PH YSI CA L BE VI E% D VOLUME 4, NUMBER 12 15 DE C EMBER 1971

Hamiltonian Theory of a Relativistic Perfect Fluid*

Bernard F. Schutz, Jr. f
California Institute of Technology, Pasadena, California 91109
(Received 4 August 1971)

The velocity-potential version of the hydrodynamics of a relativistic perfect fluid is put in-
to Hamiltonian form by applying Dirac' method to the version% degenerate Lagrangian.
There is only one independent momentum, and the Hamiltonian density is -To (— g )- ~ . The
Einstein equations for a perfect fluid are then put into Hamiltonian form by analog with Arno-
witt, Deser, and Misner's vacuum Einstein equations. The Hamiltonian density splits into
two pieces, which are the coordinate densities of energy and momentum of the fluid relative
to an observer at rest on the hypersurface of constant coordinate time.

INTRODUCTION which is just the normalization constraint on the

four -velocity.
The velocity-potential version of perfect-fluid The dynamical field equations are five coupled
hydrodynamics as formulated by Seliger and nonlinear first-order equations,
Whitham, ' generalized to relativity by Schutz, and '
independently discovered by Schmid, ' can be re-
U'Q, = -p, (5a)
garded as a nonlinear relativistic field theory for Uv~ 0 (Sb)
five coupled scalar fields, whose Lagrangian den-
sity is simply the pressure of the fluid. The the-
U'P, =0, (5c)
ory is degenerate: Not all the generalized momen- U'0, =T, (5d)
ta are independent, so they cannot be solved for
the generalized velocities. In this paper we use
U'S, =0, (5e)
Dirac's4 algorithm for degenerate theories to cast (where T is the temperature) plus one nonlinear
the equations of perfect-fluid hydrodynamics into second-order equation,
Hamiltonian form, whose Hamiltonian density is
the energy density of the fluid. We then match the ( p, U'). „=0.
theory to the Arnowitt-Deser-Misner'6 (hereafter
referred to as ADM) canonical theory for the vac- There are really only two independent equations
uum gravitational field. among the three Eqs. (5a), (5c), and (5e) because
The independent variables of the theory are the of Eq. (4), so that there are five independent equa-
velocity potentials: five scalar fields (It), a, P, 6, tions altogether.
and S. Here $ is the entropy per baryon, while the
These equations follow from extremizing the ac-
others have less obvious interpretations. ' ~') The tion,
fluid's four-velocity is a combination of the poten- I= PV-g d4x. (7)
tials and their gradients, '
First-order changes in p are computed from the
U„= p '(p„+np, + 8S„), equation
where p. is the specific enthalpy of the fluid, 5P = p, 5p —p, TOS,
p =(p+p)&p. . which expresses the first law of thermodynamics.
(Here p, is the rest-mass density, p is the density Equation (4) is used to obtain 5p in terms of the
of total mass-energy, and p is the pressure. ) independent variations of P, n, P, 6, and S.
When one formulates these equations in terms of
Through the equation of state,
a Hamiltonian, one singles out the time coordinate
P =P(p, S), (3) for special attention, thereby destroying the equa-
all thermodynamic quantities are expressed in tions' four-dimensional symmetry. In what follows
terms of S (one of the velocity potentials) and p. we will therefore use the ADM notation appropriate
In its turn, p, is a function of all the velocity po- to such a (3+1)-dimensional split of spacetime:
tentials through the equation The four-dimensional metric 4g & is replaced by
the three-dimensional metric g, , =~g, , (whose
= -g '(P, +nP, + 8S ~)(P „+nP, + HS „), inverse is g" e4g"), by the lapse function N
(4) =(-'g o) '~', and by the shift functions N, = g„.

3559
3560 BE RNARD F. SCH UTZ, JR.
Derivatives covariant with respect to g, , are de- only one independent momentum is not clear. It
noted by V, or by a subscripted slash (e.g. , h„.a). would be a mistake to conclude that a perfect fluid
Dots (e.g. , h, , ) denote partial derivatives in time. has only one dynamical "degree of freedom": that
The action (7} becomes such constraints as zero viscosity and conserva-
tion of entropy have wiped out the other degrees
I= pN ~ d'xdt,
of freedom. The relationship between independent
so the Lagrangian density of the fluid is L = pNg'~' momenta and degrees of freedom is not well under-
In all but the last section of this paper, we will stood. In the velocity-potential representation one
treat the metric 4g 8 as a constant, not as part of must specify six independent functions on an initial
the dynamics of the fluid. It will suffice until then Cauchy hypersurface in order to determine the
to take as the fluid Lagrangian density '
future evolution of the perfect fluid. This indi-
cates the existence of three dynamical degrees of
L =pN,
freedom.
so that the action can be written in the standard What seems to be the case here is that two of the
way, three second-order dynamical equations (one for
I= Ld three-volume each component of velocity} have been replaced by
dt. four first-order equations [the four independent
CONSTRAINTS ON THE MOMENTA
equations among Eqs. (5}]. Hidden among the four
potentials o. , P, 8, and S are two dynamical vari-
Let q, stand for the five fields p, n, P, 8, and ables and their momenta. Since all four are treat-
S. The momenta conjugate to q, are ed as coordinates here, they appear to have no
independent momenta among them.
'd L ~pN
There are some tantalizing suggestions that this
Bq Bq
may be just the hint of a deeper canonical relation-
They are explicitly ship among the potentials. Seliger and Whitham'
show that one can modify the formalism slightly
p@ = -poU N, and introduce a function X such that da/d7 = 83C/BP
po'. 0 and dP/d7= -MC/Sn. Moreover, Schmid3&'~ points
pe (10) out that P obeys the relativistic Hamilton-Jacobi
P'= nP~
equation
P'= @
A, +%8+A =, p
Since only one momentum is independent, there
where
are four constraints on the momenta (the Dirac
rp equations), e' =g a(oP „+BS „)(nP 8 + 6S 8)
is positive definite because the vector nP „+8S„
is spacelike (it is orthogonal to U ). We have
cp =P =0, nothing more to add to these considerations here,
~3=p -&p = 0, so we return to the Dirac method.
y =p' —ep~=p THE HAMILTONIAN AND THE EQUATIONS
OF MOTION
There are no arbitrary functions of time in vel-
ocity-potential hydrodynamics: What gauge free-
The Hamiltonian density is defined in the conven-
dom exists lies only in the choice of initial values
tional way,
for the potentials. Consequently, we do not expect
any of these cp's to be first-class: None of them H =Qp'q, —L (12a)
has a vanishing Poisson bra, cket [see Eq. (16)]
with all the others. =p~(p+ap+ GS) -pN (12b)
That there is only one independent momentum is
surprising. One might expect at least three (for
=-T ON. (12c)
the spatial components of velocity), if not more. Although p, p, and S appear explicitly in H, we
The mathematical reason seems to be that, of all still have (8H/aq, )~, = 0, so that we can differenti-
the field equations, only Eq. (6) is second order in ate H with respect to p' and q, while holding q,
time derivatives. Equation (6) is obtained by vary- constant.
ing in the Lagrangian, and p@ is the only inde-
fII} Because of the cp equations one cannot solve for
pendent momentum. all the q, 's in terms of p''s. Instead, one intro-
The physical reason (if one exists) that there is duces additional variables A, (which Dirac' calls
 HAMILTONIAN THEORY OF A RELATIVISTIC PERFECT FLUID 3561

u, } in place of the q, 's. If one varies Eq. (12a) with tonian variables now are p Xy A2 X3 AQ p 0 p,
respect to q. and p', the A.,'s serve as Lagrange (9, S.
multipliers which ensure that variations in the The power of the Dirac approach is that the
q, 's and p"s maintain the y equations. Then one Poisson-bracket version of Hamilton's equations,
gets q=[q, H],

qa
~H
+A. m (13) j=[P,H],
gapa gapa
can easily be generalized to the degenerate case.
~L H Before applying this to fluids, however, we must
+X (14) define a Poisson bracket for fields in a curved
Bq Bq Bq
three-dimensional space. The conventional defin-
(A sum on m from 1 to 4 is implied here and ition from particle dynamics,
throughout. ) For the perfect fluid, Eqs. (13) can
be solved for the A. 's to give

is not sufficient when A and B are functions of the

, = (j) = 9, ~3= p, &~=S. (15) spatial derivatives of the fields q, and p'. In the
A. A.
2

Appendix we generalize this definition to fields.

Thus in this case the A. 's are self-consistent: For the perfect fluid (five scalar fields} the result
Equation (14) implies nothing new. So the Hamil- 1s

&A 5B BA 5B eA & 5B BA
. ",", .'" "-" '", '
& 5B
. ..
BA & 5B BA
„,
8 5B
.
(16)
)

where A and B are any functions of q„p', and y brackets. Equation (18) can be stated concisely
their spatial derivatives (of any order), and where as
5B/6q, is the spatial variational derivative
5B aB
f =[f,H')+— (19)
8B aB
+V V; ~ ~ ~

by defining a generalized Hamiltonian,

In the Poisson bracket all q's and p's are treated
as if they were independent: The y equations are O'=H+A. (p (20)
used only after the Poisson bracket has been com- Because the y's are all zero, H' is numerically
puted. equal to H.
Dirac' shows that the time derivative of any func- The equations of motion are a special case of
tion f of the q's, p's, and their spatial derivatives Eq. (19),
(and possibly explicitly of time) can be expressed
in the form P =(p. »')„,
f =[f,H]+[f, A. y ]+- P, q
(18) p = -poU'p, N,
p = ( poU'oN)u, (21)
In the second term, one is to regard A. as inde-
pendent of q, and p', but dependent upon position.
For example, one contribution
from a term like
to that term will be
p' = -p, U" S,N,
p' = (poU'~&}i~ po&& -~
The first is the continuity equation, Eq. (6). Upon
a~ ~ 8~ @
~a +' ' ' application of the y equations and the continuity
gapa gapa gapa equation, we see that the remaining four equations
8 8 Pm are the four independent velocity-potential equa-
m
m
~~
5 m
gp a ]c tions among Eqs. (5}.
gapa
One must also demand that the y equations be
In Eq. (18) one must treat H a, s a function only of
maintained in time, i.e. , that
the independent momenta [cf. Eq. (12b)]. Contribu-
f
tions to from the other momenta come from the [y, H') =0. (22}
3562 BE RNARD F. SCH UTZ, JR.
These equations are just the four independent vel- R' independent of N and N, Dirac' shows that this .
ocity-potential equations, Eqs. (5): There are no will also be true of the Hamiltonian density for
Dirac g equations; i.e., there are no equations any field. In our case, we split up H'g /' in two
from Eq. (22) that involve p's and q's without X's steps: (i) Differentiate with respect to 4g„, while
or q's, which would thus be constraints like the holding p, and q, constant,
cp equations. For example, y, —the constraint on
P" —is preserved at zero by the equation U'p „=0, aP( ~g } 1 4 )1/2

¹,
TOv(
which is obtained from the original variational gP Ij gtf I/

principle by varying n. This equation can be re- and (ii) convert derivatives with respect to g„,
arranged to read to derivatives with respect to N, g j with the
formulas given by Schutz. We obtain '
~2=N'ii, ' g"i3,;(4., +oP, , + eS, ).
;+ P0N O,
(23} gH I +/2
g1/2N(T oi +Ni T 00)
BN;
This is not really solved for A, in terms of p~ and gll2N giJ To (28)
q, because p0 and p on the right-hand side implicit-
ly depend on all the A. 's. Nevertheless, all the A. 's =g p g"( p ii+np, +&S, ), (3o)
do have unique solutions (through the velocity-po- Rnd
tential equations) in terms of p0 and q, . This gH i~/2 1/2N 2T 00
means that there are no first-class y equations &N
(as we guessed earlier) and no arbitrary functions g1/2N 2
4+ 0 T 0 (31)
of time in the solutions. i
H l(2 N; BH'g
(32)
COUPLING TO GRAVITY N N aN,.
Equation (32) implies
Until now we have treated the metric tensor g„z
as a constant because we were interested in the
canonical theory of the fluid. The fluid is, how-
H'=N
aH'
+¹ $
aH'
+A. y (33)
ever, coupled to the gravitational field, and one Since s(H'g'/')/BN; is manifestly independent of N
ought to treat the full dynamical system, fluid and N, , differentiation of Eq. (33) shows that
plus field. &(H'g'/ )/SN is also independent of N and N,
The Hamiltonian density of the free gravitational The two pieces of H' have straightforward physi-
field is" cal interpretations, as is shown by Schutz. Let '
Hc ——NR +NiR (24) g = -N'go™be the unit normal to the spacelike
hypersurface. Then the two pieces of H'g'i' are
with'
aH'g'i2
R' = -2m" Ij &
(25) (34)
R' = g'/ [ R2+2g '(-2'-4/' —ii" ii;, )], (26) Rnd
r l//2
Rnd gH
J~ ~aj =g
BN;
&11
g 4PO gmn)gik i
Ng1/2(4+0 (27} 1

They are, respectively, the coordinate densities

Here 'R is the scalar curvature of the hypersur-
face, Rnd g'J ls the momentum canonical to gcj
Since the Lagrangian density of the fluid, pNg' ',
does not depend upon time derivatives of the met-
ric, the full Hamiltonian is
Hc + 16''g'i
Note that Hc splits into two pieces, with R0 and
of energy and momentum measured by an observer
at rest in the hypersurface.
By analogy with Eq. (16) we may define a general
Poisson bracket for any two functions of m", g; j,
q„P', and their spatial derivatives (but not of N
(28) or N;, which are arbitrary functions that contain
coordinate information but have no dynamical
content),

sg /i2
' 6&" »" i2

~A. 6B aA 5B aA gB a aA e
+ (36)
q. &p' &p' &q, q, , &x' Op' ep', ax' Oq.
 HAMILTONIAN THEORY OF
Then the time derivative of any such function that
does not depend explicitly on time is
A =[A, X] (37)
=[A, NR +16vNh]+[A, NR +16vN;6"]+[A, X (p
(38)
In particular, the ADM form of the Einstein field
equations follows by using g„. and z" for A,
g„=g,, (vac), (39)
v" = ir"(vac) +8m Ng't'(T"
, N'N-'T") (40)
where (vac) indicates the terms that are there in
the vacuum case (see ADM). These must be sup-
plemented by the constraint equations that come
from varying the Lagrangian density (2 = -3C
+ v"g;, + Q, p'j, ) with respect to N and N, (which
are not Hamiltonian variables),
g'+16m@ =0, (41)
g'+ 16m 6" = 0. (42)
Equations (39)-(42) are identical to those derived
by Schutz' for a general stress-energy tensor.
The constraints, Eqs. (41) and (42), must be
maintained in time; i.e., we must have
[Ro+ 16vg, 3C] = 0 (43)
and
[R'+16v6", K] =0. (44)
In the vacuum case these are the Bianchi identities.
In our case the Bianchi identities reduce these to
the equations of motion, T"'. =0. These four
, Let us suppose that the field variable is a vec-
equations can be used to replace the four indepen-
dent velocity-potential equations among Eqs. (5), 'o
which themselves guaranteed the maintenance of
the y equations. Therefore, the full canonical set
of equations is
(45a. )
(45b)
(45c)
(45d)
with either the constraints (41) and (42) (main-
tained in time) or the y equations (11) (also main-
tained in time).
CONCLUDING REMARKS
The direction of any further analysis of these
equations must depend upon the application they
are intended for. It would in principle be possible
to reduce the 12 gravitational variables (w" and
g;, ) to four in exactly the same manner as ADM.
Solving the constraint equations would then involve
A RELATIVISTIC PERFECT FLUID 3563
the fluid variables Q and p @, but the coordinate
conditions would be unaltered (as was pointed out
by ADM).
Methods very similar to these have been used by
]. the author to derive the Hamiltonian density and
from it a conserved energy density for the pulsa-
tions of and gravitational radiation from a differ-
entially rotating relativistic star. These results
will be published elsewhere.
ACKNOWLEDGMENTS
I would like to thank Vincent Moncrief for guid-
ing me through Dirac's theory. I am greatly in-
debted to my many conversations with Yavuz
Nutku, whose assistance was invaluable in prepar-
ing this paper.
APPENDIX: POISSON BRACKETS FOR
FIELDS IN CURVED SPACES
For a system with n degrees of freedom, the
Poisson bracket (P.b. ) of two functions of p' and
q, is
(46)
A classical field has an infinite number of degrees
of freedom, one (or more} for each point in space.
Functions like A and B may be functions not only
of the fields p' and q. , but also of their spatial
derivatives. In this case, a simple definition like
Eq. (46) above is not sufficient.
tor field q; with canonical momentum p' — = BL/8 . j,
Our results can be extended in a straightforward
manner to cases where the field is a higher-rank
tensor or a scalar.
Because the field variables at different points
are independent, we wish the P.b. of two functions
to be nonzero only if they are evaluated at the
same point. Accordingly we define the canonical
P.b. 's,
[q, (x), q, (x')] =[p'(x), p'(x')] =0, (47a)
[q, (x), p'(x'}] = -[p'(x'}, q, (x}]
= -0';6'(x —x') . (47b)
Here 0', is the derivative of Synge's world func-
tion" Q(x, x'} with respect to x' and x", with the
index j'raised by the metric at x'. Because of
the 5 function the only properties of 0',
. that we
will need are" (1) its limit as x' approaches x,
(48a)
and (2) the same limit of its covariant derivatives,
3564 BERNARD F. SCH UTZ, JR.
lim Vk Q', = —lim Vk Q';, (48b) to covariant differentiation,

where is a covariant derivative at x' and acts

Vk.
V, [n',. 5'(x —x')] = -V, ,[n',. 5'(x —x')] . (50)
only on primed indices, and vice versa for Vk. We define the differentiated canonical P.b. 's,
The 5 function is normalized to proper volume,
[q;(x},V), p'(x')] = -V~, [n', 53(x —x')], (51a)
5' y 'd'y=1, (49a)
[V,q,.(x},p'(x'}] = -V [n",. 5'(x —x'}], (51b)

and has the usual property and so on for higher derivatives. The Poisson
bracket [,
]is the bilinear antisymmetric tu)o-
point differential operator whose domain is all
&x'
5'(x —x') = — . 5'(x —x') . (49b) C' functions of p', q;, and their covariant deriva
tives and which obeys relations (47) and (51).
Equation (48b) permits us to generalize Eq. (49b} By application of the chain rule we find

[A(x), B(x')] =
BA, i
, ~B. (x')
(x)[q, (x), p'(x')],

+,&A ;
(x)[p'(I), ,
q, (x')]
dB
(x'), +
&A , ,
(x)[q,.(, (X), p'(x')],
dB
. (x') + ~ ~
~, (52)

(x), (x')V, [n', 5'(x -X')]+, (x) (x')V, [n', 53(x -x')]

~A
eq,.
dB,
(x),
ap']k
,'
(x')V,' . [n';5'(x
A
—x')]+, 3, 'dB
ep' (x) eq„, (x')V, [n', .5'(x —x')]+ ~ ~
(52)

This is the usual definition of a Poisson bracket neither A nor B depends upon derivatives of q,.
in classical field theories. But for the purpose of and p'.When such derivatives are involved, the
practical calculations it is useful to obtain a one- left-integrated P.b. is the P.b. of A at the point x
point P.b. by integrating. The left- (ri ght-) inte- with the entire field B: Values of B at other points
grated P. b. is the integral of the P. b. over all x' influence the bracket through the spatial deriva-
(x). We denote these by «f, ] and ]„, respec-
tively. Integrating Eq. (53) on x' and using Eq.
[, tives of B at x. Note also that the integrated
P.b. 's are independent of any coordinate system.
(50} gives The following interesting properties follow di-
rectly from the definition of the integrated brack-
;[A, B] = f (A(it), B(x')](g')'~'d'x' (54) ets:
aA gB aA gB 1. -„[A, B] =
[B,A]„-. - (57a)
p p qi 2. «[A, B] = -«[B,A] (57b)
~A 5B &A 5B
qi]k k~P kg qi
i ~ i
P (k if and only if both A and B are independent of de-
(55) rivatives of q; and P'.
where 5B/bq, is the variational
6B BB — BB
. derivative
3. „A, B g'~ dx= A, B„~dx
&B
&q ~q
V),
~q ]k
+ Vi' V))
~q. (k[
—' . (56) (57c)

Although one cannot generally integrate a tensor

if «[A, B] is a scalar (if not, the integrals are un-
over a curved space, as we have done in Eq. (54), defined) .
in this case the 5 function limits the integration to 4. V, -„[A, B]= „-[V,.A, B], (57d)
only one point, so that the integral is unambiguous.
This integrated P.b. is the generalization of the v, [A, B]„-=[A,v, B]„. (57e)
simple P.b. , Eq. (46), to which it reduces when The integrated brackets fit into the Hamiltonian
 HAMILTONIAN THEORY OF A RELATIVISTIC PERFECT FLUID 3565

theory because the canonical equations are (for a For a degenerate field theory these become
system whose momenta are all independent)
q, =„[q,, H]+ „[q;, X y ], (65a)
q; = 5H j6p', (58a)
p' =:[p*,H]+;[p', &.q. ]. (65b)
p'= 6H/6q; . (58b)
In these equations A. appears inside the inte-
They translate to (from now on we will use only grated bracket because it is generally a function
the left-integrated brackets) of position. To compute a bracket that has A. in- „
side, one treats A. as a function of x independent
q, = „-[q,, H], (59a)
of p' and q;. For example, the variational deriva-
p'= -„[p', H]. (59b) tive of Eq. (56) is
By property 4 above these imply ~~m'Pm a Xm(j('m ~mV'm
li7 + ~ ~ ~

&q.
q;)„= "„[q;)„H], (60a)
P Ik x[p Ik ~ H] ~ (60b) (66)
which in turn imply
Conservation laws for the Hamiltonian can be
A = )[A, H] (61) derived here too. They are especially simple in
the case where H depends on no derivatives of p'
for any function A (not necessarily a scalar) of
and only first derivatives of q;, and where cp is
q;, P', and their spatial derivatives that does not
explicitly depend on time. independent of any derivatives. The equation
Property 2 implies that in general H+ 0. This maintaining the (It) equations is
is to be expected: Energy can be transferred from q. =;[q. H]+;[q. &.q'. 1
point to point. We should only expect that
= „-[(p„,H]+X „-[(p„, (p ]=0. (67)
H g'i'cPx=0, (62) The time derivative of H is
J all space
which is true because of properties 3 and 1. Thus, y ].
H = r[H, H) + „[H, z
in general, there exists a canonical Poynting vec- Using the properties of the integrated bracket, our
tor S' such that assumptions about H and y„, a.nd Eq. (67), we cs.n
H+V;S'=0. show that this becomes

For the simple case where H depends on no deriva- H+V; S'=0, (68)
tives of q; and p' higher than first order (which in- with
cludes almost all physical systems), the Poynting
vector is
S'= —,BH
~P'
. +x
~P
~P
BH
(69)

S'= eH eH eH aH
(63) But by Eq. (65a) this is just
qg ~P Ia ~P quota
'dH
Ss
For a degenerate system (momenta not all inde-
pendent) the equations of motion are almost as which is the canonical flux in the nondegenerate
simple. Dirac4 shows that for a system with a case as well.
finite number of degrees of freedom, In the body of this paper we will consistently use
the left-integrated Poisson bracket, which we re-
q, = [q;, H] + A [q;, (p ], (64a) fer to simply as the Poisson bracket, denoted by
p'=[p', H]+~4p', q. ]. (64b)

*Work supported in part by the National Science Foun- 2B. F. Schutz, Jr. , Phys. Rev. D 2, 2762 (1970).
dation [GP-27304, GP-19887, and GP-28027]. ~(a) L. A. Schmid, in A Critical Reviewer of Thermo-
tPresent address: Department of Applied Mathematics dynamics, edited by E. B. Stuart, B. Gal-Or, and A. J.
and Theoretical Physics, Cambridge University, Cam- Brainard (Mono, Baltimore, 1970). {b) L. A. Schmid,
bridge, England. Pure and Appl. Chem. 22, 493 (1970).
~R. L. Seliger and G. B. Whitham, Proc. Roy. Soc. P. A. M. Dirac, Proc. Roy. Soc. (London) A246,
(London) A305, 1 (1968). 326 (1958).
3566 BERNARD
~R. Arnowitt, S. Deser, and C. W. Misner, in Gravita-
tion, edited by L. Witten (Wiley, New York, 1962),
Chap. 7~
6A similar formu1ation of Einstein's equations, though
not as complete in its discussion, can be found in P. A.
M. Dirac, Proc. Roy. Soc. (London) A246, 333 (1958).
We follow the notation of Schutz, Ref. 2. In particu-
lar, Greek indices run from 0 to 3, while Latin indices
run from 1 to 3, with these exceptions: a and b label the
velocity potentials and run from 1 to 5, while mand n
F. SCHUTZ, JR.
labe1 the Dirac p equations and run from 1 to 4.
Actually, Diracs proof (Ref. 4) is for a system with
finitely many degrees of freedom. We have generalized
it' to fields using the results of the Appendix.
~B. F. Schutz, Jr. , in Proceedings of the Pittsburgh
Conference on Relativity (Springer, Berlin, to be pub-
lished) .
See Appendix B of Schutz, Ref. 2.
~~J. L. Synge, Relativity: The General Theory (North-
Ho11and, Amsterdam, 1966).

PH YSICAL REVIEW D VOLUME 4, NUMBER 12 15 DECEMBER 1971

New Equation of Motion for Classical Charged Particles*

Tse Chin Mo and C. H. Papas
Electrical Engineering Department, California Institute of Technology, Pasadena, C~llfo~nig 9g1O9
(Received 15 March 1971)

With the intuitive new ideas that (1) in classical electrodynamics, radiation reaction should
be expressible by the external field and the charge's kinematics, (2) a charge experiences,
in addition to the Lorentz forces, another "small" external force e&F" u& proportional to its
acceleration, and (3) inertia plus radiation is balanced by these two external forces, we pro-
pose the new equation of motion,
Plu —( e /3')Fext uyu~u = eFe~tuy + e(F ext uy
where mass conservation requires e& —2e3/3m. (The particle's spin is not considered in this
work. ) This equation for a classical charge is free from all the well-known difficulties of the
Lorentz-Dirac equation. It conserves energy and momentum in a modified form in which the
energy-momentum tensor contains a part tl"'(x) made of a new field-charge interaction Q" (x),
in addition to the conventional "local" part made of F",et(x) and F",«(x) only, and therefore it
no longer satisfies the conventional "local" conservation laws. It predicts correct radiation
damping, as demonstrated here by applying it to various cases of basic physical importance.
Also, it implies that a massless particle follows a null geodesic and cannot interact with
the electromagnetic field whether it be charged or not; this implication may add a new de-
gree of freedom to the charge-conservation law.

I. INTRODUCTION radiation is created from an "acceleration self-

The equation of motion of a charged particle has
been a subject of interest for many years. ' The
equation now generally accepted was obtained by
Dirac by decomposing the energy-momentum ten-
sor of the retarded self-field into a sum that re-
normalizes mass and a difference that gives reac-
tion. ' An explanation and rederivation based on
an absorber mechanism was provided by Wheeler
energy" which becomes more and more negative
and is unphysical. It is the purpose of this work to
obtain a new equation that is free from these dif-
ficulties, agrees with existing laboratory results,
and predicts new phenomena that can distinguish
the new equation from the old one and test its va-
lidity.
II. THE NEW EQUATION

and Feynman. ' However, as is well recognized,
the Lorentz-Dirac equation has certain inherent
difficulties. First, it involves the derivative of
the acceleration and hence needs one extra condi-
tion, in addition to the Newtonian initial conditions,
to determine the motion. Second, it gives runaway
solutions which can be avoided only by artificially
presenting a preacceleration. 4 Third, in certain
cases it implies that the external energy supplied
to the particle goes only into kinetic energy, and
By following the old idea of expressing the radi-
ation reaction of a charged particle only by its
kinematical quantities, it is not possible to con-
struct an equation that includes reaction in a form
simpler and more satisfactory than the Lorentz-
Dirac one. However, in classical electrodynamics
in an inertial frame' the only field that can accel-
erate a charged particle and make it radiate is the
external electromagnetic field Fl,'„'t . Accordingly,
radiation reaction should be expressible by F"„",