Академический Документы
Профессиональный Документы
Культура Документы
C. FRONSDAL
University of California, Los Angeles, U.S.A.
(Received February 23, 1978)
References 145
Selected Notation
.Ce==an algebra of distinguished observables, either as an abstract Lie algebra or as an alge-
bra of functions on phase space with the Poisson bracket.
ccu’0m- a subalgebra of ,d.
.?I,, .d, ~7the representations a -+ a*, a ‘--f -_*a.
.cJ” .y the real vector space dual of .sl.
add, ads _ the adjoint and co-adjoint actions of .d.
0, h, c = elements of .d or of &**.
E : function on G, related to E” and to Exp.
E” -2 *-representation map.
Exp(f) = a formal series, a function on W.
e 2 basis element for the center of H,.
E(2) =: two-dimensional Euclidean algebra or group.
f, .f“, &, y’l, . . . - functions on W.
f:.&, $1, . . - functions on G or on G-homogeneous spaces.
3&w. the inverse Weyl application.
G _ a connected Lie group with Lie algebra .N’.
G0 = a subgroup of G.
H, = Heisenb:rg algebra or group of dimension 2n+ 1.
AC, X0 = Hilbert spaces.
K = character for sp,,.
3: = character for G or Go.
LA(A~-0,1,2orA-1,2,...)=abasicfor.dorcoordinatesfor.d*.
I(a) ;: vector field for the left regular representation of G.
M, -= metaplectic algebra
A
Q, Q, . . . = Casimir elements.
I.(U) = vector field for the right regular representation of G.
7i, C ;- the representations g H E(g)*, g I+ *E(g-‘) of G.
,T 7 the irreducible representation of G with character x, or Ihe representation of G induced
from a character x of Go.
Tw 7 representation of G by co-adjoint action in L2( IV).
T, T’, . . ~.. various other group representations
W 7 phase space, or an orbit of ads in .d*.
X,,X , . . -= *-polarizations, subspaces of CJj(W).
2x,,?>. ,... - space of functions on either .d or G.
<E, Q;, dual form on .3-j’,&* (5 a point in .d* or in W).
SOME IDEAS ABOUT QUANTIZATION 113
Introduction-Summary
Correspondence principle. Between the two types of descriptions of physical systems-
one by classical, Hamiltonian mechanics, the other by quantum theory-there can be
no well-defined correspondence ; for it is evidently not possible to maintain that two
theories-one classical, the other quantum-describe the same physical system, unless
one or the other or perhaps both are approximate. The most optimistic point of view is
that the quantum theory is an exact description of the physical system; t&s then becomes
the dejinition of the physical system. Adopting this view, and regarding every classical-
mechanical description as an approximation, we conclude that the problem ofquantiza-
tion is ill-posed because the physical system to be quantized is not fully defined in terms
of its approximate classical description.
Consider now a physical system, approximately defined by classical equations of mo-
tion; is it possible to define it more precisely by providing more information about it?
Is it true that a deeper understanding of the system leads to a reduction of the ambiguity
that is intrinsic to quantization? An example will suffice to justify an affirmative answer.
Think of two interacting point particles in a homogeneous, otherwise empty universe
and suppose that the classical equations of motion are given in terms of an inertial coordi-
nate system. The fact that the universe is precisely homogeneous cannot be inferred from
the equations of motion, it therefore constitutes additional information. The required
exact description of this system by quantum theory must incorporate exact translational
invariance and this amounts to a restriction of the class of possible quantizations.
How can one obtain information about a physical system besides that contained in
the equations of motion? Usually, every C” function on phase space is regarded as a clas-
sical observable ; but perhaps some observables are more fundamental than the others.
Bayen et al. [l] have proposed that one attempt to identify an algebra (JS’)of distinguished
observables, consisting of C” functions on phase space (IV) with the Poisson bracket.
This algebra would include the generators of exact symmetry groups and others whose
geometrical significance were expected to be retained after quantization.
In the usual Weyl quantization scheme (see 8 1) the Heisenberg algebra H,, (2n = di-
mension of W) plays a very fundamental role (see $6 2 and 3). However, it is easy to imag-
ine physical systems for which no canonical realization of H, exists, such as those with
a compact configuration space (see 3 4 for examples). It was proposed’ that the role that
is played by H,, in the Weyl quantization be assigned to the algebra d of distinguished ob-
servables.
The second condition ensures that this *-product is &-invariant in the sense that, for
every a in d,
{a,f*.?> = {aJ>*f+.f*{aJ> (v)
All *-representations examined up to now are analytic and normalized in the sense of
DEFINITION 2. A *-representation 8 of G on Wis called analytic if it coincides, in a neigh-
borhood U of the identity of G, with E(g-‘) * d,g, where E(g-‘) is analytic in U. It is
called normalized if, in addition,
El,, = 1, /(a)El,, = -a/ih (vi)
for each a in d
The first condition of Definition 1 implies funder some additional differentiability
assumptions) that 6’ is an eigen-distribution on G. (See 0 7.) In the case of a normalized,
analytic *-representation this suggests that 8 coincides with @g-l) d,g, where the function E
is analytic in some open, dense domain in G.
Suppose that 6’ is a normalized, analytic *-representation. We may take the neighbor-
hood U in Definition 2 within the domain of the exponential map. For g in U, write g = co
with a in d; then we have a convergent series:
where T,,(a) is an n-linear function from ti into Cm( IV). In particular, TO = 1 and T,(a)
= a. A mapping from the space of formal *-polynomials into C”(W) is defined by (v) and
(a*)” = T,(a), n = 0, 1, . . . (viii)
In this way the domain of the *-product defined by (iv) is extended to the linear span
of the functions T,(a). In all cases studied up to now it turned out that each T,(a) was
the restriction to W of a polynomial on JXI*of degree < n.6
*-polarizations. Suppose now that phase-space (W) has been equipped with an alge-
bra d of distinguished observables, and with a *-representation 8 of G on W, where G is
a connected Lie group with Lie algebra ,&. Then we say that the system has been *-quan-
tized. It turns out that there is a close relationship between *-quantization, pre-quanti-
zation in the sense of Kostant [2] and Souriau [3], and quantization in the conventional
sense. (This is explored in $0 10-15.) The relationship between *-quantization and pre-
quantization is based on the identification, first proposed by Souriau7 of phase space
with an orbit of the co-adjoint action of G in the vector space dual d* of PI. It must
be stressed, however, that *-quantization is intended to be a complete and autonomous
theory, equivalent to quantum theory, whereas pre-quantization is only preparation for
quantization. Jn both cases one deals with the space Cm(W) of functions on phase space;
.-___ _
6 Compare Ref. [I], I, Section 12.
’ Ref. 9, p. 105.
I16 C. FRONSDAL
that is, functions of 2n = dim W variables, and in both cases one considers a reduction
of the number of independent variables, though the purpose is very different. For simpli-
city, we now assume that I is surjective.
Consider the endomorphisms of Cm(W) defined by
f H a*f-f*b, (a, b) in d@d.
These form a representation of &Q&; for simplicity we suppose that it is irreducible.
The representation of ~2 given by f I-+a *f is then isotypical; that is, a direct sum ‘or in-
tegral of equivalent representations. We now give a precise formulation of the idea of
reducing the number of independent variables.
DEFINITION 3. A *-polarization is a topological subspace of Cm@‘), stable and irre-
ducible under the algebra of endomorphisms given by f H a *f, a in d.
For simplicity we now consider *-polarizations that can be defined in terms of a subal-
gebra do of d and a character X on &, as follows:
-JL = (SE WW>; f *a = K(a)f, Vaod,} (ix)
The analogy between *-polarizations and polarizations in the sense of Kostant [4] and
Souriau’ is very evident; nevertheless, a very important difference should be noted. Here
there is no limitation corresponding to the restriction to invariant polarizations. Associa-
tivity of the *-product guarantees the stability of X, under f H a *f, for every a in A and
for any choice of &,. (See $3 12, 13.)
In $5 lo-15 we explore, by working out two examples in full detail, the relationship
between *-polarizations, polarizations and induced representations in the sense of Mackey.
It turns out that the map G intertwines between an induced representation of L$ (formu-
lated in terms of a space of functions on G, a subalgebra d, and a character X on &,,)
and the action on X, given by a ++a *f, a in d. This interwining operator may be viewed
as an extension of the (generalized) Weyl application (defined for operators) to wave
functions. (See 9 15.)
Symmetry andpositivity. One should have liked to be able to classify, up to “physical
equivalence”, those *-representations that correspond to “realistic physical systems”.
Ill-defined as these concepts are, it is possible to adduce some justification for restricting
the attention to a mathematically well defined class of *-representations. Consider any
real function in the image of 8; we require that its spectrum (in the sense of Bayen et al.‘)
be real. This leads to the
DEFINITION 4. A *-representation & is said to be symmetric if the domain oft” is closed
under the involution f H f*, where f**(g) = fpj, and if
This implies that the image of 8 is closed under complex conjugation in Cm(W) and
that f *f’ = f’*f- The spectrum of a real function is then closed under complex conjuga-
tion, but not necessarily real.”
To prepare for the possibility of an interpretation of some functions on Was proba-
bility densities we introduce the
DEFINITION 5. A Cm function Q on W is said to be positive if there is a function f such
that e =T*f.
If Q is positive we write Q > 0; whether Q is a positive valued function on W, or not,
is another question.
DEFINITION 6. A trace on d is a complex, linear function on the image of 6, with val-
ues in CU { co}, such that tr(i) = tr(e) and tr(@) > 0 for every positive Q.
If 6 is symmetric, then there is an obvious candidate for a trace on 8, namely
case a single, regular orbit is sufficient. Our preliminary study of the noncompact group
SL(2, R) in $21 shows only that many representations can be obtained on each regular
orbit; the “accommodation” between infinitesimal characters and orbits, suggested by
the work of Kirillov [6] and others for nilpotent groups, yields at most a technical simpli-
fication. In the case of the singular orbits of SL(2, R) we have been unable to discover
any *-representations that are both analytic and irreducible. Much work remains for non-
compact groups, especially in connection with the singular orbits.
Ge??eratingfi~nctions. Let there be given a Lie group G with ~2 as its Lie algebra, an
orbit Win d* and an irreducible representation T of G with character x. Is it possible
to find a general algorithm for the construction of a *-representation c” of G on W with
trace x? Ideally, 8 should be symmetric and positive whenever T is unitary, and tp should
be analytic and normalized whenever possible. A very general algorithm has been found
(P 22) although it is not known whether it is the best one possible. In the semisimple case
the *-representation can be constructed directly in terms of x; in general one needs a char-
acter for an extension of G.
Here E = {$‘, . . . . E2”1\ = (q’, . . . . qn;pl, . . . . p”> are natural coordinates for W = R2”,
fi a
and i = {[I, . . . . i2”, are the corresponding operators in X’,: i1 = x', pt' = i -a;l-,
algebra, as well as the associated abstract Lie algebra, will be called the H&&erg al-
gebra and denoted H, . Every Weyl application has the following property, for all a in H,,:
I’ See footnote 2.
I2 The reason why one refers to (6) as an invariance property is that it implies Eq. (i) of the introduc-
tion, which shows that the Heisenbxg group is a group of symplectic automorphisms of the Moyal algebra.
120 C. FRONSDAL
4. Substitutes for the Heisenberg algebra. The symmetric Weyl application is invariant
under algebra M,,; the more general Weyl application is less invariant but retains the
invariance under the Heisenberg algebra H.. The invariance algebra of the Moyal product
has a geometrical significance and is doubtless of physical importance, but must it include
H,? We shall consider quantization schemes-generahzed *-products and generalized
Weyl applications-invariant under any Lie algebra &, but first let us consider some
elementary physical examples.
First example, d = E(2). The physical model is a “free” particle in one dimension,
confined to the interval - 1 < q < 1 of the real line, with Hamiltonian function H = p2/2.
The Heisenberg algebra is inappropriate in this case because the function p is not a “good”
observable: it does not generate a group of canonical transformations. Geometrically,
the reason is that the classical trajectories generated by the function p crash into the bound-
aries at q = t_ 1; that is, {f, p} is different from zero at q = rt 1 (for most functions f).
A possible choice for the algebra & ofgood observables is the span of (LA} = {Lo, P’, P’},
where
LO = (I-qyp,, L’ = p’ = q, L2 = p2 = (1 -qy. (8)
This algebra is isomorphic to E(2). Suppose one simply replaces q H x, p H (E/i) a/ax
in (S), retaining the order of the factors. Then the invariance condition (4) holds for
a = a, LA. The operators iA can be made self-adjoint if we take the following inner product:
The Schrodinger equation has the usual form, but the boundary conditions @(& 1) = 0
that lead to quantization of the energy levels are satisfied by every normalizable wave
function and need not be introduced in an ad hoc manner as in the usual treatment.
Second example, s2 = SO(~). This algebra is appropriate if phase space is a 2-sphere.
The physical model is a relativistic particle confined to a circle-the equator of S2. The
meridians are the momentum fibers and their termination of fusion at the poIes reffects
the limitation on velocity implied by special relativity. In the limit (velocity of light =)
c + co, so (3) contracts to E(2). Quantization on this compact phase space will be carried
out in 0 20. Conventional Weyl quantization is evidently not applicable.
Third example, .&’ = so(2, 1). The physical model is the one-dimensional Kepler
problem. Configuration space is the half line r > 0. The Hamiltonian may be regularized
in the same way as in three dimensions-one replaces the time parameter by the mean
anomaly and obtains the following equation of motion
hold for all k in C, a in ,r4, fin 17(g). Under some restrictions, the *-product can be ex-
tended from n(9) to Cm( 6’). An algorithm has been given for the construction of all
invariant *-products on any regular orbit of any semisimple Lie algebra.15 The general
result for H, is given by Eqs. (l), (2) and (5). A similar formula for E(2) will be derived
later. We now develop an important tool, the *-exponential function Exp.
6. The functions Exp and E. Given an invariant *-product on W one defines a function
Exp: JZZ-+ n(P), where 8’ is the ring of formal power series on d*, by
m
If {eO>denotes the local group having d as its Lie algebra, and if eaeb = ec denotes the
Campbell-Hausdorff application, then
Exp(a) * Exp(b) = Exp(c). (17)
We assume that (16) converges for a in some neighborhood “Z of the origin of & so that (16)
defines, for every a in +Y,a function Exp(a) on W. The value at l in W will be denoted
Exp[(a). It will also be assumed that (17) makes sense when Exp(a) is considered as a func-
tion rather than a formal power series.
Let G be a connected Lie group having d as its Lie algebra. The canonical, local
diffeomorphism from % to U (an open subset of G including the identity) allows us to
regard Exp(a) as a function on U. All eIements of G will be taken to belong to some open
subset Vof U, small enough so that all products that appear will be in U. With this under-
standing we define
E(e”) = Exp(a), Me”) = Exp&) (18)
and rewrite (17) in the form (compare Introduction, Eq. (iv))
E(e”) *E(eb) = E(e”eb). (19)
Another property of Exp, that is an immediate consequence of (15), is invariance under ad,,
precisely (compare Introduction, Eq. (iii)):
(ad,(a)Exp)(b) = (a, Exp(b)) = ad>(a)Exp(b). (20)
Finally, the function Exp satisfies a set of differential equations; these will be discussed
next.
The function Exp and E are important for two reasons. Firstly, it turns out that any
function from % to Cm(W) that is ad, invariant in the sense of (20), and satisfies the dif-
ferential equations as well as some boundary conditions, is the function Exp associated
with some invariant *-product on W. l6 This provides a method for construction of all
invariant *-products on W. Secondly, it will turn out that the function E is an essential
ingredient in the formulation of the generalized .Weyl application. In fact, the function
E for H, is essentially the kernel of the integral transform (l), and our study began with
the idea of generalizing the Weyl application.
7. DQj5erential equations for Exp. Let
Q = g,,LALB, g,,eR, A,B= I,..., m, (21)
be an element of the center of the enveloping algebra of &, and Qw the function on W
(34)
2 in R2, 0 < t < 2~. With this choice it is easy to calculate the *-product for E(2).
9. A *-product for E(2). Using the parameterization (31) we can write I(b) and r(b)
in the form
-Z(b) = b~(X~a,-x,a,-xAa)+X~b~a+x,b/\a (35)
r(b) = bo(x,20-xo23+x~2)+x,b~ 2-xob~a. (36)
Here aP = 2/2x,, ,U =O, 1,2,3andxAa =x,2,-x,2,,b.2 = b,2,+b,2,. NowEq.
(23) gives
b*Exp(a) = ifi[bo(x,2,-x,2,-xA2)+x,b*2+xobA2]Exp(a). (37)
When for Exp one takes the simple choice (33) then this can be written
Lo* = Loll - (ifia,yp+ifiPA a, (38)
6. P* = 6. P[I-(ifi20)2]1~Z+ifib~P20. (39)
These equations are to be understood as operator identities, derived for application to (33),
but valid for application to any function on W of the form f = ly(a) Exp(a)da; that is,
to any function in the domain of 6’. (See the Introduction.) Since &’ is essentially surjec-
tive in this case, (38) and (39) are applicable whenever the right hand sides make sense.
Thus we arrive at pseudo-differential operators.” In the case of E(2) there exists no
invariant *-product such that b* is a differential operator. This appears to be typical
except for nilpotent groups. The Moyal product for H,, is generated by differential opera-
tors if the function F of $ 1 is a polynomial. For so(2, 1) there exists only one invariant
*-product that is generated by differential1 operators. [We say that a *-product is gener-
ated by differential operators, resp. pseudo-differential operators, if the operators a* are
differential, resp. pseudo-differential, operators.]
” We use this term in the sense of a formal series of differential operators of unbounded order.
SOME IDEAS ABOUT QUANTIZATION 125
Il. Qantization for HI. Preliminaries. The basis is denoted (Lo, L1, L2} = (E, q,p}.
The orbit W is defined by e = 1. On this orbit the Poisson bracket and the co-adjoint
action are
{a, b} = UAb, #a = aA 2, (40)
where ar\ 2 = a, 22-a,2, and 2, = 2/2L*. The symplectic two-form, defined by
~(#a, *b) = {b, a}, is exact, o = dcc, and the potential I-form is a = -qdp+dm, where m
is a differentiable function on W. The vector fields Z(a) and r(a) are (with 2* = 2/2b,)
l(a) = -aA2!-(1/2) (a/\b)2’, r(a) = aA2”-(l/2) (aAb)2’. (41)
The differential equation for Exp, and its solution, are
[f(G)-e/%]Exp(b) = 0, (g k e = l), (42)
Exp&~) = F(b, , b,)e(t,b)“fi, <E, b) = LAbA. (43)
The function F was described in 5 1. If F = 1, then we can use (23) to obtain e* = *c = 1
and
4*
*4 I= (I-+ (jW)al2p, f..)= p ‘f: (%/2) 2/2q. (44)
Three different methods of quantization will now be shown to lead to the same result.
Polarization-foliation (Kostant, Souriau). Define, for a in HI,
23 The &functions that appear in Eqs. (54) and (67) do not indicate a necessary relationship between
the character and the orbit. We have elected to relate these a priori independent parameters in order to
simplify the expressions for Exp. In general, the &functions relate the inducing character to the *-polari-
zation character; see Eq. (79).
128 C. FRONSDAL
The integration extends to the double covering of E(2) and d,g = dtdx, dx, is the Haar
measure. Inserting (65) and (34) we get 25
f'(E)
- W-R'>_fM (67)
where fA has the form (59) with y+(p) = &(p- 1,). Transferring the operators (64) to WA
one obtains full agreement with (60).
13. Comparison. The results of *-polarization and of induction are identical, and
differ only superficially from that of the method of Kostant and Souriau. The appeal
that the latter has for geometers is that the entire quantization procedure takes place in
~~..
24 This is another example of Eq. (ix) of the Introduction.
25 See footnote 23.
SOME IDEAS ABOUT QUANTIZATION 129
the original geometrical setting of classical phase space. Real polarizations do not always
exist, but the introduction of complex polarizations by Kostant has widened the field of
applications to include a large class of solvable groups. Cases are known, however, when
no invariant polarization exists, either real or complex. [The most interesting case is
the quantization of the Kepler problem on an orbit of the dual of so(4,2).] Invariance
of a polarization is simply the requirement that the space of functions annihilated by the
polarization be stable under the action of the operators d(u), a in .&. [In the case of HI,
with polarization *p, this is the space of functions of the form (46); in the case of E(2),
with polarization *P’, it is the space of functions of the formf(l’, v) = exp(-m/z%)y(p).]
A *-polarization is a space of functions that is required to be stable under the action of a*,
a in d; but this is guaranteed by the construction-it is simply the statement that a* com-
mutes with *b (for all a, b in ~2); that is, associativity of the *-product. The alternative *-PO-
larization for E(2), that follows Eq. (60) above, is based on a subalgebra that does not
satisfy the condition of invariance that is required for a Kostant-Souriau polarization.
We believe that the method of *-polarization will encounter no obstacles, even in cases
when invariant polarizations do not exist.
The idea of the method of *-polarization is the following. The application &[ of Z?
defined by a H a* is a Lie algebra isomorphism. [The Lie bracket [a, b] corresponds
to the *-bracket, or generalized Moyal bracket, [a * b] = (a * b-b *a)/%.] The same is true
of the application d, or d defined by -a H *a. Thus, we have a representation D of
&‘@d by (a, b) H a*-*b. Let us assume that this representation is topologically irre-
ducible so that D = D, x D,, where D,, D, are irreducible representations of d deter-
mined by &i, d,., respectively. The problem is to find a space X of functions on W on
which .zJ’~acts irreducibly. Now X is not stable for the action of ZZ’,, but it is highly pla-
usible that two representations of .@‘, given by the action of -02, on X, resp. Y = X*a
(where a is some fixed element of d), are equivalent, the intertwining operator being
closely related to the operator *a. If this is the case, then the choice of X is of no impor-
tance so long as JZZ~acts irreducibly in it.
The most straightforward construction of a space X with the desired properties is
to choose a suitable subalgebra ZZ’, of JZI and reduce JZZ’,,to scalars. Let a +* K(a) be
a character for .sl,, and let X, be the space of measurable functions on W such that
*-polarization, faithfully to the point of reproducing precisely the formulae (48), (49)
and (59) (60). This suggests that the Exp transform intertwines between the representa-
tions of &‘@d given by (d, b) e, Z(d)+r(b) and (a, b) H a*- *b, the former acting on
a space of functions on the group and the latter on a space of functions on W.
Before turning to a formal discussion of the Exp transform in the general case, it is
convenient to recall the essentials of the method of induced representations. We consider
induction from one-dimensional representations only.
14. Induced representations. Let G be a connected Lie group and G,, a closed subgroup.
Let g, H x(gO) be a character for Co and let -?X denote the space of measurable functions
on G such that, for all g, in G, and all g in G
(70)
is a G-invariant scalar product on 2;. The measure d,g is the right-invariant Haar measure
on G.
Let J: G/G, H G, z H gZ be a differentiable map, such that g, belongs to the left
coset z of G, in G. For any g in G we write g = go(g) *g,(g), or simply g = go *g,; both go
and gZ being determined by g. Any solution of (69) can be factorized as
15. Exp transform and Weyl applicdtion. Given an invariant *-product on an orbit W,
and the associated function (or distribution) E on G, consider the application (compare
Introduction, Definitions 1, 2):
(76)
SOME IDEAS ABOUT QUANTIZATION 131
The image of _?xby Q is therefore the *-polarization XK defined by (68), X being the infini-
tesimal character related to x. Similarly, for J7 in 2%:
That is, the mapping (76) intertwines between the induced representation T,(g) and the
representation dejned by the action E(g)* in X,. The condition that 8 be defined densely
in _%Xis verified in some case’s (compact groups, some discrete series of representations
of semisimple groups). In other cases it is necessary to consider a space of characters,
endowed with a measure dK and associated d-function distribution. The application
b, = SlgX then takes the form
8,: f” r-+f), f’(5)%, xo) = [&Mg-‘Mg (79)
G
where x0 is a unique character, determined by the choice of the *-product, i.e., by the
function E. Examples of (79) were given by Eqs. (54) and (67).
Returning to (78) we note that d defi:es a generalized Wigner application, and its
inverse a generalized Weyl application. Let A be an operator of the form
Note that this is not inve$ble;Jf 2, i are the operators &Q(Z) = GA(z),(&Z))(z)
= &(z+~mod2n), then Aw = Bw = 1. Thus a new degree of freedom appears on
quantization; this is because the function E defined by (34) is defined on the double co-
vering of E(2). (Invariant *-products for which the functions E are defined on the N-fold
covering of E(2), including E(2) itself, do exist, but are somewhat more complicated,
see 0 16.)
The Exp transform establishes a link between *-products and unitary representations
constructed by the method of induction. In fact, it was pointed out by Bayen et aZ.[l],
in connection with spectral theory, that a link exists between invariant +-products and
representations. The problem that will-be discussed next is whether a given *-product,
defined most conveniently (as in the Introduction, Definitions 1 and 2) by a function
or distribution E on a group G, is associated with a unique representation of G. One
should also like to know under what conditions this representation is unitary and/or
irreducible, etc.
Physical applications will not be considered in this paper. It may be worth while to
emphasize, however, that the results of this section may turn out to by very useful indeed.
One of the main reasons why the Moyal formalism is sometimes found to be somewhat
unwieldy is that the Wigner-Weyl correspondence deals exclusively with operators and
not with wave functions. The “Exp transform,” or b-application, remedies this defect.
to exist, between *-products and unitary representations, suggests that one look for a ses-
quilinear form (f,f’), invariant under the mapst7
C(g) : f I+ ml *L (88)
T,(g): f Hf*m-l). (89)
That is,
@(g)*.L %)*f’) = UJ-‘) = (J-*%)X*%$ (90)
Such a form exists for *-products that satisfy a condition of symmetry. An invariant
*-product is said to be symmetric iff
E(g) = E(g-‘) (91)
[the bar denotes the complex conjugate]. If (91) holds, then
f7 = f’*f (92)
holds for a very wide class of functions; in particular, for functions of the form (functions
in the image of 8’):
[Here and below, drg denotes the right invariant measure on G.] We shall limit ourselves
to symmetric *-products and to functions that satisfy (92).
THEOREM. Let an invariant, symmetric *-product be given on an orbit W, and let dl
denote th.e G-invariant Liouville measure on W; then
(94)
is a hermitean sesquilinear form, invariant in the sense that Eqs. (90) are satisjed for g in G.
Proof: That (94) is a hermitean sesquilinear form satisfying the first of the condi-
tions (90), follows immediately from (92). As to the second condition:
(f*%W'*Ek)) =
W
sEtg-‘kf$-‘*-%iM
s
= t.hf’><gWt = ff,f’>.
W
27 Ne continue the formal development in a notation that is strictly appropriate for the case of analy-
tic *-representation only. (Introduction, Definitions 1 and 2).
z8 Compare Introduction, Definitions 5-7.
134 C. FRONSDAL
then
x(g) = iE(g)dE
= trE(g). (97)
(W.F)(g,) = _&I g)
and the inner product (f, f’& . Then
All this suggests that a *-product quantization scheme may be physically acceptable
iff the trace is the character of a unitary representation, and that two such schemes are
physically equivalent if and only if they have the same trace. (Introduction, Definition 8.)
17. Example, HI. The general form (43) for Exp(b) defines E(eb) globally,
E(eb) = F(b, , b,)e(e,6)/i6, (102)
since HI is an exponential group. [That is, b H eb defines a global diffeomorphism of
the algebra onto the group.] Symmetry of the *-product means that
WG, W = F(-b,, -U.
Integrating (102) over phase space, we find, since F(0, 0) = 1:
which agrees with (103). Note that the trace is independent of the choice of the factor F
in (101). This was to be expected, since HI has only one unitary irreducible representation
(up to projective equivalence). Different choices of F are believed to lead to physically
equivalent quantum ‘theories.
18. Example, E(2). The general form (30) for Exp(a) gives, globally, on the N-fold
covering of E(2),
E(etLo~“‘) = c,(t)cos((~, x)/X)-z&(t)sin((E, x)/x), (107)
provided that c,(t+Nx) = c1(t) and c,(t+Nrr) = (-)N~l(t). The *-product is symmetric iff
--
c,(t) = Cl(--t), c,(t) = CJ-t). (108)
Integrating (107) over the orbit, we find
N-l
(109)
(111)
0
we obtain
TA(etto&$): GA(z)H e R’[x,cos(2z+2ftl)+x,sin(2z+2~+~)]/i~~(z+~) (112)
and this is in agreement with (109), provided R’ = Q*1/2 and provided that we choose
cl(m) = C”. (114)
19. Compact case. We shall find a necessary and sufkient condition for the existence
of an invariant *-product with trace 3: on an orbit W. It will be demonstrated that the
function E is always analytic in both variables; that is, on WxG.
Given any invariant *-product on an orbit W, let T be the representation of G defined by
T(g): f I+ E(g)*f*E(g-‘).
Invariance of the *-product gives
F’Wfl(5) = fk- ’8.
Now let Tw be the representation of G defined by the co-adjoint action in L2(W, dt)-
where d5 is the invariant Liouville measure; then
Tc Tw.
Suppose now that the *-product is symmetric, and let Tl be the representation g H E(g)*;
then T? is the representation gH *E(g-‘) and Tl@Tl* = T. Thus, we have proved the
PROPOSITION.If an invariant *-product on an orbit W is symmetric, and if Tl is the re-
presentation g t+ E(g)*, then
Tl@Tl* c T,, (115)
where Tw is defined by the co-anjoint action of Win L2( W, dt).
The representation Tw is a countable direct sum of irreducible representations:
Tw =&T,. (116)
CZ=O
[a # B does not imply that T,, Tp are inequivalent !] We introduce an orthonormal basis
on L2( W, dl):
{Yg} (a = 0, 1,2, . . . . m = 1, 2, . . . . d,), (117)
adapted to (116) and consisting of functions that are polynomials on W. [This means:
they are restrictions to W of polynomials in the linear coordinates on &*.I
SOME IDEAS ABOUT QUANTIZATION 137
On the other hand, let T’ denote the unitary representation of GX G defined by the
left and right translations on G, in the Hilbert space L’(G). If 2 is the complete set of
characters of irreducible representations of G, then the reduction of T’ is given by
(118)
Given W, we choose a x in Z and attempt to construct an invariant *-product associated
with TX. In view of the proposition, we must evidently choose x so that T,@T,* = Tw.
This insures that the complete reduction of T,QT; is obtained from (116) by restricting CI
to a finite set N:
TX@ T; = @Ta. (11%
-EN
s dt = 1, s dg = 1, (122)
W G
20. SU(2), the function E. As an example of the construction of a-products and quan-
tization on a compact phase space, we consider the 2-sphere and the natural action of
W(2) induced by the homomorphism W(2) -+ SO(3). A physical model was proposed
in 0 4. A basis for & = SO(~) is denoted (LA} (A = 1,2,3) and an element a of so (3)
is written a = aAL* in terms of real parameters aA, A = 1,2, 3. The commutation re-
lations are [L1, L’] = 2L3 with the cyclic permutations. The orbit W is given in terms
of a fixed, real positive number R by
TW = @ T,, (126)
a=O,l,...
where T, is the irreducible representation of dimension d, = 2cr+ 1. The condition (I 15)
of the proposition is satisfied for every irreducible Tl . Ordinary spherical harmonics,
denoted {Y$“) (m = -c1, --a+ 1, . . . , cw;ilr = 0, I, . ..). and normalized so that
dt = dL’dL2/4aRL3, s dt = 1, (128)
W
(with 8, E a/&P):
&LA)\
= +(XA~IJ-X~~_J-(XA~)A. (130)
GA) I
The differential equation for E, and the definition of I, are given by
1
1
(x,$-x&J2+2Z(21+2) E(x) = 0, 6 = 4fi21(1+ 1). (131)
1-
D c.v
A solution that is regular on SU(2) exists if and only if 21 is integer. [We take 21 to be
non-negative without losing generality.] The regular solutions are linear combinations
112
of the hyperspherical harmonics (a = 0, 1, . . . ,21; m = = - a, . . . , a):
YoC,i”(X)
here normalized to
=
1 (132)
s
G
y,aim(x)Yp”‘(x)dx = P’Pm (133)
dx = (d3x/xo)/4x2, s
G
dx = 1. (134)
The gem-al expression for E is
sE,(ddt = W+ I)-‘xdg),
W
(136)
where
xl(g) = sin(21+ l)t/sint
is the character of T,,.
Closed forms for E can be obtained in three ways, by appropriate choice of C, and
summation, by direct solution of (131), or by means of a generating function (§ 22).
With e > 0, take
C,m[l/(22-a)!](fi/R)aQ;‘+‘(cothe) (137)
to obtain, with 2Z+ 1 s 1:
E(g) = (sinh ilc/sinh 5) (sinhe/sinh &I)
cash 5‘ E x0 cash e + (St* z/iR) sinh e (138)
R/fi = lcoth 12~- cothe.
140 C. FRONSDAL
The second relations defines c and the third guarantees the boundary conditions. The
limit ,o + co gives
c, = (2@)! @,R)
(21-R)!
I
E(g) = (x0 1-3 *ZjiR)” 1
(139)
R = 2%. I
The price of this simplification is that I is now fixed by R and k. This type of relationship
or accommodation between infinitesimal character and orbit is interpreted, in the context
of the geometrical quantization scheme of Souriau and Kostant, in terms of integrality
of a de Rham cohomology class. Others have suggested that it amounts to a quantifica-
tion of ?i. We have stressed throughout that no accommodation is required; thus R and
I are in principle independent of each other. It ist rue that (138) impIies the inequality
R < 2Z?i,but this is entirely due to the choice (137) of the coefficients C, in (135). With
the general formula (135) one can obtain all irreducible representation of SU(2), besides
the trivial representation, as an analytic, normalized *-representation on a single orbit
with R # 0.
A simpler road to (138) or (139) is also mar! interesting. An invariant 2-point function,
corresponding to the infinitesimal character Q of G = SU(2), is a function from GX G
into C, satisfying (131) in each argument, and invariant in the sense that f(ggl, gg,)
= f(g,, g2). Evidently, h(gi’gJ = f(g, , g2) defines a spherical function (= a central
eigenfunction) on G. Let g, , g, be parameterized by x and y; then f(g, , gJ depends on
the invariant x. y E xpy, only, and (13 1) reduces to an ordinary, second order differen-
tial equation. The general solution is, with il = 21+ 1 and cosh5 = x *y,
f(g, , g2) = (C, .&‘+ C- e-“?/sinh [. (140)
Now, in (140) let us substitute
-+
IY,) I+ {coshe, @lR)sinhe), @E c. (141)
Let E: (g, l) H Et(g) be the function on GX W obtained from f by this substitution;
then E is an Ad,-invariant solution of (131). Imposing reality conditions and boundary
conditions we easily recover (138). This construction can evidently be generalized to
compact, semisimple groups.
21. SL(2, R). Some of the ideas that were presented in B 16 can probably be imple-
mented rigorously for some types of Lie groups. The main difficulty is the paucity of in-
formation concerning the properties of the mapping d in the general case. In this section
we study another tractable example, selected with a view to illustrating the obstacles
that stand in the way of a general formulation of the theory. Tentative conclusions and
conjectures, suggested by this and the other examples that we have studied, were discussed
at the end of the Introduction.
SOME IDEAS ABOUT QUANTIZATION 141
Algebra, dual, orbits. We use a notation that exploits the relationship between SL(2, R)
and SU(2); many formulae of $20 can be taken over with little or no change. The stan-
dard basis for the Lie algebra sl(2, R) is denoted {LA} (A = 1,2, 3). The commutation
relations are obtained from those of su(2) by the substitution L’, L2, L3 H iL1, -iL2, L3.
The same substitution gives the generator Q of the center of the enveloping algebra:
Q = - (L1)2 - (L2)2 + (L3)2. (142)
The orbits of the co-adjoint action in the real vector space dual are:
W, = (5~s~‘~; LA(l) = 0 (A= 1,2,3)},
W,, = {TV&*; Q(t) = R2 2 0, L3(E) > 01,
W,_ = (E E -01*; Q(l) = R2 2 0, L3(t) < 0},
WR = (5 E &*; Q(l) = RZ < O}.
The orbits are thus parametrized, mainly, by a complex number R E RuiR. All orbits,
except W,, + , W,, _ and W,, , are regular.
Group manifold. The group can be parametrized by
g = &J+x1o1+x2~~++xj$, Xi--x:-x:+x,2 = 1 (143)
where {oAA)(.4 = 1,2, 3) are real 2 by 2 matrices. The SL(2, R) group manifold is identi-
fied with the surface xb- x:-x: +x2 = 1 in R4. We use the convention x0, x1, x2, x3
= x0, -x1 2 -x2, x3 ; then the vector fields of the left and right translations are given
by Eq. (130), and (131) holds provided we interpret e.g. c x: E xpxP. The value of the
Casimir element Q = 4ZP Z(Z+1) is one of two invariants that may be used to classify
(up to a trichotomy) the integrable representations of the universal covering G of SL(2, R).
Representutions [8]. To every pair (I, k) of complex numbers, with - f < Re 2 and
-+ < Rek < $, is associated a representation D(Z, k) of the universal covering G
of SL(2, R),in which the Casimir element has the value 4fi2Z(Z+l), and the spectrum
of $L3 is k+Z. This representation is irreducible unless Z-k or Z+k (or both) is integer.
If Z-k is integer, but Z+k is not, then D(Z, k) is the sum of two irreducible representa-
tions D+(Z, k) and D-(I, k). In the reverse case D(Z, k) is the sum of two irreducible re-
presentations D+( - I- 1, k). The representations D+ (a, k) and D- (CT,k) are chara-
cterized by extremal weights; the spectra of i L3’ being (r + 1 +N and a-N, respectively.
If I+ k are integers, then D(Z, k) = D’(Z, k)OD-(-Z- 1, k)@D(Z), where D(Z) is the
finite-dimensional irreducible representation where the spectrum of f L3 is -Z,
-Z+l, . . . . +Z. The true representations of SL(2, R) are those for which k = 0 or i.
The unitary representations have k real and (Z+$)2 real; in particular, the principal se-
ries has (Z+$)” < 0, the supplementary series has If Ikl < 0, the discrete series are
of the type D’(a, k) with a > - 1 and D-(a, k) with a < 0. Our problem is to determine
which of these representations can be related (in the sense described formally in 9 16) to
*-representations, and on which orbits. We attempt to answer this question by exam-
ining the ad&-invariant eigendistributions on G.
142 C. FRONSDAL
Invariant eigendistributions. Let 9’ be the space of formal power series in (LA}, and
n(Y) the projection of P’ on an orbit. The most general solution of Eq. (131), by a fun-
ction E: G +17(Y) is given by the last expression in (135) with the sum of a extended
to N and C, # 0, CIin N. We choose an orbit W, 1 in C and the Cdl’s,In principIe, the
spectrum of $L” is determined; it is the Fourier spectrum of Exp(L3); but in practice
we do not know how to determine it. A fortiori we don’t know how to choose the C, so
as to obtain the spectrum appropriate for D(I, k) or one of its irreducible constituents.
However, the series can be summed for special values of the C,‘s; in fact, both methods
already used for SU(2) can be adapted to obtain closed expressions for E(x); and the
analysis can be completed.
The most general cIosed expression for E(x) found in 9 20 is Eq. (140). Adapting
it to SL(2, R) we have
E(x) = (C+t?fC_e-“?/sinhI;,
(1W
coshc = xOcosh~+(Z*~/iR)sinh~.
At the identity in G, [ = Q, so E(x) is analytic near the identity provided n # 0. The
normalization conditions are
+)/id = 1, (~E(x)/~x,.& = LA/%; (145)
that is
C+$.E+C_e-“e = sinhe,
(146)
n(C+ e’,”- C_ eeLe) = coshe+(R/fi)sin@.
The special case of symmetric *-products is characterized by E(g) = E(g-‘). There are
four possibilities :
I. R, e, R, C,, C- real.
If.R,e,inreal, C, = C_.
III. iR, ip, il real, c+ = -C_.
IV. iR, ie, 2, iA, iC, , iC_ real.
The trace. We consider now the orbits W,, , fixing +R > 0. We shall determine
the trace; that is, the distribution on G that is given formally by
restricting ourselves to symmetric *-products, cases, I, 11. Let E be the sign defined by
lim Etanhc = 1.
!L’,-‘UZ
The leading term in Et(x) for large JL”j is then
C,(cosh[)?-‘, if C, f 0,
C_,(cosh&“-‘, if C, = 0.
SOME IDEAS ABOUT QUANTIZATION 143
By definition, Rel > 0. The above integral does not converge (in the usual sense) unless
C, = 0, ReA > 0, and then only on compact elements. We restrict ourselves to this case.
When C, = 0,
sinhe
E(x) = @a(-) m, cothQ+++cI = 0. (148)
For g compact set x,, = cost, sgn(j;_* ,$1x: -x:-x.$~~/~ = sink Now 7t. L’ E (@sint}
(1 < ,LJc 03); coshr = costcoshe-ipsintsine, and
where the contour y is the image in the complex c-plane of the straight line
(costcoshe-i(sintsinhe)p} (1 < ,u < co)
in the cosh[ plane, with endpoint co such that coshTo = cosh(e - it). Now cosh5 and
.ssinh[ are in the same quadrant of the complex plane. At the identity this gives E = sgne
and thus by (148):
R
& ,-+A+1 < 0, co = e-it,
We have thus identified the trace as that of D’(I, k) for R/ii > A+ 1 and as that of
D-( - I- 1, k) for -R/A < 1+ 1. [Of course, the sum must be understood as the limit
as t tends to R from the upper half of C.]
The limitation obtained here, namely jRl/ti > A+ 1, just as an analogous limitation
found for SU(2), is no more than the effect of choosing the expression (137) for the C,‘s.
We may, for example, generalize Eq. (140) as follows
l da(e) = 1,
q>o
and da need not be positive. To remove the restriction it is sufficient to support da on two
points.
The simplest case is obtained by accommodating the infinitesimal character to the orbit,
144 C. FRONSDAL
(153)
If I = -i this simplifies, and if C, = 1 (a = 0, 1,2, . ..) the sum can be carried out,
giving
-1
~~i/ificosX/Z.
,
this may be recognized as the +-exponential associated with the Moyal *-product.30 The
integral over Wo+ converges in the sense of a distribution to (for x compact):
which is the sum of the characters for D+ (-d, i) (Sp f L3 = $ +N) and D+ (-i , -b)
(Sp&F = ;+N).
Summary. Our study of SL(2, R) is far from complete. In the case of regular orbits
it was found that some classes of unitary, irreducible representations can be realized
as *-representations on each orbit, no accommodation being required. This could be
verified because we possessed *-exponential functions in a closed form, suitable for in-
tegration over phase space and subsequent identification with the characters. In the case
of the singular orbits we know of no example of an analytic *-representation that has been
shown to be equivalent to an irreducible representation. No *-representation corresponding
to the principal series of SL(2, R) has been found.
22. Generating furzctions. An algorithm has been found for the explicit construction
of a large class of *-representations. It was found by reducing the Weyl application (9 l),
and the construction of *-representations for SU(2) (Q 20, last paragraph), to common
terms. The following is a somewhat oversimplified account of the method. Details will
be reported elsewhere.
The essential step, in all cases except the semisimple case, is the construction of a mi-
nimal extension G’ of the connected group G by an involution I. Choose a point E,, on W,
and let d, be the stabilizer of E. in & under the co-adjoint action. Assume that d, is
the set of fixed points of I. We have G’ = GuGI; the remarkable fact is that central eigendis-
tributions on GZtum out to be much less singular that on G. Let C = {g1g-I: g in G},
and choose a central eigendistribution (T, defined on GI, equal to 1 on C. It may then
be shown, under some additional conditions, that the map h c, 5(/s), defined by
<E(h), a> = r-a@)4(4,
is a diffeomorphism of C on W. Finally, set
&h,(g) = a(g@*
Acknowledgments
It is a pleasure to thank M. Flato and A. Lichnt5rowicz for hospitality at Collbge de
France, R. Blattner for helpful discussions during which he suggested the connection
between *-polarizations and induced representations, and especially M. Flato and D. Stem-
heimer for constructive criticism and assistance.
REFERENCES
[l] F. Bayen, M. Flato, C. Fronsdal, A. Lichnkowicz, and D. Stemheimer: Deformation Theory and
Quantization, I and II. UCLA preprint ll/TEP/ 7 and 8. Annals of Physics, to lx published.
[2] B. Kostant: Lecture Notes in Mathematics 170, Springer-Verlag, 1970, 87.
[3] J.-M. Souriau: Structure des syst2mes dynamiqaes, Dunod, Paris 1970.
[4] B. Kostant: Proceedings of the International Congress of Mathematicians, Nice, Septembx 1970.
[5] A. Lichrkowicz: Lett. Math. Phys. 2 (1977), 133.
[6] A. Kirillov: Eliments de la The’orie des RepGentations, Edition Mir, Moscow, 1974.
[7] J. E. Moyal: Proc. Cambr. Phil. Sot. 45 (1949), 99.
[S] A. 0. Barut and C. Fronsdal: Proc. Roy. Sot. A 287 (1965), 523.