom Schwarm
spreche ich in Zukunft vom einzelnen Elektron und demgemsili von Wabrscheinlichkeit statt von Hauufigkeit. Etwas Paradoxes liegt erst in der
Aussage, dafi <is die Komponente eines gewissen Vektors, des Impulsmomentes, in bezug auf die x-Richtung ist. Denn dies involviert doch,
wenn wir ein rechtwinkliges Koordinatensystem s y s im Raume ein-
* Obwohl also 6%noch wieder zerlegt werden kann, sin$ doch die so entstehenden Teilstrahlen nicht homogener a19 Gz selbst. Das ist genau wie bei
einem Lichtstrahl, der durch zwei gegeneinander verdrehte Nicols hindnrchgegangen ist: er ist von derselben Beschaffenheit wie Licht, das nur durch den
zweiten Nicol hindnrchging.
+ hG,
hat,
(8)
Wie vertragt sich das mit dem Umstand, daD 6, so gut wie G,, G ~ G,
, nur
1 f i i g ist? Aber in einem vorliegenden reinen Fall haben
der Werte
die hier auftretenden GroSen uberhaupt keine mit Sicherheit angebbaren
Werte, so daI3 zunachst der Sinn der Gleichung (8), wenn er in der
ublichen Weise auf die Wer t e der physikalischen GrbIjen bezogen werden
SOU,ganz im Leeren h k g t . Sie wird einen Inhalt erst gewinnen, wenn
wir die physikalischen Grblen durch solche mathematische Entitaten
darstellen, welche Nultiplikation mit reellen Zahlen und Addition untereinander zulassen. - Und was soll es zweitens heilen, daD dieser Vektor
mit den Komponenten 6,, 6y,6, ,,Impulsmomentu ist? Damit wird
offenbar ein bestimmtes Verhalten dieser GrbDen gegeniiber einem das
Elelitron einbettenden Nagnetfeld (H,, H,, H,) ausgesagt. Wenn wir
uns das Elektron ganz naiv als ein rotierendes Kiigelchen vorstellen, in
welchem das Verhsltnis von Ladungs- und Massendichte uberall konstant
ist, so ergibt sich in der Hamiltonschen Energiefunktion die HBfte
des Terms
y (Hz bz -I- H y 6 ~$. Hz 62))
(9)
872
+CGz.
54
10
H. Weyl,
bedeutet. Ich nehme dabei die Dimensionszahl 'y1 des Vektorraums, die
Zahl der Variablen x, endlich, obschon sie in den meisten Fallen unendlich groD ist. Alles Gesagte 1'it sich aber analogisch auf den unendlich dimensionalen Vektorraum iibertragen. Im oben besprochenen
Beispiel des Elektrons ist, wie sich zeigen wird, .n = 2.
Der einzelne reine F a l l wird durch einen Vektor
Tom
B e t r a g e 1 i n u n se re m !?a-dimensionalen V e k t o r r a u m g e g e b e n ,
die einzelne physikalische GriiSe CG wird reprgsentiert durch eine
Hermitesche Form A in diesem Raume. Mittels Einfiihrung eines
geeigneten normalen Koordinatensystems el, e,, . . ., en bringe man A(?)
auf Hauptachsen :
A(?) = a,z,Z:,
+ anxnZn
(7 = I, el + x, e, + ... + sn en).
+ aaz3Z,f
* - -
(10)
+ la
55
Quantenmechanik und Gruppentheorie.
11
Es ist klar, da5 man Hermitesche Formen addieren und daD man
sie mit reellen Zahlen multiplizieren kann, ohne dadurch aus ihrem Be,-$& herauszutreten. Die kalkulatorischen Anforderungen, die wir am
SchluB von 9 2 erhoben, sind erfiillt.
Wenn die Werte, deren die physikdische Grate a fahig ist, sehr
dicht Iiegen oder gar eine kontinuierliche Skale bilden, wird man nicht
fragen nach der Wahrscheinlichkeit, mit welcher sie einen bestimmten
Wert annimmt, sondern rnit der sie in ein bestimmtes W e r t i n t e r v a l l
a 5 a 5 a' hineinfdlt. Nach unserer Anweisung haben wir dann im
Hauptachsensystem diejenigen Eigenvektoren ei aufzusuchen , deren zugehiirige Eigenwerte ai in jenes Intervall hilieinfallen ; sie spannen den
Teilraum 8;' auf. Die gesuchte Wahrscheinlichkeit ist die auf diesen
Teil der Indizes i sich erstreckende Summe
C
Xi&
(11)
der quadrierte Betrag der senkrechten Projektion des den reinen Fall
Die Formen (11) sind
darstellenden Vektors F auf den Teilraum %.;'
es, welche v. N e u m a n n a. a. 0. als ,,Einzelformen" E$ einfuhrte.
Liegen mehrere Gro5en a, p, . . . vor, deren zugehorige H e r m i t e sche Formen v e r t a u s c h b a r e Koeffizientenmatrizes besitzen, so lassen
sie sich alle simultan durch Einfuhrung eines geeigneten normalen
Koordinatensystems e, auf Hauptachsen transformieren. Die korrespondierenden Eigenwerte zu ei mogen ail hi, . . . hei5en. F = ei stellt einen
reinen Fall vor, in welchem jede der betracbteten GroOen m i t Sicherheit
eben bestimmten Wert hat, namlich a den Wert a,,
den Weft bi usw.
Die klassische Physik nimmt an, da5 es sich fur a l l e GriiBen so verhiilt,
und sie la5t nur die reinen Falle e,, e2, .. ., e,,
die besonders ausgezeichnet sind und in denen alle GroBen einen bestimmten Wert haben,
als rebe Falle zu und fafit die anderen bereits als Gemenge von ihnen
auf. Sobald aber zwei physikalische Grij5en auftreten, deren Matrizes
nicht vertauschbar sind , entfallt diese Moglichkeit : In einem reinen
Falle, in welchem die erste Gr6De einen m i t Sicherheit angebbaren M'ert
hat, bestehen fur die Werte der zweiten GriTDe nur Wahrscheinlichkeiten.
Das ist in Einklang mit H e i s e n b e r g s Anschauungen, wie er sie kiirzlich in dieser Zeitschrift (43, 172, 1927) entwickelte.
56
12
H. Weyl,
(12)
oder als Hermitesche Formen geschrieben:
2, Zl - 5,s2,
x15, 5,q, i (zlz, - 2, Z,).
Jede yon ihnen, ja auch das zu einer beliebigen anderen Richtung r mit
den Richtungskosinus a, b, c (aa ba ca = 1) gehorige
+ +
S, = as,
+ b S , + cS,
(1
I/
(13)
;'-ic,
B +i ca
hat die Eigenwerte
1. Der reine Fall, bei welchem cix mit Sicherheit
den Wert
1 hat, ist durch den Vektor el gegeben. Lra reinen Fall
T = (zl,za) sind die Wahrscheinlichkeiten iir Gz = &1 bzw. gleich
] s l l a l x21a. Wir suchen eine Richtung r auf, deren zugehoriges G, in
diesem Falle mit Sicherheit den Wert f 1 hat, d. h. fur welche der
Vektor (sl1 s,) in die zum Eigenwert
1 geharige Hauptachse von
s, fitllt:
(7J
- ic)
- a 2, = 2+.
sl:zs=b+ic:l-a=
l+a:b-iic.
a ist der Kosinus des Winkels 6 zwischen der 1'- und der s-Richtung.
Wir finden
ls,la:l$g(a= b a + c a : j l - u ) a =
l-aaa:(l-a)a,
= 1 + a : 1-a
4
= ~ o s4~ ~ : s i n a ~ .
Darin liegt keinerlei neuer Ansatz. Wenn das Gemenge ein games
Kontinuum reiner F d l e enth<, verwandeln sich die Summen in Integrle.
57
13
12
- -n( a i l
+ + *..+
am).
Die Summe der Glieder in der Haugtdiagonale, die S p u r der H e r m i t e schen Form, stellt sich iibrigens dadurch als eine Invariante gegeniiber
unitiiren Transformationen heraus.
Ein weiteres Problem dieser Art ist das folgende: a und p seien
zwei bzw. durch A und B reprilsentierte Grollen. Ich fiihre die beiden
aus den Eigenvektoren el und e: von A bzw. B bestehenden normalen
Koordinatensysteme ein :
A ( T )=
2 azsiiiil B ( T )= Ct t+xriif
=2
2xFe;),
<
<
58
14
H. Weyl,
(14)
Die Wahrscheinlichkeit, mit der in einem reinen Fall
die GriiSe p
einen der Werte b,, b,, ..., b, annimmt, ist andererseits gegeben durch
die Einzelform
x%Tzr=
u
F! =
x f i k x i z k ;
i, k = l
i=l
Nach unserer Anweisung hat man in 2':' alle Variablen xi auller den
ersten 8 gleich Null zu setzen und dann uber den Teilraum
zu
mitteln. Dabei ist
l/e (fur i = 7E - 8 )
(Xizk) =
0 (fiir alle anderen Paare i, k )
oder
1
(Xi$
= -eki
(i, 7c = 1, 2, . . ., w).
<
So kommt
Halt man das Interval1 a d fest und will nur die r e l a t i v e n Wahrscheinlichkeiten miteinander vergleichen, die verschiedenen Intervallen b b' ent1
sprechen, so kann man den konstanten Faktor - weglassen. Die Summe
8
rechts ist die Spur von E$F!'. Weil einer Hermiteschen Form die
Abbildung mit derselben Koeffizientenmatrix unitar-invariant assoziiert
ist, hat neben der Spurbildung auch die Zueammensetzung der Matrizen
von H e r m i t e schen Formen einen invarianten Sinn. Infolgedessen genugt es, die Einzelformen Ef,Fbb) in ir.gend e i n e m normalen Koordinatensystem zu kennen, um daraus die gesuchten relativen Wahrscheinlichkeiten vermittelst der Formel
w = Spur (E:'.P:)
zu finden. Diese Art von Fragen iiber Gemenge zieht v. Neumann
a. a. 0. allein in Betracht. Sein Schluhesultat ist mit unserem naturlicb
inhaltlich identisch, aber seine Formel ist komplizierter. I n der ganzen
59
Quantenmechanik und Gruppentheorie.
15
8 5.
g b e r G r u p p e n und i h r e u n i t a r e n D a r s t e l l u n g e n . Fur
die unitiiren A b b i l d u n g e n gilt ein analoges Theorem, wie das von
der H a u p t a c h s e n t r a n s f o r m a t i o n der Hermiteschen Formen: Zu
einer gegebenen unitairen Abbildung la0t sich ein solches normales
Koordibatensystem ei finden, in welchem die Abbildung durch die Gleichungen
z; = eksk
(15)
wiedergegeben wird. Die E i g e n w e r t e ek sind Zahlen vom absoluten
Betrag 1, ihre Phasen y k , ek = ezVk, heil3en die D r e h w i n k e l der uni&en Abbildung. Analoge Bemerkungen, wie fur die Hauptachsentransformation der H e r m i t eschen Formen, greifen Platz betreffs der Eindeutigkeit, mit welcher Eigenwerte und Eigenvektoren bestimmt sind,
sowie betreffs der simultanen Uberfiihrung mehrerer unitiier Abbildungen
in die Normalform (15).
Bus einer G r u p p e u n i t a r e r A b b i l d u n g e n abstrahiert man das
Gruppenschema, indem man die Abbildungen zu Elementen gleichgiiltiger Beschaffenheit degradiert und nur auf die Art ihrer Zusammensetzung achtet. Die abstrakte Gruppe ist a.lso ein System von Elementen,
innerhdb dessen durch ,,Kompositionn:aus zwei Elementen a, b in bestimmter Reihenfolge ein Element a b des Systems entspringt ; in solcher
Weise, dal3
60
16
H. Weyl,
Die Gruppe der unibren Abbildungen erscheint dann ds eine Verwirklichung oder D a r s t e l l u n g der abstrakten Gruppe, welche dadurch zustande kommt, daI3 jedem Gruppenelement s eine unitire Abbildung U(s)
in solcher Weise zugeordnet ist, da13 allgemein
U(s) U ( t ) = U(St)
(1 6)
gilt [es folgt daraus sofort U(1) = I]. Da das Gruppenschema aus der
Darstellung abstrahiert wurde , ist die Darstellung g e t r e u , d. h. verschiedenen Elementen entsprechen verschiedene Abbildungen U, oder,
was dasselbe besagt, V ( S )ist = 1 nur frir s = 1. Die Gruppe der unitaren Abbildungen ist r e d u z i b e l , wenn in dem Vektorraum %, in
welchem sich die unitiren Abbildungen abspielen, ein h e a r e r Unterraum 8, existiert mit einer Dimensionszahl n,
0, aber
R , der
gegeniiber allen U(s) invariant ist. Die Vektoren, welche zu allen in
gelegenen senkrecht sind, bilden einen linearen Unterraum
;,
und es ist & = %Irnf % n- m in dem Sinne, daD sich jeder Vektor auf
eine und nur eine Weise in zwei Komponenten spalten lafit, von denen
die erste !&, die zweite Rn-m angehiirt. Weil die U(s) unitire. Transinvariant: Die Darforma.tionen sind, lassen sie auder !& auch 8n-.rn
stellung zerf a l l t in eine mdimensionale und eine (n - m)-dimensionale.
Wshlt man das normale Koordinatensystem e, so, d d die ersten m Grundvektoren den Raum Brnaufspannen, die letzten n-rn aber den Raum
Rn- m , so kommt dieser Zerfall an den Koeffizientenmatrizen V(s) unmittelbar zum Ausdruck. Man kmn sich danach auf die Ahfsuchung
der i r r e d u z i b l e n D a r s t e l l u n g e n beschriinken. Fur irreduzible Darstellungen gilt der wichtige Satz: 1st die unitire Matrix A mit allen
U(s) vertauschbar: A U(s) = U(s)A, so ist A = E 1 Multiplum der Einheitsmatrix 1. [Es ist dabei sogar unwesentlich, daS die U ( s ) eine Gruppe
bilden.]
Wir denken in erster Linie an e n d l i c h e G r u p p e n . Zu jeder
Gruppe gehort eine bestimmte , G r i i D e n a l g e b r a U . Eine GriiDe im
G r u p p e n g e b i e t wird dadurch gegeben, daJ3 jedem Gruppenelement s
eine Zahl t ( s ) zugeordnet wird. Die GroDen haben demnach so viele
Zahlkomponenten, wie es Gruppenelemente gibt, sie sind sozusagen die
Vektoren im Gruppenraum, in dem jedes Element eine Dimension, einen
Grundvektor hedeutet". Die GriiSe g mit den Komponenten g(s)! die
l ( s ) . s bezeichnet werden mag, erscheint in
danach symbolisch mit
>
am
<
an-
61
17
(17)
8
(18)
Die Summe in (18) erstreckt sich uber alle Paare von Eiementen t, t',
deren Kompositum tt' = s ist. Man kann sie als einfache Summe iiber
aJJe Gruppenelemente t, aber weniger symmetrisch auch so schreiben:
(18) ist also d a s M u l t i p l i k a t i o n s g e s e t z d e r GroDen im Gruppengebiet. Die GroDen kiinnen danach, in genauer Anschmiegung an die
zugehorigen Matrizen , ad dier t werden, m i t 2 ah1en mu 1t i p l i z i e r t
und u n t e r e i n a n d e r m u l t i p l i z i e r t werden; in solcher Weise, daD die
wichtigsten algebraischen Axiome erfiillt bleiben. ( N u das kommutative
Gesetz der Multiplikation und das Axiom, welches Nullteiler ausschliellt,
gelten nicht.)
Die GroDe 6 h e s t r e e l l , wenn &re Komponenten der Gleichung
E(s-9 =
geniigen.
(19)
(19)
Denn aus
U ( S ) U ( S - ~=
) 1 zusammen mit U*(s) V(s) = 1
folgt 8*(s) = U(s-l). Darum gilt, unter der Voraussetzung (19),
8
Den Bereich der reellen Graden verlaBt man nicht durch Addition, Multiplikation der GrtiDen untereinander und durch ihre Nultiplikation mit
reellen Zahlen*.
Fiir iins kommen vorzugsweise die A b e l s c h e n G r u p p e n in Betracht, bei welchen die Komposition der Elemente kommutativ ist :
* Von der natiirlichen nnd wichtigen Rolle, welche diese Begriffe in der
Darstellnngstheorie spielen, die sich nachher auch als die grundlegenden in der
Quantenmechanik herausstellen werden, kann man nnr einen Eindruck gewinnen
dumb das Stndinm dieser Theorie. Es sei insbesondere verwiesen auf: F. P e t e r
and H. W e y l , Math. Ann. 97, 737, 1927.
Zeitachrift fur Phyeik. Bd. 46.
2
62
18
H. Weyl,
2'
= E (s)
. %.
U ( s ) U(t) C Z U(st).
(21)
19
u (s)
U ( t ) = 6 (s, t) U(St),
wo 8 eine von s und t abhangige Zahl vom absoluten Betrag 1 ist. Die
b g a b e einer GroDe im Gruppengebiet ist relativ auf die benutzte Eichung ;
wird die Eichung gemgo der Formel U ( s ) --f E (s) U ( s ) verandert, so
mussen die Komponenten 6 (s) jeder Gr6lje t ersetzt werden durch E-1 (s)t (s).
Das Hultiplikationsgesetz lautet
5 (s) =tt'r=
,86 (4 t'>6 (t)'I(0.
Die Beschreibung (19) der reellen Gr6Den ist nur dann zutreffend, wenn
die Eichung SO eingerichtet wurde, da13 U(s-1) = U-1 (s) ist. Fur eine
irreduzible Darstellung gilt nach wie vor der Satz : 1st die feste Drehung A
mit allen U ( s ) vertauschbar, A-1 U(s)A = U(s), so ist A? 1.
Die eindimensionalen Darstellungen verlieren jetzt jedes Interesse ;
denn die einzige eindimensionale Drehung ist die Identitkit. Aber im
gegenwartigen S b n e gibt es auch fur A b e l s c h e G r u p p e n m e h r dimensionale i r r e d u z i b l e D a r s t e l l u n g e n . Nicht freilich, wenn die
Abelsche Gruppe z y k l i s c h ist, aus den Wiederholungen eines einzigen
Elementes a besteht :
1, a, aa,. . .,ah-1
(ah = 1).
Denn ist A die a korrespondierende Matrix in der Darstellung, so ist
h-
(22)
64
20
H. Weyl,
Die Eichung ist so gewiihlt, daD Ua((a) oder U(u) U(a-1) = 1 ist und
Entsprechendes fir die iibrigen Elemente gilt. Die ,,reellen Gr61Jenu
Q l f b f r l b S D
(23)
sind also jene, deren Komponenten Q,El 7, g reelle Zahlen sind. Ihre
Algebra ist die einfachste nicht-kommutative, welche existiert : die der
Q u a t e r n i o n e n (genauer derjenigen Quaternionen, von denen die
skalare Eomponente reel1 ist, die drei vektoriellen rein imaginir). I n der
Darstellung 'illerscheint die Gr6De (23) als die Matrix
AB = EBA.
(24)
Wir haben uns zu uberlegen, in welcher Weise sie eri.illt s e b kann.
Bildet man auf beiden Seiten die Deterrninante, so ergibt sich E~ = 1,
E ist also eine n-te Einheitswurzel. Ferner erhalt man durch Induktion
fur 7c = 1 , 2 , 3 ,...:
AkB = $BAk,
(25)
ebenso
AB1 = E~BIA.
Kombiniert man beide Gleichungen, indem man die zweite auf Ak und B
statt auf A und B anwendet, so erhalt man die allgemeinere Regel
AkBE = ,$'BI.Ak.
(26)
Weiter notieren wir die Gleichung
k(k+l)
(AB)k =
.BkAk.
(27)
65
21
d u z ib le n A b e l s c h e n D r e h u n g s g r u p p e i n a D i m e n s i o n e n s i n d
demn,aoh T e i l e r v o n n.
Liegt eine endliche Abelsche Gruppe in abstract0 vor, (20), so wird
man zur Aufsuchung ihrer getreuen irreduziblen a-dimensionalen Darstellungen folgendermaDen verfahren. Fur jedes Basiselement, z. B. a1 = a
yon der Ordnung h, = 12, eicht man U ( a ) 3 A in solcher Weise, d d
~h = 1 ist. Nachdem dies geschehen, eiche man U ( s ) fur das Element ('20)
durch die Festsetzung
V(s) = A,Z1 A?. . . A;f.
&A, = 6ikAkAi
Da aus ('28)
',htAk
das ist
(i> k; i,
= 'ik
JC
= 1, 2, . . .,f)
&tk
(28)
Ahl,
i
A , = shiA
ik k
(29)
Die Forderung, daD
Gleichung aus
66
22
H. Weyl,
Die H e r m i t e s c h e Form auf der linken Seite kann aber nur dann
identisch in den st verschwinden, wenn alle ihre Koeffizienten Xu11 sind.
So ergeben sich die Bedingungen der scbiefen Symmetrie
C" = - c.
Cki = - Cik,
Setzt man C = iA,so ist A eine Hermitesche Matrix. Resultat: Mit
j e d e r H e r m i t e s c h e n F o r m A i s t i n u n i t l r - i n v a r i a n t e r Weise
(vgl. 1) d i e i n f i n i t e s i m a l e u n i t a r e A b b i l d u n g
verbunden. Der Satz von der Hauptachsentransformation der H e r m i t e schen Formen stellt sich dadurch als der infinitesimale Grenzfall des
entsprechenden Theorems fur unitare Abbildungen heraus. Diejenigen
infinitesimalen Abbildungen. welche alie Strahlen ungeandert lassen,
haben die Form
aF = icy,
az
mit einem reellen Zahlfaktor c. Der homogene Standpunkt verlangt also
hier, daD A nicht von A
c 1 unterschieden wird.
Indem man in jedem Zeitelement d z die gleiche infinitesimale
unitare Abbildung (29) wiederholt, erhalt man durch Integration von (29)
a(z) = r U ( z ) .
I n jedem
Die Konvergenz kann ebenso leicht bewiesen werden wie im eindimensionalen Fall, wenn C eine Zahl ist. Auch ergibt sich die Potenzreihe
z
za
U ( t ) = l + - Cl !+ - C ' 2+!
a * . .
(30)
67
Quantenmechanik und Gruppentheorie.
23
cine andere Methode ist die sukzessive Approximation ; sie setzt nicht
voraus, daD C von z unabhangig ist. Als nullte Approximation wird
das fur z = 0 vorgegebene T genommen, allgemein wird die 1-te aus der
l)-ten Anniihemg mittels der Gleichung
(7.
ZU
(31)
Bei zeitunabhangigem C kommt wieder die Gleichung (30) heraus.
E i n e Z w i s c h e n b e m e r k u n g : Es wurde erwiihnt, daO in der Physik
meist F o r m e n m i t u n e n d l i c h v i e l e n Varia.blen eine Rolle spielen.
Die Theorie der H e r mi t e schen Formen von unendlich vielen Veranderlichen unter dem EinfluD unitirer Transformationen wurde von H i l b e r t
und H e l l i n g e r entwickelt, unter der Voraussetzung, daO die Form beschrankt ist, d. h. dad eine Konstante M existiert, unter der die Werte
der Form ihrem absoluten Betrage nach fur alle Vektoren vom Betrage 1
bleiben". Die in der Physik vorkommenden Formen geniigen dieser
Bedingung nicht. Eine Erweiterung der Theorie, welche den physikalischen
Anforderungen geniigt, hat v. Neumann a. a. 0. in Aussicht gestellt.
Es ergibt sich hier die Aufgabe, das Analoge fur die unitaren Abbildungen
zu leisten. Fur sie wird die Theorie wesentlich befriedigender ausfallen,
weil keinerlei spezielle, die Konvergenz garantierende Voraussetzungen
zu machen sind, wie es die Hilbertsche Annahme der Beschranktheit
war. Denn der Begriff der unitaren Abbildung bringt es mit sich, daD
in der Matrix die Quadratsumme der absoluten Betrage jeder Zeile und
jeder Spalte konvergiert, namlich = 1 ist. (Die mathematische Durchfiihrung soll an anderer Stelle gegeben werden.) Der integrale Standpunkt
ist in begrifflicher Hinsicht dem infinitesimalen immer uberlegen, er liillt
zugleich die naturlichen Grenzen der differentiellen Begriffsbildungen
erkennen. In diesem Sinne ist es zweckmaBig, da rnit einer GroDe a ja
h e r auch &re reellen konstanten Multipla 3Ga als physikalische GriiBen
auftreten, diese zu ersetzen durch etka, die H e r m i t e schen Matrizen k A
68
24
H. Weyl,
durch die unitiren etkA. Mit A erscheinen sie zugleich auf Hauptachsen
transformiert, wobei an Stelle der aP die Zahlen egk% als Eigenwerte sich
ergeben.
Doch nun zu den u n e n d l i c h e n G r u p p e n ! Eine unendliche
Gruppe kann diskontinuierlichen Charakter haben, wie die in der Lehre
von der Eristallstruktur auftretende Grnppe der Gittertranslationen des
Raumes, deren Komponenten in bezug auf die drei Achsen 5, y, a game
Zahlen sind. Es kBnnen auch gemischte kontinuierlich-diskrete Gruppen
vorkommen, wie die Gruppe aller Raumtranslationen, deren x-Komponente
ganzzahlig ist. Doch haben wir jetzt insbesondere die kontinuierlichen
Gruppen im Auge. Eine solche denkt man sich nach S. L i e erzeugt
durch ihre i n f i n i t e s i m a l e n Elemente. 1st die Gruppe eine f-parametrige kontinuierliche Mannigfaltigkeit 8,so sind die infinitesimalen
Elemente die Stellen auf der Gruppenmannigfaltigkeit, welche der
Einheitsstelle 1 unendlich benachbart sind, oder die von 1 ausgehenden
Linienelemente. Sie bilden also eine f - dimensionale l i n e a r e Mannigfaltigkeit. Halten wir uns sogleich an die Darstellung, an die konkreten
unitiren Abbildungen statt an die abstrakten Elemente, so haben wir
mithin eine f-dimensionale lineare Schar schiefer Matrizen vor uns :
(32)
innerhalb deren C,, C,, . . ., C, e k e willkiirlich gewiihlte Basis ist und
die Zahlparameter dd,, d 6 9 , . , ., asf aller reellen Werte fabig sind.
Setzt man in (32) d d i = aids und iteriert diese infiniteshale Abbildung, die man sich im Zeitelement d z vollzogen denkt, so gelangt man
nach Ablauf der Zeit z, wenn an Stelle von q z jetzt wieder cii geschrieben wird, zu
g1ci + 02 c2 t ...tf C f .
u(d,, 6 2 , * * - 9 6,) =
(33)
69
25
zu]: Gruppe. Fuhrt man Pie beide hintereinander aus, das eine Ma1 in
der Reihenfolge d,d', das andere Ma1 in der Reihenfolge d',d, so ist die
Differen2 der dadurch aus F entstehenden Vektoren
A&!= d d ' r - d ' d t = r ( C C ' - C'C) d z d z ' .
Diese infinitesimale Abbildung ist der gesuchte Kommutator. InnfoIgedessen muD mit C und C' auch immer CC' - C'C der Schar g angehiiren.
An der Basis formuliert, heifit das, daO die Matrizen C,Ck - CkC, sich
linear mittels reeller Zahlkoeffizienten aus C,, C,, . . ., Cf kornbinieren
miissen. Diese von L i e aufgestellte Bedingung, deren Herleitung leicht
streng zu machen ist, ist nicht nur notwendig, sondela auch hinreichend *.
Die Gruppe ist A b e 1s c h , wenn der Kommutator irgend zweier
Elemente gleich 1 ist. I n diesem Falle mussen die Yatrizen C, den
Gleichungen
tick- ckc$= 0
genugen, d. h. sie miissen vertauschbar sein.
Natrizen A und B gilt
eA+B
(34)
Fur zwei vertauschbare
Ie A . e B ;
das ergibt sich genau wie fur Zahlen. Die Gleichung (34), d. i. die
Vertauschbarkeit der infinitesimalen Elemente, geniigt also, wie das
eigentlich selbstverstandlich ist, um den A belschen Charakter der ganzen
Gruppe sicherzustellen, es gilt auf Grund von (34)
* - - i zf) = u(6,
Z1,GS
zz1- - -,6, f tf).
Jede f-parametrige A belsche Gruppe ist danach isomorph mit der Gruppe
der Translationen in einem f-dimensionalen Raume. Die C, spielen eine
analoge Rolle wie die Basis bei den endlichen Abelschen Gruppen.
Wir werden es zwar mit einer Abelschen Gruppe zu tun haben,
aber die Abbildungen sind als solche des S t r a h l e n k o r p e r s zu verstehen.
Oberdl ist das Zeichen = zwischen unitaren Abbildungen durch N zu
ersetzen. An Stelle der Bedingungen (34) treten danach solche von
der Form
u(G1i6 3 , * .
*i
6,) r ( z , ,
c, c, - c, c,
= ic,, 1.
Der Kommutator
+ - -.+ z f C f
+ +
G~~Z,..~.
PI v
70
26
H. Weyl,
c,ll
zv = h (6,z),
!4 v
welche eine von der Basis unabhangige Bedeutung hat, nenne ich die
K o m m u t a t o r f o r m . Wendet man (26) an fur eine gegen 00 konverA
B
gierende Zahl k = 7 = m und 1 -, 1 - an Stelle von A und B,
m
m
SO erhiilt man im Limes als den Kommutator irgend zweier Elemente der
Gruppe U(G,, G,, . . ., 6f)= U(G) und U ( Z ) :
U ( G )U ( z ) U - ~ ( G U
) - l ( z ) = e(h(6, T)). 1.
(35)
[Um der Leserlichkeit willen schreibe ich oft e(z) statt e i Z . 1 Dieselbe
Einsetzung in (27) mit nachfolgendem Grenzubergang liefert noch
U(G
matrix
j/ ,
die sich
i (P,QV
und
pupv
Qv
- QVPv)= 1, i (PC1
PVP'U
= 0,
- QvP,) = 0 fur p
QpQv-
v,
Q y Q / l = 0 fur alle p, v.
Die
U(G) = e(G,P, f G2P2f
* * *
-k GfPf)
(36)
7 1
2i
und
(37)
9 7. E r s a t z d e r k a n o n i s c h e n V a r i a b l e n d u r c h d i e Gruppe.
Das E l e k t r o n . Unsere Xntwicklungen sind bis zu dem Punkte gediehen, wo die Verbindung mit der Quantenmechanik in die Augen
springt. Liegt ein mechanisches System von f Freiheitsgraden vor, so
geniigen ja die H e r m i t eschen Matrizen, welche die kanonischen Variablen
repr8sentieren, gerade den Relationen @6), bis auf den Faktor h/2 z, von
dem noch die Rede s e h wird und den wir einstweilen in die Mafieinheiten
hineinstecken. Nehmen wir die Zahl der Freiheitsgrade f zunachst = 1
und bezeichnen in der iiblichen Weise die kanonischen Variablen mit p , y,
&re reprasentierenden Formen mit PI Q, so sagt die Relation
i ( P Q - QP) = 1
(38)
aus, dafi die beiden durch die Matrizen i PI i Q gekennzeichneten infinitesimalen Drehungen des StrahlenkBrpers vertauschbar sind. Die durch
sie erzeugte A be lsche Drehungsgruppe besteht aus den Drehungen
U ( 6 , z) = e(P6
+ Qz)
(39)
(6,z reelle Parameter, die sich bei Zusammensetzung additiv verhalten).
Die reelle GrBlle im Gruppengebiet, deren Romponenten (G, z) der
Gleichung (19) oder
(6,z) = (-- 6 , - z)
(40)
+=
F = Jfe(Pci
+ ~ z ) t ( z)dadt.
~ ,
(41)
-03
72
28
H. Weyl,
f ( P , 4) = [ S e e d
+ gz)t(a, z ) d ~ d ; z .
(42)
-00
Die obertragung auf f Freiheitsgrade ljegt auf der Hand. Insbesondere sahen wir, wie a u s d e r F o r d e r u n g d e r I r r e d u z i b i l i t i l t
im F a l l e d e r k o n t i n u i e r l i c h e n G r u p p e n d i e c h a r a k t e r i s t i s c h e
k a n o n i s c h e Paarung e n t s p r i n g t . Fur endliche Gruppen freilich
existiert nicht ein so einheitliches Schema. Das ist im Einklang mit den
physikalischen Tatsachen. Denn aus den Entwicklungen von P. J o r d a n **
ging bereits hervor, daB beim magnetischen Elektron G , so gut wie dZ als
**
a,
13
29
(43)
wird. Die mit A vertauschbaren Matrizen U unserer Gruppe haben notwendig die Gestalt
1 : :,il;
U also = A .
Es gibt Gruppenelemente, deren Matrix B nicht mit A
vertauschbar ist. Wir wissen, dai3 in der Gleichung
AB =EBA
74
30
H. Weyl,
hat. Die Zahlen b, b sind vom absoluten Betrag 1. Wir wghlen ein
bestimmtes solcbes B, das gem59 Ba = 1 geeicht sei: bb = 1. A d e r dem kann man b zu 1 machen, indem man das bisherige normale Koordinatensystem el, e, durch e,, be, ersetzt; (43) wird dadurch nicht angegriffen :
(45)
Jede Matrix U unserer Gruppe, welche mit A vertauschbar ist, ist N 1
oder %A. Wenn sie nicht mit A vertauschbar ist, hat sie die Form (44),
und demnach ist ihre Zusammensetziing U B mit dem durch (45)gegebenen
bestimmten B eine Diagonalmatrix. Als solche ist sie mit A vertauschbar,
also ~1 oder N A . Das Kesultat ist, da13 jedes U z einer von den
vier Matrizen 1, A , B, A B ist. Es l i e g t i n d e r T a t d i e V i e r e r g r u p p e v o r u n d d i e D a r s t e l l u n g 23 derselben.
S 8. O b e r g a n g z u S c h r s d i n g e r s W e l l e n t h e o r i e . In iihnlicher
Weise, wie soeben der Fall 9% = 2 behandelt wurde, wollen wir jetzt
zeigen, dn9 die z w e i p a r a m e t r i g e n k o n t i n u i e r l i c h e n G r u p p e n nur
eine r irreduziblen Darstellung in unserem Sinne (auOer der identischen)
fahig sind. Wir erhalten jene Gruppen durch Grenzubergang aus den
z w e i b a s i g e n en dlichen. Die irreduzible Abelsche Drehungsgruppe
rnit der Basis A, B babe die Dimensionszahl w. I n der Kommutatorgleichung
AB =E B A
(46)
ist E eine 92-te Einbeitswurzel. Diese Gleichung gilt es jetzt naher zu
untersuchea. Die Kommutatorzahl E sei eine p r i m i t i v e vz-te Einheitsnwrzel, d. h. ~m sei die niederste Potenz, welche = 1 ist; rn ist Teiler
von 12. Die Drehungen A , B sind von einer in 12 aufgehenden Ordnung:
An
1, B N 1, und die Matrizen kijnnen daher so geeicht werden, dafi
A = Bn = 1 ist. Durch geeignete Wahl des normalen Koordinatensystems sei b auf Hauptachsen gebracht; die Glieder in der Hauptdiagonale, Oi, sind lauter rz-te Einheitswurzeln. Die Gleichung (46)
liefert fur die Koeffizienten von A = IIatkII:
(47)
75
Quantenrnechanik und Qruppentheorie.
31
76
32
H. Weyl,
kijnnen wir dazu benutzen, um auch die zweite Teilmatrk in die Einheitsmatrix umzuwandeln; und so fort bis zur (m- 1)-ten. Die damit erzielte
Normalform wird nicht zerstijrt, wenn die Variablen jeder Klasse untereinander der gleichen unitireit Transformation unterliegen. Diese Transformation kann man schlieLllich, wie man aei8, noch so bestimmen, da5
die letzte Teilmatrix A(m) eine Diagonalmatrix wird. Nunmehr nehmen
wir eine Umnumerierung vor, indem wir zunachst aus jeder Klasse das
erste Glied auslesen, darauf aus jeder Klasse das zweite usf. Dann
zerfkillt A in d Teilmatrizen, die sich langs der Hauptdiagonale aneinanderreihen. Wegen der rorausgesetzten Irreduzibilittt ist nur eine davon
vorhanden: d = 1, n = m. Wir haben die Normalform (die nicht ausgefiillten Felder stehen leer ") :
II
A=
A:
Xi
= Xk-11
B:
Z;
&kxkX;k.
Bt:
Xi
= EktZk.
A':
Xi
= Xk-8,
(48)
Jetzt last sich in aller Strenge der Grenzubergang zu kontinuierlichen Gruppen vollziehen. Es sei (39) die kontinuierliche zweiparametrige irreduzible Abelsche Drehungsgruppe. Die Basis iP,i Q sei
nach (38) normiert. Wir identifizieren in unserer Betrachtiing A mit
dem infinitesimalen e (EP),B nit e (7 &), C; und 7 reelle infinitesimale
Konstanten. Es ist e (6P) = AS, e (z &) = Bt, wenn im Limes s t = G,
t = e (6 kz). e (z &)
t q = z wird. E fallt mit e (6 7) zusammen, ~ k ist
ist die Reprasentation der physikalischen Gr6Se eizq; diese ist also (bei
beliebigem reellen z) der Werte fihig e i z c k , wo 7G die ganzen Zahlen
durchlauft. Mit anderen Worten: die GroSe p ist der Werte kE f i i g ,
ihr W e r t b e r e i c h d a s z u sa m m e n h a n g en d e K o n t i n u u m d e r r e e l l e n
Z a h l e n v o n - cc b i s
00.
(Dabei ist 76 freilich mod.%, kE mod.%t
zu verstehen; aber %C; ist ein Multiplum von 2 x/ql folglich im Limes
unendlich groll.) Darum schreiben wir jetzt q an Stelle von k& unter
77
33
cl zugleich eine Variable verstehend, welche den Wertbereich der physikalischen Gr6Be 4 durchlauft, und
@ (4) an Stelle von sk 9 (q) ist
cine willkiirliche komplexwertige Funktion, welche der Normierungs-
fg.
gleichung
(49)
*nterworft?n ist. lhre Werte sind aufzufassen als die den verschiedenen
Werten von q entsprechenden Komponenten eines ,reinen Fdles" in demjenieen normalen Koordinatensystern, das aus den Eigenvektoren der
GroBe q besteht. - An Stelle der zweiten Gleichung (48) erhalten wir
& Limes
9' = @V2: @'(q) = e i " q . Q ( q ) :
(50)
das ist. die unitire Abbildung P2, welche die GroBe ei2q darstellt. Der
gleiche Grenzubergang an der ersten Gleichung liefert die unitare Abbildung
ef = 1c, uo: @'(q) = @ (a - O),
(51)
welche eiirP reprasentiert. Beide Abbildungen sind in der Tat mitar,
weil sie die Gleichung (49) invariant lassen; sie bilden, den verschiedenen
Werten von 6 bzw. z entsprechend, zwei einparametrige Abelsche Gruppen
linearer Funktionaltransfoi-mationen :
+ zg)
U,.
Q (4)+ Q'(4) = e - l / z ( b Z . e i z n @ (a - 4.
Geht man endlich auf die infinitesimalen Operationen zuriick
was
freilich im allgemeinen nicht zweckmSiDig ist -, so bekommt man &
Reprasentation von
-a
(52)
D am it s i n d w i r b e i d e r S c h rB d i ng ers ch en F a s s u n g an g el an g t .
Die Eigenfunktionen Qn (q) seiner Wellengleichung haben danach die Bedeutung, dall sie die unitare Transformation angeben, welche zwischen
den beiden Hauptachsensystemen der GrijDe q und der Energie l3 vermittelt. I m Hinblick auf den ersten Teil ergeben sich daraus die bekannten P a u l i schen Ansatze fur ihre W ahrscheinlichkeitsbedeutung.
Die Obertragung auf mehrere Freiheitsgrade ist muhelos durchfuhrbar. D i e K i n e m a t i k e i n e s S y s t e m s , d i e d u r c h e i n e k o n t i Xeitschrift flir Phymk. Bd. 46.
78
34
H. Weyl,
+-
+J;iqip+4
-m
'p
U, aber in
1Yi.p.
e-
9 (PId F .
-s
Es ist also
'p (PI7
,
(53)
III.T e i l . D a s d y n a m i s c h e P r o b l e m .
fj 9. D a s O e s e t z d e r z e i t l i c h e n V e r a n d e r u n g . D i e Z e i t g e s a m t h e i t. Die bisherigen Ansatze beanspruchen a.Ugemeine Geltung.
Nicht so gunstig steht es mit dem d y n a m i s c h e n P r o b l e m , das eng
mit der Frage nach der Rolle zusammenhangt, welche R a u m und Z e i t
in der Quantehphysik spielen. In der F e l d t h e o r i e werden Zustandsgr6Sen behandelt, die in Raum und Zeit ausgebreitet sind, die M e c h a n i k
im engeren Sinne hat es nur mit der Zeit als der einzigen unabhBngigen
Veranderlichen zu tun. Die unabhangigen Veranderlichen sind keine
* Nach einem wichtigen Sakt von P l a n c h e r e l (Rend. Circ. Mat. Palermo 30,
330, 1910) und Titchmarsh [Lond. Math. SOC.Proc. (2) 23, 279, 19241 hat
diese TransEormation fur alle absolut quadratisch integrierbaren Funktionen einen
klaren Sinn und erhalt (bis auf den Faktor 2 n) das Quadratintegral.
79
36
Hier ist iE die infinitesimale unitare Abbildung, welche mit der die
Energie reprasentierenden Hermiteschen Form E gekoppelt ist, k das
Wirkungsquantum. Die mit dem Vorriicken der Zeit um d t verbundene
d y ) - A(y) irgend einer E e r m i t e s c h e n Form A (F)ist,
Anderung A (y
wie man leicht ausrechnet,
2nidt
d A = -( E A - A R ) .
(W
h
dE ist = 0. Bringt man die H e r m i t e s c h e Form E der Energie auf
Hauptachsen :
Z(h) = E,x,X, +E,z,X,+
E,,XnZnl
50 bezeichnen die Nummern 1 bis n die mbglichen Quantenzusttinde, Ei
die zugehbrigen Energiestufen, und in den Gleichungen (54) separieren
sich die Variablen:
dxv
2aiEw
0 . -
3*
80
36
H.Weyl,
(56)
Die Hermitesche Form
A(.$)= X a , , z p ,
ist nach Ablauf der Zeit t ubergegangen in
81
Quantenmechanik und Gruppentheorie.
37
durch
f =fpl d f
gemaB unserer Festsetzung die beiden Ableitungen d- = fq
dP
&l
reprgsentiert durch
+-
F, = i I l e ( G P + z & ) . G ~ ( G , z ) d G d z ,
-OD
+-
Fq = i j I e ( G P + z Q ) . z $ ( G , z ) d G d t ,
- 0 2
+
&. e(GY f z Q ) - e ( 6 P + z Q ) . @ = - G. e(GP + z Q ) ,
+
also
Fq = i(PF-FFP).
-Fp = i(QF-FQ),
Das dynamische Gesetz (54)lg5t sich daher, wenn f ( p , q) die Energiefuriktion ist, nach (55) so fassen:
Daraus sieht man: wenn u und b zwei reelle Zahlen vom Produkt h / 2 1~
sbd, reprleentieren a P und 59 GraDen, welche k a n o n i s c h sind i n
82
38
H. Weyl,
+ zca + + wa.
v2
2xmc
sind sie dimensionslose Gr6Den und zugleich mit der von uns befiirworteten Normierung der kanonischen Koordinaten in Einldang. Es
handelt sich darum, die Abbildung oder Hermitesche Form zu konstruieren, welche dieser GroDe entspricht im Raume der Funktionen
$(s, y, 2). Als Musterbeispiel dime der eindimensionale Fall. Es id
1927, S. 163,
83
Quantenmechanik nnd Gruppentheorie.
cd
39
oo
1%[ e - a 2 b V 1
X
f a 2 e - ~ l r ~ d=
u Gat,$,
"
0
(68)
@(+4&)
=+yvj(2:-~1
G a ( 6 ) d d = J G a ( ~ - - B ~ ( B d(59)
&
- m
-m
Im Limes fur
cd
- G (G) =
1
XG
-r ( G )
40
H. Weyl,
+ i6 = 0
+-
W++'(d = @ * ( 4 + + J ' ( x - t ) @ ( w t ,
(61)
- 00
>
+*
D(+)uber die obere Halbebene, haben wir schlieDlich als die der GrtiSe
f1 +us zugehbrjge Hermitesche Form:
+ m
Wenn es sich um ein einzelnes Teilchen handelt und eine (in der Einheit
m ca gemessene) potentielle Energie V ( s ) da ist, besteht das Eigenwertproblem darin,
+ J J m- E )
+--I w)
~YWX
+ m
f m
DW
5m ) d
-m
s a
10
+'T
-m
= 1. Die
85
Quantenmechanik und Gruppentheorie.
ES ist
41
+ +
+ +
e-
1~
86
42
B. Weyl,
(62)
aaai
dw
&T+du"+--=cpd Wa
F = Y, . F(s, Z),
in welchem Y, eine nur von der Richtung u : v : w abhangige Kugel0
funktion n-ter Ordnung sein solt, so geniigt im oberen Halbraum X
der nur von s und 1 abhangige Faktor F der Gleichung
>
daF
= n(12 + 1)E;
und die Operation P bedeutet den ubergang von den Rsndwerten ihrer
normalen Ableitung zu ihren eigenen Randmerten. Vielleicht ist es
87
43
+ F-
+ up*,
S2
(W
- Indem
- r)2= S2 + G2 - 8 S G cos 6,
+1
-1
Wenn das Einkbrperproblem vorliegt, wird man, auf die Gefahr hin,
eine Serie falscher Eigenwerte einzuschmuggeln, P iterieren und dadurch
zu einer reinen Differentialgleichung kommen. Fur das nichtrelativistiscbe
Wasserstoffatom sind die Eigenfunktionen cpp,(u,w, w), die durch die
Fouriersche Transformation aus den Schrbdingerschen Eigenfunktiohen
(x,y, z), den Laguerreschen Polynomen, hervorgehen, in
meiner Dissertation angegeben *. Sie kbnnen auch sehr schlin direkt auf
dem hier skizzierten Wege gewonnen werden. Im Mehrkorperproblem
versagt die Iterationsmethode.
Coulombsche Krrtfte zwischen m e h r e r e n beweglichen
Teilchen. Dem reziproken Abstand l/r19 zweier Teilchen 1 und 2
entspricht im Gebiet der Impulsfmktionen sp (u~,
w,, w,; aUp
, 'u2, us),wie
man auf die gleiche Weise erkennt, die Abbildung
+,,
+m
* Math.
88
44
H. Weyl,
Die Beneichnung soll naturlich nicht ausschlieden, da4 y auch von den
Iqpulsen der ubrigen Teilchen abhangt, diese werden aber von der Transformation nicht mit betroffen.
Ha t h e m a t ischer Anhang.
B e w e i s d e s S a t z e s von d e r . H a u p t a c h s e n t r a n s f o r m a t i o n
e i n e r u n i t a r e n Abbildung. 1st die unitire Abbildung A = 11 aik 11
gegeben, so bestimmen wir einen Vektor F' # 0, der durch A in ein
Multiplum von sich selber ubergeht:
n
(63)
Wahlen wir
...
...
0
0
O
893
. . .0
a93
a39
a2n
a3n
............
a
. . . . . . . . . . . .
0
an2
an3
ann
Durch Induktion in bezug auf die Dimensionszahl 12 ist damit der Beweis
vollendet.
Lie& die unitire Abbildung A in der Normalform vor, init den
Termen at in der Hauptdiagonale, so geniigen der Gleichung (63) offenbar
alle und nur diejenigen Vektoren, welche sich a m Grundvektoren ec zusammensetzen, fiir die at z e ist. Daraus geht bervor, da8 die vers c h i e d e n e n Eigenwerte a', a", ... mit ihrer Vielfachheit und die zu-
89
45
in
8 1 die Redc
war,
11
(ai - ak)bik = 0.
(64)
90
H. Weyl, Qnantenmechanik und Gruppentheorie.
46
Den S a t z uber die lineare Transformation einer nichtausgearteten schiefsymmetrischen reellen Bilinearform
f
(65)
beweist man so. 'Man fasse das einzelne Zahlsystem (zl,zzl ..., q)als
einen Vektor auf und bezeichne (65) als das schiefe Produkt [ g g ] der
beiden Vektoren g und 9 = (yi). Man wghle einen Vektor el qk 0. Nach
Voraussetzung ist [el r] nicht identisch in r gleich 0; ich kann also einen
zweiten Vektor e, so finden, daD [ele,] == 1 ist. Die simultan zu erfiillenden Gleichungen
k, TI = 01 [e,rl = 0
haben mindestens f - 2 linear unabhhngige Lasungen e,, . ep Auch
zwischen ihnen und el, ea findet keine lineare Relation statt. Denn ist
..)
+6ae, +
$- * " +&re/ = O!
so folgt durch Bildung der beiden schiefen Produkte [el?] = lZl
e , ? ] = - ,&, dal3 = & = 0 wird. Man kann also el, e,, ..., ef als
Koordinatensystem, als Vektorenbasis verwenden. In den darauf bezuglichen Romponenten &, qi der beiden Vektoren g und 9 laute das schiefe
F = 61el
6ae3
Produkt
i,
R=
= [eiekl:
y,1 = 0,
y 1 2 = 1; y , 3 = 0,'",
yal = - 1, yza = 0; y23= 0, " ' 1
ylf
= 0,
y2p=
0.
Wegen der schiefen Symmdrie sind infolgedessen auch alle yil, yi2
mit i = 3, .. ., f gleich 0 ; und die Matrix der y i k zerfhllt in das zweireihige Quadrat
1I-Y
und eine (f
- 2) - dimensionale
schiefsym-
91
J. v. Neumann
in Berlin.
.1
ist in der neuen Quantentheorie von fundamentaler Bedeutuiig, sie ist e8,
die den ,,Koordinaten-Operator" R und den ,,Impuls-Operator'bP im wesentlichen definiertl). Mathematisch gesprochen, liegt darin die folgende Annahme: Seien P,Q zwei Hermitesche Punktioneloperatoren des Hilbertschen
Raumes, dam werden sie durch dis Vertauschungsrelation bis auf eine
Drehung des Hilbertschen Raumes, d. i. eine mitiire Transformation U,
eindeutig festgelegtg). Es lie& im Wesen der Sache, da13 noch der Zusatz
gemacht werden mull: vorausgesetzt, daJ3 P, Q ein irreduzibles System
bilden (vgl. weiter unten Anm. 8 ) ) . Wird nun, wie es sich durch die
Schrodingersche Bassung der Quantentheorie als besonders giinstig erwies,
der Hilbertsche Raum als Funlrtionenraum interpretiert - der Einfachheit
halber etwa als Raum aller komplexen Pnnktioneii f ( q ) (- co < q < m )
+m
-, so gibt
-m
92
57 1
Operator 2rr( 1 ist. Die beiden Seiten konnen also nur gleichgesetzt werden,
wenn ihre Definitionsbereiche (d. h. der der linken Seite) naher umschrieben
werden. Dieser Schwierigkeit kann man folgendermafien aus dem Wege gehen:
Durch formale Operatorenrechnung folgt
der Vertauschungsrelation
(P(x) analytisch, P'(z) seine Ableitung, vgl. Anm. I))
rni
--2 ; i P Q
--BQ
P ( P )e
= P(P+
,8 I),
Ini
-ao
Jb
Diese Gleichung ist von Weyl aufgestellt nnd als Ersatz der Vertanschungsrelation vorgeschlagen wordend). Ihr grol3er Vorzug besteht in folgendem:
Es ist unter Urntiinden moglich, mit Hilfe der Operatoren P,Q einparaB a6
2376
-aP
U ( a )= e h
--8Q
, V(@=
) e
metrige Scharen
sind, uncl dem Multiplikationsgesetz
U ( a )U ( P )= q a
+P ) ,
V ( a )V ( P )=V(a
+ B>
t-ap
U ( a )V ( P ) = e
J(P) U ( 4
2ni
u(4: f ( d - f f ( q t 4 ,
W j : f(a)+e
T P C l
f ( a )-
93
572
J. v. Neumanii.
unitlire, also beschrankte und uberall definierte Operatoren, ao daLl ihr Sinn
ein vollig klarer ist.
Es bliebe daher zu zeigen, da13 die einzigen irreduziblen Losungen6)
der Weylschen Gleichungen die Schrodingerachen (d. h. die am Anm. ")
sind. Beweiaansiitze hierfiir gab Stone (vgl. Anm. ") an, jedoch ist bisher
ein Beweis auf dieser Grundlage, wie mir Herr Stone freundlichst mitteilte,
nicht erbracht worden.
Im folgenden sol1 der genannte Eindeutigkeitssatz bewiesen werdeu.
Wir werden Rogar alle (auch die reduziblen) Losungen angeben konnen.
2. Sei 8 der Hilbertsche Raum (etwa durch alle Bolgen komplexer
m
Betrag"
If l
1xf( I
q ) " d q ) einfiihren';).
--m
..
n=1
N , N - k c a ?n=1lt=l
9t=l
94
Eindeutigkeit der Sohrodingersohen Operatoren.
573
Orth ogonalsystem :
( A ( a ) (a)f, g> = ( B (a)f, A (a)*g j =1 2 4
=
?&=l
Es
U ( a )U(B)= U ( a
+s>
2
V ( 4W)= V ( a B ) ,
~ ( 8=) e""P(B)
~ ( a )
~ ( 0 ) .
x (u,p ) = e-"
idup
~ ( aV )( P )= e'
V ( P )~ ( a )
+ +
fi(aJb)x(Y, d ) = e
f l ( a Y,B 9)
zusammenfassen. Infolgedessen ist S (0,O) die Einheit, rind daher X (- a,- p )
zu E ( a , 8) reziprok, also X (01, ,f?)*= X(- a, - /I).Es sollen nun Linearaggregate der X (a,,8) betrachtet werden, diese werden folgendermafien defkiert: Sei a (a,p ) eine iiber die game a, P-Ebene absolut integrierbare
Funktion, dann ist wegen der Schwarzschen Ungleichheit
I(&
NG B ) fl.19 I = I fl.1
(a,B> f, 9 ) I I
I
9I9
JS.(.,
B ) (8( a , B ) f, s) dadP
I I I.
A = S J a ( ~ , p ) X ( a , Bd ) w L
obwohl die Definition eigentlich
lautet. a ( a , 16)
0)
( A f ,9 ) = S , f a ( ' J
p ) ('
heiBe der Kern von A .
f , g) dadP
Vgl. aiich a. a. 0. A m . '), Math. Annalen 102 (1930), S. 94, Anm. as).
95
574
J. v. Neumann.
+i(KV-pU)
a ( a - u,
- w)
bzw. e
-$. i (K W-p
21)
a ( a - u,p -w).
a),
( A B f >9 ) = ( B f J
=
=
ss
(a,B )
(A
SSJS7) ( a , 8) +,
JJ
[JS e+i
A*g)
( a>
B ) f ,9 )
acyp-scv,
p- 6 4
S> f , g j d a : d B d y d d
(Y
a ( y --
01,s
9)
'
J , f P ( " , B ) a ( a , B ) ( S ( a , 8f ), 9 ) d ~ d B = O ,
wenn P(a, p) ein Linearaggregat von endlich vielen e i ( r a + ' P )ist, also fur
jedes trigonometrische Polynom mit einer Periode p > 0 in a, ,f?. Da der
zweite Yaktor absolut integrierbar ist, und der dritte beschrtinlrt, konnen
wir mit dern ersten ( P ( a ,p ) ) Grenziibergange ausfiihren, falls dieser dabei
gleichm5Big beschrllllrt bleibt. So konnen wir die Klasse der P( a , ,b) sukzessiv erweitern: 1. zu allen stetigen Funktionen mit einer Periode p > 0
in a,/?,2. zu allen beschriinkten stetigen Bunktionen, 3. zu allen beschriinkten
Funktionen der ersten Baireschen Klasse. Wenn also 9? ein beliebiges (endliches) Rechteck in der a , /I-Ebene ist, so konnen Wir P ( a , p ) in % gleich 1
und aul3erhalb = 0 setzen, as wird:
JJa(.,B) ( m a , P 1 > f , q ) d a ~ p = O
Dt
96
575
ss e*
- ss -9
i (a9 -B 5 )
=e
=e
=e
5%-
a E
-a a%- & pl
-)u'-*v'
=2ne
-*uz-*u'
71
ss e
ss e
e-iu2-&p
-*a*-+p
.e
-4 p+ + ( a tu-ip -i V ) -4
-+(E-&
711
dE dy
d ldr
(a+u-ib-(v))g-3(,~-~i(a+u-iP-io))B
d 5 dq
dy
JJ e-+"*-+y'dx
-*.1-&p
Hierin ist der erste Faktor konstant und der zweite der Kern von A ,
also gilt
-4t P -$ 3- A .
AS(u,v)A=2n;e
Wir betrachten nun die Losnngen der Gleichung A f = 2 m f , da A
linear-bemhrinkt ist, bilden eie eine abgeschlossene Linearmannigfaltiglreit
im Hilberbchen Raume, '$R. Jede von ihnen hat die Form Ag (mit
1
ist
(man setze in der obigen Gleichung u = w = 0). Die ZLI 5132 orthogonale
abgeschlossene Linewmannigfaltigkeit sei 8,die Elemente f von Zn sind
durch Orthogonalitlit zu allen Elementen von 92 gekennzeichnet, d. h, zu
allen A g . D.h.: immer ( f , A g )= 0 , oder: immer ( A f , g )= 0 , oder: Af=O.
Wenn f , g zu 9
' 2 gehoren, so kt
1
( X ( a , P ) f y S ( y , b ) g ) =m(s(@Yp)Af,S ( Y , W g )
1
4 n 2 ( A8 (- y,
L-
-6 ) x ( a ,B ) A f , 9
1 e+i(aa-SY)
=4aQ
(AS ( @
- Y,B - 8) A f , 9 )
97
576
J. v. Neumann.
Sei nun
. . ein normiertes Orthogonalsystem, welches in 91
vollsthdig ist (d. h. %Q aufgespannt, die Zahl seiner Elemente ist endlich
oder abzahlbar unendlich). Aus ('pm,p
' ,) = d,,, (d. i. 1 fur m = n, 0 sonst)
folgt
-& (a- UP- a (1-SP+& i (ad -P u)
( # ( a , PI F,,,, 8 ( Y , 6) 9,) = e
~ , nDie durch alle S ( a , ,6) p
' , ( n fest, a, ,6 variieren) aufgespannte abgesohlossene
Linearmannigfaltigkeit heil3e p, nach der obigen Formel sind fur m n
!$, '@fl zueinander orthogonal, Die
p2, . mogen eusammen die abgeschlossene Linearmannigfaltigkeit Q aufspannen, die zu Q komplementare
abgeschlossene Linearmannigfaltigkeit sei 5 .
Da, jedes 8 ( y , 6) die 8 ( a ,S ) v,, (bis auf Zahlenfaktoren) in ebensolche
transformiert, bildet es !&, auf ein TeiI von sich ab; da dasselbe fiir
S ( y , S)-'=
S (- y , - 6) gilt, ist ?j3, genau invariant. Also ist auch S
und S invariant. (5 umfaBt alle p,,,also alle qn, also YR, daher liegt 5
in (57. Somit gilt in Z stets A f = 0. Nun gelten alle msere uber A migestellten Betrachtungen schon in &, d e n die X (a,8) konnen als Operatoren in Z angesehen werden, da dieses ihnen gegeniiber invariant ist. Da
nun in Z A = 0 ist, kann nach unserem Beweise in 7X niemals f + 0 sein.
Also enthalt S nur die 0, G ist der Hilbertsche Raum, d. h.: Q,,'j&, . . .
spannen den vollen Hilbertschen Raum auf.
Der Hilbertsche Raum erscheint somit als in eine endliche oder ahzahIbar unendliche ZahI von (paarweise orthogonalen) Unterriiumen Ipl, !&,...
zerlegt; jeder derselben ist gegeniiber allen S (g, ,8) invariant, es geniigt
also das Verhalten der S ( a , /I)(d. i. der U ( a ) , V (/I))
in einem jeden derselben gesondert zu ermitteln, urn uber sie restlos informiert zu sein. (Im
Palle der Irreduzibilitat darf es natiirlich nur ein p,, geben, und dieses ist
dann der volle Hilbertsche Raum.) In
wissen wir nun uber die 8 (a, )
die f olgenden Tatsachen :
Wir nennen Pn 9,S(CX,/~)P),,fa,@. D a m gilt:
$2
..
und die Linearaggregate endlich vieler fm,@ (die beliebig wahlbar sind!)
liegen in 'g iiberall dicht.
Wenn wir nun zeigen konnen, daB irgend zwei solche p, in deren
jedem eine Schar von unjtaren Operatoren S ( a , 6 ) und Punkten fa,@ mit
den obigen Eigenschaften gegeben ist, isomorph sind, so sind wir am Ziele.
Isomorphie bedeutet : Existenz einer ein-eindeutigen, linearen und Iangentreuen Abbildung der beiden !'$ aufeinander, die die fa,@ und S(a, @)des
! in dieselben des anderen iiberfiihrt.
einen $
98
577
f ( q ) - e e " a @ f ( q(GI
)
stattp!), daher
Anm.6) angegeben: f ( q ) - f ( q + a ) ,
ist &'(a, p ) : f ( q ) - + e iB(g+$) f ( q
$q
+m
f(q>
-4q'-
&4' =
+ a ) , und,
( 6 )drj'.
y,(q) = n-'
-11-42
, und
= 2 TZ f
99
578
. ..)
Preiheits-
Auch hier ist der Eindeutigkeitsbeweis mit unseren Methoden aus 2. bis 3.
ss
'I)
(1926),
11.
S. 737-755.
(Eingegaagen am 31. 8. 1930.)
100
J U N E 1, 1932
PHYSICAL REVIEW
VOLUME 40
1
N classical statistical mechanics the relative probability for the range
$1 to Pl+d+l; $2 to P z S d P z ; . * . ; $n to $,,+d@, for the momenta and x1
to xI+dx1; x 2 to x 2 + d x z ; . . . ; x, to x , + d x , for the coordinates' is given for
statistical equilibrium by the Gibbs-Boltzmann formula
P(x1,.
= e-fltdxl.
. .dx,dpl. . .dpn(l)
(3)
In quantum theory there does not exist any similar simple expression for
the probability, because one cannot ask for the simultaneous probability for
the coordinates and momenta. Moreover, it is not possible to derive a simple
expression even for the relative probabilities of the coordinates alone-as is
given in classical theory by e-flV(? ' . ' =n). One sees this by considering t h a t this
expression would give at once the square of the wave function of the lowest
state [ +,,(xl . * . x,) when p = to is inserted and on the other hand we know
that it is not possible, in general, to derive a closed formula for the latter.
The thermodynamics of quantum mechanical systems is in principle,
however, given by a formula of Neumann,' who has shown that the mean
value of any physical quantity is, (apart from a normalizing constant depending only on temperature), the sum of the diagonal elements of the matrix
l2
Qe-flH
(4)
101
750
E. WIGNER
I t does not seem to be easy to make explicit calculations with the form
(4)of the mean value. One may resort therefore to the following method.
If a wave function +(xl . . . x,) is given one may build the following
expression2
P(x1,
. , x n ; PI,
=
(')hr,
* *
Jm
Pn)
--
* *
.Jdyl.
* '
dyn+(xl+
y l *
. . xn
yn>*
+(xl - y l . . . x, - yn)e2i(~,~i+...+~nY,)/h
(5)
and call i t the probability-function of the simultaneous values of X I . . . xn
for the coordinates and $1 . . p , for the momenta. In (5), as throughout
this paper, h is the Planck constant divided by 2s and the integration with
respect to the y has to be carried out from --OO to 43. Expression ( 5 ) is
real, but not everywhere positive. I t has the property, that it gives, when
integrated with respect to the p, the correct probabilities I$(xl . . x,)
for the different values of the coordinates and also it gives, when integrated
with respect to the x , the correct quantum mechanical probabilities
Is
Jw. . . J $(xl
. . . ~,),-;(~~~:~+...+~n+n)/hdx~ . . . dx,
--m
l2
for the momenta pl, . . , p,. T h e first fact follows simply from the theorem
about the Fourier integral and one gets the second by introducing x k + Y k
=Uk; Xk-yk=vk
into (5).
Hence it follows, furthermore, that one may get the correct expectation
values of any function of the coordinates or the momenta for the state $ by
the normal probability calculation with ( 5 ) . As expectation values are additive this even holds for a sum of a function of the coordinates and a function
of the momenta as, e.g., the energy H. I n formulas, i t is
--m
--m
(6)
This expression was found by L. Szilard and the present author some years ago for another
purpose.
102
751
QUANTUM CORRECTION
(7)
ap
- = -
at
. , x,,;$1,
p k ap
c--+c
n
k-1
mk a x k
, $,) is given by
axl+. . * + X n v
axlXl.
-.
( h / 2i) A
* .+Xn--l
xl! . . . A,!
aXnXn
a A I+
aplkI.
* . .+A,p
. . ap,~n (8)
where the last summation has to be extended over all positive integer values
of X I , . . . , A, for which the sum Xl+Xz+ . . * +X, is odd. In fact we get for
aP/at by ( 5 ) and (7)
(9)
+ hi
-"v(~l
yl,
- V(x1- yl, . . *
* 1
xn
xn
+ yn)
- yn)]+(x1+
.,
* * 1
xn
Yn)*+(X1-
yl,
* *
'xn
- yn)
103
752
E. WIGNER
(10)
which is identical with (8) if one replaces now the differentiations with respect
to y k by differentiations with respect to x k . Of course, ( 8 ) is legitimate only if
it is possible to develop the potential energy V in a Taylor series,
Eq. (8) shows the close analogy between the probability function of the
classical mechanics and our P : indeed the equation of continuity
-a p- -at
ap
av aP
c---+
c--P k
m k
a x k
k a x k
a p k
differs from (8) only in terms of a t least the second power of h and a t least the
third derivative of V . Expression (8) is even identical with the classical when
V has no third and higher derivatives as, e.g., in a system of oscillators.
There is an alternative form for aP/at, which however will not be used
later on. I t is
J W
+ j,,
(11)
* *
, P,
+ j,)J(x.,,
* * *
, x,;
--m
'
'
'
xn; j ~ ,'
' 1
jn)
- T/' (
,, . . . , xn - y,) ] e-
( 2 i I h ) (Y I i
,+. .S + Y j~ d
(114
28
104
753
QUANTUM CORRECTION
*
. . d y n [ ~ 2 + ( x l + ~ ~ , * , ~ n + y n )+(x1-~1,
..
axk2
*
xn-yn)
(12)
which is just what we need for the left side of (11). By integrating the second
term on the right side of (11)
- V(x1 - 81 . . . xn - zn) 3 e - 2 i ( z
ir+.
,+gn
id / h
J.
(i/h)
* . J d y l * * . dyn+(xl+ y l . . xn
yn)*+(xl- y l . * . xn - yn>
e2i(piyl+...+~n~n)h.
[v(xl y1 . . . xn yn) - ~ ( 3 ~ 1yl * * . xn - yn)] (12a)
and this gives the second part of the left side of (11).
3
So far we have defined only a probability function for pure states, which
gives us the correct expectation values for quantities f(P1 * . * Pn)+
g(x1 . .
If, however, we have a mixtureI4e.g., the pure states $1, $z,
$3, *
with the respective probabilities w1, w2, wS, . . * (with wl+wz+wa+
* . . = 1) the normal probability calculation suggests a probability function
(13)
P(x1, * * 9 xn, 91, * . * , Pn) = C wx Px(x~,
* * - 7 xn, * *
Pn)
A
where PAis the probability function for $A. This probability function gives
obviously the correct expectation values for all quantities, for which (5) gives
correct expectation values and therefore will be adopted.
For a system in statistical equilibrium at the temperature T = l/kP the
relative probability of a stationary state +A is e-oEA where E Ais the energy of
$x. Therefore the probability function is a part from a constant
Cf. e. g., R. Courant und D. Hilbert, Methoden der mathematischen Physik I. Berlin
1924. p. 62, Eq. (29).
J. v. Neumann, Gott Nachr. 245, 1927. L. Landau, Zeits. f. Physik 45,430 (1927).
754
E. WIGNER
e-BEh#(xl
Now
C fiA
* 1* *
- yl . . . xn - yn)e2'(pi"~+...+P"~")/h,
~ , ) * j( E A ) fix
(211 *
(14)
. Vn)
P(x1 *
* *
x n ; PI
J:. .
dy,
' '
Pn)
. . . ~ y , e ~ [ ( z i + U ~ ) P i + . ~ ~ + ( ~ n + ~ n ) P n l ~ ~.,.=,
[ e -+un;z,-ul...zn-vn
~~~~~+~~
-m
,e - i [ ( z , - U , ) P , + .
. . + ( ~ n - ~ n ) ~ n l/ h . (15)
. . x,+y,; x l - y ~ .
*
which is equal to
(16)
where
c =
2 -+ V(x1,
Pk2
k=l
'
, x,).
(17)
2mk
(18)
cp(
106
755
QUANTUM CORRECTION
+ yl, . .
xn
+ yn)'6(xl+
yl, x1
- yl)
&(xn
+ yn, xn -
yn).
As 6 (also 6',
(1/2")
C ( - ~ ) r e ( x l , . , x,,)r/r! = e-b6/2*
(19)
(19)
(20)
By replacing all operators by symbolic integral-kernels one gets for the
x l + y l , . . . , x,+y,; xl-yy,, . . . , x n - y n element of the operator (20)
h2
- -6(-
2y1)
2mk
. . . 6',(-
2Yk)
'
. . 6 ( - 2y,)
(XI
- y1,.
* *
, xn -
y,)r-p.
Now
1
2"
mk
dWk
2mk
dWk2
g-8f(Sl....
e(x1,
,lk-Wk/2,..*
,zn)
- e-bt(Zk,."
* *
,Zk+Wk/2.".
. , Xk
vZ?t)
+Wk,
. . , x,)
*
107
756
E. WIGNER
In principle it is possible to calculate in the same way the terms involving the
higher powers of the second part of H also, the summation over r and the
quantities corresponding to our p can always be performed in a very similar
way. In practice, however, the computation becomes too laborious. Still i t is
clear, t h a t if we develop our probability function for thermal equilibrium in
a power series of h
= e-pr
+ h/i + h2jz + . .
(22)
(we can omit the factor 1 / 2 " before e-oc, as we are dealing with relative probabilities anyway) all terms will be quite definite functions of the ;b, V and t h e
different partial derivatives of t h e latter. Furthermore it is easy t o see, t h a t
f k will not involve higher derivatives of V than the k-th nor higher powers of p
than the k-th. These facts enable us to calculate t h e higher terms of (22) in a
somewhat simpler way, than the direct expansion of (18) would be.
T h e state ( 2 2 ) is certainly stationary, so t h a t i t would give identically
aP/at=O when inserted into (8). B y equating the coefficients of the different
powers of h in aP/at to zero one gets the following equations :
*
(23, 0)
(23,1)
av a j 2 x---h2 a%-@
c - _ _ _+ c--h2
- z-.-----0
8
P k
d3V
af2
m k dXk
k f i
dxk
a p k
a3v
a3e-Pt
aXk2dXi
apk'api
aXk3
24
apk3
(23,
( 2 3 , 2)
2)
av
z---+C--=O.
k
p k
dF
mk
3%
aF
axk a p k
, x m ; P I , . . ' , pn)
= F(
c
k
Pk2
I
_
2mk
108
(25)
as a solution of (23, 2) and it is also clear, that this is the solution we need.
The first two terms of fz we have already directly computed ( Z l ) , the third
arises from terms with the second power of the second part of i?. Similarly
f4 is for one degree of freedom ( n = 1)
64m2P-2 eojd = H 4 ( q )[p2V2/72- pV//120]
. . , x,; P I , . . . , pn)
e-af(l
+ k2gz + k4g4 + . . . )
(27)
are rational expressions in the derivatives of V only (do not contain V itself)
and all terms of g k contain k differentiations and as functions of the p are
polynomials of not higher than the k-th degree. The first term in (27) with
the zero power of h is the only one, which occurs in classical theory. There is
no term with the first power, so that if one can develop a property in a power
series with respect to h, the deviation from the classical theory goes a t least
with the second power of k in thermal equilibrium. One familiar example for
this is the inner energy of the oscillator, where the term with the first power of
h vanishes just in consequence of the zero point energy. The second term
can be interpreted as meaning that a quick variation of the probability function with the coordinates is unlikely, as i t would mean a quick variation, a
short wave-length, in the wave functions. This however would have the consequence of a high kinetic energy. T h e quantum mechanical probability is
therefore something like the integral of the classical expression e-ot over a
finite range of coordinates of the magnitude -h/p where p is the mean momentum ~ ( k T r n ) The
~ / ~correction
.
terms of (27) have, among other effects,
109
758
E. WIGNER
..
Sdpl
..
'
dp,P(zl
. . . x,; p 1 . . . p,,)
(28)
(29)
I-
'
24
(30)
J
This formula also is correct only within the second power of h ; in order to
derive i t one has to perform again some partial integrations with respect to
the x. Eqs. (28), (29), (30) have a strict quantum mechanical meaning and
i t should be possible to derive them also from (4).One sees that the kinetic
energy is in all cases larger than the classical expression +nkT.
5
110
Q U A N T U M CORRECTION
759
ticles, as, e.g., a gas. There is reason to believe however, t h a t because of the
large radii of the atoms this is not true and the corrections due to Fermi and
Bose statistics may be neglected for moderately low temperatures.
T h e second virial coefficient was first calculated i n quantum mechanics by
F. London on the basis of his theory of inneratomic force^.^ He also pointed
out that quantum effects should be taken into account at lower temperatures. Slater and Kirkwooda gave a more exact expression for the inneratomic
potential of He and Kirkwood and Keyes' calculated on this basis the classical part of the second virial coefficient of He. H. Margenaus and Kirkwoodg
performed the calculations for the quantum-correction. T h e present author
also tried to calculate it by the method just outlined. H e got results, which
differ from those of Margenau and Kirkwood in some cases by more than
100 percent.1 I t does not seem however to be easy to compare these results
with experiment, as the classical part of the second virial coefficient is a t low
temperatures so sensitive to small variations of the parameters occurring in
the expression of the interatomic potential, t h a t i t changes by more than
20 percent if the parameter in the exponential (2.43) is changed by 4 percent
and i t does not seem to be possible to determine the latter within this accuracy.
111
October 1946
Physica XII, no 7
ON THE PRINCIPLES
O F ELEMENTARY QUANTUM MECHANICS
by H. J. GROENEWOLD
Natuurkundig Laboratorium der Rijks-Universiteit te Groningen
Summary
Our problems are about
a t h e correspondence a +--+ a between physical quantities a and quant u m operators a (quantization) a n d
p t h e possibility of understanding t h e statistical character of q u a n t u m
mechanics b y averaging over uniquely determined processes as in classical
statistical mechanics (interpretation).
a and p are closely connected. Their meaning depends on the notioneof
observability.
We have tried t o p u t these problems in a form which is fit fcr discussion. We could not bring them t o a n issue. (We are inclined t o restrict
the meaning of CL t o t h e trivial correspondence a + a (for lim fi -+0) a n h
to deny t h e possibility suggested in p).
Meanwhile special attention has been paid t o t h e measuring process
(coupling, entanglement; ignoration, infringement; selection, measurement).
For t h e sake of simplicity the discussion has been confined t o element a r y non-relativistic quantum mechanics of scalar (spinless) systems with
one linear degree of freedom without exchange. Exact mathematical
rigour has not been aimed a t .
Physica XTI
26 *
112
406
H.J . GROENEWOLD
(1.01)
When the same system is treated in elementary quantum mechanics, the (real) quantities a are replaced by (H e r m i t i a n)
operators a, which now represent the observables. I n the equations
of motion the P o i s s o n brackets (1.01) are replaced b y the operator brackets
[a,b] =-(ab-ba)
F,
h
(h=--, h P l a n c k s constantof action). (1.02)
2x
113
ON T H E PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
407
= 8,VP
the completeness b y
2 cp,cpf
= 1.
(1.03)
'pv.
The
(1..04)
(1.05)
(1.06)
(Because the right hand member is invariant under unitary transformations of the ?
,, this definition is independent of the special
choice of the complete orthonormal system of p,). This gives
Wkvpa) = 9: a 'pv.
(1.04) and (1.05) can be written
Trk, = a,,
C k,, = 1
P
(1.10)
(1.11)
(1.12)
114
4C8
H. J . GROENEWOLD
Tr(k,ukv#p*)
= 6p#,6,*,
C k,Tr(k,,a)
= a (for every
(1.13)
a).
(1.14)
(1.13) and (1.14) show t h a t every operator a (with adjoint at) can
be expanded into
a = C xvpkPuwith a,,,= Tr(k,,a).
(1.15)
P.
CI Tr(k,b) Tr(k,,a)
Further
Tr(ab) = Tr(ba).
(1.17)
When a is a H e r m i t i a n operator
a+ = a, ,:a = a,,,
(1.18)
(the asterik denotes the complex conjugate), the system of eigenfunctions 9, with eigenvalues up
acp,
= a,cp,
(1.19)
X k p = 1.
(1.21)
(In some cases the sum diverges and the right member actually
should symbollically be written as a &function). The statistical
operator cf the mixture is (in the same way as it would be done for
115
ON T H E PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
409
k = Z: k,kp,
(1.22)
T r k = 1.
(1.23)
(we will always write 1 for the right member, though in some cases
it actually should be written as a %function). For brevity we often
speak of the state (or mixture) k.
An arbitrary non-negative definite normalized H e r m i t i a n operator k (Trk = 1) has non-negative eigenvalues k,, for which Z k,= 1
CL
and corresponding eigenstates with projection operators k,. Therefore k can according to (1.20) be expanded in the form (1.22) and
represents a mixture of its (orthogonal) eigenstates with weights
given by the eigenvalues.
The statistical operator k,, of a pure state is from the nature of
the case idempotent (k2 = k,,). If on the other hand an idempotent
Y
normalized H e r m i t 1 a n operator k is expanded with respect to
its eigenstates k,, with eigenvalues .k,, we get
k2 = k, k t
= k,;
Trk
1, C k , = 1 ,
P
(1.24)
k = k,.
(1.25)
This gives
k - k2 = X k, (k, - k:)
+ Q C K,k,(k, - k,)2.
(1.27)
1,s
116
410
H.J. GROENEWOLD
117
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS 41 1
118
412
H . J . GROENEWOLD
p and q are the generating elements of the commutative ring of i-,!assical quantities n(q,p), p and q the generating elements of the
non-commutative ring of quantum operators a. The non-commutability (1.30) of p and q entails t h a t the quantities a($,q) cannot
unambiguously be replaced by a(p,q). The ambiguity is of the order
of fi. The classical quantities a(p,q) can be regarded as approsimations to the quantum operators a for 2im F, -+ 0. The former can
serve as guides to get on the track of the latter. Problem u1 asks for
a rule of correspondence a@,q) -+ a, by which the quantum operators
a can be uniquely determined from the classical quantities a(p,q).
In practical problems no fundamental difficulties seem to occur
119
ON T H E PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS 4 13
/(a),
if a t+ a, then /(a)t-+
I1
a and b t+ b, then a
b t+ a + b.
/(a) is defined as the operator, which has the same eigenstates as a
with eigenvalues /(a,), where ap are those of a. Then I seems to be
obvious. The observable /(a)can be measured simultaneously with
if u
t+
120
414
H . J. GROENEWOLD
E x ( a ) = 2 Tr(kk,,)a, = Tr(ka)
(1.31)
Tr(k(a
have that
(1.33)
(1.34)
Because this has to hold for all states k, a and b have to satisfy rule
11. When I1 is given up for certain pairs a,b, the additivity of the
expectatiori values of these pairs has also t o be given up.
I n 4.01 it will be shown that, if I and I1 shall be generally valid,
the symbols a have to be isomorphic with the operators a. But then
there is no reason to introduce the former, their task (if any) can
be left t o the latter. Accordingly for the sake of brevity we shall
henceforth speak of the (quantum) observable a.
When on the other hand, the symbols a are intended as real commuting quantities, the general validity of I and I1 cannot be maintained. As long as the symbols a are not further defined, problem
c ( ~
comes to searching for a one-to-one correspondence a +-+a
between the commutative ring of real symbols a and the non-commutative ring of H e r m i t i a n operators a. There may be no, one or
more solutions. After the pleas for I and for 11, one might be inclined to maintain 1 and to restrict 11. In 1.13 we meet with a particular case (problem as! for which I1 has to be maintained and therefore I has to be restricted. Because we are further exclusively interested in problem as, we will not examine the possibility of solutions for which II is restricted.
121
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
415
Trk = f d6 A(().
(1.35)
+ k, ++ k,(E)+k,(E). (1.36)
Tr(k1kz) = J d E
h(E)b(E).
(1.37)
122
41 6
H. J . GROENEWOLD
1t-+
1.
111
t+
h
O N THE P R I N C I P L E S OF E L E M E N T A R Y QUANTUM MEC.HANICS 4 17
k(s)
t-+
(1.41)
if
dl H(dt)
at
Because
d
- Tr(ka) = Tr - [ H , k ] a
at
- ( ( H ( S ) k, ( k ) ) ) .
+ k at
(1.43)
at
(1.44)
and correspondingly
(1.45)
the dynamical time dependence can be shifted from the wave functions rp and the statistical operators k (S c h r o d i n g e r representation) and the distribution functions A ( ( ) to the operators a
(H e i s e n b e r g representation) and the superquantities a ( c ) .
Instead of (1.40), (1.43) we then get
(1.46)
(1.47)
27
124
~
~~
418
~~
H. J. GROENEWOLD
The equations of motion for the hidden parameters may be of a different form. When all parameters (inclusive the hidden ones) are
continuous, their equations of motion have to satisfy t h e condition
that when inserted in
(1.49)
(where the last term stands symbolically for a sum over all separate
parameters 5), they must give (1.47).
We may summarize that, in order to give a statistical description
of the 1st kind, one would have to determine (only formally for
type St, also experimentally for type S:) the parameters 6 (inclusive
the hidden ones) and the density function, the (one-to-one or one-tomany) correspondence a -++ a(5) (problem as) and the equations
of motion for the hidden parameters (if there are any such), all
with regard to the imposed conditions.
1.15 Correspondence a +-+ a&). Because a non-H e r m i t i a n
operator a (with adjoint at) can be written as a complex linear combination of H e r m i t i a n operators
a = 4 (a
1
+ a+)+ (ia - iat),
2i
the generalization of the correspondence a t-f a ( [ ) to non-H e rm i t i a n operators is uniquely determined. Now take the non-H e rm i t i a n transition operators k,, which according to (1.13), (1.14)
form a complete orthonormal system in the ring of operators a.
For the corresponding functions KP,,(t)
we get corresponding to
( l . l l ) , (1.12); (1.13),(1.14) and (1.15) (andusingII1, W a n d (1.03))
the relations
(1 .SO)
J-E JZ&)
= $w,
xP A&)
(1.51)
= 1;
J (it kCU(6)h P V ( t )=
appn
44 ,
(1.52)
(1.53)
(6([
4 5 ) = c a v p h,(E)
PJJ
with a,,,=j 4 q ( 5 ) 4 5 ) .
(1.54)
I25
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
419
(the avp are the same as in (1.15)). These relations show, that the
functions a(E,) can be regarded as elements of a (generalized)
H i 1b e r t space, in which the k,,,(E) form a complete orthonormal
system ;(1.52) expresses the orthonormality, (1.53)the completeness.
We now show that the correspondence a c+ a(<) has to be a
one-to-one correspondence. Suppose for a moment there are operators kpvto which there correspond more than one functions k&),
which we distinguish by an index p, k,, t+ kp,;,,(t).Then the expression
cJ
kp;
p,,
evaluated with (1.52) gives k,,,; ,,,,(?), evaluated with (1.53) it gives
k,,; p ( < ) . Therefore hpv;,,,,(t)and kpv;
have to be identical. To
each operator a and only to this one there has to correspond one and
only one superquantity a ( [ ) .As a consequence the superquantities
a(c) must depend on the same number of parameters (at least if
they are not too bizarre) as the operators a, i.e. on twice as many as
the wave functions cp.
Thus to each (normalizable)real function a(E) and only to this one
there corresponds one and only one H e r m i t i a n operator a, which
represents an observable quantity (with respect to OJ. In other
words every real function a(<) is a suparquantity. Bec6use this also
holds for the (real and imaginary parts of the) parameters E themselves, none of them can be hidden in the sense defined above. (An
observable quantity may occasionally be inobservable in a measuring device adepted to an incommensurable quantity; in this sense a
parameter may occasionally be hidden). In particular all parameters
must obey (1.48).
Comparing (1.15) and ( 1.54) we see that the correspondence
a +-+ a(E)can be expressed by
,,(El)
with
a ( t ) = Tr(m(t)a),a =/dSm(E)n(F),
4 5 )=
kp,,k;JF);
m+(t)= m(5).
(1.55)
(1.56)
PI,
Tr(m(t)m(E))= Yt - E),
J d E Tr(m(4)a) Tr(m(6)b) = Tr(ab) (for every a and b)
(1.59)
(1.60)
H. J. GROENEWOLD
(1.60) is equivalent to
k;, = ukLvu?,
U U ~=
1 ; k;, = utkLvu
(1.61)
The
127
~
421
by
v(E ; q) = c K J E ) K?(q) ;
45;
q) = v*(E ; q)
(1.62)
(1.63)
cL,v
satisfies
jd5
~ ( 5q); = Jdq
v(E; q) = 1 ;
(1.64)
/ d ? v ( S ; q ) ~ ( E ;q)=S(E-E),
/ d E v ( S ; q ) v ( S ; q ) = S ( q - q ) . (1.65)
The rings of a(!$)and of a(?) are not necessarily isomorphous.
When they are, we must have
/ldqdqv(E; q>.(E;
v(E ; q) v(E;q) = 4 4 ; q)
a(? - 3)
(1.67)
5).
(1.68)
and similarly
vK; i)v(E; q) = v(; q) S(S-
( E ) - q)
v(< ; q) = W - 5 (d)
v(k
q) = h
(1.69)
(1.70)
where q(5) and E(3) are single valued functions. Because (1.69) and
(1.70) have to be identical, q(E) and <(q) have to be inverse t o each
other with unit functional determinant
(1.7 1)
(it should be remembered that we symbolically write 6 or q for what
might be a whole set of parameters or q). With (1.69), (1.70) we
get for (1.62)
(1.72)
n(4) = a(q(E)); a(q) = a(q(5)).
This shows that the transformation between two isomorphous representations ~ ( 5 )and a(q) can be regarded as merely a transformation of the parameters. It further follows that, if the dynamical
conditions for (1.49) are fulfilled by one of these representations,
they are also fulfilled b y the other one. Therefore isomorphous representations can be regarded as equivalent.
128
422
H. J. GROENEWOLD
When the solution v(E;?) of (1.64), (1.65) is not of the form (1.69),
(1.70), the representations a([) and a(?) are non-isomorphous.
1.17 Parameters. In 4.03 we derive a correspondence, satisfying
I11 and IV, in which the two independent parameters (denoted by
# and q), which run continuously between - 00 and 00, correspond to the operators p and q.This choice of parameters might seem
the most satisfactory one, as it is adapted t o the fundamental part
played by the momentum and the coordinate. (By the way, because
momentum and coordinate cannot simultaneously be measured, p
may be regarded as occasionally hidden in a coordinate measurement, q similarly in a momentum measurement -or in a somewhat
different conception p may be regarded as occasionally hidden in
q-representation, q in #-representation; both p and q may be regarded as occasionally partially hidden in other measurements or representations).
In 1.16 we have seen that for each choice of a complete orthonorma1 system of k J p , q ) , satisfying (l.SO),(1.51); (1.52), (1.53), there
must for every other representation with parameters 5 be a similar
system of k&) with the same set of indices p,v. That makes us expect that when E stands for a set of not too bizarre continuous parameters, the latter can like # and q be represented by two independent
real parameters 7 and s. We do not examine the validity of this expectation (which would be very difficult).
1.18 Bracket expressions. When these parameters Y and s are also
independent of time, the consistency relation for (1.47), (1.48) and
(1.49) reads
(1.73)
(1.75)
129
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
423
( ( Y , S ) ) ( f J ( Y , S ) , p(r,s)) =
(1.76)
if
f-3
a(r,s),b
t3
b(r,s) and p
t+
c---f
$(Y,s), q -++q(r,s),
V)
a(fi,q) and b
+-+
++
b(fi,q),then [ah]
f----f
( 4 f i , q ) , b ( f i > q ) )V.
130
424
H.J . GROENEWOLD
for operators of occuring types, a restriction of the admitted operators could not help. Therefore a solution of problem a5 with # and q
as parameters, which satisfies the dynamical conditions, is impossible, just as a solution of a3,which describes the quantization of the
classical equations of motion b y the same rule as the quantization of
the classical observables.
This is in point of fact all we have been able t o prove. Though p
and q may seem the most satisfactory choice of parameters in a
description of type S*, the formal disproof of just this description
does not involve the impossibility of any description of type St. A
complete proof of the impossibility of a description of type S j does not
seem simple and neither does the construction of such a description.
For a pair of continuous time independent parameters Y and s
condition V would have to be satisfied. When the commutator of
r and s commutes with r and s, V is self contradictory just like V'.
I t is doubtful whether V can be consistent in other cases. A pair of
continuous time dependent parameters r(t) and s ( t ) must at every
time t be unique single-valued functions of the initial values r(to)and
s(to) at an arbitrary time to. Then instead of the time dependent r(t)
and s ( t ) the time independent r(to)and s(to) can serve as parameters.
Therefore, if our expectation about continuous parameters is justified, the difficulty for such parameters lies mainly in the consistency
of V. It is difficult to see how parameters with entirely or partially
discrete values or of too bizarre continuous type could give a satisfactory description of type S'.
There are still more difficulties for a description S' as we will see
in a moment.
1-19 Q.uasi-statisticaZ description. Whereas it is doubtful whether
the dynamical condition V can be fulfilled, conditions III and I V can
be satisfied without much difficulties. With a solution of the latter
conditions only, one can construct a quasi-statistical description of
the 1st kind Q', which looks very similar to a formal statistical
description of the 1st kind S,l, but in general does not satisfy the
dynamical (and, as we will see in a moment, other necessary) conditions. A solution of 111 and I V gives according to (1.39) the correct
average values. But the real distribution function k ( ( ) corresponding
to a H e r m i t i a n non-negative definite statistical operator k of a
quantum state (pure state or mixture) is in general not non-negative
definite.
131
ON T H E P R I N C I P L E S OF ELEMENTARY QUANTUM MECHANICS
425
Kl(s)
132
426
H.
GROENEWOLD
Ji
= Tr(k, k).
(2.03)
PPV
c x,
(2.04j
=1
Cr
Dev(k;a) = X X , , U ~ - ( ~ ~ , ~ U , ) ~= ~
P
(2.05)
~ X , , X ~ ( U , - U ~ ) ~ .
PI"
133
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
427
(2.02) gives
= Z K,Dev(k,; a)
Y
(2.07)
+ Q X k, k,(Ex(k,; a) - Ex(k,;a))2.
(2.08)
Id
Dev(a) = E x ( ( a - E ~ ( U ) =
) ~Ex(a2)
)
- (Ex(&)). (2.09)
In an ensemble, in which this deviation is zero, a must have the
same value in all samples. Then it follows that for every function / ( a )
E x ( f ( a ) )= f(W4).
(2.10)
Whereas in general a has a proper value only in a sample and in an
ensemble only an average value (expectation value), one can speak
of the proper value of a in an ensemble if the deviation is zero.
In quantum mechanics it is not entirely clear what is meant by
the square or another function of an observable. In order to discuss
things, let us have recourse for a moment to the notion of 1.10 and
let a stand for the observable represented by a(a t--f
a ; problem
x4). Then (2.09) is only identical with (2.02) for all states k if
134
428
H . J. GROENEWOLD
a2 +-+
135
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
429
t2k-1.
(2.12)
For the sake of simplicity we impose the condition that different time
intervals do not overlap
t, > tk-1.
(2.13)
Then the couplings between the various pairs of adjacent systems
can successively be treated separately.
I n 1 we choose a complete system of orthonormal wave functions
cp;,(t). The time dependence can be described with the help of a
H e r m i t i a n operator HP(t) according to
(2.14)
of
d,(t)
(H12(t)- H':(t))cp;,(t)= c p ; , ( t ) G p 2 ( 4 .
(2.15)
tza
(2.16)
given by
(2.17)
136
430
H.J. GROENEWOLD
(2.18)
(2.17) becomes
(2.19)
Yl,(~l)cp2,(~1~*
(2.20)
The system of cpZP(tl)need not be complete.
For t > t,, after the coupling has been dissolved, 1 and 2 have
separate H a m i 1 t o n i a n operators Hl(t) and H2(t).The orthonormal functions cplp(tl) and yZp(t2)then transform into the orthonormal functions
t
and
(2.21)
(t
4).
(2.22)
Q t2m-2).
(2.23)
137
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
43 1
The statistical operator of the first m systems after the last interaction then becomes with (2.23)
xl,(t,)klp(t)k2~Y(t)
klz,.,(t) =
* *
*kmpv(t) (
h m q 6 t 6 t2m-2). (2.25)
(z+I)
= C .lpp(to)k~pp(t) *
...m(t)
=T
~k1z2...m(t)
.k~-l)PP(t)kp+l)PP(t) *
*kmpp(t)
(2.26)
(Trl denotes the trace with respect to the variables of 1). More general the state of a selected series 11, I,, . . . lk (1 <I1 < 12 < . . . Ik <m)
out of the chain of the first m systems irrespective of the state of the
other systems is described b y the statistical operator
khl,..&(t) =
xl,,(to)khPP(t)kliPPIL(t)
* k2k/.4,LL(t)
(t
h k 7 4 ).
(2.27)
correlation with other systems has also partially destroyed the correlation between the selected systems themselves. According t o the
remaining correlation only individual pure quantum states cpl,(t)
of the systems ZI,Z2,. . .Ik with the same Greek index occur together.
We denote a state of a group of systems, which has come about
by interaction with other, afterwards ignored, systems as an infringed state. ((2.25)is the entangled state (2.27) the infringed state).
We consider two particular cases of infringed states. First we put
m = n and let the selected series consist of the systems 1 and n only,
(2.27) then becomes
kln(t)
x l ~ ~ ( t , ) ~ l ~ ~ ( t ) k ,(t
, ( t>) t2n-3).
(2.28)
138
432
H. 5. GROENEWOLD
(2.29)
k1P) = C Xlp&)klpp(t)
(t
tl),
(2.30)
which determines the infringed state of the original object system after the interaction with the measuring instrument, irrespective of
the final state of the latter (and of the observer).
2.04 The measurement conclusion. When the original object system
and observer are connected b y a measuring instrument, which consists of an unramified chain of one or more interacting systems, it
follows from (2.28) that the conclusion about the original object
system, which the observer can infer from his final perception,
certainly cannot go further than to indicate which of the pure
quantum states cpl,(f) is realized. According to the quantum notion
of observation 0, the observer can in principle actually infer that
conclusion under ideal conditions and he cannot infer more under
any condition. This rule establishes the connection between the
mathematical formalism and the observers perceptions. The rule
does not follow from the formalism. The formalism is in harmony
with the rule. The rule justifies the representation of the formalism
in terms of pure quantum states.
The conclusion derived from the measurement thus consists in
indicating which pure quantum state of the mixture (2.29) or (2.30)
of the extended or original object system is realized after this
measurement. I t could indicate equally well the realized pure quantum state of an arbitrary system or group of systems of the measuring instrument. For a great number of measurements on identical
object systems with identical initial operators the statistical probability of realization of a pure quantum state with index p is
139
~~
433
XpklpY(t)k2pv(t).-
entangled state
.1
infringed state
selection
J.
ignoration
measured state
PlV
X&l,(t)
kl,(t)
(2.3 1)
[HQl+*)VLa&)l = 0.
(2.32)
In the pure quantum state cpl,(t) the observable al(t)has the value
alp(t).A measurement, which decides which of the states (pC(t) of I
is realized, also determines the value of al(t). I t can be regarded as
a measurement of the observable a@). This establishes the experimental meaning of the value of an observable. Meanwhile, remembering 2.01, one should be careful in regarding alp(t) as the
proper value of a&),
If all eigenvalues of al(t) are different
(2.33)
alp@)# a,&) for p # v,
the value of al(t) on the other hand uniquely determines the pure
quantum state of the system 1. Therefore, instead of indicating
which state cpr,(t) of 1 is realized, the observer can in the ideal case
(2.33) equally well (and otherwise less well) record the value of
Physica XI1
28
140
434
H . J. GROENEWOLD
= %z
-qzp),
(2.34)
(2.35)
we denote the measurement as a (pointer) reading. l is called the
scale system. The measuring result of a reading can be expressed by
the value of the coordinate of the scale system.
A simplified model, which gives such a coupling between the
systems (I - 1) and I, that the values of the observables a(z--l)(t)
are measured by the values of the coordinate ql, is obtained)
with a H a m i 1 t o n i a n operator of the type
H(l--l,&)
= w%-l)(t))
+ /(a(L-l)(t))Pz.
(2.36)
141
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
435
(2.18) gives
(2.38)
We suppose that the wave function of the initial state of
ql-represent ation
Tzokz; to) = s(qr - qro),
so that qrhas the initial value qlo
Z reads in
(2.39)
(2.40)
(2.41)
(2.42)
(2.43)
QlCptJtl)
(2.44)
= qlpcpzp(t,)*
(2.45)
v,
(2.47)
pl=-7-.
aqr
-4rJ.
(2.48)
142
436
H. J. GROENEWOLD
2.08 Observational connections. For a relation between observational d a t a we need a t least two measurements. We consider two
succeeding measurements during the time intervals (to,tl)and &ti)
with
t;, > tl
(2.49)
performed on the same system 1. The first measurement measures
the states cpl,(t) or a corresponding observable al(t),the second one
measures the states cp;,(t) or a corresponding observable a;@).
If the first measuring result indicates the final pure quantum state
cpl,(t) (tl < t < th), the statistical operator at the beginning t;l of
the second measurement is klpP(t;),
which is expanded according to
k1,Jth) = c X;,,,*,G>
with
l4,,M
,,
4pp,v~p4d
= Tr(k;y,,.(t~)kl,,(t~).
(2.50)
The statistical probability, that, after the first measuring result has
indicated the pure quantum state cpl,(t) (tl < t Gth), the second
measuring result will indicate the pure quantum state &,.(t) (t ti) is
4,,,43
= T%,VG)
klpp(&))
I ma cpl,(th) 12*
(2.5 1)
This conditional probability is actually the most elementary expression contained in the formalism, which denotes an observable
connection and which has a directly observable statistical meaning.
When the functions &(t) coincide with the cp(tlp),i.e. when a;(t)
and al(t)commute, (2.51) becomes
4 p p , v , J ( t ; ) ) = SJp
(2.52)
(2.53)
The probability of finding system 1 in a state kl and 2 in a state k2 is
(2.54)
PlV
When kl and k2 coincide with the projection operators kl, and k2,
143
ON T H E P R I N C I P L E S OF ELEMENTARY QUANTUM MECHANICS
437
k;2 =
Xppkl,,,
k2,p
(2.55)
instead of the state (2.53). I n this way the correlation between 1 and
2 would be partially destroyed by the omission of the non-diagonal
terms. In the mixture (2.55) the expectation value of the states kl
and k2 would be
(2.56)
instead of f2.54). It has been emphasized by F u r r y 3) (*in a somewhat different form and particularly against our common opponents,
cf. 2.1 1) that only if neither kl nor k2 coincides with any of the klPP
or k2, respectively, (2.56) can be different from (2.54). Because the
latter case hardly occurs in the relevant applications, one is apt
to make the mistake of replacing (2.53) by (2.55) (and to draw unjustified conclusions whenever this case does occur).
If 1 and 2 had been coupled with one or more further systems
3,. . . . according t o
k123.... =
xvp k l p k2p
bP..
*
(2.57)
PL.V
and these further systems had been ignored afterwards, the infringed state of 1 and 2 would correctly be given b y (2.55) indeed.
This infringed state is quite distinct from the entangled state (2.53).
2.10 Multilateral correlation. I n (2.53) the transition operators
klPvand kZPV
belong to two systems of orthonormal wave functions
y I pand y 2 p ,which span the (generalized) H i 1b e r t subspaces R1 and
R2. An interesting case 4) is that for which k12can similar to (2.53)
also be expanded with respect to the transition operators llpaand lzPu
belonging to any two systems of wave functions + l p and +2p in
R1 and R2, when one system is chosen arbitrarily variable but
orthonormal and complete, the other system suitably to the first
(2.58)
144
438
H. J. GROENEWOLD
(2.60)
xp 'Plp 'Pzp.
P
Xvp
+ .
(2.61)
(2.62)
I xpv j2
= 0 ( I = 1 and 2).
(2.63)
145
ON T H E PRIXCIPLES O F ELEMENTARY QUANTUM MECHANICS
439
2 xvp7-r2(k,
hup) klpv
(2.65)
= ,A 11,.
PIV
and
(2.67)
and similarly for interchanged indices 1 and 2. (2.66) and (2.67) must
be identical according t o (1.08). Because one of the systems l l p uor
l,, is arbitrarily variable and complete in R1 or R2, we must have
(remembering (2.64))
xpIvxvp = x 2 x p # p ;hp&p=h~hpp (. = A
> 0).
(2.68)
= x ; ~ u p = x ~ p J l ~ p l = v .(2.69)
x v p-- X*y x p , [ x p I
h p=
A c x p ( 4 f p Yip)
(P2/.47
(2.70)
P P
A PL.
c xvp Trl(kl,v1lup) k z p v ,
UP
(2.72)
146
440
H. J. GROENEWOLD
and
x v p T 7 1 P l p v llup)
(2.73)
= AupTr2 (k2vp f 2 p A *
Therefore
P*V
(2.74)
P*O
PIV
(2.75)
hup11pu12pu13pu.
PIU
12up)
(CYCJ.)
(2.76)
I xpv j 2
= I hpu 1
or TrJ(kJpv
11,) = 0
(2.78)
= Tr3(k3pv13op) = 1 or 0. (2.80)
This would mean that the systems of llpu,12p0 and l,, should (but
for a simultaneous change of enumeration of the Greek indices of
the three corresponding operators and but for unimodular constants)
,
and k3pv.This is against the
be identical with those of k l p ukZpv
assumption. Multilateral correlation between the states of 1, 2 and 3
is therefore impossible.
147
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
441
k123....=
x ~ v p k l p v k 2 p v k 3 p v -*
Aupllpo12p013pu*
P.0
PPV
- - - (2.81)
*
Then
(2.82)
( k l p p 'lpp)
xpp7-rdk2,p
'2pp)
= AppTr2(k2pp '2pP)
(2.83)
= ~ p p W k 1 p hpp p )
and
hPP)= h;pTrdkrpp
x;p%(klPP
Ilpp)
(I = 12%
(2.84)
so that
x P p = & A,
or Tri(klpPllp,,)
= 0 (I = 1 and 2).
(2.85)
Because one of the systems lip,,is arbitrarily variable the latter alternative is excluded and because the traces in (2.83) are non-negative
we must have
XPP = hpp.
(2.86)
Further we have similar to (2.65)
7.Yl(klpp11pp)k2pp
= I2pp
(2.87)
from which we derive
llpp1luu
c~
~ l ~ ~ l ~ ~ ~ l p p kzpp
~ ~ ~ l ~ ~ (2.88)
l p p ~ l u u ~
and
1lpp
a,,
Tr1 W l p p
=
P
lpp)
$xJ
k2pp
(2.89)
(2.90)
= 6pu
(2.91)
148
442
H. J. GROENEWOLD
The sums (which are identical) denote the projection operator of the
In the mixture (2.92)all states
(generalized) H i 1b e r t subspace RI.
in R , have the same probability x2 = A2. If R1 coincides with the
entire (generalized) H i 1 b e r t space of wave functions of 1, the infringed state (2.92) becomes entirely undetermined.
If in dealing with the entangled state (2.58) one would make the
mistake pointed out by F u r r y (cf. 2.09), one would get
x2
kl,
kz,,
= A2
C h p p I,,,.
(2.93)
In dealing with (2.82) we have seen that (2.93) cannot hold. (2.85)
does not express a correlation between pure quantum states of 1
and pure quantum states of 2 (in the way a member of (2.93) would
do).
If, however, (after the interaction between 1 and 2, which establishes the state (2.58)) one of the systems, say 2, interacts with a
measuring instrument, which measures the states 12pp, the infringed
state of 1 and 2 together after the latter interaction 1s
12 c 11pp12pp.
(2.94)
P
149
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
443
type. When the measuring result selects for 2 the state l,, the
state of 1 is 1lPP.
After the interaction between 1 and 2 has taken place, an observable bl of 1 with eigenstates 11, can be measured in two different ways: either by a direct measurement on 1, or by measuring
an observable b2 of 2 with eigenstates 12,, (corresponding to llpP)
by a direct measurement on 2 (then 2 can be conceived as a part of
the measuring chain). At a first glance it might seem surprising and
perhaps even paradoxical that it is still possible to decide which
observable of 1 will be measured by a measurement on 2 after all interaction with 1 has been abolished 6, and that it is possible to measure independently two incommensurable observables al and
bl([al,bl]# 0) by applying the two measuring methods side by
side ') ". (Of course one should care for not making the mistake of
(2.93), which would naturally lead to paradoxical results).
When the eigenstates of a, are kl, and those of bl are 11,, a
measurement of al selects a state out of the left member, a measurement of bl selects a state out of the right member of the expression
(2.92) for the infringed state of 1. The probability that one measure
ment selects the state kl,, if the other selects the state llpp(or
opposite) is according to (2.51)
(2.95a)
150
4 44
H. J. GROENEWOLD
When the measurement of bl has selected the state llPp,the probability that the measurement of c1 will select the state mlrris
(2.95~)
Thus we get two different probabilities for the same event. This is
not unfamiliar in statistics, because the probabilities are (always)
conditional. They have onIy a meaning for a great number of combined measurements of al, bl and cl. The probability of finding a
state kl, is x2,the probability of finding a state 11, is h2,the probability of finding a state ml,is then according to (2.95b) or (2.95~)
x2 X Tr(kl,, rnlT7)or 1 X Ty(llP,,
fi
ml,).
(2.96)
15 1
ON T H E PRINCIPLES O F ELEMENTARY QUANTUM MECHANICS
445
through the diaphragm and the change of momentum of the diaphragm. The slits can b e taken so far apart, that exchange effects
can be neglected.
(2.97) is of the form (2.60)with (2.59), as can be seen by expanding
(2.97) with respect t o e.g. coordinate or momentum eigenfunctions
of 1 and 2
1
i[ P
Q 6 (qz - - 2
p 12 = .\/h dE e
8 q1-E 2)
(2.98)
3. Operator relations.
3.01 Exponentials. I n the ring of operators a generated by two
non-commuting ,He T m i t i a n basic operators p and q, for which
[p,q] = 1, i.e. pq - qp
>
= 7- (tt
a
(3.01)
0),
- e ~ P e ~ q e -(3.02)
~ .
152
446
H.J. GROENEWOLD
With (xp
+ Yq),
UXP
(XP
+ Y@I
(3.03)
= xyf -P
i
- ,T (xp+yq)
e-((x+X)P+o)+YWl)
n
eT(XP+YW e- 5 W Y Y z )
(3.04)
.-x i
(tP+9q)
i
i
ex
h + Y d ex ( t P + T d -
(xp+yq)
(.?j-Yt)
(3.05)
Lh2/ j d q dq ex
(xP+rd
b)J
(3.06)
J- jjdE
drl e-
b (tP+W)
h2
(3.07)
Further analogous to
3.08)
we have
=//dc
d-q e-x
(tp+~q)
(tp+vq)
a(6) s(-q) = a.
(3.09)
In the same way as (3.08) and (3.06) show that every (normalizablej
function a(+,q) can be expanded into a F o u r i e r integral
(3.10)
(3.09)and (3.07) show that every operator a (with adjoint at) can be
expanded into
153
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS 447
(3.11)
UtU = 1,
(3.12)
a = UtaU; 9
= Uy, pt = qtut
(3.13)
(3.14)
(a&
(3.16)
- T yeF( a f y q ) - 6(x)6 ( Y ) ,
h
(3.18)
154
448
H . J. GROENEWOLD
(3.19)
and
(3.20)
(3.18), (3.19) and (3.20) are entirely analogous to (1.13), (1.14) and
(1.15). (3.18) and (3.19) respectively express the orthonormality and
the completeness of the systems of operators
i
1 p(.P+Yq)
4. Correspondence.
4.01 v o n N e u m a n ns rules. We now examine the rules of
correspondence I, 11, 111, 11 and V. First I and 11.
We show that if between the elements a of one ring and the elements a of another ring there is a one-to-one correspondence
a c-+
a, which satisfies v o n N e u m a n ns rules (cf. 1 .lo)
if a t+ a, then / ( a ) +-+
if a
a and b
t-+
t+
b, then a
/(a),
+ b +-+a + b,
I
I1
+ b)2 - a2 - b2 = ab + ba +-+ab + ba
(4.01)
2aba
t-+
(4.02)
+ b a y - b(2aba)- (2aba)b =
_-
Therefore we have
ab - - a
t+
f (ab-ba),
(4.04)
155
ON THE PRIKCIPLES OF ELEMENTARY QUANTUM MECHANICS
449
t+
if 4 p , q )
+-+
++
[a,bl V'
is self contradictory.
With
p2 ++ x1, q2 ++ x,; p 3
we find from
4 (P2,q) = p
4 (p2,p)= 0
++
Yl,
4"
+-+
(4.06)
Y2
Q [x,,ql = Pt
f-+
$ [Xl,Pl
(4.07)
=0
p 2 f+
p2
+ c1, q2
++
q2
(4.08)
c2
and from
(4.09)
(and similar relations for q3 and y2) that
dl,
q3 ++ q3 4-3C2q f d, (4.10)
+(P3,q2)=pZqt-f~[(P3+3c,P+d1),(q2+c2)1=4(P2q+qP2) fclq@* 1)
(4.11)
Pq2
3(pq2+q2p)+C2P
+-+
and
f (p3,q3)= p2q2
= B (P2$
++
Physica XI1
3(p3
ClCZ.
(4.12)
29
156
45 0
H . J. GROENEWOLD
a @2q,pq2)= P2q2
=
+-+
C2P2
ClC2.
C2P)l
(4.13)
a =lf
' q P(#,q) m(+,q) a@Jq) (4'14)
with a transformation nucleus m@,q),
which satisfies (1.57), (1.58) ;
a(fiJq) = Tr(m(fi,q)a),
(1.59), (1.60)
Trm@,q)= 1,
(4.15)
(4.16)
fJ@ a4 P(Pd m@4)= 1;
Tr(m@,q) m(?',q')) = P-'(ib,q) ' ( p
' ( q -q')J
(4'17)
/ l d p dq P(P,q) Tr(m(pJq)a) Tr(m@,q) b, =
= Tr(ab) (for every a and b), (4.18)
When we replace in (1.56) the complete orthonormal systems
Kz,(p,q) of (1.54) and k,, of (1.15) by the complete orthonormal
systems
i
1 -2
-e
A ( x P I - 9 4 ) of (3.10) and eT(Xp+yn) of (3.20),
h
we find a solution
m(p,q)=
dy e t ( z p + y q ) , - ~ ( z f i + ~ q )
(4.19)
P(PJq) =
1
'
(4.20)
157
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
451
ab =
I/. . . .
dy dx dy d p dq dp dq
/dx
h4
. e -h ((x+x)p+(r+r)s)
@Y-YX)
- ++Yq+xP+Yq)
q=y+y,
x-x
5 = ___
2
Yi
=-y -y,
a=
6 =
+ P
2
p - p)
7 1
+ q
4=-
= q - 41,
(4.24)
this becomes
(4.25)
The expressions in brackets at the end are a symbolical representation of T a y 1 o r expansion. With the substitution
i4.26)
F-. x, y1-+ y , a+ p , T+ q
we get by partial integration
(4.27)
158
452
H. J . GROENEWOLD
. (ab - ba)
i
4 (ab + ba) and 2
the correspondence
b($,q)
t+
i
b($,q) t+ - (ab - ba). (4.29)
2
--)
if a
t+
(4.32)
only holds if
(4.33)
First take for n a homogeneous polynomial in p and q of degree n.
An elementary calculation shows that the condition
(4.34)
or
(4.35)
159
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS 453
(4.38)
If a@,q) or b(j5,q) is a polynomial in j5 and q of a t most 2nd degree,
we have a special case for which the bracket expressions ((a,b))and
(a,b) coincide.
The correspondence (4.21) is a solution of 111 and IV. We have
not investigated the possibility of other solutions with the same
parameters P and q.
5. Quasi-distributions.
5.01 Proper and improper representations. With W e y 1s correspondence (4.22) as a special solution of
1t+1
if k t+ K(p,q) and a
I11
t+
a(P,q),
IV
Wklk2)
l
I^Sdp d q k,(P,q) k,(P,q) = 0,
h,
(5.01)
the product klk2or K,(p,q)K,(p,q) vanishes in the proper representation, but in the improper representation it need not. The equations
160
454
H.
J. GROENEWOLD
(5.02)
Because the wave functions cpp are only determined but for a
factor eiIRYP (y real), the K,,(p,q) are only determined but for a factor
ei (Yp-Yv). The distribution functions, which are thus obtained with
W e y 1s correspondence 2) become identical to those given by
W i g n e r 10).
5.03 Proper value. In a distribution k or @,q) a quantity a or
a($,q) can be regarded to have a proper value if the condition (2.10)
Tr(kf(a))
= f(Tr(ka))
(5.03)
or
(5.04)
161
~~
455
and k an eigenstate of a. For any H$,q) the validity of (5.04) requires that K(9,q) is of the form
Ya($,q) -
(5.06
5.07)
Qp(d
-Y(P)
dY V Y q - P) e h
for x
= 0.
a(xP f yq)(Pp = 4 P ) Y p .
(5.08)
(5.09)
I62
456
H. J. GKOENEWOLD
(5.10)
is the mass, w the classical circular frequency of the binding. We
consider 9 and q as new canonical coordinates and omit the dash.
In q-representation the normalized stationary solutions of the
wave equation
iiz
(5.11)
(5.11)
are
~ ~ (=4 )
-__
_ _1
2h
QH, __
(TZ=
4/2n!
The H e r m i t i a n polynomials
function
H,
(5.14)
457
(5.15)
This gives
(x
2
LLL;(z;
(P2+4'))
ei (tn--n)
(arc tan
2 --wi)
. (5.16)
_ _ --
at
- wq, = wp,
at
(5.17)
with solutions
= a cos (ot-
(5.18)
164
458
H . J. GROENEWOLD
(5.19)
By ignoration of one or more systems of the measuring chain the
non-diagonal functions (p # v) are dropped and only the diagonal
functions remain. Instead of (5.19) we get
(5.20)
(The latter &-function is actually a remainder of the ignored distribution functions). The effect on (5.19) of ignoration of other systems
is formally the same as that of integration over fi with density
function 1 / h . This illustrates even more plainly than before (cf. 2.07)
how the correlation between pl and other observables is completely
destroyed by the reading of ql. So far there is no difficulty with an
interpretation of the 1st kind. We are only concerned with the value
of ql, which is a proper value and uniquely determines the distribution (5.20). The value of f l is indifferent. As soon as inference is
made about other systems in the chain with overlapping k,+(p,q),
correct results are only obtained after the integration over pl (with
density function l/h) has been performed (cf. 1.19). In a description
of the 1st kind this integration could only be interpreted as an
averaging over a great number of measurements. But the integration has already to be performed in a single reading and therefore an
interpretation of the 1st kind is excluded.
165
ON THE PRINCIPLES OF ELEMENTARY QUANTUM MECHANICS
459
WPQ
Q( ~ , ! ? ;
I #2,!?2)=
(5.22)
(properly instead of \5.21) we should use eigendifferentials). The infringed distribution after a measurement of q2 or f 2 can be found
from. (5.22) by integration over f12 or q2 respectively with density
function I/h. This gives
(5.23)
or
(2.24)
respectively. For the distribution (5.21)this becomes
(5.25)
The correlation between f i l and f i 2 or q1 and q2 respectively has entirely disappeared.
If the state of 2 is entirely ignored, the distribution of the infringed
state of 1 can be found from (5.22) by integration over p 2 and q2
with density function l/h. This gives
h
(5.26)
For the distribution (5.21) the result is llh, the infringed state is
entirely undetermined (the normalization can be understood from
(5.26)). A measuring result q2 = qZpor p 2 = p2,, selects from (5.23)
166
(5.27)
or
(5.28)
For (5.25) this gives
1
-S(P1
k
P2,-P).
(5.29)
REFERENCES
167
I: 99 1
QUANTUM MECHANICS AS A STATISTICAL THEORY
BY J. E. MOYAL
Communicated by M. S. BARTLETT
for considering dynamical parameters as statistical variates, not only for large aggregates, as in classical kinetic theory, but also for isolated atomic systems, is quite fundamental in justifying the basic principles of quantum theory; yet paradoxically, the
expression of the latter in terms of operations in an abstract space of state vectors is
essentially independent of any statistical ideas. These are only introduced as apost hoc
interpretation, the accepted one being that the probability of a state is equal to the
square of the modulus of the vector representing it; other and less satisfactory statistical
interpretations have also been suggested (cf. Dirac (I)).
One is led t o wonder whether this formalism does not disguise what is an essentially
statistical theory, and whether a reformulation of the principles of quantum mechanics
in purely statistical terms would not be worth while in affording us a deeper insight
into the meaning of the theory. From this point of view, the fundamental entities
would be the statistical variates representing the dynamical parameters of each
system; the operators, matrices and wave functions of quantum theory would no longer
be considered as having an intrinsic meaning, but would appear rather as aids to the
calculation of statistical averages and distributions. Yet there are serious difficulties
in effecting such a reformulation. Classical statistical mechanics is a crypto-deterministic theory, where each element of the probability distribution of the dynamical
variables specifying a given system evolves with time according to deterministic
laws of motion; the whole uncertainty is contained in the form of the initial distribu.
tions. A theory based on such concepts could not give a satisfactory account of such
non-deterministic effects as raclioa,ctive decay or spontaneous emission (cf. Whittaker ( 2 ) ) . Classical statistical mechanics is, however, only a special case in the general
theory of dynamical statistical (stochastic) processes. I n the general case, there is
the possibihty of dEusion of the probability fluid, so that the transformation with
time of the probability distribution need not be deterministic in the classical sense.
In this paper, we shall attempt to interpret quantum mechanics as a form of such
a general statistical dynamics.
I. QUA.NTUM KDTEMATICS
2. THE EXISTENCE OF PHASE-SPAUE DISTRIBUTIONS I N QUANTUM THEORY
168
100
J. E. MOYAL
S in the Hilbert space of the state vectors. The probability of finding sc in a state @ is
then equal to the square of the modulus I a$l2 of the projection a, of @ on the corresponding eigenvector 1C.i. A complete or irreduciblerepresentation for a given mechanical
system is given by a set of commuting observables s such that their eigenvectors pi
span the whole space, i.e. such that any ?b, = Ca,$i. Hence we obtain directly from
i
II.the joint distribution of the variables s. It is known, however, that these s are not
sufficientin themselves to specify the system completely; we need, in addition, another
complementary set, say r, which does not in general commute with s; for example,
a complete representation is given by either the Cartesian coordinates q or their conjugate momenta p, but the complete dynamical specification of the system requires
both qs and ps. Hence, the phase-space distributions of complete sets of dynamical
variables, which are required for a statistical theory, are not given directly by y?.
It has been argued(3) that such distributions do not exise, because of the impossibility of measuring non-commuting observables simultaneously. This argument is not
conclusive for two reasons; one is that the impossibility of physical measurements does
not preclude us from considering the proposition that there exists a well-defined probability for the two variables t o take specsed values or sets of values; in fact, the theory
of probability is introduced to deal with suoh situations where exact measurement c
ir
impossible (see Jeffreys ( 4 ) ) .The other reason is that it is possible in principle to form
operators G corresponding to functions #(r, s) of non-commuting observables; the
expectation value of G in a state 9 is then given by the scalar product ($, GP). But the
j o h t distribution of r and s can be reconstructed from a set of such expectation values,
e.g. the values of all the joint moments ykSn. The formalism of quantum theory allows
us therefore to derive the phase-space distributions indirectly if a theory of functions
of non-commuting observables is specified and conversely.
There are serious difficdties to be met, however, in defining these distributions
unambiguously. This may be seen, for example, in the case of the harmonic oscillator.
The energy eigenvalues form a discrete set E,, = (n 8)hv, The corresponding eigenfunctions uu,(q),un(p)are sets of Hermite functions, continuous in p and q. Hence any
joint distribution for p and g in a state consistent with the individual distributions
must extend continuously over the whole ( p , q ) plane, while any joint distribution
for the energy H = +(p2/m+27rmvq2)and the phase angle 0 = tan-lp/q consistent with
probabilities an@ for En,wiU be concentrated on a set of ellipses
+(p2/mi27rmvq2)= (ni-9) hv.
We are thus forced to the conclusion that phase-space distributions are not unique for
a givert state, but depend 0% the variables one is going to measure. In Heisenbergs words (61,
the statistical predictions of quantum theory are thus significant only when combined
with experiments which are actually capable of observing the phenomena treated by
the statistics. Since the introduction of statistical concepts in atomic theory is
justsed by an analysis of the interaction between observed system and observer,
it is perhaps not surprising that different distributions should arise according to the
169
101
We denote by r a set of commuting observables or operators giving a complete representation, s the complementary set, such that s do not commute with r and that r
and s together form a basic set of dynamical variables, characterizing a given system;
r and s me their possible values or eigenvalues (these are, of course, ordinary commuting variables). The most natural way of obtaining the phase-space distribution
P(r,s) is to look for its Bourier inverse, i.e. the mean of exp{i(mi-Bs)} (known in
statistical terminology as the characteristicfunction). On forming the corresponding
in
operator
(3.1)
M(7,B) = exp {i(m+ 8s)) = 2 --J
(71-+ B s ) ~ ,
n
(3.2)
From well-knownformulae for Fourier inversion, the phase-spacedistribution function
is then
(3.3)
(3.4)
for discrete eigenvalues ri , 8k (Cram& ( 6 ) ) $ .
The operator (3.1) takes a specially simple form for canonically conjugate coordinates
and momenta q, p (pq-qp = &/i),
M(7,0) = e ~ < k Oe i 8 4 e h P = e-i%7PeiOCe?IiTP
(3.6)
(cf. Kermack and McCrea (7)).From the second expression for M,we find
(3.6)
co t o + a.
$ The term distribution function is used in this paper t o denote the probability density of
conthous eigenvalues, and the finite probability of discrete eigenvalues.
170
J. E. MOYAL
102
and hence by Fourier inversion
F ( P ,P) =
,s$*(s
1
- &) eiT9 @
+ liw
(d7,
!I
(3.7)
(3.8)
M ( T , ~=) JL-)
= h-f
ss
[@*(q)$(p)
&@/fi]
e+fi7i7eeib+W)pdq
fs
ef(Wa*/a~
&[@-*(q)#(p)
e { ~ / f ei(TP-k&)dpdq,
i]
(3.9)
(3.10)
PHASE-SPACD EIQENPUNCTIONS
@ = Z%@z
1
(4.1)
in the expression (3.3) for F(Y,s),we find for the latter the expansion
(4.2)
(4.3)
(4.4)
(4.5)
171
(4.6)
(4.7)
(4.8)
(4.9)
In the general case, this follows from the fact that (4.3)and (4.4) or (4.5) form a unitary
transformation from a vector, say @lk, of oomponents ?,41*,@k in the product space of
the vectors ?,h*with the vectors @, to ,f& The veotors @lk form a complete orthogonal
(and self-orthogonal) set, and these properties are invariant under a unitary transformation. Furthermore, it is easily seen from their definition that the fir, form a
Hermitian matrix with respect to their subscripts I, k
fik(r,
s, = ff21(r,
(4-10)
We shall see later (557 and 8) that the& can be interpreted as the eigenfunctions of
characteristic equations for the phase-space distribution functions, corresponding t o
the eigenvalue equations of the 9 s ; we therefore call them piwcse-spaceeigenfunctions.
In the case of the canonical coordinates and momenta q and p , relations (46)-(4.9)
can be proved by elementary methods (cf. Appendix 2), and the f z A ( pq)
, have the
explicit expressions, corresponding to (3.7) and (3.8)
1
t i k ( p ,9) = % J h ( q - iw + ~ p $lc(q 467)a7,
(4.1 1)
(4.12)
(4.13)
(4.14)
172
J. E. MOYAL
104
6 . MBANVALUES,
The mean value of an ordinary function #(r,s) taken with respect to the phase-space
distribution P(r,s) is
B = JJ#(Y,
$1 P(r,s)dras
= [/[[Q(r,
s ) ($,ei(*+OQ$)e47r+0@drdsd~d6
(5.1)
where y(7,6) is the ordinary Fourier inverse of G(r, s)
y(7, 8) = S f a c r , s)e-q7?+")drc~3.
(5.2)
(5.3)
which is thus the operator corresponding t o the ordinary function G(r,s) in our theory.
It now ~OUQWS
that the matrix Gl,c of G in any representation of eigenvectors $-1 can be
obtained by integration of the ordinary function G(r,s) with respect to the corresponding
pkse-space eigenfunctionflrc(r,s)
Z > G$k)*
(5.4)
Since file is a Hermitian matrix with respect to I and k, we see at once from (5.4)that
Glkwill be Hermitian if Q(r,s) is real.
The operators and matrices corresponding to any function of the basic variables
r , s are thus uniquely defined by the phase-space distributions. I n other words, our
theory of phase-space distributions is equivalent t o a theory of functions of noncommuting operators. Inversely, this theory of functions defines the phase-space
distributions uniquely.
In the special case of functions U(p,q) of canonically conjugate coordinates and
momenta, (6-3) coincides with an expression derived by Weyl(9)on group-theoretical
for P ( p ,q) is
considerations.A n alternative expression corresponding t o (3-10)
e ~ ( n / i ) a ~ / aGo(q,p),
~aa
(5.5)
where G,(q, p) is obtained directly from the ordinary function G ( p ,q) by writing all
the operators p to the right (e.g.cpp'n), and this order is maintained when applying the
operator e$(fi/i)@/aPagto Go(cf. Appendix 3 for the proof; see also McCoy(10)).The form
of the usual operators of quantum theory: energy, angular momenta, radial momenta,
etc., are not changed when they are derived by this method from the correspon&g
classical functions of p and q.
173
Quantum mechanics as
CL
statistical theory
105
PROUESSES
(6.1)
where R is the distribution of p , q at t conditional in p,, qo at to. X is therefore the
temporal transformation function, and must express the laws of motion of the system.
While Faand P depend on the initial and h a 1 states of the system, K must be independent of these states, and depend on the inherent dynumical properties of the system.
Hence the assumption that K is homogeneous, i.e. invariant for a translation of the
origin in t, and dependent only on the interval t-to (as long as there are no external
time-dependent forces acting on the system).
K gives the transformation for fhite intervals. We now derive the corresponding
infinii;esimal transformation. The characteristic function A for the differences q - 6,
p - 7 conditional in 5 , is~
A(T,8 I 7,E; t - t o )
ss
e'[o(q--8+T@-'l)lK(p,
q 1 7, E; t -to) dpdq.
(6.2)
We make the second assumption that in the stochastic processes of physics, the probability of a transition from [, 7 t o q [, p + 7 in a small interval t - tois of the order of
t-io, For t = to, obviously K = 6 ( p - ~ ) 6 ( q - [ ) and A = 1. Hence (A.-l)/(t-to)
tends to a finite limit L when t +to
(6.3)
We shall call L the derivate characteristic function. If M(r,8; to)is the characteristic
function at to
M(T,8; to)=
fs
ei(Tq+@)
Po(7,
[;to)dTd6,
(6.4)
t o )Fo(7,E;
$0) d
~dt.
174
J. E. MOYAL
106
Hence
(6.5)
(6.6)
(first suggested to the author by Prof. M. S. Bartlett). ( 6 6 ) and (6.6) express the
idnitesimal transformation corresponding to (6.1) in terms of characteristic functions;
they can be inverted to express this transformation directly in terms of distribution
functions. This may be achieved in two ways; if L admits a Fourier inverse
~ ( pq I ,7,&)=
jJ~(7,
B I7 , t )
e([7($--23)+0(5--4)3d7d~,
(6.7)
$(P,
q; t ) =
p(P, I
(6.8)
(6.9)
(where the anr(r,t)are called the derivate moments of the system), then
the differentid equation of infinite order
_F1
satisfies
(6.10)
This reduces to an equation of finite order if the expansion (6.9) for L terminates,
i.e. if the derivate moments vanish above given powers o f p and q.
7. EQUATIONS
OF THP
In order t o derive the equations of motion for the quantum phase-space distributions,
we look for the time derivatives of their cheracteristic functions. We find from the
Poisson-bracket form of the quantum equations of motion
-=JP*(s)[M,WlP(s)dq
aiv
=~~~*(p)[MH-HMlP(q)dl,
at
(74
where M(7,O) is the characteristic funotion operator (3.5), and H the Hamiltonian
operator, expressed from (5.3)by
(7.2)
175
507
~ ( c Tp,) being the Fourier inverse of the aorresponding classical Hamiltonian B ( p ,q).
Hence, using expression (3-5), we obtain
(7.3)
(7.4)
where a/apH, a/aqHin the right hand of (7.4) operate only on H and a/apR, a/aqF
on F . The comparison of (7.3) with (6.5) gives the derivate characteristic function
L(T,O
I Ip,d = ~ [ H ( p + g n s , q - ~ ) i ~ ) - - , Y ( p - - t n e , ( l + ~ ) i 7 ) ] .
(7.5)
(7.5)
6-
(7.7)
(7.8)
which is easily shown equivalent to (6-10)with derivate moments
(7.9)
Inversely, the quantum equations of motion, and in particular the Schrodinger
equation, may ba derived from the equations above for P(p,q;t ) (cf. Appendix 4).
There is thus complete equivalence between the two.
Pinally, we may notice the analogy between the right-hand side of (7.8) and the
olassical Poisson bracket. This may be generalized in the following way. It may be
shown by a method similar t o that leading t o (7*8),that the commutator ifi[RG - GR]
of two operators R, G obtained (e.g. by (5.3) or (55)) from the ordinary functions
R ( p ,q), @ ( p ,q) is identical with the operator corresponding (by the same rules) to
(7.10)
In other words, (7.10) is the analogue of the classical Poisson bracket when the laws
of quantum mechanics are expressed in phase-space, and the commutator is the
176
108
J. E. MOYAL
corresponding operator in a q- or ^-representation. It is also seen from this that operators whose classical analogue is 0 may correspond to non-vanishing phase-space
functions in the present theoryf-
( 8-1)
Substituting in (7-7) and identifying term-by-term, we see that the/#c are the eigenfunctions of the homogeneous integral equation
(8-2)
(8-3)
i, le
7i
Calculating the mean of the commutator [G, M] from the two sides of (8-4)
J(g) [GM - MG] uk(q] dq = (y? - 7,,) j*J e^> gik(p, q) dp dq
/* /*
)N
'
(8/ 0 <6)l i
'
177
109
9. TRANSFORMATION
EQUATIONS FOR PINTTI IXTERVBLS
tO)d/POdqO>
(9.1)
(9.2)
(9.3)
(9.4)
(9.5)
where
=J/"lO(p,q
IpO,qO;t-tO)g,k(pO,qO)dpOdqO,
(9.6)
The expansion of Kl0 in terms of the gilt :K,, = Zhkqik has coefficients
htk =h p G o ( PQ
,
= W C ( P ,4; t - t o ) ,
= E l 0 = K(P,5?Polqo; t - t o ) = F,I=g,l,(Po,c1O)Y~~;G(Ip,4;
t-to).
6k
(9.7)
We have thus found an expression for the transformation function R in terms of the
gik and yilc;horn it we see that K satisfies the iteration relation
'(pB,qZ(pO,qO;
t-tO)
(9'8)
The transformation (9-1) form therefore a continuous unitary group. Stochastic processes satisfying the iteration relations (9.8) are known as Markoff processes (cf.
Hostinsky (11);see also Jeffreys (12)).
The energy eigenfunctions f $ k ( p ,q ) of a conservative system are easily seen from
(9.4) and (9.6) to satisfy the homogeneous integral equation
f , k ( p , q) = ~ - w $ - W f - ~ o ) / f i
(9-9)
178
J. E. MOYAL
110
=K
(Po,40 I PI 4; to -t)-
(9.10)
ss
e-c(Tp--70po)$*(q
457 I q, +h0)
d7a~~.
(9.12)
It is seen from the expansions (8.3) and (9.10) of S and K in terms of the energy eigenfunctions that
&A
4 IPo,4 0 ) =
(10-1)
(10.2)
S(rln,h I Po,Qo) d71d5, * * + dT,,dl,,.
This is easily verified by substituting from (8.3)for Sand comparing with (9.10)for K .
Binoe S has a simple expression, obtained directly from the Hamiltonian, (10.2)
supplies also a convenient method of approximation for K when the energy eigenfunctions are not known exactly. We have thus a new method of solving problems in
*.*
179
111
The transition probabilities ckn(t) from state Ic t o state n are the diagonaI coefficients
Cbn c&akn whose expression in terms of R will clearly be
I
///k(P,
I
= aZnaktb
The present theory should help to elucidate the question whether quantum mechanics
is deterministic in the classical kinetic theory sense?, since it permits a direct comparison between the two. The infinitesimal time transformation of quantum phasespace distributions (7.8) may be written in the form
(11.1)
where @lap,3/84] is the phase-space differential operator giving the classical Poisson
bracket. The corresponding transformation of classical kinetic theory is given by
(11.2)
Its deterministic character may be seen from the fact that the characteristics of this
first order partial differential equation are simply the classical paths in phase-space.
Alternatively, we may say that P is an integral invariant of the transformation
generated by the operator {a/@,2/84); an element Soof phase-space will transform to
8, in the interval t, and
r
(11.3)
This no longer holds in the case of quantum theory; the transformation generated by
the operator (2/7i)sin +Ji{a/ap,a/i3q}is equivalent to {a/ap,a/Q}when applied t o H p , Hq,
but not in general when applied to HP,so that while Sowill transform into X,exactly
as for the corresponding classical system, yet generally
n
(11.4)
Hence the present theory leads t o the conclusion that quantum theory is not generally
deterministic in the classical sense.
In the correspondence principle limit, when h+O, the quantum equation (11.1)
is seon t o reduce t o the classical equation (11.2); this will equally well be the case if
the Hamiltonian H ( p , q ) is a second degree polynomial in q and p , leading t o the
amprising conclusion that systems such as a free or uniformly accelerated particle, or
a harmonic oscillator, are deterministic in quantum theory: this should not be taken
too seriously, since even small perturbations or non-linear terms would, according
t o ( l l s l ) , destroy this deterministic character.
The phase-space transformations with time of quantum theory form a continuous
Unitary group, which reduces therefore to the group of contact transformation of
Cf. in this connexion Whittaker (2), Jeffreys (12) and SO Reichenbech (25).
180
J. E. MOYAL
112
classical mechanics in the correspondence principle limit and for the deterministic
quantum systems whose Hamiltonian is a second degreepolynomial;the transformation
function K of 3 9, which is the probability distribution of p and q at time t conditional
in po,qo, a t time to, degenerates in the classical limit t o a singular distribution, with
complete concentration of the probability mass on the classical path in phase-space
issuing from p,, qo; K may then be expressed as a product of delta functions
I n the theory of discrete Markoff processes (where the random variables have only
a discrete and finite set of possible values) characteristic roots of modulus 1 for the
transformation matrix correspond to deterministicprocesses (non-degenerateprocesses
involving roots of the form [ e-N-fo)l < 1).Yet we saw abovethat the quantum mechanical
process is not deterministic in the classical sense. The explanation of this discrepancy
must await the further study of unitary-Markoff processes of this type.
A possible field of application for the statistical approach t o quantum mechanics lies
in the kinetic theories of matter, where the joint distributions of coordinates.and
momenta are required. As a first step in this direction, we shall study the equilibrium
distributions in large assemblies of similar systems.
The notion of Gibbss ensemble is translated into the quantum theory of statistical
assemblies by introducing mixed states, where the assembly has a probability P,
t o be in a state y?n and the average of any dynamical variable G is given by the diagonal
sum
= c ($It, G$,)P,
(12.1)
.n
(12.2)
181
113
x
n
= JL-~N
e+(wozuanm,auu[7kt(qu)
#,(Pu) e ~ z u ~ u ~ ~ n ~ ,
(12.4)
(12.6)
It has been held that the existence of Gibbs's ensembles 'is rather surprising in view
of the fact that phase-space has no meaning in quantum mechanics' (Dirac(13)).
This apparent paradox is removed by the statistical approach to quantum theory,
which leads, as seen above, t o an interpretation of ensembles closely analogous t o that
of classical statistical mechanics.
13. PEASE-SPACB DISTRIBUTIONS OF ONE MBMBER OF A STATI9TIU& ASSEMBLY
We consider now an assembly of similar particles in weak interaction. For a given
energy
- - En of the whole assembly, we find complexions anwith a, particles of energy
k
lC
B.E. case:
A,, = ( W - t 2
pr%(ql) Ul(ClZ)
P
F.D. case:
* . - %7&)I,
(N!)-t2 k P[u,(q,)
ul(q,) ... zlk(qN)],
P
(13.1)
B.E. case:
F.D. case:
C~ =
(01
(13.2)
when all ad = 0 or 1,
ifanyui>l.
8
182
J. E.MOYAL
114
The phase-space distribution p(pu,qu) and eigenfunctionspdn( p g ,qr) for the assembly
are obtained by substituting from (13.1) in (12.3), (124). It is easily seen that in the
M.B. case pa, is a product of diagonal eigenfunotionsfii(p,q ) of the individual particles
only, while in the B.E. and F.D. cases, non-diagonal eigenfutlctions occur too.
The phase-space distribution for one particle is obtained by integrating over the
coordinates and momenta of the remaining particles
f(pJ.2ql) =
**
2(N-1)
/ d p u , 4~7)drpZ&?Zd~3d!?3
...@ I V ~ ~
(13.3)
becauseJJf&pdq
eigenfunctions
f (Pl,Q1) = z %h(Pl,
Sl)*
(13.4)
.1
It is easily shown that the nc are simply the average frequencies of the occupation
numbers a, of (13.1). Introducing a canonical ensemble, where the Pn of (12.3) are
proportional to e-Ed6T, we obtain
(13.6)
By substituting from (13.2) for the C, .the n$ can be calculated by the method of
sums-over-states (Schrodinger(u)),
leading to the well-known expressions
1
nc = (l/f
es4kT-y
(13.6)
(13.7)
M.B. case: y = 0; B.E. case: y = 1; F.D. case: y = - 1,
which can be substituted in (13.4) to give an explicit expression for the phase-space
distribution of one member of an assembly, As usual in equilibrium theory, all results
are independent of the type of ensemble provided that the dispersion of the total
energy is sufficiently small.
The effect of degeneracy of the individual energy eigenstates i8 to introduce nondiagonal terms in (13.4). As a result, the lzi must be multiplied by the corresponding
order of degeneracy wi,while the fic must each be replaced by
(13.8)
whers the indices k, 1 refer to the degenerate phase-space eigenfunctions at the ith
level, supposed orthogonal.
The foregoing may be used to justify the introduction of ensembles in quantum
theory. If we think of an ensemble as an assembly of similar assemblies, then the distribution of one assembly will have the diagonal expansion (12.3) for the same reason
that the distribution of one particle in an assembly has the diagonal expansion (13.4),
even if the ensemble is in a pure state. If the ensemble consists of an infinite number of
distinguishable assemblies, then the coefficients P, of the expansion must be M.B.
factors e-E.T(-Q being now the energy of one whole assembly) rtnd we thus have a
canonical ensemble.
P 15
"*
2(N-2)
/P(PcJ q g ) h
d q 3 *.'
(14.1)
*NdqN*
In the M.B. case, the integration of each eigenfmction pa, yieIds onIy products of
diagonal eigenfunctions of the form ,fi{(pl,Q )fiFe(p2,
q2). In the other two cases, it is
seen that if i p k, there will be in addition non-diagonalterms (obtained by permuting
the two particles) fik(pl,ql)fw(p2,q2), preceded by a + sign in the B.E. case, a - sign
in the F.D.case. Other non-diagonal terms in pm cancel by integration as in the case
of a single particle. Hence we can write for all three cases
f i k ( p 1 , 4 1 ) fk.l($b
qa),
(l4.2)
where y has the same meaning as in (13.7). The coefficients of this expansion are easily
found t o be for a canonical ensemble
(14.3)
Carrying out the summations in (14.3) by the ' sum-over-states' method, we find that
the non-diagonal coefficients (i Ic) are
nik
(14.4)
= n(nk,
where the niare the average frequencies of the a$,as given in (13.6),while the diagonal
coefficients are
(14.6)
M.B. case: mi$= n!, B.E. case: niC= 2 4 F.D. case: nii = 0.t
The last (F.D.
case) is of course a result of the exclusion principle. Substituting in
(144)we have
(14*6)
f k i r 41,Pz,4 2 ) = 2: %nkfii(%k41)f k k ( p Z , & ) +Y n i n l c f i k ( p l , nl) f k k ( p 2 ,
4, 16
4, k
(14.7)
t Strictly speaking, the right-hand sides of (14.4) and (14.6) should be multiplied by a
normalizing factor (1+yx.n:).
i
8-2
184
J. E.MOYAL
116
6,Ic
where
Qnk,
(14.8)
It is this dependence which gives rise to the exchange energy between the particles
when they interact.
15. LIMITATIONS
OF THE STATISTIUAL APPROAOH
TO QUANTUM THEORY
The results obtained so far seem t o offer a, fairly complete scheme for treating quantum
mechanics as a form of statistical dynamics. It is important now to return to the
difficulties mentioned a t the beginning of this paper, and discuss the limitations of
this approach.
First, we notice that phase-space eigenfunctions must generally take negative as
well as positive values, since they are orthogonal. Only one eigenfimction (generafly
the ground state one) may possibly be non-negative for all values of the dynamical
variables, except for singular eigenfunctions involving delta functions, such as the
momenta eigenfunctions (4.13). Hence, on taking for example Cartesian coordinates
and momentap, q as the basic system, the phase-space distribution in the nth energy
eigenstate formed according to the method of 6 3 would be the diagonal eigedunction,
fnn(p,q), which can be negative, and is therefore not a true probability. This is not
really surprising, because we have seen in 9 that the dynamical equations are those of
a, Markoff process. The existence of eigenfunction solutions for the fundamental equstions (9-8), (9.9) of Markoff processes is well known (see Hostinsky(rl)),and it is also
hiown, that these eigenfunctions are not generally probabilities by themselves. Probability distributions are expressed as non-negative linear combinations of these
eigenfunctions.
In the language of quantum theory, we may say that true probabilitg distributiolzs
of any given set of non-commuting variables do not exist for every state; the physical interpretation would be that where the distribution, as calculated by the method of 3, can
take negative values, it is not an observable quantity, This is a restatement of the
necessity, already discussed in 5 2, for postulating tho existence of Werent phase-space
distributions according to the basic set of dynamical variables. Take, for example,
a system composed of one proton and one electron. The distribution P(p,q)corresponding to the @(q)of a Gaussian wave-packet is positive for allp and q, and is hence
an observable quantity. On the other hand, there would be no observable ( p , q )
I17
aH
-=
ae
0,
aH - constant = w .
-
ag
The transformed equationof the motion (7.8) can be written
-+o-+-sinaF
at
aF
ae %
Y-a -a 1
EIP
2 ae2aQs
= 2ip
= 0.
( p constant),
= &@+0/4~
(15.2)
(18.3)
(15.4)
Hence, if P>O for all 0 a t t = 0, it must be non-negative for all t. This proof was
suggested t o the author by Prof. M. S. Bartlett.
Binally, we may discuss the meaning in the present theory of observables having no
classical analogue. gS2-5 on quantum kinematics are framed SO as to apply t o such
observables as well as to those having a classical analogue. The phase-space distributions represent for both types the joint distributions of eigenvalues for non-oommuting
Bets, and are subject t o the same restrictions. The quantum equations of motion in
phase-space, on the other hand, were expressed only for Cartesian coordinates and
momenta, so as t o bring out the relationship with the theory of general stoahastic
186
118
J. E. MOYAL
processes. It is clear, however, that they can be extended to general quantum observables, say r and s. If P(r,6 , t ) is their joint distribution, then as in $ 7 , aP/at is obtained
(16.6)
16. PRAUlIUfi
The foregoing restrictions are necessary a0 long as we require probabilities in phasespace. They may be relaxed in practical applications of the theory, where we introduce
phase-space distributions as aids to calculation, and where the observable quantities
we wish t o calculate are necessarily non-negative, independently of whether the phasespace distribution takes negative values or not. It is not difEcult to see that the phasespace distributions and eigenfunctions obtained by the rules of $$ 3 and 4, though not
necessarily non-negative, obey the other fundamental rules of probability theory,
i.e. the addition and multiplication laws. Bartlett (15) has discussed the introduction
of such negative probabilities as aids to calculation, and has shown that they can
be manipulated according to the rules of the calculus of probabilities (with suitable
precautions) provided we combine them in the end to give true (non-negative)probabilities. He remarks that where negative probabilities have appeared spontaneously
in quantum theory, it is due t o the mathematical segregation of systems or states
which physically only exist in combination .
Now this relaxation will be possible in practical applications, because the phasespace distributions oontsin more information than is generally required for comparisonwith observations. For example, ifwe wish t o calculate the way the distribution
in space p(q;b) of a wave-packet varies with time, we may use the method of $10,
The statistical approach to quantum theory involves the introduction of an additional postulate on the form of the phase-space distribution, which is equivalent to
a theory of functions of non-commuting observables. The choice of this postulate is
not unique. Dirac (16) has given a, theory of functions of non-commuting observables
which differs from the one obtained in 5 of this paper; it has the advantage of being
187
Quuntum mechanics
a statistical theory
119
bdependent of the basic set of variables, but, as might be expected from the foregoing
discussion, it leads to complex quantities for the phase-space distributions which can
never be interpreted as probabiliti6s.t
It is natural to ask therefore whether any experimental evidence is obtainable on
this subject. In so far as observable results calculated by such theories are equivalent
to those obtained by orthodox methods, e.g. transition probabilities, or distributions
of coordinates only, this is obviously impossible. However, though the simultaneous
measurement of coordinates and momenta is not possible for single particles, there is
Borne hope that experiments on large number of particles might be devised to verify
the phase-space distributions predicted by the theory. Alternatively, one might
hope t o verify the corresponding theory of functions of non-commuting observables
if experimental evidence became available on 8ome Hamiltonian involving products
of and P.
GGS
APPENDICES
Appendix 1, #pace-conditionul averages of the momenta and the uncertainty relations
The space-conditional moments
are the means of pn when 4 is given, They may
be obtained either from expression (4.14)for P(p,q)
F$
(A 1.1)
*w
M(7 1 4r) =
4) eZ7g%J = @*(a$(q+ *W/**(q)
@(d*
On writing
$(q) = p"q) eis(Q)lfi
the logarithm of M(7 I 9) or 'cumulant' function (KendaU(18))
(A 1.2)
(A 1-3)
(A 1.5)
t Note added in proof. Reference should ako be made to a recent paper by Feynman (26)
giving an alternative approach.
$ The double bar denotes EL conditional moment, while t b single bar - denotes a mean over
the di8tribution of bokh p and q.
188
J. E. MOYAL
120
as
a)
='
(A 1.0)
leading to the interpretation of the argument of the wave-function @(q) as the potential
#(q) of the apace-conditional mean F(q). The conditional mean-square deviation ist
(A 1.7)
We note also that the asymmetry of a distribution depends only on its odd c u d a n t s ;
hence the asymmetry of the conditional distribution of p depends entirely on S(q).
Pormulae (A 1.6) and (A 1-7)lead directly to Heisenberg's inequality for the meansquare deviations of p and q. Let a , ,8 be any two random variables with zero means.
We have the well-known Schwarz inequality
--
(a2/q
= rauflB
PI.
(A 1.8)
Now take a = $(q), where we suppose to become random when we allow q to vary;
take also p = q. Then from (A 1.8) above, and assuming (as can be done without loss
of generality) that B = = 0, we obtain
(A 1.9)
~ , e s 6 1 q P & 7 >ag2.
(A 1-10)
(A 1.11)
Since
the sum of the two inequalities (A 1.9) and (A 1.10) gives Heisenberg's inequality
r;.;
z (p4)2+
agz.
(A 1.12)
This derivation of Heisenberg's inequality was pointed out to the author by Prof.
M. 8. Bartlett.
t The fact that 41, can be negative according to (A 1.7) results from the possibfity of th0
formal expression for P ( p , 9) being negative in certain states. The restrictions thus imposed on
the interpretation of P ( p , q ) as a probability me discussed in 16.
121
@7') W$((I
uk(y)Uzd(y)
axdv = WS~S,,
f $87') e--2(7-7')gdTdT'dfldq
(A 2.1)
A 2.2)
(A 2.3)
(A 2.4)
Appendix 3. Operators corresponding to functions of canonically conjugate variables
The proof of (6.15)follows from expression (3.10) for the phase-space distribution
(A 3.1)
(A 3-2)
190
J. E. MOYAL
122
4) = I;@ n k ) P n
n
(A 3-31
(A 3.4)
(A 3.5)
This could also be derived from (A 3.2) (cf. McCoy (lo)).
Appendix 4. Transport equations and the Xchrodinyer equation
The 'transport' equation of any quantity g(p,q,t) is defined as the equation
governing the time variation of the mean z(q,t) at every point q (space-conditional
mean). It is obtaineclfrom (7.7) or (7.8) by integrating over themomentap and making
use of the expressions in Appendix 1 for the conditional moments of p . In the case of
a particle of mass nz, charge e in an electromagneticfield, whose classical Hamiltonianis
(A 4.1)
(A&, t ) being the vector, V(q,, t ) the scalar, potentials) integration of (7.8) and substitution of = aS/aq, from (A 1-6)lead to the continuity equation
(A 4.2)
where p(qt) is the distribution function of the coordinates. Multiplying (7.8) by P k
and integrating gives the transport equation for
(A 4.3)
Substituting in the above from (A 1.6) and (A 1.7), and combining with (A 4-2), we h d
(A 4-41
Hence the quantum-mechanical equivalent of the classical Hamilton-Jacobi equation
(A 4.5)
Substituting p = $@* and S = &/2ilog(+/$*) mcl adding and subtracting (A 4.2) and
(A 4.5) we fmd the Schrodinger equation of a charged particle in the field
(A 4.6)
191
Quantum mechcmics
us u stutistica2 theory
I23
(A 5.1)
Changing to the variables
6 = ar + bs,
A=-+-
2a
2b'
7 8
111 = --2a 2b'
(A 5.2)
5.2)
(A 5.3)
I should like t o acknowledge my indebtedness to Profs. P. A. M. Dirac, H. Jeffreys
and the late R. H. Fowler for their criticisms, suggestions and encouragement in
carrying out this work, and my gratitude to Prof. M. S. Bartlett for many invaluable
discussions and the communication of his various results referred to in the text.
M. J. Bass and Dr H. J. Groenewold have studied the same subject independently
(cf. Bass (19)(201, Groenewold (21)),and I have benefited from discussions and correspondence with them. The papers of Powell (22),Stueckelberg(23).Dedebant (24) and
Reichenbach's book (25) also have a bearing on the questions disoussed in the present
paper (I am indebted to Prof. Bartlett for these last references).
SUMMARY
192
J. E. M o v a
124
REFEREN CE S
1946).
(26)
FEY",
DEPARTMENT
OF MATHEMATICAL
PHYSICS
T m QUEEN'SUNIVERSITY,
BELFAST
INTRODUUTION
The calculation by the usual perturbation methodg of transition probabilities between
1.
~ ( pq ),= %J+:k(q
(1.1)
(1.2)
we have an expansion for P ( p ,4; t )
(1.3)
in terms of the energy phase-space eigenfunctions
(1.4)
These functions form a complete orthogonal system in phase-space which is also 'selforthogonal' and hermitian with respect to the indices 7c, n, i.e.
ss
f k , J ~ ~ m dq
~ d ~ ~ - l ~ k ~ Jlfkndpdp
~ ~ ~ n ~=
~ &, k w
f7m = f ; k s
(1.5)
/k(p'
q)fkn(p,9)
(1.6)
(1.7)
-f When no limits are specified, all integrals are t o be taken
from
- co t o + m.
194
546
M. S. BARTLETT
AND J. E. MOYAL
where alapH, a/aqH operate only on the classical Hamiltonian H ( p ,q ) of the system;
a/8pP,a/aq, on P ( p ,q, t). This is seen t o be an extension of Liouvilles theorem
(1.8)_
and reduces to the latter in the correspondence principle limit (h+O) and for systems
whose Hamiltonian is a polynomial of the 2nd degree or less inp andq. For suchsysteme
(and they include the free and the uniformly accelerated particle, and the oscillator)
the transformation with time of P ( p ,q; t ) is of the deterministic type of classical
kinetic theory, each element of the distribution transforming in phase-space according
t o the laws of classical mechanics. A direct veriiication of the deterministic character
of these systems in given in Appendix 1 and in 5 2.t
Equation (1.7) specifies the infinitesimal transformation with time of P ( p ,q; t).
The transformation over a finite interval t - tocan be given in terms of a transformation functionK(p,ql po,qo;t--to);
(1.9)
K is interpreted as the probability of p , q a t t conditional in p o , qo at to, and may be
expressed either in terms of the phase-space eigenfuimtions by an expansion similar
t o (1.3)
(1.10)
fic7~(220, lo) fzn(l., 4)ei(Ek-EdcL-o)fi,
(1.10)
q I Po, 4 0 ; t - t o ) =
k, n
u,(qo) tP((q)
n
e-~~~(~-Lo)/~~,
(1.11)
(1.11)
(1.12)
The phase-space theory of quantum mechanics may be applied to calculate the
transition probabilities of a perturbed system. If K is luzown for the perturbed Hamiltoiuan H , and we wish to calculate the transition probabilities from the lcth unpertu.bed
state in an interval t, we take as initial distribution the lcth diagonal phase-space eigenFo(po,qo) = fL?(l)o, go).
function corresponding to the unperturbed Hamiltonian So):
The transformed distribution after an interval t is then from (1.9) and (1.3)
(1.13)
The transition probability from state lc to state n is then given exactly by$
(1.14)
547
EARMONIC OSCILLATOR
(e-Q') = (2nnfn!)-te-~~'H,(q),
(2.2)
(2.3)
a a
(@)
(z)
IC
fIc&,
z * ) = ( - ~ / * ) ? L + / c (nzn!ii!)-+ezz*
?L,
(e-?.
(2.4)
(&Y
(W?Le-W>
dw (2n)-l ei(7c-W do
= +( - 1)"(k!/n.!)a,-~(r-n)e-tw~~-k(W)dw (27r)-lei(k-?@d0.
(2.5)
- 1)"
(%!)-I
e*w
(aY
-
(W?&
e-w)
dw d8/2n
= &(- 1 ) n e - h L , ( W ) ~ W ~ B / ~ T .
An expression for these eigenfunctionshas been found independently by Dr EL
wold (2)
I
(2.6)
I. Qroene-
196
M. S. BARTLETT
AND J. E. MOYAL
548
[(q2-+qi)coswt - 24q0]}.
(2.7)
-w,Y I P O ?
40;
t)
J exp{i(sinwt)-l[~(q~~s~~t-pssinot-q,)
= (it+sinwt)--1J
~ d ~ ,
= sinoJt8(qcoswt-psino~t-q,,) 8(q,~oswt+13~sino~t-q),
(2.8)
a product of delta functions expressingtlie contact transformation of classicalmechanics
eigenfunctiom
for the oscillator. The expansion (1.10) of K in terms of the l~hasc-spaoe
gives the following interesting formula for tlie associated Laguerre functions
(32n)-1 r, (k! I % ! ) W;f(k--r~) e-al/)uL;;-lc(w w--8(1c--1L)
e-B~t~L?~-l~
$(7+ic)(o-uu-d)
0)
1c
15, n
=~
- w0)&(a-0,- ~ +t 2 ~ 7 ~ )(2.9)
)
( L U
where the 8-functions are norinalized over the ranges of zu and 0 , r being an integer
such that 0 < 8, -+ wt - 2 m < 27r.
In Appendix 3, the above results are applied to derive the equilibrium phase-space
distribution of members of a statistical assembly of oscillators.
3. TRANSITION
PROBABILITIES OF
A PERTURBED OSCILLATOR
We shall now apply the method outlined in $ 1to calculate the exact transition probabilities of a perturbed oscillator for a perturbing potential of the forin V = p%)~
where &(t) is an arbitrary function of the time. On using equation (1.14), the transition
probability from state k to state n is given in terms of the variables w = 2(p2 + q2) and
8 = t a r 1 ( p / q )used in $ 2 by
4)
pkTL(t)= 87i/
2n
~ f k k ( ~eo)
0 JL??'~,(~J
)'
dw,d60J
(3.1)
where wo,8, are the initial values a t t = 0, w and 0 those at time t obtained from the
classical solution for the perturbed oscillator
q = qocoswt+posinwt-w
Hence
where
8(7)sinw(t-7)d7,
(3.9)
197
649
It is easily seen that e = &?hai s the non-fluotuating part of the work done by the
perturbing force, while q5 is the phase change of the oscillator. We now substitute these
solutions in (3.1) and also introduce two auxiliary variables y , t which we shall equate
to 1after all the indicated integrations and differentiations have been carried out. This
yields a closed expression for the transition probability
where Io(s)is the modified zero-order Bessel function of the first kind (I&) = Jo(is)).
The integration is easily carried out, for example, by substituting for I&) i b Taylor
expansion and integrating term by term, whence
(3.3)
Carrying out the indicated differentiation on y, [ n ) one finds that
(3.4)
With the use of the e-qression for the Laguerre polynomials generating function (3)
(3.5)
probability generating function (p.g.f.) may be calculated for the p
G(r,0) = C 2 1 , OrcrlL
~~
--
(3.8)
1- r O
The coefficient of 816 in the Taylor expansion of G ( r ,0) in powers of 0 only, will be the
p.g.f. CrlC(r)
for transitions froin the lcth state; that of B I c P in the expansion in powers
of both 0 and T will be pkn(t). Eor n. 2 k
(12-k)!(ri+v)!(-2)~
). (3-7)
198
550
11.,S. BARTLETT
AND J. E. MOYAL
Since, as is obvious from (3-3)or ( 3 4 ) , the pkn satisfy the principle of detailed balance,
i.e. pkn = Pnk, the probability of transitions t o states n < Ic is simply obtained by interchanging the indices n and k: in the right-hand side of (3-7).
The p.g.f. Cr0(7)and the probabilities pOn(t)for transitions from the gromcl state are
simply those of a Poisson distribution
(3.8)
These exact results may now be compared with the approximate ones obtained by
the standard perturbation method. The Taylor expansion of the approximate expression for the p.g.f.
QL(T,l9)
--
(1-7)(1-0)
(3.9)
(3.10)
correct to the first power in E , whose expression is identical with the first approximation
o j Me perturbation melhod. Expression (3.7) shows that in general the probability of
a 2nth-pole transition for small E is of the order of (e/fiu)L,i.e. only dipole trailsitions
have an appreciable probability. The perturbation method equates t o 1the exponentid
factor exp { - E / ~ J } in the exact expressioii (3.7). This procedure is justified only for
small B ; as the perturbation eneryzJ increases, however, multipole transitions become progressively moreprobable. In order to find the most probable oiies from the ground state,
let us substitute the continuous variable x for n in (3.8)
The way in which the phase-space method may be used to solve wave-paoket problem
is easily exemplifiedin the deterministic cases of the free particle and the particle under
199
551
force. I n both these cases, the Schrodhger equation is most easily solved in
Domenturn space. We describe the particle at t = 0 by a Gaussian wave-packet
ww
(A. 1.1)
correspondingto the minimum uncertainty uo,so
= 4% on its position and momentum
(of. Kennard (5)), so and uobeing respectively the mean square deviations of p o and qo
&Ddtalring the origin at the mean of p o and qo. The corresponding phase-apace distribution is by (1.1)
$ o ( ~ o )=
~ O ( P 0 , Y O=
)
~ X [P-P~/*S;I
,1 ~ ~ ~ ( ~ o - ~ ~ ~ ) ~
b-fi1-lexp - B(Pt/44-48/4)1.
(A. 1.2)
The solution is then obtained simply by substituting in (A. 1-2)the classical solutions
for p and p in terms of po, qo, t. In the case of the free particle, this gives for the distribution at t
R P ,4; f) = ( T WexP [ - 8(P1so)2- & @ / g o -P~1mf10o)21.
(A. 1.3)
It is easily shown that this corresponds by (1.1)to the wave-function
(A. 1-4)
$ ( p ; t ) = (2nst)-fexp [ - (;u/2~,)~+ip~t/2mKj,
which is the solution of the Schrodinger equation
=
(A. 1-6)
with 4 0 ( p o )as initial. wave-function a t t = 0 (cf. Darwin (6)).
Similarly, for the particle under constant force my we find for the same initial dis-
(A. 1.6)
which again corresponcls t o the solution
(A. 1.7)
of the Schrodinger equation
p2
-$+-2na
mg?i,a$
$84
= --z 211
i at
(A. 1.8)
(A. 2.1)
200
562
M, S. BARTLETT
AND J. E. MOYAL
Substituting from expressions (1.4) for fss(p,q) in terms of (8 I q ) and (1-12) for .K in
terms of (ql I q2), we have
= q1- *fir1,
etc., has been
made. Heiice
(A. 2.2)
P S L S 2 = I (81 I 6 2 ) 15
i.e. the two expressions for the transition probabilities are formally equivalent.
(3) YJime-spacedistribution of a memboy of u statistical assembly of oscillators
The equilibrium distribution of the coordinates and momenta of one member of a
statistical assembly of similar particles is expressed in I, p. 114, as a sum of cliagond
phase-space energy eigenfunctions relative t o the inciividud particles
f(13,q) = c c nlLf,clc(lJ,q ) ,
(A. 3.1)
16
where C is a normalizing constant and the u,, = e-c7Jkz' in the case of a Maxwell-Boltzmann assembly. On substituting from (2.6) for an assenibly of oscillators, this becomes
f(w, 8) = (47r)-1(1 - e--JLw/'c'l' ) c ( - 1)ne-!lr-'lRo/kT~ 1E.(w)
72
(A. 3.2)
where the last line follows from (3.5). Transforming back from w and 0 to p and 4, we
finally find EL Gaussian distribution for the coordinates and momenta of a member of
an assembly of oscillstors
p and q are chosen as the basic system, then tho energy distribution becomes continuous,
with a m.s. deviation of (+%J)~ for each of the energy eigeifunctions, leading thus to
the first value cg = E2.
20 1
T v a i d i o i a pobubilities of quaiztum-mecha?ziculoscil1aiol.s
553
SUMMARY
The phase-space method in quantum theory is used to derive exact expressions for
the transition probabilities of a perturbed oscillator. Comparisonwith the approximate
results obtained by perturbation methods shows that the latter must be multiplied
by an exponential factor exp ( - e/%w),where E is the non-fluctuating part of the work
done by the perturbing forces; mi long 8s 6 is small, exp (- E/&w) 1 and only dipole
transitions have a11 appreciable probability. As the perturbation energy increases,
however, this is no longer true, and multipole transitions become progressively more
probable, the most probable ones being those for which the change in energy is approximately equal to the work done by the perturbing forces.
N
REFXRENCES
( I ) Mouffi,
(2)
(3) Sz~a6,G. Orthogonal polynomials (New York, 1939), pp. 97, 102.
(4) JAHNXRP,
E. and EMDE,
F. Tables offunctions (3rd ed.) (Leipzig, 1938), p. 19.
( 5 ) I ~ N N AE.
R H.
D , 8. Phps. 4 4 (1927), 326.
( 6 ) DARWIN,
S. (2. PTOC.
ROY.XOC.A, 117 (1928), 258-93.
(7)COULSON,C. A. and RUSHBROOKB,
G. S. Proc. Cambridge Phil. 800.42 (1946),286-91.
DEPARTMENTOF M~THEMATICS
UNIVERSITY
OF MANOHESTER
202
311
Progress of Theoretical Physics, Vol. 11, Nos. 4-5,
April-May
1954
*) Extending the meaning of the word picture t o include any way of looking at the fundamental
laws which makes their self-consistency obvious , according to Dirac?).
**> In this paper, however, we shall not try any such suggestion, confined merely in the reformulation
203
312
T. Takabayasi
languages in still another way : That is, we can transform the Schrodinger equation into
Liouville-like equation for a distribution function in phase space which is produced as a
certain fourier transform of a bilinear form of the wave function, leading to the picture
of certain Markoff-like process of an ensemble in phase space for quantum-mechanical motion.
This method, which we will refer to as the method of the p h s c S ~ O C L c7tscmbh (abThe purpose of
breviated as ps. en.), was initiated by Wignerfi and later by Moyal.
the present paper is to develop this method into a consistent formulation of quantum
mechanics by establishing the srLbsidiafy conditions for a ps. en. to represent a ~ Z W P state,
and also to prove thereby the equivalence of this formulation with the cs. en. formulation
formerly mentioned.
I n the phase space formulation the knowledge involved in the phase of the original
wave function is reflected upon the momentum distribution in such a manner that the
phase space distribution function (abbreviated as ps. df.) implies the representation of a
state symmetrical in coordinates and momenta. But the manifold of the ps. df. covers
wider possibilities than that of the original wave functions. Now, according to our prescripwith positive coefficients of relevant
tion, a mixirig of states corresponds to a sz~~,~7~positioiz
distribution functions which as well satisfies the same Liouville-like equation, because the
latter is linear in the df. Accordingly a ps. df. in general would be the representative
of a mixed state, in so far as it satisfies certain positivity condition . Thus in this
formulation of quantum mechanics it is an essential problem to obtain the subsidiary
conditions* that a ps. df. should particularly correspond to a pure state. W e explicitly
obtain these conditions, which must be of some non-linear relations ( 4). This is made
tractable by first replacing the usual pure state condition for the density matrix (4.2) by
Zocal n~lntions (4.4). Transforming the latter we acquire the pure state conditions on
the ps. df., which consist of the condition of irrotationality of mean momentum field, (4.19),
and the condition (4.6) which we call the quantum condition . T h e latter will further
be transformed into a series of relations between distribution moments in respect to momentum
components of successively higher orders.
Now, in virtue of these pure state conditions, we can prove the equivalence and correspondence between the cs. en. formulation and the ps. en. formulation (S 5 ) . For
instance, the quantum potential in the former may be looked upon as an apparent force
appearing as a result of projecting the ps. en. onto the configuration space.
The CS. en. formulation and ps. en. formulation, though they are equivalent and
transmutable to each other, are of very different characters. The effect of quantum fluctuations is represented with fluctuations of continuous trajectories5 due to quantum potential
in the former, while in the latter with Markoff-like transitions, the properties of which
we shall examine in detail ( 3 ) . The ps. en. seems to be one step superior to the
cs. en. in that it correctly yields quantum-mechanical expectation values as the mean values
over the ensemble for wider class of dynamical quantities, yet it must be emphasized that
*) Moyal) unnoticed the presence of these conditions, while other authors have been unable to obtain
their expression.
204
it cannot do so for aZL hermitian quantities and that we must in return allow of ~ccgntive
2).
The ps. en. formulation is formally consistent within its limited range of applicability
and accompanied with the picture working along classical lines, but we cannot take the
picture too realistically, just as in the case of the cs. en. formulation4. Instead, these
formulations provide coiicrrtc nm@scs us to thc drgm in which thc stntisticnZ pivprutics
pobnbiZtties
(s
qunntum mi~chnriic~
mu bc uirdcrstood along- n i y stntisticnl scheme bnsrd o n some
hid&n vni-inbks.
The ps. df. is a 7 ~ a Zquantity produced as a bidiircnr- form of the wave function,
as is needed for the representation in terms of it to have classical pictures, but that would
just mean greater complications in mathematical treatments usually. It is well known that
We
the method of ps. df. is useful for the treatment of quantum statistical
shall, however, show how this method of ps. en. can effectively be applied to pure state
problems for a few elementary examples ( 6).
I n the last section we consider the positivity condition and also express the pure state
condition in an alternative form.
c,
p(x, x)={)*(XI,X),
(2.1)
normalizable :
s p p= (rn,p(s,
(2 2 )
s p (f
X)dX=l,
432 0,
x) satisfying
A.
-
(2 * 3)
**
with
where
#,,(x)and
[I
*) The region of applicability of our formulation is rather limited. When a vector potential is acting,
the Markoff-like picture to be stated in $ 3 requires certain extension.
**I In this paper we indicate an abstract operator by attaching an underline.
205
T. Takabayasi
341
@>,u
=Sp(_P
4).
(2.5)
f(x,p ) by
the transfor-
mation,
(2.6)"
and will use the latter for the specification of the mixed state.
written in the rrcijvocnb form :
(2.7)*
where p(p, p') is the momentum representation of the density matrix. Eqs. ( 2 . 6 ) and
( 2 . 7 ) show that the function f(x,p ) is the fourier transform of ,o(x, x') or p ( p , p')
along its antidiagonal ; in other words, if we consider the density matrix to be a function
of ' mean coordinate ' and ' relative coordinate ', as
?(XI
v) = p ( x - ! / / 2 ,
x+v/2),
( f ( x ,p ) d x d p = 1 .
(2.8)
(2.9)
W e may thus imagine, corresponding to a mixed state, an ' ensemble ' with the ' probability
distribution ' in phsr spncr, f ( x , 21) , though we must then allow of negative probabilities. The function f(x,p ) , which we call the phase space distribution function (ps. df.),
can be regarded as the Y L ~r'smtativcz
I
of statc .gwi?m*iricti//'Ti cooidimtc niicr' mo?nnrrituiir.
From ( 2 . 6 ) or ( 2 . 7 ) we get
( 2 * lo)
(2.11)
*) In this paper integral usually means a definite integral over whole coordinate space or whole momentum space. In such case we shall hereafter omit to note the boundaries =tx.
206
(A),,,,
=Sp(pA(:z,
provided -A (- x_,
23)
22)) ={
. ~ . ~.
(2.12a)
A(%,
- p- ) =Al(:x)
+A&)).
(2.12b)
This means that our phase space ensemble (ps. en.) correctly gives the quantum-mechanical
expectation values as the average values over the ensemble for any quantity of type (2.12b).
However, for a more general function A(%, 1)) involving cross terms of x and p
there exist ambiguities in the definition of the corresponding quantum-mechanical function
A
x., p- ) of non-commutable operators x and p , so that in such a case our ps. df. may
- (-
A\(%,
- p)
function A ( x ,$3).
(x ,p
) corresponding to the same c-number
ties A
~-
p
.- ) =
(Q,
(2.13)
Lir,
r ) ei(StT19
no dr.
(2.14)
(A(x,
- - -p ) ) q , , =( A(%, p > f ( x P)
, dx
49,
(2.15)
dQ dr.
( 2 . 1 6 j7)
Thus it may readily be found that the ps. df. does not give mean values correctly for
quantities such as, e.g., the commutator [ -f i r , x,],
the square of the energy N (except fot
the case of free particle), or the magnitude* of angular momentum Z.
I n other words
our ps. df. does not reflect quantum-mechanical probability distributions correctly for
H or -ld (a component of angular momentum), in contrast to the case
quantities such as of
_A,(?)
or
A_@)
(2.17)
V.J
in terms of
(2.18)
g(x,p ) = (pI{@)(xlp).
Here the latter distribution function g ( ~p,) , though not real, gives the mean values
correctly for the well-ordered functions of the form, A(%,
p.
_ _ 1))
_ = Cn,&,,,
n,
*)
For the components of angular momentum, the ps. df. yields expectation values correctly.
207
T. Takabayasi
346
p
-
Taking a function of z
- alone, or of
->-
Z'(X)~O
A,(%) or A , ( p ) ,
SO
Q(p)ZO.
and
(2.19)
s P ( X >d32 = s a (P)4!J=1,
we can construct many ps. dfs. which are compatible with those given
Indeed the ps. df.
fo
( G P ) =J'@)
Q @)
(2 * 2 0 )
+A (GP)
(2.22)
SA(x9 P ) d P = I f , ( % , P)dx=O,
(2-23)
P>=l'(x)
in which
f,is
*) Such propetty cannot genetally be conserved with the passage of time in classical as well as in
quantum ps. en. In the former, however, this property persists in the special case of stationary canonical
ensemble, while in the latter it is not so. This is due to the difference of the ' Liouville ' equations for ps.
dls. in both cases (see 4 3 ) .
208
_
The Formulation of Quantum Mechanics in terms of Ensemble in Phase Space 347
remarked that there can exist many distribution functions in phase space other than/~(03, p) ,
which satisfy (2-8) and ( 2 - 9 ) and give expectation values correctly for any quantity of
type (2-12b). Examples are
/.(5. 2>)=cos(/2-r p P,) f(<K, p),
and
We have considered the df. defined in OC p space, but it may also be possible to
introduce a distribution as a function of another variables and TJ, where or -q represents each a complete set of commuting observables, in such a way that it correctly gives
the quantum-mechanical probability distributions for quantities of type ^4](?) and AV(TJ)',
this distribution*, however, differs from the original one, /(fiC, p) In this sense the ps.
en. has not the meaning invariant to such pairs of unitary transformations11' (ac, p-+, 7j).
3.
(a)
We shall now represent the temporal change of state in terms of ps. en.
tion of motion for (i,
ifi d[>/dt=Hp_pH,
The equa-
(3-1)
is written, in ^-representation, as
(3-2)
since we take hamiltonian, ff=p~/2m+ V(af) , corresponding to a particle in a scalar
potential V(OC). For the intermediary function
(3-3)
(3 2) is written as
which, by fourier transformation, leads to the equation of motion for our df.:
d//dt+p/m-rf=A[f\,
(3-5)
with
4/1= !/(* P-PO/to 2>'> */.
(3 6)
(3 7)
'
*)
/(I,
It is given by
7?) = (^)-
209
T. Takabayasi
348
(3.8)
where
(3-9)
The appearance of the fourier component, V ( 2 p ) , may more easily be understood by starting from the momentum representation of ( 3 . 1 ) ,
(3.10)
where & ~ t ' / 4 and
,
the integral operator
1
'
1
means the summation over positive odd integers. Using this form,
is expressed as an infinite series of differential operators"'):
(3.11)
(3 12)
If
.I.,
I*'(%)
11'""f]=DI.'.P,
f.
(3 13)
*) T o next page : T h e circumstance that in case of potential quadratic (including below quadratic) in
the ps. en. moves purely classically is closely connected with the fact that in this case, speaking with the
language of matrix mechanics, the Hamilton's or Newton's rguntfou of uzotioit for y-number coordinates is l h e o ~ ' ~ ) .
Generally in quantum mechanics formeily the same equation of motion as in classical theory holds for y-number
quantities, but, if this equaticn of motion be linear, there does not appear anyj./.oduci o f y - n u / A - , / x , the rule on
which specifies the essential diserence of quantum mecbanics from classical theory, and theiefore each matrix element
changes like clzssical quantity. I n this case the diserence of quantum mechanics from classical theory can only
prescnt itself in the dehr.ition of the initial c o d c i o n s which is repiesented by the commutation relation in
matrix mechanics, and by the subsidiary conditions for pure state (to be stated in 4) in out formulation.
It is t o be noted that the above case includes a generaiized/bfZ.:,(f 'oscillator ' with the hamiitonian, I;r=a(t)&l-
B(452
210
!](I.,
p)nr=o,
(3 * 14)
serves to ensure
(3.15)
(ii) The transition probability depends upon the amount of jump alone, irrespective of
the value of momentum before or after the jump. (iii) The external field V ( X )acts
so as to induce indeterministic transitions in particle momentum, but the transition probability] itself is perfectly determined by this potential. (iv) As is seen in (3.8) J ( x ,
p ) is of a form of a perfect sinusoidal wave in x-space with no damping in far away
irrespective of the form of V ( X ) . The amplitude of this wave is the fourier component**
V(2p);while its wave-length is Z/2$, so that for larger jumps of momentum the Xspace oscillation of J is more rapid, having therefore less effects when we consider the
Since
space average over the ensemble. (v)
jpJ(x, p)dy,= -v
v,
(3 * 16)
the rate of average change of momentum agrees with the classical value. In each individual
case, however, particle R uctuates, performing jumps in momentum, which may well be far
greater than the mean value, and may thus, for instance, penetrate a potential barrier.
(vi)
The various
transition
moments
of
**) The circumstance that the probablity of momentum transition from 270 to 1, is detetmined by
b 7 ( 2 ( p-20))
in out stochsstic picture is somewhat similar to the usual perturbation-theoretical result of quantum
mechanics, where the probability of transition from the state of momentum p o to that of 2, is proportional to
I~~l~-PcI)~?.
21 1
T. Takabayasi
350
(3.17)
%f,
transition moments of even orders vanish while those of odd orders survive, in contrast to
the usual Brownian processes. In the latter, the transition moments in and above third
orders are assumed to vanish, resulting in the differential equation of Fokker-Plan~k-Kramers~
type for the df.; the diffusion takes place in a manner essentially determined by the second
order moment, the distribution always diffusing monotonously and irreversibly. O n the
other hand in our case the moments of third and higher odd orders give rise to a quite
The non-vanishing of third or higher order moments
different type of diffusion *I*****).
means that in our process the probability that the value of momentum changes by a
finite amount in a small time interval cannot be regarded as small.
W e have thus obtained a stochastic picture of distinctive features for a quantummechanical change of state.
Though this picture cannot be taken as a real one, for
instance, on account of (i), it may be said to represent quantum fluctuations in a picturesque manner.
(b)
dP,
(3 * 18)
(3-19)
212
The Formulation of Quantum Mechanics in terms of Ensemble in Phjase Space 351
This kernel K(xtlxoto)is nothing but the transformation function which transforms the
x,, (particle position at t o ) diagonal to the one making x, (posirepresentation making tion at t) diagonal, i.e.,
(3-20)
(3 * 2 1 )
j K(#t Ixoto)K(
xoto
IXZ) Cdi, =b (x-x)
,
(3.22)
K(xt,J3C,to)=i;(X-xo).
(3 *23)
SO
As K ( x / ~ z also
~ ~ satisfies
~)
the Schrodinger equation, it can be determined by solving
that equation under the initial condition ( 3 . 2 3 ) . For a conservative system K may be
written as
K ( x t INot,) = ( x I E - - - l o ) f i l
x,),
(3.24)
K( xll%,to) =K * (xoto
1xt).
(3 * 25)
Now, corresponding to ( 3 * 1 9 ) , the kernel for the time development of the density
, t) is given by
matrix p ( ~ x,
j K(at Ixoto)
K * (Xt I%;to) dx&lx;,
and accordingly the relation of our transformation function T with
(3.26)
K is found
to be
(3.27)
From the properties of K, (3.21)-(3.25), we can find the corresponding properties of
T. First, T is real, satisfies the iteration law :
T ( ~ p t l ~ o , o t o ) = 5 T ( ~ p d lt~i )i T
p (, x , ~ tilxo~?oto)dxl
i
d~i,
(3.28)
T ( x pt lx,p , to)T(x,p,t
o pp t ) d X 0 dpo= 8 (x-x)
6(P-P),
(3 * 29)
T(xpt,~x@,dO)
= ~ ( x - x o ) ~ ~ P - P* o )
(330)
(3.31)
213
T.
352
Takabayasi
~ ( ~ ~ ~ l ~ o P , ~ o ) = ~ 7 1 ( ~ 0 1 3 o ~ o l ~ ~(3~* 32)
) ,
and normalized :
j T ( x pt \ ~ , , pt0)dx
,
dp= 1.
(3 * 33)
Eq. ( 3 . 2 8 )
(3.34)
1,
TI(%]>
I z o p o ) = -6(p-pa)V.6(z-zo)
//2
+6(~~-z")-~(~,~)-.z)o),
J7zlY(z-x:o)
y ( x ! 2)-2)0)
iS(:X:
--X:o)
1J ( z p-p')
,
J(z,'p/-.p0)dp/.
When the potential t'(x) is, in particular, a polynomial in or below szcond power, 1
' is the solution
of the classical Liouviile equation with the initial condition (3.30), and hence is given by
T ( z pI
z,,)
where (z,,
ticle we have
zoP0'0)
=6(x--2,(zoPof0))6(P--Pt( Z 0 2 ) 0 ~ " ) ) ,
Y'CZ~HI x o 2 , o t 0 ) =6(2)-j,o)6(zx;-
-ieo-2)0/~1.
(3.35)
(t.--!,J),
(3.36)
which is quite difFeferent fiom the kernel for the usual diffusion, while the kernel for probability amplitude is,
as i s well known, of diffusion type with imaginary time coefficient :
(3.37)
But the fouriet transCotmation of (3.36), taking account of the relation (3.27), gives
214
(3.38)
(4.1)
This relation, being non-linear in f, is a necessary condition for the correspondence of f
to a pure state, but not a sufficient one.
The condition that a mixed state should in particular fall into a pure state can be
expressed, in terms of the density matrix, as po==p,
- i.e.,
(4 * 2 )
Transforming this relation according to ( 2 . 6 ) we may immediately obtain the pure state
condition for
but then the result is not of a convenient form (see 7)*; so we will prefer
another way.
Now, if (4 2) is satisfied, there exists a suitable complex function S ( X ) that makes
p(x, x;) written as
(4 * 3)
on account of ( 2 . 1 ) , ( 2 . 2 ) , and ( 2 - 3 ) .
It is because ( 4 . 2 ) with ( 2 . 3 ) requires
ZU,=J,,~(with certain fixed ? i n ) , when p ( ~ x)
,
is put in the form (2.4).
But if
and
p.
215
__
354
T. Takabayasi
dfl 3,>
&,>
d(>I
l -- ' i) -- - = [i
3xt dxk'
dxtdxk'
d .log!
/ a .
--
dxt' dXi
logo 1 = 0 ;
\dxt'
(4-4)
9
'd \-)(
( 1----9 , 19-V }p
I/ I---\2 dx
dJ1 \2 dx
j1
J_ + J_V,
dx* dyk r
(4-5)
which, by the fourier transformation, separates into the following real and imaginary part
equations :
where the notation such as /i*/"2 means the convolution with respect to p, i.e.,
, p") 3(p-p'-p"}dp'Jp".
(4 8)
We have thus obtained the subsidiary conditions for the correspondence to a pure state,
which have following properties :
(i) The condition consists of six relations (4 6) symmetrical in i and k and three real
anti-symmetrical ones (4 7) . Since f is a. scalar, (4 -6) is a tensor equation and (4 7)
a vector equation, both covariant to coordinate transformations.
(ii) They are ' kinematical ' relations independent of the dynamical characteristics of the
system. Planck's constant which appeared in the time development equation (3 5) with
(3-11) also enters the first condition (4-6), both in the form =/52/4. Thus in our
*)
216
The
Formulation of Quantum Mechanics in terms of Eusemble in Phase Spcae 355
formulation of quantum mechanics Plancks constant has to play such ~ z u q h d dr-odcs. We
may call ( 4 . 6 ) the quantum condition** for the ps. en. formulation, and ( 4 . 7 ) the
irrotationality condition (see below).
(iii) T h e conditions are surely not linear in f, but are again integral equations quadratic
satisfying both (4 - 6 ) and ( 4 . 7 ) , the disin f. If we have two distributions f,and f..
tribution obtained by their superposition, (which corresponds to mixing of states), no
longer satisfies these conditions, as it should not.
W e could adopt, in place of ( 4 . 4 ) , the similar relation for p ( p , 2)) :
(4.9)
which, by fourier transformation using (2 . 7 ) , leads to relations :
(4 * 10)
(4.11)
(b) Our next task is to re-express the conditions (4 * 6) and (4 7) in another forms.
For that purpose, first, we integrate ( 4 6) and ( 4 7) throughout over the 23-space, which
procedure we shall call the $rq>ctioiz of the relations onto the coordinate space , and
employ the factorization formula for the convolution :
(f1*f*)@=
SfidP,Sf?dP.
(4.12)
(ai=apxi)
(4.13)
and from ( 4 . 7 )
sip. pk-
a$&
= p(ajpk-
akc<)
(4.14)
&(x) = S P i f ( X l P)dh
e k ( x ) = ~ $ ~ p k f ( x ~
P)dP
(4.15)
(4.16)
are distribution moments with respect to momentum components of first and secohd orders,
**)
It is to be noted that the quantum condition in the old quantum theory determines stationary
(pure) states, while our quantum condition selects pure states out
of mixtures.
217
T. Takabayasi
356
respectively.
pi(,)
=I</P,
p,P&)
Pi
and
P, 9,
=P%k/PI
(4.17)
curl
a.p;, = -E
(p= log P)
a?'$/aX,ax,,
(4.18)
p=O.
(4.19)
Pi 2k,
Eq. (4 a18) means the relation that the dispersion tensor of momentum, pip,at a point
should be connected with the space derivative of the space density at the
~Tcz).
point*, while (4.19) implies the irrotationality of the mean momentum field
Next, we multiply ( 4 . 6 ) by Y j and then again project the result onto the coordinate
space to obtain
f'fj I'k
P
i P j k -P $ f k -Y*
l'ijk
ak
Y+a,
uj.a, P ) ;.
(4.20)
By use of (4 18) and (4 19), this can be transformed into relations symmetrical** in
i, j , and k :
__
(C$i
aj a,)Vi-ai 3.j$ k } ,
cyc
pipjpk-$i*p;j*pk=-E{
(4.21)
which are ten relations for the (symmetrical) moment tensor of the third order:
.__
j-)ijk
= Y ' p i p j p k = ~ p i p j p f ( x ,
p)dp.
(4 22)
a k ai'$V>
(4.23)***
CSC
Kl-(K)c
at a space point is
*) It is to be noted that tke mean deviation of momentum components
not necessarily positive for our ensemble whose df. f ( ~p,) is not necessarily positive, and is equal to
--Pd?$/dx~? for pure states.
is symmetric b y virtue of (4.19).
In the right side of (4.23) the first summation contains 4 terms, the second summation 6 terms,
and the third summation 3 terms.
**)
***)
djd$k
218
by
xdx,--xi.n:,= --E
a9s3/ap, apk?
(a-log
Q)
(4 * 24)
curl,,X
-- = 0,
(4-25)
where
(P)= S X i f dx,
T A P ) =Q i / Q ,
Qi
xi.;(-@)
g,,+( P ) = [ x i x kf dx.
=g l k / Q ,
(4.26)
Finally there is a problem more general than that of the pure state condition, i.e.,
to give a measure to the degree of mixture for an arbitrary distribution
may be achieved by the introduction of entropy C defined by
n=l
({F--P)
(p
--l)-}
_
f ( ~p,
) . It
p(x,x,t ) =Q(x,t)S*(x,t ) .
(4.27)
This determines the wave function sh up to an arbitrary phase depending upon time only ;
that is
+(x,2 )
can be written as
+(a% t ) =$ho(x,
t)Cf*@),
where A ( t ) is an arbitrary function.
Schrodinger equation :
Furthermore,
(4 * 28)
+= 0,
~ [ + ] = ( z / i .a p t - z p n . A + v(x))
because the above
satisfies
(4.29)
+(x,t ) in
(4.27)
219
T. Takabayasi
358
K [ + ( x ) ] + * ( -K*[+*(x')]+(x)
xf)
=o.
Consequently we can uniquely determine the wave function +(x,I ) that corresponds to
n~y
' (4.29).
the given df. from (4.27) and the ' s z ~ j & m ~ i ~ tdoiiditio?z
s', =&'S/ti
(R,S
: real)
(5 * 1)
U ( x )=R (x)2,
(5.2)
p (x)= V S ( x ) .
by
(5.3)
p=o
(5.4)
fL (x,p ) =K (33)
6 (p- V S ( X ) ) .
as,w+1/2]%. (vs)t+v- (
a(R')/a2+div(R%S/m)
and
(5.5)
S,as
~ 7 2 A4 R / R = O ,
(5.6)
(5.7)
=o,
of which the latter gives the equation of continuity for the cs. en. :
aP/at+div(Pu)=o,
=P/m)
(QJ
(5.8)
while the former leads to the equation of motion for a particle of the ensemble :
wz dv/dt=
-V(V+ V ' ) ,
(5.9)
with
V'= - Z2/21n.AR/R.
(5.10)
220
Thus the temporal development of the ensemble can be regarded as is built up through
the process in which each point of the ensemble moves along a continuous path with the
momentum (5.3) at each instant, accelerated not only by the external potential '.I but
I n place of ( 5 . 9 ) we may also
also by the additional ' quantum potential ' (5.10).
adopt the relation of momentum conservation :
(5.11)
where
c$k
',
c<,+
=~ 2 / 4 7 P
~ ~a?
. (log
u)/aA-<ax,.
(5 12)
(or (5.11)),
(A)
with a wave function $ b ( ~ 2), satisfying the Schrodinger equation, by means of (5.2)
and (5.3). Conversely, any ensemble of trajectories that satisfies (A) corresponds to a
quantum-mechanical change of state as follows : Given a solution of (A), P ( x , t ) and
p ( ~t ),, we can determine R and V S by (5.2) and (5.3), and so S ( X ,t ) itself can
also be determined up to an arbitrary function of time, A ( t ) . But this arbitrariness is
excluded by imposing on S the ' supplementary condition ' that S should satisfy (5.6) ,
which condition is clearly compatible with the equation of motion (5.9).
Thus we
uniquely get the wave function 9 =(Zcis!~ satisfying the Schrodinger equation*.
(b) W e shall now explicitly show the correspondence between our ps. en. representation
and the cs. en. representation outlined just now in two steps.
i) I n the first place, given a phase space distribution f ( ~ p, , t ) corresponding to
a change of pure state, the corresponding cs. en. can be produced by 'projecting' the ps. en.
onto the coordinate space. This means that we introduce the cs. en. which consists of the
density and momentum fields, P(x, I ) and @-(x,t ) , derived from the ps. df. f ( ~p,, t ) ,
by (2.10) and (4.17) with (4.15); in other words we eliminate the momentum dispersion
at each space point x in the ps. en., adopting the average momentum and the total
density at each point X. W e can then show that the cs. en. obtained satisfies the condition (A) and represents the same quantum-mechanical change of state.
T o show this, first we project the quantum Liouville equation (3.5)
coordinate space to obtain
a P p +div ( P i / m )= 0,
(5 * 13)**
**)
22 1
T. Takabayasi
360
derived by projecwhich expresses the continuity equation (5.8) for the cs. en. (P,
tion.
Next we project the momentum conservation relation for the ps. en.,
(5.14)
obtainable from (3.5), and get
(5 15) *)
Here we take into account one of the pure state conditions (4.18) for the original phase
space distribution, then the second term in (5.15) may be rewritten as
(5.16)
This means that the contribution from the convection term in the ps. en. picture transforms
to the convection term in the derived cs. en. picture plus the extra momentum flow such
as is ascribable to the occurrence of the ' quantum stress ' (5 12). Now (5.15) with
p>,
( 5 - 1 6 ) is exactly the momentum conservation (5.11) for the derived cs. en. (P,
and so we can also obtain the equation of motion (5.9).
Furthermore another one of pure state conditions, (4 19), for the ps. en. immediately
5).
Thus we can
warrants the subsidiary condition ( 5 - 4 ) for the derived cs. en. (P,
conclude that the cs. en. produced from the original ps. en. by projection is in fact apossibk
one satisfying (A).
&
(x), satisfying
ii) Conversely, if we are given a cs. en. specified with P ( x ) and
(A), we can consider many a phase space distribution f ( ~p ,) which can yield that CS.
en. (P,
5)by
projection.
- Pip,,~ ~ j ~ ~ p according
,,*..,
f(x,
p ) ziniqu&, such that
5 we determine
successively quantities,
to ( 4 * 1 8 ) , (4-21)
higher moments, since now every order moment is specified for the df. This ps. df.** is
6)
*) In this equation for the time change of mean momentum, the effects of the stochastic transitions in
momentum induced by the external potential Y are reduced to the classical value --1Ual.*/&rd.
**) The explicit form of this df. can be written down as
1;
i.e., rS=$.
222
Formulation
The
of Quantum Mechanics in Terms of Ensemble in phase Spcae 361
I n all above the equivalent correspondence between ps. en. and cs. en. has perfectly
been verified.
(c) I n a previous paper), we inquired whether the quantum potential (5.10) in the cs.
en. formulation could be analysed into a mechanism like any Markoff process underlying.
The problem is now explained more clearly : From the viewpoint of the ps. en. formulation, the quantum potential can be regarded as an ctfpn7*ent f o 7 - c ~appearing as the result
of projecting on to the coordinate space the ps. en. that satisfies the pure state condition
and changes according to a sort of Markoff process. W e cannot, however, regard the latter
picture as a literally real one any more than the picture of trajectory ensemble under
quantum potential, on account of the inevitable appearance of negative probabilities. Furthermore it is also to be noted that the ps. en. could not yield mean values correctly for
certain quantities, and that the pure state condition was in a sense ad hoc.
Notwithstanding, our formulation of quantum mechanics in terms of Markoff-like
picture might further tempt the idea of some hidden mechanism of irregular external disturbances which vanish in the average yet make the particle momentum fluctuate, acting
on particle irrespective of its momentum [see (ii) of 3 (a)].
But the features of stochastic
transitions stated in 5 3 (a) do not allow to construct any such model ~~~zlisticnZ@.
Recently Weizel attempted to derive the quantum potential from certain stochastic
process based on some model. H e proceeded in a considerably different fashion, but the
nature of his method may also be illuminated from our viewpoint which may be more
far-reaching than his method, standing upon the systematic formulation of quantum mechanics
in terms of the ps. en.
By the way, the mean kinetic energy of particle for the observer moving with the
{s(s2},
--x
(d) W e have explained that a cs. en. can be looked upon as the projection of a ps. en.
for pure state. It is to be remarked that as a result of such contraction the cs. en. gets
free from an unrealistic property of the ps. en., i.e., negative probabilities, but at the same
time it partly loses the property of the ps. en. to give correctly the quantum-mechanical
expectation values as the ensemble averages for most of usual quantities: The cs. en.
defined with density and momentum fields, P(x) and B S ( X ) , yields the mean values for
a quantity
A(%,
- _ _p )
as
223
T. Takabayasi
362
which agrees with (2.12a) for quantities of zeroth or first order in p but not for those
of second or higher orders4).
As for the ps. en., relations known to hold between quantum.mechanica1 expectation
values of some physical quantities usually hold also with the understanding that the average
over the ps. en. is to be taken, since it yields the expectation values correctly as the
ensemble averages for most of usual physical quantities. A n example is the uncertainty
relation stated in 4 (a); another one is the Ehrenfests theorem : Integrating (5 15)
throughout over the x-space we get
n2(xi)f/dt?=d(%)f/dt=- (a v/ax,>,,
which is valid for general mixed states. I n the particular case of pure states, this relation
may also be written, in terms of the quantities in the cs. en., as
?nL(xPdx=at2
virial
dt
theorem
pp dx=-
Pl7
v dx.
d(xp),/dt= l / m . (p2),+( X V V ) p
and also the variational theorem, into which we do not enter here.
i) We shall ask whether there can exist any pure state distributions having no correlation at all between particle position and momentum. Such distribution must factorize as
f(x,w)=fYx>a (2,)
(6.1)
and therefore satisfies one of the pure state conditions ( 4 - 19) from the outset.
must further satisfy ( 4 - 6 ) , we obtain
---aik,
Since it
(6.2)
(6.3)
c con st.
exp[-xaikxi
d, b
xk++ri
x,],
(6.4)
224
(6.6)
where each as should be positive so that P should be normalizable.
Next*, re-expressing ( 6 . 3 ) in terms of the fourier transform for Q(p),
(6.7)
x (0)= 5 Q (P)eil'q 4~
we get
(6.8)
This is a equation similar to ( 6 . 2 ) , and can be integrated into
log
where 7,'s are the new components produced by the coordinate transformation (6.5) applied
to ~ I f s . W e have now
Q ( p )=const. exp[-x
(1/4Ea,) (pp'-b;)'],
(6-9)
f ( ~p ,) =const.
exp[-
(6.10a)
(6 1Ob)
a&= l/A',
namely it is Gaussian for each freedom of position and momentum components in a suitable
coordinate system, with the relation (6.10b). It has the minimum ' uncertainty product ':
((Xi-
(xi),>2>,.(
(9r(fJJ2),=&.
The wave function corresponding to (6.10) can be obtained according to the procedure of 4 (c): First we derive
F(X,?/)=
if(%,
=const.
p)c-+l,Y'h
exp
dp=Y(x) .x( - 7 J / A )
{a,(~l-ani')'+(1/4) CI!*Y:'+(i/$) hi'y:}],
[-x
a
p ( x , X,)=const.
exp C [ - a I / 2 -
{ (T:-LZ;)'+
(~,:--n',)'}
+( i / A )
(bi(.~;-x~;)],
SO
(6.11)
*) From here on it would be simpler to proceed as follows: The other form of the pure state condiwhich is similar to (6.2), and the condition (4.18) further
tion, (4.10), yields 8: log Q/8ji 8 j k = - f l L k / Z
requires (6.10b).
225
T. Takabayasi
364
which is nothing but the general form of the 'minimum wave packet'.
ii) Next we shall ask whether or not the form of the distribution (6.10) (i.e., the
property of having no correlation) be conserved during the course of time. First, taking
up the case of free particle*, we suppose that a distribution of type (6.10),
(6.12)
(here considering one-dimensional case for simplicity) occurs a t a time t=t,. I n this case,
as was stated in 3 (a), each point in the ps. en. moves classically with its respective
constant velocity, and the distribution at time t is given by
(t-to), p )
f ( x , p . t )= f o ( x - - p / ? / I -
(6.13)
But. this is no longer of type (6.10), and indicates that the correlation grows with time.
T h e phase space distribution (6.13) gives at once the space density in the well-known
form
showing that the distribution in coordinate space diffuses with the passage of time. Thus
the so-called "diffusion of wave-packet" in this case is, from our viewpoint, simply a
result of the fact that each point of the ps. en. performs the purely classical motion with
its Y L J S ~ P C ~momentum"'.
~ V ~
I n other words the simple circumstance that the particles
which move faster cover the greater distances brings about the spreading of the packet
and at the same time introduces correlation.
(b)
i)
Linear oscillator
The hamiltonian is
H ( x , p ) = 1/2. (p?/m+ ? ? Z ~ I J ~ X ? ) ,
and the quantum Liouville equation is identical with the classical one :
af/&
+ p / ? n -af/ax--nto'x af/ap=o,
(6.14)
f=F(p'+
(7/iwx)',
-$ sin o t + m o x cos o t ) .
(6.15)
Now in case of oscillator, if there can exist a pure-state distribution having no correlation
during the course of t i r x , such distribution must be of the form (6.12) and at the same
time consistent with the form (6.15). This determines the parameters in (6.12) such
that
*)
226
6=
-711tW(Z0
sin cot,
(ao:
const.)
(6.16)
f(x,
1
Is, t ) =--exp
mi
WUU
sin tot)'
7lLto-h
(6.17)
This distribution takes its maximum value at the point ( ~ = n ,cos cut, $= -a,,ui~o sin tot)
in phase space, decreasing around it in Gaussian manner. W i t h the passage of time the
distribution rotates, kcLyiiig its f077n y<qidZy, along an ellipse around the origin of phase
space with the angular frequency w .
T h e corresponding wave function'') is obtained by inserting ( 6 . 1 6 ) into ( 6 . 1 1 ) as
7fl(U
$b(.r, t ) =exp --
26
(x-a,
I.
(6 18)
nzw
A ( t ) = - 1 -a0'sin
4
2oi---cof.
2
(6 * 19)
ii) As previously stated, for quadratic potential such as in case of oscillator, the ps. en.
develops classically. O n the other hand, in such potential, also in the cs. en. formulation,
we have particular solutions'' which consist of purely classical trajectories, with quantum
potential vanishing, and are determined by the following equations :
(6.20)
(6.21)
(6.22)
Now, such a cs. en. corresponds to a ps. en. with the df.
(6.23)
which can be simplified in one-dimensional case as
f ( x , p , t ) =z~(t).s(p-as/ax),
(6.24)
The expression
because from (6.21) we have S(x--y/2) - S ( X + J / ~ )= --y aS/ax.
(6.24) (together with ( 5 . 5 ) ) shows that for such particular quantum-mechanical motion
the ps. en. and the cs. en. become identical with each other.
I n the case of oscillator, a solution satisfying (6.20),
by4'
P= const/sinor,
S= (mo/2sin
(T
(or) { (X'+X,?)COS
=t -t o)
(UT- ~ Z X , } ,
(6-25a)
( 6 * 25b)
227
T. Takabayasi
366
(6-26)*
1-his expresses an ensemble of trajectories flowing out X , at to with a7y velocity. The
corresponding wave function P r l z~6 is nothing but the propagation kernel K(xtlxoto).
I t is to be noted that the transition probability of
3 (b) is a different thing from the
above df., and is given by (3.35) in this case, implying the distribution at z when a
particle starts from xo at to with a de$nite momentum p,.
(6.27)
from which we can obtain the corresponding ps. dfs. as
(6.28) **
Each of these distributions has a constant density on an energy surface, as it should for
a stationary distribution in case of any quadratic potential, and takes negative as well as
positive values, giving the quantum-mechanical expectation value of energy correctly as
( H ) r ,= $ K ( xp),f,(
5,p)dx
dp=
(?I+
1/2j30=E,=
(H)n.
.-
However, it cannot give the expectation value correctly for any power of energy, H( I J > ~ ) ,
because fn distributes over an ~ L Z in phase space.
The ps. distribution that yields the probability distribution correctly for H is clearly the
-
md
*) In this case the wave function and therelore the ps. df. are not normalizable; the const. in (6.22)
(6.26) are in reality infinitesimal.
**)
-?%.(
k=O
( - I ) ~ L X ~L L ( x , p ) (n=o, 1,
m).
z-..)constitute
228
H ( x , 9)= (12
+1/2)E w ,
p ) K6(H(nr, p ) - ( 7 1 + 1 / 2 ) h J ) .
Such distribution cannot, however, correspond to any state, as is understood, from the fact
that it does not satisfy the Wigners condition ( 7 . 1 ) .
8 7.
(a) E.P. Wigner pointed out* that following two important conditions are further
n~wssary,besides the conditions ( 2 - 9 ) and ( 2 - 1 9 ) , for a ps. df. f(x,p ) to be a
permissible one corresponding generally to a mixed state. They are
(7.1)
(7.2)
The condition ( 7 . 1 ) which is derivable by applying Schwarz inequality to ( 2 . 6 ) with
(2.4) indicates that a possible df. should extend at least over a phase volume ( h / ~ ) ~ ,
expressing in a certain degree the uncertainty principle for the general case of a mixed
state. The left side of ( 7 - 2 ) , being
+x
Wn4
(7.3)
(where run is the quantity defined by ( 2 . 4 ) ) , becomes unity for a pure state and is
smaller for a general mixed state. This quantity may be taken as giving a measure to
the degree of mixing, though it is different from entropy defined in 4 (b).
The Wigners conditions (7.1) and ( 7 . 2 ) must be regarded, together with ( 2 . 1 9 )
as a part of the positivity condition stated in 2, which means the condition to be
imposed upon any possible ps. df. corresponding to ( 2 . 3 ) . These conditions (2.19) (7.1),
and (7.2) may not yet constitute the sufficient one, since the positivity condition means
that every nun be non-negative and so it would require infinitely many inequalities of a type
such as ( 7 . 2 ) . Indeed, the condition ( 7 . 2 ) , which is expressible in terms of density
matrix as
j p ( x x)p(x x) dx dxi 1,
(7 2)
jp(zx)p(xa)dx5 p ( x x)= P ( x ) .
(7.4)
The latter condition ( 7 . 4 ) is also a iiccessnry one, and is expressed in the language of
the ps. df. as
235f(x
p ) f ( d ~ ) c o s [ ~ /( zp. - p ) (XI-
*)
x)]& dp dx
229
T. Takabayasi
368
2 (f(xp)dp=Y(x).
(7.5)
The left side of (7.5) is non-negative, and therefore (7.5) means a severer condition than
the first relation of (2.19), and at the same time involves ( 7 - 2 ) . Similarly the second
relation of (2.19) can better be replaced by
(b)
(7.6)
(7.2)
as a necessary condition for a ps. df. to correspond to a pure state. This condition has
a different form from those obtained in 4. Now we can show that the pure state condition can also be expressed as a series of infinitely many relations each of which is of a
type similar to (7.2O), the series including (7.2) itself as its first relation.
For that
purpose we start from the original form of the pure state condition for density matrix,
( 4 . 2 ) , in place of (4.4). Re-expressing (4.2) in terms of ps. df., we obtain
I(
23
x--,p
2
(- 2 i/%-p?/)
dp.
(7 7)
This complicated relation involving two continuous parameters x and 7~ can be split into
a series of infinitely many simpler relations by the following two-step procedure. First,
integrating (7.7) over the whole x-space, we obtain
dp.
(7.8)
(7.9)
I n similar fashion, we multiply (7.7) successively with xd x i ,
xj =c,,..., and then
integrate the results over the whole x-space to get a series of relations involving
$-differentiation of sdccessively higher orders. This series of relations, (7.8), (7.9),-..,
may be regarded as equivalent with the original (7.7).
230
Then
(7.10)
(7-11)
.................................
Similarly (7.9) yields
P j X $f ax d p =
[Xi f
ax
dp,
(7.12)
(7.13)
\ ....................................
W e have thus found that the pure state condition can also be expressed with a set of infinitely many relations,
(7.2),
(7.10),
and
(7.11),**.; (7.12),
(7.13),..-;
ones,
(4.6),
(4.19) ; or
(4.19),
(4.18),
(4.21),,-..
But both forms should be equivalent with each other, which fact may be understood if
we return to the expressions in p, though it may be difficult to show it directly.*
(c) W e could take either (4.2) or (4.4) as the pure state condition in terms of
density matrix. W e have, however, still another relation valid for a pure state,
(7.14)
though it is a necessary but not sufficient relation for the pure state condition. It can
be shown that the relation (7.14) is derivable from (4.4) by the Taylor expansion of
the former.
p(xx)= ~p(xx>ro(xx)dx,
(7.15)
see also
(c).
(7-16)
23 1
T. Takabayasi
370
('.c+si=7?i)
- (-
4,-. ...*
> 9
(7.17)
with
P7y.'f3=
j p p p2?-?ps'.xf (x,P)
4%
where the summation is to be taken for zero or integer values of each I ' ~ , s,, with r,+s,
= i i i fixed.
The lowest relation (/z=Z)
of (7.17) is nothing but (4.13) (i.e.,(4.18)),
and the next one involves the fourth order moment.* Thus we see that the relation (7.17)
or (7.16) consists of just a h a y of the former relations (4.19), (4.18), ( 4 . 2 1 ) , . . .
of the pure state condition corresponding to (4.4).
The relation (7.15) is expressed in terms o f f as (7.5) with equality sign, which
can be split into a series of relations, (7.Z0), (7.12),... by the procedure of taking
moments with respect to x . These relations, on one hand, naturally coincide with a part
of relations obtained in (b), and, on the other, must be derivable from (7.17), since
(7.15) was a result of (7.14).
Our method is essentially the transcription of .von Neumann's density matrix methodqa)
through a particular fourier transformation of a specified representation of the latter which
formulates quantum mechanics generally for mixture and characterizes a pure state by a
subsidiary condition, thus leading to formulating quantum mechanics in a closed form in
such a way that it is associated with the phase-space ensemble picture.
Really the method was described for the case of single non-relativistic particle, but it
would also be interesting to apply this method to the case of a many particle system.
I n conclusion the author would like to express his sincere thanks to Professor S.
Sakata, Professor K. Husimi and Dr. S. Nakajima, and also to Professor E. P. Wigner,
for kind interests and valuable discussions on this work.
232
V ( x )=Yo for
J(x,
ii)
1x1
2 n/2,
arid O for
and so
p ) = -2Vo/nZj.sin(jn/Z)
sin (2jx/R).
Therefore particle momentum can jump only by an amount of some ' resonance value ' Akn/2.
iii) For Coulomb potential V= - e ' / y , we have Y ( p )= - (%/2&)
1/p', so we obtain
In this case momentum cannot jump in the direction perpendicular to the force.
D (U,
72? 11?)
p . ad[)- p * D (721
(4.4),
722 7 1 3
is satisfied, we can
0,
(B.1)
7z7
??,)D'(Izi71: 72<)p=o,
JZ:
D' (7~:
n,') Gap',a*'n'=astnfq with
where D ( M 7z2
, 72),
&,l,N( and 1/11 being non-negative integers*).
Now, if (4.4) holds for any values of x and K', we must have
ai=a/a.Vi,
(B.2)
a:=
a/
(ai p a ~ p -pa, a:
for
CI
= 0,
p)
,vl=o, 1, 2,..-00,
(B.3) can further by replaced by
for
(B.3)
?il=1, 2, 3 , m . e
03.4)
Co.
*) It is to be noted that the compatibility of the pure state condition (4.4) with the equation of motion
(3.2) for p can here be shown explicitly, as we can derive
alat
from (3.2) and (B.2).
(a,pafkp-at
ak'
P)
=o
233
T. Takabayasi
3 72
SLYL'S
of
~ L & Z ~ ! O I Z SOIJ
tht do-
xomZ, (B.4).
[a<,o a;
p-p
a<a;
(/].r=J,=O,
03.5)
,4rs,r,
= 0,
P.6)
La,
a,'
-f ) a k a:
the nine comp/:lx relations, (B.5), are not independent of each other but consist of six sym~
metrized T E equations,
[(a, p a:
fl
+ ak a; ,4- ,)(a, a;
p i - a k ail P ) I , ~ = ~ , = O ,
(J3.7)
{'-a& 8;
(B.8)
;[(a,
f)
8;
//-ah
4
0
8;
fJ)-//(ai
8;
p)]~~21=0.
[ai a, ,o a;
[)
- ,O aia,
a;
p]X.z, =o
03.9)
(b.10)
Here, using the relation
with
u,,
T? 73=
-~'(na~z-T
-a&]
l)
1'?7.3-
i p(Yj7-??.3)
+Ik)]=o,
which may further be separated into real and imaginary part equations.
( B - 11) *
Thus we have
obtained the pure state conditions in the form of relations between distribution moments
234
1953).
8a)
9)
10)
11)
12)
13)
14)
15)
16)
.:2)/2>.ln=
S n t $ t f ( % P ) dx dP.
Further, if we take this hermitian quantity not belonging to (2.12b) for the _.
-4 in (2.3), we obtain another
one of positivity conditions,
XS?
pi? f(2, p )
dx d p l - S / 4 .
T o page 346 : A quantum-mechanical state compatible with given P ( z ) and @ ( p )cannot be determined uniquely even when the state is restricted within pure cases (see K. Husimi, Kagaku 5 (1935), 370).
3) To page 352 : T h e transformation function T for the case of forced oscillator was given by T.
Nishiyama, Prog. Theor. Phys. 8 (1952), 655.
4) To $ 5 (c): Recently Weizel applied his method to many particle system (ZS. f. Phys. 136 (1954),
2)
582).
5)
as
of mixture, since it means that a permissible ps. df. f(x, p) should be expanded as the superposition with
fil satisfying
as
f ( x , P ) =-ptf,,(2P
, ),
235
PHYSICAL REVIEW
VOLUME
109,
NUMBER
MARCH
15.
19.58
236
FORMULATION
OF QUANTUM
MECHANICS
2199
goes to zero,
-[A$]+
h
(A$) =
Jr. *JT
J*. :J fI
X A (Tk,uk)B(kjlIk)dl.. .dndql*. *
237
2200
XJ/*(Qk+yk)+(Qk-yk)dYl
dyn. (3)
. .dfln,
Xf(qddfi1.
qk-Sk)
=S, *S,
+m
+m
. .dwn.
* *
If we think of G ( q k f s k , W I ) as the kernel of a homogeneous, linear integral equation, we see that it has at
least one solution, i.e., G ( w t , qk.-Sk) and its eigenvalue
is unity. By a slight modification of the arguments of
Courant and Hilbert,Kwe know
+=
S, . . *S,
+m
[ G ( q k , W k ) I * d q l . . dqndwl...dWn?C
i-1
1 I2
SC-..~(qk,qk)dqldqn,
f=Cpwpf,(qhflk)-
s, *-s,
+a0
+m
f ( q k , f i l e ) d q i * --dq&i.
.dfi,=l,
Pp=Ci
up$+
j= C wP~,i*apif;;,
P.Li
238
FORMULATION OF QUANTUM MECHANICS
2201
where we define
(iii)
1,.
if and only if
f;j*fk,,,=O
(iv)
Jjj,=6ij,
~ d qf l . .;. d qj n d*q i . . . d p ~,
we obtain
C C p
wpapi*,apil
+m
. .
[Rjm]=J
--m
R(qk,Pk)
Xfj,m*(Qk,P$dql...dqndpl...dp.,
. *dqn.
62- 2s= ( h / i ) 6 ,
239
2202
uu= u,
which is just von Neumanns characterization of a
pure state. The physical interpretation in the two
cases is similar. I n matrix language, it characterizes a
projection operator onto some state, while our condition
may be thought of as characterizing sort of a smearedout projection operator for a region of phase space. I t
represents a modification of the classical delta function
which projects onto a phase-point.
(gk)
[>
(qk,pt)B(qk,flk)
X d p l . .dqndpi. . .dQ,.
4,BI=WA,B),
if A is a polynomial, at most quadratic, where ( A , B )
is the classical Poisson bracket.
The large-scale experimental validity of classical
mechanics tells us that quantum theory must, in some
sense, correspond closely to classical mechanics. We
have altered the classical concept of a moving point in
phase space to that of a quasi-probability distribution
which changes in time. This distribution (see Sec. 11)
is imagined to be concentrated about the classical
point, so that a crude measurement will be unable to
differentiate between the two theories. T o insure this
lo W. 0. Kermack and W. H. McCrea, Proc. Edinburg Math.
SOC. 2, 224 (1931).
ai
6j= si(aj/ai).
We must have, by the correspondence principle,
= - cqk,6gl.f
=-h{qk,8g)
CORRESPONDENCE PRINCIPLE
A B = c*
a,r
Also
dl
=-(qk.f)=qk.-.
=61{(lk,H),
.f
f.
Sg = -Hst/h.
af
-=
at
--Cf,HI,
A
240
FORMULATION
OF QUANTUM
V. FORMAL PROPERTIES
2203
MECHANICS
aik*aim=L5knr.
i-1
(A,B).f=( f , A ) . B ,etc.,
C ~ , ~ ~ ~ ~ I I + C ~ , C ~ , ~ 10,1
which. as
JW*=1, implies
VI. MEASUREMENT
X~k*(qj+yj)~k(4j-yyj)dyl'
w=Ad',
'dyn,
=0,
or
6(R.f )=R.6j = R . [
and
[R,j ] . 6 g = O 1
CR,fl=O.
then fi=f11. This means that if f represents an ensemble composed of equal numbers of systems in N
orthogonal states, then we get the same f no matter in
which way we make up the orthogonal states. To see
this, we expand fix as
X=R. fAh.
We shall say that the
{ f h ~ ) form
a "complete" set if
l=hnCAf A A ' f
(conservation of probability) for all quasi-probability
distribution functions f.
11H. effreys and B. S. Jeffreys, Methods of M a t h a t i c a l
Physics [Cambridge University Press, New York, 1950), Sec. 10.14.
24 1
2204
S=
Substituting for C(S) and equating these two expressions, we see, when the appropriate interchange of
limit processes is permissible, that we must have, as 9
is arbitrary by the relations of Sec. I11 (vi),
C ~ X A ( ~ A ) = - - 0 5 ASR
1
27rh
J+[l-exp(iM/tz)]
--ao
is
X ( ~ ( - ~ ) R ( ) ( q k , f i k )
V!
>a*
XdFx(qk,pk),
W ~ U ~ A * U ~ . ~ A ~ ( ~ E , P ~ ) .
(Stieltjes integral),
P.XP
5=C ~ ~ a , ~ * a ~ ~ . f ~ ~ ( p b , p k ) .
PJ
R().
5 = (R,R(d).
5
for all 5,as we would expect from the analog pointed
out in Sec. 111.As MoyaP has shown, R()=
1, so that
we may use the above relation to construct successively
the Rev).
This result gives an explicit method of solving the
eigenfunction problem for the measurement of R. We
use the above equation to compute the R ( v )and then
1* M. G. Kendall, The Advanced Themy of Slabistics (Charles
Griffen and Company, Ltd., London, 1947), Chap. 4.
242
F O R M U L A T I O N OF Q U A N T U M MECHANICS
where
with
[Tj,frj]=O,
P=R.jpp,rr,
Q=Sfpp,ra,
P.
C
P.
fpp, gm.
5.
0s ,A R
osuss
(R()S())
=hn
P~fpp.~.(qk,p,t).
all ( P . 4
(R(),S())=hZ(
C
Pfpp.ra,
.I1 ( P . 0 )
C Ujpp,m)
811 cpvu)
= (R(Y)SOI)),
~ j = T j . f + j ,
G ( T , ** * , T N )
is given by
(G)=
2205
s..-s
G(TI.. . , T N ) ~ F ( T.I.,TN).
,.
entire range
of the Th
[R,S]. 3=0
for all 5. That is to say, if two quantities are simultaneously measurable, their operators commute, a
well-known result of the standard formulations.
Let us define an Nth order cosine bracket as
243
2206
Ol
n-4
=exp[- ( i / A ) ( 2 H / u ) tan(fsw)]/cos(fsu),
It is a matter of straightforward
calculations,9 to
"
show that for the one-dimensional harmonic oscillator, where w = 2nv.
the energy eigen-quasi-probability-distribution-func- We now obtain the various H'"' from G (s) by the
tions are:
relation
~~
fn ( H, @ ! H a = [(
- 1) */ (27rn!)-&,(4H/hv)
X exp ( -PH/hv)d (2H/h v) d8,
0 = tan-l[p/(2?rmvq)],
H = (p2/2m) 27r2mvzqz.
+m
fn = ( 2 ~ h ) - 1 1
-m
We obtain by differentiation
244
Ptoc. Cccmb. Phil.
Pt-i&
800.
581
in Qreat Britaan
/yJ;m
mz,P)f(z,P,t)dzdP
J-;aJ-;m
f(Wht)dzdYP,
(1)
where 3,the energy of the system, is time-independent provided H(z,p), the Hamiltonian, is. We seek an integral equation forf such that (1)will be automatically true.
Consider
(2)
245
D. B. FBIRLIE
582
Then
l ( z J p J t ) d z d=
p
Iw
IW
H(z,p)f(z,p,t)d~d/p
-m
-00
= E/~m/~af(z.pJ
t )dz
dp.
(3)
We postdab that there are no intrinsically distinguished points in phase space; i.e.
any local continuous mapping of x J pto x,p should leave either all points or none
fixed. These mappings are restricted to be pure translations by Brouwers Translation
Theorem (3). (Iam indebted to Dr P. H. H. Fantham for this reference.) Thus we may
have
l(z~pJt)
= Ef(XJpJt)J
(4)
or more generally
I(%,pJt, = E f ( z+ p + bJ t,
(4 a)
rn possible local relations between I andf. In other words the relationship is chosen
to be translation invariant. We further require that if H = constant there is no
restriction onf. Thiswill be so if #(a)= S(a- 1) and the more generalform of identifica-
x f ( 7 , h)g(,uJv)d7dhdpdu
The undernoted identities, which follow by formal calculation from the above
definitions (11,are quoted for convenience.
[A,B].C= [C,A].B= [ B J C ] . A J
= -LBJA]J
=
( A J B ) . C= ( C J A ) . B Jetc.,
=
[aJIBJC]I+[C,[A,B]l+[BJICJa]]
[ AJ (B,c)]
= ([A
J B]Y
c, + ( [ A1,
J
>
(5)
B).
(6)
Consider two eigenfunctionsf, and f, with real eigenvalues E, and E,, respectively,
which satisfy (6). (The eigenvalues must be real since they represent the possible
246
583
energy values of the system.) Then with the aid of ( 5 ) we may deduce the following
equationa
H.U,f,*)+iCfiffI) = 2E,f,.ff,
H - {(f?fj)- iV?.fjI) = 2Ej.f:
-f5-
(7)
(8)
Let us examine how far (7) and (8) are consistent with the a~sumptionthat equations
of type (6) hold for both (f,fj*) +i[fif3and its complex conjugate, i.e.
(Wfifr*>
+~[fd?l))
+ i[-rr((fifr*)
+i[f,ffl)l = 2-Wfif:) + i[fJ?I)>
(fwi*f,)-icf t?fjI))+i[rr((f?.fj)- i[f?f,l)l = 23j{(f,*f,)- iCf?hl).
(9)
(10)
Equations (9) and (10) are consistent with (7)and (8) upon integration.
We may then label the eigenfunctions of (6) according to their eigenvalues and those
of their complex conjugates as
(11)
(fif;)+ iCfiff1 = kfi5Y
where k is a normalization constant.
By definition
(12)
fS = fji.
Rearrangement of (9) and (10) with their complex conjugate equations yields
(13)
(14)
Equation (14)has been given by Moyal(5)and is already implicit in previous work (7):
equation (13) does not appear to have been stated before.
Comparison of (7)with the complex conjugate of (8)provides the further consistency
requirement
fi.f j* = 0 unless Ei= Ej.
(15)
Similarly, fi.f j* = 0 unless f
notation introduced above
fil.fmr
and!?
=0
(16)
This is the orthogonality condition for the eigenvalues of (6). ( 1 1) may be rewritten as
(17)
{LfjJiJ +iCfji,fi;l)-fm~
or
*0
+Xfir,fmjI= K4mfij-
(18)
247
584
D.B. FAIRLIE
This is the relationship which is adopted by Baker ( I ) as an initial postulate, in the
case where i = j = 71 = m. He shows that this law implies that f&, p ) may be Written as
(19)
i.e. in the form of Wigners distribution function (7).The generalization to off diagonal
functions is obvious and yields
(20)
The orthogonality condition (16), when applied to (20) gives the usual orthogonality
requirements for the functions $$1 namely
/Ia
II.,*(Z t )
(21)
@,@,
t )dx = &j.
This implies
Distribution functions me conventionally normalized to have unit integral so that
they may be interpreted as probability densities: the choice K = h leads to no loss of
generality of the theory and permits the observance of this convention. These la&
relations may be used to deduce the following equation for
+:
(2)
(23)
(23)
or
(24)
Thus af,/at may be expanded in terms of the eigenfunctions of (6) with eigenvalue Ei.
suppose
afcr = CC&.
a
(25)
Since
(26)
248
585
In consequence of the orthogonality condition (16) cin = 0 unless I =j . We may drop
the third suflix and write the consequence of choosing Z = j EM
(27)
ci5 = C?$
a2
- _ - c&.
(28)
The result of differentiating (18) and substituting (28) gives a further condition on cij,
i.e.
C ~ + c 1 3 = Ck5
( d l k, andj).
(29)
Put k = 1 in the above relationship: then ckk = 0, or (apt)f k k = 0, i.e. the real eigenfunctionsf, are stationary. This permits the expansion of H in terms of thefis aa
H = =-Mi,
(30)
since H is time-independent and real, and this resolution of H guarantees the consistency of (18). Put k = j in (29); then
c,+c,
(31)
= 0,
i.e.
ckl+ck*E= 0.
Thus cM is pure imaginary, and must be written in the form
(32)
cm = i ( F ( k )-F(Z))
(33)
in order t o satisfy (29). This is aa far aa we can go in the determination of ckl without
involving some dynamical principle; we have not used the fact that t is a time variable
except in supposing the independence of E and H of it. We appeal to the correspondenceprinciple to verify that the simplest non-trivial choice of arbitrary function
in (33), namely
i
C u = - (Ek- El)
(34)
will do. The factor I/&is necessary to give cw the dimension of an inverse time.
Then equations (13) and (14) take the form
(H,fij)= (&+Ej)f<j,
(35)
[wd= 5 afxjz
(36)
and the second equation in the limit h + 0 is just the classical Liouville equation, thus
verifying (34).
Finally, &s an example of the simultaneous solution of the above system we consider
the harmonic oscillator Hamiltonian
H = 4(p2+x2)&w.
(37)
Ifwewrite
fM(X,$,t) = ~ k l ( X , P ) e x p i ( k - W
(38)
= 0,
(39)
(40)
249
D. B. FAIRLIE
586
where Ekhas been set equal to (kf4) ?iw.These equations have the solution
(41)
where jYck(w)
is the associated Laguerre function, w = 2(p2 +x2) and tan 6 = p / x . This
solution has been obtained before, (2,4) by calculation of (20),using the known
harmonic oscillator wave functions.
REFERENCES
(1) B-,
c m e
250
Plzysica 83A (1976) 210-212
The Wigner function, which is commonly used as a joint distribution for non-commuting
observables, is shown to be non-negative in all quantum states when smoothed with a gaussian
whose variances are greater than or equal to those of the minimum uncertainty wave packet.
The Wigner function, introduced by Eugene Wigner in 1932l) to study problems of statistical equilibrium, has long been used in quantum thermodynamics*).
More recently, it has been introduced in quantum optics to study coherence
properties of light3), and it is currently applied in the study of plasmas4) as well.
The function serves as a joint probability density for non-commuting observables
despite the fact that it is known to take negative values in many states. Its use as
a probability is frequently defended by the assumption that the function will be
non-negative when employed in a way that does not violate the uncertainty principle. Mori, Oppenheim, and Ross, for example, conjecture that the Wigner function is everywhere non-negative when integrated over regions of phase space of
the order of ?i3N9. Thus (restricting attention to one dimension for simplicity) it
is frequently supposed that
(1)
for all quantum states, p(q), and for all pointsp', q', where W ( p , q) is the W i p e r
function:
+ + ~ hd)t .
A result weaker than eq. (l), but of the same import will be proven here.
(2)
25 1
21 1
D ( p , 4)
1/A2,
P' - h/2
Ip I
p'
+ h/2;4' - h/2 2
5 9'
+ fi/2
otherwise.
= 0,
x exp (-itp - p ( p - $ ) * / A
-a
(q - q')2/fi) d t dq dp 2 0, (3)
+ tth,
(4)
f(u)
+ u (aq'/Zi + ip'/A)]
(7rha)f
--
2 X W
f ' * ( y ) yn dy. ( 5 )
n=O
252
212
N.D. CARTWRLGHT
References
E. P. Wigner, Phys. Rev. 40 (1932) 749.
J.VIieger, P.Mazur and S.R. de Groot, Physica 27 (1961j 353, 957 and 974.
H.Mori, 1.Oppenheim and J.Ross, in Studies in Statistical Mechanics, Vol. I, J. de Boer and
G. E.Uhlenbeck, eds. (North-Holland, Amsterdam, 1962).
1.Bialynicki-Birula and Z.Bialynicki-Birula, Phys. Rev. A8 (1973) 3146.
R. J. Glauber, Phys. Rev. 130 (1963) 2529.
C.L.Mehta and E.Wolf, Phys. Rev. 134 (1964) A1149.
M.Lax and W.H.Louisel1, I.E.E.E. J. of Q. Electronics QE-3 (1967) 47.
J. P. Hansen, Phys. Rev. A 8 (1973) 309k
E.L.Pollack and J.P.Nansen, Phys. Rev. A 8 (1973) 3110.
L. J. Roszman and C. F.Hooper, Phys. Rev. A7 (1973) 2121.
5 ) H.Mori, 1.Oppenheim and J.Ross, loc. cit. n2.
253
PHYSICAL REVIEW A
FEBRUARY 1977
It is pointed out that the Wigner function f ( r , p ) is 2 / h times the expectation value of the panty operator that
performs reflections about the phase-space point r, p. Thus f( r, p) is propor&ioualto the overlap of the wave
function Jc with its mirror image about r, p; this is clearly a measure of how much Jc is centered about r, p . and
the Wigner distribution function now appears phyeically more meaningful and natural than it did previously.
(5)
f(r,p ) =
aJ
d s e-zips/$(r- s)*+(r+s) ,
(5')
(1)
(2)
This Wigner representation has proved useful
for studying the passage from quantum to classical
mechanics and establishing quantum corrections
to classical results, and generally it enhances understanding by favoring the use of classical intuition in quantum problems.2
At first sight the constructions (1)and (2) seem
rather a d hoc and devoid of any deep physical o r
mathematical significance. A somewhat more
meaningful expression for f(r,p ) was provided
by Moyal: namely
(3)
where 6 and 9 are the position and m2mTntum
operators, respectively, satisfying [R,PI =iti.
The fofma(3) is conspicuous to statisticians:
et(kR+sP)
I$) appears as a characteristic function, being the expectation of the operator that
corresponds to the function ei(h+sp)in Weyls rule
of a s s ~ c i a t i o n . ~
Here we wish to point out t h a t f ( r , p ) has a much
more direct physical meaning, in that it is the expectation value of the parity operator about the
phase-space point r ,p.
To show this, let us first rewrite
($1
f(r,p)=(2/h)(+)n,,l+),
(5'')
( 5)
where lr) and Ip) a r e eigenstates of fi and p,
respectively. Let u s now consider the special
case r = 0, p = 0, and denote IIr=o,Bsp II; we have
(6)
(6')
(6')
From (6) or (6) it is immediately apparent that lT
is the parity operator (about the origin): it changes
$(r)into +( -r) and &p) into $( -p), o r equivalently
(note that n-=n),
nrin=-8, nr;n=-P.
(7)
We now observe that n, may be obtained from n
by a unitary transformation
nrp=D ( r , p ) n D ( r , p ) ;
~(~,p)~~i(~iP-rB)/n
15
-
( 9)
(4)
(4)
( 8)
here
449
(10)
254
450
ANTOINE ROYER
D(Y,
p)-l~(Ei,$ ) ~ ( r , p )= q R + Y , P + p )
(11)
(nJ=
1.
(12)
n,#l@;#)=*l@;p)2
(13)
a r e functions that a r e either symmetric o r antisymmetric about Y ,p. They may be obtained by
displacing in phase-space functions of the same
symmetry about the origin, i.e.,
I@$P)
= W r ,P) @*)9
(14)
=h(1f nrp)
(P:J2=
[ F ( k P)
, being a power series in f? and $1, whence
Eq. (8), in view of (5) and (6) [and noting that
D ( r , p Y = D(-Y, -PI].
Using (71, (8)) and (lo), we readily verify that
nr@(R
- r)II,,=-(I? - r) ,
rI,,(P - p)rI,,= -(P - p) ;
15
-
(15)
N1,
= D ( r , P)P*D(Y,
where P*=+(l n ) projects on the space of functions symmetric (antisymmetric) about the origin.
We have
(16)
p*l,+P;#=l,
(17)
(18)
+=
+:p+
$;p
(19)
where
$:p) e p $) .
By (16) we have
(20)
I1 $@: II * .
($k#
I$) = ($:@ I$:p)
Then by (4),
(la),
(21)
and (21),
II - II$#; II9 .
(22)
II~:@1I2 I G@112
11 $:# 1 C 1, implying in turn,
+
(23)
in view
This implies
of (22), t h a t f ( r , p ) is bounded by the values -2/h
and 2/h:
-2/h <f(r,p) c 2/h.
(24)
I$)
I#+) + c- I@-)).
We then have ($1 II,, I$) = c: - c? and ($1
(25)
= D(r,P)(c+
$) = cz + cf.
255
WNALS OF PHYSICS 111,
61-110 (1978)
F. BAYEN
DCpartement de Mathimatiques, Uniuersite de Paris 6, 75230 Paris Cedex 05, France
M.
FLATO AND
c. FRONSDAL
A. LICHNEROWICZ
AND D. STERNHEIMER
Physique Mathdmatique, College de France, 75231 Paris Cedex 05, France
Received May 19, 1977
INTRODUCTION
The usual probabilistic interpretation of quantum mechanics contrasts with the
deterministiccharacter of classical mechanics. The axiomatic settings of the two types
of mechanics are relatively disjoint; consequently the paradigm of quantization- and
to some extent also the reverse process of passage to the classical limit-does not
seem natural.
This situation has encouraged attempts to interpret quantum mechanics as a
statistical theory over phase space. Already in 1932, Wigner [I] introduced a phase
space distribution function, related to Weyls quantization procedure [2-4]. A most
interesting development is due to Moyal [5], who introduced the sine-Poisson
bracket, now called Moyal bracket, for functions on phase space. It is this bracket,
61
00034916/78/1111-0061$05.00/0
AU
256
62
BAYEN ET AL.
and not the Poisson bracket, that correspondsto the commutator bracket of quantum
mechanics. Essential aspects of quantum mechanics can be given a classical formulation in terms of the Moyal bracket and the question thus arises whether this structure has a natural place in classical mechanics.
Recently some of us 161 studied deformations (in the sense of Gerstenhaber [7n
of the Lie algebra N of differentiablefunctions on a symplectic manifold with the
Poisson bracket, in terms of I-differentiable cochains (bidifferential operators of
order
in each argument). Such deformations are trivial in the flat case (manifold
R2 with the ordinary symplectic structure) but are interesting in other cases, as was
illustrated by some physical applications [6c]. Vey [8] studied deformations of the
algebra of polynomials (with the ordinary product) and derived, in the flat case, a
nontrivial deformation of the Poisson Lie algebra in terms of differentiable cochains
of increasing order. The Vey bracket turned out to be identical with the Moyal
bracket. Vey also demonstrated the existence of such deformations of the Poisson
bracket on general syrnplectic manifolds with vanishing (de Rham) 3-cohomology.
Other mathematical properties and physical applications were sketched by some of
us [9].
These developments encourage attempts to view quantum mechanics as a theory
of functions or distributions on phase space, with deformed products and brackets.
W e suggest that quantization be understood as a deformation of the structure of the
algebra of classical observables, rather than as a radical change in the nature of the
observables. Incidentally, the nontriviality of the deformations throws some light on
the nontrivial nature of the correspondence principle. As will be shown in the companion paper [lo], our treatment of deformations of classical mechanics is a viable
alternative to conventional quantum mechanics. This suggests the possibility of
developing new methods for quantum theories, especially quantum field theories.
This article will emphasize mathematical aspects of deformations; physical applications are presented in the companion paper. In the first two sections we introduce
the notion of Poisson manifold (a collection of symplecticleaves), on which a Poisson
bracket is defined by a [possibly degenerate) 2-tensor A, and examine an important
example (the coadjoint representation of a Lie algebra). In Section 3 we treat the
case of flat Poisson manifolds (which have connections without torsion and curvature such that the covariant derivative of A vanishes) and show the unicity of Moyaltype deformations that are formal functions of the Poisson bracket. We next deal with
infinitesimal deformations of the Poisson bracket on general symplectic manifolds,
giving a simple proof of the nontriviality and making more precise the relation to
symplectic connections. h Sections 5 and 6 we determine all derivations (infinitesimal
automorphisms) of the deformed structures, both those that do, and those that do
not, depend on the deformation parameter fi. In the second case the result is a finitedimensional Lie algebra, a fact that is key to the selection of the proper quantization
procedure for a physical system. Section 7 is concerned with the unicity of the Lie
algebra deformations. It is shown that, in the flat case, there is only one nontrivial
choice to make at each order of fi2. In Section 8, we present a fairly general procedure
for constructing deformations in the case of nonflat symplectic manifolds.
257
63
i ( [ A , B ] ) p= (-1j~*+*i(A)di(B)p+ ( - l ) p i ( B ) d i ( A ) p
where i( ) is the interior product. For p = 1, [A, B] = Y(A) B, where
(1-6)
Y is the Lie
[ A , B] = (-l)P*[B, A ] .
( 1-2)
CJ,A ]
=0
( 1-31
(1-4)
where
a,
258
64
BAYEN ET AL.
(b) Introduce on W a 2-tensor I.1 and, on the space N = C K (W, R), the bracket
If u, v , w
EN
i(A)(du A do)
= A(du, du),
u, ZI E N.
(1-5)
we have
DEF~NITION
1. A structure of Poisson manifold is defined on a manifold W of
dimension m by a 2-tensor A such that [A,A ] = 0. The Poisson structure is called
regular if A has constant rank 2n (An # 0, An+l = 0, everywhere); in this case
h = m - 2n is called the codimension of the manyold.
For a Poisson manifold (W,A), (1-5) defines on N a Lie algebra structure, the socalled Poisson Lie algebra.
(c) A symplectic structure is defined generally on a manifold W of dimension
2n by a closed 2-form F of rank 2n. We denote by p: TW -+ T*W the isomorphism
of vector bundles defined by p ( X ) = -i(X)F; this isomorphism is extended to the
tensor bundles in a natural way. Let I.1 be the 2-tensor p-*(F) of rank 2n; the Poisson
F ) is defined by (1-5) and we have [A,A ] = 0. A symplectic structure
bracket of (W,
is nothing other than a regular Poisson structure (W, A) of codimension 0. Moreover,
if A is a p-tensor, we have
P"4
= dPW.
(1-6)
It is well known that there are, on a symplectic manifold ( W,A), atlases of canonical
charts {x'} = {x", x"} (a = l,..., n; G = cy. n); for such a chart, the only nonvanishing components of A are
La+
-AGa
= 1.
= 1.
In particular, ha
= 0 and xh = const along a leaf:
(1-7)
259
65
More generally, there are on the Poisson manifold (W, A ) atlases of charts (xi>
such that the components of A in the charts are constant. Such a chart is called a
natural chart for the manifold.
r.
ALGEBRA
(a) Let sfbe a Lie algebra of dimension m over 88, d*the dual vector space
of d , , ) the bilinear duality form. Denote by {LA}( A = 1,..., m) a basis for &
and by (A,) the dual basis for d*;
then
<
[LA,LB]= c;BLc
where {C;} is the structure tensor of the Lie algebra d. The Jacobi identity can be
expressed by
scycp = 0
where S is the summation over cyclic permutations of (A, B, C). Denote by
components of 5 E d* and introduce on d*the 2-tensor A defined by
(2-1)
f A
the
(2-2)
We have
[ A , A ]= 0
(2-3)
260
66
BAYEN ET AL.
.$ E d*.
The set of points 4 such that r ( 0 = r is an open submanifold W of d*,
the
complementary set of which in d* is a cone; the restriction of A to W defines on
W a structure of regular Poisson manifold.
(b) Let G be a connected Lie group having d as Lie algebra; G acts naturally
on d* by its coadjoint representation Ad*G. We recall some well-known facts
concerning the Lie algebra ad*& of the group Ad*G. Let a E SP and exp(tu) the one
parameter subgroup of G defined by a. The action of this group on 4 E d*gives
[ ( t ) = (Ad* exp(ta)) 4. It follows from the definition of the coadjoint representation
that, for each b E SP,
(2-4)
We deduce from (2-4) the map a E d I+ "a E ad*&, where #a is the linear vector
field on &* given by
(2-5)
This map is a homomorphism of Lie algebras admitting as kernel the center of d.
The tensor field A on &* is invariant under the Lie algebra ad*&; this is a direct
consequence of the Jacobi identity.
(c) Choose a point
4 of the
67
HE.Therefore M(5,) has even dimension 2r(5b) and the orbits depend only on the
Lie algebra a
' and are independent of the choice of the group G.
Let U be a domain of d* such that M(E0)n U # 0. Introduce a local chart
{fA, 5.1 (a = I, ..., 2r(5,)) of d * with domain CJ such that M ( t 0 ) is locally defined
by f A= 0. For [ E M(t0)A U,we have in this chart AAA(()= 0 and the restriction
of A to M([,) defines a 2-tensor AM(Eo)
of rank 2r(tO)on Meo)with components
(A"}. We deduce from (2-3):
VL4(Cu)
44k")I
=0
3. COMPOSITION
LAW * A
AND
(a) A pat Poisson manifold is a regular Poisson manifold that admits a Poisson
connection without curvature. Let ( W , A ) be such a manifold and define the covariant
derivative V in terms of a Poisson connection without curvature; we introduce the
bidifferential operator Pr, of order r in each argument, defined by the following
expression on the domain U of a chart { x * ) :
*.*
(3- 1)
We set Po(u,v) = uc. For r = I , we obtain the Poisson bracket P, with P'(u, t;) =
1;) is symmetric if r is even, skew symmetric if r is odd.
k t E(N; A) be the space of the formal series in A E C, with coefficients in N. Given
a formal series f(z) with constant Coefficients, such that f ( 0 ) = 1, we substitute
P
' for 'z in the development of ~ ( A z ) ; we obtain a bilinear map (u, v) E N A N - +
u -A u E E(N; A). This map can be extended in a natural way to E(N; A). If
1
.
f(z) =
2 ar(zr/r!),
a, = 1,
r-0
we have
, , *
* A L'
C Ar[ar/r!)Pr(u, c )
(u, c E N ) .
(3-2)
r=O
This defines a format associative deformation of the usual associative algebra defined
on N by u, PI+ uti if we have formally:
(u * A v) *A w = u *A (v 'LA w)
(u, c, w E N )
We obtain immediately:
c A'
m
(u
*A
P) *A w =
t-0
L'), 11')
(3-3)
262
68
BAYEN ET AL.
and
0)
(u * A
10)
C At C
r+a=t
t-0
T,(u,U, W) = &(u,
U, W)
(t = 0, 1,..-)
(3-4)
Tt(u,u, w) =
(ap,/r!s!)
PT(PS(U, u),
w),
rf8-t
V,(u, u, w) =
r+a-t
vjl...j,,Rl...L,u
(r')
SI-0
vj,,+l...j,ll
.z, M'.
(3-5)
(3-6)
ul-,u, = a,,a,,
For z
= 0,
(r'
(3-7)
263
69
from (3-7) by induction that necessarily at = 1 for each t. Conversely, if such is the
case, (3-3) is satisfied. We have thus obtained:
THEOREM
3. r f (W, A) is afrat Poisson manifold, there is onb one formal function
of the Poisson bracket (up to a constant factor and a linear chmge of variable) that
generates a formal deformation of the associative aIgebra defined on N by the usual
product: it is the exponentialfunction.
Symbolically, we can write this deformation, for instance,
u *A v
exp(hP)(u, 0).
(3-8)
sinh(AP)(u, v )
(3-9)
[u, t:] =
C (vr/(2r-+ I)!)
(3-10)
P2rL1(u,v )
r-0
where we h u e set Y = h2. The first term of (3-10) is P(u, u). We see that (3-9) or (3-10)
defines a formal deformation of the Poisson Lie algebra N.
(b) Let (W, A) be any regular Poisson manifold. Consider the Chevalley
cohomology of the corresponding Poisson Lie algebra N ; it is the cohomology,
with values in Nitself, defined by the adjoint representation: ap-cochain is an alternate
p-linear map of N p in N, the O-cochains being identified with the elements of N. The
coboundary of the p-cochain C is the ( p 1)achain 2C given by
ac(uLl * * - - >
up>
= (1/p!)
$ ~ ~ > ~ ~ A .9
c(uA, ,.--,UA)]
UA,]t UAz
?*--*
uAp)
(3-1
where uA,E N. We note that the space of the I-wcycles of N is the space of the derications of N,the space of the exact I-cocycles is the space of the inner derivations.
A pcochain C is called local (support preserving) if, for each ul E N, such that
u1 :V = 0 on a domain U,we have C(u, ,..., u,) Iu = 0. If C is local, aC is local.
We obtain thus the so-called diagonal complex in the terminology of GelfandFuks. A p a h a i n C is called d-differentiable (d 2 1) if the cochain is local and if its
restriction to each domain U of Wis a d-differentiablep-mchainof N(U) = C*(V; R)
in a clear sense. Such a m h a i n is &fined by a multidifferential operator of maximum
264
70
BAYEN ET AL.
f(z) =
2 (ay/(2r+ I)!)
+=o
Z*~+I
(ao = I )
we have:
0:
Y)
C v*(aY/(2r+ I)!)
PPr+I(u,V )
(V
= A').
(3-12)
r-0
We search for functions f such that (3-12) determines a formal deformation of the
Lie algebra N; (3-12) satisfies formally the Jacobi identity if and only if, for t = 1, 2,..,:
Dt
aPsEr.a= 0
r+b-t;r.r)O
where, for u, V , w
N,
THEOREM
5. If (W, A) is spat Poisson manifold, there is only one formal function
of the Poisson bracket (up to a constant factor and a linear change of variable) that
265
71
generates a formaI deformation of the Poisson Lie algebra N of the manifold it is the
sinhfunction. The corresponding deformation is nontrivial evenfor the order 1 .
Thus we obtain only the deformation M(u, v; v = A2) given by (3-9), which is
therefore nontrivial. It is remarkable that for h = %/2, we obtain a bracket given
many years ago by Moyal [5]. We often suppress the mention of the variable v and
call M(u, v ) the Moyal-Vey bracket of the flat Poisson manifold (W, A, I'); this
bracket depends on the choice of the flat connection F. (For W = R2 and = 0,
the unicity of the sine-bracket as a function of P has been noticed in 1964 by C . L.
Mehta).
P'(u, r);"
V2L...,,~ (u, c E N )
(3- 13)
and set
n
zFI
t'
1 ( P / r ! )Pr(u, u).
(3- 14)
r-0
If (3-14) is limited to the order 2, we see that the associative property (3-3) is satisfied
up to the order 3 if and only if Vil,zuis always symmetric, that is if r i s without torsion.
If (3-14) is limited to the order 3, then (3-3) is satisfied up to the order 4 if and only
~ t l , , i , u is completely symmetric, that is if I
' is without forsion and curvature (flat
symplectic connection). In this case (3-14) satisfies (3-3) to all orders.
4. DEFORMATIONS
OF THE POISSON
LIEALGEBRA
OF A SYMPLECTIC
MANIFOLD
(4-1)
If T is a completely symmetric covariant 3-tensor, we have:
(4-2)
Consider the 2-cochains Sr3 defined by:
(4-3)
266
72
BAYEN ET AL.
In the special case when Pis flat, Sr3 coincides with P3. In the general case, we deduce
from * = *
SSF\P(u, v), W) =
On the other hand, we deduce from the properties of the Lie derivative by Xu :
It follows that:
(v e N).
(44)
(u, v e N).
and:
(JS?(JT.) r)Wj
(4-5)
267
DEFORMATION TKEORY AND QUAiNTIZATION, I
73
PROPOSITION
6 . The cohomology 2-class defined by the 2-cocycle Sr3 does not
depend on the choice of the symplectic connection
r.
Consider two symplectic connections r and r' and take for T the 3-tensor
Proof.
defined by the difference between the two connections. We obtain:
(s",t - S,",(U,
u) = A i1j1Ai2kA
'"'"{(P(xu)
T)ilirir ( y ( X u > Ojz,da
+ (g(xu>
(p(xv)Thl~z,sl
r)f1i2i3
+ Ai1j~Aiz'~~/li~f3(.k9(X,)
T)ilizil(P(Xu) T),,,ds.
Therefore we have:
s,,- s,"
a(&
+ $A?-)
(4-7)
(4-8)
where K is a differential operator of order <2.
Proof. Let r' be an arbitrary symplectic connection; @ - S& is exact and,
according to Proposition 4 of Section 3b there is a differential operator C of order <3
such that:
Q3
-9
., =
EC.
-+ .
**-
This tensor defines, by means ofp, a symmetric covariant 3-tensor still denoted by TThe terms of order 3 in aC can be written:
(4-9_)
Let BAD) be defined locally by:
268
74
BAYEN ET AL.
and consider the terms of order 3 in aBT(u,u). It follows from (4-1), (4-2), and (4-6)
that we obtain the two first terms on the right-hand side of (4-9). We see that
Q3 - S$ - aBT is an exact 2-cocycle defined by a bidifferential operator of maximum
order 2. We are led to introduce the symplectic connection such that the difference
- r' is defined by the 3-tensor T. It follows from (4-7) that $ - Sr3 aBT is
an exact 2-cocycle defined also by a bidifferential operator of maximum order 2.
The same is true for Q3- Sr3, and the connection is a solution to our problem.
Conversely we will see in Section 5 that aBT is a bidifferential operator of maximum
order 2 if and only if T = 0. It follows that the symplectic connection satisfying (4-8)
is unique.
THEOREM
8 (Vey). Let ( W ,A ) be a symplecticmanifoldsuch that H3(W;W)
There exists a formal deformation of the Poisson Lie algebra N :
'I)
Q(u, v; v) =
vT
T.=O
(2r
+ I)! QZr+l(u,c)
(u, v E N )
{O).
(4-10)
5. DERIVATIONS
INDEPENDENT
OF v OF A
{Du,
V}
+ {u,Dv} - D{u,
U} = 0
(5-1)
269
DEFORMATION THEORY AND QUANTIZATION, I
75
(5-2)
+ k(Z)F = 0
or
(5-3)
9 ( Z )A = k(Z)(I.
Du
= 9 ( Z )u
4c
W
u~
where Z E L, c E W, and 71 is the symplectic volume element; rhese are nonlocal derications (for c # 0).
We suppose now that W is noncompact. The results concerning the compact case
are similar, since it is possible to prove that the nonlocal derivations do not appear
in the following study.
(b) Consider an infinitesimal Vey deformation of the Poisson Lie algebra of
the symplectic manifold ( W, A )
(5-4)
where Q3 is a 2cocycle satisfying Proposition 7; we have
(5-5)
where is a symplectic connection and where Kis a differential operator of order G2.
An infinitesimal automorphism D-independent of v-of the bracket (5-4)is defined
by an endomorphism D:N + N such that, for every u, v E Nand v E C, we have:
(5-6)
The space of these infinitesimal automorphisms admits, for the natural bracket of
the endomorphisms, a structure of Lie algebra. For v = 0, we obtain aD = 0;
270
76
BAYEN ET AL.
that is, D is a derivation of the Poisson Lie algebra. It follows from the above theorem
that necessarily
= 9(Z)
-+ k ( 2 )
(ZE Lc)
(5-7)
Q3(Du, u)
+ Q3(u,Du) - DQ3(u,U) = 0.
(5-8)
U)
K(u) Iu = Hk'Vk,u
+ JkV,u + CU.
(5-9)
Here c is a constant since aK is null on the constants (as noticed at the end
of Section 3b).
We obtain easily according to the Ricci identity
~ K ( uv)(
, u = di5 ViHkZ(VktuV ~ U V~LVV ~ U-) A"HkZ(VkiuV i j ~- V k i ~VljU)
- A"HkzRl,rk(V,u V ~ V V,V V ~ U )
(5-10)
r.
E ~ ( uU)
,
~ K ( D uV, ) f ~ K ( uDu)
, - DaK(u, v);
that is:
(5-11)
We have the following lemma:
LEMMA
10. The bidzflerential operator Ez dejined by (5-11), where ZELc, is
of maximum order 2 on each argument.
Proof. Computing (5-1 1) with the help of (5-10) and the identity
(P(9
VkV, - V , V , - W ) )
2.4 =
(%TI
D k J
v,u
we see immediately that all terms in Ez of order >2 in each argument cancel out, and
the lemma is proved.
27 1
77
(d) We have
(5-12)
Introduce the symmetric covariant 3-tensor T = U ( Z )r a n d the differential operator
Br corresponding to and T defined in (4-4).It follows from (5-12) that:
a3D
= -2BT
(5-13)
- 4k(Z) Sr3 + Ez
(5-14)
Introduce the symmetric covariant 3-tensor T = U(Z) rand the differential operator
We have at this point:
where we can choose arbitrarily the symmetric 3-tensor 7 related, according to (4-1),
to the third derivatives of u at x. Lemma 10 gives &(u, v)(x) = 0 and we deduce
from (5-14):
79i1fzfa 2k(ZX9(X,)
Ti1i2fa[(Y(x*)
r),,f,f,l(X)
= 0.
9(XJ T
(5- 15)
.Y(X,)T = 0
for every u E N.
Consider an element v e N such that (a,,u)(x)
According to (4-2), we have at x:
(Wi,i,iJ(x)
(5-16)
= 0, (a,u)(x)
being arbitrary.
=0
272
78
BAYEN ET AL.
qz)r = 0.
r.
(5-18)
r,
U ( Z )H = 0
(5-19)
[ 9 ( Z )J , A ] = 0.
(5-20)
and
The vector field 9 ( Z ) J defines a symplectic i.t.
We note that the same argument proves also the uniqueness of the connection defined
by Proposition 7.
We denote by Ls,, the symplectic subalgebra of the Lie algebra of the affine infinitesimal transformations corresponding to the connection
Let L(Q3 be the Lie
algebra of infinitesimal automorphisms (independent of v) of the bracket (5-4) (Vey
infinitesimal deformation). We have proved:
r.
THEOREM
11. The Lie algebra L(Ql) of infinitesimal automorphisms of the bracket
Q1which do not depend on v is a subalgebra of the algebra LSAof the symplectic affine
infinitesimal transformationsof the symplectic connection r associated with the 2-cocycle
Q3. Therefore L(QJ is finite dimensional.
r>
273
79
i([A,[Z, , J ]
+ Z,])(du
dv) = 0:
PROPOSITION
13. A vector field 2, + vZ,preserves an infinitesimal Vey deformation Q,
Z" =
(5-2 1)
V'Z,
r=0
r,
DEPENDING
ON Y OF
6. DERIVATIONS
Consider linear maps D,: N
---f
A VEY
LIE ALGEBRA
D,=
1 V'D,
8-0
(6-1)
(6-2)
(a) Taking the term of (6-2) independent of Y, we see first that Do is a derivation
of the Poisson Lie algebra. There is a Z, = Z E Lc such that
Do
For the term linear in
Y,
= g(Z)
+ k(ZJ
(6-3)
274
80
BAYEN ET AL.
If we set T = 9(Z)
+ 4k(Z)Sr3 - Ez) = 0,
(6-4)
+ QZr+'(~,Dv) - DQ2'+'(u,
0).
(6-5)
(6-6)
Since Q2r+l is null on the constants, we can define a differential operator
order (2r 1) by
(U E
N).
QZ+'
of
(6-7)
(6-7)
(6-8)
For r = 1, aQa = 0 and (6-3) can be written with the above notations:
a(D, - (1/3!) Q",) = 0.
We thus obtain:
PROPOSITION
15. The derivations Dv = Do
by:
Do = y(ZJ,
Di = (1/3!)
Q"Z,
+ p(zi)+ k(Z3
(6-9)
275
81
where S is the summation after cyclic permutation on u, u, w. According to the defination of QF+', we have on the domain U,for Z E L satisfying (6-6):
9
' SQ"+l(Q"+l(~ I , y , li ),I
9
'(l$,.+,Q$'+l)(~,li)l
wU) =
(u,v E A'),
(6-10)
where Y is the symmetrizer in (r,s). It follows from (6-8), (6-9), (6-10) that for Z EL:
1
(2t i l ) !
rz=t
1
(2r + I)! (Zs + I)! a2r+lQy+r. (6-1 1)
+ aQY1) = -
(&+1-W)
r. s>l
We deduce from (6-1 1) that 0,defined by (6-1) satisfies formally (6-2) if and only if
for each t = 1,2,-..:
(6- I 2)
D, =
V'D,
5-0
<t -2
that is:
1
D8 = (2s
+ I)! QYl+
r,;=5
(2r'
1
+ l)! QE? +
g(za)
(6-1 3)
r'.a'>l
where 2, E Lfor s
= 0,
=t
- 1:
(6-14)
where
E LE.
If we introduce (6-13), (6-14)in (6-12), we obtain:
(6-156)
276
82
BAYEN ET AL.
< t - 2:
T>i;R>l
(6-16)
T>%8>1
Therefore:
16.
and (6-14).
(e) We denote by L* the Lie algebra of the globally Hamiltonian vector fields.
It is known that [L,L] = L* and dim LIL* = b,(W), where bl(W) is the first Betti
number of W. Let E(L; v) (resp. E(L*; v)) be the space of the formal series in v with
coefficients in the Lie algebra L (resp. L*).An element 2" of E(L; v) is given by:
m
2, =
C v*z,
(2, E L).
(6-17)
8-0
E(L; v) admits a natural structure of Lie algebra given by the bracket of the vector
fields and [E(L;v), E(L; v)] = E(L*;v).
277
DEFORMATION THEORY AND QUANTIZATION, I
83
It follows from the above proposition that there is an isomorphism from the
space E(L; v) onto the space of the derivations
1
.
D, =
vfDt
(6-1 8)
f-0
defined by p: 2,
---f
D, , with:
(6-19)
If 2 belongs to L, we obtain
and we have:
(6-20)
If 2,E E(L*;v) we obtain an inner derivation of the Vey Lie algebra and if
2,
(6-21)
THEOREM
17. The Lie algebra of the derivations of a Vey Lie algebra defined
by Q is isomorphic with the Lie algebra E(L; V) bj? the following isomorphism
p : 2, E E(L; v) ++D, , where
The Lie algebra of the inner derivations is isomorphic to the Lie algebra E(L; v )
and the first cohomology space W(Q)of the Vey Lie algebra for the Chevalley
cohomology is isomorphic to E(L; v)/E(L*; v). If b,(W) = 0, all the deriuations of
a Vey Lie algebra are inner derivations.
The derivations of associative algebra deformations giving rise to a Vey Lie algebra
coincide with the (local) derivations of that Lie algebra.
278
84
BAYEN ET AL.
7.
C-EQUIVALENCE OF
DEFORMATIONS
OF h
u p = c V+C,(O)(U,v) = {u, v} + 1
W
03
r*o
r-1
VC$?)(U,
v)
(7- 1)
where the CO
; ) (r 2 1) are differentiable 2-cochains on N .
Consider an alternate bilinear map N x N- + E(N; v) which gives a formal series
in V :
c vrC))(u, v)
03
[u, v p =
= {u, u }
r=O
+ 1 v+CF)(u,u)
(7-2)
r-1
where the Cll) (r 2 1) are also differentiable 2-cochains on N. All these cochains
can be extended naturally to E(N; v). We set:
Djl)(u, u, w )
sc:(c~(u,
v),
w).
7+8*t
7.8>0
C v8T, = Id + C vsT,
8-0
=0
for t
= 1,2,
....
(7-3)
5-1
where the T,are differential operators on N; T,acts naturally on E(N; v). We have
also a natural definition of a product T, . T,and of the inverse Ti1of T, . Consider:
We have Fo = 0. We set:
279
85
and we obtain:
(7-4)
If we evaluate ST,[[u,v]:), w]il) in two different ways, we obtain:
LEMMA18. The formal deformation (7-1) being given, for each bilinear map (7-2)
a d for each formal series (7-3), we have the identitiesfor t = 1, 2,...:
D?)(u, 0,
W)
T,Df)(u, t, W )
rts-t
= - a ~ t ( u , 0, W >
+s
+s
r+r=t
~,(cl(u,v>, W >
C?(F,(U, v), w)
r+s-t
+s
+s
{Fr(u, c>,
r t r-t
C?(F~U,
01, T,,w)
(7-5)
(r, s, s 2 I).
++a+d-t
(7-6)
that is Ft = 0 (t = 1,2, ...). We have Cjl = Cjo) aTt - G,. It follows that (7-3)
determines (7-2) satisfying (7-6) in a unique way. For this map (7-2), we deduce from
(7-5):
~ j l ) ( uu,, w >
+ C. T,D$)(u, u, W) = o
rfa-t
PROFWITION
19. The formal deformation (7-1) being given, each formal series
(7-3) generates a unique bilinear map (7-2) satiflying (7-6). This map is a new formal
deformation of the Lie algebra N.
(b) We are led to the following definition.
DEFINmoN 20. Twoformal deformafions of N are called cohomoIogicaI&equivaIent
(c-equivalent) f there is a formal series (7-3) such that the identity (7-6) is formaIly
satisjied.
E(C2 -
cp + G,) = 0.
280
86
BAYEN ET AL.
The deformation [u,u]:" is c-equivalent to the deformation [u, u]:O1 up to the order
(q 1) if the 2-cocycle (CF) - C:") G,) is exact. The cohomology class defined by
this 2-cocycle is the obstruction to the c-equivalence at the order q.
We have GI = 0. An infinitesimal deformation defined by the 2-cocycle Cjl)
is
c-equivalent to the infinitesimal deformation defined by the 2-cocycle Cia) if the
2-cocycle (C:" - Cf")is exact.
We recall that, for the flat case, we have for the local Chevalley cohomology:
dim H 2 ( N ) = 1.
Therefore, at each order q, we have in the flat case only two choices of cochains C, ,
up to c-equivalence of formal deformations (the trivial, and the one given by the
cohomology class 13): altogether, we have at most 2 4 deformations (up to c-equivalence)
at order q, and 2 N =
~ K~ classes of formal deformations. However, at each level there
is only one nontrivial choice to make in this case.
8. GENERAL
DEFORMATIONS
OF THE ASSOCIATIVE
ALGEBRA
N
(A STAR-PRODUCT
IS BORN)
We saw that (3-14) defines a deformation of the associative algebra N of C" functions over a symplectic manifold Wif is without torsion and curvature. It is natural
to ask whether a formula analogous to (410) can be obtained for the associative
algebra when such connections do not exist. Vey proved the existence of a Lie algebra
deformation (4lO)-without giving explicit formulas for QZr+lwhen r > 1-under
the technical assumption H3(W; R) = 0 by tracing at each step the possible obstructions to the continuation of the truncated deformation in the (finite-dimensional)
third differentiable cohomology space. However, in the associative case the obstructions space is isomorphic to the huge functional space of the global sections over W
of the fiber bundle A3TW (skew-symmetric 3-tensor fields): the same method is
therefore impractical, except of course when dim W = 2, in which case the obstructions disappear. As a matter of fact, Vey was able to find cochains Qr satisfying the
associative property modulo terms of higher order in A only up to the order r = 5.
We shall present a procedure for obtaining an associative deformation which we
may write symbolically as exp(hQ)(u, v):
(8-1)
where Ql = P and the cochains Qrare explicitly determined, when the manifold W
can be imbedded in some RZkin a suitable way. We shall illustrate the procedure by
an example [I41that will be needed later for the hydrogen atom problem.
28 1
DEFORMATION THEORY AND QUANTIZATION, I
87
1 T 12 =
- .$ = c 7 r q n = 0.
Pl
1 ( T y = 1,
7r
(8-2)
a-1
Ir=l
Here (ra,5,) are canonical coordinates of T*[W1+land this space is endowed with the
usual symplectic structure fl =
d+ A df, . We shall denote by 1'the canonicaI
flat symplectic connection of I@,so that in the coordinates (T', &,) the covariant
the bidifferential
derivative is nothing but the usual partial derivative, and by
operator defined by (3-l)-the rth power of the Poisson bracket P of T@ in the
canonical coordinates.
Notice that p(&I T 12, T * 511 T 12) = 1;that is, W is defined by a pair of canonically
conjugate constraints. In accordance with the general theory of second-class submanifolds [&, 16a] the canonical symplectic form F of W,induced by p , defines the
Poisson bracket P of W.
We shall now define a "thickening" of W to an open subset of @ by a group G
of symplectic transformations, so that on C-invariant functions P will coincide
with P. Moreover G will be chosen as a subgroup of Sp(21+ 2, R), hence, will be
affine syrnplectic for p, so that the restriction of
to Ginvariant functions will still
be a G-invariant function and thus define a cochain @ on N(W). More precisely,
let G be the following representation of the two-dimensional (non-Abelian) solvable
group in Iw2~+2:
xzl
(T",
5")
(p",
p-'5, i- or"),
> 0,
(3
R.
(8-3)
aG
c$:fiG36*2iIW=~EN,
c $ - 1 : ~ M 6 = U o g .
(8-4)
We have
9 ( 2 ) R= 0 = 9 ( T ) A ,
S?(Z)fi= 0 =
282
88
BAYEN ET AL.
(8-5)
(summation A , B = 1,..., 21
and (8-1) defines in a natural and global way an associative *,-product on N(W),
with a corresponding Vey deformation
m
Q,(u,
V)
=
r-0
(vr/(2r
(8-7)
(8-7)
9. INVARIANT
*-PRODUCTS
(a) Let (W, A) be a Poisson manifold and N = Cm(W,C). For f, g E N,
{f,g> = A(@, dg). Every a E N defines an infinitesimal canonical transformation given
by X,f = { a , f } that preserves the Poisson bracket (Jacobi identity):
In other words
@,f* 8) = hf)* g + f
* (4 g),
tJf, g E N-
(9-1)
283
DEFORMATION THEORY AND QUANTIZATION, I
89
(b) Conversely, let & be any finite-dimensional subalgebra of the Poisson Lie
algebra N. We pose the problem of constructing a *-product on Wthat is &-invariant;
that is, a-invariant for every a E d.This, in our view, is a central problem of quantization, a problem that is discussed at length in the companion article [lo]. RecalI
that t h e Moyal *-product is invariant under the group of affine symplectic transformations, and that conventional quantization realizes this group as a group of
unitary operators. In general, the C functions on W that belong to d C N will be
taken to have significance as a preferred set of physical observables; hence we require
that the Lie algebra .d be preserved by the deformation; precisely (henceforth we
write * for * A )
a * b - b * a = iti{a, b),
Va, b E d
(9-2)
where ih/2 = X is the deformation parameter. In addition, it will be supposed that the
set of observables belonging to d is sufficiently large. Let {LA}( A = I,...y m)
be a basis for .dand let J be the map of Winto d* given by x ++ [(x) where t A ( x )=
LA(x) [17]. The action of the vector field X , in W is now transferred to the image of
J by dJ and this is just the coadjoint action ad*d of d in d*.
The image
M of W by the map J is therefore an ad*&-invariant submanifold of .a*.
The algebra s2 is sufficiently large if this manifold is diffeomorphic to the manifold
of actual interest; this happens if J W + M is locally bijective (if W is the manifold
of interest) or more generally if we are interested in constraint manifolds with constraints in Ker J. In any case we now direct the attention to MC&* and formulate
a precise mathematical problem.
(c) Let d be a Lie algebra of dimension m over R and d* its real dual. We
identify d with the space .d**of linear functionals over d*;in particular this means
that a basis { L A ) ( A =
m) for a2 is also a set of coordinates for d*.A polynomial over d* is a polynomial in L.l,...Lm
y with coefficients in @. We denote by
U(32) the enveloping algebra and by S ( d ) the symmetric algebra of the complex
extension of sd, and by 9 the algebra of polynomials over d * ,naturally isomorphic
to S ( d ) [IS]. The natural Poisson structure on d* was described in Section 2.
Let M be an algebraic submanifold of d * ,invariant under ad*d ,and let I7 denote
the projection of 8,B -+ 9/9
where 4 is the ideal of polynomials in B vanishing
on M . Thus Ker I
7 = 9 = { fS;fl
~ M = 0} and l7(9)= 9/Kerl7. Commutativity of the diagram
284
90
BAYEN ET A t .
DEFINITION
21. An invariant *-product on M is an internal composition law on
n(.9),
writtenf, g ++ f *g, associative and distributive ouer C, such that
k*f=f*k=kf,
Vk E C ; a, b ~ d ,
a * b - b * a = %{a, b},
vj; g E
{ a , f * s> = { a , f >* g + f * {a, g),
n(q.
(9-4)
10. EXAMPLES
OF
INVARIANT *-PRODUCTS
We shall make use of the formula (3-8) to obtain some examples of invariant
*-products.
(a) Let (W,F, I') be a connected, paracompact, symplectic C" manifold of
dimension 2n, fundamental 2-form F, and connection It will be assumed throughout
that the covariant derivatilie V of F vanishes. Let {xi> (i,j = 1 ,..., 2n) be a local chart
with domain U.For u, v E P(W, C) define
r.
(10-1)
with A
ifi/2 and
(10-2)
Recall from Section 3d that this product is associative if and only if the torsion and
the curvature ofl' vanish and suppose that this is the case. Denote by Inv(F, r ) the
subalgebra of the Poisson Lie algebra N = C'-(W,C) that consists of all (I EN,
independent of A, such that (10-1) is a-invariant (Section 9a). This algebra is ;SOmorphic to-and, if I' = 0, equal to-the semidirect product of Sp(2n, R) and the
Heisenberg Lie algebra H , . Of course, its representation by vector fields (considered
in Section 5 ) is the semidirect product of Sp(2n, R) and the Abelian algebra R2,.
r)
285
91
cj-0
J)
( 10-4)
( g c J).
r)to be Rn %
EXAMPLE
22. Let a2 = H , with basis xl,..., x2,, xZn+l= e; M = Im(J) is determined by e = 1 and is a single 2n-dimensional orbit of ad*& in &*. Moreover,
ad*& can be identified with W,J is the identity map, and No = N.
EXAMPLE
23. .Let srl = Sp(2n, W) with basis (xixJ}(i,j = I ,..., 2n); M is a single,
2n-dimensional orbit; No consists of all even functions of xl,..., xZnand is a *-subalgebra of N.
EXAMPLE
24. Let x*,..., x2n = 41,...,4; p, ... p, and a? = SO(n) with basis
ypj - qjpi (i,j = I,,.., n), The cornmutant 93 of d in Inv(F, I) has the basis
gigi, 2 q%i, x p i p i . The cornmutant NB of
in N consists of functions of Lij
and therefore it coincides with N o . It follows (see the Lemma below) that No is a
*-subalgebra of N . In this case M contains many orbits.
Lij
This last example has a number of easy generalizations; thus one may introduce
a symmetric, nondegenerate form g on R and take for 9 the span of g(q, q), g(q, p ) ,
g ( p , p ) , obtaining an invariant *-product on ME&*, where d is the pseudoorthogonal algebra defined by g. The success of these constructions depends on
LEMMA25. Let 9 be any subalgebra of Inv(F, r)and N g the cornmutant of 3
in N; then iVa is a *-subalgebra of N .
Proof: If u, v E N9 andb E
so that
I(
*V
* + u*{b, u> = 0
E N ~ .
L/(X)
= p-l(n.
(10-5)
= L;yX) ...
1
q(x)
g(x),
(1 0-6)
286
92
BAYEN ET AL.
p . c = L&B((Lik/pL,C,
- pkLk=)
( 10-8)
(10-9)
The *-product f * g for ft g E C(M, 6 ) was defined by (10-4); direct calculation
give
f*g =
(hk/k!)fAAkg,,...,k
(10-10)
k-0
where
fAi*-.Ak
- {LA;
=
{.-;
{LA,f} -..}
(10-11)
(d) Conversely, we may pose the question of finding necessary and sufficient
conditions on 2 in order that (10-lo), (10-11) define an associative, invariant
287
93
is completely symmetric
(10-12)
(10-13)
The last equation is the condition for (LA;dsc) to vanish; hence (M,AYf) is a
symplectic manifold with fundamental 2-tensor A , and pseudoconnection fi such
that the contravariant derivative of vanishes. Moreover, r is d-invariant by hypothesis and therefore f is also d-invariant, which is expressed by
{ L A , fC.}
- ,;BpC.D
- C;CfrBE.D
- ,jD,C.E
= 0.
(10-14)
XC}}
-{LB; {LA;
XC}} =
AMylD
(10-105)
PROPOSITION 26. The *-product dhfined 6y (10-lo), (10-11) is d-inaarianr and
associative if and only ifthe pseudo connection f satisfies (10-12)-(10-15).
f ABC
= {LC;{ L B ; {LA,f}}} = 0
iffis linear,
(10-16)
which means that the series that defines L A * g ends with the third term. If g is a polynomial in L1,...,Lm (more precisely, g ~n(9),
in the terminology of Section 9),
LA
* g = LAg + A{LA, g } + $ h y f = w C
where
KAB
+ KADaD)g
= -Lc D AL CDB.
(10-17)
(10-18 )
In particular,
LA
L B
= LALB f
AAAB
+ iA2KAB.
(10-19)
288
94
BAYEN ET AL.
PROPOSITION
27. (1) K A B is constant on each orbit; that is, {LA,K E C }
3-tensor KADCiCis completely skew symmetric [19].
0. (2) The
PROPOSITION
28. If f satisfies (10-12)-(10-15), then (10-17) is compatible with
a unique invariant *-product on M ynamely, that given by (10-101, (10-1 1).
EXAMPLE
29. Let W = IW2n,
and of A = p-'(F) are related by
= 0,
A"Fkj
82.
(10-20)
x,= XjF,,.
xi = LPX5 ;
(10-21)
= n(2n
+ 1)
zy = - B-1
c gA BzjBi
and normalize so that
z$zBklgAB= 8ikc8j2 + 82i 8:,
Z$Bi5
= 2gAB.
(10-22)
(10-23)
(10-24)
289
95
+ x'x'),
Lz= X'X',
L3 = +(x'x'
- x'x'),
(10-25)
(10-26)
((10-27)
L A
Perm
* LB = LALB + P g " B ,
(10-30)
(10-30)
(10-31)
11. CONSTRUCTION
OF INVARIANT
*-PRODUCTS
@(I) = 1,
@(a) = a/%,
gZ 0 ad&) = add(a) @,
Ker @ = bilateral ideal of V ( d ) .
0
(11-1)
(11-1)
Comersely, any C-linear map 0 of U ( d )onto B that satisfies (11-1) definesa *-product
on d* given by
f * g = @O(@,Yf)
@Ol(g)),
tlf, g E 9
(11-2)
290
96
BAYEN !3 AL.
EXAMPLE
31. The canonical bijection between B and U ( 4 ,together with a change
of scale, gives a linear map defined by @(an)= (a@)", or a * * Q = an, Va E d .
This is the quantization rule proposed by Abellanas and Martinez-AIonso [20].
.--
EXAMPLE
32. More generally, for k = 2, 3,..., let
(11-3)
A *-product on a'* will be called regular if @ is onto. More generally, let M be any
algebraic ad*&-invariant submanifold of d*,and fl(9') the associated projection
of 9. To any invariant *-product on M we associate a map Y (with properties
analogous to (11-1)).
DEFINITION
33. An invariant *-product on M is called regular if the map
= 2,
3,....
DEHNKION
34. An invariant *-product on M' is said t o be compatible with IT
if Ker 17 is *-invanant; that is, iff E Ker 17 implies that a * f E Ker 17,Va E d.
If an invariant *-product on M' is compatible with D, then commutativity of the
following diagram induces an invariant *-product on N :
Let Z(P) denote the center of the Poisson Lie algebra 9' and let Q denote the set
of homomorphisms from Z(P) to C that reduce to the identity on the scalars. For
any a E %',the family of equations q - r ( q ) = 0, Vq E Z ( q , defines an ad*,
invariant submanifold M,. If riff is the projection of B defined by M,, we have
fl, I Z(@)= v . We now restrict ourselves to projections of this type.
DEFINITION
35. Let R and R' be commutative rings and $: R -+ R' a ring homomorphism. Let A be an R-module, A' an R' module and #: A -+ A' any map that
29 1
97
preserves addition: Y(a, a2) = Y(a,) fY(a,>. Then Y is called [22] semilinear
(relatively to $), in short $-linear, if Y ( r a ) = $(r) Y(a), for all r E R and a E A .
Proof. Let q E Z(P) and f E 8 ; q - p(q) E K e r n , . Since the *-product is compatible with l7, we have (q - dq)) *f E Kern, and thus Un(q*f) = .rr(q)nn(flThus for any 4 E Z ( d ) and
U(ss), Y(4f) = f l n 0 (43) = f l n ( # 4 ) @(!)I =
3.
*(@
WP).
n,,
, VTTE V ;
(i) The *-product is compatible with
(ii) @ is +linear [+ zz 0 I Z ( d ) ] ;
(iii) q * f = qf, V q E Z(P), Vf E 9.
(1 1-5)
DEFINITION
38. A regular, invariant *-product on .d* that satisfies any one
(and therefore all) of the conditions (1 1-5) is called normal. The associated map gZ
will also be called normal in this case; hence any map ds: U ( d ) 9 is normal if
it is onto, +-linear, and satisfies (11-1).
(c) Let an invariant *-product on M, be given and consider the question of its
lifting to d * ;the following gives a partial answer that covers the most interesting
cases:
WEOREM
39. Let an invariant, regular *-product on M, be given, with the associated
supposed +linearporn the Z(d)-moduIe lJ(sa>to the @-module
map F U ( 4 l7=(P)
n,(q ($ = ?P1 Z ( 4 ) . Suppose that either a= H,, , the Heisenberg algebra, or
that .d is semisimple and that the ideal Ker Y is generated by Ker $. Then there exists
a m r d *-product on A?* with associated map 0:U ( d ) -+ 9 (boective in the semisimple case) such that 17,o @ = Y.
--f
Prmf. In both cases U ( d ) and B are free modules over Z ( d ) and Z(9) (respectively); Z ( d ) is a polynomial algebra in a finite number of Casimir elements
g= E Z ( d )1181. We denote by @, the canonical bijection U ( d ) 9obtained from
the canonical bijedon U(@ + S(d) by rescaling, and by its restriction to Z(d).
we give a proof for the semisimple case.
292
98
BAYEN ET AL.
(i) Construction of
= @ I Z(4. We need an algebra isomorphism
Z(P) such that 7r 0 4 = 4. To show that one exists let +o(&) = qe and
fl4J = qu - dq,) #(&) = qa h, = q i with X, in @. We extend 4 as a (polynomiaI) ring isomorphism Z ( d ) = @[gal
Z P )= C[q,l.
4: ZW)
--.f
xi
-+
Let S mean symmetrization with respect to the order of factors and take, for
k , I , m = 0 , 1 , 2 ,...:
(11-7)
where {Ck*mj
are polynomials in p , q, e, with Coo0 = 1, C1W = p/ifi, Cola = q/8,
Coo= el%. This *-product is &-invariant if and only if the following recursion
relation s hold :
(1 1-8)
The center of U W )is generated by e; hence (1 1-7) is +linear if and only if e *f = ef,
Vf E 3;that is, Cklm= (elifz)m Cklwhere Ckzis a polynomial in p , 4. The necessary
293
DEFORMATION THEORY AND QUANTIZATION, I
99
and sufficient conditions for (1 1-7) to be both invariant and +linear can be expressed
by 1231
(11-9)
where C(a, is a formal power series in a,/3 and the equality is to be interpreted
in the sense of equality of formal power series in a, p. Comparing the constant and
linear terms in (11-9) one sees that C(a,j?) must have constant term = 1 and no linear
terms, so that
1
where the unwritten terms are monomials inp, q of order <k inp, <l in 4,and <k
in p. q. Therefore (11-9) is regular and thus normal. Thus one finds, in the case
= HL, that invariance and +-linearity imply regularity. In the case C(a, /3) = 1
we recover the ordinary Moyal product; that special case is characterized by the
natural invariance under the semidirect sum Sp(2, R) H I .
LALBga.
(11-11)
(11-13)
Here S means symmetrization with respect to permutation of the indices. Let {Pu(u)}
be the solid spherical polynomials deiined by
(11-14)
294
100
BAYEN ET AL.
and let fa@) be the *-polynomial obtained from Pu(u) by the substitutions
LA1 L A k --t L A 1 * * LA*.The map Qi must take the form
---
- a -
P&)
with c, E C
- (01,a = 1, 2, ..., c,
c,P,(u),
Vu E d,
1. Taking a
= 2,
+ +(Q -
SLA * LB = c2LALB
cZQ)
(11-15)
3 we find in particular:
gAB,
(11-16)
These results reduce to (10-30), (10-31), when c2 = c, = 1 and Q, Q take the values
(10-28), (10-29). If, for the same values of Q,8, one takes c, = 1, a = 1, 2,..., then
the *-product given by (11-15) reduces to that calculated in Section 10: Eqs. (lo-IO),
(10-11) with (10-8), (10-9), and (10-27).
(d) The results obtained here may be summarized as follows. The definition
(11-3), with cg = 1 and T = 0,defines an invariant *-product on &
inI
the
*simplest
possible way. However, this *-product is not compatible with the projection to an
invariant submanifold of d*,
and it becomes interesting to calculate the required
modifications; that is, to find ck and T such that the *-product becomes compatible
with the projection. Corollary 37 solves this problem in the case of invariant s u b
manifolds M,, defined by fixing Z(9). Equation (1 1-15) gives the general form of the
solution for the case d = SO(2, 1). Unfortunately, these results are incomplete,
since the set (M,,} I.( E does not include all the orbits of ad*& in a'*.
DEFINITION
42. An orbit M of ad*& in d is called regular if there is a
such that M is open in M,,, otherwise exceptional.
TE
The term exceptional is justified by the following fact. Let .$ E d*and M(0 be
the orbit of ad*& through f . Then the set of all f E d*such that M ( 0 is regular is
open in .a?* and its complement is a cone. This is not to say that exceptional orbits
are uninteresting; on the contrary, the construction of invariant *-products on excep
tional orbits presents a challenging and physically interesting problem.
12. *-EQONENTIALS
Here we attempt to generalize, to any Lie algebra, the ideas that underlie the Weyl
correspondence [2], a tool that played an important role in the discovery of the Moyal
bracket. We begin with a brief review of the simplest case ( a restatement of the usual
Weyl correspondence).
295
DEFORMATION THEORY AND QUANTIZATION, I
101
e(u~+g*+vb)IffiF(a,
p, y ) dol d/3 dy.
(12-1)
Under suitable restrictions on the class ($} of functions and the class {F)of distributions, this is a bijection.'
On the other hand, let 4',p', e', be operators in L2(-co, +a)defined by
There is a common domain of analytic vectors on which q', p', e' are essentially selfadjoint, allowing us to define a group of unitary operators
(12-2)
(the last integral being an operator integral with the strong topology). Under suitable
restrictions, (12-1) and (12-4) establish a bijection A ~f between a class {A} of
operators in L2 and a class (f}of functions on d*;this is the Weyl correspondence.
Now let A, B,..., be operators in L2and A,, B, ,...,their images under the Weyl
correspondence. The Moyal *-product is then defined by
* B w = (AB)w.
(12-5)
Aw * Bw = Awe(i'/2'dBW
(12-6)
Aw
I n particular, let
296
102
BAYEN ET AL.
cr; g>.
The result (12-6) is easily generalized to a large class of functions, including all polynomials on d**
The essential point here is that the functions (U(o1,p, Y ) ) form
~
a group under
*-multiplication that is isomorphic to the operator group (12-3). We now investigate
invariant *-products on any Lie algebra from this point of view.
(b) Consider the formal group [24] with Lie algebra d.For a E d,let {exp(tu)}
t) a)-
(12-7)
Let 8 be the extension of B to the algebra of formal power series over &*.
DEFINITION
43. We call *-exponential the function Exp: d
defined by
m
--f
9,u ~ + Exp(a),
(12-8)
(a * ) = a *
--- * a
(n factors).
(12-9)
PROPOSITION
44. The function Exp enjoys the foIlowing properties:
(12-10)
(12-11)
where add is the extension of the adjoint action of I in I toformalpower series over d.
(iii) Exp(u) = 1
+ a/% + .-(12-12)
where T(,,)
is an n-linear map of d into 9.
Proof. (i) This follows from (12-7) and Eq. (9-2):
(a 6
Va, b E d.
(12-13)
297
103
(12-14)
PROPOSITTON
45. Let f: d
---f
(12-15)
m
(12-16)
where T(n)is an n-linear map of A? into 9. Then there is a unique @-linear map
@: U ( d )-+B, given by
(a * >"
T(,,l(a,..., a).
(12-17)
Proof. From (12-15) and (12-16) we conclude that 4j commutes with add and
that @(l) = 1, @(a)= a/iti; therefore, if Ker @ is a bilateral ideal of V ( d ) , we have
an invariant *-product by Proposition 30. Substituting (12-17) into (12-16), we see
that f coincides with the function Exp defined by (12-8).
Remarks. (1) The condition that Ker @ is a bilateral ideal of U ( d ) is necessary
as well as sufficient. (2) Any function f: d 3 9" that satisfies the conditions of
Proposition 45, with the exception of (12-15), defines an associative *-product on B.
E ~ ~ ~ 46.
P L Let
E d be. the Heisenberg algebra with basis XI...,
, Jczn+l = ql,...,qn;
pl,-.-,pn;e with [ q i , p j ] = W e , and the other commutators equal to zero. Let
g1,...,p" + Ql, ...,Pn = y l ,...,y2n be a canonical transformation and define
Exp(oriq* + pipi + ye) = e(a,V'+BtP'+ve)/i~.
= ih/2 and
it is &-invariant if and only if yr - xi:are constants and then it is an invanant *-product on d*.
298
104
BAYEN ET AL.
v q E z ( q , VfEP.
(12-18)
* Exp(b) = q Exp(b),
Vq E Z ( q , Vb E d .
(12-19)
We shall show that the structure equations (12-10) allow us to reduce these equations
to partial differential equations for Exp. First, an example.
EXAMPLE
47. Suppose that d has a nontrivial center do;then any rr E V defines,
by restriction, a linear map cr: do+ @. [This map defines a submanifold Muof d*;
the following applies to the problem of construction of a *-product on either M,
or Mu.]Let a normal *-product on d* be given, and let Exp be the associated *-exponential. Then (12-18) holds and implies that
U,
* Exp(b) = a, Exp(b),
VU,, E 54, , Vb E d .
Vu,
do, Vb E at'.
(12-20)
= ifi77" Exp(b)
(12-21)
It4
[The tangent space at any point b E Jaz is canonically identified with d.1 We have
'
9
ID
=a
+ t[aa,b1 f
i%
[[a,b1, b] f--'
299
105
where x is the linear operator xa = -add(b)a and the coefficientsB,, are the Bernoulli
numbers. The identity
B,xR = x/(l - e-.) between formal power series can
be used to evaluate qa. For example, if add@) is diagonable and A' = @A EAis the
eigenspace decomposition and (PA}the associated projection operators, then we have
x =
C AP, and
zz4
(12-22)
then
1 ~is a~ family
) of vector fields over d and we have
(12-23)
Let Q be any element of the center of the formal *-product algebra of the form
Q = Q AI...A,LA'*
*--
* LA-
A1
(12-24)
in Z(P) is also
(12-25)
PROPOSITION
48. Let a normal *-product on A'* be given, and let Exp be the
associated *-exponential. Let Q be any element of the center of the formal *-product
algebra of theform (12-24). Then Exp satisfies thefollowingpartial dyerential equation:
(12-26)
THEOREM
49. L e t f : A'
(i)
W)
(iii)
(iv)
form
--f
f ( b ) = (a,f(b)).
a/ih i- =
(I/n!)(ih)-nT(,,)(a,..., a).
The C-linear map @: U ( d ) + B defined by (a *)* = T(,,)(a,...7a) is bijective.
For every element Q of the center of the formal *-product algebra of the
ad..&)
f(u) = I
p=
& A x...A.LA1
1
.
-
zy=,,
* .--* L A - ,
Then @ is normal and defines a unique invariant *-product on A'*, compatible with
D=for every z E %,' and f is the associatedfunction Exp.
300
106
BAYEN ET AL.
. EXAMPLE
50. Let d = SO(2,l) and adopt the notations of Example 41. Let
b = bALAE d and let 6denote the matrix of -ada(b). We have
bAB= C;'bc ,
t2
(12-27)
or
This equation is of course add invariant and can be separated by introducing pseudcspherical coordinates: t , z = cosh 8 = b(t2Q)-112 and a polar angle. The invariance
property of Exp, namely, Eq. (12-1 l), means that Exp does not depend on the polar
angle and can be expanded in terms of Legendre functions; formally Exp(b) =
CaC',(f)P,(z). Now Exp(b) must be interpreted as a formal power series in { b ~ )
as well as in (LA>;we therefore restrict the summation to values of a E C such that
each term is a formal power series; that is
(12-28)
Here P,(b) is the "solid spherical polynomial" defined by (11-13) and (11-14); it
diEers from the usual Legendre functions by a factor t a p 1 2 and is a polynomial in
both {bA]and {LA].Of course, (12-28) must be interpreted as an equality of formal
power series. For C&) we find
pt2
2
+tan t
at
sin2 t
')
(i?i)-Z
Q] tac,,(t) = 0.
(12-29)
301
107
P,(N
(2a
3
- l)!! ba.
(12-30)
a!
Therefore, the only acceptable solution of (12-29) is the one that is regular at t = 0:
(12-31)
Q = (2fi)Z Z(I
+ 1).
(12-32)
(12-33)
In Part TI [lo] we shall rediscover this formula and obtain from it the spectrum of the
harmonic oscillator.
I n order to show the connection between (12-28) and (11-15) in more detail we
give explicitly all terms up to the third order in {bA}in Eq. (12-28):
302
108
BAYEN ET AL.
the value of Exp(a) at f E sd*. When d is Abelian and the *-product is normal,
then this reduces to the definition of the ordinary Fourier transform. In general, one
has a novel and possibly interesting notion of Fourier transform that relates f on
d to f on d*.We shall not approach the problem of determining the precise conditions under which (12-34) makes sense, but limit ourselves to some formal observations.
One has
Here and below c stands for the Campbell-HausdoriTfunction C(Q, 6). If the *-product
is normal we can restrictfand g to M,, and integrate to obtain a type of Plancherel
formula:
(12-35)
jf* g dp(0
= 0)
and
J ~ ~ e dA.
(1 2-36)
(12-36)
One can define the spectrum 2 of a to be the support of mA and the multiplicity
of a discrete point h E 2 by the integral of mAover phase space. This idea finds support
in several particular instances that are investigated in Part I1 [lo].
Let a faithful, unitary representation of the group SO(2, 1) be given, together with
a map of the type of the Weyl correspondence, that maps operators to B: A F+Aw .
A *-product is defined on 9 by A * Bw = (A& . Let H be the operator that represents the generator t l of the compact subgroup of SO(2, I), so that Hw = L1and
(etHpw= Exp(tL1). Now, since the spectrum of H/2fi consists of integers, we must
have Exp(mL1) = 1 for this *-product. Examination of (12-31) reveals that the period
of each term in (12-28) is 27m, where n is the smallest integer such that 2nZ is integer.
This indicates that the *-products that were found for SO(2, 1) are related to faithful
representations of SO(2,l) or to coverings of SO(2, 1). In particular, it is known that
(12-33) is related to a representation of the fourfold (metaplectic) covering of SO(2, 1);
that is, the twofold covering of SL(2, R).
303
DEFORMATION THEORY AND QUANTIZAnON, I
109
ACKNOWLEDGMENTS
We thank Professor J. Schwinger for helpful advice and Dr. D. Amal and Professor R. Ra&
for interesting discussions. One of the authors (M.F.)thanks the Physics Department, UCLA. for
hospitality. Another author (C.F.)thanks luniversitk de Dijon for hospitality received during the
early part of this collaboration.
REFERENCES
1. E. P. WIGNER,Phys. Rev. 40 (1932), 749.
2. H. WEn, Gruppentheorie und Quantenmechanik, Hirzel-Verlag, Leipzig, 1928; 2. P h y ~
46 (1927), 1.
3. G. S . AGARWAL
AND E. WOLF,
Phys. Rev. D 2 (1970), 2161.
4. K. CHILnr, J. Math. Phys. 17 (1976), 859.
5. J. E. MOYAL, Proc. Cambridge PhiI. Soc. 45 (1949), 99. See also E. A. Render, Ann. Phys. (N.Y.)
95 (1975), 455.
6. M. FLATO,
A. LICHNEROWICZ,
AND D. STERNHEIMER,
(a) C. R. Acad. Sci. Pans Ser. A 279 (1974),
877; (b) Compositio Math. 31 (1975). 47; (c) J. Math. Phys. 17 (1976), 1754.
Ann. Math. 79 (1964), 59.
7. M. GERSTENHABER,
8. J. VEY, Comment. Math. Helvet. 50 (1979, 421.
9. M. FLATO,
A. LrcHNeRowxz, AND D. STERNHEIMER,
C. R. Acad. Sci. Paris Ser. A 283 (1976), 19.
10. F. BAYEN,M. FLATO,C. FRONSDAL,
A. LICHNEROWICZ.
AND D. STERNHEIMER,
Deformation
theory and quantization. 11. Physical applications, Ann. Phys. (N.Y.)111 (1978), 111.
11. A. NUENHLJIS,
Indag. Math. 17 (1955). 390.
12. A. LICHNEROWICZ,
Les Varittts de Poisson et leurs Algkbres de Lie associks. J. Differential
Geometry, to appear.
13. A. Avez AND A. LICHNEROWICZ,
C. R. Acad. Sci. Paris Ser. A 275 (1972). 113; A. AVEZ, A.
LICHNEROWICZ,
AND A. DIAZ-MIRANDA,
J. DifferentialGeometry 9 (1974), I.
14. F. BAYEN,
M. FLATO,C. FRONSDAL,
A. LICHNEROWICZ,
AND D. STERNHEIMER,
Lett. Math. Phys.
I, 521 (1977).
15. P. A. M. DIRAC,
Lectures on Quantum Mechanics, Belfer Graduate School of Sciences
Monograph Series No. 2, Yeshiva Univ., New York, 1964.
16. A. LICHNEROWICZ,
(a) C. R. Acad. Sci. Paris Ser. R 280 (1976), 523. (b) Leu. Mafh. Phys. 2(2)
(1977).
Structures des Systkmes Dynarniques, Duncd.
17. This map was introduced by J. M. SOURIAU,
Paris. 1970.
18. For informationabout enveloping algebras, see J. Dixmier, Algebres Enveloppantes, GauthierV a h , Paris, 1974.
19. In the examples of *-products for semisimple algebras that have been calculated so far, K turned
Out to be a multiple of the Killing form.
20. L.ABELLANM
AND L. MARTINEZ-ALONSO,
J. Math. Phys. 17 (1976), 1363. See also M. Vergne.
Bull. Soc. Math. France 100 (1972), 301. When at = H, (with basis q1 - p I , e), the canonical
bijection between B and U(& defines a bijection Yobetween the quotient of U ( 4 by the ideal
generated by (e - 1)-a Weyl algebra [18]-and the quotient of 9by the corresponding ideal;
in the Heisenberg representation of d , Y;*gives the Weyl quantization mle. Other bijections,
delined with different vector space bases of U ( 4 , correspond to different ordexings [3] (standard.
no-4
etc.). This is one of the motivations for the restriction made from the next subsection
onward to projections defined by characters of the center of the Poisson Lie algebra 9.
21. We have found some invariant *-products that are not regular; they are related to h i t d i m e n sional representations of the algebra and are of interest on compact symplectic manifolds.
22. N.W AAlgtibre,
K I , Chap. II, Sect. 1, No. 13,3rd ed., 1962 or new edition, 1970, Hermann.
Paris.
304
110
BAYEN ET AL.
23. The general form (11-9) for the Exp of an Hl-invariant, normal *-product corresponds exactly
to the general Bassociation 1awconsidered by Agarwal and Wolf [3]. The result is easily
extended to H,.
24. E. B. DY,
Amer. Math. SOC.Truml. 97 (1953). pspehi Mat. Nauk (N.S.) 5, No. 1 (35)
(1950),135]. See also N. JacobSon,Lie Algebras, Interscience, New York, 1962; G. Hochschild,
The Structure of Lie Groups, Holden-Day, San Francisco, 1966.
305
ANNALS OF PHYSICS
M. FLATO~
AND C. FRONSDAL
University of California, Los Angeles, California 90024
AND
A. LICHNEROWICZ
AND D.
STERNHEIMER
In the preceding paper general deformations of the structures based on the classical
symplectic manifolds were examined. Quantization can be understood as a deformation
of the algebra of observables without any need for introducing a Hilbert space. By a slight
but crucial restatement of the usual interpretation of classical mechanics we find a framework
for the description of both classical and quantum mechanics, within which the continuity
of the quantization process is brought out. The spectra of some important physical observables are determined by direct phase space methods; this helps support the belief that
a complete and autonomous theory, equivalent to ordinary quantum mechanics in special
e,
but capable of wide generalization, can be constructed.
A. GENERAL
PKysrcAL CONSIDERATIONS
1. Contractions and Deformations in Physics
306
112
BAYEN ET AL.
to contraction: one determines, in a precise mathematical sense, all groups that are
close to the Galilei group and finds the PoincarC group (among a small number of
possibilities). In this sense, relativity theory is a deformation of nonrelativistic physics.
Another instance of contraction is the passage from quantum theory to the classical
limit fi + 0 (contraction parameter fi = Plancks constant). Here the inverse process
of deformation is nothing but the general problem of quantization. The radical change
in interpretation that accompanies the passage to quantum theory might discourage
attempts to apply the concepts and techniques of deformation theory; nevertheless
it is our aim to show that it can be done.
(b) The basic mathematical structures of classical mechanics are the symplectic
structures attached to phase space: the algebra ( N ) of Cwfunctions on phase space
(W) under ordinary multiplication of functions, and the Lie algebra structure induced
on N by the Poisson bracket that is defined by the symplectic form ( F ) on W. In the
preceding paper [3] we have examined the formal differentiable deformations of these
algebras; in particular, we have introduced an associative algebra (*-product algebra
or *-algebra for brevity) that is a deformation of the algebra N of functions with
ordinary multiplication. For ft g E N = CO-(W7C) we write the new (deformed)
composition law on N as
g) ~ f g. The
* corresponding Lie algebra defined by
(Ag) I+ [f*g] = (f*g - g * f ) / i h is a deformation of the Poisson Lie algebra.
A particular instance of this type of deformation of classical mechanics is familiar
and is known as the Moyal product and associated Moyal bracket.
It is our intention to demonstrate that quantum mechanics can be replaced by a
deformation of classical mechanics: a description of quantum phenomena in terms
of ordinary functions on phase space, including a complete and autonomous physical
interpretation. Naturally, this alternative formulation of quantum theory will include
some features that are not usually associated with phase space. In order t o know what
to expect it is worth while to recall the elements of the theory of Weyl [4], Wigner [5],
and Moyal [6]. See also the preceding paper [3]. (I, Section 12a).
u,
Weyl correspondence-between a
large class of operators in Hilbert space and a large family JV of functions or distri-
307
113
butions on phase space 171. The map that sends Aw to A is called the Weyiappfication
and denoted Q; one has
Q: f +f =
If@, 3) -.-.
++
e(f'P+n.O)Iifi
d35 d31)
(2-2)
a(
= v*g1= C f * g - g * f ) / i f i .
(2-6)
(2-6)
It is this bracket, and not the Poisson bracket, that corresponds to the quantum
, = (i/i+i)[f,ilW.
commutator: ~ fgjM
The fact that one has to do with a formal deformation of classical mechanics,
with Planck's constant +i playing the role of deformation parameter, is brought out
b y Es. (2-4).
Other rules of association (other orderings) between functions and operators have
been considered in the literature (see e.g., Agarwal and Wolf, Ref. 7). They are of the
type (2.2), with the difference that the measure d5 dv is multiplied by a weight function
c @T&*r,
co
-n, = 1 -t
+-1
308
114
BAYEN ET AL.
where upris a homogeneous polynomial; we shall call them Q,-laws. In the orderings
usually considered (standard, normal, etc.) Ol= exp(tiw,), with o2a homogeneous
quadratic polynomial. Its Fourier transform will thus be a formal series
a
where the T, are differential operators. If the functions f and g give the operators
under the O,-law, we shall denote by f * g the inverse Q1-image o f f i . Then
we have T,f = fw and therefore T,( f * g) = Tfif * Trig. Thus T,,realizes the cequivalence (in the sense of Definition 20 of I) between the Moyal bracket and the
O,-bracket [f* g], and similarly between the two associative algebra deformations.
All orderings correspond therefore to cohomologically equivalent deformations,
and any treatment done in one ordering can be translated into another ordering via
this equivalence. In this sense the two formalisms are equivalent, though the same
classical function will in general correspond to operators of different form in the
different orderings.
f and j
(b) Many problems of quantum physics have already been translated into
the Moyal idiom [8]. Our point of view is different: we want to make the classical
formulation autonomous in order to open up a vast field of generalizations with all
kinds of interesting applications (see Section 5). For this reason it is necessary to
deal with some problems of interpretation. We begin with an examination of the
Moyal equations of motion.
Letf be any operator without explicit time dependence, A the Hamiltonian operator,
and f, H the corresponding functions on phase space. The Schrodinger equation of
motion for f is translated by the WeyI correspondence into the Moyal equations of
motion forf, namely,
(2-7)
The Moyal bracket defines a derivation of the *-product algebra; that is,
(2-8)
whence
d
dt
d
d
cr * g ) = (z
f) * g + f * (x
g).
(2-9)
[The significance of the exception made for polynomials of order <2 is discussed
309
DEFORMATION THEORY AND QUANTIZATION. I1
115
in Section 3b.l In order words, one solves the Moyal equations of motion (2-7)
forf; g, and fg and discovers that, in general,
CTa4 f fct)do.
(2- 1 1)
The impression that this result is paradoxical must be dispelled by a proper interpretation.
Every student of the Hamiltonian formulation of classical mechanics has to overcome the difficulty that is posed by the dual interpretation of the ps and the qs.
On the one hand, these quantities are elements of the Poisson Lie algebra of functions
on phase space; this gives a meaning to the Poisson brackets {q,pI) = a j i , etc. On
the other hand, the solutions of the equations of motion are trajectories in phase
space, described by an application t E 88 H (qi((t),pi(t))E R6,and no sense can be
attached to the bracket {qi(t),pj(t)).The error is more serious than just the usual
failure to distinguish between function and value, for there is here an unfortunate
confusion between observable and state. We shall sketch a formulation of classical
mechanics that makes a fundamental distinction between the ps and the qs as
observables (elements of N ) and the coordinates of points on a trajectory in phase
space. It will be seen that this in effect prepares the way: the shock of a complete
reinterpretation of the process of measurement that usually attends quantization is
softened and the continuous passage that is implied by the use of deformation theory
is brought out. The idea is to isolate the two principal elements of the theory, equations
of motion and initial conditions, from each other and to associate the former with
observables and the latter with states.
The Hamiltonian equation of motion for an observablef,
(d/dt)f = -{H,f),
f E N,
(2-12)
defines a map of R
(dldt)f(t) = - { f f , f ( t ) )
since f c t ) is a function on W. The image of the map 0,f)
-.f(t) is a trajectory in
N through$ In particular, q*(t)and p i ( ? )are functions on W and there is not yet any
reference to trajectories in W.The entire discussion is in terms of functions on phase
space and is not directly concerned with values f ( t ) ( p , q) offct) at any Point P. 4
in W.
TO any set of initial conditions one can associate a real (pseudoprobability) distribution p on phase space, normalized so that
p d3p daq = 1.
(2-1 4)
310
116
BAYEN ET AL.
(f>t
(2-15)
EXAMPLE
1. Take H = p2/2m,f = q2; Eq. (2-12) gives the trajectory in N given
+ tjj/mj2.With p = S3( j - j)S3(q - B) one gets ( f ) t = (T + rF/m)2,
byf(t) = (4
This interpretation lends itself to statistical generalizations though it remains a m pletely deterministic as long as the first-order equation of motion (2-12) is retained,
Determinism, in the usual narrow sense, is lost when (2-12) is replaced by the nonlocal
equation of motion (2-7). It is then that phenomena such as spreading of the wave
packet appear [9].
We return now to the discussion of (2-10). In classical mechanics we are dealing
with the algebra N of functions on phase space with ordinary multiplication of functions. The Poisson bracket { H ,f} defines a derivation of this algebra and therefore
so does the time development determined by the equations of motion. In addition,
it is clear that (fg)(t) must perforce be equal to f ( t ) g(t) as long as one is dealing with
a single trajectory in phase space and f ( t ) , g(r), Cfgj(t) are identified with their numerical values at the point on the trajectory labeled by t. In Moyal dynamics, on the
other hand, the relevant structure on N is the *-product; the equation of motion
shows that d / d is a derivation of this new structure and f ( t ) * g ( t ) = Cf*g)(t).
The paradox of (2-11) means that trajectories in phase space have no invariant
meaning: if jj7 (q -+ P, 0 is a canonical transformation, then the trajectory j ( t ) ,q(t)
does not necessarily coincide with the trajectory P ( t ) , &(t). Of course, this is directly
related to the fact that the system
(j
- p)
*p
= 0 = (4
- q) * p ,
(P I ,
6)) E I w 6
(2-16)
(2-17)
This can be derived from quentum mechanics by means of the Weyl correspondence
12-1) with the following interpretations: j3 = Q(p) is a density matrix, f ( t ) = Q ( f ( t ) ) 7
f > t = tr(fj3). If j3 is pure, 9 = \ $)(+I,
then p = (1 $>(#
is called the Wigner
function for the state I.#)Remark. Equation (2-17) appears in the literature in the equivalent form (2-15).
The equality
g P p d3q = Jfg d3p d3q was discussed in {3] (I, Sect. 12d); it is
If*
311
DEFORMATION THEORY AND QUANTIZATION. I1
117
valid for the Moyal *-product (and equivalent orderings related to the Heisenberg
algebras Ifn),but not for *-products in general. We shall adopt (2-17) in the general
case.
3. General *-Products and the Notion of Preferred Coordinates
DEFINITION
2. A function F E N is a good observable if it generates, by the Poisson
bracket, a group of symplectic diffeomorphisms of W.
For a E N we have (I, Sect. 9) called a-invariant any *-product such that
(3-1)
k t z2 be the Lie algebra of all a E N such that the *-product is &-invariant. In the
case of a Vey deformation (defined by the differentiablecochains of increasing order),
we have Seen that d is always finite dimensional. [It is likely, though not proved,
that d is finite dimensional for more general deformations as well.] Therefore,
euery Vey qzuznrization scheme (one formalizable as a Vey deformation) distinguishes
ajinite subalgebra of N. This algebra is a subalgebra of the algebra Inv(F, r)that
generates the infinitesimal symplectic transformations, affine for a connection T
on W.
The Moyal *-product is invariant under Inv(F, 0,
where r is the usual Aat connection on W.If we take the natural coordinates on R2 such that the components
3 12
118
BAYEN ET AL..
Tik of I vanish, then the elements of Inv(F, I) are the polynomials of order ,<2We note that these are all good observables. Furthermore, there are enough of them
to coordinatize W; more precisely, Inv(F, )I is sufficiently large in the sense of
DEFINITION
3. A finite subalgebra d of N is called suficiently large if the map
J: W - t d*given by x A,LA(x) is injective. Here (LA}( A = I, ..., m) is a basis
f o r d and {AA} is the dual basis for d*.
313
DEFORMATION THEORY AND QUANTIZATION. I1
119
[Units have been chosen so that both physical parameters of the problem are equal
to unity.] Here
p2
=@
- 9,
I
-
(q'
4)1P
> 0,
(3-3)
and (q, fi) E (R3- (0)) x R3. The singularity of (3-2) at the origin must be removed;
this will be done by a method that goes back to Poincark and even to Kepler himself.
The equation of motion forfE P ( W , C), namely,
(d/dt)f = - { H , f } ,
(3-4)
H-EEO,
EER
and
r(E - H)M 0.
(3-5)
Here s is the mean anomaly introduced by Kepler; comparison of (3-4) and (3-5)
shows that dsldt = l/r. In this way the singular Hamiltonian (3-2) is replaced by
the smooth function
r(E - H) = Er - J.rp2- 1.
= {r(E
- H, p} = (E - &p2)q/r;
3 14
120
BAYEN ET AL.
( A = 1,..., 10).
(3-8)
( A = 11,..., 15).
(3-9)
The larger algebra includes the so(4) symmetry algebra of the Hamiltonian.
Although the Poincard algebra spanned by (3-8) is sufficiently large, there are
advantages to including (3-9) and taking so(4,2) as the algebra of distinguished
observables. Because the function r(E - H) is an element of so(4,2) (for E fixed
in W), one finds that every element of so(4,2) has simple harmonic s-dependence,
which trivializes the solution of the equations of motion [13]. In addition,
the arbitrariness in the choice of *-product is always reduced when the algebra of
invariance is enlarged.
The map J defined by (q,p) H hALA(j,if) ( A = 1,..., 15) sends W onto a sixdimensional submanifold of the dual &* of & = s0(4,2). This is an exceptional
orbit of the coadjoint action ad*d of d in d* (most orbits are 12-dimensional;
see I, Sect. lld). The construction of invariant *-products on such orbits is a very
interesting problem.
4. Spectral Theory, a Beginning
(a) For f E N and t E R consider the formal series (see I, Sect. 12)
(4-1)
Assume that there exists a p > 0 such that for I t I < p the power series in t converges
to a distribution on W.Suppose also that, for t fixed in some complex neighborhood
315
DEFORMATION THEORY AND QUANTIZATION. II
121
1 TAentlm
(4-2)
AEI
where I is a sequence in C and T,,E N for h E I. In .the special case of the Moyal
*-product, if Q(f)is a normal operator in Hilbert space, then Zis the spectrum of
this operator and the
are the projectors for the spectral decomposition. This
motivates the following definition (we return to the general case):
DEFINITION
5. Iff N satisfies (4-2) we call I the spectrum off, any h E Z an eigenvalue off, and r Athe projector associated with A.
From (4-1) one obtains
f * Exput) = ifi(d/dt)Expuf)
(4-3)
(4-4)
* T A ~ = 8AA'wA.
TA
In general, the support of dp(A) will be referred to as the spectrum off. It is the
(Fourier) spectrum of ExpCft) as a distribution in t , in the sense of L. Schwartz.
(b) Let a *-product
position law
(Lg)
(L g) - f
=2 *3
f *g
(c) Iff in (4-2) is real, then I is symmetric (that is, X E I =- E I> and +A = q
316
122
BAYEN ET AL.
Moreover, if g * g 2 0 for any real g E N (the expansion (4-2) being supposed unique),
then I and T,,
are real. We suppose now that this is the case, and define NAby ( N AE R
if the integral converges, otherwise we shall say that NAis infinite):
(4-6)
The principle of detailed balance thus tells us that NAand N,' are (up to a common
factor) the multiplicities of the states pA and p i ; that is, the multiplicities of the
eigenvalues h off and K of g.
Henceforth, we shall take the Liouville measure dp(x) on W to be so normalized
that (4-6)gives precisely the multiplicity of the eigenvalue h off. In the case of the
Moyal *-product on RZn one verifies easily that NA = tr 9(nA)if one takes
(4-7)
5. Suggestions
3 17
DEFORMATION THEORY AND QUANTIZATION. 11
I23
This bracket has the correct symmetry properties and satisfies the (generalized)
Jacobi identity.
318
124
BAYEN ET AL.
Generalizations are immediate. In the first place, A need not be constant. The antiPoisson bracket (5-1) satisfies the Jacobi identsy if and only if the (generalized)
Schouten bracket [A,A ] vanishes. The powers of P that appear in (5-2) must be understood in terms of covariant derivatives and conditions are imposed on the connection
by associativity. Applications include quantization of classical (Grassmann algebra
valued) Fermi fields.
(e) From the general considerations and the particular examples of invariant
*-products developed in I, from the remarks made about our classical spectral
theory in this paper, and from the partial success of the Kostant-Souriau geometrical
approach to group representations, it seems more than evident that *-products have
a promising future in representation theory.
(f) One of the most important theorems of classical mechanics-it can be taken
as the starting point for a formulation of statistical mechanics--is the Liouville
theorem. One form of this theorem states that the generalized velocity vector in phase
space is divergenceless. In ordinary classical mechanics (phase space W = W2
with {xi} = {ql .*. p}) this reads
-div f = (a/8xi){H,xt} = 0.
The validity of this formula follows easily from the equations of motion. In Moyal
mechanics the Liouville theorem continues to hold. The reason for this is that
[H* a]/% = {H, a} for every first-order polynomial a = aixfin 9,..., xZn. [This
statement remains valid for all the generalizations of the Weyl correspondence
considered in the literature.] It is interesting to study the validity of the Liouville
theorem for general deformations and invariant *-products.
(g) The Weyl application establishes a correspondence between functions
on phase space and linear operators in Hilbert space. To the product of operators
corresponds the *-product of functions and to the commutator of operators conesponds the *-commutator of functions (Moyal bracket). One may ask: What is the
image of the Poisson bracket of functions in the ring of operators ? That is, what is
the contraction limit fi + 0 of the ring of operators of quantum mechanics ? In other
words, we are interested in the empty box in Fig. 1. The application indicated by a
dashed line is defined by commutativity of the diagram. The physics described by the
mathematical structure in the empty box applies to certain situations in quantum
optics in the coherent states formalism.
1-
Weyl
application
Deformations
[Mcyolbracket]
m
1
Contractions
Weyl
appl icot ion
FIGURE
1
1-
319
DEFORMATION THEORY AND QUANTIZATION. I1
125
(h) Finally, it should be noted that our suggestion to regard Moyal mechanics
and its generalizations as autonomous physical theories raises the question of whether
or not all such theories can be cast in a form that fits the general axiomatic formulation of quantum mechanics.
B. CALCULATION
OF SPECTRA
6. Harmonic Oscillator
In this section the energy levels of an I-dimensional harmonic oscillator will be
computed with the help of the Moyal product. This will be achieved by considering
the function Exp(Ht) and its Fourier decomposition as explained in the previous
section. Though some of the calculations that will appear may be more or less well
known, it does not seem that such a classical treatment exists in the literature.
Here by classical we mean that we deal only with functions defined in phase space
and that no operator of the quantum mechanical Hilbert space corresponding to the
problem appears.
(a) One-dimensional case. We denote the Hamiltonian function by H; for the
harmonic oscillator
H(P,4 ) = H P 2
+ 43.
(6-1)
- (h2/4) HKXH);
(6-2)
PROPOSITION
1. For any fixed ( p , q) E W, the power series in t:
(6-3)
320
126
BAYEN ET AL.
= {(t,H) E C2 I t # (2k
+ 1)n,k E Z}.
It is obvious that the coefficients C,,satisfy the recursion relation (6-2) with C, = 1.
It follows that for any n, C, = K,,. Compare I, Eq. (12-33).
This pointwise convergence property of the series EL,,
(I/n!)((Ht/ih)*),, allows us
to prove the convergence of (6-3) in the usual Y(W) space of distributions in the
variables p and q.
PROPOSITION
2. For $xed t E ]-7r, n[ (or [ t
converges in 9'(W) for rhe weak topology to
Proof. In fact with respect to the variable H EC the series converges uniformly
on compact sets in the complex plane. It follows that when His replaced by $( p 2 qz)
the series converges uniformly on compact sets in the ( p , q) plane.
Now the map
t -l
1
t
t w (cos T ) exp
(p2
q 2 ) tan
[x +
7-
from the disk I r [ < n into 9'(R2) is weakly analytic and has a (weak) analytic
continuation in the open set U' = C - {(2k 1) T I k E Z}. We therefore define
Exp(Hr) as the distribution (6-5)in the variables (p , q)for$xed t E Li'. One easily checks
0, the distribution Exp(Ht) E Y l ( W ) ,i.e., is tempered.
that if t E U', Im t
In order to define the Fourier expansion of Exp(Ht) we remark that for fixed
(pyq) E R2 the function (6-5) is periodic in r with period h.However, this function
does not belong to L1(O, h).We therefore consider (6-5) as a distribution in the
variable r E R. In fact we have:
<
32 1
127
PROPOSITION
3. Forfixed ( p , q) E R2 - (0) thefumtjon
with
(6-7)
R = A-l( p 2 + 92) > 0 and prove the proposition for$ One first shows that f defines
a periodic distribution S, for example, by the formula:
where the test function 4 E 9(R). This allows us to compute the Fourier coefficients
u, of S and to write S = C:-o a,e"T. If we denote by S the distribution defined on
the one-dimensional torus associated with S, one has a, = (3, e-in7>. The coefficients a, are then computed with the help of the function
and a suitable contour integration in the complex plane. One finds that a,,,, =
2(-1)" e-*Ln(2R) for TI 2 0 and that all other coefficients vanish.
The Fourier expansion of Exp(fft) being defined, we can now examine the convergence (in the distribution sense, in the variables p and q ) for fixed t of the series
(6-6). This leads us to our main proposition:
PROPOSITION
4. Forfixed t E C with Tm t
i?
( k E Z)
(6-8)
the series converging in Y'(R2) for the weak topology. Moreover, i f t = =tr the series
converges (in Y'(R2)) to Tim%.
322
128
BAYEN ET AL.
Proof. We denote by Hn the Hermite polynomials of degree n and by g5, the corresponding orthonormalized function in L2(W): +,(x) = ~ - l / ~!2n)-1/2
( n Hn(x)
If T E 9"(Rz) one knows that
T=
+nl
O +n2
n,.n,EN
where the series in the right-hand side is weakly summable. Between Laguerre and
Hermite polynomials the following well-known formula holds:
n
Thus in (6-8) one can express rrn with the help of this formula. One then checks that,
up to an obvious (constant) rescaling of the variables p and q, the result is identical
to the expansion of the left-hand side of (6-8)in tensor products +nl 0 q5ns of Hermite
functions. The case t = +r is an easy exercise left to the reader.
The last point of Proposition 4 is by no means surprising since limttfr,Imt<o
Exp(Ht) = rfirfi8 holds in 9'(Rz) for the weak topology.
We may now summarize the content of the preceding considerations in the following
way. The distribution Exp(Ht) has been constructed with the help of a power series
expansion around the origin. The application t ++ Exp(Ht) may be considered as a
periodic (with period 47r) distribution-valued function defined in W. It has a Fourier
expansion in 9"(R2) and the only harmonics which occur in this expansion are
ei(n+1/2)t.
The coefficients rrz
E Y'(Rz) which appear in the Fourier series satisfy
(as is easily checked) JRt r ndp dq = 257%. We thus recover the standard result of
quantum mechanics: En = (n &)A, the multiplicity of the corresponding state
being equal to one. Furthermore, it is clear that, via the Weyl correspondence Q,
(6-8) becomes the spectral decomposition of the unitary operator Exp(&/ih) where
A = Q(H), the functions r nbecoming projectors.
(ii)
H * rrn
= rr,
* H = (n + &)fir,,
-
V n = T n ,
Tn
7Tm
= 8nrnTn.
* # = y5 * H = E#,
323
129
analytic near 0 (with respect to p and q) and belonging to Y'(R2) exist if and only
= (n
4) ti (n = 0, I,...). Furthermore, in that case $J = w n . (3) The spectra1
decomposition (6-8) allows us to compute in principle any *-function of H, such as
the resolvent ( ( H - h)*)-l. One may check that for any h E C - (ti(N f 1/2))
the series
if E
2,
((n -i
1/2) k - A ) - l
mTT,=
((H - A)
*)-l
n=O
converges in Y(P)
and that if Re h
((H - A)
*)-I
(-
+ q2) with
d6.
p 2 = Ci-1p,2,
q2 = CIC1
q,2. We also write p
y y 2 I a,B
, , y E R} is an so(2, 1) Lie algebra with respect to Moyal (and Poisson)
-+
X*f(X)
(6-9)
This proves that for any n = 0, 1,2, ..., (X*)"may be expressed as a function of X
only. Propositions 1 and 2 may be generalized as follows:
PROPOSITION
5. For any ( p , q ) E RZ2,the power series in t
(6-10)
= r/(2 I d
= 0).
= exp(Xt/iti)
if d = 0 ,
'A
if d < 0.
(6-1 1)
PRoPosmo~6. For fixed t ~ ] - p ,p[ (or I t I < p and t E C) the series (6-10)
converges in W(R23for the weak topology to one of the expressions (6-1 1) according
to the sign of d.
Proofs are analogous to those of Propositions 1 and 2. This allows us to define
Exp(Xt) for fixed t E C outside the singularities of the functions (6-11) as in Section 6a.
324
130
BAYEN ET AL.
< 0:
In case d > 0, Fourier analysis may be worked out with the same results as
previously. W e consider the case X = H. Obviously
Exp(Ht) = Exp(H1t) *
**.
* Exp(H2t)
where H1 = $(p:
~ x p ( ~=t )
n-0
( c
0 .-.0T n z ) e-i(n+2/2)t.
n,+..-+n(-n
Let Lg)denote the generalized Laguerre polynomial of degree n. The known formula:
f L k ) ( X ) L!Lm(y) = L?+D+l)(X + y )
m=O
implies that
PROPOSITION
7. For fixed t E C with Im t
= &I..
(x -)
t -1
2x
t
Exp(Xt) = (cosh T ) exp
tanh 2
[- e A ' 4 r ( h ,X ) dh
W
(6-13)
325
131
Here lFl
denotes the confluent hypergeometric function. For any h E Iw, n(h, X( p , 4)) E
9"(rWzz) and one has X * n(h, X ) = n(X, X ) * X = h ( h , X ) . Furthermore, if we
define E(A) =
n(p, X)dp, the map h H E(h) is a continuous map from R into
Y ( R Z z ) endowed with the weak topology, such that limA+m(A) = 0 and
limA-mE(A) = 1 hold in this topology. In this case also, *-functions or distributions
may be computed. For example, if T E Y'(R), T ( X * ) is defined by the formula(where
is the Fourier transform of T )
T(X*)
and the function
we have
(T,~ ( hX,) )
= (2n)-'I2(T7 Exp(-shX)j
Remarks. (1) In the previous analysis we have observed that the function Exp(Ht)
had singularities in t. This fact is not surprising. Assume for simplicity that of fi = 1
and I = 1. In the Hilbert space L2(rW)one can show that for 0 < I r 1 < T (t E R)
and f E L2(R) one has:
Thus if t -t n this expression for exp(-&t) becomes singular, though the oneparameter group r H exp(-iRr) of unitary operators is continuous and exp(-i%) =
- i S 2 = Q(-in8). (
9
denotes the Fourier transformation.) (2) In order to get
Exp(Xt) we utilized a power series. Another definition of the exponential function,
x/n)", is also available. In fact it is easy to prove that
namely, ez = limn+&
?a-7, ((1
iim
+ $)*)" = Exp(Xt)
holds for t E 02 and [ t I < n for pointwise convergence and weak convergence in
9'(Rzz). (3) The computation of the continuous spectrum (case d < 0) is not only
completely autonomous in our formalism, but also much easier than the corresponding operatorial treatment. (4) The function +H = & ( p 2 qz) "represents"
the usual compact (elliptic) generator of so(2, I), i.e., the generator with period 2nwhen exponentiated in the structure to the group SO(2, 1) (and period 47T in SL(2, R)).
The other two usual (hyperbolic) generators are represented by &pq and p 2 - q2),
and the lowering and raising generators by ( p f fq)2.For l = 1, from Remark 2
in Section 6a, one sees that ( p - iq)2 * rn* ( p iq)2 is proportional to T , , + ~ .
The multiplicity being 1 in that case, the *-exponentials of the generators of so(2, 1)
close to the direct sum of two irreducible *-representations of the metapkctic group,
the twofold covering of Sp(2, 52) = SL(2, R). The group appears thus naturally
a(
326
132
BAYEN ET AL.
7 . Angular Momentum
zz
<
c = (e*)2
c (L,
*)Z
ez - p i z ,
*f(g')
3?i2
(7-1)
where f',...,f(4)
327
DEFORMATION THEORY AND QUANTIZATION. I1
Let t, x
133
#(tyx)
2-1
= (cos T )
exp
- 2-dimensional
(7-2)
(7-3)
T h e proofs of this lemma and of those that follow are given in Appendix A-2.
We thus prove that, up to a constant, the Moyal powers of the Casimir C are the
o d d Moyal powers of the Hamiltonian Hof the I - 2-dimensional harmonic oscillator
divided by g (after replacing H by g).
The idea is now to find a *-square root y of g2 - (31 - 4)(fi2/4) = C + ( I - 2)2(fi2/4)
in order to compute Exp(ys). We consider the function y = g - ( I - 2) fi2/4g defined
in the open set U = {( p , q) E R21 I g ( p , q ) # 0); we shall prove that ( Y * ) ~=
g2- (31 - 4) fi2/4 holds in U.It is legitimate to restrict ourselves to this open set of
W21since it is an open symplecticsubmanifold of RZLwith induced symplecticstructure.
LEMMA
2. Let Q( y ) be any polynomial; denote by W the function W ( p , q ) =
Q( g2(p , q)) and by R the polynomial R(x) = Q(x2). Then in U we have
(7-4)
where A is the differential operator
A =
fi2
d2
-(x+ I-) dxd + x - ( I -4x2) f i z 4
dx2
The next step will be to show that (7-4) holds for more general functions of g2.
U p to now we have not used the fact that y depends on fi. As a matter of fact, as we
mentioned before, the *-product can be naturally extended from the differentiable
functions N to the formal series E(N, fi) in the parameter fi of the deformation with
coefficients in N. Let q5 E Cm(U)and suppose first that 4 does not depend on fi;
h 28
134
BAYEN ET AL.
* 4 will be the power series in fi which is obtained after a reordering of the powers
of Zi in the series C,"(l/n!)(ih/2pP ( y , $), i.e.,
(7-5)
We have :
y = g - [(I - 2)/4gJfi2 be defined as before and suppose that
- (0)) does not depend on fi. Denote by 4 thefunction 4( pyq) = F( g( p , 4)).
LEMMA3. Let
E C=(BB+
ien in U we have
(7-6)
Here A is the differential operator defined in Lemma 2.
We finally consider the case when F = xi-,,hkFk is a polynomial in h with coefficients FkE Cp.(W+*)and show that (7-6) holds also in this case. In the expansion of
the star product y * F( g) the coefficient of Zi" will be after reordering:
(+)n
(+)'
n(n
- 1) .--(n - k + 1)
x [Pn-'(g, Fk(g))
According to Eq. (A-3) of Appendix A-2 all terms such that n - k > 2 will cancel.
Hence we get for such an 1":
LEMMA4. In the open set U we have ( Y * ) ~= g2- [(31- 4)/4]fi2 and more
generally
( y * I n = g-lGn+l(g)
if
n 2 1.
(7-7)
One has thus shown that the nth Moyal powers of the function y( p, q ) = g( p , q ) -
t 29
DEFORMATION THEORY AND QUANTIZATION. I1
135
PROPOSITION
1. For any ( p , q) E U,the power series in s
(7-8)
Moreover the series (7-8) converges in 9j'(!R2')for the weak topology to the right-hand
side of (7-9) i f 1 s I < r r and s is fixed.
Proof. Equations (7-2) and (7-7) imply that in U
if I s I < T. Suppose now that s is fixed and I s J < T. It is clear that with respect
to x the series C ( l / n ! ) ( ~ / i Gn+,(x)
f i ) ~ converges uniformly on compact sets in R.
Thus for any b > 0 and any E > 0 there exists No = No(,b) such that for any
n 2 No and any x E [-b, b], 1 Xm,' (l/m!)(s/ifi)mGm+,(x)I
I R,(x)l < E . Let
d~E 9(R2'), K = SUPP4, b = S U P ( D . ~ ) E K g ( p , 41, and = JK g ( P , q)-' dp dq- For
any n 2 N,(EM-~I\
4 ,';I\ 6 ) we have
If I s 1
+ 2g(ih)-l]
(7-10)
330
136
BAYEN ET AL.
PROPOSITION
2. Forfixed s E C with Im s
(7-12)
Exp(ys) = g-1e--*+Wz
c ataw
rn.
n>o
Moreover, Eq. (7-10) allows one to make an estimate of the remainder of the previous
series: for any 7 E C such that I 7 I
r < p < 1 where r and p are arbitrary, any
g 2 0 and any n 2 0, the following inequality holds:
<
E Y((W2l)
This gives a meaning to the representation (7-11) of Exp(yso) by the infinite series
when s = so and defines a topology for which the series converges.
We have thus obtained the spectrum of y:
~p y = Ifi (n
+9
)
1 n = 0, 1,...I
+
33 1
DEFORMATION THEORY AND QUANTIZATION. 11
137
of the
n,,
(7- 13)
8. Hydrogen Atom
(a) In this section we determine the spectrum of the Hamiltonian H of the
hydrogen atom (H.A.). This will be accomplishedwith *-product techniques analogous
to those of Sections 6 and 7 but slightly more involved. In fact, the difficulties that we
shall meet with the Moyal *-product will compel1 us to change the *-product; namely,
to introduce a *-product already described in Sections 8 and 10 of I and associated
with the geometry of the Kepler problem. We start with the well-known classical
so(4) symmetry of the Kepler problem. We reparameterize by stereographicprojection
from the cotangent bundle T*(S3)C R8 over the three-dimensional sphere. The
"good" star product, denoted by *', will then be deduced from the Moyal *-product
in the embedding space R8 by suitable restrictions.
It appears that for any I > 2, the Kepler problem in I-dimensional configuration
space may be worked out in the same manner as in the case 1 = 3. The reparameterization will be done in that case with the cotangent bundle T*(Sz)C 8Ber+2 over the
I-dimensional sphere.
(b) Difficulties with the Moyal *-product and some notations. Let ( p , q) E Re'
= (qI2 --. q?)ll2 = I q I. We consider the following
Hamiltonian
H( p , 4 ) = p2/2 - r-1
+ +
and attempt to get its spectrum by *-product algorithms. One may at first think
of (at least) three approaches, all of which meet with difficulties if we keep the usual
Moyal product in W E .
(i) Direct exponentiation of H as in Sections 6 and 7. This seems for the
moment hopeless: successive *-powers (H*)* of H are not in general functions of H
only; these. powers become more and more singular as n increases and no simple
recursion formula seems to hold among them. For example, when I = 3 one finds
that
332
138
BAYEN ET AL.
r, = -21 * ( p + 1 ) = -2I
r(p2
ifi p q
+ 1) - -- (Z - 1 ) h2
2 r
8r
'
ifi p . q - ( I - 1 j ti2
r4= -21 r * ( p 2 - 1) = -21 r(p2 - 1) - 2 r
8r
'
S =p - q
(8-1)
- iA/2 = T - 312.
The Casimir is
c=(r,*j2-((r4*)2-(s*)2=g2-~(z-
I)#
u = &(r,
+ r4)- ~ ( -r r,)
, -I
as expected. Let E < 0 and write ea = ( -2E)ll2. The function Exp(Ts) may be easily
computed with the help of the results of Section 6. One verifies that, iff is a polynomial in the variables p and q,
ExP(--J)
* f( P,4) = W
P ,e - w * EXP(- Ts).
Thus this equation will hold on formal series and hence on all (generalized) functions
for which the above *-products are convergent. This allows us to write
EXP(-T~) * f(P , 4 ) * EXP(TJ) = f (espye-'qj.
* U * Exp(Ts) = 4e"(ro
+ I'J - Ee-d(T,,- r4)- 1
= (-2,9112
r,,- I.
r,
333
139
<
-=
< +
<
r,
cl ==
=
(c
(Mjk*)2
l<i<k<l
&42(p2
(Aj*J2
3-1
1)2
- ( t i 2 / 4 ) ( 1 - 2 ) p2 - ( h 2 / 4 ) Z(1 - I),
4(p2
1)1/2
* r * ( p 2 + 1)1/2.
One easily checks that (yo* ) 2 involves only a finite number of terms and one gets
(yo * ) 2 =
Cl
+ (h2/4)(,-
Furthermore, if E < 0 and ed = (-2E)ll2 (as before) one has Exp( -Ts) * ( H E) *
Exp(Ts) = &?8( p 2
1)1/2 * (1 - e-8(yo * j-l) * ( p 2
1)1/2 where ("/o*)-l =
2( p2
1)-lj2 * rl*
( p a I)-V. Thus the knowledge of the spectrum of yo would
give the negative part of the spectrum of H. Although here the even powers of yo
are polynomials (in p , q), the situation is not better for computing Exp(yd). No
simple recursion formula appears for the power ('yo * )2n.
In this approach the previous representation of so(Z+ 1) may be enlarged to a
representation of so(/
1, 1). Let M3,z+2= -$(p2 1) q1 - ( p q ) p j ( j = 1, ..., I),
M L r l m l=
f 2 T = p . q. Then {MuB}(1 < a < /3 < Z 2) do form a representation
of so(Z
1, 1) with respect to Moyal and Poisson brackets. However, this represenof (ii). In fact
tation is not equivalent to that obtained with the generators
in this case one gets &Mu,
* M", = -(h2/4) 12 whereas for the two representations
of so(/ -/ I, 1) with "hatted" generators the value of the same Casimk is equal to
-(fi2/4)(12
(a=,)
- 1).
A11 this strongly suggests that there is something wrong with the *-product of
W21we used at first. Hence we shall try to choose a new *-product that is suggested
by the geometry and the symmetry group of the problem: SO(I I). We are working
in 21-dimensional phase space. This leads to parameterizing the problem with a
manifold on which SO(1 f 1) acts naturally. In view of Section 8 of I, the cotangent
bundle T*(Sz)appears as a natural condidate.
334
I 40
BAYEN ET AL.
functions on
= (R2+l- {O})X RZ+l.We denote by * w the new *-product on the
manifold Wand *o the Moyal product on the open subset w,,of [w2z+2. We thus get:
<f*ok>I w.
f*wg=
(8-2)
If one chooses a chart on W one will get a *-product in an open subset of R21.
This *-product will depend on the choice of coordinates and will be different from the
usual Moyal product in R2. However, it results from our study that Q3 and Pa
(the third power of the Poisson bracket on R2?, restricted to the considered open
subset, will belong to the same (nontrivial) cohomology class, i.e., define equivalent
infinitesimal deformations.
We now give an expression for the product * w in coordinates. We choose for St
the stereographic projection from the north pole:
(8-3)
gij
dp dp,
i.I
j-1
and this will define the duality between vectors and covectors. The expression of
cotangent coordinates will thus be:
(8-4)
Formulas (8-3) and (8-4) define a chart for W on the open subset 52 = W @-YO, ..., 0, 1)) where B is the projection of the bundle Won its base F.This chart
will be denoted by c = (Q, +) where i,h is the map (v, 6 H ( p , q) defined by (8-3)
and (8-4).
The corresponding curvilinear coordinates on Podefined by the projection
4
from Woon 882 are:
$ 0
m,,
335
14 1
tion between upper and lower indices, since both p's and q's can be considered as
coordinates on W. The inverse map #-I: RZz-+ 52 may be written:
rj
= 2(p2
+ l)-'pj
rz+1=
(PZ
+ 1)-'(p2 - I),
51+1 = P
4.
5j = H p Z + 1) q5 - (Pq)pj
N(W);its expression with the help of the chart c will be the function F E Cm(W)
*
LetfE
which is defined by F( p , q) = f (r,@. The corresponding function which is defined
by F( p , q ) = f(r,[). The corresponding function P defined in 0 will be the function
P(T, .f) = F(x(n),y ( r , 0)= f(l r 1-l r, I r I 5 - I r 1-l (n * 5) T). In the specific
chart c we shall denote the twisted product *w by *'. We thus finally get for two such
functions F, G E C=(Rz2)
F*'G=(@'*;Q)l
(8-5)
W)o#-I
0.It is this
product
*'
Remark. The functions xj and y,, together with the two functions xl+l(n) =
log I r I and ~ ~ + 5)
~ (=nn, f , define a (Poisson) canonical diffeomorphism of 0
onto R2z+2:
1+1
C dx5 A
1+1
dyu =
1 dru
d5,
UP1
RE1
= M'(qj
M'(pj
qk,) = 0 7
T(1)
= (ql
*')"I
Sfk
*
(1 G-i,k
< 0.
= 1,
7
47,) = 41 ,
* p")
qk>
d p j ) = pj
T(q"
1 <.j G 1,
<j < 1,
I")'.*
<I
(8-6)
336
142
BAYEN ET AL.
and
One checks that T maps polynomials in the variables p , q into polynomials in q's with
coefficients in the ring of rational fractions in p's whose denominators are powers
of p 2 1. If R and S are polynomials we thus get
(8-7)
and one can define easily an extension of this homomorphism to the ring of polynomials in q's whose coefficients are C* functions of p . Moreover, T will map formal
series in p , ,q5 into formal series in the variables p5 , q j , and z 5 (p 2
and the
ring of formal series in qj with coefficients that are C" functions of p , into itself,
in such a way that (8-7) will hold. This relation (8-7) exhibits the "c-equivalence"
(in the sense of I, Section 7) between the associative algebra deformations defined
by the twisted products * and *'.
Remark. One should note that in relation (8-6) the product of two "position
coordinates" is a function of linear momentum also. This feature is completely
different from what one gets with the usual Moyal product, related by the usual Weyl
which is a natural product
correspondence to the operator product. The product
for the preferred set of observables for the hydrogen atom problem, though being
(as we shall see) equivalent spectrally to the product *, differs radically in its functional form from the latter. [It has the correct classical limit when fi + 0.1
We now consider the product which is induced on so(l 1) by *'. The generators
are
= q5pk - q k P 5 Mj,Z+l = A3 = - i ( P 2 - qj f ( p * q ) P f (1 <j7 k < 2).
8 = tarfl
- [pa(1 < a,
One checks that on the open set 0 one has (MaB0 #)(r,
fi I 1). We thus have on so(Z 1) the *-product mentioned at the beginning
of Section 7 ( I 1 replacing I). The product *' is so(f 1) invariant in the sense of
Section 9 of I, i.e., we have for F, G E Cm(R21)and any X = &B
AoLBMaB
where ha,DE W
*I
*I,
< +
{F*'G,X}=F*'{G,X}+{F,X}*'G.
The spectrum of the generator Ma, is known from Appendix A-I. In the particular
case in which we are primarily interested (f = 3), the well-known decomposition
so(4) = so(3) @ so(3) allows us to exponentiate easily all generators of either so(3)
subalgebra. Let =
(aj/2)(Mk, i M ~ Jwhere (jkm) is an even permutation
of (123) and
aj2 = 1. SupposefE C"(R); then one gets as in Section 6
x x;=l
~ x p ( ~=
s )jcos
$1'
exp
[-
4iX
tan
337
143
c=
M,,
M , ~= r2 - i(z
+ I)
k2
l<*CB<l+l
where
= &-(p2
1) = I [ I and where the corresponding (G-invariant) function
is T(T,0 = T(X(T),
y(n, 0)= ( 7 ~ 2 f 2 - (T .()2)1/2. One finds Sp C = (n(n Z - l)h2
n = 0, I,...). The square root (for the product *') of C (I - 1)2(fi2/4) is
y = T - [(I - 1)/4I'] h2. The Fourier-Dirichlet expansion of the *'-exponential
Exp(ys) is obtained from formula (7-1 1):
Exp(ys)
CFne
--I(n+.+)S
It
(8-8)
where
fl',
I-
417
+ 7)
-c2
(T)
(8-9)
Moreover for any s E U' with Im s < 0 the distribution Exp(ys) is tempered and
Eq. (8-8) holds in 9 ' ( I w z L ) for the weak topology if Im s < 0.
One easily checks that
NnJ = (2n
(Mjk),
(1
However, here the product *' does not induce an invariant *-product (in the sense
of Section 9 of r) on the Lie algebra so(I 1, 1). For example, one gets
338
144
BAYEN ET AL.
(d) Discrete spectrum of H. We shall solve in this section the problem of the
discrete spectrum of H, i.e., we suppose E < 0. This will be done in a Schrodingerlike approach: We look for solutions of
(H - E) * @
Q, =
0,
(8-10)
5,
where the bar denotes complex conjugation. We utilize the notations introduced
in Section 8b: ?I = (-2E)1/2, yo = B(p2
* r * (p'
T = p q. We
have
Exp(-Ts)
Thus if we set
If we introduce Y = (yo* ) - l *
+ 1)'/'
(8-12)
We now show that the spectrum of yo for Moyal *-product is identical to the spectrum
of y = F - [(Z - 1)/4r] fi2 for the *'-product. To begin with we have 7(y0 * yo) =
y * y. In fact
l<a<B<1+1
(M&*')2
?i'
+ ;?(I - 1)' = y *' y.
We know that
339
DEFORMATION THEORY AND QUANTIZATION. I1
145
Let us consider the G-invariant function l?, associated with IT', .We can develop
Prn
as a series in the variables T, 4 with respect to the Moyal *-product in W2'+2.
n',
=n'n
1-1
* yo = *do
c (n -7)
i
fi2nn
a nd finally
yo =
I- 1
f ( n + 7)
Ann -
n20
and hence only the plus sign is admissible since the left-hand side is positive.
Finally we get
1
(-2E)'/2 = e- * - n
1-1
( + -) 2
fi
340
146
BAYEN ET AL.
>0
and we
It is trivial to check that one has (u *)- = 2-1/2r1/2* ( p 2 1)112 * (yo*)-1/2 and that
if yl0 = &1/2 * ( p 2 1) * r1I2 one has yo = u * yr0 * (u *)-l. The functions yto ,
y, , T already mentioned in Section 8b satisfy the commutation relations of so(2, 1)
for Moyal bracket. It follows that y = 7(y0), y4 = T(U * y, * (u*)-l), and
7(u * T * (u*)-) satisfy the commutation relations of 4 2 , 1) for the bracket M.
Consider the abstract Lie algebra so(2, 1) with the following commutation relations
[ F 0r
,,]= iT, [r,
, T]= -iFo,
[TI, Po]= ir, . One has Ad(exp(-ifr,)) *
= (cos t ) Pa (sin t ) T. If I s I < ?T the function Exp(ys) is well defined by
its power series expansion
r,
The function Exp(y(.rr/2)) may be written with the help of Eq. (8-8):
341
DEFORMATION THEORY AND QUANTIZATION. I1
f4
* (v
*)-I
147
where
It follows that for the product * the spectra of 7(f4) and 7(T) = Tare identical.
It is easy to show that this spectrum is the real line. One first proves that if T * I#A =
+A * T =
(A E R) with
=
, then for any 01 E R the function
=
I p !ia* 4 * I p
satisfies the same equations with A fi01 instead of A. Thus one
only needs to show that there exists a function I#,, such that T * +,, = I#o * T = 0,
= $o . The function $, = 7(r(Oy T))where ~ ( 0 T)
, is given by Eqs. (6-13) satisfies
these requirements, and moreover
* q5u = 6(h - p) . Note that the spectrum
of T = p - q for the Moyal product has been obtained in Section 6b. It is of course
the real line (which shows in this case also that 7 is isospectral). If we denote the
generalized eigenvalue by h E R, we have A = e-8 and thus E = 1/2A2 > 0.
We have thus succeeded in getting the spectrum of the hydrogen atom Hamiltonian.
I n our approach the singular Coulomb potential r-l has been replaced by the (probably
smooth) velocity-dependent potential
o0
APPENDIX
(A-1)
342
148
BAYEN ET AL.
+
aL= 2-1/2(q2 +
a, = 2-1/2(q1 ip,),
N , = i&
* a,,
Nc = Zl* al .
@I2),
The functions a, and a, are (respectively) creation functions of left and right
quantas: a,. = 2-1/2(a1- iaz), a, = 2-1/2(a, iaz> where a, = 2--1P(q, + ipi),
j = 1,2. The functions N , and Nl commute (for *>and one has
H=N,+N,+
L, = N ,
1,
(A-2)
- Nl .
2
U,
2(-])k
e-(2hr,+l)
Lk(4N, + 2) e - i ( k + ( l / 2 ) )
I;;O
nn,m
4(--1Y e-2HL(l/r)(n+nr)(2(H G)
k/Z)(n-rn)@(H- LJ)
d = nn.m
a ;
0, andp E R*. Let I t
Go(x) = 1.
343
149
-+ *.. 4- ZiyTy(g2).
A2[31(Z- 1) h2/8 - ( I
- 2) p-2
+ 2(a - 4) g2];
tig"[(Z2
~cz~/.L-~((.L
A4g6 = 16azp-2g6.
Furthermore T,(gz) = S2(g2).Hence Tn(g2) = 5'2,(g2) = g-'(-fi)-'"-'
> 0 and the proof is complete.
2n+i(g) =
Next define recursively the following functions RJx) = 1, R, = A"& (n > 0).
It is straightforward to check that for any n 2 0, R,(x) = x-'Gn+,(x) where G,
is the polynomial which is defined in expansion (7-2). On the other hand
let F EP ( R + - (0));then
(A2F)(g) = ( g 2 - (31 - 4)(fi2/4))* F(g).
TO see this one only needs to rewrite Eq. (7-1) with f(gz) = F(g). It follows from
R,n+-, = A2"R1 = AR2, and R,(x) = x - [(I - 2) fi2/4x]that
((g' - (31 - 4)(fi2/4))*In
* Y ! D . q = (AR2n)(g(~,4)).
344
150
BAYEN ET AL.
This proves relation (7-4) in the case Q(y) = T J y ) for RZn(g)= g-1G2n+l(g)=
Tn(g2).The powers ((g2 - "31 - 4)/4] h2)*)" generate all polynomials in the variable
g2, whence the result.
LEMMA
3. One$rst computes P2(g,F(g)) andfinds
P2(g,F k ) ) = 2(gF"(g)
+w d ) .
It follows from Lemma 2 that in the case + ( p , q) = Q(g2(pp,q)), where Q is a polynomial, we have:
y
* d = yd - (W3)P 2 ( g ,4).
On the other hand, for such a 4, Eq. (7-5) holds (the series has only a finite number of
nonvanishing terms) and P(g,4) = 0. Hence for any n > 2 we get
P q g , +)
+ (1 - 2) n(n - 1) pn-yg-1,
4) = 0
(A-3)
in the case 4 = Q(g2). We now show that the previous equality holds for any
= F(g). Let n > 2 be a fixed integer. It is easy to prove that there exists a sequence
of polynomials (Pk)k>l such that the polynomials Rk defined by &(x) = pk(xz) have
the following property: on any compact subset of R+*, the polynomials
& , R'k ,..., RE) converge uniformly (respectively) to F, F',..., F(n)as k 4 co. Let
us write R k ( g ( p ,4)) = $k( p , 4). On any compact subset of U the expression
Pn(g, +k)
( I - 2) n(n - 1) Pn-2(g-1, 4 k ) y which is 0, converges uniformly to
P ( g , 4) (I - 2) n(n - 1) Pn-2(g-1, 4) as k 4 00. Hence this last expression is
identically 0 on U.
+
+
LEMMA
4. Let g(x) = x - [(I - 2)/4x]P.One immediately checks that (Ag)(x) =
x2 - [(31- 4)/4]iizy
which proves the lemma for ( Y * ) ~ . We prove by induction rhe
end of the lemma. One has glx) = x-lGZ(x).Suppose that (7-7) holds for n and that
x-lG,+l(x) = R,(x) is a polynomial in fi. Lemma 3 implies:
(p)n+l
= Y * ("/In
REFERENCES
1.
345
DEFORMATION THEORY AND QUANTIZATION. I1
151
3. F. BAYEN,M. FLATO,
C. FRONSDAL,
A. LICHNEROWICZ,
AND D. STERNHEIMER,
Deformation
theory and quantization. I. Deformations of symplectic structures, Ann. Phys. (N.Y.) 111 (1978),
61 ; henceforth referred to as I.
4. H.WEYL,The Theory of Groups and Quantum Mechanics, Dover, New York, 1931.
5. E. P. WIGNER,
Phys. Rev. 40 (1932), 749.
Physica
6. J. E. MOYAL,Proc. Cambridge Phil. Soc. 45 (1949), 99. See also H. J. GROENEWOLD,
12 (1946), 405.
7. The domain of the Weyl correspondence has been studied by many authors. In particular, it
e.g., K. CHI LIU, J. Math. Phys. 17 (1976), 859 and references quoted there. See also G. S.
AGARWAL
AND E. WOLF,Phys. Rev. D2(1970), 2161.
8. See, e.g., E. REMLER,
Ann. Phys. (N.Y.) 95 (1973,455. See also S. R. DE GRWTAND L. G.
S U ~ O R PFoundations
.
of electrodynamics. North-Holland, Amsterdam, 1972; AND S. R.
DE GROOT.La transformation de Weyl et la fonction de Wigner: une forme alternative de la
mkanique quantique. Presses Universitaires de Montrtal, 1974.
9. The nonlocality of the equations of motion provides the basis for the introduction of fuzzy phase
space, in the sense of E. Prugovdki, J. Math. Phys. 17 (1976), 517, 1673; S. T. ALI AND E.
PRUGOVEEKI,
J. Math. Phys. 18 (1977), 219.
:O. The map J is Souriaus moment; J. M. SOURIAU,Structure des Systemes Dynamiques,
p. 105, Dunod, Paris, 1970;See also C.-M. Marle, in Differential Geometry and Relativity,
volume dedicated to A. Lichnerowicz (M. Cahen and M. Flato, Eds.), Reidel, Dordrecht, 1977.
11. P. A. M. DIRAC,Lectures on Quantum Mechanics, Belfer Graduate School of Sciences
Monograph Series No. 2, Yeshiva Univ., New York, 1964.
Nuovo Cimento A 53 (1968), 717.
12. See, e.g., G. GYORGYI,
13. See, e.g., C. FRONSDAL
AND R. W. HUFF,Phys. Rev. D 7 (1973), 3609.
L4. See C. MARTIN,Lett. Math. Phys. 1 (1976), 155.
15. M. FLATO,
A. LICHNEROWICZ,
AND D. STERNHEIMER,
J. Math. Phys. 17 (1976), 1754.
16. I. E. SEGAL,Symposia Math., Vol. XIV, pp. 99-117, Academic Press, New York/London,
1974, and earlier references quoted therein.
17. One simply substitutes in (2-4) formal powers of the field-theoretical Poisson bracket (using
higher-order functional derivatives). This has of course to be justified. See also G . S. AGARWAL
AND E. WOLF,Phys. Rev. D 2 (1970), 2187, 2206; E. REMLER,
preprint, College of William and
Mary, 1977.
18. M.GUENIN,Phys. Left., 62B (1976),81 and Ref. 8. M. M. FLATO
AND M. GUENIN,
Helv. Phys.
Acta, 50 (1977),117.
CERN preprint TH. 21 39 (March 1976). K. DRUHL,Max-Planck-Institute
19. R. CASALBUONI,
preprint, Starnberg, December 1976. A. SUDBERY,
Univ. of York preprint, October 1975.
20. J. L. GERVALS
AND B. SAKITA,
Nucl. Phys. 834 (1971).
21. J. WESSAND B. ZUMINO,Nucl. Phys. B 70 (1974),39.See also J. Schwinger, Phys. Rev. 92 (1953),
1283 and Proc. Nut. Acad. Sci. U.S.A. 48 (1962), 603; M. FLATO
AND P. HILLION,Phys. Rev.
D 1 (1970), 1667.
22. A. ERDBLYI(Ed.),Higher Transcendental Functions, Vol. 2, McGraw-Hill, New York, 1953.
346
J. Phys. A: Math. Gen., 13 (1980) L185-Ll88. Printed in Great Britain
Abstract. In this Letter we show how for classical canonical transformations we can pass,
with the help of Wigner distribution functions, from their representation U in the
configurational Hilbert space to a kernel K in phase space. The latter is a much more
transparent way of looking at representations of canonical transformations, as the classical
limit is reached when h + O and the successive quantum corrections are related with the
power of P,
n = I, 2, . , . .
In recent publications one of the authors (MM) and his collaborators have discussed
extensively the representation in quantum mechanics of non-linear and non-bijective
canonical transformations (Mello and Moshinsky 1975, Kramer et a1 1978, Moshinsky
and Seligman 1978, 1979a, b). The representations, to be denoted by U, are given in
definite Hilbert spaces like, for example, the one associated with coordinate q ; thus the
matrix elements (4'1Ul4")related with specific canonical transformations were derived
explicitly. It is not easy though to see from these matrix elements the quantum
modifications to the canonical transformations, as the latter are formulated in phase
space rather than in Hilbert space. Thus it is interesting to discuss the representation of
canonical transformations in the phase space version of quantum mechanics that was
developed originally by Wigner (1932), with the help of the distribution functions that
now bear his name. We shall do this in the present Letter, illustrating the analysis with
the representations of some simple examples of canonical transformations.
We begin by recalling the definition of Wigner's distribution function f(q,p ) for a
given wavefunction +(q),i.e.
m
f ( 4 ,~ ) = ( . r r W - ' I m ( $ 1 4 + y ) ( 4 - y l $ ) e y p ( ~ ) dy,
(1)
0305-4470/80/060185
-+ 04$01.50
L185
347
L186
under which a classical distribution function f(q, p) would of course transform into
F(q, P ) given by
F(4,P ) = f [ Q ( 4 ,P ) , P(q,PI].
(3)
(4)
(5)
jq = ( v h ) - l
1-m
exp(2iP(Y-9))
h
dp,
m,P )
11
dq dPfk.79PNqPlIClqP),
(6)
--03
WplKJqp)= 2(.nh)-
J jdy dy (q + y l U tl q + y )
-m
x(4 -
(7)
22 ( ~ P I K I ~ P ) = S [ ~ - PQ)(I~S, [ P - P (PI].
~,
(8)
348
Letter to the Editor
L187
equations (Mello and Moshinsky 1975, Rramer et al 1978, Moshinsky and Seligman
l978,1979a, b)
Q(q , h ~a) ( ~ f l ~ 1 4 3 = ~ f ( q l U / ~ ) ,
(10a)
(10b)
Of course these equations only make sense when Q, P are well defined operators;
otherwise, more sophisticated procedures need to be used (Moshinsky and Seligman
1978, 1979a, b).
We shall now consider two simple examples of canonical transformations. The first
will be the linear one
Q = aq + bp,
P = cq + dp;
a d - b c = l , b>0,
where the constants are all real. We have then (Moshinsky and Quesne 1971)
( q l ~ l q =
) (~!.rrb)-/~
exp[(-i/~b)(aq~
- 2qrq+dqrf2)3,
(11)
(12)
which satisfies equations (10) if we note from (11)that c = (ad - l ) / h Introducing (12)
in (7) and using the relation (q1Utlq)= (qIUlq)*we immediately obtain
(qplKlqp)=
(13)
Thus for the linear canonical transformation the kernel coincides with its classical limit
(8), in agreement with the fact that for this type of transformation Poisson and Moyal
(1949) brackets coincide.
In the second example we take Q as the Hamiltonian of a linear potential (Landau
and Lifshitz 1958), and thus we have the canonical transformation
Q = (p2/2m)--Foq,
f = -p/Fo,
(14)
where m is the mass, Foa constant of the dimension of force, and {Q, P } = 1. Equation
(10a) leads then to an Airy function (Landau and Lifshitz 1958), and we also satisfy
(106) and get a normalised (Landau and Lifshitz 1958) unitary representation if we
write
(41UIq) =AW-5),
(15~)
5 = [q+ ( ~ / F o ) I ( ~ ~ F o / ~ ~ ) ~ ,
A = (2m)1/3.rr-1/2~;l/621-213,
(15b)
(15~)
(15d)
Substituting (15a) into (7) and making use of (15d)we can show straightforwardly that
for the canonical transformation (14) the kernel K becomes
(16)
We note first that when h + 0 the function becomes (Landau and Lifshitz 1958) either
very small or very rapidly oscillating except when q = (p2/2rn)- Foq. Furthermore,
with the help of ( 1 5 d ) we easily see that v-1/2 @(x)dx = 1. Thus the expression in
349
L188
{ } in (16) tends to a S function in the limit h+0, so that the kernel K goes into its
classical limit (8),where Q and P are given by (14).
To see what the quantum corrections are, it is best to apply the K of (16) to a smooth
distribution function f(q,p ) , rather than study it directly. We choose
(17)
where from (1)we will have the relation b = h / a iff is obtained from a Gaussian state in
configuration space, Again using (15d) we obtain for the new distribution function
F ( 4 ,p ) the expression
(18)
3 k - : even
where Q, P are given by (14). As indicated in (3),f ( Q ,P)is thaclassical change in the
distribution function due to the canonical transformation, and it will be the only one
remaining in (18) if h+0. Thus the terms associated with the higher powers of R2
indicate the successive quantum corrections to the distribution function when we
perform the canonical transformation.
The examples discussed in this Letter are very specialised, but they clearly indicate
the procedure to be followed in general. Among the more interesting cases where this
formalism can be applied are those of non-bijective (Kramer eta1 1978, Moshinsky and
Seligman l978,1979a, b) canonical transformations. The concepts of ambiguity group
and ambiguity spin used in the derivation of the representation U can then give
interesting insights into the structure of phase space as a carrier of canonical transformations, as will be discussed in future publications.
The authors are indebted to Professor E P Wigner for a stimulating presentation of his
distribution function formalism and helpful discussions, during his recent stay in
Mexico.
References
Dirac P A M 1947 The Principles of Quantum Mechanics (Oxford: UP)
Kramer P, Moshinsky M and Seligman T H 1978 J. Math. Phys. 19 683
Landau L D and Lifshitz E M 1958 Quanrum Mechanics (London: Pergamon) pp 73-4
Mello P A and Moshinsky M 1975 J. Math. Phys. 16 2017
Moshinsky M and Quesne C 1971 J. Murh. Phys. 1 2 1772
Moshinsky M and Seligman T H 1978 Ann. Phys., NY 114 243
-1979a Ann. Phys., NY 120 402
-1979b J. Phys. A: Math. Gen. 12 L135
45 99
Moyal J E 1949 Proc. Camb. Phil. SOC.
Wigner E P 1932 Phys. Rev. 40 749
350
MOLECULAR
PHYSICS,1982, VOL. 47, No. 5, 1001-1019
We have constructed the Wigner function for the ground state of the
hydrogen atom and analysed its variation over phase space. By means of
the Weyl correspondence between operators and phase space functions we
have then studied the description of angular momentum and resolved a
dilemma in the comparison with early quantum mechanics. Finally we have
discussed the introduction of local energy densities in coordinate space and
demonstrated the validity of a local virial theorem.
1. INTRODUCTION
This is the first of a series of papers devoted to the phase space description
of atomic and molecular systems. Phase space representations of quantum
mechanics have been extensively discussed since the classical works by Weyl
[l], Wigner [2], Groenewold [3] and Moyal [4]. They have been applied in
quantum statistical studies of transport processes and radiation (see, for example,
[S] and [ 6 ] ) , and in treatments of molecular and nuclear dynamics (for example,
[7-101). They have, however, not yet been used in such detailed theories as
the theories of atomic and molecular electronic structure.
I n this and forthcoming papers we shall investigate the possibility of extending the application of phase space representations to such theories as well.
Very accurate wavefunctions are now available for all atoms and for a large class
of molecules. These wavefunctions have always been generated in coordinate
space, but there has been a considerable interest in their momentum space
representatives as well. The use of phase space representations allows one to
include the coordinate and momentum characteristics in a single picture, and
hence it may serve to improve our understanding of the dynamical behaviour
of electrons in atoms and molecules.
The phase space formulation of quantum mechanics treats states and
transitions in an equivalent manner. Thus, there is a phase space function
associated with every quantum state and with every quantum transition as well.
This function is the Wigner function.
I n the present paper we shall only consider Wigner functions associated
with states. Operationally, such functions play the rBle of probability densities
in phase space. The values of the functions are, however, not restricted to being
positive or zero, although they are always real. Hence, one may not interpret
the functions as probability densities. Such an interpretation would of course
also be inconsistent with the uncertainty principle.
0026-8976/82/4705 1001 804.00 (B 1982 Taylor & Francis Ltd
M.P.
2K
35 1
1002
n(r, p ) = ( s )
du dv exp
[$o.
u + p . v)
1
(1)
(2)
Jf
(3)
f(r, P) dr dP = 1.
If +(r) and +(p) are the coordinate and momentum wavefunctions, respectively,
that is
(4)
$(r)= <rI+>,
(4)
(5)
+(PI= (PI$>,
(i)
f
dr$( r - r)*#(r
+ r) exp (
-:
P r)
(6)
and
(7)
and
s dP f(r, P) = $(r>+Krh
s dr f(r, P) = +(P)#+(Ph
($lSl$>= s s dr dP f(r, P M r , PI,
(8)
(9)
(10)
where a(r, p) is the Weyl transform [l] of the operator 8, as discussed in the
Appendix.
352
1003
It is the relations ( 3 ) and @)-(lo) that show how the Wigner function plays
the r81e of a probability density in an operational sense. Let us also note that
the eigenvalues of R(r, p) must be 5 1, since it is an inversion operator, Hence
[13] we get that
(11)
and consequently that f(r, P) must have support in a volume not smaller than
(h/2)3.
With the value of f at the point (r, p) being equal to 2/h times the overlap
between l # ) and its mirror image with respect to (r, p), we may say that f(r, p)
is a measure of the way the point ( r , p) supports the given quantum state.
Similarly, we may talk about the way a certain region or a certain trajectory in
phase spacesupports a state. A proper use of this kind of language leads to an
integrated description of the wave and particle characteristics of quantum states,
It is the purpose of the present paper to show how this kind of description
works for the ground state of the hydrogen atom.
T h e coordinate wavefunction for the ground state has the well known form
(13)
Thus we obtain the following equivalent expressions for the Wigner function
by using equations ( 6 ) and (7)
23
- rl)
(15)
(16)
and
fls(r,
8Z5
p) =yrl
353
1004
The integrals (15) and (16) are, however, not newcomers in the theory of
electronic structures. T h e integral (15) is, for a fixed value of r, the Fourier
transform of the product of two 1s orbitals centred around the points - r and r
respectively. Thus it is recognized as a generalized scattering factor in the theory
of X-ray diffraction of molecular crystals and as a standard integral in those
band theories of solids that are based on Fourier transform methods. Several
procedures have, accordingly, been devised for the evaluation of this integral
by methods of approximation.
Thus, McWeeny [17] and Silverstone [18] have studied methods in which
one orbital is expanded in an infinite series about the origin of the other. Such
methods are only rapidly convergent for small values of r, and hence they are
not applicable in the present context where all values of r must be considered.
Other methods implying the summation of an infinite series or the numerical
evaluation of an integral have been suggested by several authors ((19-211 and
references therein). Although very powerful for single values of r where the
accuracy can be readily assessed, these methods are again not easily applied when
r is allowed to vary freely.
The natural procedure to follow in the present context is to approximate
the function (13) by a finite series of gaussians and insert this series in (15).
The resulting integrals can then be evaluated analytically. This method is
capable of giving a good representation of the Wigner function for all values of
r and p, and it can easily be extended to other orbitals than the 1s orbital.
Generalized scattering factors have been calculated along these lines by McWeeny
[22] and Stewart [23].
We write accordingly
M
(17)
This .
gives
,
fle(M)(r, P) =
?)34
C
i=1
exp ( - c q P ) .
(18)
ci2 p d r , P)+
C
i>j=l
w#dr,
~ > - t P d rPI)
,
(19)
(20)
in which
(22)
and
(23))
354
Wignkr function for the hydrogen atom
1005
(24)
In the following sections we shall study this expression for the hydrogen
atom, using an M = 10 representation of the 1s orbital determined by Duijneveldt
[24]. This is the extremely good approximation to the true 1s orbital, leading
for instance to an energy which only deviates 0*00015 per cent from the true
value. The parameters defining the expansion are reproduced in table 1.
Table 1. Gaussian approximation to the 1s hydrogen orbital [24].
0.062157
0.1 38046
0.304802
0.710716
1.794924
4.91 5078
15.018344
54.698039
254.017712
1776.775559
2
3
4
5
6
7
8
9
10
0.107330
0.339658
0.352349
0,213239
0.090342
0.030540
0.008863
0-002094
0*000372
0~000044
M = 1 representation of the hydrogen Is orbital. As is well known, this approximation is obtained for a = 8/9n= 0.282942
3. A CHANGE OF VARIABLES
The Wigner function is a function in six-dimensional phase space. It is,
however, obvious from (24)that f18(r, p) only depends upon the three quantities
r , p and u, with u being the angle between the vectors r and p. Let us therefore
define new phase space variables instead of
= (XI,
x2, x 3 )
(25)
and
(26)
P = (Pll P 2 l P 3 )
by introducing three mutually orthogonal unit vectors [25]
,)!-(=le
1
2 stnu12 r p
e2=-
('+").
1
2 cosu/2 r p
1
e3
=w
r x p = el x e2.
(27)
355
1006
These vectors define a right handed coordinate system S in ordinary threespace. The orientation of S with respect to a laboratory system So may be
specified by three Euler angles a:, /I, y such that S is obtained from Soby
(1) a rotation about the third axis of Sothrough the angle a,
(2) a rotation aboilt the new second axis through the angle j,
(3) a rotation about the new third axis through the angle y.
The following relations are then valid :
(281
x g = r sin /I sin y - -
3
(29)
(30)
and p
356
Wigner function for the hydrogen atom
1007
p) =7ra
- exp ( - 2ar2)exp
(-$),
(33)
(34)
which is normalized such that
m m
J J
.5
F18(1)(Y,
1.5
1.0
p ) dr dp = 1.
2.0
2.5
3.0
(35)
3.5
4.0
4.0
p/0;dh
3.5
3.5
3.0
3.0
2.5
2.5
2.0
2.0
1.5
I .5
1.0
1.0
.5
.5
0.0
0.0
.S
1.5
1.0
2.0
2.5
3.0
0.0
3
3.5
c/oo
p o )=
(m,
1
+!a))
(36)
Fls(l)(~o,
po)=-= 0,6893 kl.
re2
(37)
357
1008
(38)
Figure 2. Contour map of the function FIs(r,p, u) for u=O. Starting from the nodal
curves (dashed lines, contour value (0.0)) contours have been drawn at 0.01, 0.03,
0.1, 0.2 k1(solid lines), -0.01, -0.03 ?V1(dotted lines).
358
Wigner function for the hydrogen atom
1009
4.0
4.0
3.5
3.5
3.0
3.0
2.5
2.5
2.0
2.0
1.5
1.5
1.0
1.0
.5
.5
p/a;h
0.0
0
.5
1.0
1.5
2.0
2.5
3.0
0.0
I
3.5
rho
Figure 3.
Figure 4. Contour map of the function F&,p, u) for u = s / 2 . Starting from the
maximum (yo, p,)=(1*405 a,, 0.759 a,-l ti), contours have been drawn at 0.3, 0.1,
0.06, 0.03,0.01, 0.006, 0.003 f i - l .
359
1010
This function is displayed in figures 2, 3 and 4 for u equal to 0, 7114 and 712
respectively. For u = 4 2 it is non-negative everywhere, but for all other values
of u it has negative as well as positive regions.
For several purposes it is sufficient to know the function obtained from (38)
by integrating over u. We shall call this function the radial phase space
function and designate it FIF,,(~,
p ) . It satisfies the normalization condition
(35). From (24) it is found to be
16
FlS(y,p ) = - r2 p 2
7r
i=l
32
7r
i>j-l
(-f)
(39))
where
(40)
Figure 5 .
360
1011
=(p/ao-l fiI
F d r , p1lh-l
0.0
0.3
0.6
0.9
1.2
0.0
0.0326
0.2676
0.4648
0.301 5
0.0473
-0.0047
0.0081
- 0.0031
0.0014
- 0*00063
0~00021
1.5
1.8
2.1
2.4
2-7
3.0
3.3
When the product of p and r is large, a regular albeit weak damped wave is
disclosed by figure 5. The presence of such a wave is readily understandable
from the expression (39). When both Y and p are large, the dominant terms in
(39) will be cross terms for which one 01 is large and the other a is small, since
it is only for such terms that both yu and l / ( q + a I ) become small and hence
lead to slowly decaying exponentials. Since (23) shows that IT$,( z 1 when one
a value is much larger than the other, we find that the relevant Bessel functions
in (39) approach jo(p?),and this leads to a damped wave as observed.
Applying a similar argument to the expression (24) shows that the contour
maps for the functions Fls(r,p , u ) must disclose damped oscillations in cos (Xp r)
when both p and r are large. That this is in fact the case is apparent from
figures 2 and 3. I n figure 4, cos (2p r) equals 1 for all r and p (the wavelength becomes infinite), and the phase space function is accordingly nonnegative everywhere.
The damped oscillations which we have discussed will, of course, have their
counterparts in the theory of generalized scattering factors (cf. $2). T h e
appearance of the oscillations in that context has been noticed and discussed by
Avery [26], on the basis of arguments quite different from ours.
I n closing this section it is worthwhile drawing attention to the complexity
of the exact phase space function, as compared to the simplicity of the approximate phase space function discussed in the previous section, and a natural
question presents itself. What must a wavefunction look like in order that the
associated Wigner function be non-negative everywhere ? Hudson [27] has
given a mathematical answer to this question for one-dimensional motion. His
analysis showed that the wavefunction must have the form
+ +
(41 1
36 1
1012
where u, 6 are arbitrary complex numbers with Re a > 0 and the complex
number c is chosen so as to ensure correct normalization. When Im u is zero,
this wavefunction describes a minimum uncertainty state [28] in one dimension.
The conclusions of Hudson may probably be extended to three dimensions
in a straightforward manner. That the wavefunction (31) describes a minimum
uncertainty state in three dimensions is in accordance with this assumption.
6. THEDOMINANT SUBSPACE AND THE CLASSICAL SUBSPACE
The function F18(r,p)of (39) and figure 5 was obtained from the function
Fls(r,p , u ) of (38) by integrating over u. It is interesting to integrate over I
and p instead to obtain a function Fls(u),normalized such that
n
S FIa(u)sin u du = 1.
(42))
Fdu)=3
with
10
-227,~
Ci2 $-
i= 1
C$p~~4(Uij
(43)
i>j-l
T ~ ,
q j=
4LYicij
(mi+ a j y
(44)
Figure 6 shows the functions Fls(u) and Fls(u)sin u. Both functions have
a sharp maximum at u = rr/Z.
/ - .
0.0
-4
.I
.2
.3
.4
.5
.6
.7
.E
.9
1.0
u/n
Figure 6 . The functions K a ( u )(upper curve) and Fl,(u) sin u (lower curve) as a function
of uln.
362
Wigner function for the hydrogen atom
1013
e, =
- a&$,
(45)
then a(r, p ) is the same component of the classical angular momentum vector,
that is,
(46)
1, = xiP2 - XaPi.
As is well known, the left hand side of (10) is zero when I#} is the 1s state and
a" is jl, or j3. Hence the right hand side must also be zero. That this is
actually the case is easily seen by remembering that f(r, p) is independent of the
Euler angles a, j3 and y . This makes all directions of the vector e3 in (27)
equivalent. But the direction of e3 is also seen to be the direction of the
angular momentum vector and thus each component of this vector does in fact
have a zero mean value.
As regards the length of the angular momentum vector, it is shown in the
Appendix that the classical function
e2
A2 = I1Z
+ 1,= + 1:
(471
Evaluating the right hand side of (10) with a( r, p) = Xa will accordingly produce
the value #Aa.
This interesting result allows us to resolve a pedagogical dilemma which
has bothered writers of elementary textbooks [29]. How does one bring the
fact that the angular momentum in a Bohr orbit is ti into accordance with the
fact that the angular momentum in the SchrBdinger picture is zero ?
363
1014
+ i2z +
12 = 2e,
(49)
is different from the operator that corresponds to the classical function X2 of
(47). Averaging A2 over the classical subspace described in the previous section
does in fact give the value P, as in the Bohr description.
la2
8. LOCALDENSITIES
An advantage of the phase space formulation of quantum mechanics is that
it leads to a natural definition of a local density in coordinate space for a given
operator and a given state. Thus (10) suggests that we consider the quantity
41)"
dP f(r9 p)a(r, P)
as being the local density associated with the operator a* and the state
Integration over the spatial coordinate gives the expectation value of d
<4(4lCI>=s
dr 4 r ) .
(50)
I$).
(51)
(52)
13,
:
Y
(53)
d r )=
-;
+(r)*+(r)*
(54)
For the ground state of the hydrogen atom this becomes, by observing (13),
Ep(r)=
--n1 exp r( - 2 r )
(55)
As an important example of an operator that is not a function of P we consider the kinetic energy operator
rf=-
2'
(56)
It may be evaluated when f(r, p) is known, but as shown by Ziff et al. [30] and
by Cohen [31] it may also be evaluated directly from the coordinate wavefunction
by using ( 6 ) . T h e result is
d r )=
iH IW2-kV21#I2)
(58)
364
1015
or equivalently
It is seen that
where
EK(r)=*(*lv#/2-)#yv2
16-i)#v29").
'Idr)=!d'RB(r) + EKC(r)),
EKB(r)
and
=*I V#12
- +,PO2 #.
(59)
(60)
(61)
(62)
(63 1
The expressions
EKE, eKc and cK were all discussed by Cohen 1311 (who
used the designations KA,KB,K c and K,, respectively). He showed that each
expression could be derived from phase space descriptions discussed earlier by
him [32].
For the ground state of the hydrogen atom we obtain the following explicit
expressions
(;-&) exP ( - Z y ) ,
1 1
eKA(r)=eKC(r)=G
1
KB(r)=g
exp (-2y),
1
eK(r)=-exp(-2r).
27rr
(64)
(65 1
(66)
P119. A
Let
(67)
(68)
(69)
Ex= -aEp
(70)
which expresses a global balance between the kinetic and potential energy. Due
to the obvious relation
E = E K i Ep
(71)
365
1016
-E,
(72)
Bader and his co-workers have shown that for a molecular system it is
possible to perform a spatial partitioning of the charge distribution in such a
manner that the kinetic and potential energies of the resulting fragments obey
the virial theorem (see [33]). For the ground state of the hydrogen atom they
also noted [33] that
E K B ( ~ ) = -E141z,
(73)
which is a local virial theorem if the right hand side is identified as the local
counterpart of the total energy.
Such an identification is, however, not quite satisfactory, since the natural
definition of this local energy density would be
EK=
= %B ( r, -k EP(r)
(74)
in accordance with (71) and this quantity is different from the right hand side
of (73). As a consequence EItn(r) and rp(r) do not satisfy a local analogue of
(70)It is, on the other hand, easily seen that if we work with EK(r), defined through
the Weyl-Wigner correspondence, then we obtain a completely satisfactory
local form of the virial theorem
(75)
(76)
(77)
It must be stressed, however, that this remarkable result only holds for the
ground state of a hydrogen-like atom.
In forthcoming publications we plan to study the local balance between the
potential and kinetic energy for other atomic and molecular systems.
10. DISCUSSION A N D CONCLUSIONS
The phase space description of a quantum mechanical system is an alternative to the description based on wavefunctions. It is from several points of
view a more complicated description. From other points of view it is a richer
description.
The state of a system is described by a Wigner function, I n the present
paper we have constructed this function for the ground state of the hydrogen
atom and analysed its properties. T h e analysis was facilitated by the introduction of concepts like the dynamical plane and the dynamical angle, the classical
subspace and the dominant subspace. The orientation of the dynamical
plane is closely related to the angular momentum vector ; the classical subspace
is that part of phase space to which the ground state motion was confined in
early quantum mechanics.
The Wigner function is found to be independent of the orientation of the
dyanamical plane. It attains a large, positive and constant value in the classical
subspace and it remains large and positive in an extended region containing that
subspace. Outside this region the Wigner function shows a rich structure
366
WigneY function for the hydrogen atom
1017
We are very grateful to Dr. Sten Rettrup for his kind assistance at an early
stage of this work. We also want to thank Dr. Helge Johansen for letting us
use his density contour programs and Dr. Kurt Nielsen for interesting discussions.
APPENDIX
where fr(r, p) is the inversion operator (1). a^ and a(r, p) are said to be mutual
Weyl transforms.
Whenever a(r, p) is a function of r or p alone, then d is the same function
of the vector operator P or b. In the general case one obtains the operator a"
from the function a(r, p) by the replacement of r with P and p with 8, followed
by a proper symmetrization of products of non-commuting operators. This
symmetrization is such that, if x, and p , are corresponding components of r and
p, then the operator associated with the function ~~~p~~ is
(A 2)
367
1018
i= p
(A 4)
which is just the ordinary quantum mechanical angular momentum vector.
For the square of an angular momentum component, l3 say, we obtain
fi=(22$3-d3$2,
d3$1-%$8,
4$2-2$1)
- 2Xd,X,P2.
(A 5 )
Symmetrization according to (A 2) gives the corresponding operator which we
denote Az2
X22 $12
(A 6)
+$232).
Squaring the third component of 1gives, on the other hand, the operator
$12
- 4$1$2i32
[$k, b k l
=ifi
+p i 2
= + a%?,
fa=
+ x2z + i 3 s
iz = + +
X32 = 132
12
where
I2
A12
and
e,2
e22
e2
,.
-$14d2$2.
(A 7)
(A 8)
(A ()
(A 10)
(A 11)
(A 12)
Thus, there is a difference of $ti2 between the Weyl transform of l2 and the
ordinary quantum mechanical operator P. This difference was apparently
first noticed by Shewell [35] in connection with a general discussion of correspondence and symmetrization rules.
REFERENCES
[l] WEYL,H., 1931, The Theory of Groups and Quantum Mechanics (Dover).
[2] WIGNER,
E., 1932, Phys. Rev., 40, 749.
[3] GROENEWOLD,
H. J,, 1946, Physica, 12,405.
[4] MOYAL,
E. J., 1949, Proc. Camb. Phil. SOC.math. phys. Sn., 45, 99.
[ 5 ] BALESCU,
R., 1975, Equilibrium and Nonequilibrium Statistical Mechanics (John W i b
& Sons).
161 LOUISELL,
W. H., 1973, Quantum Statistical Properties of Radiation (John Wiley Lk
Sons).
[7] BILLING,G. D., 1978, Thesis, University of Copenhagen.
[8] BROWN,
R.C., and HELLER,
E. J., 1981, J. chem. Phys., 75, 186.
[9] HUTCHINSON,
J, S., and WYATT,R. E., 1981, Phys. Rev. A, 23, 1567.
[lo] ESBENSEN,
H., WINTHER,A., BROGLIA, R. A., and DASSO,
C. H., 1978, Phys. Rea.
Lett., 41, 296.
[ll] DAHL,J. P . , 1982, Physica A, 112, 439.
[12] GROSSMANN,
A., 1976, Commun. Math. Phys., 48, 191.
368
1019
[13] ROYER,
A., 1977,Phys. Rev. A, 15, 449.
[13]DAHL,J . P., 1982,Physica scripta, 25, 499.
[15] PODOLSKY,
B., and PAULING,
L., 1929,Phys. Rev., 34, 109.
[16] DAHL,J . P.,1982,Molecular Structure and Energy Scrambling, edited by J . Hinze
(Plenum).
[17] MCWEENY,
R., 1952,Acta crystallogr., 5, 463.
[18] SILVERSTONE,
H.J., 1968,J . chem. Phys., 48, 4098.
[19] LAURENZI,
B.J., 1969,Int. J . quant. Chem., 3, 489.
[20] MONKHORST,
H.J., and HARRIS,
F. E., 1972,Int.J. quant. Chem., 6, 601.
C.,
ARRIGHINI,
G. P., and MARINELLI,
F., 1979,Theor. chim. Acta, 53, 165.
[21] GUIDOTTI,
[22]MCWEENY,
R., 1953,Acta crystallogr., 6 , 631.
R. F., 1969,J.chem. Phys., 51, 4569.
[23] STEWART,
F.B., 1971,Gaussian Basis Sets f o r the Atoms H-Ne for Use in
[24] VANDUIJNJWELDT,
Molecular Calculations, IBM Technical Report RJ 945.
[25] DAHL,J . P., Mat. fys. Meddr., 39, No. 12.
[26] AVERY,J., 1978,Acta crystallogr. A, 34, 582.
R.L., 1974,Rep. Math. Phys. (Poland), 6 , 249.
[27] HUDSON,
L.I., 1955,Quantum Mechanics (McGraw-Hill).
[28] SCMIFF,
L., 1970,General Chemistry, 3rd ed. (W. H. Freeman & Co.), p. 125.
[29] PAULING,
G. E., and KAC,M., 1977,Phys. Reg., C 32, 1970.
[30] ZIFF, R. M., UHLENBECK,
[31] COHBN,L.,1979,J. chem. Phys., 70, 788.
[32] COHEN,L., 1966,J. math. Phys., 7, 781.
P. M., 1972,J. chem. Phys., 56, 3320.
[33] BADER,R.F. W., and BEDDALL,
1932,Proc. natn. Acad. Sci. U.S.A., 18,674.
[34] McCoy, N.H.,
J . R., 1959,Am. J . Phys., 27,16.
[35] SHEWELL,
369
R.F. OCONNELL
Depamnent of Physics and Astronomy, Louisiana State University, Baton Rouge,
LA 70803, U.S.A.
M.O. SCULLY
Max-Planck Xnstitut fur Quantenoptik, 0-8046 Gatching bei Munchen,
West Germany
and
E.P. WIGNER
370
PHYSICS REPORTS (Review Section of Physics Letters) 106. No. 3 (1984) 121-167. North-Holland, Amsterdam
and
Max-Planck h.stitu~fur Quantenoptik, 0-8046 Garching bei Munchen, West Germany
R.F. OCONNELL
Department of Physics and Ashonomy. Louisiana Slate Uniuersiiy, Baton Rouge. LA 7W3,U.S.A.
M.O. SCULLY
Max-Planrk Imtitut fur Quanienoptik. 0-8046Garching bei Munchen. West Germany
and
Institute for Modern Optics. University of New Mexico, Albuquerque. NM 87131, U.S.A.
E.P. WIGNER*
Department of Physics and Astronomy, Louisiana Slate Uniwrsity, Baton Rouge, LA 70803, U.S.A.
Contents:
I . Introduction
2. Wigner distribution
2.1. Properties
2.2. Associated operator ordering
2.3. Dynamics
2.4. An example
2.5. Statistics and second-quantized notation
3. Other distribution functions
4. Distribution functions in terms of creation and annihilation
123
126
126
132
135
142
I46
Normal ordering
Symmetric ordering
Anti-normal ordering
Examples
Distribution functionson four-dimensionalphase space
5. Conclusion
References
4. I .
4.2.
4.3.
4.4.
4.5.
1%
158
161
162
163
166
166
150
152
operators
* Permanenl address: Department of Physics, Joseph Henry Laboratory, Princeton University, Princeton, NJ 08540, U.S.A.
~~~~
PHYSICS REPORTS (Review Section of Physics Letters) 106, No. 3 (1984) 121-167
Copies of this issue may be obtained at the price given below. All orders should be sent directly to the Publisher. Orders must be
accompanied by check.
Single issue price Dfl. 29.00, postage included.
37 1
M.Hillery el al., Distribution functions in physics: Fundamentals
123
Ahslracf:
This is the first part of what will be a two-part review of distribution functions in physics. Here we deal with fundamentals and the second part
will deal with applications. We discuss in detail the properties of the distribution function defined earlier by one of us @PW) and we derive some
new results. Next, we treat various other distribution functions. Among the latter we emphasize the so-called P distribution, as well as the
generalized P distribution, because of their importance in quantum optics.
1. Introduction
It is well known that the uncertainty principle makes the concept of phase space in quantum
mechanics problematic. Because a particle cannot simultaneously have a well defined position arid
momentum, one cannot define a probability that a particle has a position q and a momentum p, i.e. one
cannot define a true phase space probability distribution for a quantum mechanical particle. Nonetheless, functions which bear some resemblance to phase space distribution functions, quasiprobability
distribution functions, have proven to be of great use in the study of quantum mechanical systems.
They are useful not only as calculational tools but can also provide insights into the connections
between classical and quantum mechanics.
The reason for this latter point is that quasiprobability distributions allow one to express quantum
mechanical averages in a form which is very similar to that for classical averages. As a specific example
let us consider a particle in one dimension with its position denoted by 4 and its momentum by p.
Classically, the particle is described by a phase space distribution Pcl(q,p). The average of a function of
the position and momentum A(q,p) can then be expressed as
(1.1)
The integrations in this equation are from --03 to +w. This will be the case with all integrations in this
paper unless otherwise indicated. A quantum mechanical particle is described by a density matrix p (we
will designate all operators by a *) and the average of a function of the position and momentum
operators, A(q, p ) is
(1.2)
(Tr Q means the trace of the operator 6).
It must be admitted that, given a classical expression A(q,p),
the corresponding self-adjoint operator A is not uniquely defined-and it is not quite clear what the
purpose of such a definition is. The use of a quasiprobability distribution, P&, p ) , however, does give
such a definition by expressing the quantum mechanical average as
(1.3)
where the function A(q, p) can be derived from the operator A(d,$)by a well defined correspondence
rule. This allows one to cast quantum mechanical results into a form in which they resemble classical
ones.
The first of these quasiprobability distributions was introduced by Wigner [1932a] to study quantum
corrections to classical statistical mechanics. This particular distribution has come to be known as the
372
I 3
Wigner distribution.? and we will designate it as P,. This is, and was meant to be. a reformulation of
Schrodingers quantum mechanics which describes states by functions in configuration space. It is
non-relativistic in nature because it is not invariant under the Lorentz group; also, configuration space
quantum mechanics for more than one particle would be difficult to formulate relativistically. However.
it has found many applications primarily in statistical mechanics but also in areas such as quantum
chemistry and quantum optics. In the case where Pa in eq. (1.3) is chosen to be P,, then the
correspondence between A(9,p ) and A is that proposed by Weyl [1927], as was first demonstrated by
Moyal [ 19491. Quantum optics has given rise to a number of quasiprobability distributions, the most
well-known being the P representation of Glauber [1963a] and Sudarshan [1963], which have also found
extensive use. As far as the description of the electromagnetic field is concerned, these do exhibit
(special) relativistic invariance. Other distribution functions have also been proposed (Husimi [ 19401;
Margenau and Hill [1961]; Cohen [1966]) but have found more limited use, although, more recently,
extensive use has been made of the generalized P representations by Drummond, Gardiner and Walls
[1980, 19811. In this paper we will discuss the basic formalism of these quasiprobability distributions and
illustrate them with a few simple examples. We will defer any detailed consideration of applications to a
later paper.
We now proceed to the basic problem: how do we go about constructing a quantum mechanical
analogue of a phase space density? Let us again consider, for simplicity, a one particle system in one
dimension which is described by a density matrix 6. In this paper we will work, for simplicity, in one
dimension; the generalization to higher dimensions will be given in a few cases but is in most
circumstances obvious. It is possible to express the position and momentum distributions of the particle
as
(1.4a)
(1.4b)
where S(9 - 4) is the operator which transforms 19) as follows:
(1.5)
and similarly for 6 ( p - P ) , We introduce the function p(q, 9) defined by
(1.6)
where wA is the probability of the system being in the state
$A,
h1.7a)
and
Pmo,(p)
=(2h)-
1dx \ dx p ( x , x) exp{ip(x
- x)/h}.
(1.7b)
t We use this designation here and throughout the paper despite the strenuous objections of one of us since the majority of us feel we should
adhere to what is now common nomenclature.
373
M.Hillery el al., Distribution functions in physics: Fundamentals
I25
To show that this corresponds to the usual definition we will examine Pp(q). We have that, in the
Dirac bracket notation,
(1.8)
which is a more conventional expression for the position density. A first guess for some kind of a phase
space density might then be
(1.9)
On the other hand, we might choose instead
(1.10)
But these expressions are not equal and although either of them, or a combination of both, could be
used to evaluate expectation values of functions of 4 and 0 (provided the operators are ordered
properly, the ordering for PI being different than that for &), they do not possess what we regard as
desirable properties (see section 2). In fact, they are, in general, not real.
The association of distribution functions with operator ordering rules (or, equivalently, the association of operators with classical expressions) is one which will recur throughout this paper. Each of the
distribution functions which we will discuss can be used to evaluate expectation values of products of
operators ordered according to a certain rule. We will consider distribution functions which can be used
to compute expectation values of products of the position and momentum operators 4 and 0,and also
distribution functions which can be used to compute expectation values of prclducts of the creation and
annihilation operators, d and d. The latter are useful in problems involving electromagnetic fields.
Because the creation and annihilation operators are simply related to 4 and rj there is a relation between
these two types of distribution functions. The Wigner distribution, for example, has proved useful in
both the 6, 6 and i, 4 contexts. The basic criterion for the choice of a distribution function for a
particular problem is convenience.
In the next two sections we will continue to examine distribution functions expressed in terms of both
the position and momentum variables. The Wigner function, Pw, will be discussed first in section 2 for
not only was it the first quantum mechanical phase space distribution to be considered, but also it
satisfies a number of properties which make it quite useful in applications. First of all, we will discuss its
properties and then show that Wigners distribution function gives the same expectation value for every
function of p and q as does the corresponding operator, as proposed by Weyl [1927], for the density
matrix which describes the same state to which the distribution function corresponds. As was mentioned
before, this was first observed by Moyal [1949]. Next we derive an equation, in many different forms,
for the time dependence of P,. Finally, we apply the formalism we have developed to the calculation of
P, for the eigenstates of the harmonic oscillator and also for the case of a canonical ensemble of
harmonic oscillators at temperature T.
In section 3 we discuss distribution functions other than P, which correspond to operator ordering
374
126
schemes different from that of Weyl-Wigner. Then in section 4 we treat distribution functions in terms
of creation and annihilation operators, with emphasis on normal, symmetric and anti-normal ordering.
In particular, we emphasize the normal ordering from which arises the well-known P distribution of
quantum optics. We also discuss the generalized P representations. Finally, in section 5 we present our
conclusions.
Applications will be treated in a future paper but we would be remiss not to mention the recent
extensive review of quantum collision theory using phase space distributions (Carruthers and Zachariasan [1983]) and the work on relativistic kinetic theory- in addition to extensive discussions on the
Wigner-Weyl correspondence - by the Amsterdam group (Suttorp and de Groot [1970]; Suttorp [1972];
de Groot [1974]; de Groot, van Leeuwen and van Weert [1980]). Also, a brief overview of some
applications is presented in OConnell [2983a,bl.
2. Wigner distribution
2.1. Properties
2.1)
was proposed to represent a system in a mixed state represented by a density matrix ii In the case of a
pure state, $, it follows from eq. (1.6) that p ( q , 4)= $(q) $*(q)and hence
(2.2a)
The latter result refers to one dimension. In the case of more than one dimension, the n h must be
replaced by (d)-,
where n is the number of the variables of $ (or the number of variables of the rows
or columns of 8) and 9, y and p are n-dimensional vectors, with py the scalar product of the two. The
integration is then over all components of y. Explicitly, eq. (2.2a) generalizes to
(2.2b)
It was mentioned that this choice for a distribution function was by no means unique and that this
particular choice was made because it seemed to be the simplest of those for which each Galilei
transformation corresponds to the same Galilei transformation of the quantum mechanical wave
functions. In later work Wigner [1979] returned to this issue by considering properties which one would
want such a distribution to satisfy. He then showed that the distribution given by eq. (2.1) was the only
hhj
M.Hillery er al.. Dishibution functions in physics: Fundamentals
127
one which satisfied these properties. A subsequent paper by OConnell and Wigner [1981a] considered
a somewhat different list of properties and showed that these, too, led to the expression in eq. (2.1).
The properties for a distribution function, P(q, p), which were considered of special interest, for the
case of a pure state (generalization to the case of a mixed state is straightforward), are as follows
(OConnell and Wigner [1981al):
(i) P ( q , p ) should be a Hermitean form of the state vector $(q), i.e. P is given by
(2.3)
where a ( 4 , p ) is a self-adjoint operator depending on p and q. Therefore, P(q, p) is real.
(ii)
(2.4)
(2.5)
(2.6)
(iii) P ( q , p ) should be Galilei invariant, i.e. if $(q)+$(q+ a ) then P ( q , p ) + P ( q + a , p ) and if
$h)+exp(ipqlh) $(q) then P(q, p ) + P ( q , P - P).
(iv) P(q, p ) should be invariant with respect to space and time reflections, i.e. if $(q)+ $(-q) then
P(q, p ) + P ( - q , -P) and if $(4)+$*(q) then P(q,p)+P(q, --PI.
It should be admitted, however, that neither of these transformations is relativistic and also that they
do not yet involve the spin variable.
(v) In the force-free case the equation of motion is the classical one
(2.7)
(vi) If P,(q, p ) and P+(q,p) are the distributions corresponding to the states $(q) and 4(q)
respectively then
2.8)
Property (vi) has two interesting consequences. If we set 444)= $(q) we get
(2.9)
and, in the case of a mixed state, the right-hand side of eq. (2.9) is multiplied by ZBwk where the wk are
the probabilities for the different states (the characteristic values of b). This implies that P,(q, p ) is not
too highly peaked and rules out such distributions as P,(4, p) = 6(q - 4) S ( p - p) which would be
376
12H
possible classically. We can also choose 4 and i,b so that they are orthogonal. We then have that
(2.lo)
which implies that P(9, p ) cannot be everywhere positive. This conclusion is actually rather general.
Wigner I19791 has shown that any distribution function as long as it satisfies properties (i) and (ii)
assumes also negative values for some p and q.
(vii)
(2.11)
where A(9, p ) is the classical function corresponding to the quantum operator
to Wigners prescription, by
A(q, p ) =
dz eip2*( q - fzl&
+ $2)
(2.12)
so that
d9 dpA(q, p ) = 2.rrh Tr(A). A similar relation exists between B(9, p ) and 6.
The proof of eq. (2.11) will be shown below to follow as a particular case of a more general relation
(eq. (2.23)) for F(9,p), in terms of A(9,p) and B(q,p),where E = Ab. From eq. (2.12), it is at once
is real if
is self-adjoint
evident that the phase space description A(q,p) of the operator
(Hermitean) and is imaginary if A is skew Hermitean. Since in neither case does A(9,p) vanish, it is
evident that if it is real, its operator A is self-adjoint, if it is imaginary A is skew symmetric. It is also
evident that the phase space description of the Hermitean adjoint
of A is the complex conjugate of
the similar description of A. Similarly, if the phase space descriptions of two operators are complex
conjugates of each other, then the operators are Hermitean adjoints of each other.
By comparison of eqs. (2.1) and (2.12), it is clear that P ( q , p ) , derived from the density matrix, is
(2?rh)- times the phase space operator which corresponds to the same matrix. Also, for A = 6 and B
equal to the unit matrix, eq. (2.6) immediately follows from eq. (2.11). Furthermore, for B = 6, eq.
(2.11) reduces to
a+
(2.13)
which is equivalent to eqs. (1.2) and (1.3). This result was originally obtained (Wigner [1932]) for the
special case of A being the sum of a function of j j only and a function of 4 only but Moyal [1949]
showed it was actually true in the case where A is any function of 9 and jj, if A(9,jj) is the Weyl
operator (discussed below in section 2.2) for A(4,p ) . In addition, if we take = B = p^ in eq. (2.11) and
use the fact that, if b represents a pure state, Tr@)2= Tr b = 1, we obtain eq. (2.9) again.
(viii) If we define the Fourier transform of the wave function
d ( p )= (2&)-
d9 $(q) e-iqph
I
(2.14)
377
M.Hillcry et al., Distribution functions in physics: Fundamenfals
129
dp 4 * ( p + p) 4 ( p - p) e-ziwlh,
(2.15)
(2.16)
gives the distribution function which corresponds to the properly contracted p (in square brackets).
Actually, this is true also for the other distribution functions which will be considered in section 3.
Wigner in his 1971 paper also showed that properties (i)-(v) determined the distribution function
uniquely. OConnell and Wigner [1981a] showed that properties (i)-(iv) and (vi) also accomplish this. In
both cases the distribution function was that given by eq. (2.1).
Finally, we draw attention to two restrictions on the distribution function discussed above. First of
all, as already mentioned, it is non-relativistic. Secondly, not all functions P(q, p) are allowed, as we will
now demonstrate by turning to the question of the admissability of P and asking what condition is
necessary so that P implies the existence of the density function b, the expectation values of which are,
naturally, positive or zero. Our starting-point is eq. (2.2a) from which it follows that
dp e-zipyhP(q, p ) = p(q - Y,q + y) .
(2.17)
dp e-ip(u-u)l*P(%u + v), P) .
(2.18)
We remark that since p on the right-side of eq. (2.18) is a dummy variable it is clear that it could be
replaced by q.
Now the condition for P(q,p) to be a permissible distribution function is that the corresponding
378
130
II
II
dx
(2.19a)
for all $. Using eq. (2.18) and eq. (2.19a), it follows that the condition that P(9, p) be permissible is that
dq
dpP(q, P ) P'(q, P ) 2 0
(2.19b)
for any P'(q, p) which corresponds to a pure state. This is evident already from eq. (2.8). It also follows
Eq. (2.19b) holds, of course, for any P' which is itself
from eq. (2.11) and the fact that Tr($)?O.
permissible but the permissibility of P follows already if it is valid for all P' which correspond to a pure
state.
Eight properties of the distribution function were discussed above, eqs. (2.3) to (2.16), with the
emphasis on the use of this function to form another description of a quantum mechanical state, i.e. be
a substitute for the density matrix. Just as eq. (2.1) permits one to give a phase space formulation to the
density matrix b, we emphasize that eq. (2.12) permits one also to give a phase space formulation to any
matrix -or operator - and it may be useful to consider the properties of eq. (2.12).
In particular, we wish to derive an expression for the function F(4, p) which corresponds to the
product F' = ad of two operators and B to which the q, p functions A(q,p) and B(q, p) correspond.
We assume that the operators
and B are matrices, the rows and columns of which can be
characterized by a single variable, but the generalization to a many-dimensional configuration space is
obvious. We can write, therefore
a
a
P(x, x") =
(2.20)
Analogous to eqs. (2.17) and (2.18), eq. (2.20) can be written as (taking h
II
(2.21)
Substituting x = q + q', X" = q - q', multiplying with e-zi9'pand integrating over q' one obtains
F(q, p) = 2 (272)-'
I\ I
dq' dx' dp' dp" A($(q + q' + x'), p') B(i(q - q' f x'), p")
x exp{-4i y'(p
=
16 (27r)-'
II1I
-p )
= $(q
(2.22)
- 4i p'(q - y)}
379
M. Hillery ei al., Dbnibulion funcfiom in physics: Fundamenrals
131
This expression for F(q,p), which is a new result, also shows the similarity of the roles of p and q in
Hamiltonian mechanics. In the next subsection, another expression (eq. (2.59)) for F(q, p) will be
presented .
If we integrate F(q, p) in eq. (2.23) over q and p, we obtain
(2.24)
Hence
(2.25)
Since the 1eft:hand side of this equation is the same as (2n-h) Tr(fi), it is clear that eq. (2.25) is the same
as eq. (2.11). In the case of n dimensions, the nth power of (4/.rr*) appears in the expression
corresponding to eq. (2.23).
Eq. (2.23) provides also a means to ascertain, in terms of the phase space descriptions of and B,
whether these two operators commute, Naturally, the condition for the commutative nature is
(2.26)
Since this is valid for all p and q, the integration over the variables which are their fa!tors i? the
exponent (i.e. y' and p') can be omitted. This gives as condition for the commutability of A and B (we
replace y, y' by q, q' and p, p' by p, p'):
x exp{4i(q'p + p'q)} = o ,
(2.27)
380
132
(2.28)
A Pw(q, p )
In order to simplify this, one can multiply with exp{4i(qp t pq)} and integrate over p and 9 to obtain,
substituting also 9 and p for the integration variables y and p.
(2.29)
Both eqs. (2.27) and (2.29) are a good deal more complicated than the quantum mechanical equations
for which they substitute. It is questionable whether they are really useful. We thought that they should
be derived in spite of this because the final form is considerably simpler than the original one and
because they clearly demonstrate the essential phase space equivalence of 9 and p . It may be worth
remarking finally that in the case of several dimensions all variables should be considered as vectors, and
products like 9p or p9 should be replaced by scalar products of these vectors.
a,
(2.30)
381
M.Hillery er a/.. Disfribuion functions in physics: Fundarnenrols
133
is equal to the expectation value of the classical function A(q,p) to which corresponds assuming that
the system is described by the distribution function Pw(4,p) which corresponds to 6. This is the content
of eq. (2.30) and it is valid, as will be demonstrated below, for every state vector (I and also or any
density matrix 6
(2.31)
Actually eq. (2.31) is an easy consequence of eq. (2.30) and only the latter will be proved below.
In order to prove eq. (2.30), we start with Weyl's expansion of A(4,p) into a Fourier integral (taking
h = 1 for the purpose of this proof):
A(4, p) =
du
(2.32)
d.rcr(u, 7) ei(Oq+'p)
Weyl then defines the operator which corresponds to the exponential in the integrand on the right-hand
side of eq. (2.32) as exp(i(mj t $)}. The operator which corresponds to A(q,p) is then given by
&,j)
(2.33)
(2.34)
The integration over p gives 27-r S(y t T ) and hence the right-hand side of eq. (2.34) becomes
In order to evaluate the left-hand side of (2;34) we note that according to-the Baker-Hausdodl theorem
(Messiah [1961]), if the commutator d. = [A, B ] commutes with and B then
eA+B = eA
es e-m2,
(2.35a)
= eiud ei'P
eim/12
(2.35b)
382
I34
Fundamenrnl
(2.36)
and transferring the eid to the left-hand side, this becomes
@(x)I@(x+ T ) ) =
eim/2(e-iux
@*W@(x+ 7)
(2.37)
which is equal to the expression obtained above for the right-hand side of eq. (2.34). Thus, we have
proved eq. (2.34) and hence also eq. (2.30).
In summary, if a classical function
(2.38)
(2.41)
as can be seen by considering the d""" coefficient in (cr4 + TO)"+'".
Finally, we would like to mention the role played by the characteristic function. This is a description
of the state by means of a function of two new variables, cr and T,
C(a,T) = Tr@ &(u,T)) ,
(2.42)
(2.43)
383
M.Hillery CI a/.. Distribution functions in physics: Fundamentals
135
Here we are following the nomenclature of Moyal which has now become standard in describing this
quantity as a "characteristic function". This description stems from statistical terminology, and, in
particular, should not be confused with the sometime usage of "characteristic function" in quantum
mechanics to denote an eigenfunction.
C(a,7)is just the Fourier transform of P(q, p). To see this we note that the function corresponding
to t ( u ,7)is just exp{(i/fi)(uq + 7p)). Making use of eq. (2.11) gives
(2.44)
so that
(2.45)
We can use the characteristic function to compute expectation values of Weyl-ordered products of p
and q. We have that
(2.46)
the right-hand side of which is just the average of the Weyl-ordered product q"p".
2.3. Dynamics
We would now like to derive equations for the time-dependence of P,. As before, our detailed
considerations will be confined to one dimension but some results will also be quoted for the
multi-dimensional case. The time-dependence of P, may be decomposed into two parts (Wigner
[1932a1)
(2.47)
the first part resulting from the (ih/2rn) a2/aq2part, the second from the potential energy V/ih part of
the expression for a+/at.
From the definition of Pw, given by eq. (2.2a), it follows that
(2.48)
where we have taken advantage of the functional dependence of $ to replace dZ/aqZby a21dyz. Next we
perform one partial integration with respect to y to obtain
(2.49)
384
136
since the boundary term does not contribute. Switching back from a/ay to a/aq, we finally obtain
(2.50)
This is identical with the classical (Liouville) equation for the corresponding part of dP/at, as was
mentioned at eq. (2.7). We next calculate
(2.5I)
Assuming that V can be expanded in a Taylor series, we write
V(q + Y) =
25
W 9 )
(2.52)
A =O
-- 2i
at
?rh2
dy
A
(2.53)
where now the summation over A is restricted to all odd positive integers. It is clear that in the powers
y A in the integrand we can replace y by (h/2i)(a/ap). It then follows that
(2.54)
A again being restricted to odd integers. An alternative form for &PW/at is given by
(2.55)
where
(2.55a)
is the probability of a jump in the momentum by an amount j if the positional coordinate is 9. The first
part of eq. (2.55a) may be verified by inserting the Fourier expansion, with respect to y, of V(q + y) -
385
M. Hi/lery el al., Distribution functions in physics: Fundamentals
137
V(q - y ) into eq. (2.51). The second part is obtained by replacing the exponential by c o s t i sin and
noting that the expansion in the square bracket is odd so that the integral of the cos part vanishes.
In the multi-dimensional case where P, = P,(ql, . . . 4,; pl, . . .pn), the corresponding results are
(2.56)
where the last summation has to be extended over all positive integer values of Al, . . . An for which the
sum A 1 t A,+
* t A, is odd.
The lowest term of eq. (2.56) in which only one A is 1 and the others vanish, and which has no h
factors, is identical with the corresponding term of Liouville's equation. Hence eq. (2.56) reproduces the
classical (but non-relativistic) equation if h is set equal to zero. The h 2 terms give the quantum
correction if this is very small. We will obtain a somewhat similar equation for the 1/T dependence of
the distribution function of the canonical ensemble, which also is useful if the temperature T is not too
low so that the quantum correction is small.
Eq. (2.56) is the generalization of eq. (2.50) and eq. (2.54) for an n-dimensional configuration space.
The same generalization of eq. (2.50) with eq. (2.55) is
(2.57)
where J(ql,.. . q,; jl, . . .In)
can be interpreted as the probability of a jump in the momenta with the
amounts jl,. . .j, for the configuration q l , . . . 4.. The probability of this jump is given by
(2.58)
that is, by the Fourier expansion coefficientsof the potential V(ql,.. . 4,).
From eq. (2.56) it is clear that the equation of motion is the same as the classical equation of motion
when V has no third and higher derivatives as, for example, in the case of a uniform electric field or for
a system of oscillators. However, there is still a subtle difference in that the possible initial conditions
are restricted. This comes about because not all P(q, p ) are permissible (see eq. (2.19b)).
While we consider that the above form for the equations of motion (Wigner [1932]) are the simplest
to use in practice, we will now discuss some other forms which occur frequently in the literature.
Before doing so it is useful to take note of another relation, ip addition to that given by eq. (2.23),
which expresses the Weyl function corresponding to an operator F = AB in terms of the Weyl functions
corresponding to and B. This relation was first derived by Groenewold [1946] and was also discussed
by Imre, Ozizmir, Rosenbaum and Zweifel [1967]. They find that the function corresponding to fi is
(2.59)
386
138
where
(2.60)
and the arrows indicate in which direction the derivatives act. Also (8/8p)(d89) is considered as the
multi-dimensional scalar product of dldp and 8/89, or, in other words, it is equal to (8/dpi)(8/89,), where
i = ( l , , . . n) and n denotes the number of dimensions and, as usual, repeated indices denotes
summation.
To derive this result we first note that
(2.61)
where a is defined by eq. (2.32). This result follows from eq. (2.33) by taking the matrix element of both
sides. A similar result follows for (q@(q') except that a is replaced by p, the Fourier transform of
B(% P):
(2.62)
We can now calculate F(9, p). We have from eq. (2.12) that
(2.63)
We now define two new variables of integration r = 9' - 9 + (z/2) and r'
=9
It is possible to replace the exponential factor exp{(i/h) (U'T- mf)/2} by exp(hA/2i) so that eq. (2.64)
becomes
(2.65)
i.e. just the first expression appearing on the right-hand side of eq. (2.59). The second expression also
follows readily from eq. (2.64).
387
M. Hillery ef al., Dicbibuh'on funcfionr in physics: Fundamentals
139
We can also make use of eq. (2.64) to find an alternative expression for F(q,p) involving the Bopp
operators (Bopp [1961] and Kubo [19641)
Q=q--2i ap '
h a
2i aq '
(2.66)
so that
(2.68)
Using this result in eq. (2.64) we then have that
(2.69)
(2.70)
is just the Weyl-ordered operator A(4, j)with 9 + Q and jj P. A(Q, P) is also an operator but not on
is an operator; it operates on functions in phase space. We can,
the Hilbert space on which
therefore, express F(q, p) as
&,a)
(2.71)
(2.72)
where
h a
Q * = q t -2i- ,ap
h a
P * = P - - 2i
- ilq *
(2.73)
It is now possible to make use of the fact that the Wigner distribution is the function which is
associated with (1/27rh)i. The equation of motion for C is just
ih ab/a~
= [fi,81.
(2.74)
388
141)
or
(2.75)
where H(q, p ) is the function corresponding to the Hamiltonian operator for the system, fi. Actually,
this is an abbreviated form of eq. (2.56) as can be verified by expanding the sin into a power series. Note
that if we take the h .-+ 0 limit of this equation we obtain the classical Liouville equation
(2.76)
where { } denote Poisson brackets and the superscript c on P, indicates the classical limit. For an
H ( q , p ) which is at most quadratic in 4 and p , e.g. a free particle or an harmonic oscillator, eqs. (2.75)
and (2.76) coincide. In these systems, then, the difference between a classical and a quantum ensemble
is the restriction on the initial conditions in the case of latter (cf. eq. (2.19)).
We also want to quote two alternate forms of eq. (2.75). The first follows immediately from our
discussion of the Bopp operators. We have, using eqs. (2.65), (2.71), (2.72) and (2.75), that
ih aP,/at
[H(Q, P ) - f i ( ~ *P *. ) ]~
~ (p )9 ,
(2.77)
a result first obtained by Bopp 119611. Analogous to the definition of A ( 0 , P ) ? given by eq. (2.70).
Q and .-+ P, where 0 and P are defined in eq. (2.66).
These equations do not exhaust the possible formulations of the dynamics of the Wigner function. One
can also make use of propagation kernels. This approach is discussed by Moyal [1949] and Mori,
Oppenheim and Ross [1962].
We turn now to a consideration of a canonical ensemble. If /? = l/kT where k is Boltzmanns
constant and T is the temperature, then the density matrix of the canonical ensemble is
(2.78)
and Z ( p ) = Tr(e-OH). The unnormalized density matrix, d!, then satisfies the equation
(2.79)
subject to the initial condition fi@? = 0) = i where f is the identity operator. Eq. (2.79) is referred to as
the Bloch [1932] equation for the density matrix of a canonical ensemble. Using the product rule given
by eq. (2.59) we have that
(2.80)
A being given by eq. (2.60) so that
389
M.H i l h y el a/., Disfribufionjuncfions in physics: Fundarnenfals
141
(2.81)
This is the Wigner translation of the Bloch equation, which was entensively studied by many authors
and was first derived in this form by Oppenheim and Ross [1957]. It is useful in the calculation of
quantum mechanical corrections to classical statistical mech?nics. The initial condition for this equation
is just the Wigner function corresponding to h(p = 0)= I. Inserting I in eq. (2.12) we find that the
initial condition is just Jl(q, p)la-o = 1.
It is also worth noting that P,(q,p) does not satisfy the Wigner translation of the Bloch equation
simply because of the fact that it must be multiplied by the P-dependent factor ( 2 h ) Z ( p ) in order to
obtain O(q,p).
Finally, we emphasize that all equations from eq. (2.59) onwards hold in the multi-dimensional case,
where we simply interpret (4,p) to be (ql, . . . q,; pl, . , .p,) and the simple products in the exponents as
scalar products. The solution of eq. (2.81) in the multi-dimensional case, is to order hZ(Wigner [1932a]),
(2.82)
Actually, the Wigner translation of the Bloch equation, eq. (2.18) above, can be simplified further into
a form, analogous to that of eq. (2.56), which is more convenient for applications. This is achieved by
writing the cos term as the real part of the operator
(2.83)
where we have used the explicit form for A given in eq. (2.60), again noting that the arrows indicate in
which direction the derivatives act and that the gradient operators are 3N-dimensional. Next we
decompose 8 by means of the Baker-Hausdorff theorem (eq. (2.35a)), and using the fact that
(2.84)
it follows that we may write
(2.85)
where we have neglected terms which do not contribute in the present context. Again because of eq.
(2.y),
and also using the fact that we are only interested in the real part, it follows that the only terms
in 0 which contribute are
(2.86)
where i, j = 1, . . . n (and as usual, it is understood that ( d a q ) ( d d p ) stands for (a/aqi)(slap,)). From
henceforth, we wit1 assume that we are dealing with a system of ( 4 3 ) identical particles of mass m.
390
142
H O a = exp----ifi a a V---th Z a
{ [
2 39 apl
+?3
2m 892 2m
(2.87)
where it is to be understood that the (&as)term in the exponential operates only on V and not on R
(whereas the p/a9 term operates on 0).Also, the 8 a p term has no effect on V and thus operates only
on 0. Since all arrows now operate to the right, they will be omitted from henceforth so that we finally
obtain
(2.88a)
(2.88b)
where the 8/39 term in the cos and sin terms is to be understood as operating only on V. Such a form
was given for the first time by Alastuey and Jancovici [1980] and, in fact, their result also takes account
of the presence of a magnetic field. We recall that (3/3p)(8/89) is considered as the multi-dimensional
scalar product of 8/8p and 439, or, in other words it is equal to (a/@,) (8/89,)where i goes from 1 to n
and n denotes the number of dimensions. Hence, the explicit form of eq. (2.88a) is
(2.89)
where the last summation is to be extended over all positive integer values, as well as zero values, of
A ,, A, . . . A,,, for which the sum A I t A 2 t * * t An is even. This form for the Wigner translation of the
Bloch equation is the most convenient from the point of view of applications.
One of the earliest applications of these results was to the quantum corrections of the classical
equations of state and to similar corrections to chemical reaction rates (Wigner [1932b, 19381) and they
have been extensively used in statistical mechanics (Oppenheim and Ross [1957]; Mori, Oppenheim and
Ross [1962]; Nienhuis [ 19701, for example). However, we will defer a detailed discussion of applications
to Part I1 of our review, to be published at a later date.
1
2.4. An example
We would now like to use some of the formalism which we have developed to actually calculate some
distribution functions. The system which we will consider is the harmonic oscillator and we will consider
both pure and mixed states. We will find the Wigner functions corresponding to the eigenstates of the
harmonic oscillator and also the function corresponding to a canonical ensemble of harmonic oscillators
at temperature T.
The eigenstates of the harmonic oscillator are (Landau and Lifshitz [1965])
(2.90)
39 1
M. Hillery er al., Distributionfunctions in physics: Fundamenfah
143
where H n is the nth Hermite polynomial and a = (mw/h)lR. Substituting this expression into the
definition of the distribution function, eq. (2.2a), we find that
a2
u*.(q+Y) un(q-Y)=
1/2
(7) , , , e ~ ~ { - ~ * t ( q + Y ) ~ + ( q - y ) ~ 1 / 2 } ~ n ( a ( q + ~ ) ) . ~ . ( a ( q - y ) )
(2.91)
so that
(2.92)
We now note that
+ p2/aZh2
(2.93)
(2.94)
(2.97)
where L, is the nth Laguerre polynomial. Re-expressing a and p in terms of q and p we have
(2.98)
392
144
translation of the Bloch equation (eq. (2.88b)) which for this system results in
(2.100)
Because V is quadratic in q2 it is clear that only the leading order term in the sin2 expansion will
contribute, and since d2V/dq2= mw2, it follows that the Wigner translation of the Bloch equation for
the oscillator reduces to
(2.10 1)
To solve this equation we make the Ansatz
R(q,P) = exp{-A(P) H t
W)}
(2.102)
where A(0) = B(0)= 0, and H = (p2/2m)t imoq2. Substituting this into eq. (2.101) gives us
(2.103)
This equation can be re-expressed in the form
(2.104)
Because this equation must hold for all q and p, and the terms in the brackets are independent of q and
(2.106)
Eq. (2.105) can be integrated directly. One has that
dA
-1dp
1 - ( h ~ / 2 A
) ~-
(2.107)
or
hw In
[(1 t
A ) / (1 -
)I
(2.108)
393
M.Hiky et a/., Dishiburion funcfions in physics: Fundamentals
145
(2.109)
d#l' t a n h ( y ) = -In c o s h ( y )
B @ )= -
(2.110)
Therefore, we have
(2.111)
To complete our derivation we need to normalize the above expression. As was noted before the
Wigner function is the function which corresponds to the operator @/2?rf)).From eq. (2.78) we then
have
. .
(2.112)
p ) is just the function corresponding to e-@'. We also have from eq. (2.11) (setting
as
and B = I)
Z(B) = Tr(e-@') =
a
' I dqI dp O(q,p)
a = e-@
(2.113)
Z@)= f[sinh(hwp/2)]-' .
(2.114)
Finally we obtain for Pw(q,p), from eqs. (2.111), (2.112) and (2.114),
(2.115)
We now want to compare the two expressions (eq. (2.99) and eq. (2.115)) for P, for the pure and
mixed states, respectively. Examining the first few Laguerre polynomials
L&)
=1
L,(x)= 1 - x
(2.116)
L,(x) = 1 - 2x + x2
we see that for the ground state of the oscillator Pw(q,p ) > 0 while for excited states P,(q, p) can assume
negative values. The result for the canonical ensemble, however, is always positive. It does not have the
oscillatory structure which is present in the expressions given by eq. (2.99). The incoherence induced by
a finite temperature leads to a much smoother distribution function.
146
2.5. Statistics and second-quantized noration (Klimintovich i1958); Brittin and Chappell [1962]; Imre,
Ozizmir, Rosenbaum and Zweifel [1967])
When one is dealing with more than one particle one has to include the effects of quantum statistics.
To illustrate how these effects come in to the Wigner function we will first consider an example. We will
then show how the Wigner function can be expressed in second-quantized notation. In this form it is
easier to take the effects of statistics into account, but two of us have an article in preparation
(OConnell and Wigner [1983]) which not only will take the effect of statistics into account, but will also
include spin effects.
Let us consider two identical particles in one dimension in a harmonic potential well. We will further
assume that the particles are bosons. The Hamiltonian for the system is
(2.117)
Suppose that we want to find the Wigner distribution for a canonical ensemble of these systems at a
temperature T. We would again like to use the Wigner translation of the Bloch equation but now we
must be more careful; the initial condition is no longer so simple.
To see this we first find the density matrix for the system. The eigenstates of the Hamiltonian given
by eq. (2.117) are
(2.118)
(2.118)
(2.119)
(2.119)
(2.120)
(2.120)
d~
= 0) =
I4nim)
(4ninJ
(2.121)
(2.121)
nisn2
(2.122)
We can now make use of the identity
395
hi.Hillrry el 41.. Distribution f u d m in physics: Fundomenlrrlr
147
(2.123)
(2.123)
to give
w;-
q2)+
6(q;
42)
S@-
9111,
(2.124)
(2.124)
(2.125
If t+h is symmetric the result on the right-hand side of eq. (2.125) is +, if
0.
Therefore, h(j3= 0) is just the projection operator, pssay, onto the state of symmetrictwo-particle wave
functions. This result is also true for an arbitrary number ofparticles, N. Our result that d(g = 0) is was
derived for bosons. Similarly, if the particles are fermions O(p = 0) is PA,the projection onto the space of
anti-symmetric N-particle wave functions, but in this case, the spin variable should also be included.
Returning now to our example we want to find the initial condition for the Wigner translation of
Bloch equation, i.e. we must find the function corresponding to ts.
Making use of the two-particle
extension of eq. (2.12) we find
(2.126)
The corresponding result for fermions has a minus sign in front of the second term. This initial
condition is considerably more complicated than the initial condition, R(q,p) = 1, which was obtained
in the one-dimensional case. The situation rapidly becomes worse with larger numbers of particles.
Second-quantized notation provides, in principle, a convenient way to deal with the problems
imposed by quantum statistics, We will consider a Fock space an! designat! the vacuum state of this
space by lo), and the quan$zed field operators at the point r by $+(r) and $(r). The interpretation of
the field operators is that $'(r) adds a particle at point r to the system whereas $(r) removes a particle
at point r. They are defined as
(2.127a)
(2.12%)
396
148
where the so-called annihilation and creation operators, Cip and id, respectively (discussed in detail in
section 4), act to remove or create a particle of momentum p in a box of volume V. For bosons these
operators obey the commutation relation
(2.128a)
(2.128b)
and for fermions the anti-commutation relation
(2.129a)
(2.129b)
To every N-particle state IPN)in the Fock space corresponds an N-particle wave function given by
(Schweber [1961])
(2.130)
then, is given by
The distribution function for the state I PN)
(2.131)
(2.132)
where, in the case of a pure state,
(2.133)
with I@,) denoting the N-particle ket basis vector. An N-particle density matrix has the property that if
QN, and QN- are "-particle and N"-particle states respectively, then (@N"(bNl@N.)
= 0 unless N ' = N " =
N. Therefore, eq. (2.132) can be expressed as
397
M. Hillcry et al., Distribution functions in physics: Fundamentoh
149
(2.134)
This is the desired expression for the Wigner function in second-quantized form (Brittin and Chappell
[1962]; Imre, Ozizmir, Rosenbaum and Zweifel [1967]).
It is also possible to derive expressions for the reduced distribution functions in terms of the
quantized field operators (Brittin and Chappell [1%2]; Imre, Ozizmir, Rosenbaum and Zweifel [1967]).
The distribution function of order N, reduced to the jth order, is defined as
(2.135)
and this definition will be used for the rest of this section. This can also be expressed, by making use of
eq. (2.134), as
(2.136)
(2.137)
where N is just the number operator. We then have that, for both bosons and fermions
A] =
(2.138)
Therefore,
(2.139)
and
398
~
~~
150
(2.140)
x Tr(bN$(rl
+ iyl). . . $(rj
t iyi) N
* * *
(2.141)
. $(rl - 4~1)).
(2.142)
(2.143)
It is now possible to formulate the dynamics of this theory in a way which is independent of the
number of particles. We first go to the Heisenberg picture in which the field operators become time
dependent. We then consider the operators
1
.J d3yjexp(i(p, - y lt . . t pi -yj)/h}
it(,-,
t iy,; t ) . . . ijt(rj + +yj;t ) . . . J(rj - iyj; t ) . .
J ( r l - tyl; t ) .
(2.144)
(2.144)
(2.145)
We see that in this formulation all of the dynamical information is contained in the operators which
contain no reference to a specific particle number and also contain the information about the statistics of
the particles. Thus, in principle, the second-quantized formalism should be a useful starting-point for
the incorporation of statistics into problems involving a system of identical particles. However, it must
be admitted that - to our knowledge - no application has been made along these lines.
399
M.Hillery cf a/., Disbibution funcfionr in physics: Fundamentals
151
symmetric ordering
qnPn
+I(2
q
- m A n+
pqn),
(3.1)
in which case we would use the distribution function (Margenau and Hill [1%1]; Mehta [1964])
(3.2)
On the other hand, we may want to consider a distribution which is always greater than or equal to
zero. We will discuss a distribution which has this property shortly.
A scheme for generating distribution functions was proposed by Cohen [1966] and further examined
by Summerfield and Zweifel [1%9]. They give the rather general expression
(3.a)
(3.3b)
for the distribution function of the pure state $(4), where
(3.4)
Thus the function P is simply the original function P,,,smeared with another function g. The basic
requirement which leads to eq. (3.3) is that P transform correctly with respect to space displacement,
+(q)+ $(q - a), and transition to a uniformly moving coordinate system, $(q) + exp(-imuq) +(q).
These requirements were formulated in giving the form eq. (3.3a) to P,-and the satisfaction of the
requirements can easily be verified; eq. (3.3b) then follows.
Cohen also pointed out that it is possible to obtain distributions whose dependence upon the wave
function of the system is other than bilinear simply by choosing g(a, T) to depend upon $(q). For
example, one can choose
(3.5)
where qo is an arbitrary value of q. This choice for g(a,T) satisfies g(0, T) = g(a,0) = 1 so that the
correct marginal distributions are obtained. On the other hand, we now have the rather awkward situation
that the function-operator correspondencedepends upon the wave function. An even simpler choice is, of
course
(3.6)
where d(p),the Fourier transform of $(q) is defined by eq. (2.14). The conditions on g(q, p) which must
be satisfied so that the correct marginal distributions are obtained are
400
152
(3.7a)
(3.7b)
(3.7b)
One choice of g(q, p ) which does not satisfy eqs. (3.7) but which is interesting nonetheless is given by
(3.8)
The use of this smearing function was first proposed by Husimi [1940] and has been investigated by a
number of authors since (Bopp [1956]; Kano [1965]; McKenna and Frisch [1966]; Cartwright [1976];
Prugovecki [1978]; OConnell and Wigner [1981b]). It leads to a distribution function, pH(4, p ) , where
the subscript H denotes Husimi, which is non-negative for all p and q. One can see this by noting that
g(q - 9, p - p; a ) is just the Wigner distribution function which one obtains from the displaced (in both
position and momentum) harmonic oscillator ground state wave function
(3.9)
which we will call Pq,p(OConnell and Wigner [1981b]). If the Wigner distribution in question, P4,
corresponds to a wave function 4(9) we have
(3.10)
(3.10)
where we have used eq. (2.8).Note that in order to get a positive distribution function we had to violate
condition (ii) on our list of properties of the Wigner function. Property (vi) is also violated as was shown
by Prugovecki I19781 and by OConnell and Wigner [1981a].
We will encounter PH(9,p ) again in the next section in a somewhat different form. It is the 0or
anti-normally-ordered distribution function of quantum optics. It is one of a number of distributions
which are useful in the description of harmonic oscillators, and, hence, modes of the electromagnetic
field. We now proceed to examine these distribution functions.
40 1
M.Hillery er 01.. Dishibufionfwctiom in physics: Fundamentals
153
[19701). The latter is developed within the framework of annihilation and creation operators for bosons (see
below) but it is then possible to go to a description in terms of c-numbers (while fully retaining the quantum
aspects of the situation) by means of distribution functions. In most cases, this greatly facilitates the
calculation while, at the same time, it contributes to a better understanding of the connection between the
quantum and classical descriptions of the electromagnetic field.
A number of studies of these distribution functions have been done (Mehta and Sudarshan [1965];
Lax and Louisell [1%7); Lax [1968]; Cahill and Glauber [19691; Agarwal and Wolf [1970); Louisell
[1973]). We will rely most heavily upon the papers by Cahill and Glauber [1969] in our treatment. Their
discussion considers a continuum of possible operator ordering schemes, and hence distributions (an even
larger class is considered in Agarwal and Wolf [1970]) but we will consider only three of these. A final
section will discuss distributions defined on a Cdimensional, rather than a 2-dimensional, phase space.
We will describe the system in terms of its annihilation and creation operators
(4.la)
(4.lb)
satisfying
(4.2)
As mentioned before, it is assumed that the field operators we consider obey Bose statistics. Each
pair of Ci, d' refers to a certain function of position. These functions form an orthonormal set which is
countable if the basic domain is assumed to be finite, and continuous if infinite. We deal with a very
large, but finite, system so that the system is only approximately relativistically invariant (exact
invariance is achieved for an infinitely large system, but this would make the calculation in other ways
difficult).
The various functions of d and d' are investigated individually because the corresponding 6 and 2'
do not interact with the d and d' of another member of the set. They interact with the matter which is
in the basic domain. Thus, for example, when we apply this formalism to the case of the electromagnetic field, we investigate each mode (corresponding to a definite momentum and definite direction of
polarization) separately, and the operators associated with different modes commute (no interaction
between modes). In addition, there will be a distribution function corresponding to each mode.
The ci and d' operators act on the basis vectors In),the so-called "particle number states", and have the
properties:
d In) = d; In - 1)
(4.2a)
(4.2b)
(4.2~)
d 10) = 0 .
(4.2d)
402
M. Hillery er a/.. Disfribufionfunctions in physics: Fundamenrah
I54
(4.2e)
and if we are
If we are considering an oscillator of mass m and angular frequency w we take A =
considering a mode of the electromagnetic field of angular frequency w we set A = (h '"w/c).
We also want to consider a special class of states known as coherent states (Schrodinger [1926];
Glauber [1963a]; Glauber [1963b]; Sudarshan [1963]; Glauber [1965]). To define these we first define
for each complex number a the unitary displacement operator:
&)
= e(m5+-m-h) - e-la12/z
ea6+ e-u'h
(4.3)
where the last expression is obtained by use of the Baker-Hausdorff theorem (eq. (2.35a)) and the
commutation relation given by eq. (4.2). The operator d(a)has the property that
P ( a )8 B ( a )= 8 t a
(4.4a)
(4.4b)
The proof of eq. (4.4) readily follows from eqs. (4.2e) and (4.3). We now define the coherent state
(Glauber [1963a]; Glauber [1963b]; Sudarshan [1963]), which we denote by la), as
(4.5)
where 10) is the ground state of the oscillator. This state has the property that it is an eigenstate of the
annihilation operators with eigenvalue a. Again, this can be verified by using eq. (4.2e). Perhaps it
should be emphasized that the symbol a always refers to a complex eigenvalue whereas la) always
denotes a state, just as n denotes a real eigenvalue and In) a state, the so-called "number state". Also,
just as In) refers to a definite state of excitation of a system of one mode, (a)also refers to a state of one
mode.
The la) states are not orthogonal but they are complete (in fact overcomplete). Explicitly,
(4.6)
which follows immediately from eq. (4.5) and the fact that the number states are orthonormal.
Furthermore, it is possible to express the identity operator as
(4.7a)
where d2a = d(Re a ) d(Im a)= 4da da*. The proof of eq. (4.7a) follows by setting a = r e", so that
d2a = r dr do, and then using eq. (4.5) to get
(4.7b)
403
M.H i h y CI a/., Distriburion functions in physics: Fundomenfals
155
where we have used the fact that the angular integral simply equals 27rS,,,,. The latter radial integral
equals n!/2, so that using the fact that Z, In) (nl = 1, eq. (4.7a) readily follows. A direct consequence of
eq. (4.7a) is that the trace of any operator d is just
1
Tr(d) = - d2a ( a l a l a ) ,
(4.8)
7T
It is also of use to compare the expression for the displacement operat9r &a) to our*previousresults
and use this comparison to derive an expansion for a general operator A in terms of D-(a).This will
be of use later. First we note that if we set a = (2h)-2 (AT t iA-a) then (see eq. (4.1))
&,
71,
(4.9)
where
was defined earlier by eq. (2.43). Thus from eqs. (2.42) and (4.9) it is clear that the
characteristic function is the expectation value of the displacement operator. This in conjunction with
eqs. ( 4 3 , (4.6) and (4.8) gives
T&a)
8-(/3))
= P S0)(a- p ) ,
(4.10)
where the 6 function here is a(() = S(Re 6 ) 6(Im 6). Suppose that we can expand the operator a(d, a+)
as
(4.11)
Using eq. (4.10) we find that
g(6) = Tr(d
(4.12)
It can be shown (Cahill and Glauber [1969]) that if d is HilbertSchmidt (i.e. Tr(a+a)< a) then the
function g(5) is square integrable.
The three types of ordering of the operators a^ and 4 which we wish to consider are defined as
follows:
(i)Normal ordering - A product of rn annihilation operators and n creation operators is normally
ordered if all of the annihilation operators are on the right, i.e. if it is in the form (ci+) ci.
(ii) Symmetric ordering - A product of m annihilation operators and n creation operators can be
ordered in (n t rn)!/n!m! ways. The symmetrically ordered product of these operators, denoted by
{(ci) d } , is just the average of all of these differently ordered products. For example
(4.13a)
(4.13b)
(4.13~)
(iii) Anti-normal ordering - A product of m annihilation operators and n creation operators is
404
I Sh
M.Hillery
anti-normally ordered if all of the annihilation operators are on the left, i.e. if it is of the form i(6+).
For each operator ordering we have a rule which associates a function of a and a * with a given
operator. The rule is as follows: for any operator ordering scheme the product of m annihilation and n
creation operators, ordered according to that scheme, is associated with the function (a*)am.For
example, if we are considering normal ordering the product ( d ) i m is associated with ( , * ) a m ;if
anti-normal ordering is being considered then i m ( d + > , is associated with a m ( a * ) We
, will now make
the meaning of our rule more explicit by considering each of these orderings and its associated
distribution function.
4.1. Normal ordering
where P(a) is a c-number and the state la) is given by eq. (4.5). P(a) is called the P-representation of
the density matrix (or the distribution function representing the density matrix) of the particular mode
under study. It should be emphasized that both the real and imaginary parts of a are used as the
variables of the distribution function. Also, it is probably worthwhile mentioning again that our
discussion is restricted to a system of bosons and thus the distribution functions under study are not
applicable to, for instance, a gas of neutrinos. Also, we are dealing with a very large but countable set
since we assumed that the basic domain is finite.
From eqs. (4.10) and (4.7a) and because ( ~ l h ~ 6 l=a a*na,
)
it follows that
d2aP(a)
[2
n.m = O
c,,(a*)a
1=
d2aP(a)AN(a,a * ) ,
(4.16)
with
(4.17)
Therefore, we associate the operator A(Ci,ci) with the function AN(a,a*) in the evaluation of
expectation values with the P-representation,
We now want to derive two expressions for P ( a ) in terms of the density matrix. It is not always
possible to find a useful representation of b of the form given by eq. (4.15). For some density matrices
P ( a ) would have to be so singular that it would not even be a tempered distribution (Cahill [1965];
Klauder and Sudarshan [1968]). This difficulty will be apparent in our formal expression for P(a).
405
157
a=
e-owa
(4.18)
The corresponding function is then AN(a,a*)= exp([a* - [*a).Inserting these expressions into eq.
(4.16) we find that
d2a P ( a )eb"'-t'" .
(4.19)
The function ,yN([) is known as the normally ordered characteristic function. The right-hand side of eq.
(4.19) is just a Fourier transform in a somewhat disguised form. In fact one has that if
(4.20a)
then
f ( [ )=
(4.20b)
P ( a )= P
d2[ e"t'-"'fx,(~).
(4.21)
The problem with this expression is that xN([) can grow rather rapidly. In fact we have that because
exp([d+ - [*S) is unitary
(4.22)
which suggests the type of behavior which is possible. For example, if p = In)(nl, where In) is the
eigenstate of the number operator with eigenvalue n, then for large ( 1 we have IxN([)l ([I2". This
representation, then, is not appropriate for all density matrices, but, nonetheless, is useful in many of
the cases of interest.
Finally, we will derive an expression for P ( a ) in terms of a series expansion for the density matrix.
Let us suppose that we can express the density matrix as an anti-normally ordered series
p=
pnrndrn(d+)n.
(4.23)
n.m - 0
A(& Ci+)
406
158
M.Hillery
so that
c
m
1
Tr(&d)=;Id%
C pnmcrsa*nama*rcxr.
(4.25)
n.m=O r,s=O
(4.26)
n.m =O
The difficulties which we had when considering eq. (4.21) suggest that we will have similar problems
with eq. (4.26). In fact the problem goes back to eq. (4.23). The class of operators for which a
meaningful anti-normally ordered expansion exists is highly restricted. One can see this by considering
the representation for an operator given by eq. (4.11). Expand &I(() = exp((*d) exp(-@+) exp($l[l)
in an anti-normally ordered power series and insert it back into eq. (4.11). This gives us an
anti-normally-ordered power series for A :
a= c
m
dn,dm(i+)n,
(4.27)
n.m=O
d26Tr(AD(6)) e1612/2(-~)(~*)m
(4.28)
For these coefficients to exist Tr(Ah(6)) must be a very rapidly decreasing function of 161. Our previous
remarks indicate that this will not be true in general for Hilbert-Schmidt opTr!tors and, in fact, will not
be true in general for operators of trace class (operators, A,for which Tr([AtA]) < m) such as density
matrices.
It should be mentioned that normally-ordered power series expansions are far better behaved. A
derivation similar to the one above gives for the coefficients cnm in eq. (4.14)
c,,
liL
n! m .
IT
(r*)
(4.29)
This clearly exists for a much wider class of operators than does dnm.The cnms exist, in fact, for all
Hilbert-Schmidt operators and the series converges in the sense that if one takes its matrix element
between two coherent states, (a1 on the left and Ip) on the right, the resulting series converges to
( aIAIP).
407
M.Hillery et a/., Distribution junctions in physics: Fundamentals
(&ti+t (zci)"
5Y-I)
5:
1-0
(;) {(6+)n-lCi'},
159
(4.30)
a=
(4.32)
n.m =O
brim (a*)"a m .
A&, a * )=
(4.33)
n.m = O
Under this correspondence we see from eq. (4.30) that the function b(6)goes to
(4.34)
Comparison with eq. (4.9) shows us that this is nothing other than Weyl ordering expressed in a
different form. The distribution function, therefore, should be the Wigner function. As before we define
this as the Fourier transform of the characteristic function x(T) (see eqs. (2.42)-(2.45)) and we use the
real and imaginary parts of a = a, t iai as the variables of the distribution function, so that, analogous
to eq. (4.1), (Y = (2h)-'" (Aq t (i/A)p), where A = (rno)ln.Thus
(4.35)
where
(4.36)
It may be verified, using eqs. (2.42), (2.45) and (4.35) that
(4.37)
408
IMI
where, in the derivation of the last line from the previous line, we have used eq. (2.90) and where
a(&
d2t Tr(Ab(5))eE'a-cu.'.
(4.38)
(4.39)
IT
'I I
d2aA.(a, a * ) W ( a )= 7
IT
d2a d2t Tr(ab(5)) ef'a-co' W ( a )'
(4.40)
(4.41)
(4.42)
We also have from eqs. (4.11) and (4.36)
(4.43)
so that
-1
1
7r
d2a As(qa * ) W ( a )=
;I
1
d25Tr(a&))X(-t)
= Tr(&),
(4.44)
which proves eq. (4.39) and shows that Ar(a,a * ) and W ( a )can be used to calculate the expectation
values of symmetrically ordered operators.
We would also like to say a word about symmetrically ordered power series. Comparison of eqs.
(4.11) and (4.31) allows us to calculate the coefficients appearing in eq. (4.32)
(4.45)
409
161
These coefficients, then, will exist for all operators which have the property that all moments of
Tr(&))
are finite. While this behavior is not as good as that for a normally ordered power series it is
certainly better than that of anti-normally ordered series.
It is also of interest to examine the behavior of W(a).First we note that
(4.46)
so that ~ ( 5 is
) a square integrable function. As W ( a ) is just the Fourier transform of ~ ( 5 )it too is
square integrable. Therefore, W ( a ) is far better behaved than P ( a ) and will exist for all density
matrices.
It is also possible to express the Wigner distribution in terms of the P representation. If we can
represent the density matrix as in eq. (4.15) we then have that
(4.47)
(4.48)
Let us suppose that we have an operator given by an anti-normally ordered power series as in eq.
(4.27). The function corresponding to the of eq. (4.27) is then
A,(a, a*)=
A m
a"' (a*)" .
(4.49)
(4.49)
n.m - 0
By analogy with our discussion of the P representation (eq. (4.26)) we can then express &, 4') as
a(6,d') =
(4.50)
(4.50)
410
162
(4.53)
We have that
(4.54)
so that
(4.55)
Again by considering our derivation of the P representation we can derive an alternate expression
for A*(&,a*).Examining eq. (4.21) we see that
(4.56)
The "function" A.(a, a*)has, of course, all of the singularity problems of the P representation.
The distribution function, Q(a),has, on the other hand, no singularity problems at all. It exists for all
density matrixes, is bounded, and is even greater than or equal to zero for all a. The problems in this
ordering scheme arise in the representation of the operators.
As a final remark, we note that all of the distribution functions can be written in terms of the Wigner
distribution function (McKenna and Frisch [1966]; Agarwal and Wolf [1970]; Haken [1975]; O'Connell
[1983b]), by use of integrals or derivatives.
4.4. Examples
We would now like to calculate Q ( a ) and P ( a ) for a single mode of the radiation field of angular
frequency w . The system which we will consider will be a canonical ensemble at temperature
T = (kp)-'. Our discussion will follow that given in Nussenzveig [1973].
The density matrix for this system is
We first consider the anti-normal distribution function a@).
(4.57)
"=O
41 1
163
e-BnO)exp[-(aIZ (1 - e-Bmw)]
.
= -(I?T
f = x t iy,
(Y
=r
(4.58)
~ (from
6 )eq. (4.54). If we set
t ik
(4.59)
then
x&)
/
'\ \
dZae e * - e * u
dr
=P
To calculate x&),
e-slolz
=7T
dr
dk exp(-s(r - iy/s)'-
(4.60)
(4.61)
Therefore, we see that
x&)
(4.62)
= exp(-lfl2(1- s)/s}.
P ( Q )= 7
=
jdzf e"f""'e
e-b12/A
?TA
e-A1e12
=
IT2
/ /
dx
1
= -(grw- 1)exp[-IaI2
7T
(8"- I)] .
(4.63)
For this system P(a) is a well-behaved function, a Gaussian in fact, and has no singularities. It is
even positive definite. Q(a) is also well behaved, but this comes as no surprise. Our general discussion
had ensured that this would be the case.
412
164
Glauber in his 1963 paper. It is very well behaved but has found little use in applications. More recently
a new class of these distributions, the generalized P representations, has been used to study the photon
statistics of various non-linear optical devices [Walls, Drummond and McNeil [ 19811; Drummond and
Gardiner [1980]; Drummond, Gardiner and Walls [1981]).
The R representation of the density matrix is obtained by using the coherent state resolution of the
identity twice. One has
(4.64)
where la) is defined in eq. (4.5) and
This representation has no singularity problems, Also it exists and is unique for all density matrices
provided that R(a*,/3) is an analytic function of a* and P (Glauber [1963b]). It can be used to evaluate
normally ordered products. One has
1
The generalized P representations (Drummond and Gardiner [ 19801; Drummond, Gardiner and
Walls [198l]) are again functions of two complex variables but are not necessarily defined for all values
of these variables. To define these representations we define the operator
(4.67)
and an integration measure dp(a, p). It is the choice of this measure which determines the distribution
function. We will consider two different choices. The density matrix is then
(4.68)
where D is the domain of integration. Normally ordered products are then given by
(4.69)
Our first integration measure is dp(a, /3) = d a dp where a and P are to be integrated on some
contours C and C' respectively. This gives rise to what is called the complex P representation. Let us
consider the case in which C and C' are contours which enclose the origin. One can then show
(Drummond and Gardiner [1980]) that if the density matrix is of the form
(4.70)
413
M.Hillery et al., Distributionfunctions in physics: Fundamentals
165
where both sums are finite then P ( a , p ) exists and is analytic when neither a nor /3 is 0. Whether
P ( a , p ) exists for a general density matrix is not known. The complex P representation is also not
unique; if one complex P representation exists for a given density matrix, then an infinite number of
representation exist.
The second measure which we wish to consider is dp(a, /3) = da2dp2. Because the coherent states
are linearly dependent such a representation is not unique. In fact we have encountered one
representation of this type already, the R representation. It is possible to choose P(a,P)so that it is
real and non-negative (Drummond and Gardiner [1980]), i.e.
P(a, p ) = (1/4.rr2)exp{-fla - p*12)(%a + p*)IBI&~
t P*)>.
(4.71)
This representation, the positive P representation, is defined for all density matrices.
These two distributions have been used in problems in which non-classical photon states (states
which are more like number states than coherent states) are produced. Under these conditions the
above defined generalized P representations are better behaved than the original P representation. For
example, the P representation corresponding to a density matrix = In) (nl contains derivatives of delta
functions up to order 2n. On the other hand, the complex P representation for this state (again defined
on two contours C and C encircling the origin) is just (Drummond and Gardiner [1980])
P(a, p ) = -(1/4aZ) n! e@ (llap)
(4.72)
(4.73)
Both of these functions are far less singular than the original P representation.
The original motivation for the introduction of these generalized P distributions was connected with
their practical applicability to the solution of quantum mechanical master equations (Drummond and
Gardiner [1980]; Drummond, Gardiner and Walls [1981]). In general, using a coherent state basis, it is
possible to develop phase-space Fokker-Planck equations that correspond to quantum master equations
for the density operator (Haken [1970]; Louise11 [1973]). From this equation observables are obtained in
terms of moments of the P function. However, for various problems, as for example the analysis of
recent experiments on atomic fluorescence (Kimble, Dagenais and Mandelt19781) where we are dealing
with non-classical photon statistics (Carmichael and Walls [1976]), the Glauber-Sudarshan P function is
singular whereas the generalized P function discussed above is not. Also, use of the latter leads to
Fokker-Planck equations with positive semi-definite diffusion coefficients whereas the former gives rise
to non-positive-definite diffusion coefficients. In particular, the generalized P representations were
applied successfully to non-linear problems in quantum optics (two-photon absorption; dispersive
bistability; degenerate parametric amplifier) and chemical reaction theory (Drummond and Gardiner
[1980]; Drummond, Gardiner and Walls [1981]; Walls and Milburn [1982]). On the other hand, the
usefulness of the Wigner distribution in quantum optics has been demonstrated in a paper by Lugiato,
Casagrande and Pizzuto [1982] who consider a system of N two-level atoms interacting with a resonant
mode radiation field and coupled to suitable reservoirs. The presence of an external CW coherent field
injected into the cavity is also included, which allows for the possibility of treating optical bistability
(which occurs when a non-linear optical medium, interacting with a coherent driving field, has more
than one stable steady state) as well as a laser with injected signal.
414
166
5. Conclusion
We have given what we hope is a useful summary of some of the formalism surrounding the use of
quantum mechanical quasiprobability distribution functions. To be of use, however, the formalism
should either provide insight or convenient methods of calculation. In our next paper dealing with
applications we hope to show that this particular formalism does both in that it has proven to be a tool
of great effectiveness in many areas of physics.
Acknowledgments
R.F.OC. acknowledges support from the Dept. of Energy, Division of Material Sciences, under
Contract No. DE-AS05-79ER10459. He would also like to thank the Max-Planck Institute for Quantum
Optics for hospitality, during the summers of 1981 and 1982, at which time part of this work was carried
out. M.O.S. acknowledges support from the Office of Scientific Research, under Contract No. AFOSR81-0128-A,
References
Aganval, G.S. and E. Wolf, 1970. Phys. Rev. D10,2161. 2187,2206.
Alastuey, A. and B. Jancovici, 1980.Physica 102A. 327.
Bloch, F.. 1932, Zeits. f. Physik 74,295.
Bopp. F., 1956,Ann. Inst. H. Poincare 15. 81.
Bopp, F.. 1%1, Werner Heisenberg und die Physik unserer Zeit (Vieweg. Braunschweig) p. 128.
Brittin, W.E. and W.R. Chappell, 1962,Rev. Mod. Phys. 34.620.
Cahill. K.E.. 1%5, Phys. Rev. 138, 81566.
Cahill, K.E. and R.J. Glauber. 1969. Phys. Rev. 177, 1857, 1883.
Carmichael, H.J. and D.F. Walls, 1976,J. Phys. B9. 1199.
Carruthers, P. and F. Zachariasan, 1983,Rev. Mod Phys. 55, 245.
Cartwright, N.D., 1976, Physica 83A, 210.
Cohen. L., 1966, J. Math. Phys. 7.781.
Dahl, J.P., 1982,Physica Scripta 25,499.
De Giorgio, V. and M.O. Scully, 1970,Phys. Rev. A2, 1170.
De Groot. S.R.and L.G.Suttorp, 1972,Foundations of Electrodynamics(North-Holland. Amsterdam).
De Groot. S.R., 1974, La transformation de Weyl et la fonction de Wigner: une forme alternative de la mkcanique quantique (Les Presses
Universitaires de Montrtal, Montreal).
De Groot, S.R.,W.A. van Leeuwen and C.G. van Weert. 1980, Relativistic Kinetic Theory (North-Holland, Amsterdam).
Drummond, P.D. and C.W. Gardiner. 1980. J. Phys. A13, 2353.
Drummond. P.D., C.W. Gardiner and D.F. Walls, 1981,Phys. Rev. A24.914.
Glauber, R.J.. 1%3a, Phys. Rev. Lett. 10, 84.
Glauber, R.J.,1%3b. Phys. Rev. 131,2766.
Glauber, R.J.. 1%5, in: Quantum Optics and Electronics, eds. C. DeWitt, A. Blandin and C. Cohen-Tannoudji (Gordon and Breach. New York).
Gradshteyn, IS. and M. Ryzhik, 1980, Table of Integrals. Series and Products (Academic Press, New York) p. 838.
Graham, R. and H. Haken, 1970, Z. Physik 237,31.
Groenewold. H.J., 1946, Physica 12,405.
Haken, H., 1970, Handbuch der Physik, Vol. XXV/2c (Springer, Berlin).
Haken, H.. 1975. Rev. Mod. Phys. 47.67.
Husimi, K.. 1940. Proc. Phys. Math. Soc. Japan 22, 264.
Imre, K., E. Ozizmir, M.Rosenbaum and P. Zweifel. 1%7. J. Math. Phys. 8. 1097.
Kano. Y.,1965,J. Math. Phys. 6, 1913.
Kimble, H.J., M.Dagenais and L. Mandel, 1978, Phys. Rev. A M . 201.
Klauder J.R. and E.C.G. Sudarshan, 1968. Quantum Optics (Benjamin, New York) p. 178.
Klimintovich, I.L., 1958,Sov. Phys. JETP 6,753.
415
M. Hillery er al., Distribution funcriow in physics: Fundamenrals
167
416
E. P. Wigner
Joseph Henry Laboratories, Princeton University, Princeton, New Jersey 08544
I. INTRODUCTION
The present organization of the first-year graduate
course in quantum mechanics is largely based on the Schrodinger picture and its applications to atomic and nuclear
physics. The first widely accepted textbook on this subject
439
439
417
of quantum mechanics other than the Heisenberg or Schrodinger picture.
The phase-space picture of quantum mechanics is a case
in
Starting from the Shrodinger wavefunction, it is
possible to construct a distribution function in phase space
in terms of the c-number position and momentum variables. Using this distribution function, we can perform canonical transformations in quantum mechanics. The earliest application of the phase-space distribution function was
made in quantum corrections to thermodynamics in 1932.4
Now, this phase-space approach is an important scientific
language for many branches of physics, and there are a
number of review articles.'-'
From the pedagogical point of view, the phase-space picture of quantum mechanics is a useful tool for demonstrating the transition from classical to quantum mechanics. In
this picture, it is possible to perform canonical transformations, just as in classical mechanics.',' Canonical transformations in quantum mechanics lead to a more precise picture of the uncertainty relation, particularly for the
spreading wave packet." It also allows us to define the
uncertainty relation in a Lorentz-invariant manner. ' I
In addition, the phase-space picture of quantum mechanics is a practical research tool in modern optics. Coherent and squeezed states are of current interest,"-" and
they deserve to be included in the existing quantum mechanics curriculum in their own right. The phase-space
picture of quantum mechanics is the simplest language for
these states."
The group theory of linear canonical transformations in
phase space is a very important theoretical tool in many
branches of physics.'"*' In particular, the group of homogeneous linear canonical transformations is locally isomorphic to the ( 2 1 )-dimensional Lorentz group." This allows us to study Lorentz transformations while performing
canonical transformations in phase space that correspond
to physical processes in optics laboratories. I'
The purpose of this article is to present some of the features of the phase-space picture that can be easily accommodated in the existing quantum mechanics curriculum.
In Sec. 11, we introduce the phase-space distribution function and its general properties. In Secs. 111and IV, we discuss linear canonical transformations in classical rnechanics and in the phase-space picture of quantum mechanics.
In Sec. V, the wave-packet spread is shown to be an example of canonical transformations in phase space. In Sec. VI,
coherent and squeezed states of light are introduced as
minimum-uncertainty states. In Secs. VII and VIII, the
coherent and squeezed states are discussed in terms of canonical transformations. It is pointed out in Sec. IX that we
can study the Lorentz kinematics of the Thomas precession
in terms of canonical transformations of squeezed states.
a ( p , t ) = IX(P,t) l 2
W(x,p,t)dp,
W(x,p,t)dx,
(3)
wherex(p) is the momentum wavefunction. It is also possible to compute the absolute square of the inner product of
two wavefunctions. Let us next consider two wavefunctions $ ( x , t ) and d ( x , t ) . If W,(x,p,t) and W , ( x , p , t ) are
the PSD functions for $ ( x ) and d ( x ) , respectively,
then4-6.?3
W , ( x , p , t )W, (x,p,t)dxdp
= (1/237) l ( d ( x , t ) , $ ( x , t ) ) 1 2 .
(4)
This expression is nonnegative, but can become zero if the
two wavefunctions are orthogonal to each other, indicating
that the PSD functions are not positive everywhere in
phase space. For instance, the one-dimensional harmonic
oscillator wavefunction takes the form
(5)
for the ground and first-excited states, respectively. Then,
the corresponding PSD functions are
~ , ( x , p=
) (1/r)e-'"'+P'',
W , ( x , p )= ( 2 / r " ~ ~ + p ~ - t ) e - ' ~ ' + ~ ' ' .
(6)
These examples confirm the properties of the PSD function
given in Eqs. ( 3 ) and (4).
The time-dependent Schrodinger equation leads to the
differential equation4"
. a $ ( x , t ) = -I at
2m
(y $( + V ( x ) $ ( x ) ,
ax
x,t)
(7)
(1)
'I
W ( x , p , t )= -
$ * ( x + y , t ) t j ( x - y , t ) e'"dy.
?T
440
(2)
where m is the mass of the particle, and V ( x )is the potential. If the particle is free, the above differential equation
becomes
- W ( x , p , t )=
at
(9L
-
-W ( x , p , t ) .
(8)
440
418
In the case of the harmonic oscillator with V ( x ) = Kx2/2,
the differential equation becomes
a W(x,p,t).
+ KX -
(9)
aP
There are many other interesting properties of the PSD
function, and they are extensively discussed in the literat ~ r e . ~ -In
" this article, we are interested only in the fact
that canonical transformations are possible in quantum
mechanics through the use of the PSD function.
The probability density functions in Eq. ( 3 ) clearly indicate that x andp are the position and momentum variables
in quantum mechanics. On the other hand, they are c
numbers, allowing us to define a function over the phase
space of x andp. Does this mean that we can determine the
x andp variables simultaneously?The answer to this question is no.
The uncertainty principle does not allow us to determine
a point in phase space. However, it does not forbid the area
element that satisfies (Ax) ( Ap) > 1. Since canonical transformations are area-preserving transformations, they preserve the uncertainty relation. Indeed, we may achieve a
deeper understanding of the uncertainty principle through
canonical transformations in phase space, which are possible in the phase-space picture of quantum mechanics.
- sin(B/2))
cos(B/2)
(x)
(15)
We are now ready to formulate the group linear canonical transformations in phase space. This group consists of
translations, rotations, and squeezes in phase space. The
coordinate transformation representing translations
X = x + b , , P=p+b,,
can be written as
[ U'V ]X P
auav auav
ax ap ap ax
(11)
the Poisson bracket remains invariant. For one pair of canonical variables, this leads to the condition that the Jacobian determinant be 1:
I- ax
I
(16)
(16)
(17)
(14)
(19)
aP
-
aP aP
This means that the area element in phase space is preserved under canonical transformations.
The elongation along the x axis is necessarily the contraction along thep axis. The squeeze along the direction that
makes an angle 4/2 with the x axis is
(sin #)sinh(v/2)
cosh (7/2) - (cosd)sinh(7/2)
0
o\
0 .
1
(20)
r
Since a canonical transformation followed by another
one is a canonical transformation, the most general form of
441
441
419
using the generators of the transformation matrices. If we
use T(bl,bz)for the translation matrix given in Eq. (17), it
can be
- i ( b , N ,+ b.N.1
(21 1
T(u,v)
~
where
O O i
N,O 0 0
( 0 0 3 .
0 0 0
N2= 0 0 i
( 0 0 3
: :),
Since the PSD function is real and defined over the phase
space ofx andp, we can perform area-preserving canonical
transformations as in the case of classicalmechanics. Let us
consider first linear canonical transformations applicable
to a function of x and p. This is then a matter of converting
the matrix generators given in Sec. 111, into differential
forms.
The generators of translations are
N I-i-
[NiJfzI = 0.
The rotation matrix is generated by
0
-i/2
0
L=(if
(22)
N2=
. a
-I--,
aP
while rotations around the origin are generated by
ax'
(30)
(31)
(23)
The squeezes along the x axis and along the direction that
makes an angle of 45" are generated by
and
R(0)=,-IeL.
(24)
These generators satisfy the following commutation relations with N , and N2:
[ N , , L1 = ( i / 2 ) N 2 , [ N , , L 1 = ( - i / 2 ) N 1 .
(25)
Indeed, L, N , , and Nz satisfy the closed commutation relations. They generate the two-dimensional Euclidean group
consisting of rotations and translations in two-dimensional
spa~e.~~.~~
The squeeze matrix of Eq. (19) can be written as
s(0,~)
= e-'''Km,
where
i/2
0
K,=(:
(26)
-?
3.
: ;),
[ K , , LI =
\K,,L ] = iK,.
(28)
This set of commutation relations is identical to that for the
generators of the ( 2 + 1)-dimensional Lorentz
group. 19,2425 The group generated by the above three operators is known also as the symplectic group Sp(2 ),Icz1
and
its connection with the Lorentz group has been extensively
discussed in the literature.
If we take into account the translation operators, the
commutation relations become
] ( - i/2)N2,
[ K , , N l l= ( i / 2 ) N l , [ K I , N 2=
[ K z , N l l= ( i / 2 ) N 2 , [K,,N21 = (i/2)Nl.
(29)
These commutators together with those of Eq. ( 2 8 ) form
the set ofclosed commutation relations (or Lie algebra) of
the group of canonical transformations. This group is the
inhomogeneous symplectic group in two-dimensional
space which is often called ISp(2).'h.Z".2'.26
442
- iL,
a
ax
N I -- - i - - ,
N2=x,
(33)
(32)
respectively.These operators satisfy the commutation relations for the generators of the group of linear canonical
transformations given in Sec. 111. Therefore, we can continue using the matrix formalism of classical mechanics in the
phase-space picture of quantum mechanics.
Next, let us see why canonical transformations are not
discussed in the Schrodinger picture of quantum mechanics. In the Schrodinger picture, the wavefunction is a function ofx orp, but not both. On the other hand, the transformations corresponding to those applicable to the PSD
function are possible on the Schrodinger wavefunction. Indeed, the following transformations on $ ( x , t ) , through the
definition given in Eq. ( 2 ) ,lead to the transformations given in Eqs. ( 3 0 ) - ( 32) 16:
- iK,,
These operators are all Hermitian in the Schrodinger picture. Therefore, linear canonical transformations in phase
space correspond to unitary transformations in the Schrodinger picture of quantum mechanics.
The question then is whether these operators satisfy the
commutation relations for the generators of the group of
linear canonical transformations. The answer to this question is no, but almost yes. They satisfy ail the conputation
relations except one. The operators N , and N2 do not
commute with each other, while N , and Nz do:
A
[ N , , N , ]= - i, while [ N , , N , ]= 0.
(34)
This causes a factor of modulus unity when the translation
alongp is commuted with the translation along the x direction, and the group of linear canonical transformations in
the Schrodinger picture is not the same as that in classical
mechanics.
Let us next look at the subgroups. The group generated
Y . S. Kim and E. P. Wigner
442
420
by N , and N2 is the translation subgroup of the group of
linear canonical transformations in phase space. xhis is an
Abelian group. On the other hand, the operators N , and N2
do not form a closed Lie algebra. They have to be supplemented by an identity operator to form a group. This is
known as the Heisenberg group. This means that the Heisenberg group is represented in phase space as the translation group. The operators N , , N,, and L form thn txodimexsional Euclidean group. On the other hand, N , , N,,
and L need the identity operator to form a group.
The operatorsly,, K,, and L generate the group of homogeneous linear transformations. The coxnterpaxts in the
Schrodinger picture, unlike the case with N , and N2,satisfy
the same set of commutation relations given in Eq. (28).
Furthermore, the Lie algebra of Eq. (28) is identical to that
of the (2 + 1)-dimensional Lorentz group, where K , and
K2are the generators ofboosts along the x andy directions,
respectively,while L generates rotations around the z axis.
This correspondence allows us to study the group of Lorentz transformations using the phase-space picture of
quantum mechanics.
g(k) =( b / ~ r ) ~ ~ e - ~ ~ ,
(35)
at t = 0, the time-dependent Schrodinger wavefunction becomes
tl(x,t) = (b/.n)4[1/(b+it/m)12
- x / 2 ( b + t:/mI
(36)
In order to obtain the PSD function, we solve the differential equation for W ( x , p , t )given in Eq. (28) for the free
particle. The solution is
IP
W(x,p,t)= W x - pr /m,p,O).
(37)
If the initial momentum distribution is Gaussian as is given
in Eq. (35),
W(x,p,t)= (I/n-)exp{-
[(~-pt/m)~/b+bp~]).
(38)
This distribution is concentrated within the region where
the exponent is less than 1 in magnitude. This is the region
of uncertainty, which these days is called the error
We can choose the coordinate system in which b = I .
Then, the above PSD function is a circle at t = 0. As time
progresses, the circle becomes a tilted ellipse while preserving its area,. as is described in Fig. 1.
In the Schrodinger picture of quantum mechanics, as
time progresses, the momentum distribution u ( p , t ) remains unchanged, while the spatial distribution p ( ~be)
comes widespread. This is called the wave-packet spread.
On the other hand, in the phase-space picture of quantum
mechanics, the uncertainty is defined in terms of the error
box. The volume of the error box, which measures the uncertainty, remains constant. Thus we achieve a deeper understanding of the wave-packet spread in the phase-space
picture.
The above-mentioned elliptic deformation is a canonical
transformation. Indeed, the transformation matrix is
(39)
This transformation is generated by
G = ( i,
0 0
which leads to
(40)
(41)
1 2 [ si n h ( ~ / 2 ) ] )- (cos(a/2)
- sin(a/2))
(0
1
sin(a/2)
cos(a/2)
cosh(q/2) sinh(v/2)
sinh(qV2) cosh(v/2)
(cos(a/2) - sin(a/2))
x sin(a/2)
.
cos(a/2)
(42)
where sin a = tanh( 77/2). On the right-hand side, each of
the three matrices is a canonical transformation derivable
from the generators given in Sec. 111. Thus the left-hand
side is a canonical transformation matrix.
443
42 1
er the transformation from the Cartesian coordinate system of q andp into a polar coordinate system, as is indicated in Fig. 2,and the canonical transformation
n = i ( q 2+ p 2 - I ) , 4 = tan-'(p/q),
(43)
with
(alNla) = aa*.
(49)
In order to see the uncertainty relation associated with
the photon number, we note that each number state can be
represented by a harmonic oscillator wavefunction:
(44)
where the variable q is not the x variable for spatial displacement. This variable, together with its conjugate p , is
related to the photon number n through the canonical
transformation of Eq. (43).If we write a and at as
(45)
If a and at are the annihilation and creation operators, respectively, with
[a,at] = 1,
(46)
from this commutation relation, it is possible to define the
number operator N = ata, where
N l n ) = nln).
(47)
The coherent state la) is not an eigensate of this number
operator, but is an eigenstate of the annihilation operator,
satisfying the eigenvalue equation
ala) = ala).
(48)
The probability of the coherent state la) being in the nphoton stateis (aa*)"/[n!
exp(aa*)], whichmeans that
the number of photons in the coherent state has a Poisson
distribution. The expectation value of the number operator
is
IP
(50)
(51)
then Eq. (48) is a first-order differential equation in q, and
its normalized solution is12
(qla) = ( 1 / ~ ) 1 / 4 ~Ilrn(a)l'e(q-{Za)'/2.
(52)
The parameter a is a complex number that can be represented in the two-dimensional complex plane. This wavefunction of the Gaussian form gives the minimum-uncertainty product.'.''
It is possible to change the width of the above Gaussian
form without changing the minimality in uncertainty. l4.I5
While it is posible to write the coherent-state representation of Eq. (45)as
la) = exp(aat - a*o)lo),
(53)
we can consider the state
I W = S(0la),
(54)
with
S(6)=expC:[6(atat) - 4 * ( a a ) l ) .
If 6 is a real number 7,S ( 5 ) can be written as
(55)
h
VII. CANONICALTRANSFORMATIONS OF
COHERENT STATES
It is straightforward to evaluate the PSD function for the
coherent state given in Eq. (52).The result is5-7311.'2
Fig. 2. Polar coordinate system in phase space. Since both q and p are c
numbers, it is possible to make the canonical transformation given in Eq.
(43). The minimum uncertaintv in n and 1 means the same uncertaintv in
q and p The coherent or squeeied state is a minimum-uncertainty stite.
444
(56)
444
422
This is the vacuum or zero-photon state. We can obtain the
circle of Eq. (57) by translating the above circle to (a,b).
This translation is a canonical transformation.
If we multiply a by e, the circle of Eq. (57) becomes
rotated around the origin, and the resulting equation is
( q - a)2 (p - b ) 2 = 1,
(59)
where
cos(8/2) - sin(8/2))
(3
(61)
0
(62)
Indeed, every coherent state is a translation of the vacuum
state.
This two-dimensional Euclidean group is isomorphic to
the internal space-time symmetry group of massless particles and has been discussed extensively in the literature.24*25.29
It is interesting to note that we can study this
group in terms of canonical transformations of coherent
states of light.
+ ( p - b) = 1,
(64)
where
t)
cos(8/2)
= (sin (
12)
- sin(8/2))
cos(e /2 )
(3
t.)(
+
+
(sinO)[sinhA
(tanhv)(coshA- l ) c o s e ]
(sinhA)cosB + ( t a n h v ) [ l + (coshA- I ) ( c o s ~ ) ~ ]
(66)
(a1,
(67)
where
Y.S . Kim and E. P.Wigner
445
423
R
(sin8)[tanh(A/2)][tanh(7/2)]
tan - =
2
1 + [tanh(R/2)][tanh(~/2)](cos8)
Since the circle centered at the origin is invariant under
rotations, the effect of R ( R ) is the same as the identity
transformation.
Indeed, if we apply the above three successive squeezes
on the circle of Eq. (57) not centered at the origin, the net
effect is
( q - f ) z + ( P - g ) z = 1,
(68)
where
cos(R/2) - sin(W2))
();
sented by 2 X 2 matrices, satisfy the same set of commutation relations as that for the (2 + 1)-dimensional Lorentz
group applicable to the space of (x,y,t). Therefore, it is
possible to carry out calculations of Lorentz transformations using the 2 x 2 matrix representation of Sp(2). 19*34 In
fact, a detailed calculation of the Thomas rotation angle
has been carried out within the framework of this 2 x 2
repre~entation.~
In this section, we shall discuss a simpler
Lorentz kinematics that contains all the essential features
of the Thomas prece~sion.~
The (2 1)-dimensional Lorentz group consists of
boosts along the x and y directions and rotations around
the z axis. Indeed, the 2 x 2 rotation matrix R ( 8 ) of Eq.
(14) corresponds to the rotation matrix applicable to
(x,y,t):
I + (cosh/l-l)cosz$
$(1 -coshA)sin(2$)
(sinhA)sin4
cos 8
R(e)=(si;e
;),
-sin 8 0
(69)
:(l-cosh/l)sin(2$)
(sinh/l)cos$
1 + (coshA- l ) s in Z 4 (sinhA)sin$
cosh A
(sinh A)cos $
(70)
(71)
sinh A = tanh 7.
The resulting transformation matrix is
cosh 9
(72)
Wecanreversetheordertoget thesameresultonthemo[S,(R)Sy(7/)]-SY(/z)S,(7)
=R(fl),
mentum by performing the transformation S, (A)S,(,7)
on the particle at rest. However, are the transformation
with
matrices identical to each other? The answer to this question is no:
sin R = (tanh 7).
446
(73)
446
424
It would be very difficult, if not impossible, to carry out
this experiment in special relativity. On the other hand, this
experiment is possible in optics laboratories. For squeezed
states, we can use the following 2 x 2 matrices for S, (7)
andS,(A):
S,(A)
cosh(A /2)
=(sinh(A
.
/2)
sinh(A /2)
cosh(A/2)
(74)
Sy(A)Sx(7)
e7/* cosh(A/2) e- sinh(A /2)
e7sinh(A/2)
e-q/2cosh(A/2)
Then these matrices satisfy Eq. (73) with
=(
(75)
(76)
Each of these 2 X 2 matrices represents a concrete measurable operation in modern optics laboratories. To a coherent state with a = ( a + ib)/,h, we can apply two repeated squeezes, namely, S, (A)S,(7)and S, (A)S,(?).
The difference between these two operations is the rotation
R ( a),which represents a phase change in a according to
Eq. ( 5 6 ) . In principle, this angle can be measured in optics
laboratories. If the present pace of development continues
in optical t e c h n ~ l o g y ,it~ may
~ . ~ ~be possible in the near
future to include this or a similar experiment in an advanced laboratory course in the physics curriculum.
D. Han, Y. S. Kim, and Marilyn E. Noz, Linear canonical transformations of coherent and squeezed states in the Wigner phase space, Phys.
Rev. A37,807-814 (1988).
Vladmir 1. Arnold, Mathematical Methods of Classical Mechanics
(Springer, New York, 1978). This book is an English translation by K.
Vogtman and A. Weinstein of Arnolds original b w k in Russian, Matematischeskie Mefody Klassicheskoi Mekheniki (Nauka, Moscow,
1974).
hEugene P. Wigner, *Quantummechanical distribution functions revisited, in Perspective in Quantum Theory, edited by W. Yourgrau and A.
van der Merwe (MIT, Cambridge, MA, 1971 ), pp. 25-36; R. F. OConnell, The Wigner distribution function-50th birthday, Found. Phys.
13, 83-93 (1983); P. Carruthers and F. Zachariassen, Quantum collision theory with phase-space distributions, Rev. Mod. Phys. 55, 245285 (1983); N. L. Balazs and B. K. Jennings, Wigner function and
other distribution functions in mock phase space, Phys. Rep. 104,347-
391 (1984).
447
447
425
Carlton M. Caves, Kip S. Thorne, Ronald W. P. Drever, Vernon D.
Sandberg, and Mark Zimmerman, On the requirement of a weak classical force coupled to a quantum-mechanical oscillator, Rev. Mod.
Phys. 52,341-392 (1980).
*R. Dirl, P. Kasperkovitz, and M. Moshinsky, Wigner distribution
functions and the representation of a non-bijective canonical transformation in quantum mechanics. I. Phys. A 21, 1835-1846 (1988).
Whilt his nonlinear transformation is possible in the phase-space picture of quantum mechanics in which the variables q, p. n, and are c
numbers, it is not possible to construct Hermitian operators for n and 4.
See W. F. Louisell, Amplitude and phase uncertainty relations, Phys.
Lett. 7, 60-61 (1963); Leonard Susskind and Jonathan Glogower,
Quantum mechanical phase and time operator, Physics 1, 49-61
(1964).
291twas shown in Ref. 24 that the internal space-time symmetry group for
massless particles is locally isomorphic to the two-dimensional Euclidean group. See D. Han and Y. S. Kim, The little group for photons and
gauge transformations, Am. J. Phys. 49,348-351 (1981 ); D. Han, Y.
S. Kim, and Marilyn E. Noz, Internal space-time symmetries of massive and massless particles, Am. J. Phys. 52, 1037-1043 (1984); Y. S.
Kim and E. P. Wigner, Cylindrical group and massless particles, I.
Math. Phys. 28,1175-1179 (1987).
For papers on the generation of squeezed states, see Horace P. Yuen and
Jeffrey H. Shapiro, Generation and detection of two-photon coherent
states in degenerate four-wave mixing, Opt. Lett. 4, 334 (1979); R. E.
Slusher, L. W. Hollenberg, B. Yurke, J. C Metz, and 1 F Valley, Observation ofsqueezed states generated by four-wave mixing in an optical
cavity,Phys. Rev. Lett. 55, 2409 (1995); R. M. Shelby, M. D. Levenson, S. H. Perlmutter, R. G. DeVoe, and D. F. Walls, Broadband
parametric deamplification of quantum noise in an optical fiber, Phys.
Rev. Lett. 57, 691 (1986); Ling-An Wu, H. 1. Kimble, 1. L. Hall, and
Huifa Wu, Generation of squeezed states by parametric down conversion, Phys. Rev. Lett. 57,2520 (1986); Mari W. Maeda, Prem Kumar,
and Jeffrey H. Shapiro, Squeezing experiments in sodium vapor, J.
448
448
426
Negative probability
Richard P. Feynman California Institute of Technology
Some twenty years ago one problem we theoretical physicists had was
that if we combined the principles of quantum mechanics and those
of relativity plus certain tacit assumptions, we seemed only able to
produce theories (the quantum field theories) which gave infinity for
the answer to certain questions. These infinities are kept in abeyance
(and now possibly eliminated altogether) by the awkward process of
renormalization. In an attempt to understand all this better, and
perhaps to make a theory which would give only finite answers from
the start, I looked into the tacit assumptions to see if they could be
altered.
One of the assumptions was that the probability for an event must
always be a positive number. Trying to think of negative probabilities
gave me a cultural shock at first, but when I finally got easy with the
concept I wrote myself a note so I wouldnt forget my thoughts. I
think that Prof. Bohm has just the combination of imagination and
boldness to find them interesting and amusing. I am delighted to have
this opportunity to publish them in such an appropriate place. f have
taken the opportunity to add some further, more recent, thoughts
about applications to two-state systems.
Unfortunately I never did find out how to use the frecdorn of
allowing probabilities to be negative to solve the original problem of
infinities in quantum field theory!
It is usual to suppose that, since the probabilities of events must be
positive, a theory which gives negative numbers for such quantities
must be absurd. I should show here how negative probabilities might
be interpreted. A negative number, say of apples, seems like an absurdity. A man starting a day with five apples who gives away ten and is
given eight during the day has three left. I can calculate this in two
235
427
236
Richard P. Feynman
steps: 5 - 10 = - 5 ; and -5 + 8 = 3. The final answer is satisfactorily positive and correct, although in the intermediate steps of
calculation negative numbers appear. In the real situation there must
be special limitations of the time in which the various apples are
received and given since he never really has a negative number, yet
the use of negative numbers as an abstract calculation permits us
freedom to do our mathematical calculations in any order, simplifying
the analysis enormously and permitting us to disregard inessential
details. The idea of negative numbers is an exceedingly fruitful mathematical invention. Today a person who balks at making a calculation in this way is considered backward or ignorant, or to have
some kind of mental block. It is the purpose of this paper to point
out that we have a similar strong block against negative probabilities.
By discussing a number of examples, I hope to show that they are
entirely rational of course, and that their use simplifies calculations
and thought in a number of applications in physics.
First let us consider a simple probability problem, and how we
usually calculate things, and then see what would happen if we allowed
some of our normal probabilities in the calculations to be negative.
Let us imagine a roulette wheel with, for simplicity, just three
numbers: 1, 2, 3. Suppose, however, the operator, by control of a
switch under the table, can put the wheel into one of two conditions,
A, B, in each of which the probability of 1, 2, 3 are different. If the
wheel is in condition A, the probabilities of 1, p I A= 0.3 say, of 2 is
p l A = 0.6, of 3 is p 3 A = 0.1. But if the wheel is in condition B, these
probabilities are p l B = 0.1, pZB= 0.4, p3B= 0.5, say, as in Table 13.1.
Table 13.1 Probability table for roulette wheel with tv'o conditions
-~
2
3
Condition A
Condition B
0.3
0.6
0.1
0.1
0.4
0.5
We use the table in this way: suppose the operator puts the wheel
into condition A 7/10 of the time and into B the other 3/10 of the time at
random (that is, the probability of condition A, PA = 0.7, and of B,
P, = 0.3.),then the probability of getting 1 is Prob. 1 = 0.7 (0.3) 0.3
(0.1) = 0.24, etc. In general, of course, if a are conditions and pi, is a
conditional probability (the probability of getting the result i if the
condition a holds), we have (pi,= Prob (if a then i)):
[1]
428
Negative probability
237
where Pa are the probabilities that the conditions a obtain, and Piis
the consequent probability of the result i. Since some result must occur
in any condition, we have:
[2]
where the sum is that over all possible independent results i. If the
system is surely in some one of the conditions, so if:
a
then:
c31
xipi= 1
2
3
Condition A
Condition B
0.3
0.6
0.1
-0.4
1.2
0.2
which would be all right. We have also allowed pZBto exceed unity. A
429
238
Richard P. Feynman
The sum of these is 1.00 as required, and they are all positive and can
have their usual interpretation.
The obvious question is what happens if the probability of being in
condition B is larger; for example, if condition B has probability 0.6,
the probability of result 1 is negative 0.4(0.3) + 0.6 (-0.4) = -0.12.
But suppose nature is so constructed that you can never be sure the
system is in condition B. Suppose there must always be a limit of a
kind to the knowledge of the situation that you can attain. And such
is the limitation that you can never know for sure that condition B
occurs. You can only know that it may occur with a limited probability (in this case less than 3/7, say). Then no contradiction will occur,
in the sense that a result 1 or 2 or 3 will have a negative probability
of occurrence.
Another possibility of interpretation is that results 1, 2, 3 are not
directly observable but one can only verify by a final observation that
the result had been 1, 2 or 3 with certain probabilities. For example,
suppose the truly physically verifiable observations can only distinguish two classes of final events. Either the result was 3 or else it was
in the class of being either 1 or 2. This class has the probability
p 1 + p 2 , which is always positive for any positive P,,P,. This case
corresponds to the situation that 1, 2, 3 are not the finally observed
results, but only intermediaries in a calculation.
Notice that the probabilities of conditions A and B might themselves be negative (for example, PA = 1.3, PB = -0.3) while the
probabilities of the results 1, 2, 3 still remain positive.
It is not my intention here to contend that the final probability of a
verifiable physical event can be negative. On the other hand, conditional probabilities and probabilities of imagined intermediary states
may be negative in a calculation of probabilities of physical events or
states.
If a physical theory for calculating probabilities yields a negative
probability for a given situation under certain assumed conditions, we
need not conclude the theory is incorrect. Two other possibilities of
interpretation exist. One is that the conditions (for example, initial
conditions) may not be capable of being realized in the physical world.
The other possibility is that the situation for which the probability
appears to be negative is not one that can be verified directly. A
430
Negative probability
239
n2r)
c41
n=l
f(x) =
1p n sin nx
c51
n=l
or:
CSl
The easiest way of analyzing this (and the way used if P(x,t) is a
temperature, for example) is to say that there are certain distributions
that behave in an especially simple way. If , f ( x ) starts as sin n x it will
remain that shape, simply decreasing with time as P-"*'. Any distributionf(x) can be thought of as a superposition of such sine waves.
But f'(x) cannot be sin nx if f ( x ) is a probability and probabilities
must always be positive. Yet the analysis is so simple this way that no
one has really objected for long.
To make the relation to our previous analysis more clcar, the
431
240 Richard P. Frvnman
various conditions c1 are the conditions n (that is, the index ct is
replaced by n). The a priori probabilities are the numbers pn. The
conditions i are the positions x (the index i is replaced by x) and the
conditional probabilities (these do not satisfy equation [2], for we
have particles lost off the end of the rod, and the state of being off
the rod is not included among the possibilities i) (if n then s at time t )
are:
pi#-+ px,n = e-lt sin nx
432
Negative probability
24 1
order Ap, which at the same time can determine that the position x is
within Ax unless Ax > h/Ap.
It is possible, therefore, that a closer study of the relation of classical
and quantum theory might involve us in negative Probabilities, and so
it does. In classical theory, we may have a distribution function F(x,p)
which gives the probability that a particle has a position x and a
momentum p in dx and dp (we take a simple particle moving in one
dimension for simplicity to illustrate the ideas). As Wigner has shown,
the nearest thing to this in quantum mechanics is a function (the
density matrix in a certain representation) which for a particle in a
state with wave function $(x) is:
r
c71
(If the state is statistically uncertain we simply average F for the various possible wave functions with their probabilities.)
In common with the classical expression, we have these properties.
1 F(x,p) is real.
2 Its integral with respect to p gives the probability that the particle
is at x:
JF(X3P)dPlW = $*(x)$(x)
I31
c91
I&).
[10]
where wM is a weight function depending upon the character of
the physicai quantity.
The only property it does not share is that in the classical theory
F(x,p) is positive everywhere, for in quantum theory it may have
negative values for some regions of x,p. That we still have a viable
physical theory is ensured by the uncertainty principle that no
measurement can be made of momentum and position simultaneously
beyond a certain accuracy.
The restriction this time which ensures positive probabilities is that
the weight functions w,(x,p) are restricted to a certain class - namely,
those that belong to hermitian operators. Mathematically, a positive
probability will result if wp is of the form:
w ( x , ~=
) JX(x - Y / ~ ) ~ + X * ( XY/2)dY
c113
433
jF(xa,pa; xb,Pb)w,(xb,Pb)dxbdPb
Xb@b)
= 6(P,
+ Pb)d(Xa - xb)
which means that the particles a, b, have the same position and opposite momenta so that a measurement of bs position would permit a
determination of as and a measurement of bs momentum would determine us (to be the opposite). This particular F is entirely positive and
classical in its behaviour, so that letting w,(xb,pb) be 6(xa - b) x
6 ( p a - Q ) would not lead to negative probabilities directly, for equation [l] gives P(x,,p,) = 6(x, - b)6(Pa + Q ) in this case, but further
use of such a P in subsequent interactions has the danger of producing negative probabilities. We have become quite used to the rules
of thought and limitations of an experiment, which ensures that they
never arise in quantum mechanics.
It is not our intention to claim that quantum mechanics is best
understood by going back to classical mechanical concepts and
allowing negative probabilities (for the equations for the development
of F in time are more complicated and inconvenient than those of $).
(The classical equations for F for a particle moving in a potential are:
aqx,p,t)lat = - p / m . aFpX
+ vyx)aF/ap
+ JG(x,Q)F(x,p+ Q)dQ
so instead of the momentum changing infinitesimally during an infinitesimal time, A f , it may jump by an amount Q with probability
when it is at x:
434
Negutive psobabilitv
243
[12]
virtual interaction between two currents, j and j'. I t is obviously relativistically invariant with the usual symmetry of the space j,,jy,jzand
time .j, components of the current (in units where the velocity of light
is c = I ) . The original starting Hamiltonian with only transverse
components does not look invariant. Innumerable papers have discussed this point from various points of view but perhaps the simplest
is this. Let the photon have,fotir directions of polarization of a vector
.~,y,z,f,
no matter which way it is going. Couple the time component
with ie instead or e so that the virtual contribution for it will be
negative, as required by relativity in equation [12]. For real photons,
then, the probability of a t-photon emission is negativc, proportional
to -l(fkJi}l2 the square of the matrix element of j , between initial
and final states, just as the probability to emit an x photon is
+ ~(f~,~i)~2. The total probability of emitting any sort of photon is the
algebraic sum of the probabilities for the four possibilities:
c131
It is always positive, for by the conservation of current there is a
relation of .if and the space Components of j, k, j , = 0 if k, is the fourvector of the photon. For example, if k is in the z direction, kz = w,
and k , = k,, = 0 so j , = j , and we see equation [13] is equal to the
usual result where we add only the transverse emissions. The probability to cmit a photon of definite polarization e, is (assume e, is not
a null vector):
- I <fI.i,p,IQI
2/(e,e,)
435
244 Richard P. Feynman
plm
(exp( -
+ mwx
2Rw
for the ground state of an oscillator does not mean that for that state
we can indeed measure both x and p simultaneously.
As another example we will give an analogue of the Wigner function
for a spin half system, or other two-state system. Just as the Wigner
function is a function of x and p , twice as many variables as in the
wave function, here we will give a probability for two conditions
at once. We choose spin along the z-axis and spin along the x-axis.
Thus let f++ represent the probability that our system has spin up
along the z-axis and up along the x-axis simultaneously. We shall
define the quantity f++ for a pure state to be the expectation of
$(l cr= a,
cry), where ax,cry, and oz are the Pauli matrices.
For a mixed state we take an average over the pure state values.
Likewise f+- is the expectation of $(I + az - o.r - a,), ,f-+ is the
expectation of a(l - az + ox - ay) and f-- is the expectation of
1
z(1
- a= - ax cry).
Understanding that this probability can be negative, we shall train
+ + +
436
Negative probability
245
+
+
+ +,
437
++
438
through hole 1. (b) The probability of going through hole 1 in phase. (c) The
probability of going through hole 1 but out of phase. (d) The total probability
to arrive with holes out of phase.
+ + +
439
248
Richard P. Feynman
were available to it. The total probability is the sum of these P for
every pair of ways. If the two ways in P,coming and going are not
the same, P is as likely to be negative as positive.
The density matrix, pij, if the states are i is then represented instead
by saying a system has a probability to be found in each of a set of
conditions. These conditions are defined by an ordered pair of states
coming in i and going in j , with probability p(ij? equal to the real
part of (1 + i)pii. The condition that all physical probabilities remain
positive is that the square of p ( i j ) not exceed the product p(i,i)p( j , j )
(equality is reached for pure states).
Finally, suppose that, because of the passage of time, or other
interaction, or simply a change in basis, the state i has an amplitude
Smiof appearing as state m, where S is a unitary matrix (so the new
density matrix p is given by S p S ) . We then discover we can find
the new probabilities p(rn,n) by summing all alternatives ij of p(i,j)
times a factor that can be interpreted as the probability that the state
coming in i, going in j turns into the state coming in m, going in
n. This probability is:
h
Z(SimSjn
*
sj*nsim)
+j<sj*,s,
- Si*,Sj,)
With such formulas all the results of quantum statistics can lie described in classical probability language, with states replaced by
conditions defined by a pair of states (or other variables), provided
we accept negative values for these probabilities. This is interesting,
but whether it is useful is problematical, for the equations with amplitudes are simpler and one can get used to thinking with them just as
well.
My interest in this subject arose from many attempts to quantize
electrodynamics or other field theories with cut-offs or using advanced
potentials, in which work apparently negative probabilities often
arose. It may have applications to help in the study of the consequences of a theory of this kind by Lee and Wick.
440
ABSTRACT. We prove the existence of star-products and of formal deformations of the Poisson
Lie algebra of an arbitrary symplectic manifold. Moreover, all the obstructions encountered in the
step-wise construction of formal deformations are vanishing.
0. INTRODUCTION
In the Hamiltonian formulation of classical mechanics, a phase space is nothing else but a
symplectic manifold. Passing to quantum theory in the classical way implies a fundamental change
in the nature of observables and makes the interpretation of the classical theory as a limit
of the new one uneasy in many respects. An important aspect of quantization is its relation to
deformations of classical theories. In that spirit, Hato, Lichnerowicz and Sternheimer have
proposed building up quantum mechanics on an ordinary phase space in such a way that
quantization manifests itself in a deformation of the algebra of observables. The value of the
parameter of deformation is closely related to the Planck constant and letting it tend to zero
gives back classical mechanics as a limit case. An account of the deformation approach to
quantization can be found in (1,6].
The algebra of observables is the space of smooth functions over a symplectic manifold with
its natural structure of associative algebra and the appropriate deformations of this structure are
called the star-products. The first star-product appeared as the inverse Weyl transform of the
product of operators (Moyal [9]).It was rediscovered by Vey [lo] who also proved the existence
of nontrivial deformations of the Poisson Lie algebra structure for a symplectic manifold with a
vanishing third De Rham cohomology group. The result was extended to associative deformations
by Neroslavsky and Vlassov [8] under the same assumption. In the mean time, various classes of
manifolds where this assumption is not necessary have been exhibited [4].
The cohomological obstructions appear as follows: a star-product or a deformation of the
Poisson bracket are usually constructed step by step.
In passing from step k to step k t 1, one encounters a Chevalley or a Hochschild cocycle which
should be a coboundary to allow the construction to continue. The work of Vey and Neroslavsky
and Vlassov consists of confining this cocycle in the De Rham cohomology.
We show in this paper that there exists no obstruction at all: each formal deformation of order
487
Letters in Mathematical Physics 7 (1983) 487-496.0377-9017/83/0076-0487
0 I983 by D. Reidel Pllblishing Company.
$01.50.
44 1
k of the Poisson bracket extends to a formal deformation and a similar result holds true for starproducts.
The basic tools are, first, cohomological properties of the Nijenhuis-Richardson bracket,
showing in particular that the bracket of a one-differentiablecocycle with an arbitrary cocycle is
always exact. Secondly, if $. is a conformal nonsymplectic vector field for the symplectic formF
of M,homogeneity with respect to allows us to avoid the obstructions. This was fust observed
in an analytic setting in [2]. An algebraic interpretation led to the proof of the existence of starproducts for exact symplectic manifolds [4,5].A further refinement combined with gluing
allows us to use this type of argument for nonexact F.
1. N O T A T I O N S A N D D E F I N I T I O N S
We will mainly use the notations and defMtions of IS]. Some of them have just to be pre'cised.
Let Mbe a smooth connnected Hausdorff second countable manifold equipped with a symplectic
form F. We suppose dim M > 2. We denote by A(M) the space of smooth forms on M and by X(M)
the space of smooth vector fields on M. As usual, we set N = Ao(M> and L x denotes the Lie derivative in the direction of X E J(Wacting on A(M).
If V and V' are vector spaces, A p ( V. V') is the space of (p + 1)-linear alternating maps from
V p + ' into V' and A ( V, V ' )is the direct sum of theAP(V. V'))s (p 2 -1). For simplicity, we set
A"( v)= A p ( V , V ) and A( V ) = A( V , V). It is known that (A( V), [ ] ), where [ J is the NijenhuisRichardson bracket, is a graded Lie algebra, the degree of C E A P ( v > being p . If 9 is a Lie algebra
structure on V, the Chevalley coboundary operator of the adjoint representation of (V, 9 )is up
to +1 the adjoint action of B onA(V).
We denote byAI,@(M), Aq(M)) [re~p.A~,,,~,(N)]the space of all CA(JCO, Aq(W)[resp.A(N)I
which are local [resp. local and vanishing on the constants]. We set also A,(N) =E(A(N), u] and
Av,ioc, ncQ = E(A lot, nc(N)r V ) ,where E( V , v) denotes the space of formal power series in v with
coefficients in V.
The mapping p: K(M) + A' (M): X + -i(X)F induces an isomorphism between the spaces of
contravariant and covariant tensor fields on M. One sets A = p-'F and, for u EN, Xu = p - l du.
For C AP(X(M), 1 \ 4 0 ) , set p*C(uo, ..., u p )= C(X,, , ..., X ) and, if q = 2, p'C = (A, p*Cj.
In particular, P = p*F is t h e Poisson bracket ofM (A(M) being identified once and for all with
a subspace ofA(X(M), N) in the natural way). Denote by Z&nc(N) the space of Chevalley twococycles of (N, P).Then a being the coboundary operabr and r a symplectic connection of
(M, F), each C EZ&, nc is of the form C = rS: + p*sZ + aE (r E IR, R E Az (M) closed and
E EA&.,(N));S; is a cocycle of the form p'ar,where aFisa cocycle of the Chevalley
cohomology of the representation (A(M), X +Lx) of K(M). Further details may be found in [3,5,71*
PROPOSITION. 1.1. Thereexistsalinearmap r : A~,,,.,(N)~A:,(JC(M),N)such t h a t p * o r =
1 on A' (M), @$) = ( A, 4r ) and r 0.&4Lc,.,(N) C im a' (ll denotes the appropriate
identical map; a' is the coboundaty operator of the cohomology of the representation (A(WY
7 0 p* =
--*
Lx)of Jc(M).1
n,
442
where
Observe that p*: A2(M) -+A{oc,n,(N)is injective. Choose then any right inverse of p*, r1 on El
and r2 and E2 and set rS: = ( A , aPr
), o LL*~,,~(M)
= 1,r o a = a' 0 r1 on El and r = r2 on E z .
This defines a linear map T : Afoc,nc(N) +A&,(K(M). N), which obviously verifies the first three
required properties.
For the last property, one notes that by 15, prop. 2.21, p*A(M) n aA&c,nc(N>= ap*A1(M,
and that onA'(M), 7 0 a 0 p = 7 0 p* 0 a'= a'.
h this paper, we fur once and for all a r such as in Proposition 1.1.
LEMMA 1.2. LetRiEAIoc,n,(~,SiEA~~,,nc(N)andaclosed
C2EA2(M) begiven. Letalso
Rf, S,!, T'EAlo,(3C(M);N) besuch thatRi=p*Ri, Si=p*Sfand Ci[Ri,Si] =p*T'. There exisfs
T E A lo, .,(M such that
(1)
whenever 52 = di(X)F on an open subset U CM.
Proof: If 521 = di (X')F and U n U' # @, then X ' - Xis a symplectic vector field on
U n U'.It is, thus, locally of the form Xu.Since p*i(X,) = i(u)p*, the right-hand side of (1)
vanishes for X = Xu by the graded Jacobi identity.
2. ONE-DIFFERENTIABLE DEFORMATIONS OF P
Recall that a formal deformation of order k of P is an element
ovkCk of A:,lm,nc(w such
that C, = Pand such that IL,, L,I vanishes at order k (i.e., the components [LV,L"] I of
[L,, L,] are vanishing for 1 Q k). A f o m a l deformation o f P is an element L, E A :,lot,
such
that Co = P and that IL, L,r = 0.
nc(w
(2)
whenever 52 = di(X)F'on the open subset LI C M.
(ii) JL; is a cocycle for L, and its f v s t term is p*R. More generally,
489
443
is a formal deformation in p of L,
(iii) In particular, for each t 2 1,
(3)
From Lemma 1.2, we obtain the existence of a globally defined IL; verifying (2) for 1 = k. Then
applying (2) and the graded Jacobi identity:
in view of (3). Hence, (i), (ii) and (iii). For (iv), one has simply to note that the NijenhuisRichardson bracket of one-differentiable cochains is one-differentiable.
Let us say that a formal deformation L, of order k is a driver of a formal deformation 1: if
&, = Ev at order k.
THEOREM 2.2. Every one-differentiable formal deformation of order k of P is a driver of a onedifferentiable formal deformation of P. In particular, there exist one-differentiable formal
deformations o f P with driver P + vc(*52, where is an arbitrary closed two-formof M.
Pro05 Let L, be a one-dimensional formal deformation of order k of P. Let 0 Q 1< k be @en.
Suppose that there exists a one-differentiable formal deformation l; of P such that C,!= Ci for
i < 1. Then C:+, - C1+,is a one-differentiable cocycle. It is thus of the form p*C2 for some closed
two-form 51. By theorem 2.1
490
444
is a one-differentiable formal deformation of P such that
result follows by induction on 1.
= Ci for i
< 1 + 1 . Taking L;
= P,the
Let U be open and suppose that Flu = di(t)F for some t E JC(U). It has been seen in [6] that
+ aE = ~ * T ( C 'where
)
T(C')is a cocycle for a'.
(4)
Thus L E + p l - a o p* o i(t) o T is surjective. Moreover, if the right-hand side of (4)is vanishing,
r = 0 and C' = 0. Hence, the injectivity.
Let ll Ao(E(N v)) be defmed by
and set 0 = 211 - 1.As can easily be seen, for every Tv EA,(N),
THEOREM 3.3. Let r E & be given. There exists a unique formal deformation L, of P of driver
F) + a,o = 0.
((5)
5)
49 1
445
(a, is the coboundary operator associated to the adjoint representation of (E(N, u). Xu)).
hoof: (i) If U is open and Flu = di(t)F for some E E WU), after its definition in Theorem 2.1,
iLv(L; F)reads
Observe, moreover, that L t = p*i(.$)F = p*i(t)T(Co).Hence, the above expression can be written,
for k > 0,
(ii) Let 8, and LL have driver P + vrS$ and satisfy (5). Assume that k > 1 and that C;=Ci
for i < k. Then
(&
+ (2k t i)n
- ap*i(E)T)(c;, -
c,) = 0.
It follows from Lemma 3.2 that c; = ck on U.Thus c; = ck, hence, the uniqueness by induction
on k.
(iii) Observe that (5) means that for every open subset U on which F = di ($)F, gU = c(*i($)r(&)+@
is a derivation of Xu,,. Set C, = P,C1 = rS:, L: = C, + YC, and 9 = L Et p*i(&-(S$) + 0. One has
and
b y Lemma 3.1.
Let now k > 1 and suppose that the Ct's EA&c,n,(N)(i < k ) have been constructed such that
L: - I = &<kdCi is a formal deformation of P of order k - 1 and that 9 E-' =
&<kvip*i(E)7(Ci) + 0 is a derivation of S$-' at order k - 1. Set
492
446
Then Jk is known to be a cocycle. As seen in [5], it has a decomposition. J , = p'@ $. p*\II when
CP and \k are three-cocycles of the cohomology of the representation (A(M), X + Lx) of sC(M).
S t J ; = ( A , a ) +9.By Lemma 1.2,there existsd E A ~ , . n c ( ~ s u cthat
h
reads
(Ls + 2k + 2)Jk = 2
(6)
It follows that p'@ and p*@ are coboundaries [5], prop. 2.3, and thus J , = 2 aC for some
c E4 0 c , ncQ.
Substituting 2aC to Jk in (6) and observing that aLs = L$J t a, awe see that LtC + (2k + 1)C - Ak,
(7)
In view of the properties of 7, the cocycle B has a decomposition B = r'S;
B' = r(B - r'S;) is a cocycle. Taking
+ p*B'
where
b-' .I
= .9
493
447
vkp*i(E)T(Ck) is a derivation of order k of L:Iu. Hence, the existence of e,, by induction on k.
h t T E A L ,nc(N)
or T = rII(r E IR). Denote by ad Tits adjoint action on A(E(N, v)). It is
easily seen that, for t > 1, if& is a formal deformation ofP, then so is
00
Ad(exp vfl".)L,=
,tk
Z(ad
qk.C,.
k=O
Proof: (i) Let L, be a formal deformation of order k of P. Let us.assume that .C, is one-differentiable at order t - 1, where i < k. As
is one-differentiable, we see that C, is of the form 6%; t p*q t dT for some B E IR, some q E A2 (M)
and some T E A ; c , n c ( w . If 0 = 0, then Ad(exp vfT)E, is a formal deformation of order k ofP
and is one-differentiable at order t. Moreover, L, is a driver for EL if and only if Ad(exp vtT)EV,
truncated at order k , is a driver for Ad(exp vrT)LL). Thus, replacing L, by
for suitable Ti's E A t , nc(N)if necessary, we may assume that either L, is one-differentiable or
that there exists s < k such that Ci is one-differentiablefor i < s and that C, - r S i is onedifferentiable for some r &.
(ii) In the first case, we may conclude by Theorem 2.2. Let us now deal with the second case.
Let 1: be a formal deformation of P with driver P + vrS; and define formal deformations
L:, ...,J$ of P inductively by Lo,= E$ and, for i > 0,
L; = l L v ; ( L y ; T(C; - C y ) ) .
It is easily seen that L; = 1, at orders. If s = k, the proof is achieved. Suppose then that s < k
and that we have found a formal deformation f : of P such that 2: =L, at order I , with s < 1 < k.
Then the cocycle C;+l - Cr+l is of the form r'SF + p*R + aT for some r' E IR,some closed
R E A2(M> and some E E A ; , ,,.(N).Replacing .Cb by Ad(exp $+'QE>, we may assume that
T = 0. On the other hand, set
494
448
It follows from Theorem 3.4 that [C, c] is exact for each two-cocycle C. Moreover,
- diff, nc(h) denoting the cohomology of one-differentiablenc cochains, it can be shown that
4.
STAR-PRODUCTS
Recall that a star-product of (M,fl is a formal deformation Ak= X;;=ohkCkof the associative
algebra (N A),with driver M
. t XP, where for k > 0, the Cks are local, vanishing on the constants and such that ck(v,
u ) = (-l)kCk(U, v ) for all u , v EN.
THEOREM 4.1. Every star-product of order 2k is a driver of a star-product. I n particular. e v e v
symplectic manifold admits a star-product.
hoof: Let
= X i < 2 k i i C i be a star-product of order 2k ( k > 0). Then
(8)
is a formal deformation of order k - 1 of P.
In the proof of the Neroslavsky-Vlassov theorem [8], which asserts the existence of a starproduct when the third De Rham cohomology space of M vanishes, the key steps are the
following.
= MA+ VZk+C2k+l is a star-product O f
(i) There exists CZk+l
Akc,nc(w such that
order UC t 1; C2kt1 is determined up to an arbitrary one-differentiable element ofAk,,nc(N).
extends to a star-product of order 2k + 2 if and only if 8, f V k C Z k + 1 is a
(ii)
formal deformation of order k of P.
(iii)
[CZP+lczq+11
(9)
p+q= k
449
~~
REFERENCES
1. Arnal, D., Cortet, J.C., Flato, M.,and Stedeimer, D., Star-Products: Quantization and
Representation Without Operators, in E. Tirapegui (ed.), Field Theoly Quantization and
StatisticalPhysics, Reidel, Dordrecht, 1981, pp. 85-1 11.
2. Cahen, M. and Gutt, S., Lett. Math. Phys. 6, 395-404 (1982).
3. De Wilde, M., Gutt, S., and Lecomte, P., A propos des deuxieme et troisi6me espaces de cohomologie de lalgibre de Lie de Poisson dune van& symplectique, Ann. Inst. Poincar6, to
appear.
4. De Wilde, M. and Lecomte, P.,Letr. Math. Phys. 7,235-241 (1983).
5. De Wilde, M. and Lecomte, P., Existence of Star-Products on Exact Symplectic Manifolds,
to appear.
6. Flato, M., and Sternheimer, D., Deformationof Poisson Brackets, in J. Wolf et al. (eds.),Hamonic
Analysis and Representation of SemiSimple Lie Croup, Reidel, Dordrecht, 1980, pp. 385-448.
7. Gutt, S., Ann. Inst. Poincare33, 1-31 (1981).
8. Lichnerowicz, A., Ann. Znst. Fourier 32, 157-209 (1982).
9. Moyal, J.E., Proc. Cambridge Phil. SOC.4599-124 (1949).
10. Vey, J., Comm Math. Helv. 50,421-454 (1975).
496
450
J. DIFFERENTIAL GEOMDRY
40 ( 1994) 2 13-238
1. Introduction
au a v
{ u , { ~ , ~ } } + { ~ , { w , ~()u }
, v+) ){=wO,.
An important particular case is a symplectic manifold. In this case the
matrix t has maximal rank 2n equal to the manifold dimension. The
inverse matrix o i j defines the exterior 2-form o = $a,,
d x A dx which
is closed in virtue of Jacobi identity.
In [ 11 it has been proved that, if the tensor t has constant rank 2n >
d i m M , there exists a symplectic foliation of the manifold M , a Poisson
manifold with this property being said to be regular. The leaves F of this
foliation locally are symplectic manifolds, and a Poisson bracket is defined
by the symplectic form o (closed 2-form of the rank 2n = dim F) defined
on the leaves.
In the same paper [I] the question of deformation quantization of Poisson and in particular symplectic manifolds is considered. The problem is
to define an associative multiplication operation * , depending on parameter h (Planck constant), of two functions so that the space C(M) with
Received November 1 1 , 1992 and, in revised form, April 27, 1993.
45 1
214
B. V. FEDOSOV
c
00
a = a(x , h ) =
hka,(x),
k=O
452
GEOMETRICAL CONSTRUCIlON OF DEFORMATION QUANTIZATION
215
. .
453
B. V. FEDOSOV
216
It is easily seen that the multiplication (2.2) does not depend on the choice
of a basis in T,M and is associative.
Taking the union of the algebras W, , x E M , we obtain the bundle of
formal Weyl algebras whose sections are "functions"
(2.3)
2k+1>0
(2.4)
454
GEOMETRICAL CONSTRUCTION OF DEFORMATION QUANTIZATION
217
6 a = d xk
(2.5)
A-
aa
ayk '
6 * a = yk i ( G a) a ,
(2.8)
a = 6S-'a
+~
' 6 +aa m ,
455
B. V. FEDOSOV
218
wi,,,q/
aa = dx' A d,a .
(2.9)
aa = da
+ [ ( i / h ) T ,a ] ,
~a = aa
+ [ ( i / h ) y, a3 = da + [ ( i / h ) (+r
,4,
Lemma 2.4.
6a = -[(i//z)w,,y'dx', a ] .
Let d be a symplectic connection. Then
(2.13)
dSa + 6da = 0 ,
(2.14)
a2a = d ( a a ) = [(Z/h)R,a ] ,
where
R = i R i j k r iyy jd x k A d x I ,
456
GEOMETRICAL. CONSTRUCTION OF DEFORMATION QUANTIZATION
219
yo = 0.
(2.16)
the curvature of the connection D .
Lemma 2.6. For any section a E W 8 A we have
(2.17)
D2a = [(i/h)52,a].
D2a = [ ( i / h ) i 2 ,a ] = 0.
-1
Sr = R + a r + ( i / h ) r .
457
B. V. FEDOSOV
220
Then for the curvature we shall have $2= --o , so that SZ will really be a
scalar form.
Theorem 3.2. Equation (3.3) has a unique solution, satisfiiing the con-
dition
6% = 0.
(3.4)
r = 6-l R
+ 6-l
( a r + ( i / h ) r 2 ).
6 A = a A + ( i / h ) [ r, A]
(3.6)
and the initial condition
(3.7)
6 - A = 0.
From (3.6) and (3.7) it follows that A vanishes. Indeed, applying 6-l to
both sides of equation (3.6) and using (3.7) we shall get similar to (3.5)
A = 6- ( a A
+ ( i / h ) [ r ,A ] ) ,
= 0.
Here we have used (3.5), condition (3.4), and the Hodge-De Rham decomposition.
For checking (3.6) by taking into account that 6 S r = 0 , we obtain
S A = -6R - S ( a r )
+ [ ( i / h ) r 6, r ] ,
since
SR = $ R i J k ,yd x J A d x k A d x ,
45 8
GEOMETRICAL CONSTRUCTION OF DEFORMATION QUANTIZATION
22 I
R i j k r+ R i k l j Riijk= 0
for the curvature tensor. Further, S ( a r ) = - a ( S r ) according to Lemma
2.4. Thus,
S A = a(sr>+[ ( i / h ) r ,R + d r + ( i / h ) r 2 1 .
(3.8)
We have d R = 0 according to the Bianchi identity for the curvature tensor, d d r = [ ( i / h ) R ,r ] invirtue of Lemma 2.4 (Ricci identity), a ( i / h ) r 2=
[ a r , ( i / h ) r ] .Taking into account that [ ( i / h ) r ,( i / h ) r 2 ]= 0 , we get that
the last two terms in (3.8) would equal to 0 , and this proves equality
(3.6). q.e.d.
Note that iterating equation (3.5) we can effectively construct the form
r and, consequently Abelian connection D . The first two terms are
i j k
r = $R,,,y y y d x
+ &,,Rijkiy
i j k m
y y y dx
+.(.,
(3.9)
+ [ ( i / h ) r ,a ] .
(3.10)
a = a.
+ ~ - ' ( a a+ [ ( i / h ) r ,a ] ) ,
'
A = 6a - d a - [ ( i / h ) r ,a ] = Da
459
B. V. FEDOSOV
222
between the left-hand and the right-hand sides of (3.9) satisfy the equation
6 A = a A + [ ( i / h ) r ,A]
(3.11)
6 - ' A = 0.
(3.12)
- a + a,
according to equation (3. lo). The last expression is equal to 0 by HodgeDe Rham decomposition, since 6-'a = 0. q.e.d.
It is easily seen that for any a ( y , h) E W with fixed xo E M there
XO
exists a flat section a(x y , h ) E W (not unique, of course) such that
4
a(x, y , h ) = a(y , h ) . This fact implies that the centralizer of WD in W
coincides with the center Z of W . In other words, if a section b E W
commutes with any flat section a E W-, , then b E 2 . Similarly, the
centralizer of WD in W 8 A is 2 @ A .
Iterating equation( 3.10) we can effectively construct the section a E WD
by its symbol a, = a ( a ) :
a = a,
-L
24 R i j k [ o l m a ~ a O Y i Y j Y k +
, a .
-+
(3.14)
CL
* b = a(a-'(a)0 2 ( b ) ) .
Using (3.13) it is easily checked that such defined *-product satisfies all the
460
GEOMETRICALCONSTRUCTION OF DEFORMATION QUANTIZATION
223
4. An action of symplectomorphisms
f'
4.m)
( f a > ( x> Y h ) =
> f Y h )>
being a differential of the map f. Since f is a diffeomorphism, both
Y
8=d
the property
of the uniqueness of Theorem 3.2 implies that 5 = D , WE = WD, i.e., 1;
is an automorphism of the algebra of quantum observables WD. q.e.d.
46 1
B. V. FEDOSOV
224
(4.1)
462
GEOMETRICAL CONSTRUCTION OF DEFORMATION QUANTIZATION
(4.3)
Eja
=U
D(U-'o a o U ) 0 U - '
= Da
[DU o U-' , a ] .
3 =y+Ay
=7
- D U O U-'
+ (DU
22s
fi
U-'),,.
-iQ- = - SiZ + D
h
h
1 ),.
-D(DU
U - l ) + (DU 0
u- 1 )2 = 0.
The scalar form (DU o U-l)o belongs to W28 A' n Z , i.e., begins with
the first power of h . Thus we obtain that the curvatures of these two
connections D and 6 differ by an exact 2-form belonging to ( W .@ A 2n)
2.
?j=D-[Duou-
,.I.
Ay = f
I-+
? - r E W2@ A '
D(Ay)+ (i/h)(Ay)2= 0 .
DU
which is equivalent to
U-' = -(i/h)Ay ,
463
B. V. FEDOSOV
226
DU = -(i/h)Ay
(4.6)
U.
o = -{(
which is fulfilled according to (4.5).
Let us show that condition (4.5) is also sufficientfor the solvability of
equation (4.6). Rewrite (4.6)in the form
SU
+ S ) U + ( i / h ) A ~U
= (D
and apply the operator S-' to both sides of the equation. Taking Uo = 1
and using the Hodge-De Rham decomposition, we get
U = 1 + S-'{(D
(4.7)
+ S)U + ( i / h ) A yo U } .
6-'A
(4.8)
=0
D A = - ( D A Y ) 0 U - -A?
h
h
'
DU
h { ( D A Y )+ ;(A?)'}
i
U - -Ay
h
{ DU + i A y
U},
(4.9)
D A + ( i / h ) A y o A=O.
= 0.
Indeed, equa-
464
GEOMETRICAL CONSTRUCTION OF DEFORMATION QUANTIZATION
227
A = 6 - ' ( ( D 6 ) A + ( i / h ) A yo A ) ,
Thus the iterations yield a trivial solution.
The solution of equation (4.6) is not unique. Let V be another solution.
Then for U-' o V we get
D(U-'
U-'
+ u-'
u-'
( i / h ) A yo V
U-'
V )= -u- 1 ODUO
ODV
( i / h ) A yo V
= 0.
(4.10)
H = -ihlnU = -ihln(l+(U-1))
=-
(_l)k+'
ihx
(U-Ilk
k= 1
6H =
ad((i'h)H)
Ay
exp(ad((i/h)H))- 1
+ (D + 6)H.
H = 6-l
Since Ay
(exp(ad((i/h)H))
ad((i'h)Hi)
Ay+(D+6)H) .
-1
% , so that the
465
B. V. FEDOSOV
228
+ M , then
A
/
:
W
+
W
,
mapping
the
subalgebra
WD
onto itself,
an automorphism
can be associated with f in the following way.
Let a be a symplectic connection on M , and D be the Abelian connection in the bundle W , corresponding to a by Theorem 3.2. Let,
further, 8 = f,a and 5 = f , D be pushforwards of d and D under
diffeomorphism f. Since Weyl curvatures of D and 5 both are equal
to - ( i / h ) w according to Theorem 3.2, by Theorem 4.3 there exists the
section U E Wf of the form (4.1) such that the connection D goes to the
connection D under conjugation autornorphism (4.2). So automorphism
A / defined by the relation
A/:a
(4.1 1)
U O
( ~ ao U-'
)
(4.12)
=id,
if f,f2f3= id. However, we can state that the left-hand side of (4.12) is
the conjugation automorphism by the section U = Ul o f,,(U, o f2*U3)E
W + of the form (4.1).
Lemma 4.4. Let conjugation automorphism (4.2) map the subalgebra
WD onto itselj Then locally there exists a function a, such that U e p E WD.
Proof: For any U E WD we have Aa = U o a o U-' E WD. Then
[DUOu-',u o a 0 u-']= 0.
Thus, the form v/ = DU 0 U - ' commutes with any section U o a o U-' E
WD, i.e., it is the central form. This form is closed since
dv
=D(DUo
u- 1 ) = (DUOu- 1)2 = W A
= 0.
D(Ue-')e'
U - ' = DU
U-' - dv, = 0,
which means that the section Ue-' is flat in W + and then, according to
Lemma 4.2, it automatically belongs to W .
466
GEOMETRICAL CONSTRUCTION OF DEFORMATION QUANTIZATION
229
In this section we shall consider the case M = R2" with the standard
symplectic structure. The standard symplectic form has constant coefficients, therefore we can take the operator of exterior derivation as a
symplectic connection. The Abelian connection D in the bundle W =
W(R2"), corresponding to d by Theorem 3.2, has the form D = -6 + D .
In this case the isomorphism a-' and the multiplication * have a very
simple explicit form
-1
:a(x,h) H a ( x + y , h)=
" 1
-a(a)a(x, h ) y a ,
a!!
lffl=O
(5.1)
Definition 5.1. The trace in the algebra WD(R2")is the linear functional defined on the ideal WrmP(R2"),consisting of the flat sections
with compact support by formula
(5.2)
Thus the trace has values in Laurent formal series in h with negative
powers of h not greater than n , i.e.,
M
k=O
Lemma 5.2.
(5.3)
where a E W',mP(R2n),b E W(R2n).
Prm$ Since a ( a o b) = a(a)* a ( b ) ,according to (5.1) it is sufficient
to check the equality
dkb
aka
akb
aka
i z n axil . . . axik a x j l . . . a j k con = L 2 n axjl . . . a x j k axil . . . a X l k
which is easily verified by integrating by parts. q.e.d.
9
467
B. V. FEDOSOV
230
We shall prove that property (5.3) implies the invariance of the trace
under isomorphisms A,, considered in $4. For an open contractible set
0 c R2" we shall denote the algebra of flat sections with support in 0
by Wr"'(0). Let f be a symplectic diffeomorphism, defined on 0 and
mapping it onto the open set f (0),and let A : Wrmp(0)-+ Wrmp(f (0))
be an isomorphism corresponding to f by formula (4.1 1).
Theorem 5.3. For any a E WrmP(0),
tra = tr(Afa).
(5.4)
W )= ( i / h ) [ H ( t )401
(5.5)
d
i
dt
h
wherefrom it follows that tr a( t ) = const.
Lemma 5.4. Let ft be a family of symplectic difleomorphisms of the
open set 0 and let
- tra(t) = - t r [ H ( t ) ,a ( t ) ]= 0 ,
A , : W:mp(0)
c W,""'(R2")
W;"'(ft(O))
(5.6)
Uf0 (A*.)
u-' .
ir(t) =
ft
by gt , we shall have
Since the section a is flat, aa/ax' = a a / a y ' , so that the last expression
can be written in the form
468
GEOMETRICAL CONSTRUCTION OF DEFORMATION QUANTIZATION
23 1
Here we have used the fact that for a symplectic map f the expression
o~jui(afk/ax')(ag~/ax')v'
is symmetric with respect to u , E R2" .
Substituting into (5.7), we get equation ( 5 . 5 ) , where H ( t ) belongs to W,+
but is, generally speaking, not a flat section. We will show that it is possible
Applying the operator D to the both sides of
to pick H ( t ) E WD(R2").
(5.5) and taking into account that Da(t) = 0 we get ( i / h ) [ D H ( t,) A,a] =
0 . Being fulfilled for any a E Wrmp(0), this equation means that D H ( t)
is a central (i.e., scalar) 1-form ly , which is closed because
d y =Dry = D(DH) = 0 .
Hence, y = d p , where qt = q p , ( xh)
, is a scalar function, which is
uniquely defined if subjected to the normalizing condition q , ( J ; ( x o )=
) 0,
where xo E 0 is any fixed point. Replacing H ( t ) by H ( t ) - qt we do
not change equation ( 5 4 , as p, belongs to the center, and, on the other
hand, D ( H ( t )- pt) = 0 , i.e., H ( t ) = pt is a flat section. q.e.d.
Let us proceed to prove the theorem. According to the lemma, it is
sufficient to construct a family of symplectic diffeomorphisms fi , so that
= id , and
= f . Besides, we may confine ourselves to a sufficiently
small neighborhood 0., of an arbitrary fixed point xo E 0 . A general
case would be obtained by using a partition of unit subordinated to a
sufficiently fine covering of a compact set supp a .
The desired deformation fi is constructed in two steps. At the first
step, consider the linear part L f of the map f at the point xo given by
fo
fi
(5.8)
ijaS(z)
x = z +o
-,
aZJ
f'(X)=Z
--o
ij
-.as(z )
aZ J
469
232
B. V. FEDOSOV
From ( 5 . 8 ) we get
(5.9)
Z=
f(x)
dS(z) =
1
pi(
f q x ) - x) dz .
It is easy to see that the 1-form on the right-hand side of the second
equation of (5.9) is exact. Indeed, its exterior differential is equal to
o i j ( d f - d x )A (df
+ dx) = w i j d f iA df - wijdx A dx
+ o i j d f A dx - oijd x i A df .
f is a linear transformation f ( x ) = Afix with a symplectic matrix A:. , formulas (5.8), (5.9) give the Cayley transformation.
Remark. If
In this section we construct a trace in the algebra W,(M) on an arbitrary symplectic manifold M . The basic tool is a localization, i.e., a representation of the algebra W(M) by a compatible family of the algebras
of quantum observables in R2 . We shall denote the standard symplectic
form on R2 by oo and the Abelian connection -6
d in W(RZ) by
Do.
Let { 01}be a locally finite covering of the manifold A4 by local Darboux charts, { p i ( x ) }be a partition of unity subordinated to this covering,
and : Ol 4R2 be coordinate maps. For a given symplectic connection
a and the corresponding Abelian connection D in the bundle W = W ( M )
consider the algebra W,(M) of flat sections determined on M and its subalgebra Wzmp(Ui), consisting of flat sections with supports in Oi . Using
the constructions of $4 we may define isomorphisms
x,
A i : W;mP(Oi)
---t
WDOc o m p ( ~ i ( ~ i ) )
470
GEOMETRtCAL CONSTRUCTION OF DEFORMATION QUANTIZATION
233
Di
from the connection Zi to the connection Do using the conjugation automorphism. Finally we get
A ~ w;""(o~)
:
3 a ++ U,o (xj,u) o ui-'
E
w~~~~
Do (x;( 0 ; ) )
'
w comp
A . .= A.AT :
(x,(o,n oj))-+ w ~ ~ n
~oil).
~
(6.2)
IJ
I J
Do
From (6.2) it immediately follows that A i j satisfies a cocycle condition
A i j A j k A k i= id
(6.3)
n O, n o k ) .
Cjj,= C a - l ( p J = 6'
(p)= fJ
-I
(1) = 1 .
(6-5)
must be fulfilled.
Theorem 6.2. A trace in the algebra WD(M)does exist.
PrmJ For a given coordinate covering {Oiland a partition of unity
{ P i W ) take
tra=Ctrai =CtrAi(jioa),
i
47 1
B. V. FEDOSOV
234
where a E W,(M) and the traces tra, are given by formula (5.2) in
Wcomp(R2n).We should check the correctness of the definition, i.e., indeDO
pendence of the choice of a covering, a partition of unity, and coordinate
isomorphisms A; , and then prove property (6.5).
Let us prove the independence of the choice of coordinate isomorbe a flat section with a support in the cophisms. Let a E W,*(O)
ordinate neighborhood 0 c M , x and x be two coordinate diffeomorphisms 0 --,R 2 , and A and A be the corresponding coordinate isomorphisms, mapping W,(O)onto W,op(x(O)),
W T ( x ( 0 ) )respectively. Then the symplectic map f = x x - : ~ ( 0
+)
~(0)
and the
are decorresponding isomorphism Af : Wcomp(x(0))-+ W:mp(x(0))
DO
fined. According to Theorem 5.3 we have
tr AfAa = tr Aa .
(6.7)
i ,;
=trboa,
which proves the theorem.
C(pi a )
0
( p j 0 b ),
472
GEOMETRICAL CONSTRUCTION OF DEFORMATION QUANTIZATION
235
7. Generalizations
The above constructions of the deformation quantization and the trace
in the algebra of quantum observables allow different generalizations.
Quantization with coefficients in Hom(E , E) . An evident generalization consists in considering matrix coefficients. No change is necessary,
except that in the definition of the trace (5.2) a matrix trace under integral
sign must be taken. A less evident generalization is obtained if we admit
that the coefficients ak , i, ...i ,j , ...j , ( x ) of series (2.3), (3.4) take values in a
bundle Hom(E , E) , wheri E is a vector bundle over M . Let us consider
this case in more detail.
Let as be a symplectic connection on M , and aE be a connection in
the vector bundle E . Then d = as8 1 1 c3 dE defines the connection in
Weyl algebras bundle W 8 Hom(E , E) ; we will denote this bundle by W
as before for short. We shall look for Abelian connection D in the bundle
W in the same form as in (3.1). The same equation (3.3) is obtained for
r , R being now equal to
(h/2)R: dx
dx
+ :RijklyiyJdxk A dx ,
where the first term is the curvature of dE ,and the second one is the same
as in (2.1 1); the superscript s means symplectic. Theorems 3.2 and 3.3
are completely valid in this case.
As to the action of symplectic diffeomorphisms, the results of $4 are
also valid with some modifications. Let &: M --+ A4 be a symplectic
diffeomorphism of M , and f , * E = (&-)*E be a pushforward of the
bundle E under 4 , i.e., (&-)* is an induced bundle. Let a fiberwise
homomorphism a, : ( f , , E ) + E be given as well. Then the formula
(7.1)
So, if lifting (7.1) is given, the pushforwards and pullbacks of the sections
and the connections a and D are defined as in 54.
In the case of coefficientsin Hom(E , E) localization, considered in $6,
is constructed as before with some modifications. More exactly, not only
413
B. V. FEDOSOV
236
and then a trace can be defined as before by formulas (6.6), (5.2) with the
matrix trace under integral sign in (5.2).
Deformation quantization of regular Poisson manifolds. As mentioned
in the introduction, a regular Poisson manifold has a symplectic foliation.
It means that it is possible to introduce local coordinates (Darboux coordinates)
I
2n
x ,x , ..* > x , x
2n+l
,x ;
2n = rank(tij), m = d i m M ,
- 1,
ti+n, i
- -1;
i = 1 ,2 ,
,n ,
and the rest of its components are equal to 0 . The leaves F of the foliation are locally defined by equations x k = const , k = 2n + 1 , - m7
dx A dxn+i defines a symplectic structure on the
and the form o =
leaves. Thus the regular Poisson manifold can be locally considered as a
rn
family of symplectic manifolds depending on parameters x 2n+ 1 ,
,x .
The quantization construction, given in 552, 3, smoothly depends on parameters and is local, so it is evidently valid for the case of regular Poisson
manifolds.
More precisely, the construction looks as follows. We consider a tangent
bundle T F along the leaves and the exterior algebra A, = A(T*F). A
homomorphism i* : T * M .+ 7F is defined, induced by a local embedding of the leaf i: P --,M . In Darboux local coordinates we introduce
the natural basis of vector fields e k = a / 8 x k ( k = 1, - , 2 n ) tangent
to the leaves and the dual basis 8 = (8, . . . , 1 9 ~in~ )T*F . Instead of
series (2.4) we will now consider the series
x:=l
--
* * .
(7.3)
474
GEOMETRICAL CONSTRUCTION OF DEFORMATION QUANTIZATION
237
6*a = v z(e, ) a .
(7.4)
V,e
=e
rjke
the Jacobian matrix of the transition diffeomorphism between two Darboux local charts
$ = ax'"/axS
has a triangular form, because
a*'"/axi=O,
a=2n+l,..* m,
2
(4)
475
238
B. V. FEDOSOV
V F . A partial solution of
system (7.6) is
(7-7)
(the general solution is obtained by adding to (7.7) any completely symmetric 3-tensor).
Thus, we obtain the connection a along the leaves in the bundle T F
such that its restriction to any leaf gives a symplectic connection on the
leaf. According to (7.7) it smoothly depends on the coordinates x 2n+l , ...
x m , which are parameters, defining the leaf. Theorems 3.2 and 3.3 give a
References
[ l ] F. Bayen, M. Flato, C. Fronsdal, A. Lichnerovicz & D. Stemheimer, Deformation
fheory and quanfizafion,Ann. Phys. 111 (1978) 61-151.
[ 21 M. De Wilde & P. B. A. Lecomte, Existence of star-product and of formal deformations
in Poisson Lie algebra of arbitrary symplectic manifold, Lett. Math. Phys. 7 (1983)
487-496.
[3] B. Fedosov, Formal quantization, Some Topics of Modem Math. and Their Appl. to
Problems of Math. Phys., Moscow, 1985, pp. 129-1 36.
[4] -,
Quantization and index, Dokl. Akad. Nauk. SSSR 291 (1986) 82-86, English
transl. in Soviet Phys. DOH.31 (1986) 877-878.
[ S ] -,
An index theorem in the algebra of quantum observables, Dokl. Akad. Nauk
SSSR 305 (1989) 835-839, English transl. in Soviet Phys. DOH.34 (1989) 318-321.
[6] A. Masmoudi, Ph.D. Thesis, Univ. de Metz (1992).
[7] D. Melotte, Invariant deformations of the Poisson Lie algebra of a symplectic mani-
fufd and star-products,Deformation Theory of Algebras and Structures and Applications, Ser. C: Math. and Phys. Sci., Vol. 247, Kluwer Acad. Publ., Dordrecht, 1988,
961-972.
181 H. Omori, Y. Macda & A. Yoshioka, Weyl manifolds and deformation quantization,
Advances in Math. (China) 85 (1991) 224-255.
Moscow INSTITUTE
OF PHYSICS
AND TECHNOLOGY
476
PHYSICAL REVIEW D, VOLUME 58, 025002
David Fairlie
Department of Mathematical Sciences, Universily of Durham, Durham, DHl3LE. United Kingdom
Cosmas Zachod
High Energy Physics Division, Argonne National Laboratory, Argonne. Illinois 60439-4815
(Received 26 November 1997; published 8 June 1998)
The Wigner phase-space distribution function provides the basis for Moyals deformation quantization
alternative to the more conventional Hilbert space and path integral quantizations. The general features of
time-independent Wigner functions are explored here, including the functional (star) eigenvalue equations
they satisfy: their projective orthogonality spectral properties; their Darboux (supersymmetric) isospectral
potential recursions; and their canonical transformations. These features are illustrated explicitly through
simple solvable potentials: the harmonic oscillator, the linear potential, the Poschl-Teller potential, and the
Liouville potential. [S0556-2821(98)00714-0]
PACS number(s): 11.15.Tk,03.65.Db, 04.20.Fy, 05.30.-d
I. INTRODUCTION
f(x,p)=&
dyJI*(x-iy)e-PJI(x+iy).
(1)
These functions are not quite probability distribution functions, as they are not necessarily positive-this is illustrated
below. However, upon integration over p or x , they yield
bona fide positive probability distributions, in x or p . respectively.
Wigner functions underlie Moyals formulation of quantum mechanics [3], through the unique [4,5] one-parameter
(ti) associative deformation of the Poisson-brackets structure
of classical mechanics. Expectation values can be computed
on the basis of phase-space c-number functions: given an
operator A(x,p), the corresponding phase-space function
A ( x , p ) obtained by p e p , XHX yields that operators expectation value through
(2)
assuming the usual normalization J d x d pf ( x . p )= 1 and further assuming Weyl ordering, as addressed by Moyal, who
took matrix elements of all such operators:
X exp[ i T( p - p )
(3)
(4)
Recalling the action of a translation operator exp(aa,)h(x)
=h(x+a), it is evident that the * product induces simple
Bopp shifts:
(5)
etc., where a and a here act on the arguments o f f and g,
respectively. This intricate convolution samples the Wigner
function over the entire phase space, and thus provides an
alternative to operator multiplication in Hilbert space.
Antisymmetrizing and symmetrizing the star product,
yields the Moyal (sine) brackets [3]
+ ia(x-x)]
(6)
58 025002-1
477
~~~
(7)
respectively. Note [7,8] that
dpdxf*g=] dpdxfg.
(8)
(8)
(9)
In general, a systematic specification of time-dependent
Wigner functions is predicated on the eigenvalue spectrum
of the time-independent problem. For pure-state static distributions, Wigner and, more explicitly, Moyal showed that
(10)
(10)
{ { H ( x 4 J ) . f ( x 7 p )=
} }0;
H ( x , p ) * f ( x , p )= Ef(x,p).
(x- 5 Y )
I'=
I'=
*(
2 a
d y [ ( p - i 5 2 x ) 2 / 2,+V(x+5y)]e-'"**(x-5y)*[x+5y)
2 a
d y e C t Y p I i( e y + i s 2 x ) ? /
=1
x+ 5 Y )
,m+V(x+5.)]**(x-sy)*(x+syi
dye-IYp$*(x- i y ) E + ( x + s y ) = E f ( x , p ) ,
(12)
2a
since the action of the effective differential operators on
f *H= 2 n
] dye-I'p[ ( 3)-
**
(1 1 )
i 2,) * /
2m
+ V ( x-
5 ) ] 5 ) +(
y
$* ( x - y
x+ i y )
(13)
= Ef(X,P).
I
(14)
On account of the integration property of the star product,
Eq. (8). the left-hand side of this amounts to
J d x d p H ( x , p )* f ( x , p ) . Implicitly, this equation could have
been inferred from the Bloch equation of the temperatureand time-dependent Wigner function, in the early work of
[9]. *-genvalue equations are discussed in some depth in the
second reference of Ref. [5] and in [ l o ] .
By virtue of this equation, Fairlie also derived the general
*-orthogonality and spectral projection properties of static
Wigner functions [7]. His results were later formalized in the
spectral theory of the second of Ref. [5] [e.g., Eq. (4.4)].
Consider g corresponding to the (normalized) eigenfunction
$g corresponding to energy E , . By lemma 1 and the associativity of the * product,
02:im2-2
478
FEATURES OF TIME-INDEPENDENTWIGNER FUNCTIONS
(15)
f*H*g=Eff*g=E8f*g.
Then, if E , f E f , this is only satisfied by
duced by Fairlie appears local, but is, of course, highly nonlocal, by virtue of the convolving action of the * product.
Precluding degeneracy, for f = g ,
f * H *f = E,f
(16)
f*g=O.
d x d pf g
= 0,
*f = H *f *f ,
(18)
(17)
f*faf.
and demonstrates that all overlapping Wigner functions cannot be everywhere positive. The unintegrated relation intro-
(19)
(20)
I
(21)
and, for
* genfunctions of lemma
1,
(22
Bakers converse construction extends to a full converse
of lemma 1 , namely, the following lemma.
Lemma 2. Real solutions of H ( x , p ) * f ( x , p ) = E f ( x , p )
[=f ( x , p ) * H ( x , p ) ] must be of the Wigner form,
f =Jdye-YPJI*[x- ( h l 2 ) y ] J I [ x +( h / 2 ) y ] / 2 ~ such
,
that
HJI= EJI.
As seen above, the pair of *-eigenvalue equations dictate,
for f ( x .p ) = J d y e - Y P Y ( x ,y ) ,
(23)
This constrains T ( x , y ) to consist of bilinears JI*[x
- ( h l 2 ) y l $ [ x + ( h l 2 ) y l of unnormalized eigenfunctions
$ ( x ) corresponding to the same eigenvalue E in the Schrodinger equation with potential V. W
These two lemmata then amount to the statement that, for
real functions f ( x . p ) , the Wignerform is equivalent to compliance with the *-genvalue equation (real and imaginary
part).
111. EXAMPLE: THE SIMPLE HARMONIC OSCILLATOR
The eigenvalue equation of lemma 1 may be solved directly to produce the Wigner functions for specific potentials,
without first solving the corresponding Schrodinger problem
(as in, e.g., [l 11). Following [7],for the harmonic oscillator,
H = ( p 2 + x 2 ) / 2 (with h= 1, m= l ) , the resulting equation is
025002-3
479
THOMAS CURTRIGHT, DAVID FAIRLIE, AND COSMAS ZACHOS
By virtue of its imaginary part (nd,-pd,)f=O, f is seen to
depend on only one variable, z = 4 H = 2 ( x 2 + p z ) , and so the
equation reduces to a simple ordinary differential equation;
(25)
2l l L ( z ) = O ,
zd:+(l-z)d,+E-
These states are real, like the Gaussian ground state, and are
thus left-right symmetric * genstates. They are also transparently * orthogonal for different eigenvalues, and they project
to themselves, as they should, since the Gaussian ground
state does, f o * f o a f o . It will be seen below that even the
generalization of this factorization method for isospectral potential pairs goes through without difficulty.
(26)
$E(x)=
2;;
I+-m
dXe'F(X-X)e'EX=Ai(x-E). (33)
f, = e-2HL,(4H),
Lo= 1, L , = l - 4 H ,
L2=16H2- 16H+2,. . . .
(34)
(27)
Note that the eigenfunctions are not positive definite, and are
the only ones satisfying the boundary conditions,f(O) finite
andf(z)+O, as z+m.
In fact, Dirac's Hamiltonian factorization method for algebraic solution caries through (cf. [5]) intact in * space,
Indeed,
1
H = -2( x - i p ) * ( x + i p ) +
v?
(35)
-,
2
(28)
where z = x + p 2 . Hence, the Wigner function is again an
Airy function, like the above wave functions, except that the
argument has a different scale and shift:'
whose imaginary
Part ( i d ~ - p d x ) f ( x J ' ) = o gives f ( x - p )
1
at= v? ( x - iP).
22~3
(29)
a*fo= - ( x +
y9
ip)* e - ( x 2 + p 2 ) = 0
(31)
(36)
(30)
(32)
22~3
f ( x , p ) = %A' I ( ~ ~ ~ ( z - E )Z;;~i(22J3(X+p2)=
E))
The Airy functions are not square integrable, so that the conventional normalization f*f= ( 1/2~r)f does not strictly apply. On the other hand, the energy eigenfunctions are nondegenerate, and the general corollary 1 projection relations
f n * f b a 6,,bf, still hold for the continuous spectrum:
'This case is similar to the Gaussian wave function, i.e., the harmonic oscillator ground state encountered above, whose Wigner
function is also a Gaussian, but of different width. S. Habib kindly
informed us that this solution is also given in Ref. [13], Eq. (29).
025002-4
480
FEATURES OF TIME-INDEPENDENT WIGNER FUNCTIONS
(44)
RECURSIONS
H*f=Q**Q*f=
are
* genfunctions
provided
(45)
f*Q**Q=Ef
(38)
(39)
@A=
(40)
This Riccati equation, familiar from SSQM, can be Darboux
transformed by changing variable for the superpotential
Wx).
(41)
V = 1,
(47)
(42)
&
w,
(43)
(48)
results via Eq. (40) in the symmetric, reflectionless PoschlConTeller potential [15], V= 1 - 2/cosh2[(&&/fi)].
versely, starting from this potential,
2
V(x) = 1 -
7)
coshz( &x
025002-5
(49)
48 1
PHYSICAL REVIEW D 58 025002
(50
so that
(51)
V= I.
- iyp
fo(xp)
f o ( N ; x , p ) = [ s e c h ( n / R ) * ] N - ~ fl;x,p)[*
o(
sech(x/h)]-l.
(58)
dy 2 cosh(x/fi-y/2)cosh(x/h+y/2)
COS(YP)
dycosh(2x/fi)+cosh(y)
(57)
where the integral only need be evaluated from the above
fo( 1 ;x,p). Alternatively,
sin( 2xplh)
sinh(2x/R)sinh( v p )
(52)
Q*fo= p--d
i;l ; 11
- i tanh - + - a p
fo(x,p)=O.
(53)
2N-1
=(
W ( n ) = n tanhi
9)
m)p*fo(N-
(59)
l)*P.
by virtue of
Q(N-l)*fo(N-I)=O=fo(N-l)*Q*(N-l). (60)
(54)
(54)
Q*(N)*Q*(N- l)*fo(N-2)*Q(N-
is found recur1)*Q(N),
(61)
V(x)=n-n(n-
and so forth. Thus, the entire Wigner *-genfunction spectrum of H ( N ) is obtained with hardly any reliance on Schrodinger eigenfunctions.
l)/cosh2
to its contiguous
(55)
which has one more bound state (shape invariance). Recursively, then, one may go in N steps, with the suitable shifts
of the potential by 2n- 1 in each step, from the constant
potential to
which yield trivial Jacobians (dXdP =dxdp{X,P}) by preserving the Poisson brackets
( 7).
V ( N ; x ) = N Z - N ( N + I)/co~h -
(62)
(56)
au
{U,V},p=-----
av
ax ap
au a,
ap ax
(63)
025002-6
482
FEATURES OF TIME-INDEPENDENT WIGNER FUNCTIONS
Equivalently,
(66)
(65)
{x.pI= { X . P ) ,
(67)
Thus,
(68)
The pair of Wigner functions in the respective canonical variables, f ( x , p ) and
'I i
3(X,P)=2 a
dYq* X--Y
")
e-jyPq x+-Y
2
(69)
(70)
where 0 is with respect to the variables X and P .
To find this functional, let X = $ ( X I+ X 2 ) and Y = X , - X I , so that J d X l J d X 2 = J d X J d Y . Noting that
(:ii
'I!* X - - Y
'I! X + - Y
dPeiYP3(X,P),
(71)
(72)
483
PHYSICAL REVIEW D 58 025002
H ( x , p ) * Z ( x , p ; X , P ) = ~ ( x , p ; X , P ) O 7 - I ( X , P ) ,(73)
as
follows
from
H(x, - id,)exp[iF(x,X)]
=li(X,idx)exp[iF(x,X)]. If F satisfies a O-genvalue equation, then f satisfies a *-genvalue equation with the same
eigenvalue, and vice versa.
Note that, by virtue of the spectral projection feature (16),
(19), this equation is also solved by any representationchanging equal-energy bilinear in real Wigner * genfunctions of H and H,
(74)
for arbitrary real g ( E ) . Such a bilinear transformation functional is nonsingular (invertible) if and only if g ( E ) has no
zeros on the spectrum of either Hamiltonian?
As an example, consider the linear potential again, which
transforms into a free particle (H=P) through
1
F = - -X3- x x * p = - x ,
x=P-X2.
(75)
By direct computation,
(76)
Note N,= l/Gfor the free-particle energy eigenfunction
normalization choice .U,(X) = (2 n)-In exp(iEX). Thus, indeed, the free-particle Wigner function F,(X,P) = 6 ( E
- P ) / ( 2 ~ r )transforms into
f(x,p)=G
dPdXZ6(E-P)
2 ~ 3
= -Ai(223(x+p2-~)),
2n
(77)
II
(n+p)*B(x,p;X,P)=B(x,p;X,P)OP
(78)
(79)
The exponent of the integrand turns into i y ( E - x - p 2
- hyZl12).
025002-8
484
FEATURES OF TIME-INDEPENDENT WIGNER FUNCTIONS
(80)
Hence,
I!
&'
dpe'[(-2/3)X3-2(X+PZ-P)Xl*
(81)
'I
(82)
I
*E(x)=
$m
K , d e x ) ,
(85)
xr
.[
- i ( p + i t . ) i fi) - i ( p + it.) - i fi
2
2
(86)
(83)
This follows, e.g., from a result in [22], Vol. 11, p 51, Eq.
(27):
(87)
(84)
The solutions are Kelvin (modified Bessel) K functions, for
O<E<m,
025002-9
485
THOMAS CURTRIGHT, DAVID FAIRLIE, AND COSMAS ZACHOS
aE(p 2i + i E ) = 4
Many, if not all, properties of the Liouville wave functions may be understood from the following integral representation "241, Chap. VI, Sec. 6.22, IZq. (lo)]. Explicitly
emphasizing the abovementioned nondegeneracy,
(-i(p+ie)+ifi)
2
(90)
= [ E - ( p f i ~ ) ~ ] @+~i e( )p.
(88)
(89)
(91)
The following
(92)
The right-hand side involves a special case of Meijer's C function,
(93)
(cf. [22], Sec. 5.3). which is fully symmetric in the parameter subsets {al ,...,a,,}, {a,+,
{b,,,+] ,..., b q } . It is possible to reexpress the result as a linear combination of generalized hypergeometric functions of type
, F 3 , but there is little reason to do so here. This transform is valid for %>O, and is taken from [26], p. 711, Eq. (55); The
transform is complementary to [27], Sec. 10.3, JZq. (49), in an obvious way, a K transform which appears in perturbative
computations of certain Liouville correlation functions [21].
The result (91) may be written in slightly different alternate forms
f(x,p)=
4 7r'
- sinh(7rG)
gT3
[ e4'l i&-ip
-i&-ip
GGx
2
'
2
1- 2 i f i - 2 i p
4
i&+ip
-i
1+2i&+2ip
4
1 -2i&+2ip
4
(94)
4There is an error in this result as it appears in [27], Vol. 11, Sec. 10.3, Eq. (58), where the formula has a2z2/4instead of a2zZ/16as the
argument of the G function. The latter argument is correct, and appears in Meijer's original paper cited here.
025002- 10
486
FEATURES OF TIME-INDEPENDENT WIGNER FUNCTIONS
by making use of the parameter translation identity for the G function 221, Sec. 5.3.1, Eq. (9)]:
(95)
Yet another way to express the result utilizes the Fourier transform of the wave function, Q. (86). in terms of which the
Wigner function reads, in general,
(96)
The specific result (86) then gives, as e+O,
(97)
However, this is a contour integral representation of the particular G function given above. Because of the E prescription, the
contour in the variable z=k/2+ie runs parallel to the real axis, but slightly above the poles of the r functions located on the
real axis at z = p - fi,z = p + fi,z = - p + fi,and z= - p - fi.Changing variables to s = i i z yields
(98)
ffLiouuille*f(x,P)
jpz-E-zd:+s~
COS
dp f(x,p)=O.
(101)
The first of these is a first-order differential-difference equation relating the x and p dependence:
1
e - 2 a ~ ( x , p=) 2ip I f ( x , p + i ) - f ( x , p - i) I .
(102)
d:)fW
+ y [ f ( x , p + i ) + f ( x , p - i ) ] = 0.
(103)
(99)
For real E and real f ( x . p ) . the imaginary part of this
genvalue equation is
(-pd,+e2
sin d , ) f ( x , p ) = O ,
*-
( 100)
(104)
487
THOMAS CURTRIGHT, DAVID FAIRLIE, AND COSMAS ZACHOS
We leave it as an exercise for the reader to exploit the recursive properties of the Meijer G function and show that this
difference equation is indeed obeyed by the result (91). Rather than pursue this in detail, we turn our attention to the
transformation functional which connects the above result for f to a free-particle Wigner function.
Given Eq. (90). it follows that
(105)
and hence NE=[4rrfieTJE sinh(rrfi)]ln/2rr, if we choose a S ( E , - E , ) normalization for the free-particle plane waves as
well as for the Liouville eigenfunctions. Therefore, lemma 3 yields
IN
l ( x , p ; X , P ) = - dYdy exp[-iyp+iP Y - ~ F * ( X - ~ / ~ X - Y / ~ ) + ~ F ( X + ~ / ~ , X + Y / ~ ) ]
2rr
1
= 7 [ 4 7 r f i e n E s i n h ( r f i ) ] \ dYdy exp - i y p + i P Y-iex-yn sinh
(2n)
1
=7 [ 4 n f i e T E
4rr
sinh(rrfi)]
(106)
(107)
(108)
(109)
(110)
(111)
(112)
(113)
(114)
025002-12
488
(115)
I
T h e s u m o f E q s . (110),(111),and(115)showsthatEq. (109)
is, indeed, satisfied.
Integrating over X and P the product of Z ( x , p ; X , P ) and
the free-particle Wigner function, as given here by
(4w4?)-'6(P - fi),yields another expression for the
Liouville Wigner function which checks against the previous
result, Eq. (91). Using Eq. (92) and the parameter translation
identity for the G function, this other expression is just Eq.
(94).
Supersymmetric Liouville quantum mechanics is obtained
by carrying through the Darboux construction detailed above
(with fi = 1 = 2m), for the choice
W ( x )= e x .
(116)
H' = p Z + e Z x + e x ,
( 120)
1 1"
x [ i K l n - , d e " ) - i K ~ z + , d e ~ ) l , (121)
and may be obtained from the previous E>O eigenfunctions,
as c ~ . ~ ( x ) = ( ~ / f i ) ( ~ ~ + ~ ) c ~ . ~ ( x ) .
For both Hamiltonians, the Wigner functions are straightforward to construct directly, once again leading to the K
transform (92) and particular Meijer G functions. W e find it
sufficient here to consider only the ground state for H ,
(117)
and the allowed spectrum is O<E<m, including zero energy, for which there is a bounded wave function normalized
as part of the continuum,
(118)
(122)
a single modified Bessel function. It smoothly [29] satisfies
[p - i W ( x ) ] *f o = 0 and, hence, the *-genvalue equation
H*fo=O.
ACKNOWLEDGMENTS
(119)
025002-13
489
THOMAS CURTRIGHT, DAVID FAIRLIE, AND COSMAS ZACHOS
025002-14
490
Modern Physics Letters A, Vol. 16, No. 37 (2001) 2381-2385
@ World Scientific Publishing Company
THOMAS CURTRIGHT
Department of Physics, University of Miami, Box 248046, Coral Gables, F L 33124, USA
curtright@physics.miami. edu
COSMAS ZACHOS
High Energy Physics Division, Argonne National Laboratory, Argonne, IL 60439-481 5, USA
zachosOhep. anl.gov
Received 31 October 2001
A concise derivation of all uncertainty relations is given entirely within the context of
phase-space quantization, without recourse to operator methods, to the direct use of
Weyls correspondence, or to marginal distributions of x and p .
Keywords: Uncertainty relations; star product; Wigner functions; phase-space quantization; negative probability.
PACS Nos.: 03.65.Ta, 03.65.Db, 03.65.Fd, 11.15.K~
Phase-space quantization is the third autonomous and logically complete formulation of quantum mechanics beyond the conventional ones based on operators
in Hilbert space or path integral~.l-~
It is free of operators and wave functions:
observables and matrix elements are computed through phase-space integrals of
c-number functions (classical kernels) weighted by a Wigner function (WF).334
This is a phase-space distribution function which is not positive semi-definite, and
constitutes the Weyl correspondent of the density matrix in the conventional formulation,
(1)
Operators of the conventional formulation, when properly ordered ( e g Weylordered), correspond uniquely to phase-space classical kernel functions, while
operator products correspond to *-products6 of these classical kernels, the *-product
being a noncommutative and associative operation encoding quantum mechanical
action. The above wave functions, however, may be forfeited, since the WFs are
determined, in principle, as the solutions of the celebrated *-genvalue functional
equation^.^-'^ Connections to the original, operator, formulation of quantum mechanics may thus be ignored.
2381
49 1
2382
(2)
which is the cornerstone of phase-space quantization. Its mechanics is reviewed in
Refs. 7, 10 and 15. An alternate, integral, representation of this product i d 6
f*g =
dudvdwdz f ( X + u , p + v ) g ( x + w , p + z )
(3)
which readily displays associativity. The phase-space trace is directly seen in this
representation to obey
Idpdxf*g =
dpdx f g
dpdxg* f .
(4)
(4)
The WF spectral propertied are reviewed and illustrated in Refs. 8 and 10. For
example the trace-normalization condition,
(5)
(5)
492
Negative Probability and Uncertainty Relations
2383
(6)
An arbitrary phase-space function cp(x,p) can thus be expanded as cp(z,p) =
c m n f m n ( Z ,P ) .
Here, a concise proof of all uncertainty relations is provided completely within
the autonomous framework of phase-space quantization, unlike extant discussions
of such correlation inequalities, which rely on the operator formulation of quantum mechanics. It is stressed that, in the following, no operators occur, only the
*-product operation, and x and p are c-numbers. The controlling fact is that expectation values of arbitrary real *-squares are positive semi-definite, even though the
Wigner distribution f(x,p ) itself is not. Specifically, for any complex phase-space
function g(x,p ) , and any (real) Wigner function f(x,p ) representing a pure state,
the following inequality holds:
Em,,
(9* 9 ) =
dP W g * * g ) f 2 0 .
(7)
The * is absolutely crucial here, and its removal leads to violation of the inequality, as can easily be arranged by choosing the support of g to lie mostly in those
regions of phase-space where the Wigner function is negative. (The only pure state
W F which is non-negative is the G a u s ~ i a n . ~In- ~Hilbert
~ ~ ~ ) space operator formalism, this relation (7) would correspond to the positivity of the norm. By (4),
J d p d z ( g * * g ) f = J d p dz(g* * g) * f,i.e. inside a phase-space integral an ordinary
product can be extended to a *-product, provided it not be part of a longer string.
For example the one *-product of the left-hand side cannot be eliminated, because
of the extra ordinary product with f .
To prove the inequality (7), it suffices to recognize that, for a pure state, its (real)
Wigner function can be expanded in a complete basis of Wigner *-genfunctions of
a convenient Hamiltonian," f =
ctcnfmn, for complex coefficients c,,s.t.
(cnI2= 1 , to satisfy (5). Then, it follows thatl8>l9
En
En,,
f *f
=f/h.
(8)
Consequently, given the relations (4),(g * f ) * = f*g*, and the associativity of the
*-product,
=h/dxdp(f *g*)*(g*
f)=h
dxdp)g* fI2
(9)
493
2384
T. Curtright 4 C. Zachos
This expression, then, involves a real non-negative integrand and is itself positive
semi-definite. (Similarly, if f l and f 2 are pure state WFs, the transition probability between the respective states3 is also manifestly non-negative by the same
argument: J d p d x f l f 2 = (27rti)2Jdxdplfl* f 2 I 2 . )
Given ( 7 ), correlations of observables follow conventionally from specific choices
of g ( x ,p ) . For example, to produce Heisenberg's uncertainty relation, one only need
to choose
g =a
+ bx + cp ,
(10)
a*a+b*b(x*x) + c * c ( p * p )
+ (a*b+b*a)(x)
(11)
for any a , b, c. The eigenvalues of the corresponding matrix are then non-negative,
x*x=x2,
p*x=px-iti/2,
p*p=p2,
x*p=px+ih/2,
(12)
( A x ) 2= ( ( x- (xu2)7
(API23 ( ( P - ( P H 2 )
(13)
(14)
and hence
(15)
A x A p 2 h/2
AUAV2
- J ~ ( U * W -w
* u ) ~( ~( u- ( t i ) ) * ( .- (v))
+ ( W - ( w ) ) * ( u- ( u ) ) ) ~ .
(16)
(16)
(Au(w- ( v ) )- k i A ~ ( u ( u ) ) *) f = 0 ,
where k is the sign of i(u* v
Refs. 8-10 and 15.
(17)
(17)
494
Negative Probability and Uncertainty Relations 2385
Acknowledgment
This work was supported in part by the U.S. Department of Energy, Division of
High Energy Physics, Contract W-31-109-ENG-38, and the NSF Award 0073390.
References
1. J . Moyal, Proc. Camb. Phil. SOC.45,99 (1949).
2. F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz and D. Sternheimer, Ann. Phys.
111,61; 111 (1978).
3. For reviews, see M. Hillery, R. OConnell, M. Scully and E. Wigner, Phys. Rep. 106,
121 (1984); H.-W. Lee, ibid. 259,147 (1995); N. Balasz and B. Jennings, ibid. 104,
347 (1984); R. Littlejohn, ibid. 138, 193 (1986); M. Berry, Philos. Trans. R. SOC.
London A287, 237 (1977); M. Gadella, Fortschr. Phys. 43, 229 (1995); L. Cohen,
Time-Frequency Analyszs (Prentice-Hall, 1995); F. Berezin, Sou. Phys. Usp. 23,763
(1980); P. Carruthers and F. Zachariasen, Rev. Mod. Phys. 55, 245 (1983); A. M.
Ozorio de Almeida, Phys. Rep. 295,265 (1998).
4. E. Wigner, Phys. Rev. 40,749 (1932).
5. H. Weyl, 2. Phys. 46,1 (1927); also reviewed in H. Weyl, The Theory of Groups and
Quantum Mechanics (Dover, 1931).
6. H. Groenewold, Physica 12,405 (1946).
7. D. Fairlie, Proc. Camb. Phil. SOC.60,581 (1964).
8. T. Curtright, D. Fairlie and C. Zachos, Phys. Rev. D58,025002 (1998).
9. M. Hug, C. Menke and W. Schleich, Phys. Rev. A57,3188 (1998).
10. T. Curtright, T. Uematsu and C. Zachos, J. Math. Phys. 42, 2396 (2001), hepth/0011137.
11. N . Seiberg and E. Witten, JHEP 9909,032 (1999); for a review, see L. Castellani,
Class. Quantum Grav. 17,3377 (2000), hep-th/0005210.
12. T. Yoneya, Mod. Phys. Lett. A4, 1587 (1989); A. Jevicki and T. Yoneya, Nucl. Phys.
B535,335 (1998), hep-th/9805069; N. Seiberg, L. Susskind and N. Toumbas, JHEP
0006,044 (2000), hep-th/0005015.
13. M. Bartlett, Proc. Camb. Phil. SOC.41,71 (1945).
14. R. Feynman, Negative probability, in Essays in Honor of David Bohm, eds. B. Hiley
and F. Peat (Routledge and Kegan Paul, 1987); for a popular review, see D. Leibfried,
T. Pfau and C. Monroe, Physics Today, 22 (April 1998).
15. F. Hansen, Rep. Math. Phys. 19,361 (1984); C. Roger and V. Ovsienko, Russ. Math.
Surv. 47,135 (1992); C.Zachos, J . Math. Phys. 41,5129 (2000), hep-th/9912238.
16. J. von Neumann, Math. Ann. 104,570 (1931).
17. R. Hudson, Rep. Math. Phys. 6,249 (1974).
18. T. Takabayasi, Prog. Theor. Phys. 11,341 (1954).
19. G. Baker, Phys. Rev. 109,2198 (1958).
495
JOURNAL OF MATHEMATICAL PHYSICS
JUNE 2001
Tsuneo Uematsub)
Depanment of Fundamental Sciences, FIHS. Kyoto University, Kyoto 606-8501, Japan
Cosmas Zachosc)
High Energy Physics Division, Argonne National Laboratory,
Argonne, Illinois 60439-4815
(Received 15 January 2001; accepted for publication 26 February 2001)
In the context of phase-space quantization, matrix elements and observables result
from integration of c-number functions over phase space, with Wigner functions
serving as the quasiprobability measure. The complete sets of Wigner functions
necessary to expand all phase-space functions include off-diagonal Wigner functions, which may appear technically involved. Nevertheless, it is shown here that
suitable generating functions of these complete sets can often be constructed, which
are relatively simple, and lead to compact evaluations of matrix elements. New
features of such generating functions are detailed and explored for integer-indexed
sets, such as for the harmonic oscillator, as well as continuously indexed ones, such
as for the linear potential and the Liouville potential. The utility of such generating
functions is illustrated in the computation of star functions, spectra, and perturbation theory in phase space. 0 2001 American Institute of Physics.
[DOI: 10.1063/1.1366327]
1. INTRODUCTION
General phase-space functions f(x,p) and g ( x , p ) compose noncommutatively through Groenewold's *-product,' which is the unique associative pseudodifferential deformation* of ordinary
products:
(1)
This product is the cornerstone of deformation (phase-space) quantization^-' as well as applications of matrix models and noncommutative geometry ideas in M-physics.6 Its mechanics, however, is not always straightforward.
The practical Fourier representation of this product as an integral kernel has been utilized
widely since Baker's' early work,
1
f*g=
(2)
The determinantal nature of the star product controls the properties of the phase-space
"Electronic mail curmght@physics.miami.edu
b)Electronicmail: uematsu@phys.h.kyoto-nJ.ac.jp
"Electronic mail: zachos@hep.anl.gov
0022-2488/2001/42(6)/2396/20/$18.00
2396
Downloaded 16 Oct 2002 to 146.133.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jrnp/jrnpcr.jsp
496
J.
2397
(3)
The above-mentioned *-product and phase-space integrals provide the multiplication law and,
respectively, the trace in phase-space q~antization,~
the third autonomous and logically complete
formulation of quantum mechanics beyond the conventional formulations based on operators in
Hilbert space or path integrals. (This formulation is reviewed in Refs. 2 and 5.) Properly ordered
operators (e.g., Weyl-ordered) correspond uniquely to phase-space c-number functions (referred to
as classical kernels of the operators in question); operator products correspond to *-products of
their classical kernels; and operator matrix elements, conventionally consisting of traces thereof
with the density matrix, correspond to phase-space integrals of the classical kernels with the
~ celebrated
Wigner function (WF), the Weyl correspondent of the density m a t r i ~ . ~The
* -genvalue functional equations determining the Wigner functions and their spectral properties
(e.g., projective orthogonality) are reviewed and illustrated in Ref. 4.
The functions introduced by Wigner and Szilard correspond to diagonal elements of the
density matrix, but quantum mechanical applications (such as perturbation theory), as well as
applications in noncommutative soliton problemsL3often require the evaluation of off-diagonal
matrix elements; they therefore utilize the complete set of diagonal and off-diagonal generalized
Wigner functions introduced by MoyaL3 For instance, in noncommutative soliton theory, the
diagonal WFs are only complete for radial phase-space functions (functions *-commuting with the
harmonic oscillator Hamiltonian-the radius squared), whereas deviations from radial symmetry
necessitate the complete off diagonal set.
As for any representation problem, the particular features of the *-equations under consideration frequently favor an optimal basis of WFs; but, even in the case of the oscillator, the equations are technically demanding. It is pointed out here, however, that suitable generating functions
for them, acting as a transform of these basis sets, often result in substantially simpler and more
compact objects, which are much easier to use, manipulate, and intuit. In the following, after some
elementary overview of the Weyl correspondence formalism (Sec. II), we illustrate such functions
for the harmonic oscillator (Sec. III), which serves as the archetype of WF bases indexed discretely; it turns out that these generating functions amount to the phase-space coherent states for
WFs, and also the WFs of coherent state wave functions (Appendix A). Direct applications to
first-order perturbation theory are illustrated in Appendix B.
For sets indexed continuously, the generating function may range from a mere Fourier transform, illustrated by the linear potential (Sec. IV), to a less trivial continuous transform we provide
for the Liouville potential problem (Sec. V), where the advantage of the transform method comes
to cogent evidence.
Throughout our discussion, we provide the typical *-composition laws of such generating
functions, as well as applications such as the evaluation of * -exponentials of phase-space functions (Appendix C), or *-versions of modified Bessel functions (technical aspects of integral
transforms of which are detailed in Appendix D). Appendix E provides the operator (Weyl-)
correspondent to the generating function for the Liouville diagonal WF introduced in Sec V.
II. OVERVIEW OF GENERAL RELATIONS IN THE WEYL REPRESENTATION
Without loss of generality, we review basic concepts in two-dimensional phase space, ( x , p ) ,
as the extension to higher dimensions is straightforward. In addition, we first address discrete
spectra, E n , n =0,1,2,3,. . . , and will only later generalize to continuous spectra.
In the Weyl corre~pondence,~
c-number phase-space kernels a ( x , p ) of suitably ordered
operators d(X,P) are defined by
(4)
Conversely, the ordering of these operators is specified through
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright,see http://ojps.aip.org/jmp/jrnpcr.jsp
497
2398
(5)
An operator product then corresponds to a star-composition of these kernels,'
(6)
Moyal' appreciated that the density matrix in this phase-space representation is a Hermitean
generalization of the Wigner function:
(7)
where the $m(n)'sare (ortho-)normalized solutions of a SchrGdinger problem. (Wigner" mainly
considered the diagonal elements of the density matrix (pure states), usually denoted as f,
=f,, .) As a consequence, matrix elements of operators are produced by mere phase-space
integra~s,~
(8)
The standard machinery of density matrices then is readily transcribed in this language, e.g.,
the trace relation?
(9)
(10)
(11)
For complete sets of input wave functions, it also follows that3
(12)
(13)
the coefficients being specified through (1l),
(14)
Downioaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jrnp/jmpcr.jsp
498
J. Math. Phys., Vol. 42, No. 6, June 2001
Further note the resolution of the
2399
id en tit^,^
E f,,(x,p)=
n
1
-=
2nfi
(15)
-.h
For instance, for eigenfunctions of the Hamiltonian H(X,P)with eigenvalues E n , the corresponding WFs satisfy the following star-genvalue equations (also see Refs. 1 1 and 4). with
H ( x , p ) , the phase-space kernel of H(X,P):
H*fmn=En.fmn
fmn*H=Emfmn.
(16)
a
ifi x f ( x , p ; t ) = H * f ( x . p ; t ) - f ( x , p ; t ) * H .
By virtue of the *-unitary evolution operator (a *-exponential
(17)
2).
(18)
the time-evolved WF is obtained formally in terms of the WF at t = 0,
(19)
f(x,p;t)= u;(x,p;t)*f(x,p;o)*u,(n,p;t).
(These associative combinatoric operations completely parallel those of operators in the conventional formulation of quantum mechanics in Hilbert space.) Just like any star-function of H , this
*-exponential can be computed,6
exp,(itH/fi)= exp,( i t H / h ) * 1 = exp,( i t H / f i ) * 2 7 r f i E f n n = 2 7 r h x e i r E n l f i f n n . (20)
dx d p f E l E 2 ( x , P )= ~
( E I-
~ z )
(21)
(22)
(23)
More generally, (10) extends to
1
fElE2*fE;E;=
-E ; ) f E ;
E2
(24)
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
499
J. Math. Phys., Vol. 42, No. 6, June 2001
2400
(25)
and hence (20) extends to
(26)
111. GENERATING FUNCTIONS FOR THE HARMONIC OSCILLATOR
Consider the harmonic oscillator,
(27)
where, without loss of generality, parameters have been absorbed in the phase space variables:
rn = 1, w = 1 . Further recall that the normalized eigenfunctions of the corresponding operator
Hamiltonian 3-1 are ~ ~ ( ~ ) = ( ~ 2 " n ! ) - " ~ e - ( " ~ for
) " the
~ Heigenvalues
~ ( x ) , E,=fi(n+ 112). Define a radial and an angular variable,
(28)
so that
(29)
Groenewold,' as well as Bartlett and Moyal,17 have worked out the complete sets of solutions
to Moyal's time-evolution equation (17), which are all linear combinations of terms exp(it(m
-n))fmn . They solved that equation indirectly, by evaluating the integrals (7) for time-dependent
Hermite wave functions, which yield generalized Laguerre polynomial-based functions. More
directly, Fairlie' dramatically simplified the derivation of the solution by relying on his fundamental equation (16). He thus confirmed Groenewold's WFs,'.17
(30)
The special case of diagonal elements,
(31)
constitutes the time-independent "*-genfunctions" of the oscillator hamiltonian kernel4 [i.e., the
complete set of solutions of the time-independent Moyal equation H*f -f * H = O , where H * f ,
= E d , . Incidentally, (10) restricted to diagonal WFs closes them under *-multiplication,'2
fm*f,=
SmJm /(2rrfi).]That is to say, "radially symmetric" phase-space functions, i.e., functions that only depend on z but not 0, can be expanded in terms of merely these diagonal
elements-unlike the most general functions in phase space which require the entire set of offdiagonal f,, above for a complete basis. Note, however, that all *-products of such radially
symmetric functions are commutative, since, manifestly,
(32)
Downloaded 16 Oct 2002 to 146.139.180.65, Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
500
J. Math. Phys., Vol. 42, No. 6, June 2001
2401
Moreover, the *-exponential (20) for this set of *-genfunctions is directly seen to amount to
(33)
which is, to say, a Gaussian in phase space.' As an application, note that the hyperbolic tangent
*-composition law of Gaussians follows trivially, since these amount to *-exponentials with
additive time intervals, exp,(rf)*exp,(Tf )=exp,((t+T)f)?
(34)
We now introduce the following generating function for the entire set of generalized Wigner
functions:
(35)
Utilizing the identity" 8.975.2,
(36)
(37)
Thus,
(38)
(38)
&(ae-"+pe")-ap-
Since
pel') = v ~ ( r y + p )n\ j z i p ( a -
p).
(39)
(39)
(40)
As the name implies, from G( a , P ; x , p ) , the f m n ' s are generated by
(41)
These functions *-compose as
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
501
2402
e
G ( ~ , P ) * G ( E . O ~= G ( E , P ) .
(42)
dxdpG(a,P)=eap.
(43)
(44)
and
(45)
Consequently,
dxdpH*G(a,P)=h
(46)
(47)
In general, matrix elements of operators may be summarized compactly through this generating
function in phase space.
This generating function could be interpreted as a phase-space coherent state, or the offdiagonal WF of coherent states, as discussed in Appendix A,I9
G( a.P;x,p) = exp,(pa t)foexp,( a a ) ,
(48)
(49)
(50)
This formalism finds application in, e.g., perturbation theory in phase space, cf. Appendix B.
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jrnp/jmpcr.jsp
502
J. Math. Phys., Vol. 42, No. 6, June 2001
2403
The linear potential in phase space has been addressed" (also see Refs. 19 and 4). We shall
adopt the simplified conventions of Ref. 4,i.e., rn = 1/2, fi = 1. The Hamiltonian kernel is then
H(x,p) = p2+x,
(51)
(52)
indexed by the continuous energy E . The spectrum being continuous, the Airy functions are not
square integrable, but have continuum normalization, Jdx I(Ii,(x) i,bE2(x) = S(El - E 2 ) , instead.
Thus, (21) et seq. are now operative. The generalized WFs are"
(53)
The *-exponential (26) then is again a plain exponential of the shifted Hamiltonian kernel,
exp,( i t ( x + p 2 ) ) = 2 a
x+p2- 2
= exp(it(x+p2+ t2/12)).
(54)
(Thk could also be derived directly, as the CBH expansion simplifies dramatically in this case, cf.
Appendix C.) As before, the *-composition law for plain exponentials of the hamiltonian kernel
function follows,
(55)
Since the complete basis Wigner functions are now indexed continuously, a generating function for them must rely on an integral instead of an infinite sum. The simplest transform is
possibly a double Fourier transform with respect to the energy indices [but note the transform
factors exp(iEIX), exp(-iE2Y) may also be regarded as plane waves]. Suitably normalized,
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.orgljmp/jmpcr.jsp
503
2404
(56)
The phase-space trace is
dxdpG(X,Y;x,p)=2.rr6(X-Y),
( 5 7 ) (57)
=
6 ( E ,- E ; ) ~ E ; Ethe
~ ,*-composition
and, given (24) for these functions, ~ E ~ E * * ~ E ; E (;1/27r)
(58)
A less trivial system with a continuous spectrum is the Hamiltonian with the Liouville
potential.2032'In the conventions of Ref. 4 (fi= 1, m = 1/2), the Hamiltonian kernel is
H=p2+eZx,
(59)
-excoshX+iXp)=K-jp(ex).
(61)
7,/
1
f ~ , ~ ~ ( x , p ) = dy e - 2 i P y ~ ~ K ~ ~ e ~ - Y ) ~ ~ K ,(62)
~ ( e x ' Y ) .
This Wigner function amounts to Meijer's G function,
1
~E~E~(X9
, P ) =&inh(.rr\jEl)sinh(n\jE2)
(63)
dsinh(m6)sinh(rrJE(q))
f ~ ( k E) ( , ) ( ~ , P ) =
X I
cosh Y ' p
dXdY eJkXe'qYcash x) K2,,(eXJ4 cash X cash U,
(64)
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
504
J. Math. Phys., Vol. 42, No. 6, June 2001
2405
(65)
The form and construction of this G are consequences of (61), as detailed in Appendix D.
However, the *-composition law of this particular generating function is not so straightforward. It is singular, as a consequence of the general relation (24) and the behavior of the integrand
in (65) as k,q-+O.
The singularity may be controlled by regulating the *-product through imaginary shifts in the momenta,
It follows that one derivative with respect to either of X or Z suffices to eliminate the divergence at E = O ,
Unlike the situation in (58), here the right-hand side vanishes at X = Z . More symmetrically,
6-10
-tanhX-
cosh X + cosh Z
By some contrast to the above, Eq. (65), an alternate generating function for just the diagonal
WFs, f E E = f E , could be defined through the spectral resolution of the * - K function,
(66)
(67)
and y - + x - Y ,
(UT)J dY e -liPY,
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
505
2406
(68)
Finally, simplifying the right-hand side gives
(69)
As a side check of this expression, (69), note that it must satisfy
~ * G ( z ; x , p=) G ( z ; x , p ) * ~ =( -
a;+ e Z z ) G ( z ; x , p ) ,
((70)
70)
which follows from the spectral resolution evident in (66). Indeed, since e-'d,KIp(e')
= i p e - Z K , p ( e Z ) - K , p + l ( e z )and
, ( - d ~ + e 2 Z ) K , , ( e L ) = p 2 K , , ( e zthese
) , relations are satisfied,
(71)
Parenthetically, as an alternative to the ordinary product form in (69), the phase-space kernel
(Note:
Do not shift the integration parameter y by the phase-space variable n before the star products are
evaluated.),
*exp(iyp-eZcoshy).
(72)
i:
(73)
- -eZX-I+iyp
The ordinary product form in (69) and the *-exponential form in (72) reveal that G(z;x,p)
= G ( z ; x , - p ) , so one may replace exp(iyp) by cos(yp) in the second line of (72). Given these,
there are several ways to verify (70). These relations and the star-product expressions for the
kernel in (72) are isomorphic to those of the corresponding operators, as discussed in Appendix E.
The *-composition law of these generating functions follows from (24) and Macdonald's
identity,
'I i :
G ( u ; x , p ) * ~ ( u ; x , p )2=
This also follows directly from the explicit form (69). Again, this is isomorphic to the corresponding operator composition law given in Appendix E.
From the orthogonality of the &'r, :he diagonal WFs may be recovered by inverse transformation,
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
506
J.
2407
(75)
This representation and the specific factorized x,p-dependence of B can be of considerable use,
e.g., in systematically computing diagonal matrix elements in phase space.
In illustration of the general pattern, consider the first-order energy shift effected by a perturbation Hamiltonian kernel H I . It is, cf. Appendix B; Eq. (Bll),
(76)
Choosing
H , =e 2 n x e i s ~ R ,
(77)
yields
(78)
Now.
(79)
xr(n-if+iplr(
n-if-ip)
(80)
Thus,
xr[
n+i&-ip
2
n-i&+ip
2
)r[
n-ifi-ip
2
)r(
1.
(81)
(82)
Downloaded 16 Oct 2002 to 146.139.180.65.Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
507
2408
12-n
+ i&
2 -n -
i&\
(83)
In principle, any polynomial perturbation in either x or p can be obtained from this, by differentiation with respect to n and s. (Retaining a bit of exponential in x would be helpful to suppress
the region of large negative x).
ACKNOWLEDGMENTS
We wish to thank D. Fairlie and T. Hakioglu for helpful conversations. This work was
supported in part by the US Department of Energy, Division of High Energy Physics, Contract No.
W-31-109-ENG-38; NSF Award No. 0073390; and by the Grant-in-Aid for Priority Area No. 707
of the Japanese Ministry of Education. T. U.and T. C. thank Argonne National Laboratory for its
hospitality in the summer of 2000.
APPENDIX A: *-FOCK SPACE AND COHERENT STATES
Dirac's Hamiltonian factorization method for algebraic solution of the harmonic oscillator
carries through (cf. Ref. 2) intact in *-space. Indeed,
1
2
H=-(x-ip)*(x+ip)+
(A1)
-,
2
motivating definition of
(A2)
Thus, noting
(A3)
provides a *-Fock vacuum, it is evident that associativity of the *-product permits the entire
ladder spectrum generation to go through as usual. The * -genstates of the Hamiltonian, such that
H * f = f * H , are thus
(A5)
These states are real, like the Gaussian ground state, and are thus left-right symmetric *-genstates.
They are also transparently *-orthogonal for different eigenvalues; and they project to themselves,
as they should, since the Gaussian ground state does, fo*fo=fo/2nfi.
The complete set of generalized WFs can thus be written as
1
f
In"
=-
'
( a *)"fo(* a ) m , m , n = 0,1,2,3;
JiG
..
(A6)
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
508
J. Math. Phys., Vol. 42, No. 6,
June 2001
2409
The standard combinatoric features of conventional Fock space apply separately to left and
right (its adjoint) *-multiplication:
(A7)
(A8)
Up to a factor of e~p((lal~+IP1~)/2),
this is also the WF of coherent states la) and (PI.= As
indicated in the text, this coherent state is identifiable with the generating function G for the
harmonic oscillator.
APPENDIX B: STATIONARY PERTURBATION THEORY
Perturbation theory could be carried out in Hilbert space and its resulting wave functions
utilized to evaluate the corresponding W F integrals. However, in the spirit of logical autonomy of
Moyal's formulation of quantum mechanics in phase space, the perturbed Wigner functions may
also be computed ab initio in phase space,"*25without reference to the conventional Hilbert space
formulation. The basics are summarized in the following.
As usual, the Hamiltonian kernel decomposes into free and perturbed parts,
H=H,+hH,.
(B 1)
are solved upon expansion of their components E and f in powers of A, the perturbation strength,
(B3)
(B4)
Note the superscripts on E and f are order indices and not exponents. Resolution into individual
powers of A yields the real equations:
(B5)
(B6)
(B7)
E* yields
(B8)
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright.see http://ojps.aip.org/jmp/jmpcr.jsp
509
2410
and, by (B5).
E * H *E= E: E *A;
1
(B9)
(B10)
(B11)
the diagonal element of the perturbation. For the off-diagonal elements, similarly left-*-multiply
(B6) by f i
J", *H,*ff,
By completeness,
+.tt*H,
*t= E Y m * f ! , + E f f m * E ,
((BIZ)
B12)
fk , i f 0, resolves to
(B13)
(b14)
(B15)
and hence
(B16)
(B17)
so that
(B18)
Finally, use of (1 l), yields
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
5 10
J. Math. Phys., Vol. 42, No. 6,June 2001
~ ~ , ~ , , , = ( 2 7 r hd x) ~d p1 Pml*Pm*Hl
241 1
*t
~g-~lj,
(B19)
((B20)
B20)
By (8). it can be seen that the same result may also follow from evaluation of the WF integrals of
perturbed wave functions obtained in standard perturbation theory in Hilbert space.
For example, consider H I = v2 x = a +at . It follows that EA = 0, and
4,
(E,O- E:)
11
dxdP(Jm+lPm+.l,n+mPm,"+l)
=a n , l ~ J m + l ~ r n + l , n -
Jn';+la r n , n + ] ) ,
((B21)
b21)
fA= &(t1,nft.n-
1)-
J n ? i t t , n + l +t+l,n).
((B22)
B22)
The *-exponential (54) of the Hamiltonian kernel for the linear potential is also easy to work
out directly, since the combinatorics in *-space are identical to the combinatorics of any associative algebra. In particular, the Campbell-Baker-Hausdorff expansion also holds for *-exponentials,
exp,(A)*exp,(B)=exp,(A+B+
(C1)
where C represents a sum of triple or more nested *-commutators (Moyal Brackets, [A,B],
-2it'p
=A*B-B*A).
Now, choosing A = i t x and B = i r p 2 + i t 2 p + $ t 3 , yields [A$],=
- i t 3 , [A,[A,B]*]*=2it3, [[A,B],,B],=O, and hence C=O.
Consequently,
(C2)
But further note exp,(ux)=exp(a), and also exp,(bp2+cp+d)=exp(bp2+cp+d). This reduces the
*-product to a mere translation,
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/j/jmp/jmpcr.jsp
511
2412
exp,(ax)*exp,(bp2+ c p + d ) = exp(ax)*exp( b p 2 + c p + d )
= exp(ax
+ $ iad,)exp(
= exp(ax+ b ( p
bpz cp + d )
+ t ial2+ c ( p+
i a )+ d )
=exp(ax+ bp2+ ( c i n b ) p + d - a a z b + ) i a c ) .
(C3)
(C4)
Consequently,
+ +
(54')
follows.
The proof of
(73)
is similar. Choosing now A = - ) e y - ' e Z x and B = i y p , it follows that [A&],= -2yA, so that
only those multiple Moyal commutators survive which are linear in A. This means, then, that in
the Hausdorff expansionz6 for Z ( A ,B)=In,(exp,(A)*exp,(B)), only B and terms linear in A
survive. Hence, Z reduces to merely
(c5)
The Hadamard expansion in B ] , means successive right *-commutation with respect to B as
many times as the regular power expansion of the function in the parenthesis dictates. Consequently,
(C6)
On the other hand,
exp( - +eY-ze2x)*exp(iyp)=exp( - +eY-z+2x)exp(iy(p+ i i x / 2 ) )= exp( - + e Z x - z + iyp),
(C7)
and the identity is proven.
APPENDIX D: CONSTRUCTION OF THE GENERATING FUNCTION FOR THE
LlOUVlLLE WFS
From (60) and (61), it is evident that the Liouville wave functions can be generated by
dk
exp( - e x cosh X) =
I - d z z z cikX
$Edx).
(D1)
where E ( k ) = k 2 . Therefore, the usual wave function bilinears appearing in the WFs are generated
by (recalling that the Q s are real)
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright. see http://ojps.aip.org/jrnp/jmpcr.jsp
512
J. Math. Phys., Vol. 42,No. 6,June 2001
2413
dk
-ikX-
iqY
4!'E(k)(X--Y)4!'E(q)(x+Y).
(D2)
coshX)exp( - ex+Ycosh Y)
(D3)
Evaluation of this expression yields just a factor multiplying a modified Bessel function,
dy e-2ipyexp( -eX-YcoshX-ex+Ycosh Y )
coshY jP
=2 coshX) K2ip(e"J4coshXcosh Y).
(D4)
Thus, a generating function for the complete set of Liouville Wigner functions is
(-)
2 cosh Y
T coshX
dk
jP
iqY
fE(k) E(q)(X,P),
((65')
65')
as in the text.
APPENDIX E: OPERATOR ORDERING AND EQ. (69)
((69')
69')
what is the operator corresponding to it? According to Weyl's prescription, Eq. ( 5 ) , the associated
operator is
d T d u dx
dp B( z;x,p)exp( i T( P- p ) + ia(X - x ) )
1
(El)
The integrals over x and p may be evaluated separately, if the u contour is first shifted slightly
above the real axis, u-+c+ i c , thereby suppressing contributions to the x-integral as x-+ --oo.
Now s=)eZx-' gives
5 13
2414
-I
-2e
-i(z+lnZ)u/Zr
( - i( a+ ie)/2).
dX e-ezCoShX
2774X-
dp K,,(e')exp( - iTp) =
T)= 77e-e2CoSh
r.
(E2)
(E3)
so
'I
'Osh
877
'exp( i TP+irk').
((E4)
E4)
'I
B(z;X, P ) = - d T
8n
-ez
(I
cosh r
exp( i TP)
=-
8T
COsh
exp( i ~ p )
(E5)
This is the operator correspondent to (72); it reflects the Weyl correspondence through which it
was originally defined (although, technically, it was taken out of Weyl ordering above, merely as
a matter of convenience, not a bona-fide change of representation).
This form leads to a more intuitive Hilbert space representation. Acting to the right of a
position eigen-bra, (xlX= (xlx, while the subsequent exponential of the momentum operator just
translates, (xlexp(i.rP) = (x +
So the full right-operation of 6 is
TI.
(E6)
Inserting 1 =I& Ix)(xI gives B(z;X,P)=Jdr Ix)(xl@(z;X,P), and leads to a coordinate
space realization of the operator involving an x,y -symmetric kernel,
B(z;X,P)=
1
1
dx dy Ix)(yl exp( - - e X + y - z - - ex-y+z2
2
I
2
-e-x+y+z
).
037)
This operator is diagonal on energy states: by Macdonald's identity (67), and the reality and
orthogonality of the wave functions,
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/j/jmp/jmpcr.jsp
514
J. Math. Phys., Vol. 42, No. 6, June 2001
2415
(E8)
This is in agreement with the corresponding phase-space expression, (66).
The composition law of this operator also parallels its phase-space isomorph, (74),
(E9)
'H. Groenewold. Physica (Amsterdam) 12, 405 (1946).
'F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, and D. Stemheimer, Ann. Phys. (N.Y.) 111, 61 (1978); 111. 1 1 1
(1978); Lett.Math. Phys. 1, 521 (1977).
'J. Moyal, Roc. Cambridge Philos. SOC.45, 99 (1949).
'T. Curtright, D. Fairlie, and C. Zachos, Pbys. Rev. D 58, 025002 (1998).
'For reviews, see M. Hillery, R. OConell, M. Scully, and E. Wigner. Phys. Rep. 106, 121 (1984); H.-W. Lee, ibid. 259,
147 (1995): N. Balasz and B. Jennings, ibid. 104, 347 (1984); R. Littlejohn. ibid. 138. 193 (1986); M. Berry, Philos.
Trans. R. Soc. London, Ser. A 287, 237 (1977); M. Gadella, Fortschr. Phys. 43, 229 (1995); P. Carmthers and F.
Zachariasen, Rev. Mod. Phys. 55, 245 (1983); L. Cohen, Time-FrequencyAnalysis (Prentice-Hall, Englewocd Cliffs, NJ,
1995): F. Berezin, Sov. Phys. Usp. 23,763 (1980); A. M. Ozorio de Ahneida, Phys. Rep. 295, 265 (1998).
6N. Seiberg and E. Witten, 1. High Energy Phys. 9909,032 (1999); for a review, see L. Castellani, Class. Quantum Grav.
17, 3377 (2000) [hepW0005210].
7 G . Baker, Phys. Rev. 109, 2198 (1958); formally underlain by J. v Neumann, Math. Ann. 104, 570 (1931).
'D. Fairlie, Proc. Cambridge Ph~los.SOC.60, 581 (1964).
9F. Hansen, Rev. Mod Phys. 19, 361 (1984); C . Roger and V Ovsienko, Russ Math. Surv. 47, 135 (1992); C. Zachos, 1.
Math. Phys. 41, 5129 (2000) [hep-tW9912238].
"E. Wigner, Phys. Rev. 40,749 (1932).
"I. Dahl, in Energy Storage and Redistribution, edited by J. Hinze (Plenum, New York, 1983). pp, 557-571, see also L.
Cohen, I. Math. Phys. 17, 1863 (1976); W. Kundt, Z. Naturforsch. A 22, 1333 (1967).
"T. Takabayasi, Prog. Theor. Phys. 11,341 (1954).
13R. Gopakumar, S. Minwalla, and A. Strominger, J. High Energy Phys. 0005,020 (2000) [hep-W0003160]; J. Harvey, P.
Kraus, F. Larsen, and E. Martinec, ibid. 0007,042 (2000) [hep-th/0005031]; K. Dasgupta, S. Mukhi, and G. Rajesh, ibid.
0006, 022 (2000) [hep-W0005006]; A. Polychronakos. Phys. Lett. B 495, 407 (2000) [hep-th/0007043]; J. Harvey,
hep-WO 102076.
I4H. Weyl, Z. Phys. 46. 1 (1927); also reviewed in H. Weyl, The Theory of Groups and Quantum Mechanics (Dover, New
York, 1931).
"T. Curhight and C. Zachos, J. Phys. A 32, 771 (1999).
16J. Gracia-Bonda, Phys. Rev. A 30, 691 (1984).
I7MvI.Bartlett and J. Moyal, Proc. Cambridge Philos. SOC.45, 545 (1949).
"I. Gradshteyn and 1. Ryzhik, Table oflntegrals. Series, and Products, 5th ed. (Academic, New York, 1994).
I9V. Dodonov and V. Man'ko, Physica A 137, 306 (1986).
"E. D'Hoker and R. Jackiw. Phys. Rev. D 26, 3517 (1982); E. Braaten, T. Curtright, G. Ghandour, and C. Thorn, Ann.
Phys. (N.Y.) 153, 147 (1984); T. Curtright and G. Ghandour, Phys. Len. B 136, 50 (1984); E. D'Hoker, Phys. Rev. D
28, 1346 (1983).
"C. Grosche and F. Steiner, Phys. Lett. A 123, 319 (1987).
22G. Watson, A Treatise on the Theory ofBessel Functions, 2nd ed. (Cambridge University Press, Cambridge, 1995).
23T.Curtright, quant-pN0011101.
24V.Han, Y. Kim, and M. Noz, Phys. Rev. A 40, 902 (1989); M. Hennings. T. Smith, and D. Dubin, J. Phys. A 28,6779
(1995); 28, 6809 (1995); J. Klauder and 8. Skagerstam, Coherent States (World Scientific, Singapore, 1985).
"L. Wang and R. O'Connell, Found. Phys. 18, 1023 (1988).
26W. Magnus, Commun. Pure Appl. Math. VII, 649 (1954).
Downloaded 16 Oct 2002 to 146.139.180.65. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jmp/jmpcr.jsp
515
VOLUME 57, NUMBER 5
PHYSICAL REVIEW A
MAY 1998
Modified spectral method in phase space: Calculation of the Wigner function. I. Fundamentals
M. Hug, C. Menke? and W. P. Schleichl
Abfeilungfur Quantenphysik, Universitut Ulm, 0-89069 Ulm, Germany
2Abteilung Numerik. Universitui Ulm, 0-89069 Urn, Germany
(Received 30 October 1997)
We present a method for the direct computation of the Wigner function by solving a coupled system of
linear partial differential equations. Our procedure is applicable to arbitrary binding potentials. We introduce a
modified spectral tau method that uses Chebyshev polynomials as shape functions to approximate the solution.
Since two differential equations are solved simultaneously, the resulting linear equation system is overdetermined. We approximate its solution by a least-squares method. We prove the stability and convergence of our
scheme. As an application, we compute numerically the Wigner function for the harmonic oscillator. Our
calculations show excellent agreement with known analytic results. [S1050-2947(98)04704-0]
An alternative formulation of standard quantum mechanics i la Schrodinger or Heisenberg is the phase-space approach pioneered by Wigner. Here the standard definition of
a Wigner function [l] is based on the density matrix. Hence
once we know the density matrix we find the Wigner function by performing a Fourier integral. However, one might
ask if phase space alone defines the Wigner function T,
Does there exist a set of real partial differential equations in
the phase-space variables 9 and p which determine VI? This
would avoid the detour through the density operator and the
Fourier integral. Indeed, a set of two real linear partial differential equations [2,3] defines the Wigner function. This
set is rarely investigated in the literature [4]. although it offers new insights into the Wigner function. To derive solutions for these phase-space equations is the topic of these two
papers.
According to Heisenbergs uncertainty principle, position
and momentum cannot be measured simultaneously and precisely. So, strictly speaking, one cannot define a phase-space
probability distribution in quantum mechanics. However,
several quasiprobability distributions [1,5,6] have been proposed and widely used that carry some of the key features of
classical phase-space densities. The most famous quasiprobability distribution is the Wigner function [7]. This function
has found a wide range of applications (in the fundamentals
of quantum mechanics, quantum optics, nuclear physics, and
plasma physics, to name just a few areas). The Wigner function shows many features of a classical phase-space density.
In particular, this quasiprobability has quantummechanically correct marginal distributions [8]. However,
because of the uncertainty principle the Wigner function differs in two important points from a classical phase-space
distribution [9]. One is a required symmetrical ordering of
observables [l]. Second, the Wigner function may contain
areas of negative probabilities. These negativities have
been considered prominent features of nonclassical behavior
[lo]. Quite typically, the negative areas occur as oscillations
arising from interference effects [I 1,121.
In this paper we introduce a spectral tau method [I31 to
compute the Wigner function directly for energy eigenstates.
We apply proposals of Fairlie [2(a)] and Kundt [2(b)]. They
1050-2947/98/57(5)/3188( I8)/$15.OO
57
-
3188
516
51
...
1
x'-[QQo
3189
$ (Qi+Q2)],
where
Qo(1)
k (Qz-Qi).
and
V(x)=v(Qox+
i(Qi+Qz)>
(2)
in the phase-space variables' positions q and momentum p ,
respectively. Here M and fi denote the mass of the particle in
the potential and Planck's constant, respectively.
In general, these equations are of infinite order in p . However, in the case of a polynomial potential U only a finite
number of derivatives contributes. In any case the solution
V ( q , p ) is symmetric in the momentum variable p . that is
( 4 .- P) = W q , p ) .
and
1
and
(3)
P2
((4)
4)
((55))
This is the set of equations we solve using our modified tau
method. In the present paper we truncate this set after the
second derivative of the potential in order to outline our
modified spectral method. In paper I1 [14] we extend this
technique to treat arbitrary high orders. We conclude this
section by noting that indeed these definitions limit the range
of the variables x and y to the phase-space square
[ - 1.11 x [ - I,l].
d2
-+ V( Q) - E - - 2
2
8 aQ
P
ysPO
and the new position variable
L*=f,
((6)
6)
where L is an arbitrary differential operator, f an inhomogeneity, and CC, the solution of that problem in a domain D . The
spectral method starts from the approximate solution
N
((7)
7)
of Eq. (6), that is a linear superposition of N known trial or
shape functions 4,. Here the coefficients a , are the constant
517
M. HUG, C. MENKE, AND W. P. SCHLEICH
3190
coefficients to be determined. When we substitute the approximate solution (cIN into Eq. (6),the residual
(8)
of the equation is in general not equal to zero. The aim of the
method is to find $ N , i.e., the unknown coefficients a i , such
that R is small. We achieve this goal by requiring that the
integral
Rq,=O
jD
(9)
of the weighted residuals over D vanishes. Here the quantities cpv with v = 1, ...,N are called weight or test functions.
From Eq. (9) we find, with Eq. (8), the N equations for linear
operators L,
of the N X N matrix
ra= b
(10)
for the vector ( a ) i = a i containing the N unknown coefficients a i of the spectral ansatz (7).
The spectral method uses the same orthogonal functions
4i= pias trial and weight functions leading to the relation
j p X = 4const.
whereby 15, and x denote the Kronecker symbol Eq. (CI)
and a positive weight function depending on the particular
choice of the lp, , respectively. Making use of these orthogonality relations only pairs lpiqvwith i= v give a nonvanishing contribution to the system of equations (10).
Usually the trial functions satisfy the same boundary conditions as required for the solution $. When this is not the
case the method is called the tau method. Then certain equations of system (10) are discarded and replaced by the
boundary constraints. It is worth mentioning that for sufficiently smooth solutions the spectral tau method yields results that are much more accurate than any finite-difference
scheme with the same number of unknowns, see Ref. [16].
Since in our specific problem there are no boundary conditions, the tau method is the basis of our approach. In contrast to usual applications of tau methods we have to solve
57
-
two real partial differential equations simultaneously corresponding to the two operators Lev,, and Ldd. This leads to
an overdetermined linear system consisting of 2 N equations
for N unknowns. Here we have two choices: Either we find a
way to reduce the system to a N X N system of full rank, or
we keep all the equations with more than one nonvanishing
coefficient, and apply a least-squares method to the full system. The second approach is valid, provided that the error
introduced by the least-squares approximation is not larger
than the spectral error. That means that in practice we do not
require that the weighted residuals Eq. (9) vanish, but we
minimize them in a least-squares sense. Our numerical experiments show that this approach does indeed yield solutions of very high accuracy for our model problem, where
the exact solution is known and can be used to check the
numerical solution.
IV. TRUNCATION OF THE SYSTEM
(1)
and
(12)
This is the set of equations we use throughout this paper. We
note that Eq. (11) is closely related [7] with the Liouville
equation of classical statistical mechanics.
V. IMPLEMENTATION OF OUR METHOD
In this section we present an implementation of our modified spectral tau method and proceed in three steps: First we
formulate our spectral ansatz Eq. (7) for the two coupled
differential equations (11) and (12). Since our problem involves two differential equations we have to evaluate two
518
57
residuals Eq. (8) in the second step. The last step is to build
up the linear system of equations Eq. (10).
We emphasize that the set of equations (11) and (12) can
be solved analytically [4] for the harmonic oscillator
V(Q) = Q2/2, and thus provides a good test of our numerical
results. When we include higher-order terms of the differential equations (1) and (2), the procedure remains the same.
Each additional order leads to one additional term in the
matrix coefficients of system (10). Since this does not increase the number of equations or the number of unknowns,
it is computationally not more expensive to include higherorder terms (see Ref. [14]).
3191
(15)
where the coefficients for k = 0.1,. .. read
NY
a;.Zk=E
s = k 2(2s+1)aj,&+l
(16)
(17)
We now present our modified tau method for the secondorder differential equations (11) and (12) for the Wigner
function. We follow Ref. [17], and use Chebyshev polynomials in our approach. They are recommended as the best
choice whenever a nonperiodic problem occurs or whenever
one works with general functions that do not allow us to use
special geometries. Since we seek a universal method valid
for an arbitrary potential, we face just such a situation.
Chebyshev polynomials give good results, and converge fast
enough under almost all circumstances. Furthermore, they
are easy to handle, because all terms, products, and derivatives can be reexpressed in terms of Chebyshev polynomials.
We assume a solution of the form
(18)
(13)
which is a special case of the ansatz Eq. (7). Here we denote
the Chebyshev polynomial of order j by Tj, and refer for
their properties to Ref. [18] and Appendix B. As usual in
Chebyshev series, the primes indicate that the first term in
each sum is multiplied by the factor ).
To reduce the number of unknowns we take advantage of
the symmetry of Eqs. (1 1) and (12) in y . following from Eq.
(3). For any approximation l v N of the Wigner function that
has a continuous derivative this symmetry is equivalent to
von Neumann boundary conditions
(14)
on the x axis. When we substitute ansatz (13) into condition
(14) we find-xpressing
with the help of Eq. (B2) the derivative of a Chebyshev series in another Chebyshev seriesthe constraint
We now insert our spectral ansatz Eq. (13) for V into the
two coupled differential equations (11) and (12) for the
Wigner function and evaluate the two residuals of these
equations. For the sake of simplicity we here only present the
main ideas and results and refer to Appendix B for the calculation.
Since we have represented the Wigner function T by a
Chebyshev series, it is convenient also to express the potential
(19)
as a Chebyshev series. For polynomial potentials this representation is exact, while for other smooth potentials the approximation can be made as accurate as desired. This we
achieve by an appropriate choice of the upper limit N u .
When, according to Appendix B, we express all the terms in
Eqs. (11) and (12) by Chebyshev polynomials, for the residual of Eq. (1 1) we obtain the expression
(20)
whereas the residual of Eq. (12) reads
519
3192
57
-
(21)
\
2kSNy
The coefficients of the series follow from Eq. (B3), and are
given by
N*
a;,2k=
X 2 ( j + 1 +2 s ) a , +
2kSNY-2
ISN,-2
1+ 2 s . 2 k .
S=O
NY
++,=2
2(2s+2)aJ&+2
.1
=k
(22)
and
this yields the conditions
NY
r=k
(23)
NY
4(2t+1)(2~+2)a,,&+~.
S=I
contribute. To satisfy Eq. (23) the prefactors of these integrals, that is, the prefactors belonging to the product
T , ( x ) T , ( y ) , must vanish. This condition leads to the homogeneous system of equations
(24)
kSN,
520
57
3193
if
(25)
_L
fi,Qn
1/JL
V
"=.?
and
2/u + 1 = k - 1 - 2n
n iN
and
if
and
2 W {0.2
V,}
and
if
- 2 - 2n
(26)
and
and
2n{0,2..,;Vj,-2}
Here the terms Vf and Vf* follow from Eq. (B4). We note
that the coefficients Tffi*1 (odd in /z) result from the residual /?i whereas the coefficients rjjf'* (even in /i) originate
from /? 2 .
The linear system (24) obviously has the trivial solution.
To find nontrivial solutions, we need at least one additional
equation. We do this by setting the value of the solution at an
arbitrary point in the computational domain, such as the origin, to a nonzero value, for example unity. We then obtain
from Eq. (13), recalling the properties r 2 y+i(0) = 0 and
72/0) = (- 1V, the additional equation
(27)
Since the original equations (11) and (12) are linear, we can
arbitrarily choose the value of the Wigner function and the
solution can be normalized after the computation of the unknown coefficients.
We approximate the solution of the overdetermined linear
system (24) by a standard least-squares algorithm [19]. This
introduces an additional error to our solution since then the
52 1
3194
51
-
residuals are not exactly zero, but are minimized in a leastsquares sense using the L2 norm. In general, the norm of the
residuals decreases with increasing N , and N , , and therefore
can be used as a measure to check the accuracy of the approximation. In Sec. VII we present numerical calculations
for our model problem. As the results are of high accuracy
for appropriate N , and N , , the least-squares approach can be
justified.
VI. CONVERGENCE OF OUR METHOD
In this section we analyze the convergence of our integration scheme for the set of differential equations (11) and
(12). Throughout the section we make heavy use of results of
Canuto et al. [16]. We discuss error estimates, stability, and
consistency. For an introduction into the elements of functional analysis and definitions such as stability, consistency,
and well-posed problems, we refer to Appendix D. In the
present section we just discuss the results qualitatively.
The convergence properties and the errors of a Chebyshev
series approximating a given function have been studied extensively [ 161. We make use of these results in our approach
to control the error of the approximation to the potential
V ( x ) by a Chebyshev series. However, these properties are
not applicable to the approximation of the solution of Eqs.
(11) and (12), since here the approximated function is not
given explicitly but implicitly as a solution of differential
equations. In order to apply the convergence theorems of
Chebyshev series we have to ensure that the approximation
converges toward the solution of the differential equation.
Hence we have to show that the differential equation has a
unique solution. Furthermore we have to investigate whether
our approximation converges and represents this solution in
the limit N + m . Therefore the question of convergence of a
numerical approximation to the exact solution is central to
our scheme.
According to the Lax-Richtmyer theorem [20], stability
and consistency imply the convergence of a well-posed problem. The most straightforward technique for establishing the
stability as well as the consistency of the spectral schemes is
the so-called energy method [ 161. To apply this technique in
our specific problem, we have to prove the inequalities
and the H i norm additionally contains all the first derivatives, that is,
(30)
for each ( c r H;(R).
~
For the second-order equations we
have m = 2 , and hence the error is proportional to 1/N. The
second term on the right-hand side contains all derivatives up
to second order. This has the consequence that other parameters such as the size of the integration domain D can also
influence the error estimate. We will see this in Sec. VII
when we investigate the harmonic oscillator.
To prove the stability of our scheme, we simply have to
follow these instructions with one modification. Usually one
has to solve only a single equation with boundary conditions.
In the context of Wigner function, however, we have to actually solve two equations simultaneously. The first-order
equation is hyperbolic and the second-order equation is elliptic provided d 2 V / d x 2 > 0 . It is easy to see that a hyperbolic differential operator does not fulfill inequality (28).
Nevertheless we can apply the energy method to the system
of equations (1 1) and (12). To see that, we add Eq. (1 1) to
Eq. (12). and arrive at the equivalent system
(28)
LzV= -+V(x)-E
and
(29)
where C , and Cz denote positive constants. These inequalities have to be valid for all functions ( I . E D ( L ) and $
E H i , where D ( L ) is the domain of the differential operator
L . For a definition of the Sobolev spaces H i and H i , we
refer to Appendix D. Here it will be sufficient to note that the
is simply the scalar product
norm 11
and
+ V (x) - E -
Po
Qo
-y
1 dV(x) d
+PoQo
dx
JY
(32)
(32)
Indeed, both equations are now elliptic for d 2 V ( x ) l d x 2 > 0 .
522
MODIFIED SPECTRAL METHOD I N . . . . I. . . .
57
Now we can start to prove stability, because both equations lead to the same coercivity evaluation as we show in
Appendix E. In addition we also prove that inequality (28)
holds under certain conditions for an appropriate function
q ( x , y ) which is even in y . According to Appendix E, for
positive d 2 V / d x Zwe obtain the estimate
i
dzvj IIW
+/: 1
3195
o.2
32Qo2 min
(Lzvv,w,~
( T + V ( x ) - E Pzx d x d y
-1
+ B,
(33)
-0.2
a,.
-/
/a2[[E-V(x)]-
(34)
ligibly small. In this case inequality (33) is valid. This argumentation holds if the domain CL is chosen in such a way that
most of it lies outside the classical region. Hence the choice
of 0,i.e., P o , Q , , and Q 2 , influences the stability. For other
choices of the computational domain we have to require that
the right-hand side of Eq. (33) does not become negative.
Our proof of stability might fail, if d 2 V / d x Z becomes
negative. Since we have only binding potentials, there is always a phase-space region, where the curvature is convex
and therefore d 2 V / d x 2 is positive. In this case we can decompose all integrals containing d 2 V / d x 2 into positive and
negative parts in exactly the same way as in Eq. (34), and we
TABLE I.Mean and maximal absolute errors for the Wigner function approximation of the eighth energy
eigenstate of the harmonic oscillator depending on the number N,,= N,= N , of Chebyshev polynomials in
our ansatz (13). We applied our modified spectral method on the computational phase space domains,
D I = [ - 4.5,4.5] X [ - 4.5,4.5], D, = [ - 4.0,4.0] X [ - 4.0,4.0], and D, = [ - 3.5,3.5] X [ - 3.5,3.5]. I n the left
part of the table we compare the mean absolute errors with increasing N,, and on the right part the maximal
absolute enors which are located in the edges as shown in Table 11. Note that to find an error of the same
order of magnitude we have to increase the number of Chebyshev polynomials N,, with increasing size of the
computational doman.
Maximal absolute error
Dl
0 2
1.4X100
4.3X10-
l . l ~ l O + ~ 3.3x10-
4.1X10-
6.7X10-
4.1X10-*
1.7X10-10
3.7X1O-l4
9.3X10-2
D3
1.6X10-2
2.6~10-~
6.0X10-9
4.1X10-3
2.2X10-15
Dl
6.6X1OfZ
5.4~10~
2.OX1OZo
l.0X10-5
1.7X10-9
D2
9.8X10
6.4x10-1
9.2X10-6
1.4X10-9
6.6X10-
D3
3.2X10-1
1.7~10-~
4.6X10-
3.0X10-12
1.3X1O-l4
523
M. HUG, C. MENKE, AND W. P. SCHLEICH
3196
of
TABLE 11. Dependence of the absolute errors (times
the Wigner function approximation Eq. (13) of the eighth energy
eigenstate of the harmonic oscillator on the location ( Q , P )in the
phase-space domain D ,with N,,=50. We note the dramatic error
located in the comer (Q,P)=(4.5,4.S).
30.4
3.7
1.7
1.9
0.9
0.0
0.9
1.8
2.7
3.6
4.5
57
1.5
10.8
0.0
3.7
12.9
4.0
0.3
8.3
5.9
9.8
3.7
19.7
0.0
0.9
1.8
2.6
10.3
0.2
19.7
4.0
4.0
0.2
0.2
9.1
0.1
1.7
8.3
10.8
10.3
1675.0
9.1
0.2
0.3
1.5
2.6
2.7
3.6
4.5
1.7
10.3
5.9
4.0
Q
have to require that the negative parts do not dominate.
Again this is influenced by the choice of Q, and Q2. Thus
one has to be very careful in choosing a proper domain. To
obtain stability with the energy method, one has to choose a
domain where the positive parts of d 2 V / d x 2 outweigh the
negative parts, and where the additional negative terms do
not dominate.
To complete our proof of stability, we have to show the
inequality (29). For each term of (L2'P,$), this can be done
in a straightforward way by using the triangle inequality,
partial integrations, the Cauchy-Schwartz inequality, and
Hardy's inequality. With that estimation, stability and consistency, and hereby convergence of our method, are
achieved, if the coordinate range is chosen carefully. The
error estimate (30) holds.
We conclude this section by noting that higher-order
terms lead to more complicated investigations because more
cases occur. Nevertheless the convergence and stability results remain valid under conditions analogous to Eq. (33).
(35)
where L, is a Laguerre polynomial and
R 2 = Q2+ P2.
In Fig. 1 we show the Wigner function of the eighth eigenstate together with the contour lines of q ( Q , P ) , According
to Eq. (35) the latter are circles centered at the origin of
phase space.
Multiplying our numerical solution based on the method
of Sec. V with a normalization factor enables us to compare
it with the exact solution and to calculate the absolute errors.
524
57
3197
(A2)
VIII. CONCLUSION
We have presented an alternative approach for the calculation of the Wigner function. Our method avoids the solution of the Schrodinger equation for the wave function. Instead, we apply a result from Fairlie and Kundt [2] and
others [3] and solve two coupled real linear partial differential equations. Our numerical method is a modification of the
spectral tau method, which uses Chebyshev polynomials as
shape functions. In contrast to usual applications, no boundary conditions are available for the equations. Since two differential equations have to be solved simultaneously, the resulting linear equation system is overdetermined. Its solution
is approximated by a least-squares method. We could prove
stability and convergence of our scheme by making use of
results of Ref. [ 161. Generalizations of our approach to any
similar system of partial differential equations are straightforward.
Our numerical results for the harmonic oscillator are in
excellent agreement with the known analytic solution. The
mean absolute error is of order
for an appropriate
choice of the domain and the approximation order N,,= 70.
Our modified spectral method is equally applicable to any
potential V, and can be extended to the higher-order differential equations. This will be the subject of the following
paper. Our results encourage further research on potentials
for which no analytic solution can be calculated.
a.
fiI;=Ei
for the density operator which in this case equals the projeccorresponding to the energy
tion operator p= I &)(
eigenstate I + F)
with energy eigenvalue ?
!. and the Hamiltonian H = p2/2M+U(q). When we perform the WeylWigner correspondence Eq. (A2) of both sides of this operator equation, with Eq. (Al) we find the relation
$4
(Hpl(q,p)=W'(q,p).
ACKNOWLEDGMENTS
We thank R. Baltin, M. V. Berry, J. P. Dahl, U. Leonhardt, H. Leschke, S . Schneider, and R. Seydel for many
fruitful discussions. This work was partially supported by the
Deutsche Forschungsgemeinschaft. M. H. is most grateful to
the Studienstiftung des deutschen Volkes for its continuous
support.
APPENDIX A: A DERIVATION OF THE PHASE-SPACE
EQUATIONS
(A1)
of the density operator is that it yields the correct quantummechanical marginal probability distributions of position and
momentum. Moreover, it enables us to evaluate the expectation value
or
In this appendix we first briefly review some useful properties of Chebyshev polynomials, and then apply them to
express all terms of the coupled differential equations (11)
and (12) in terms of Chebyshev polynomials. Since the operators L , and L , , are linear, this can be done term by term.
Here we demonstrate this procedure for one term only, and
then outline how to proceed to evaluate the residuals. The
525
51
-
3198
c.
N" N"
Vy=4x
( l + l + 2 t ) ( l + 2 + 2 ~ ) V f + 2 + 2 ~ .(84)
1=0 s = t
kSNy
(B1)
where we defined T - , ( x ) = T S ( x ) . Products of the form
n'T,(x) are given by
kGNy-2
2(2r+
I,Zk
1)a:,2~+I
with
NY
ay,21 + 1 = I = I
(B2)
that is a Chebyshev series of degree n - 1 with coefficients
2(2s +
)'J.&+2
a:=x 2 ( i + 1 + 2 s ) ~ ~ + ~ + ~ ,
(B3)
S=O
for i=O,l, ... and with aj=O for i>n. Hence the differentiated Chebyshev series is a sum containing the coefficients a j
of the original Chebyshev series. Note that, for higher derivatives, Eq. (B3) is a recurrence relation that connects the
coefficients of the n-times differentiated Chebyshev series
with the coefficients of the ( n - 1)-times differentiated series. Finally these coefficients of the n-times differentiated
series consist of n-fold sums containing the coefficients ai of
the original series.
with
NY
a;,21+2=s = 1 + l 2 ( 2 s + l ) a l . 2 r + l = ~ ;
that is,
a%+ 1 = O
2. Evaluation of residuals
g*=x '
a2
c.
where
N"
NY
- 4 t x= k
5=1
(2r+1)(2s+2)~,,,+~.
(B6)
N"
v;I= t = O
(B5)
where
'r
From Eq. (B2), we find
ar,!2kT~(x)T2k(y),
JGN,
2kSNy-2
2( I
+ 1 + 2t) v;+ + Z f ,
' $ V;'ar~k[T~+l(x)+T~-f(x)lT2k(y)
I G N ~
2kGNy-2
fSN,-2
and hence
h
51
differential equations (1 1 ) and (12) we can proceed in exactly the same way, and find the residuals R , and R , presented in Eqs. (20) and (21).
APPENDIX C: DERIVATION OF THE LINEAR SYSTEM
3199
sJ'y=
1 j=v
otherwise,
(c1)
which we use in two dimensions as a product of two Kronecker deltas. With this notation, for the residual R , , Eq.
(20), we find
(c2)
(c3)
Here the constant c 2 = l / d when both coefficients v and p are positive, c,=2/~r' when one of them is zero, and c2=4/7rZ
when both are zero.
Equations ('2%)and (C3) represent system (10) of equations, r a = 0. Note, however, that it contains the unknown coeffiexplicitly, we have to express the coefficients of the
cients a,,k only implicitly. To obtain the matrix coefficients
To do so we have to change the order of summation
etc., in terms of the unknowns
differentiated Chebyshev series
in the double sums in Eq. (B6), and the equivalent equation for J2q/dx2. We start from
ry;!
and
Here J= Ij-ll or J=j + l , and we have used Eq. (B3) twice. For convenience, we have shifted the upper limit of the sum for
a?u.zp by k . This does not change the sum since u , , ~ = O for k > N , . When we interchange the order of summation we obtain
and
527
M. HUG, C. MENKE, AND W. P. SCHLEICH
3200
Ny+k
s=k
l=k
4(2t+1)(2s+2)~,~+2.
2 ( j + 1+ 2 t ) = ( s + I ) ( j + l ) + S ( S +
I)=($+ I)(j+s+I )
f=O
and
Thus the equations of the weighted residuals [Eqs. (C2) and (C3)] take on the forms
and
When we shift some indices of the first equation by T=j+ 1 + 2 s and E=2k+2+2s, it finally has the form
(c4)
Similarly we shift some indices of the second equation by T = j + 2 + 2 s and E = 2 k + 2 + 2 s . With u=T-2-2s
2 p = R-2-2s, we can furthermore collect the prefactors
and
4( s
This finally yields
and
528
51
3201
N,
(C5)
From Eqs. (C4) and (C5), we immediately can read off the
matrix elements ry;l of the system of equations for a j , k .We
finally arrive at Eqs. (25) and (26).
APPENDIX D: ELEMENTS OF FUNCTIONAL ANALYSIS
In this appendix we briefly review fundamental definitions of the numerical functional analysis and the energy
method. This summary provides the theoretical background
for the convergence analysis of our spectral scheme (Sec.
VI) .
11 Q N f
-fNll<
E(N)
for every N . Here the norm ((I(is the usual Hilbert space
norm. These two conditions are a mathematical formulation
of the fact that the result must be independent of the path
taken in approximation scheme (D3).
We can define stability by requiring that there exist two
positive numbers r and s such that the inequality
F EX ,
1. Definitions
According to the Lax-Richtmyer [ Z O ] theorem the convergence of a numerical solution to the exact solution of a wellposed problem is implied by stability and consistency. What
do well posed, stability, and consistency mean? To motivate
the expressions used here, we return to Eq. (6),
L*=f,
(D1)
(D2)
for the approximate solution J(, with the variable approximation coefficient N = 1.2 ... . Here the functions *,f , $,
and f , are elements of the Hilbert spaces X, Y , X,, and Y,,
respectively. We investigate the approximation scheme
The space
fined by
x+
(D3)
L
xN-'
yN
and
In this subsection we summarize results that were presented in full detail in Ref. [16]. The most straightforward
technique for establishing the stability of spectral schemesthe so-called energy method-is based on choosing the solution itself as the weight function discussed in Sec. V. This
approach is successful if the spaces of the trial and test functions coincide, and if the spectral operator is positive with
respect to a suitable inner product.
529
57
-
3202
4 (I,Il2=(L(I,>(I,)?
Applying this result to the Chebyshev polynomials of degree N guarantees the stability and consistency of our spectral scheme in the sense of the energy method, if N is sufficiently large. Under this hypothesis, convergence is a
consequence of the Lax-Richtmyer [20] theorem. Then the
corresponding error estimate reads
(D6)
a).
(D4)
(D5)
FE
H;(a)
(E1)
This theorem is a consequence of the Lax-Milgram theorem.
Finally one can prove that the weak solution is indeed a
strong solution, that is, it satisfies L(I,=f.
dT
1 d V d 9
Po d x dy
(E2)
is zero because the integrand is an odd function in y . Therefore the relation ( L + T , ' P ) , = ( L 2 ~ , Y ) , holds for the operators
L , and L 2 in Eqs. (32) and (31). Consequently, it is sufficient to investigate L2 only.
When we integrate the last integral of the scalar product
(E3)
with
530
51
3203
Vr need only to be in H ; ( n )
d2X
= ( 1 2 2 ) ( 1 -y2)2x5< 3x5
dx
and
2
dY
= ( 1 2 y 2 ) ( 1 -x2)2,y5< 3 x 5 ,
following from the definition Eq. (22) of the Chebyshev weight, we obtain
(E4)
which holds true whenever 'P E H!JCl) is a symmetrical function in y and d 2 V / d x 2 is positive. This relation yields
(E5)
with
The first term on the right-hand side we can easily estimate with the help of the Poincark inequality [16] ll'PllL2<CllV''ll(L;)2
which holds true for all
V E H!JfL),
53 1
3204
51
-
vanishes on some one-dimensional curve on the domain Q. The Wigner function T satisfies this assumption since it vanishes
for sufficiently large arguments x and y . Hence for positive d Z V / d x 2we find from Eq. (E5) the final estimate Eq. (33).
(E6)
with B2 from Eq. (ES). In Sec. VI we use this result with B=B2/4 to discuss the stability of our scheme with the energy
method. Equation (28) and hence Eq. (E6) requires that the sum of the last three terms have to be greater than a constant times
11~11.
APPEM)IX F A USEFUL INEQUALITY
which is crucial for the estimate Eq. (E6). This inequality is valid for a function 4'' E H : ( Q ) symmetric in y and d2V/dx2>0.
For a similar inequality, however, restricted to one dimension and without boundary terms, we refer to Ref. [16]
We start the derivation by performing the differentiation on the left-hand side of the inequality, and find
When we then perform an integration by parts in the same way as done in Eq. (E4). we arrive at
(F1)
With the help of the identities
and
following from the definition Eq. (22) of the Chebyshev weight, we obtain inequality (E4) when we omit the first integral of
Eq. (Fl) which is positive and use that ( 1 - x 2 ) 2 and ( 1 -y2)' are bounded from above by 1.
532
57
-
. . . I.
3205
533
PHYSICAL REVIEW A
MAY 1998
Modified spectral method in phase space: Calculation of the Wigner function. 11. Generalizations
M. Hug,' C. Menke? and W. P. Schleich'
'Abteilung f i r Quantenphysik, llniversitat Ulm, 0-89069 Ulm, Germany
'Abteilung Numerik, Universitut Ulm, 0-89069 Urn, Germany
In paper I [l] we laid the foundations of a method calculating the Wigner function of an energy eigenstate directly
from phase space. We introduced a modified spectral method
for the solution of the two coupled partial differential equations defining the corresponding Wigner function in phase
space. We approximated the solution as a finite sum of
Chebyshev polynomials in the two phase-space variables position and momentum. To keep the analysis of our method
simple, we restricted the discussion in paper I to secondorder derivatives. In the present paper we focus on the generalization of our modified spectral method to differential
quotients of arbitrary high order in the two partial differential
equations for the Wigner function.
By the extension of our method to the full system, we are
able to calculate the Wigner function of any energy eigenstate in any smooth potential. High-order differential terms
which occur in differential equations describing a physical
system are often neglected because it seems to be too complicated to handle them. Our method turns out to be a very
powerful tool to solve such equations, since high-order derivatives do not increase the number of unknown coefficients
and therefore the numerical costs. We derive a sum factorization formula which shows that the nth-order derivative
only contributes by a simple product of n terms to the system
of equations for the unknown Chebyshev coefficients, and
not by an n-fold sum as one might expect. This formula is
universal since it remains valid whenever any spectral
method which uses Chebyshev polynomials as shape functions is applied to any high-order differential equation. Generalizations of our method to any other high-order differential equations are therefore straightforward.
This paper is organized as follows: After a short summary
of the fundamentals in Sec. 11, in Sec. I11 we analyze the
higher-order terms which we neglected in paper I. This
analysis makes use of the sum factorization formula mentioned in the preceding paragraph. We give the proofs in the
Appendixes. In Sec. IV we apply our method to evaluate the
Wigner function of the Morse oscillator, which requires including higher order terms to obtain satisfying results for the
Wigner function, and to an asymmetric double-well potential
where neither an analytic expression for the Wigner function
1050-2947/98/57(5)/3206(19)/$15.00
57
-
In this section we briefly summarize the fundamentals introduced in paper I [l] that are actually necessary to understand the analysis of this paper. In particular we define the
phase-space equations for the Wigner function, and sketch
our spectral ansatz and the central ideas of our method.
Note that we still concentrate on energy eigenstates of
one-dimensional potentials whose phase-space description
leads to a two-dimensional phase space spanned by dimensionless position Q and momentum P . We compute the
Wigner function on a square domain in phase space
[ Q l . Q 2 ] X [ - P o , P o ] , with Q l < Q z and Po>O. For our
method it is convenient to map this domain onto the square
O=[-l.l]X[-l,l].
and to formulate our theory in the
mapped position variable x E [ - 1,1] and the momentum
variable y E [ - 1.11. Once we have found the Wigner function in these variables it is straightforward to invert this linear mapping.
In phase-space language the Wigner function T ( x , y ) of
an energy eigenstate with energy E in a one-dimensional
binding potential V ( x ) is defined as the unique solution of
the coupled partial differential equations
(1)
where
and
= 0,
(2)
with
3206
534
57
(3)
where T , denotes the Chebyshev polynomial of order j . The
end of the method is to find the unknown coefficients aj,k in
such a way that the approximated solution of our ansatz fits
the exact solution of Eqs. (1) and (2) best. The logical train
of thought of the modified spectral method is to insert this
ansatz into Eqs. (1) and (2). We call the deviation of these
expressions from the exact solution the residuals
R , = Ldd'PN and R 2 = Leven'P~.We obtain the equation system for the unknown coefficients a j , kby requiring that the
integrals
(4)
of the weighted residuals over 0 vanish. Here x denotes the
Chebyshev weight, which allows us to use the orthogonality
relations of the Chebyshev polynomials to derive a linear
system of equations
ra=O
(5)
r'&
,,
3207
m-l
(6)
and for the even ones,
"-1
(7)
Due to the symmetry relation "(x, - y ) = ' P ( x , y ) following
=0, and confrom Eqs. (1) and (2), the coefficients aj.2K+1
are zero.
sequently a;,$+, and
535
3208
51
-
--
C'
(8)
For the evaluation of the higher-order terms of Eqs. (1) and (2). we have to calculate the product
(9)
/GN,-m
where V;" is determined by Eq. (8), and the coefficients u r k by Eqs. (6) and (7). Here we used
(10)
m )Eqs. ( 1 ) or (2) contribute to the residual
where s and r are integers. Therefore the terms [ d m V ( x ) / d x m ] ( a m V / a y of
(11)
where Eq. (9) generates the summands
/ G N U- ( 2 m- I )
(12)
2kSNy-2m
/ G N U- Zm
536
51
.. .
. 11.
...
3209
(0,
jfk
(13)
(15)
and
(16)
jsNx
kSNy
r y : /a j, k =o ,
ry;/ with
(17)
537
M. HUG, C. MENKE, AND W. P. SCHLEICH
if
57
2p
= j - 1 - 2n
+ 1 = [lc - 11
n~lN
U { O , l , ...,N ,
- 1)
'Zii+l{l
.....N,}
(18)
f Vi
if
and
2p = k
ZPE{O.Z
.....IVY)
(19)
if
.C' ( - l ) j + k a 2 j , 2 k .
F N x
ksN,
Since the original equations (1) and (2) are linear, we can
arbitrarily choose the value of the Wigner function at the
origin and the solution can be normalized after the computation of the unknown coefficients.
We note that the coefficients rr;:p+' (odd in p) result
from the residual R , , whereas the coefficients r?? (even in
p) originate from R 2 . Thus the original two differential
equations still occur separately in the resulting system of
equations (17). We further note that this equation system
contains all orders of the differential equations (1) and (2).
538
57
321 1
V(Q)=~h(l-e-Qa)z,
(20)
IV.APPLICATIONS
In this section we apply our algorithm to two potentials: a
Morse oscillator and an asymmetric double-well potential
that is given by a fourth order polynomial. We use these two
potentials to investigate different aspects of our method. To
obtain satisfactory results for the Morse potential we have to
include higher-order terms of Eqs. (1) and (2). In order to
obtain a reference solution, we use an analytical expression
for the wave functions of the energy eigenstates, and numerically perform the Fourier transform defining the Wigner
function using this expression. This allows us to obtain the
Wigner function with high accuracy. Thus we can estimate
the errors of our method together with those errors that are
due to the truncation of higher-order terms in Eqs. (1) and
(2). For the asymmetric double-well potential no analytic
solutions are known, and even semiclassical approximations
fail for low excited states. Nevertheless these states are most
interesting since they are most affected by the barrier of the
potential. By qualitative arguments we verify that our
method leads to a very good approximation even in cases
where other approximations are not valid.
A. Morse oscillator
0.2
0
-0.2
539
57
-
3212
N,= SO
m
Dl
2
3
4
7.4x 10-2
7.0X
2.8X
3.1X lo-*
1.ox 10-2
3 . 4 10-3
~
1 . 7 10-3
~
1 . 3 10-3
~
5
6
7
8
9
10
D2
8.9X
2.9X lo-,
2.7X lo-*
3.8X
8 . 9 10-3
~
2.9X 10-3
6.8X
3 . 9 I~ O - ~
1 . 1 ~
2 . 5 10-4
~
(2X-2n-l)r.(n+l)
"l)=
&r.(2A-n)
L r ) denotes a generalized Laguerre polynomial. The energy
eigenvalues read
TABLE 11. Mean absolute errors for a series of Wigner function
approximations of the ground state and the first excited energy
eigenstate of the Morse oscillator with constant number of Chebyshev polynomials N,= N,= N,= SO calculated on the computational domains D , and D4 . The series index m on the left denotes
the truncation order of the Eqs. (I) and (2). For the ground state and
the first excited state we chose the same computational domain D3
to compare the errors.
Mean absolute error
SO
N,=
m
Ground state
0 3
2
3
s.sx
4
5
LOX
6
7
8
9
10
3.3X lo-'
10-3
2 . 6 1~0 - 4
1 . 7 10-4
~
2 . 1 ~
2 . 6 10-5
~
2 . 6 10-4
~
2.1x10-~
4.5I
~O - ~
4.0X
3 . 3 10-2
~
l.lXIO-l
1.4~
4.9xi0-~
1.1x10-~
3.9~
5.ix10-~
D4
4.5~
3 . S X lo-'
9 . 7 10-3
~
2.8X10-3
9.4~
2 . 5 lo-'
~
7.9xi0-~
2 . 3 1~ 0 - ~
S.OXIO-~
TABLE 111. Mean absolute errors for the Wigner function approximation of the second excited energy eigenstate of the Morse
oscillator with constant truncation order m = 10 for various numbers
of Chebyshev polynomials N,,= N,= N , .
Mean absolute error
Nea
Dl
D2
20
30
2.1x 10-2
3.1x 10-3
2.6X 10-4
2.sx
9.4x 10-2
40
so
En=
(,+
;)& (,+
-
1.ox
S.6X
I . I X 10-4
y,
with n=O,l, ...,A - 1. In our specific case we choose the dimensionless parameter A = 4.This parameter designates the
relation between the height and width of the potential, and
therefore the number of bounded states. This potential and
all its squared eigenfunctions are sketched on the left walls
in Fig. 1, where we also show the corresponding exact
Wigner function.
We approximated the solutions for the Wigner function of
the second excited state shown in Fig. 1 with our modified
spectral method on the two computational domains
D , = [ -2.O=sQe8.0] X [ - 2 . 5 C P S 2 . 5 ] and D 2 = [ -4.0
CQC8.0]X[-3.0CP<3.0].
In both cases we used
To compare the results with the exact
Nq=N,=N,=50.
Wigner function, we reproduce in Fig. 2 only the approximations on the domain D , , and cut off the outer parts from
the computations on the larger domain D,. On the top we
show the exact solution, denoted by m = m. Below we reproduce a sequence of calculations on both domains in decreasing (from top to bottom) order rrt of the partial derivatives
incorporated from Eqs. (1) and (2). We compare the results
for the two domains D , on the left, and D , on the right. In
the case m= 10 we have taken into account all terms of the
original system up to the tenth order. Nevertheless, on the
left-hand side one can see some tiny errors near the boundary. On the right-hand side these errors are outside the region
of interest. Because the larger errors are located in the corners of the computational domain, we obtain better results
when we calculate the solution on a larger domain and cut
off the outer parts. For m = 8 the result on the larger domain
is still excellent. The left-hand side also shows qualitatively
good results. It seems to be sufficient and necessary to include the eighth-order derivatives to obtain a qualitatively
correct impression of the Wigner function. The inclusion of
the tenth-order terms does not yield a significant improvement. For the two discussed regions the visible errors mainly
correspond to numerical errors. When we truncate more
terms of Eqs. (1) and (2), the truncation error becomes dominant, as we see in the next example below where m = 5. Here
we also observe that the solution on the two domains become
different. The calculation on the domain D , shows larger
deviations from the exact Wigner function than the approximation on D,. We also observe this effect for the last two
pictures at the bottom for m = 2, where we have truncated
Eqs. (1) and (2) after the second order, which is the approxi-
h
57
...
3213
FIG. 2. Table of contour plots showing approximations for the Wigner function of the second excited energy eigenstate of the Morse
oscillator shown in Fig. 1 with a constant number of Chebyshev polynomials N,= N,= N , = 50. The thicker contour lines denote the zero
level, the thin solid contours denote positive values in steps of 0.03, and the dashed contours the negative values, respectively. We calculated
each approximation on two domains D ,(left) and D 2 (right). On the top, the exact solution for the Wigner function ( m= m ) is shown twice,
where the left figure clarifies the scaling. Below, a sequence of calculations on both domains is shown, where from top to bottom the
truncation order m = 10, 8, 5, and 2 of the original equations (1) and (2) is decreased. Although the computations on the right-hand side were
carried out in the larger domain D,, here we show only the approximation on the smaller domain D , to compare the solutions. Note that the
truncation after the second order-the approximation we used in paper I to introduce our method-is not suitable for the Morse oscillator.
At least all terms up to the eighth order have to be included to obtain satisfactory results.
54 1
3214
51
-
So far we can qualitatively estimate the error of neglecting higher-order terms in the original equations (1) and (2).
We do not need very many terms to obtain all the essential
information about the Wigner function. A truncation order of
m=8 is sufficient for excellent agreement with the exact
solution. This leads to the conclusion that we can extract all
relevant physical information from the low-order terms of
Eqs. (1) and (2), at least in the case of the Morse oscillator.
When we keep m = 10 fixed, we can analyze the convergence properties for increasing N,,. We illustrate this for the
second excited state for the domains D ,and D , in Table 111.
As expected, the absolute errors decrease with increasing
N,,, the number of Chebyshev polynomials in our ansatz.
Finally we show the solution for the third (the highest)
excited state in the Morse potential with h = 4 . Here the
Wigner function spreads out over a large area of phase space,
since the energy is close to the dissociation energy. However, we obtain qualitatively excellent agreement even with
N,,=50 as we can see in Fig. 3.
B. Asymmetric double well
Our second example is an asymmetric double-well potential
(21)
It is reproduced in Fig. 4 for the parameter values k,=O,
k,=3.5, k3=0.25, and k,=0.5. We have chosen this set of
parameters to make our data comparable with former applications (Ref. [5] and references therein) where it was frequently used as a test problem for a numerical calculation of
the position eigenfunctions h ( Q ) . We note that Ref. [ 5 ]
also offers a valuable scheme to generate the energy spectrum of an arbitrary smooth potential.
Since we can easily apply our method to the onedimensional time-independent Schrodinger equation
542
MODIFIED SPECTRAL METHOD IN
P
...
. II.
...
3215
PN
02
-0 2
CQ S
16.01 X [-3.O<PC
3.01,
since they are confined to the left and right wells, respectively. Therefore the Wigner functions of these states are
quite similar to the Wigner function of the Morse oscillator
ground state. This is different for the second and third exE
cited states, where the energy level is still below the barrier
but the wave function is spread out over both wells. This can
also be seen in the Wigner function especially for the third
excited state. Above the barrier (fourth and fifth excited
states) the wave function becomes more and more similar to
an oscillator wave function. The Wigner function also exhibits such a behavior. The oscillations for large values of the
momentum variable P give a hint of the existence of an
inflexion [6] of the classical trajectory which is a consequence of the barrier.
In Fig. 6, for the third excited state, we directly compare
the numerically integrated marginal distribution of our approximated Wigner function with the position probability
distribution of Fig. 4 which we also calculated numerically.
Since we obtained the wave functions from a onedimensional problem where the highest derivative is of order
2 it is obvious that the latter error is much smaller than the
error arising from applying our method to the twodimensional fourth-order coupled differential equations and
then integrating the approximated Wigner function numerically. Nevertheless when we compare the two resulting
curves we obtain an excellent agreement in Fig. 6. This also
holds true for all other eigenstates. This is not a proof but a
strong indication that our method yields an excellent approximation of the Wigner function for the asymmetric
double-well potential. This is a remarkable result since most
former approximations of the Wigner function are semiclassical expressions and therefore only valid for higher excited
states. Here we obtain very good approximations even for
the lower excited states.
V. CONCLUSION
W e have presented an alternative approach for the calculation of the Wigner function. For that purpose we solved
two coupled linear partial phase-space equations which define the Wigner function directly. These equations are of
infinite high order in the derivatives, and finite only for polynomial potentials. For the first time one 'can compute the
Wigner function for arbitrary potentials as a solution of these
equations. Our numerical method is a modification of the
spectral tau method which uses Chebyshev polynomials as
543
M. HUG, C. MENKE, AND W. P. SCHLEICH
3216
57
02
-0.2
FIG. 5. Approximations for the Wigner functions of the lowest-energy eigenstates of the double-well potential proposed by Ref. [ 5 ] . The
two figures on the top shows the fifth and fourth excited states the middle the third and second excited states, and the bottom the first excited
state and the ground state, respectively. The potential (thin solid line) with the energy levels (dashed line) of the energy eigenstates and the
approximated position probability distribution (thick solid line), that is, the square of the approximated position wave function of each
particular state (see Fig. 4), are shown projected on the left-hand wall for each energy eigenstate. To keep the scaling consistent we
compressed the potential by a factor of 32 and the position probability distribution by 3.5. The meaning of the contour lines below the
three-dimensional representation of the Wigner function is the same as in Fig. 1.
544
57
3217
In this appendix we formulate the sum factorization formula which we apply to the m-fold sums arising from a
recurrence relation to arrive at expressions (6) and (7). The
coefficients of the mth derivative of a Chebyshev series can
be calculated from the coefficients of the original Chebyshev
series by a recurrence relation which reads, for coefficients
with even indices 2 k ,
FIG. 6. Comparison of the marginal distribution of the Wigner
function (dotted curve) of the third excited state of the double-well
potential, obtained by numerical integration of the approximated
Wigner function, with the probability distribution (solid curve) arising from numerical integration of the time-independent Schrodinger
equation.
ACKNOWLEDGMENTS
We thank R. Baltin, M. V. Berry, J. P. Dahl, U. Leonhardt, H. Leschke, S. Schneider, and R. Seydel for many
fruitful discussions. This work was partially supported by the
Deutsche Forschungsgemeinschaft. M. H. is most grateful to
(A1)
and for odd indices,
(A2)
This recurrence relation expresses the coefficients of the
m-times differentiated series by the coefficients of the
( m - I)-times differentiated series. By induction follows that
the coefficients of the m-times differentiated series are
m-fold sums containing the coefficients a,,k of the original
series. Note that in the recurrence relation the even subscript
2k changes to odd subscripts 21+ 1, and the odd subscript
2k+ 1, to even subscripts 2t+2, respectively. Remember
also that due to the symmetry
(A3)
in y following from Eqs. (I) and (2) for the Wigner function,
the coefficients a,,*k+ I in our spectral ansatz Eq. (3) vanish
for all k=0,1,... .
We observe that Eq. (1) contains only odd derivatives
with respect to y . Due to the symmetry condition (A3), q N
only consists of Chebyshev polynomials of even order in y .
We find from the recurrence relations (Al) and (A2) that odd
derivatives of even-order Chebyshev polynomials can be
written as the sum of Chebyshev polynomials of odd order.
Therefore Eq. (I) yields only terms with Chebyshev polynomials of odd order. Similarly Eq. (2) produces only terms
with even-order Chebyshev polynomials, because it contains
only even derivatives with respect to y. Thus the two Eqs.
(I) and (2) still occur separately in the final equation system
(17) for the coefficients of q N .
Making use of the symmetry condition (A3) together with
the recurrence relations (Al) and (A2), we can explicitly
write down the multiple sums describing the relationship between the coefficients uyk of the mth derivative of the
Chebyshev series of our ansatz and the original coefficients
a1.k. Starting from Eq. (A2), for derivatives of odd order
2 m - 1 with m 2 2 we obtain
545
N..
N..
2m
(A4)
where we have written the product in a closed form and simultaneously shifted the indices of summation by k , that is,
k . Therefore, the upper limit of the first sum should be N , - k, but we can also write N , due to the condition u ~ , ~ = O
if k > N , . Similarly, for derivatives of even order 2 m with m 2 2 , from Eqs. (Al) and (A2) we obtain
f,=
ru-
N.
(A5)
2m-1
(A6)
and for the 2 m - 1 inner sums of the even order 2 m [Eq.
(A5)1,
Since this expression contains multiple sums it seems to be
impracticable to apply it to an arbitrary order m . But the
inner sums with summation indices t 2 , f 3...
, can be factorized. This is shown in Appendix B. With the abbreviation
2m- I
K~~~
:= 2tl+2k+2m,
1
( 2 m - 1)!2-
m- I
a=-(rn-l)
?,ll-4(k+~)21.
(A71
m-
where we have finally shifted the summation index by K = t l + k + m = K m I 1 / 2 . Since a,,k=O for k > N , we do not have to
change the upper limit of the sum. Similarly with Eq. (A7), we find, for Eq. (A5),
546
57
3219
m-I
m-l
(t1+7n-a)(tit21c$m+a)=(ti
tm-u-l+l)(tl
d
Mtl
tni-a-1+9k+2a+1
v
An analogous result holds for the right-hand side of Eq. (A7). Thus Eqs. (A6) and (A7) are rewritten as
(B1)
and
(B2)
where now the multiple sums of a product are represented as a decomposition into a product of independent sums.
We now have two possibilities to carry out the proof by induction. We can strictly separate the even case Eq. (BI) and the
odd case Eq. (B2), and prove each of these cases independently. This turns out to be very complicated. We find it easier to start
with an even-order N=2m-2, and increase N by 1. Then we have to insert different formulas for N and N + 1 = 2 m - 1,
namely, Eq. (Bl) for N and Eq. (B2) for N + 1. The second step is to start with an odd N = 2 m - 1 using Eq. (B2), and
investigate N + 1 = 2 n using Eq. (BI).
547
3220
51
-
In this subsection we prove the sum factorization formula by induction. As mentioned in Appendix B 1, we have to make
a distinction between even and odd integers N . We start this proof as usual, and show that the formulas hold for N = 1 and 2.
The induction step has to be performed twice. First we start with an even N = 2 m - 2 and deduce the formula Eq. (B2) for odd
N + 1 = 2 m - 1. Second, we have to start at this f l = 2 m - 1 to prove the formula Eq. (B1) for even fl+ 1 = 2 m .
(a)First step of the induction. The starting point is a convenient formulation of Eqs. (Bl) and (B2). When N = 2 m - 1 is
odd, we can rewrite Eq. (B2),
N
,,N
(N-1)/2
l,+[(N-l)/2]-u
(b3)
whereas for even N = 2m - 2 we have
(B4)
By simple inserting we check that Eq. (B3) holds for N = 1 and Eq. (B4) holds for N = 2 . For N = 1, Eq. (B3) reduces to
(1
f2=0
1
2I
2 I n (2tl+2-e+2k+e)=11
@=I
ll+o-u
u=-0
h=O
(2X+2k+2a+l),
and we immediately see that the right-hand side equals the left-hand side. For the case N = 2 we need some more transformations to see the coincidence of the left- and right-hand sides of Eq. (B4). We split up the left-hand side,
(1
Iz=O
fz
fI
f3=0
fI
12
2'(2t3+2k+2)(2t2+2k+3)=2
(2t2+2k+3)(2t3+2k+l)+2
Iz=O
t2=0 f3=0
into two sums. In the first double sum we introduce the new indices
interchange the order of summation
fI
fz
II
F2=t2
c
12
(2t2+2k+3)(2t3+2k+3),
r3=0
11
Z E = E E .
Iz=O
13=o
13=o IZ'f3
c -c
12-1
lI
2-
fz=O
11
(2F2+2k+3)(2F3+2k+3)+2
13=-1
f I
fI
- - (2T2+2k+3)(2F3+2k+3)
rz=O
f3=iZ
fI
E -E
=2-
I*=O
(2F2+2k+3)(2T3+2k+3),
t3=-I
which we can easily combine to one double sum. When we shift back the index F3=t3- 1 , the right-hand side of Eq. (B4)
with N = 2 results:
[$:
fI
(2X+2k+l)
(2h+2k+3)
)(h=O
( b ) Starting from N even. We carry out the main step of the proof by induction by starting from the left-hand side of Eq. (B3)
for N + 1, which is in this case supposed to be odd:
N+ 1
2N"n
I ~ . . . I ~ + ~
(2tN+,-,+2k+e)
e = ~
(B5)
The validity of this equation is proven in Appendix C. Now we can insert the induction hypothesis Eq. (B4) for even N , and
find
548
51
11.
,,
3221
that can easily be verified to be the right-hand side of Eq. ( B 3 ) formulated with N + 1, which we assumed to be odd. Hence we
have shown that if Eq. ( B 4 ) holds, then Eq. ( B 3 ) also holds.
(c) Starringfrom N odd. This case is more complicated and needs some transformations. When we start from N odd, the
hypothesis of the induction is Eq. ( B 3 ) . The left-hand side of Eq. ( B 4 ) for N + 1, which is now even, reads
N+2
2N+'
I ~ . . . I ~ + ~
e=2
N+2
zNn( 2 t ~ + 4 - ~ + 2 k + e )
(2fN+4-~+2k+e)=-
(B6)
(B6)
We prove this equality in Appendix C. When we shift the index of the product
k + 1, this reads
r=
\ /
?=e - 2
II+N
This formulation allows us to apply the hypothesis of the induction equation ( B 3 ) , and we find
(N-l)n
1, + [ ( N - l ) R ] - r
f,
+ [ ( N - 1)/2]+[(N+ I ) R ]
(2h+2F+2u+l)
[22,+2+2k-(N+
1)+3]
IN?=o
(N- l ) R
I 1 +[(N+ I)fZ]-u-
(N+1)!u = - ( N + I ) / Z
F= k +
(2h+2k+2u+3)=-
(~
N+L)R
U+
1 we obtain
f,+[(N+l)f2]-u
(N+1)!; = - [ ( N - ~ ) R ]
h=O
a=
(2X+2k+26+
l),
X=O
which is indeed the right-hand side of Eq. ( B 4 ) for N + 1 (even). Thus we now have shown that if Eq. ( B 3 ) holds, then Eq.
(B4) also holds. Together with the results of the former subsection where we showed that if Eq. (B4) holds, then Eq. ( B 3 ) also
holds, and with the special cases N = 1 and 2 , the proof is now complete.
APPENDIX C: PROOF OF EQS. (B5) AND (B6)
In this appendix we prove formulas (B5) and (B6). These are two different formulations but essentially identical. In
Appendix B 2 c we used the formulation
N+Z
(~+1)
i2. . I ~ e
+=~
2
N+2
(2fN+4-,+2k+e)=
i2...fNfl
2
f2...tN+2
e=3
fl (2fN+3-,-+2k"+F)=-
1i 2
Il+N
(2~,+~-,+2k+e)
N+ I
zN+'
(2tN+Z+2kpN+2)
(C1)
!N+2='
N+I
Q=l
3 ( (C2)
(
(N+ 1)
(2tN+,-Q+2k+e)
~ = 2
(C3)
549
M. HUG, C. MENKE, AND W. P. SCHLEICH
3222
51
and shows how the multiplication of N+ 1 with a product can be written as a discrete sum of N diverse products. Note that
n$=; = 1 for a>/3. With Eq. (C3), we can rewrite the left-hand side of Eq. ( C l ) ,
N+2
( ~ + 1 )
e=z
I ~ . . . L ~ + ~
(2tN+4-e+2k+e)
\
1 N+2
f u rI
(C4)
where we have written the term u = 2 separately. We now rename the variables in each summand in such a way that the
separately written factor t ~ + 4 - , , is labeled with the same index t N + 2 in each summand. To accomplish this we shift, all indices
t , with v > N + 4 - u by 1, that is, tN+4-e is renamed by t N + 4 - ( e + lif) e G U , which is the case in the first product. Hence Eq.
(C4)now reads
1 N+Z
x ( 2 t N + 2+ 2k - N - 2 + 2 u)
(C5)
f u rI
i N+2
N+2
(2t~+2+2k-N-2+2U),
where we now are able to combine the first and second products. The factors containing the indices t 2 ,. . . . t N +I are common
to all summands = 2,...,N + 2 . W e therefore can factor them out. Hence the right-hand side of the last equation reads
N+2
(2tN+4-,+2k+e)
550
MODIFIED SPECTRAL METHOD I N .
51
. . . 11.
3223
,, ,
It+N
In this appendix we finally show the formula Eq. (C3) which we used in Appendix C. We carry out our proof by
mathematical induction. Equation (C3) formulated for any integer N reads
N+Z
(N+ 1)
II ( 2 t N + 4 - e + 2 k + e )
e=2
z2[g2
N+2
u-I
[2tN+4-,+2k+(e+l)l
(2tN+,-,+2k-N-2+2a)
[ II
e:::
(2fN+4-~+2k+e) ,
I
(D1)
(D1)
where we recall that H g = m .= 1 for a l p , The starting point of the induction is to verify Eq. ( D l ) for N=O, that is,
2
( 0 1 )( 2 t 2 2k + 2 ) =
E ( 2 t2 + 2k - 0 - 2 + 2 a ) .
u=2
After this trivial step we start for the further proof from Eq. ( D l ) , which is our induction hypothesis. We substitute N+ 1 for
N in Eq. ( D l ) and obtain
N+3
(N+2)
e=2
(2t~+,-,+2k+e)
(D2)
We now have to reduce the left-hand side of Eq. (D2) to the left-hand side of Eq. ( D l ) . We do this by rewriting the product
( N + 2 ) (2tN+3 + 2k + 2 ) = ( N + 1 ) (2tN+3 + 2 k + 3 ) + ( 2 f N +3 + 2 k - N + 1 ) .
When we use this relation and shift the index of the product
(:::
II
Ni-3
JJ
( ~ + 2 ) (2tN+,-,+2k+e)=
F= e-
(2tN+,-,+2k+e)
e=z
(N+2)(2tN+3+2k+2)
N+2
( 2 t N + 4 - c + 2 k + e + 1)[(N+ 1 ) ( 2 t , + 3 + 2 k + 3 ) + ( 2 t N + 3 + 2 k - N +
I)].
~ = 2
When we define
F= k + 5 we now can replace the left-hand side of the hypothesis Eq. ( D l ) by the right-hand side, and obtain
55 1
3224
N+2
I N+2
57
\
Since we can include the term ( 2 t N + , + 2 F + 2 ) into the first product, and since the last summand is the term u= 1, we can
combine this to
With
C?= a+ 1, we arrive at
which is the right-hand side of our formula Eq. (D2) for N+ 1 after resubstitution of