Professional Documents
Culture Documents
Flato 1970
Flato 1970
Flato 1970
I. INTRODUCTION
In recent years the conformal group has gained much interest and extensive
study in the literature. Many interesting ideas in quantum field theory as well as
new concepts in strong interaction physics were formulated and examined with
the help of the conformal group.
Why is this group so attractive to the theoretical physicist? Considered as
a space-time symmetry group, and in the most orthodox way, this group has a
striking advantage: It is the most general group which (beyond singularities)
leaves invariant the light cone. This last fact led physicists to try using this group as
a possible generalization of the PoincarC group-the symmetry group of the special
theory of relativity.
An immediate difficulty arisesif one takes this last point of view [I]: It can be
proved that the group of all one-to-one applications of the four dimensional
Minkowski space into itself which conserve the causal order of pairs of vectors
in the space, is the semidirect product of the positive dilatations by the orthochron-
ous Poincare group, often called the Weyl group.
However [2], one can utilize the conformal group (which contains the Weyl
group) without violating “physical causality”: It is simple to realize that to any
bounded region M containing the origin in Minkowski space there corresponds
78
CONFORMAL COVARIANCE 79
was found. However this representation was not integrable on the Poincare
subgroup and therefore the spectrum of the energy-momentum vector was
discrete.
This difficulty was overcome in Ref. [8], in which a local representation is con-
structed of a contraction algebra of SU(2, 2), giving rise to a discrete mass spectrum,
and integrable to the Poincare subgroup.
But let us come back to field theory and to our first point of view on the conformal
group.
The fact that this group can in a sense be considered as a “good” space-time
group brings us to examine rigorously the covariance of various wave equations,
Lagrangians, and action integrals under this group. To do so, one should of course
first define rigorously what is meant by covariance under this group in field theory.
Once this is done, one can classify various field equations with respect to their
covariance type and then one can try to deduce some consequences.
It has always been mentioned that, from the above point of view, only equations
describing the motion of zero-mass particles are distinguished, as they are the
only ones which are covariant under the conformal group. Indeed, as the conformal
Lie algebra contains among its generators, in addition to the Poincare generators,
that of homogeneous dilatations and four generators of local (point-dependent)
dilatations, only zero mass will be invariant under such dilatations (the mass
having the dimension of [cm]-‘).
However, it is trivial to remark that, if one interpretes the action of the five extra
generators as just the possibility of changing the unit scale in a space-time dependent
manner (a kind of passive transformation), no difficulty is caused for massive
particle equations as far as conformal (‘extended covariance” is considered. This
very evident point of view in classical field theory, which requires only redefining
the mass in the transformed equation, finds a quantum field counterpart in
the present paper.
If one sticks to the orthodox interpretation, and wants to put the accent on
zero-mass particles in conformal theories, one can of course argue that in reactions
with very high momentum transfer one can expect approximate conformal co-
variance to prevail, even for massive particles, since in such situations the mass of
the hadrons is presumably negligible.
If one adopts this generally accepted point of view, then one can look upon the
usual relativistic quantum field theory as a very badly broken conformal covariant
(or, for simplicity, Weyl covariant) field theory. Such an approach can give, among
other things, a semitheoretical connection between the amount of symmetry-
breaking on one hand and the observed hadronic mass spectrum on the other hand.
Irrelevant to which approach one is trying to push, we shall study in what follows
some general aspects of the conformal group considered as a space-time covariance
group in classical and quantum field theories.
CONFORMAL. COVARIANCE 81
[This already implies that S must in fact be related to the conformal group via some
covering group, viz., SU(2,2), and that S must depend on the point x, since it is
trivial on the translation subgroup and conformal group is simple.]
We shall see that these properties determine S(t, x) almost completely. More
exactly, if we write (as in Ref. [lo]) Q(a, x) = X-l(1 + 2ax + a2x2), QkJ =
595/61/1-6
82 FLATO, SIMON, AND STERNHEIMER
exp(kJlog Q), where .I is the matrix (‘I!1 -I),,I of the order of S and I1 is the identity
matrix, and recall that a real irreducible representation is said to be absolutely
irreducible if its complex extension is still irreducible, we have the following
entirely geometrical result:
which determines its vaiues for any t E V by (2.1) and (2.2). Here n E R, and is
called the conformal degree of S, y = i(i2 = -1) in the complex case, y = J in
the real case,k E R, and k = 0 in the real absolutely irreducible case;&(a, x) is the
uniquely defined analytic (for x in the connectedcomponentof the origin in M - Ct ,
and therefore also outside singularities in M,) unimodular solution of some matrix
d#erential system &!$,(a, x) = NU(a, x) &(a, x) with initial value &(a, 0) = Z,
the identity matrix.
First we note that since S(A) = S(A) S(e, x), where e denotes the identity of V,
S(e, x) = I. More generally, S(T, , x) = I, where T, denotes the translation
x + x + b. Therefore (still denoting by h the multiplication by X E R+) we have
by (2.1) that S(& x + b) = S(X, x) and thus is independent of x, whence
S@, x) S(4 = W) m x >, since LI and X commute. Therefore S(A, x) = f(A)1
in the absolutely irreducible case (with real- or complex-valued f according to the
real or complex character of S) and S(A, x) = f,(h)Z + f2@)J in the remaining case.
If we setf(A) = fi(X) + if2(X) in the latter, we see from (2. 1) that we must have in
all cases the functional equation f(N) = f(A)f(x’), the only analytic nonzero
solution of which is f(A) = A-(n+ik) (n, k E R), whence
with initial value &(a, 0) = I, and an equation similar to (2.1), since we also have
J&z + a’, X) = J&z, A,z) $I@‘, x). One may check that the Frobenius integrabil-
ity condition
is satisfied. Moreover, since the matrices iV, are traceless and we have, from (2.6),
the integral being taken along any path from 0 to x (in MO), e.g., a path composed
of segments parallel to the coordinate axes (which enables us to define So as a
product of ordinary matrizers of the kind c:, (I + A$ dxl) with x,’ < x1 < x1”),
the Frobenius condition enabling us to pass from one path to another.
Moreover, we see that 2(u2xU + au) a,S, = (&.Q)(a,So) = 0, which means that
So is constant along lines “orthogonal” (for the hyperbolic metric) to the hyper-
surfaces Q = const.
The singular points for the transformation A, (and for the calculus developed
here) are on the surface J&z, x) = 0, i.e., the cone (x + u/$)~ = 0 if a2 f 0, and
the plane ax = - Q if u2 = 0. The cone splits A4 into three connected components
(including MO), while the plane gives only two. But if we consider the algebraic
surface B = 0 in the complex space M, (eight-dimensional on R), we see that it is
of real dimension six and its complement is here connected. This enables us to
define arbitrary powers of LI (and log a), as well as So , on all M - Ct from their
definition on MO (where Q > 0). We can do this by analytic continuation to a
connected (and simply connected) Riemann surface defined, e.g., via a cut in i&f,
which rests on the surface Ct : Q = 0, the choice of the cut fixing uniquely the
definition of these functions.
The solution So thus obtained will also depend analytically on the four-parameter
a, and a second-order calculation gives
S,(u, x) = I + 2d‘~‘~~~ + 4~“a”(.N~,~,~ - gm.Mve) x”xp + O(l a j3). (2.7)
CONFORMAL. COVARIANCE 85
THEOREM. For pure spin representations (0, j) @ (j, 0) and (j/2, j/2), the con-
formal degreen is related to the spin j via the relation
(-1)s” = (-I)@
if we impose the mild assumptionthat the extended S defined by (2.1) and (2.2) is
real in M - C, and has an analytic extension to complex Minkowski space.
Indeed (0, j) is the symmetrized tensor product @j(O, l/2) of 2 j times the basic
representation (0, l/2), and similarly the contragredient (j, 0) is @(l/2,0), while
(j, j) = (j, 0) @ (0, j). Equations (2.6) and the expression for N,, show us that a
similar decomposition holds for the general S(t, x), using those related with the
representations (4, 0) and <o, 4). If we then characterize the S = (0, 4) represen-
tation by the relation S*(A) all,!@) = f’l:u”, where u@ are Pauli matrices, then
one can see [12] that in MO one has
and therefore the unique analytic solution of (2.6) is given, for (0, j) @ (j, 0), by
&(a, x) = Q(u, x)-j [G>“j(Z + upx,&7”) 0 @“j(Z + ~~x,3W)*-~]. (2.8)
Whichever cut is chosen in M, , @I2 will be pure imaginary in any connected
component of M - Ct, where Q < 0. Comparing (2.8) with (2.3), and since the
86 FLATO, SIMON, AND STERNHEIMER
quantity between square brackets in (2.8) is real, we see that if we want S(u, X) to
be real in M - C1 we must have (- I)n-j E R. The same result is obtained in the
case (j/2, j/2), whence the theorem.
This implies that n - j is rational, with odd denominator. If we further require
that UN determinations of S (and not only one) be real in M - Ct , then n - j must
be an integer; under this (quite natural) assumption, we have n E 4Z and n is an
integer or half-integer together with the spin.
Remarks. (1) The reality of S for (j, j’) @ (j’, j) (j f j’) and the similar
construction of S, from the basic matrices show us that in this case also n E i&Z,
whence n E 4Z also for (j, j’), since the (real) matrices S of a scalar restriction are
the scalar restrictions of the complex matrices S.
(2) In the case of a tensor field, one form (which is unique up to equivalence)
of S, is determined by the coordinate transformations [lo]
so = .Q(a,x)(*-P)($)” (fg-)”
[Due to the definition of U, the generators give the opposite Lie algebra from that
obtained through the geometrical action of V by the Cartan calculus; the opposite
group representation U’ can be obtained by writing the group action on the right,
CONFORMAL COVARIANCE 87
and thus defining (U’(t) * v)(x) = Q(t, x)-” v&t), where x -+ xt through t E V:
the generators obtained are the opposite of (2.9) and satisfy the usual commutation
relations (cf., e.g., Ref. [12], in which S = --D); the final results of this part would
not be changed should the opposite group be represented.]
A long but straightforward calculation shows that, defining A, = exp(@A,), we
have in this representation
Now given any unitary representation U of %?on the Hilbert space H of states, we
define y’ = U(t-l)q~ and write the covariance condition for states and fields,
whence
and is, when limited to B, the exact formulation of the second Wightman axiom.
This justifies the definition of v’ and the covariance condition (2.12), which can
be considered as a preformulation of the axiom. [In the case where we write
on the right the group action on M: x + xt = x’, we should of course define
88 FLATO, SIMON, AND STERNHEIMER
4. Two remarks should be made here. First, one cannot have here,
owing to the restriction to 9, a pure Wightman theory in a straightforward manner,
since the spectrality condition cannot be obtained by the restriction from V to 9
(the spectrum of P2 cannot contain a positive mass without containing a continuum
of all positive masses). A more refined interpretation of global and local dilatations
can, however, be given (using the behaviour of the constant field P”) which gives a
reasonable physical interpretation to this mathematical disease.
Second, we notice that the change of variable in the integral (x to tx) used to
derive (2.13) is legitimate only when t is nonsingular. Moreover, in the right side
of (2.14) we have the factor SZ+; therefore, in order to obtain (2.13) we must check
that L+(t, x) a(x) is still a test function; the factor S,, will also give a similar condi-
tion (due to relations like (2.8)) with, eventually, other powers of L?. Therefore,
01 should vanish for Q(t, x) = 0 and relation (2.14) will only make sense when
applied to such test functions, or rather test forms, a(x) dx. (Since tx = co for
52 = 0, this condition is quite natural.) This may lead to some difficulties, especially
concerning the topology of the spaces of functions or distributions, if a global
formulation is desired.
5. The factor Q-” in (2.14) is related to the fact that our fields have dimen-
sion. To make its interpretation clearer, let us consider the (constant) squared
mass field P”(x) : a(x) -+ P2 s a(x) dx, where P2 = V(P,,Pu), V being some unitary
representation of V. We then have
where S,, is unimodular (and n E +Z, for real Lorentz representations at least).
The transformation law of ds2, given by dsf2 = JZ2 ds2, leads us to identify -n
with the sum of the physical dimensions, in centimetres and seconds, of the field.
Physical quantities should transform according to their physical dimension. We
shall thus postulate, e.g., that e2/fic is invariant under V?,that mass will transform
according to m + m’ = Q(t, x)m when x -+ x’ = tx, etc. The transformation law
of mass (similar to what we have seen in the quantized case) will moreover enable
us to give a sense to conformal “extended covariance” of the equations for massive
particles (and not only zero mass particles, as is usual).
We shall now investigate under which conditions a Lagrangian field equation is
conformally covariant. We shall restrict ourselves, in stating the theorem, to
Lagrangians Z(I,!I, a,#, m) depending on one field #, its first derivatives, and a
mass. The more general situation, in which we have several fields (and eventually
coupling constants), is a straightforward extension of our case (we must not forget
to transform the coupling constants according to their dimensions); this remark
should be kept in mind, especially when the conjugate If of a field occurs together
with the field $ itself in the Lagrangian.
LZ’(#‘, a,‘#‘, m’),, = Z(XF%,b, h-(n+l)aulCf, A-lm), = X-45?(~, a,$, m), . (3.2)
90 FLATO, SIMON, AND STERNHEIMER
Then the field equations derived from 8 will be conformally covariant, with
where
R, = ((w,v + ~JW a~/WP). (3.3)
Remarks. (1) For several fields &, we have to introduce their degrees n, and
sum over k to obtain R, ; in particular, 5 is associated with 4.
(2) We have seen that when x + x’ = (A, a)x, then the field must cotrans-
form according to #‘(x’) = Q,P(a, x) &(a, x) 4(x): (3.2) requires n = p, which we
assumed in the theorem.
To prove the theorem, we shall first consider only infinitesimal transformations
A = 1 + E, au intmitesimal, (1 = 1 + 4~’ A&, and thus S(/T) = I + &py-aiBy.
In infinitesimal form (first-order development), (3.1) and (3.2) give
the field cotransforms (to first order, and with the help of (2.7)) as
The field equations will thus be conformally covariant, with conformal degree n,
if and only if, given any domain A’ C M (on which the inverse transformation
x’ -+ x is nonsingular), when #’ varies so that S#’ = 0 on the boundary of A’, we
have S J,,, A?‘&‘) d?x’ = 0.
This variation of t,P corresponds to some similar variation of # on some domain
A C M. Now from (3.9, writing for simplicity 9, = Y(x’) for x’ = x, = A,x =
Q(a, x)-l (x + ux2), and denoting by T,(a) the operator
the (unique, and known a priozi) solution of which is given by the (operational)
multiplicative integral 9, = c (1 + T,(u) duU) 9, , with 2Z0 = P(x), and is
defined (e.g.) as
all these cases, n is found to be necessarily real (this difference from the case in which
only global dilatations occur, and the imaginary part of n is not restricted, is essen-
tially due to the fact that, contrary to A”, Sz” does not commute with derivatives).
(a) R, = 0.
(1) The Dirac equation (~“a, + m)# = 0. The Lagrangian is 9 = $(+l, + m)#
and the conformal degree n = 312.
(2) One may check that the continuity equation for currents a,Ju = 0 is con-
formally covariant if and only if n = 3. In particular, R, = 0 implies (under the
hypotheses of the theorem) conformal covariance of the Lagrangian, which in turn
implies n = 3 for the associated conserved currents.
(3) The Maxwell equations written auFuY = 0 and &‘“~auFw = 0 are con-
formally covariant, the field F@” having the conformal degree n = 2. The Lorentz
gauge condition on the vector potential will not be conformally covariant (because
of the last example) but can be replaced by a nonlinear covariant condition such
as B,(A,A”A~) = 0, the A,, having (as expected) the conformal degree n = 1.
(4). The first-order, zero-mass Weinberg equations 114, 151, that write
a~lF>Yl”‘pj”~ = 0 for integral spin j, and aa~@“‘~~j = 0 for half-integral spin j
(Dirac-Fierz equations). In both cases, we find [14] n = j + 1.
(b) R, = a,R f 0.
(5) The real field Klein-Gordon equation (!J + m”)# = 0. We have
These examples show that the necessary and sufficient condition obtained is
indeed more general than the sufficient conditions obtained earlier by Gross and
Wess [ 171 (R, = i3,R) or by Mack and Salam [18] (R, = 0).
We find [13] that they are conformally covariant, but with conformal degree
rr = 2 - j (this formula generalizes the results obtained for j = 4 and j = 1).
4. CONCLUSION
generated by scale and conformal generators are related (through their relation to
the trace of the symmetric energy-momentum tensor). We did not utilize this fact
in proving our general theorem, since we think that this is a quite general geo-
metrical theorem which is independent of the class of dynamical theories which is
treated.
7. We have already seen that in all cases that were treated by us it was
possible formally to extend the notion of conformal covariance from zero-mass
particles to massive particles, just by substituting (in classical theories) m’ = Szm,
a fact which corresponds to a space-time dependent choice of the scale of length
and time.
In second-quantized field theories the situation is very similar. We know that in
any representation of the conformal Lie algebra, e-aDPPeaD = e-2aP2. Moreover,
we have proved that for the particular representation of the conformal Lie algebra
given by (2.9), we have exp(--(a”& + X0)) P2 exp(u”d, + ho) = QP2Q. This
means that even for second-quantized theories with massive particles, we can in a
way generalize the concept of conformal covariance.
One should first remark that, at least as far as the particular representation
considered above is concerned, we can, without restricting generality, argue only
on the Weyl group instead of V, as the situations are analogous.
Now as for the Weyl group, the unitary irreducible representations are character-
ized by one value of the spin W2/P2 (W, is the usual Pauli-Lubanski vector) and
give a continuum (or zero only) for the spectrum of P2. If we say that each unitary
irreducible representation of the Weyl group characterizes Oozeparticle, the mass of
which is measured in all possible scales, then we can choose the “unit particle”
characterized by 01= 0, and other equivalent representations (transformed from
the former by the inner automorphism defined by e-uD) will describe other particles
with the same spin having mass “measured” by the value of e-OL.
96 FLATO, SIMON, AND STERNHEIMER
However, this interpretation is not very popular in the literature. In this connec-
tion, as well as for the problem of under what condition for the second-quantized
theory Weyl covariance will imply conformal covariance, it will be quite interesting
to find all those unitary irreducible continuous representations of the conformal
group which, when restricted to its Weyl subgroup, remain irreducible. Until now,
we have found,only a partial answer to this question (which includes most, if not all,
physically relevant cases). [Note added in proof: in a recent preprint, J. Mickelsson
and J. Niederle, from Prague, have answered this problem, suggested earlier by
one of us (M. F.)].
For zero-mass particles it was conjectured [19], and proved later [20], that the
most degenerate discrete series of unitary irreducible representations of the
conformal group (constructed, e.g., as ladder representations) remain irreducible
under the Poincare subgroup, thus characterizing a zero-mass particle. This last
fact is quite important for physical applications.
REFERENCES
595/61/1-7