t

PHYSICAL REVIEW VOLUM E i 86, NUM B ER 5 25 OCTO B ER 1969

Propagation and Quantization of Rarita-Schwinger Waves in an

External Electromagnetic Potentialt
GIORGIO VELO AND DANIEL ZWANZIGERt
llePorltnenl of 1%ysics, Sew York Uninersily, New York, iV ew York ZOOID
(Received 25 April 1969)

The Rarita-Schwinger equation in an external electromagnetic potential is shown to be equivalent to

a hyperbolic system of partial dif'ferential equations supplemented by initial conditions. The wave fronts
of the classical solutions are calculated and are found to propagate faster than light. Nevertheless, for
suKciently weak external potentials, a consistent quantum mechanics and quantum field theory may be
established. These, however, violate the postulates of special relativity.
.I

I. INTRODUCTION free retarded propagator is causal, so that every finite

order of perturbation theory is causal.
HE problem of finding a suitable wave equation
Nevertheless, we 6nd that in the weak-field case
for electrically charged higher-spin particles has
Lsee Eq. (2.16)j a positive definite conserved inner
been with us for a long time. Since the pioneering work
product exists in some, but not all, Lorentz frames. In
of Fierz and Pauli, ' the commonly accepted method,
these frames a consistent quantum mechanics and
which avoids algebraic inconsistencies, is to find
quantum field theory may be formulated. The equal-
Lagrangian equations of motion whose solutions
time anticommutator is not local in other frames.
correspond to free particles of unique mass and spin,
The main lesson to be drawn from our analysis is
and then to account for electromagnetic coupling by
that special relativity is not automatically satisfied
substituting
'Lclo ~
$8o+8Ao
by writing equations which transform covariantly. In
addition, the solutions must not propagate faster than
into the free Lagrangian. A familiar example of this light. There are simple algebraic criteria on the coeK-
method is the Rarita-Schwinger (RS) Lagrangian' for cients appearing in the partial differential equations
spin-2 particles, which does indeed avoid immediate which determine the velocity of propagation of the
algebraic inconsistencies. ' However, a more subtle type signals. These criteria must be applied to other higher-
of inconsistency appears when the RS field with an spin equations which describe interactions, as we have
external potential is quantized. done here for the RS equation. In this direction, we
In the present article, we show that the difficulty is have verified that the first-order DuKn-Kemmer-Petiau
already present in the RS equation interpreted as a formalism for spin 0 and 1 and the YVentzel formalism'
classical field equation, because the solutions propagate for spin 1 are causal, even though constraints are
at velocities exceeding the speed of light for arbitrarily present.
weak external fields. More precisely, we show that the
RS equation is equivalent to a system of hyperbolic 2. EQUATION OF MOTION AND
partial differential equations, supplemented by initial ITS WAVE FRONTS
conditions. Elementary methods then allow one to We begin with the RS Lagrangian density
determine the wave fronts and ray velocities of the
solutions to the hyperbolic system. One finds that the Z=lt (r —a)4. (2.1)
propagation of RS wave fronts in an external potential
resembles the propagation of light in an anisotropic Here/ is the RS vector-spinor P„, with g"=f"tj; and'
crystal. There is an ordinary ray which travels at the 8
speed of light and an extraordinary ray which always Ir„=i +eA„, (2 2)
travels in some direction at a speed exceeding that of Bx)"
light. The violation of causality occurs even though the
with A„a given classical four-vector potential. The
~ On leave
Bologna, Italy.
of absence from the Istituto di Fisica dell' Universita, matrices I'& and 8
are given by
t Research supported —by the National Science Foundation. (P.~) x ~5e Px~ ~
f Address during 1969 1970:Department of Physics, Weizmann
Institute, Rehovoth, Israel.
' M. Fierz and W. Pauli, Proc. Roy. Soc. (London) A173, 211 = g."y Ir —(p„sr"+sr„y")+p„y Iry", (2.3)
(1939).
' W. Rarita and J. Schwinger, Phys. Rev. 60, 61 (1941). B."= —mo. „"= SIS (g„"—y„y") . (2 4)
3For an alternative spin-$ wave equation see G. S. Guralnik
and T. W. B. Kibble, Phys. Rev. 1398, 712 (1965); 150, 1406(K) ' G. Wentzel, QNantIsns Theory of Fields (Wiley-Interscience,
(1966); S. N. Gupta and W. Repko, ilrid. 165, 1415 (1967). Inc. , New York, 1969), p. 90.
Other references may be found here. 6 Our conventions are k=. c=1, g&"=diag(i, —1, —1, -1),
4 K. Johnson and E. C. G. Sudarshan, Ann. Phys. (N. , Y.) 13, yl'y" +y"y&= 2g&", a= a„y&, and y'= y'y'y'y', so (7')'= —1, &""=1,
126 (1961). '(y&y" —
g&"= — p"y&), P = y, and 0.' = (y y') for = 1, 2, 3.
Z

1337
j.338 G. VELD AN D D. ZKAN Z I GE R

This Lagrangian is one of a class of possible Lagrangians So far we have established that every solution of the
for spin a which differ by the substitution f„~f„ RS equation satisfies the primary and secondary
+tsar„y P. Variation of the Lagrangian with respect to constraints (2.6) and (2.10) and the new equation of
the 16 components of ll and p independently yields the motion (2. 12) or (2.13). Conversely, as shown in
equations of motion Appendix A: (a) Equation (2.13) preserves the con-
straints (2.6) and (2.10) (i.e. , every solution of (2.13)
(I ~ —B) Q, =o which satisfies the constraints (2.6) and (2.10) at a
y (1.~ —B)„"=O. (2.5) given time satisfies them for all time), and (b) every
solution of Eq. (2.13) which satisfies the constraints
Let us analyze Eq. (2.5). Because has more compo- f (2.6) and (2. 10) at a given time is a solution of the
nents than needed to describe a spin-~ particle, some
of the 16 equations (2.5) will turn out to be constraint original RS equation. Thus, Eq. (2. 13) contains less
equations in the sense that they do not involve time information than the original RS equation because it
derivatives. In fact, from the form (2.3), we see that does not imply the constraints. However, it is a true
when «=0, Eq. (2.5) contains no time derivative, but equation of motion because it specifies the time deriva-
tive of P for any given ll.
yields, instead, the primary constraint equation
We will now show that, for sufficiently weak fields,
(se —htr) /=0, (2.6) Eq. (2.13) is a hyperbolic system of partial differential
equations. For this purpose it is sufficient to compute
where' ~= (srt) Q= Q ') i = 1, 2, 3, and
the normals e„ to the characteristic surfaces, ' which,
h= tr ac+Pm. (2 7) for a linear system of the form
Moreover, from the form (2.3) we see that (8/8t)P 8
never appears at all in Eq. (2.5); nor is determined by P (1st) x
Bx"
the primary constraint (2.6), which only involves P,
P, and lie. To obtain an equation for P, one must are determined by
differentiate Eq. (2.5). This may be done covariantly
by multiplying Eq. (2.5) successively by p" and sr", D(st) = (I' )„),ss„i = 0. (2.14)
i
which yields, respectively,

2(y sr' —sr) /+3m' /=0, (2 8)

This determinant is a polynomial in the components of
e„, so it is sufhcient to evaluate it for n„ in the future
m(y sr' sr) P—iey'y F"—
)=0, . (29) cone, and by Lorentz invariance we may take I„
where = (ss, 0,0,0). Taking the coefficient of it/ctt in Eq. (2.13)
Ii=Ii "=8 A" — 8"A one has after slight rearrangement
pd pd &~ vg p x
v

D(~) = (g„&o+', iem F„'-» -g, )

~

Comparing Eqs. (2.8) and (2.9), one finds the covariant

Xy'(go"y'+ ',iem 'g '7'y Fa"-)n
secondary constraint ~

= ttr4/1 —(-', em —')' B'$4

'm
, 'iey'y —
F—a
P, (2.10) or, in covariant form,
which determines P. Another useful relation is D(N) = (st')')I'+ (-'em ')'(F" ss)'$'=0. (2.15)
sr f= —(y sr+ 2m)aiem '7'y F-".f, (2.11)
This equation determines the normals to the character-
which follows upon inserting (2.10) into (2.9). istic surfaces passing through each point.
After these preliminaries, we proceed to the main Before analyzing Eq. (2.15), it is convenient to
subject of this section, which is to determine under introduce the term "weak-held case" to refer to the
which conditions a solution to the RS equation exists situation in which there exists, for each space-time
and to Qnd the velocity of propagation of signals. For point, a Lorentz frame such that the inequality
this purpose we substitute Eqs. (2.10) and (2.11) for
and sr f
back into the original RS equation. (-'em ')'8'(1 (2.16)
The resulting equation,
is satisfied, and otherwise we refer to the "strong-field
(y sr m)f„+(sr„+—
army„)aaiem 2P'y F" &=0, (2.12) "
case. To avoid inessential complications we will
can be put into Hermitian suppose in the weak-field case that there exists a single
form by again using Eqs.
(2.10) and (2.11): s R. Courant and D. Hilbert, Methods of Mathematical Physics

(y sr m)P„+(sr„+ 'my„) ',—

iem 'y'7 -F"
, P-— (Wiley-Interscience, Inc. , New York, 1962), Vol. 2, pp. 590, 596.

13).
If this criterion were applied directly to the original RS
+saiem 'F„e»'(sr+ 'mp) P+ , , 'iem-- equation, we would 6nd that every surface is a characteristic
surface, corresponding to the fact, which we know already, that
XF„" »'(y sr+2m) ',iem 'y'y F" /=0 -(2. there are constraints.
186 RARITA —SCHWINGER WAVES IN EI. ECTROMAGNE TI C POTE N TIAL 1339
common Lorentz frame such that (2.16) holds at every satis6es the equation of continuity
space-time point. It is easy to verify in the weak-field
case that Eq. (2.15) has eight positive and eight
negative roots rP, for any given n= (e*'). This establishes (3 2)
hyperbolicity of Eq. (2.13) in the weak-field case, and
allows the definition of "spacelike 'surfaces" and Hence, for solutions P that vanish su6iciently rapidly at
"future and past cones" with respect to Eq. (2.13).r infinity, one may define the scalar product
These di6'er, however, from the familiar spacelike
surfaces and light cones of special relativity. In the
strong-field case, Eq. (2.13) ceases to be hyperbolic and
is not suitable for the description of wave phenomena.
(« «) I

f pE.A«", (3.3)

Hence, we restrict our considerations to the weak-field which is conserved and independent of the surface of
case. integration Z. For a surface t= const, the inner product
Ke remind the reader that, for hyperbolic equations, becomes6
the maximum velocity of propagation of signals is the
slope of the characteristic surfaces. Significantly, the (y, ~y,) = y, ~&or,y, ex= y, ~ay, dx
characteristic surfaces determined by Eq. (2.15) are
not all tangent to the light cone and, catastrophically,
spacelike characteristic surfaces pass through every = Ee" e.-( e)'( e.)j&x, (3.4)
point where F„„is nonvanishing. Consequently, signals
are propagated at velocities greater than the speed of
light. To see this, we show that there are timelike where f"= (P, g) and Ao=yol'0. Since the zeroth
normals n„satisfying (2.15). In fact, choosing component of f& does not appear in the inner product,
= (1,0,0, 0), the second factor of Eq. (2.15) becomes we retain only the space components and use nonrelativ-

—(-'em-2)2B2=0
istic notation from now on. Of course, P
may always be
1 (2.17) retrieved from the constraint (2.10) when the weak-Geld
assumption (2.16) holds:
and whenever F„,/0, there exists a Lorentz frame where
(2.17) holds. (I+~em 2e B)p=e Q+xaem '(ty'B —o»(E) g. (3.5)~

Ke remark parenthetically that the propagation of

RS waves, according to (2.15), resembles light propaga- The inner product (3.4) allows a quantum-mechanical
tion in an anisotropic medium. There are ordinary rays interpretation in terms of positive probabilities only if
corresponding to the first factor of Eq. (2.15), and
it is proved to be positive de6nite. Because e 0. =3,
this form is in fact inde6nite. However, only those
extraordinary rays with wave-front velocity exceeding
functions g are allowed that satisfy the constraint
c corresponding to the second factor. It might be hoped
that the constraints (2.6) and (2.10) eliminate the (2.6), implied by the RS equation
extraordinary ray, but we prove the opposite in v /=0, (3 6)
Appendix 3, so it is true that wave fronts do indeed with
travel at the extraordinary-ray velocity which exceeds c. (3.7)
As a Anal remark, we observe that the hyperbolic
system (2.13) possesses "spacelike surfaces" Z on which Thus, positivity need be established only on the
initial values f(Z) may be specified arbitrarily. Although subspace delned by the projector
these spacelike surfaces differ from those of special
relativity, nevertheless, in Lorentz frames where Kq. Q;;=8;;—v;t(v vt) —'v;. (3.8)
(2.16) holds everywhere, the surfaces t= to are "space- LNote that v vt= (n m)'+m'+3m' is a positive invert-
like" and the Cauchy initial-value problem may be
posed for any it (to). We will see in the next section
]
ible operator. With the help of Q, the inner product
(3.4) becomes
that, in such frames, a conventional quantum-mechan-
ical interpretation of the RS equation can be formulated. 0 I«& ft'& A'& 0=« (3 9)
with
3. CONSERVED POSITIVE DEFINITE V= ~ (3.10)
sg ~ ~
INNER PRODUCT
In Appendix C, the positivity of Q A Q is inves-
An important feature of the RS equation is the
tigated, and it is found that Q Ao Q is in fact positive
existence of a conserved current or bilinear form. Let
in those Lorentz frames where inequality (2.16) holds.
Pi and P2 be solutions of (2.5); then the four-vector It is indefinite, however, in other frames. 4
density
YVe now have all the elements for a quantum-mechan-
g„(*)=it, (&)r„P,(&) =~t, (&)(I'„)„Q„(&) (3.1) ical interpretation of the RS equation. The inner
1340 G. UELO AN D D. ZKANZI GE R
product is defined by Eq. (3.4) and the equation of equality holds:
motion is provided by (2.13), with wave function P
restricted by the conserved subsidiary conditions (2.6)
and (2.10). However, we recall that this holds only f'(t, x)A'g (t, x)dx
when the weak condition is satisfied and only for
surfaces t= to which are "spacelike" for Kq. (2.13). ft (t, x)A'D(t, x; t, x')A'g(t, x)dxdx', (4.6)
4. QUANTIZED FIELD FORMULATION
where we have set t'= t in (4.5). Because of the form of
—
Because the RS equation is linear corresponding to A', LEqs. (3.4) and (3.10)j, only space components
the fact that charged particles interact only with a appear in Eq. (4.6) and, if we define the space-space
given classical vertex potential, and not with each
—
other the quantum mechanics established in the last
equal-time anticommutator
sections may be easily formulated in terms of a quantum Cg (t, x, x') = g(t, x; t, x') j,, , (4.7)
field. I.et to be an early time before the onset of the
C

potential, and let then we can write, in nonrelativistic notation,

u;(to, x), u,'(tp, x), (4.1)

satisfying ft(t, x) A'g(t, x)dx
( p+Pm)u;(to) = (p'+m')'"u, (to)
ft(t, x).A'C(t, x, x') A'g(t, x')dxdx'. (4.8)
(a p+Pm)u, '(to) = —(p'+m')'"u, '(to),
be a'set pf functions, subject to the constraints (2.6) Because f and g are arbitrary in the subspace defined by
and (2.10), which is complete and orthonormal on the Q t Eq. (3.&) ], and because, by virtue of the expansion
surfac'e t=t, with respect to the scalar product (3.4). (4 4)
For the initial values u;(to, x) and u (tp, x), the equation ~ ~
(4.9)
pf mptipn (2.13) determines corresponding solutions we have
u, (t, x), u (t, x) at all space-time points. Let a, ' and
b " be Fermi annihilation operators satisfying the usual Q. A'Q= (Q A'Q)C (Q A'Q) (410)
anticommutation relations. Then the fields defined by
In Appendix C, the inverse of Q A'Q is calculated,
p(t, x) =P(a u, (t, x)+b tu'(t, x) j (4.2a) (4.10), we conclude that the equal-time
anticommutator coincides with this inverse. Therefore,
we have
yt(t, x) =Pea 'u (t, x)+b "u't(t, x)) (4 2b) C= (Q A'Q)-', (411)
or explicitly
satisfy the RS equation everywhere, and, for early
times, coincide with the canonical free in-delds. Cv= ~v —2~;~, +-'. m- (2~;+n;pm)
Having delned the fields by an asymptotic condition, X(1+3em-'e B)- (2~, + mp~, ).
we will now compute the equal-time anticommutator.
%'e begin by considering the anticommutator for This expression coincides with that pf Jphnspn
arbitrary times, Sudarshan, 4 who obtained it using Schwinger's canonical
quantization principle. They observed that it is an
~(*p ) = 8 (~), 4'(*')), (4.3) indefinite operator in some I.orentz frames, although
it should be positive by virtue of its form (4.3) and (4.4).
which by virtue of the expansion (4.2) takes the form However, our derivation and our analysis of Sec. 2
~(*,")=PL, (~),t(")+, (*)u,'"(*') j. (4.4)
show that expression (4.12) represents the equal-time
anticommutator only on surfaces which are "spacelike"
with respect to Kq. (2.13). For other surfaces the equal-
As usual, for any solution f(t, x) of the RS equation, we time anticommutator can only be obtained by solving
have the identity the equations of motion and will be nonlocal because
of the faster-than-light propagation of disturbances.
The foregoing holds in the weak-field situation, namely,
f(t, x) = 5(t, x; t', x')A'f(t', x')dx', (4.3)
when there are Lorentz frames in which (2.16) holds.
In the strong-field situation, Eq. (2.13) ceases to be
which holds for any t', with A' given by Eq. (3.4). hyperbolic and no quantum theory can be constructed
Consequently, for any solutions f
and g, the following at all.
186 RARITA —SCH WI NGE R WAVES IN ELECTROMAGNETIC POTENTIAL 1341
ACKNOWLEDGMENT Taking the discontinuity of the primary constraint
(2.6}, we have
One of the authors (G.V.) would like to thank
Professor Bruno Zumino for his kind hospitality at the
I W —Inc W=O
or
Department of Physics of New York University. e. ng w —0 (B3)
APPENDIX A Taking the three-gradient of the secondary constraint
(2.10), and equating discontinuities, we find that w
We prove that the constraints (2.6), must satisfy
= (m —hn) . /=0,
X— y w+ 32iem -'y F—d w=0
and (2.10), The discontinuity of the equation of motion (2.13)
~v'7 F" 4=0,
v0+— yields
with X=~3em 'i, are preserved by the equations of nw„+m„3iem 'y'y Fe w+3i.em 'F„e yy'ir .m
motion (2.13), and that the RS equation is satisfied.
Since Eq. (2.13) is of first order, it is sufFicient to show ,iem '-F " yy'n ,iem 'y-'7 F" Ye=. 0
+ (B5)
that if X and 0 vanish at t=0, their time derivatives By choosing w„=e„f, Eq. (B3) is automatically satis-
X' and 0' vanish at t= 0. Note first that Eq. (2.13) may fied, and Eq. (B4) becomes an eigenvalue equation for
be written f which has a nonzero solution when "Eq. (B2) is
satisfied, which is true by assumption. . Then' Eq. (B5)
q„+[m( q„+ ,'l—
iF„" q-q5)+ (~„+XF„"qq'q ~)]n
is satisfied identically.
—m-'[~ + XF '~~5]~ ~=0, (Ai) This shows that there are disturbances', compatible
with with the constraints, which do propagate along the
v' = (I" —&) 4'
— extraordinary ray.
Consider the two equations obtained by contracting
Eq. (A1) with y& and also by taking the zeroth compo- APPENDIX C
nent of Eq. (A1). Noting that X= — y'yo, and assuming We first verify that C, given by Eq. (4.12), satisfies
that X and 0 and their spatial derivatives vanish at Q. C= C and C= (Q. A'- Q) '. The positivity of
t=0, one finds that Q A .Q will follow from the positivity of C. From
0'=0, m- q =0 (A2) the identity

at t=0. Comparing with Eq. (A1), we see that p;=0, (hn —m) ~
(2m+nPm) =3m'+2en B, (Ci}
so the RS equation is satisfied at t=0. From X= y yo — one easily verifies that v;C,';=0, with e; given by Kq.
and the second of Eqs. (A2), we deduce that X'= 0 also. (3.7). Hence, with Q given by Eq. (3.8), one verifies
Hence, the constraints are preserved in time and our the equality Q C=C. Then we have, with A' from
proof has shown that whenever the constraints are Eq. (3.10),
satisfied, the RS equation holds.
~ ~
~ ~ ~ ~
QQ ~

APPENDIX B =[I—vt(v vt) 'v]

The discontinuities of the derivatives of a solution [I—v" (3m'+en B)—'(2m+ mpn)]
to a erst-order system of hyperbolic equations are =[I—vt(v vt) 'v]=Q. (C2)
known to propagate along characteristic surfaces. ' We This shows that C is the inverse of Q A'Q in the
will show that discontinuities which propagate along subspace Q. It is sufficient to show that C is positive on
the extraordinary ray are compatible with the con- the subspace Q, and for this purpose we may replace ~
straints (2.6) and (2.10). For this purpose, we assume by eh and he on the left- and right-hand sides, respec-
that the external potentials and fields are continuously tively. This gives
I
differentiable functions. Let be a continuous solution
C=Q {1+-, 'n{(2n. m+3Pm)[(3m'+2en —'
of Eq. (2.13) whose derivative has a discontinuity given B)
by —(2n ~+3Pm) 2](2n ~+3Pm)}n} Q.
= e„w,
(m-„u)
We assume that 3m'+ 2en 8)
0. (weak;field condition),
where (f) means the discontinuity iri f, w is continuous, and therefore from the fact that
and e„satisfies
(2n. ~+3Pm)') (3m2+2en B),
~~+-,'em-2(F' ~)2= 0. (B2} it follows that C)Q. Hence, C is positive definite
on
9 Reference 7, pp. 618 and 619. the subspace.