ANNALS OF PHYSICS: 2, 525-603 (1%‘)

Classical Physics as Geometry

Gravitation, Electromagnetism, Unquantized Charge, and Mass as

Properties of Curved Empty Space*

CHARLES W. MIsNERt AND JOHN A. WHEELERS

Lorentz Institute, University of Leiden, Leiden, Netherlands, and Palmer Physical

Laboratory, Princeton University, Princeton, New Jersey

If classical physics be regarded as comprising gravitation, source free elec-

tromagnetism, unquantized charge, and unquantieed mass of concentrations
of electromagnetic field energy (geons), then classical physics can be described
in terms of curved empty space, and nothing more. No changes are made in
existing theory. The electromagnetic field is given by the “Maxwell square
root” of the contracted curvature tensor of Ricci and Einstein. Maxwell’s equa-
tions then reduce, as shown thirty years ago by Rainich, to a simple statement
connecting the Ricci curvature and its rate of change. In contrast to unified
field theories, one then secures from the standard theory of Maxwell and Ein-
stein an “already unified field theory.” This purely geometrical description of
electromagnetism is traced out in detail. Charge receives a natural interpreta-
tion in terms of source-free electromagnetic fields that (1) are everywhere
subject to Maxwell’s equations for free space but (2) are trapped in the “worm
holes” of a space with a multiply-connected topology. Electromagnetism in
such a space receives a detailed description in terms of the existing beautiful
and highly developed mathematics of topology and harmonic vector fields.
Elementary particles and “real masses” are completely excluded from discus-
sion as belonging to the world of quantum physics.

“I transmit but I do not create; I am sincerely fond of the ancient.“-Con-

fucius.
I. IS THE SPACE-TIME CONTINUUM ONLY AN ARENA, OR IS IT ALL? CLASSI-
CAL PHYSICS REGARDED AS COMPRISING GRAVITATION, ELECTROMAG-
NETISM, UNQUANTIZED CHARGE, AND UNQUANTIZED MASS; ALL FOUR
CONCEPTS DESCRIBED IN TERMS OF EMPTY CURVED SPACE WITHOUT
ANY ADDITION TO ACCEPTED THEORY; THE ELECTROMAGNETIC FIELD
AS THE “MAXWELL SQUARE ROOT” OF THE CONTRACTED CURVATURE;
UNQUANTIZED CHARGE DESCRIBED IN TERMS OF SOURCE FREE MAX-
* Part VI of a critique of classical field theory. Part V appeared in Phys Rev. 97, 51 (1956).
t National Science Foundation Predoctoral Fellow.
1 Holder for part of the period of this work of a John Simon Guggenheim Fellowship at
the Lorents Institute of the University of Leiden.
525
526 MISNER AND WHEELER

WELL FIELD IN A MULTIPLY-CONNECTED SPACE; UNQUANTIZED MASS

ASSOCIATED WITH COLLECTION OF ELECTROMAGNETIC FIELD ENERGY
HELD TOGETHER BY ITS OWN GRAVITATIONAL ATTRACTION; HISTORY
OF IDEAS OF PHYSICS AS GEOMETRY; SUMMARY OF PAPER

Two views of the nature of physics stand in sharp contrast:

(1) The space time continuum serves only as arena for the struggles of fields
and particles. These entities are foreign to geometry. They must be added to
geometry to permit any physics.
(2) There is nothing in the world except empty curved space. Matter, charge,
electromagnetism, and other fields are only manifestations of the bending of
space. Physics is geometry.
To understand how far one can go in regarding classical physics as geometry
is the object of this paper. Nothing will be said here about the fascinating
issue’ of quantizing this classical pure “geometrodynamics” (Table I).
In describing classical physics (in the sense of Table I) as geometry, we invent
no new ideas. We accept Maxwell’s 1864 electrodynamics of empty space, his
formulation of the stress-momentum-energy tensor of the electromagnetic field,
and Einstein’s forty-one-year old description of gravitation in terms of curved
space. Restricting attention to classical concepts (Table I) we take as source
of metric fields, gfiV, exclusively electromagnetic fields, Fap = (c2/G1”)j,g-and
electromagnetic fields that are themselves free of all sources’:
i See, however, Misner (1) and Wheeler (2) for a partial discussion of some features of
this problem.
2 We accept the following familiar conventions: Greek labels refer to four dimensional
space; Latin labels refer to three dimensional space. The fourth coordinate, x0(= T =
ct = “cotime” in flat space) receives the label 0 to prevent confusion with the occasional
use in special relativity of x4 to designate ict. The proper distance, ds, or proper interval of
cotime, dr, between two neighboring events is given by

@s)~ = - (&)2 = gas dx” dxa.

In flat space and Euclidean coordinates, the metric tensor is diagonal with -1, 1, 1,l in the
diagonal. Many of the considerations of this article deal with space like manifolds, on which
it is a great convenience to have a positive definite metric, as given by the present conven-
tion (see also Pauli, Landau and Lifschite, Jauch and Rohrlich). The determinant, 1 gaB 1,
of the metric in four space is designated by g, and the determinant, ( g;k 1, of the metric on
a spacelike manifold is designated by 39. Other important quantities include the bending
coefficients,

rng,y = $5@gdax~ + a&a28 - agae/ae ;

the Riemann curvature tensor, with its twenty distinct components, R,,, , where

R,Q~,,= ar,,p/ax” - ar,fl/ax’ + r~r,~q - r,,Mr,,v;

the symmetric Ricci tensor or contracted Riemann tensor,

R,, = R,“w;
 CLASSICAL GEOMETRODYNAMICS 527

(3!)-l[aPrsl(afs,/az6) = 0 (half of Maxwell’s equations), (1)

(-g)-“‘(a/asP)(-g)l’~~ = 0 (the other half), (2)

“Oga,? = R,,a - %sc& = 2jmajtt - %&s&j”‘)

(curvature of metric by Maxwell stress-momentum-energy density). (3)
These equations describe electromagnetism and gravitation as a coupled but
closed dynamical system.
Solve Eqs. (3) for the reduced electromagnetic field, f<, , in terms of the con-
tracted curvature tensor, Ro,s . Substitute the resulting expressions into Maxwell’s
equations. Thus re-express the content of the Maxwell-Einstein equations in a
purely geometricalform. This program was carried out by Rainich in an important
paper (zI)~ that has long lain neglected. The result is simple. (1) The symmetric

the curvature invariant, R = Rmu; the generalization of the notion of d’hlembertian of

the gravitational potentials,
“Og,,” = R,. - %swR;

and the electromagnetic potentials, A, , such that

FOP = aAP18xu - aAJax@.

The alternating quantity that is often written in the form e,ora is not a tensor and is here
written in the form [C&S]. It changes sign on interchange of any two indices, and [0123]
has thevalue unity. The dual, (*F),, , of an alternating tensor, Fe@, is defined by the equa-
tion

(*F),,y = ?,~(-g)““G~~~]g~*g’“F.a .
Associated with the geometrized electromagnetic field quantities, f.~ = (Gl’*/c2)F,~ ,
are the geometrized electromagnetic potentials, a, = (G1’2/~2)A, , which are dimension-
less. In flat space and Euclidean coordinates,

dxl = dxi = displacement in x-direction,

dx” = -dx0 = interval of cotime,

A1 = A, = x-component of usual vector potential,

A0 = -A0 = usual scalar potential, V(es volts),

Fx3 = -Faa = x-component of magnetic field,

Flo = -Fol = x-component of electric field

3 Even Rainich’s later book, (4) does not summarize this paper, primarily because he
was motivated by a dierent view of classical physics than that under investigation in the
present article. We undertook the problem of expressing (I), (2), and (3) in “already unified
form”, and one of us (C. M.) independently derived Rainich’s results, before becoming
aware of his valuable contribution. The possibility of such an “already unified theory” was
first suggested to us by Dr. Hugh Everett.
528 MISNER AND WHEELER

TABLE I
THE DISTINCTION BETWEEN CLASSICAL AND QUANTUM PHYSICS
AS ENVISAGED IN THIS PAPER*

Classicalphysicsas Description in terns of the geometry Quantum physics; not discussed in

defined here of empty curved space this paper

Gravitation Defined by curving of geodesics Gravitons; photons; spin; neutri-

in a Riemannian space nos; quantization of charge;
Electromagnetism Determined by curvature, and qumtization of mass; electrons,
its rate of change, in this same mesons and other particles;
Riemannian space (Fig. 2) characteristic fields that do not
Unquantised charge Manifestation of lines of force have zero rest mass, apparently
trapped in a multiply con- associated with some of these
nected topology (Fig. 3) particles; particle transforma-
Unquantized mass Geons: semistable collection of tion processes; also all phenom-
electromagnetic or gravita- ena where quantum fluctua-
tional wave energy held to- tions in the metric are more
gether by its owngravitational important than any static
attraction gravitational fields.

G = 6.67 X 10-a cm3/g sec2 and c define no char- G, c, and h define the characteristic units
acteristic length, mass, or time. Electromag- first introduced by Planck: L* =
netic field F,, (in gauss or electrostatic volts (hG/c3)“* = 1.63 X lo-33 cm; T* =
per cm or (g/cm sec2)1/2) translated into L*/c; and M* = (he/G)“* = 2.18 x
purely geometric quantities f,, (in units of 10-6 g.
cm-i) by multiplication with G”z/c2 =
l/3.49 X lot4 gauss cm.
a The unquantieed classical charge and mass in the table have no direct relation what-
soever with the elementary masses and charges that are seen in the quantum world of
physics.

Ricci tensor, R,o , can be expressedas the “Maxwell square”, as in Eq. (3), of an
alternating tensor, .fV,, if and only if this tensor (Fig. 1) (a) has zero trace and
(b) has a square which is a multiple of the unit matrix:
R = R," = 0,
R/R,' = 6,y(R,,R"r/4).
We therefore demand these conditions of the Ricci tensor. (2) Then this con-
tracted curvature tensor determines the local value of the reduced electromag-
netic field tensor, jr7 , uniquely up to an arbitrary angle, CY,by way of an equation
which we write symbolically in the form4
4 We wish to express our appreciation to Professor V. Bargmann for bringing Eqs. (5)
and (6) to our attention two years ago, noting that their gist had been independently dis-
covered by several investigators, and expressing their content in essentially the above
exceptionally simple form. An early proof is given by Rainich himself (5). The result is
implicit in Theorem V of a study by Synge (6); see also Synge’s book (6), a paper by
Bonnor (7), and the thesis of Louis Mariot (8), for which we are indebted to M. Mariot.
 CLASSICAL GEOMETRODYNAMICS 529

fc, = (&faxwellroot)ar cos a + ~~~~~~~~~~~~~~~~ sin a. (6)

(3) The expression for the electromagnetic field in terms of the Ricci curvature
is substituted into Maxwell’s equations. The laws of electrodynamics thereby
take on the following purely geometrical character. First, out of the derivative

FIG. 1. Simplification of the analysis of the Maxwell stress-momentum-energy tensor by

passage to a locally Lorentz frame of reference in which E and H are parallel. Left: Electric
and magnetic vectors in the original reference system. Calculate the energy flux, c(E X H)/
4~, and the energy density, (E2 + H2)/83r, and their ratio, the velocity, v. Right: View the
fields in a frame of reference moving with this velocity. The energy flux must vanish. There-
fore E’ and H’ must be parallel. Let their common direction be called the z’ axis. There
is a Maxwell tension (E’2 + H’*) along this axis, and equally strong Maxwell pressures
along the two perpendicular y’ and z’ axes. Therefore the stress-momentum-energy tensor
has the form

(P/8d

where F’z is an abbreviation for the invariant

F’a = E’2 + H’Z = [(E’Z - H’S)2 + 4(E’.H’)2]1/2
= [(E* - Hz)2 + 4(E.H)21L’* = [(E* + H*)z - 4(E x H)2]i/*.
This tensor has two important properties (1) its trace is zero (2) its square is a multiple of
the unit matrix. Both features are invariant to change of coordinate system. They hold
whether the Maxwell tensor is diagonal or not.-Conversely, consider a real symmetric
tensor which enjoys the properties (1) and (2). One can find a coordinate system with a
favored direction, z’, which puts it in the above diagonal form. In particular, one can find
at once the invariant field magnitude,
F’ = [(Sk Maxwell tensor)z/(unit matrix)]1’4.
Then pick any angle a, and define vectors E’ and H’ that point in the favored direction, z’,
with magnitudes
E’ = F’ sin (Y, H’ = F’ eos CY.

The vectors E’ and H’ are determined uniquely apart from the single freely disposable
parameter, a. Transform the electromagnetic field so defined back to the original reference
system. Operating on this field, the Maxwell prescription for the stress energy tensor will
produce the symmetric tensor with which one started.-These proofs break down when the
electromagnetic field is a null field, with E perpendicular to H and equal in magnitude to
H. but the statement in the text is still true.
530 MISNER AND WHEELER

of the Ricci tensor form the vector CY,dejked by the equation

a, = (- g)‘12[7XyV]RXB;“RB”/R~~R~*. (7)
(The null-case where R,aRY6 vanishes requires special treatment). Second, de-
mand that the curl of this vector shall vanish:

%;rl - ffq;, = &,q - c&J,7= 0. (8)

(4) This diferential equation (8), plus the algebraic equations (4) and (5), sum-
marize in complete geometrical form both the whole of source-free Maxwell electro-
magnetism in curved space, and Einstein’s laws for the production of curvature by
thisJieZd. These three equations, (4), (5), (8), comprise what we shall call “already
unified field theory”. Electric and magnetic fields are not signals to invent a
unified field theory or to introduce one or another new kind of geometry. The
“already unified field theory” of Maxwell, Einstein, and Rainich, summarized
in this paper, describes electric and magnetic fields in terms of the rate of change
of curvature of pure Riemannian geometry, and nothing more.
The nature of this unification can be stated in mathematical terms as follows:
Maxwell’s equations are of the second order, and so are Einstein’s; the two sets
of equations can be combined into one set of equations (8) of the fourth order.
In more physical terms, the electromagnetic field leaves an imprint5 upon the
metric that is so characberistic (Fig. 2), that from that imprint one can read
back to find out all that one needs to know about the electromagnetic field.
Given a purely metric field that satisfies Eqs. (4), (5), and (8) of already unified field
theory, one finds the electromagnetic field as follows. First, calculate everywhere the vector
field 01,, of Eq. (7). Second, from some standard point 0 calculate the integral

Since alp is curl free, the integral is independent of path, so long as alternative paths are
continuously deformable into one another. (When instead the space is multiply connected,
new considerations will be needed.) We therefore have a dimensionless number or angle, o(,
defined as a function of position, up to an additive constant, LYE. Finally, we substitute this
angle into Eq. (6) to find the electric and magnetic field at every point in space.

We find that long established theory has a well defined means to describe
gravitation and electromagnetism in terms of empty curved space. What about
charge?
Einstein emphasized that the field equations of electromagnetism and general
relativity have a purely local character. They relate conditions at one point to
conditions at points an infinitesimal distance away. They tell nothing about the
topology of space in the large. Einstein was led by Mach’s principle (9) to consider
5 We are indebted for this phrase to Professor Peter Bergmann.
 CLASSICAL GEOMETRODYNAMICS 531

a space not topologically equivalent to Euclidean space, a spherical or nearly

spherical universe. But Einstein confesses his indebtedness to a thinker who had
still more far reaching ideas. Riemann6 in his famous inaugural lecture envisaged

FIG. 2. Relation between the electromagnetic field and geometry, schematically repre-
sented. Above: lines of force. Middle: Maxwell stress tensor due to these lines of force. This
stress tensor serves as source of the gravitational field and equals the contracted curvature
tensor of the space-time continuum, up to a multiplicative constant, according to Einstein.
Below: The metric of four-space as distorted by this curvature. In brief, the electromagnetic
field leaves its footprints on space. Moreover, these footprints on the metric are so specific
and characteristic that from them one can work back and find out all that needs to be known
about the electromagnetic field. One has a purely geometrical description of electromagne-
tism.

a connection between physics and a curvature of space that would be sensible

not only at very great distances, but also at very small distances: I‘. . . es kann
dann in jedem Punk& das Kriimmungsmass in drei Richtungen einen beliebigen
Werth haben, wenn nur die ganze Kriimmung jedes messbaren Raumtheils
6 In the opening passage of this lecture (10) Riemann declares that “the properties which
distinguish space from other conceivable triply extended magnitudes are only to be deduced
from experience.” (translation of Clifford (11)).
 532 MISNER AND WHEELER

nicht merklich von Null verschieden ist; . . .” Dying of tuberculosis twelve years
later, occupied with an attempt at a unified explanat,ion of gravity and electro-
magnetism, Riemann communicated to Bet,ti his system of characterization of
multiply-connected topologies’. What is the character of charge-free electromag-
netism in a space endowed with such a multiply connected topology?
One can outline a complete classification of the everywhere regular initial con-
ditions for Maxwell’s equations in a closed space. This analysis forces one to
consider situations-such as described by one of us (14) (Fig. 3)-where there
is a net flux of lines of force through what topologists would call a handle of
the multiply-connected space and what physicists might perhaps be excused
for more vividly terming a “wormhole”. The flux of lines of force that emerge
from the mouth of a small wormhole appears to an observer endowed with poor
resolving power to come from an elementary electric charge. But there is nowhere
that one can put his finger and say, “This is where some charge is located*.”
Lines of force never end. This freedom from divergence by no means prevents
changes in field strengths. Lines of forces which are not trapped into the topology
can be continuously shrunk to extinction, as in familiar examples of electromag-
netic induction and electromagnetic waves. However, lines of force which are
trapped in wormholes cannot diminish in number. The flux out of the mouth of
a wormhole cannot change with time, no matter how violent the disturbances in
the electromagnetic field, no matter how roughly the metric changes, no matter
how rapidly corresponding wormholes recede or approach, up to the moment
when they actually coalesce and change the topology. Either Maxwell’s equa-
7 Weyl (13) emphasizes that the field equations provide no means whatever to rule out
either multiply-connected spaces, or spaces which are nonorientable, such as a Klein bottle.
He notes (original German in 1927; translation and revision in 1949) “that a more detailed
scrutiny of a surface might disclose that, what we had considered an elementary piece, in
reality has tiny handles attached to it which change the connectivity character of the piece,
and that a microscope of ever greater magnification would reveal ever new topological com-
plications of this type, ad infinitum. The Riemann point of view allows, also for real space,
topological conditions entirely different from those realized by Euclidean space. I believe
that only on the basis of the freer and more general conception of geometry which had been
brought out by the development of mathematics during the last century, and with an open
mind for the imaginative possibilities which it has revealed, can a philosophically fruitful
attack upon the space problem be undertaken.” Einstein and Rosen (IS) proposed in 1935
to regard ordinary space as connected with a duplicated “mirror” space by short tubes.
This topology is much more particular than anything contemplated here or in the follow-
ing paper (2). Einstein and Rosen also took the electromagnetic field to have a negatiue-
definite energy density, in contradiction to experience. We learn that Professor J. L. Synge
also once mentioned in a lecture at Dublin in 1947 the idea of multiply connected space.
8 In 1895 the great physicist Henry A. Rowland said, “. . . electricity no longer exists,
for the name electricity as used up to the present time signifies at once that a substance is
meant, and there is nothing more certain than that electricity is not a substance.” (Quoted
by Darrow (15).) His words are apropos here!
 CLASSICAL GEOMETRODYNAMICS 533

FIG. 3. Symbolic representation of the unquantized charge of classical theory. For ease
of visualization the number of space dimensions is reduced from three to two. However, the
two dimensional curved and multiply-connected space is pictured as imbedded in a three
dimensional Euclidean space. The third dimension, measured off the surface, has no physi-
cal meaning. Of course the topology and geometry of the a-space receives its best mathe-
matical formulation in intrinsic terms, without this imbedding of the manifold in a space
of higher dimensionality. The 2-space is multiply-connected, but free of all singularities. An
imaginary ant crawling over the surface and entering the tunnel or handle or “wormhole”
finds there the same two dimensional space he experienced everywhere else. Electric lines of
force that converge on the right-hand mouth of the tunnel continue to obey at each point
Maxwell’s equation, div E = 0. The field is everywhere free of singularity. The lines of
force have no escape but to continue through the tunnel. They emerge from the left-hand
mouth. Outside the tunnel mouths the pattern of lines of force is identical with that due to
equal positive and negative charges. An observer endowed with poor vision sees evidence
for two point charges. He may even construct a boundary around the right-hand charge,
determine the flux through this boundary, incorrectly apply the theorem of Gauss, and
“prove” that there is a charge inside the boundary. He does not recognize that he has been
making tacit and unjustified assumptions about the topology of space. He is not aware that
his “boundary” does not bound any region interior to it. He assumes, either that Maxwell’s
equations fail in the vicinity of the charge, or that there exists there some magic substance
at which lines of force end and to which he gives the name “electricity”. But a closer in-
spection discloses that the lines of force do not end. Neither is there any violation of Max-
well’s equations for charge-free space. Nowhere can one place his finger and say, “Here
there is some charge”. Such is the purely topological picture of unquantized electric charge
which is adopted in this paper. This classical charge has no direct relation whatsoever to
quantized electric charge. At this classical level there is a freedom of choice in the strength
of the charge, and an individuality about the connection between one charge and another,
that must be entirely changed in any proper quantum theory of electricity.-The distance
along the wormhole from one mouth to the other need have no correspondence whatever
with the distance in the open space between the same two mouths. The connection can be
as short as the radius of the wormhole itself, for example, even when the openings are very
far apart, in the upper space, as one sees by bending the upper space to bring the backs of
t,he two holes into coincidence (diagram reproduced from Ref. 14).
534 MISNER AND WHEELER

tions, or Faraday’s equivalent physical picture of lines of force, plus the concep-
tion of multiply connected space, force one to the conclusion that the wormhole
flux remains invariant. This constant of the motion represents the charge.
The charge or wormhole flux is unquantized. It can have one value as well as
another. It has nothing whatsoever directly to do with the quantized charge
observed on the elementary particles of quantum physics. This circumstance is
not an objection to the concept of a classical unquantized charge. It is a warning
that quantized charge is quite another concept. This distinction is not unaccept-
able at a time when one has learned how great a difference there is between the
“undressed” and “dressed” charge of quantum electrodynamics (16). To limit
attention to purely classical unquantized charge will therefore not appear un-
reasonable in an article that restricts itself to classical physics (Table I).
Around the mouth of a wormhole lies a concentration of electromagnetic
energy that gives mass to this region of space. Mass arises even in singly con-
nected space, where there is no charge connected with the source-free Maxwell
field. The equations of Maxwell and Einstein predict the possibility of a long-
lived concentration of electromagnetic energy, or “geon”, held together by its
own attraction. Both in the multiply-connected space and in the singly-connected
continuum, the mass with which one has to do is classical, nonlocalized, and
unquantized. It has nothing whatsoever directly to do with the quantized mass
of elementary particles.
Summarized in paradoxical form, the existing well-established already unified
classical theory [Eqs. (4), (5), @)I a 11ows one to describe in terms of empty
curved space
1. gravitation without gravitation
2. electromagnetism without electromagnetism
3. charge without charge
4. mass without mass.
It has nothing at all to contribute directly to an understanding of
5. spin without spin
6. elementary particles without elementary particles,
or any other issues of quantum physics. Nevertheless, we would hardly have
taken up the analysis of classical geometrodynamics if we did not hope ulti-
mately to find out what, if anything, quantum geometrodynamics has to do with
elementary particle physics. It is our long range objective to discover if quantum
physics, like classical physics (Table I), can be expressed in terms of pure ge-
ometry, and nothing more.
It is not customary today to adopt either extreme view, either that space time
is only an arena, or that it is everything. One analyzes the states of particles and
fields into plane waves that move as foreign elements in a preassigned flat space.
At the same time one thinks of the curvature of space, not as exactly zero, but
 CLASSICAL GEOMETRODYNAMICS 535

only as very small over distances short compared to the extension of the uni-
verse. Einstein’s geometrical description of gravitation is taken seriously. His
attempts at an equally geometric description of electromagnetism-by modijy-
ing Riemannian geometry-are recognized to be incompatible (l?‘) with the
well-tested Lorentz law of force and are rejected. Particles, and fields other than
gravitation, are considered to be added to geometry, not as derived from ge-
ometry. Nature can be said to be described today in a mixed fashion, partly in
terms of pure geometry, partly in terms of foreign entities.
To go to the logical extreme, however, and think of a purely geometrical
description of nature, was not a new idea even before one knew enough to dis-
tinguish between classical and quantum physics. The distinguished mathema-
tician Clifford delivered a paper to the Cambridge Philosophical Society on Feb-
ruary 21, 1870 “On the Space-Theory of Matter,” in which he proposed that
“in the physical world nothing else takes place but this variation [of the curva-
ture of space], subject (possibly) to the law of continuity;” and later he spoke of
considerations “which indicate that distance or quantity may come to be ex-
pressed in terms of position in the wide sense of the analysis situs;” and again
about the finite volume of a uniformly curved space, but with the explicit state-
ment that “The assumptions here made about the Zusammenhang of space
are [merely] the simplest” (18). Before Einstein, Clifford, and Riemann-and
Riemannian geometry-was there ever current anything like the concept of
physics as geometry? What were Newton’s views of field theory and the idea
that empty space is the universal building material? His letter to Bentley has long
been known: “That one body may act upon another at a distance through a
vacuum, without the mediation of anything else, by and through which their
action and force may be conveyed from one to another, is to me so great an ab-
surdity, that I believe no man, who has in philosophical matters a competent
manner of thinking, can ever fall into it”g. Maxwell says, “We find in his ‘Optical
Queries’ and in his letters to Boyle, that Newton had very early made the at-
tempt to account for gravitation by means of the pressure of a medium, and that
the reason he did not publish these investigations ‘proceeded from hence only,
that he found he was not able, from experiment and observation, to give a satis-
factory account of this medium, and the manner of its operation in producing
the chief phenomena of Nature.’ ”
New insight into Newton’s ideas and their origins comes from the recent
scholarly and most interesting analysis by Fierz (20). Fierz cites especially
Patrizzi (21), who writes of space as a substance, “Spacium ergo hoc, quod ante
mundum fuit, et post quod mundus est, et quod mundum at capit, et excedit,
quidnam tandem est. . . . Quid ergo substantia ne est? Si substantia est, id quod
per substat, spacium maxime omnium substantia est.” Also some of the Vedas
9 This and the following quotation come from notes of Cajori (19).
 536 MIGNER AND WHEELER

of old India (22) suggestga that the idea is very old, that nature derives its whole
structure and way of action from properties of space.
Can space be regarded as a marvellous creation of all-encompassing proper-
ties? Independent of the origins of this idea, both ancient and modern, let us now
proceed to analyze it.
In Sec. II we recapitulate in present day notation the derivation of the equa-
tions of Rainich for already unified field theory. The starting point, the theory
of Einstein and Maxwell, deals entirely with local properties and so does Rainich’s
final system of equations.
To pass from local to global or topological properties, and still to keep the
discussion simple, we return in Sec. III to the more familiar dual language of
metric plus field. We sketch out the necessary topological background, and
introduce the theorem of Gauss and the theory of harmonic vector fields in the
necessary generality. Much of the required mathematics is most readily expressed
in terms of Cartan’s calculus of exterior differential forms. Most results we give
both in this notation and in the familiar tensor form. A few conclusions would
appear so complicated in the tensor formalism that we omit their transcription
to the conventional notation. We prove that Maxwell’s equations demand the
9% For very early ideas related to “physics is geometry” we have been referred by Pro-
fessor G. L. Chandratreya to the Indian Vedas. In this connection we wish to thank Swami
Nikhilananda who explains to us the relevant writings: “According to the Vedas, akasa
(often translated as ‘space’ or ‘ether’) is the rudimentary first element from which the other
elements, namely, air, fire, water, and earth, have evolved. These tive are the only material
elements spoken of by the Vedas. Hindu philosophers populated five elements because a
man reacts to the outside world in five ways: through his hearing, touch, sight, taste, and
smell. . . . [quoting in this connection from Taittiriya Upanishad II. i. 3.1” Thus we find
akasa as the primary element in this early (c. 700 B. C.) sketch of physics. (Brahman, al-
though preceding akasa, is pure spirit, outside the realm of physics.) We can say that in
this physics space was the primary element out of which all else was to have come, only if
we can satisfy ourselves that akasa meant something like the current word space. In this
connection we quote the most authoritative early commentators on the Upanishad just
mentioned (29). SankaracMrya (A. D. 788-820) tells us “Akasa is that thing which has
sound for its property and which affords space to all corporeal substances.” Then Sayana
elucidates: “ . . . the power of akasa to afford space to all (corporeal) things constitutes its
own peculiar nature . . . And it has sound for its property. The echo heard in mountain-
caves etc., is supposed to be inherent in akasa and is therefore said to be the property of
akasa.” Except for the curious references to sound, these explanations seem to corroborate
a tentative identification of akasa with space: a physicist might write ‘space provides room
for all things’ where the translator wrote ‘akasa affords space to all things’. The reference
to sound is understood when we recall that the five elements were chosen to correspond to
the five senses. Sanka&cMrya writes “Thence, i.e., from akasa, comes into being Vayu,
the air, with two properties, the property of touch which is its own, and the property of
sound belonging to akasa already evolved.” It is perhaps too much to expect that at such
an early date men would know that sound is transmitted through the air, and not through
empty space.
 CLASSICAL GEOMETRODYNAMICS 537

conservation of flux through each wormhole independently, thus justifying the

identification of this flux with charge.
Section IV deals with specific examples of nonsingular multiply-connected
metrics that manifest both charge and mass. First the Schwarzschild metric is
written in nonsingular form, to put into evidence the special case of a space free
of either charge or mass-in the conventional sense of those words-which never-
theless exhibits mass. Next a nonsingular form of the Reissner Nordstrijm
metric is exhibited. It describes a spherically symmetric space free of all “real”
charge and mass which nevertheless exhibits both properties. Finally, a closed
mathematical form is given for a class of metrics endowed with a plurality of
wormhole mouths, each with its own charge and mass. The initial conditions
being thus specified, the future evolution of the space with time is of course
determined by the field equations. In other words, the arguments of Einstein,
Infeld, and Hoffman apply to this situation. The entire dynamics of the system
of singularity-free charges and masses becomes a matter of pure geometrody-
namics.
Section V outlines points that require further investigation to complete classical
geometrodynamics and to extend its domain of application.
II. RAINICH’S ALREADY UNIFIED FIELD THEORY: THE MAXWELL TENSOR;
DUALITY; DUALITY ROTATIONS; INVARIANCE OF MAXWELL TENSOR TO
DUALITY ROTATION; COMPLEXION OF FIELD DEFINED; SQUARE OF MAX-
WELL TENSOR; THE ALGEBRAIC RELATIONS ON THE CURVATURE; THE
REVERSE PROBLEM-FROM THE CURVATURE TO FIND THE FIELD; RE-
SULTING DIFFERENTIAL EQUATIONS OF THE CURVATURE

THE MAXWELL TENSOR

We begin by recalling the relation between the electromagnetic field, FNy
and the Maxwell stress-momentum-energy tensor TMy. This relation is purely
algebraic. Hence we may concentrate our attention on a single point of space
time and, when convenient, use coordinates that give the metric components,
grv, their Minkowski values at that point. To keep geometry to the fore, we shall
use instead of F the “geometrized” or “reduced” field strength, f = (G1”/c2) F,
and instead of the electric and magnetic field strengths, E and H, the reduced
field strengths, e= (G”2/~2)E and h = (G”2/~2)H (dimensions cm-‘). When the
metric is Minkowskian, the reduced field tensor has the form
-e, -e, -e,
538 MISNER AND WHEELER

and the dual tensor,

*fpv = !4 -s>““b~4f”8, (11)
differs from f only by the interchange e * h, h ---f -e. Two familiar invariants
form themselves out of the field tensor:
f2 = >$S,Vf”’ ( =h2 - e2 in a Minkowski frame),
f X f = $~fP#‘” ( = 2e ah in a Minkowski frame).
The equations connecting the contracted curvature tensor with the stress
tensor, and that in turn with the field,
R,. - %gpvR = (8dW4>Tp = (87rG/c4)I(1/47r)@‘,A’,” - ~gJi’,@‘“B)l,
(13)
= 2(.f/&OL - ~g,vfnLP> = X(f)
justify one in using interchangeably for the quantity on the right the terms “re-
duced” or “geometrized” stress energy tensor, or contracted curvature tensor,
insofar as one deals with pure geometrodynamics. In the Minkowski frame of
reference, typical components of (13) are
Roe = Roe= -Ro” = (e* + h’) = (87rG/c4)(energy density);
-RIO = R1’ = Rf = 2e X h = (87rG/c4)(density of c times x-component of
momentum) = (89rG/c4)(flow of electromagnetic energy per cm2 of
area normal to x and per cm of elapsed cotime);
RI1 = R” = R1’ = ( -e2= + e2, + e”, - h”, + h”, + h2,) = (8~G/c*)
(pressure) = (8?rG/c4)(force exerted in x direction, per unit area
normal to x, by electromagnetic fields in medium at x - E, acting
on medium at x + E);
RI2 = RI2 = R? = -2e,e, - 2h,h, = (8rG/c4)(shear) = (87rG/c4)(force exerted
in x direction, per unit area normal to y, by electromagnetic forces
due to medium at y - Eacting upon medium at y + 8). (14)
The invariance of the stressenergy tensor under the interchange e + h, h -+ -e,
is apparent from (14), but doesnot show itself clearly in (13). Therefore it is pref-
erable to rewrite the reduced stress tensor, or the “Maxwell square of f”, on the
right-hand side of (13) in the more symmetrical form
5(f) = f,crj”a + *frro*fva, (144
as follows from the identity”
(15)
DUALITY ROTATIONS*
The stress tensor, or Maxwell square of f, shows a further symmetry. Note
that the operation of taking the dual of f, twice repeated, leads back to -f, so
 CLASSICAL GEOMETRODYNAMICS 539

that the square of the operation * is the negative of the identity operation. Con-
sider therefore an angle 01and define the operation e*& by the equation

e’“f = f cos c11+ *f sin (Y. (16)

In a Minkowski coordinate system this operation takes the form

h z new = h, cos CY+ e, sin QI

(same for I, y, Z) (17)
e, new= - h, sin o! + e, cos (Yi
This operation appears at first sight to be an ordinary rotation: applied to any
linearly polarized monochromatic wave, with 1 e 1 = 1h j and e 1 h, it turns
the direction of polarization through the angle cy around the direction of propa-
gation. However, when this operation is applied to less special fields, it produces
no such simply describable result. Moreover, it treats all three space axes alike.
It is not an ordinary rotation in 3-space. We shall therefore call it a duality
rotation. It has the additivity property
.a *p *@*a
e e =ee =e *ca+l9 (18)
and the special value
*r/z = *.
e (19)

The dual of the duality rotation yields the field tensor

*(e*“f) = -f sin a! + *f cos cy. (20)
The duality rotation has the following important property as a consequence of
(14), (16), and (20): The Maxwell square of a duality-rotated Jield is identical with
the Maxwell square o.f the original jield:
2(e*“lf) = 2(f). (21)

The electric and magnetic fields individually are changed, but every component
of the stress energy tensor is unaltered. In contrast (Table II) the duality rota-

TABLE II
CONTRAST BETWEEN PROPER LORENTZ TRANSFORMATIONS AND DUALITY ROTATIONS
General proper
Quantity Lorentz trans- Duality rotation
formation
Components of the Maxwell tensor or Maxwell square Transformed Unchanged
of f
The invariants, fZ and f X f Unchanged Transformed
540 MISNER AND WHEELER

tion alters the invariants of the field. Make the definition

t = e-f @2)
for a field which has undergone a duality rotation by the angle --a. Then the
invariants transform as by a rotation through the angle -2~:
t2 = (f cos Q: - *f sin CY)’
= f2 cos 2a - f X f sin 2a, (23)
h x t: = f2 sin 2a + f x f cos 2ff.
Assume that the invariants, f2 and f X f, of the original field do not both vanish.
Then choose the angle cr so that the one invariant quantity, f X I, is zero:
tan 2a = -(f X f)/f*. (24)
Then solve for sin 2a and cos 2a up to a f ambiguity and evaluate the other
invariant, finding
t2 = *[(f2)2 + (f x f)*]“Y (25)
Demand that the minus sign shall appear on the right, thus determining the
angle 2cu uniquely up to a positive or negative additive integral multiple of 27r.
Then the field tensor a represents a pure electric field along the x-axis, or a Lo-
rentz transformation thereof (Table III). We say that the original field has
received a duality rotation into an extremal field, or into an essentially electric
Jield. In a preferred Lorentz system where this field has no magnetic components,
and points along the x-axis, the field magnitude is
ez“’ = [(h’ - e’)’ + (2e.h)2]L’4
WI
= [(h’ + e2)’ - 2(e X h)2]1’4

THE COMPLEXION OF THE FIELD

Referred to an extremal, or essentially electric field, F, as standard of reference,
the actual field, f, evidently arises by a duality rotation through the angle a:
f = era&
TABLE III
TRANSFORMATIONS OF THE GENERAL (NON-NULL) ELECTROMAGNETIC FIELD TENSOR
f = (e, h) IN A LOCALLY MINKOWSKIAN REFERENCE SYSTEM
Field values At start After canonical duality rotation

At start e, h e” andh” perpendicular,and

e” greater than h”
After canonicd Lorentz e’ and h’ parallel to each e parallel to z axis;h = 0
transformation other and to the z-axis
 CLASSICAL GEOMETRODYNAMICS 541

Under a Lorentz transformation the components of the three tensors, f and f,

and *t, transform alike. The angle Q!therefore remains unchanged. It is a signi&
cant and Lorentz invariant scalar property of thejield, f. We shall call the angle CY
the complexion of the electromagnetic field.
When the field f is a null field, with
(e.h)=O and h2-e2=0
or
f X f = 0 and f2 = 0
then Eq. (24) for the angle (Y becomes indeterminate. Then the complexion is
not definable on a purely local basis.

THE SQUARE OF THE MAXWELL TENSOR AND THE ALGEBRAIC RELATIONS ON

THE CURVATURE
Now return to the case where the field is not a null field. Evaluate the Maxwell
square-or stress energy tensor-of the original field by using its equality to the
Maxwell square of the extremal field [see Eq. (21)] or the Maxwell square of the
dual of the extremal field
&i”(f) = S,“(E) = 2&l,r - 6,“(e),
(27)
5,‘(f) = 2,‘( *t) = 2*C;n,*EYC+ sl’(a”).
Now square the Maxwell square off by multiplying the first tensor by the second.
The cross terms between t and *t that arise in the calculation reduce to zero by
reason of the identity”
2.5u*fKa = 4@u%&’ = 6,“(t x r> G-w
and the extremal property, t X t; = 0. We also used in the evaluation the iden-
tity (15). We find for the square the result
%,“5,’ = 2@) (twr - *fa*t”7 - mZ,”
= 6,‘(iy
= 6,‘W + (f x VI (29)
= 6,‘[(h2 - e2)’ + (2e.h)‘]
= 6,‘[(h2 + e’)’ - (2e X h)‘].
The proof of the same result in the case of null a field is even simpler,
lo A special case of the relation
A,,B’” - *Ap~*B’OI = Ms,‘A,,B@,
whichis valid for every pair of antisymmetrictensors,A, B, in 4-space.
542 MISNER AND WHEELER

f X f = 0 and f* = 0 and the right-hand side of Eq. (29) vanishes. In summary,

the square of the Maxwell stress-energy-momentum tensor is a multiple of the unit
matrix. This beautiful and interesting relation is central in Rainich’s already
unified field theory. In terms of curvature components, it has the form
Spa&” = RraR,’ = 6,‘(~RaaRaa). (30)
Here the value of the constant is obtained by comparing the traces of the two
sides of the equation, To this relation he adds the vanishing of the trace of the
reduced Maxwell tensor.
S,” = R,” 3 R = 0, (31)
[Eq. (4)], that follows directly from the equation of definition, (13), and the
statement [Eq. (14)J that the electromagnetic energy density is positive definite:
‘Go = Roe > 0. (32)
In other words, for any time like vector v the quantity
0”&3(f)v@ (33)
is non-negative. Equations (30), (31), and (32) summarize the algebraic relations
on the curvature in already unified field theory.
It is necessary that the contracted curvature tensor R,” satisfy (30), (31), and
(32) if it is to be representable as the Maxwell square of some antisymmetrical
field tensor, f. Figure 1 derives the same conditions in a slightly different way.
It is evident from Fig. 1 that a suitable Lorentz transformation puts the tensor
R,v = ‘XPv associated with a non-null field, f, into a diagonal form,

(34)

The diagonal form necessarily has this appearance for any symmetric tensor
with a zero trace and a square which is a nonzero multiple of the unit matrix.
The tensor R,’ defines what in the language of Schouten is a two-bladed structure
(~723)in space-time at the point in question.” A rotation in the yz plane about
the x-axis leaves the tensor (34) unchanged; electric and magnetic fields remain
parallel to the x-axis (Fig. 1). That picture of parallel field vectors is also left
unchanged by any Lorentz transformation in the x, T plane. The yx plane and
the XT plane are the two blades defined by the Maxwell tensor Z&‘(f), and there-
fore in turn defined by f itself.
In diagonal form the Maxwell tensor is characterized by a single parameter.
To this parameter there are added only four additional parameters by the
‘1 We are indebted tb Professor Schouten for several illuminating discussions.
 CLASSICAL GEOMETRODYNAMICS 543

TABLS IV
PARTIAL CLASSIFICATION OF RIEMANNIAN GEOMETRY

Feature Name Remarks

R %w -0
- Uncurved space All such 3-spaces recently
classified by L. Markus, in
publication.

R,, = 0 Pure gravitation field The 10 “local” components of

R a~Ya are zero, but the
other 10 “remote action”
components of R-or8 have
to be found by solving the
differential equations
R,, = 0 for the metric. No
free components for R,,
R"=(). Roe > 0; “Electromagnetic Rieman- Five free components for
R:“R,’ ’ nian geometry” or “geo- R,, . Met.ric must be found
= b,“(fR,JW) metrodynamics” by solving field equations
= (h2 - e2)2 + (2e.h)* of already unified theory.

R’=(). Roe > 0; Null field; e.h = 0 and h* - R,, = Z,,,(f) = k,k. , where
R:“R,’ I: 0 ea = 0; a special case of k is a null vector; only
electromagnetism three free parameters in
R,, .

R,’ = 0; Go > 0; Static field; by a change of Extra non local (differential)

R,“R,’ # 0; names (duality rotation) requirements imposed on
(Y = constant for all space can be translated into a R,, in addition to the
and time condition where there is an standard field equation of
electric field but never any already unified field thoery
magnetic field; another
special case of electromag-
netism

R,, arbitrary Unrestricted Riemannian No physical laws

geometry

general &parameter proper Lorentz transformation, because rotations in the yz

and XT planes have no effect. Therefore a total of five parameters characterize
the contracted curvature tensor, Rpv , of “electromagnetic Riemannian geome-
try” or “geometrodynamics” (Table IV)--this despite the fact that the general
R,, has 10 distinct components, and despite the fact, that (29) and (31) can be
said to constitute 10 conditions on these 10 components. Evidently these non-
linear algebraic equations are not all independent.
It is not only necessary-as previously shown-but also su@cient that the
contracted curvature tensor R,, satisfy the Rainich conditions (30), (31), and (32)
544 MISNER AND WHEELER

in order that one be then able to represent R,, as the Maxwell square oj an electro-
magnetic field tensor, .fPy. Moreover, this field tensor is unique up to a duality
rotation. We call this field tensor the Maxwell root of the Ricci curvature tensor,
Rsy . We give separately the proofs for the cases where RPv is not a null tensor
and where it is a null tensor (RCyRPY = 0; only 3 free parameters left in R,,).
(1) Form the “Ricci part” ETv“” of the Riemann curvature tensor
&,F’ E >$i( - 6/R.,’ + ti/R’R,’ - &yR7’ 4- &‘R,‘). (35)
It has the same symmetries as the Riemann curvature tensor, contracts to the
same Ricci curvature tensor,
E,,*’ = --35R,’ + 2R,’ - sSRTY + 2R,” 8, = R,’ = R,,*‘, (36)
by virtue of the condition R,” = 0, and introduces the antisymmetry we need
for taking the Maxwell root. (2) Defme the extremd Maxwell root & of the Ricci
tensor-a pure electric field-up to a single f sign by the equation

f,& = -%E,,,, - ~(R,BRaB)-li2E,,,a~,y6 . (37)

Find any given component, ,$, up to a f sign by setting (u, 7) = (P, V) and
taking the root of (37). Then use (37) to determine the relative sign of different
components, & and & . The consistency of the magnitudes is guaranteed by
the Rainich conditions.
The prescription just given for the Maxwell square root is checked most easily
in a Minkowskian reference system where the Ricci curvature tensor has the
diagonal form
- (50J2
- (EoJ2
R,’ = , (38)
+ &od2
i + (tod2 i
where fol is a real positive number, known as soon as R,’ is known. Use this
number to define an antisymmetrical extremal field tensor, f, of which all the
components vanish in the present Lorentz system except to1 and .& = -lo1 .
Then the dual of this tensor has all components zero except

(*a*3 = -(*.i>az = -h .

Thus the extremal field has the properties

= magnetic component of electric component of 2

reduced extremal field reduced extremal field >

= - f (R~,R,~)~/~;
 CLASSICAL JGIWMETRODYNAMIC~ 545

t X t = 0 = (magnetic component). (electric component) ;

(*02 = +kd2 = -t2; *( x *t = 0. (39)

In this same Minkowski frame the only nonvanishing distinct components of
the tensor E of (35) are
EOlo’ = -%@1’ + Ro? = (tod2
and
E213= - %(Rt + R,2) = - (~oI)' = - (*623)‘. (40)
Therefore the general tensor component of E can be written in the covariant
form
E agya= -La&a - *Lll*tra . (41)
The product of this tensor by itself has the value
EczmE% = %od2(- EwYh"+ *taa*6v)
(42)
= uLJw2(- f&t" + *L9*tw)
by virtue of the properties (39) of the extremal field. Multiply (41) by -35 and
(42) by -$$(R”R,)-“” and add, to cancel out the terms in the dual field. There
results Eq. (37) for the components of the reduced extremal electromagnetic
field tensor, Being a tensor equation true in the simple coordinate system, it
must be true in all coordinate systems. So much for the machinery for taking
the Lorentz square root when the Ricci tensor is non-null; that is, when its square
is a nonzero multiple of the unit matrix.
Consider now the other case where the contracted curvature tensor is a null
tensor,
R,,RpY = 0, (43)
but a tensor which is not identically zero. As before, discuss the tensor in a
Minkowski reference system, where the components, g,,” , of the metric tensor
have their Lorenta values. The frame of reference is still free to the extent of a
G-parameter Lorentz transformation. Make such a rotation in 3-space (3 parame-
ters) as will diagonalize the 3 X 3, space-space part of R,, ; thus, RI2 = R23 =
R31= 0. The vanishing of the square of R,’ (Table IV),
(P) x.otentzRpaR~v
= 0, (44)
then makes conditions of the form
time-time : -Roe’ -I- RI: -I- Rzo” -t R3: = 0, (45)
time-space : -RooRol + RolRn + 0 + 0 = 0, (46)
546 MISNER AND WHEELER

spacel-spacer : -Rot + R11"+ 0 + 0 = 0, (47)

spacel-space:, : -RmRoz + 0 + 0 + 0 = 0. (48)
We conclude from equations of the type (48) that only one of the components,
Rol , Ro2, Ro3, can differ from zero. Let this nonzero component be Rol, and let
the sense of the x axis be so chosen that Rol is a negat’ive quantity, -2~~. Then
R"' is positive, corresponding to a Poynting flux in the plus x direction. Then
from Eqx. (45)-(48) plus the requirement, R00 > 0, of positive definite energy
density it follows that the Ricci curvature tensor has the form

This tensor may be written

R,, = 2k,kv , (50)
where k is the null vector
k, = (-K, K, 0, 0);
(51)
k2 = k,lc" = 0.
Being covariant and true in one reference system, the decomposition (50) is
valid in any reference system.
There is no Lorentz transformation that will diagonalize a null Ricci tensor
any more than there is a Lorentz transformation that will make a null vector
time-like, or parallelize field vectors e and h that satisfy the null condition,
(e-h) = 0, h2 - e2 = 0. The theorem that every symmetric tensor can be re-
duced to diagonal form does not hold when the metric is indefinite, a circum-
stance for the elucidation of which we are indebted to our colleague, Professor
V. Bargmann. This feature in no way prevents taking t’he Maxwell root of (49):

(52)

or
e = (O,K,O); h = (0, 0, K) (53)
as one checks by direct substitution in formula (13) or (14) for the Maxwell
square. The tensor (49) describes a flow of energy in the x-direction at the speed
of light, and (52) or (53) decompose the Poynting flux into factors, e and h.
 CLASSICAL GEOMETRODYNAMICS 547

The polarization direction alone is free in the Maxwell square root. Application
of a duality rotation of parameter (IL to the field (52) rotates the polarization
vector by the angle a about the direction of propagation, leaving unaltered its
Maxwell square, RPy .
The prescription for the Maxwell square of a null Ricci tensor can be sum-
marized in covariant form: (1) Take the “ordinary” square root, k, , according
to Eq. (50). (2) Take a four vector v which (a) has unit magnitude,
v” = v,v(l = 1 (54)
and (b) stands normal to k,
lc,va = 0 (55)
and (c) in the special Minkowski frame of (49) and (52)-for example-has the
components
v = (0, 0, 1, 0). (56)
(3) Form the antisymmetrical product
f&w = kpvv - k”V, (57T)
or-in the so-called intrinsic notation-
f = khv. (571)
(4) Then the reduced jield tensor (57) is a Maxwell square root of the null tensor
R,, , and apart from a duality rotation, is the only Maxwell square root of RPv .
The reduced field f is a null field in the Minkowski frame which we have used
by preference, as one easily shows from (51), (54), and (55). Therefore it is a null
field in any other frame of reference. The same is true of the field after it has
experienced a duality rotation.
The effect of a duality rotation on f can be stated in terms of the constituent
factors of f; thus, k is left unchanged, and v is rotated about k. However, v is
never uniquely determined by f: the new vector, v’ = v + (constant) k, gives
the same field, f, and satisfies the conditions (54) and (55) as well as does v itself.
Moreover, the descriptive word “polarization” should be taken cautiously, as
indicating in the present context only the orientation of the mutually perpen-
dicular pair of 3-vectors, e and h. It does not stand for the polarization of a
monochromatic directed wave train, which could not be determined without a
knowledge of the field at nearby points-even if one had a monochromatic
directed wave to talk about!

The Tpwo-Way Connection between Field and Curvature.

In summary, any reduced electromagnetic field tensor, fey , null or not, pro-
duces a Ricci curvature that satisfies the Rainich conditions. Conversely, any
548 MISNER AND WHEELER

Ricci curvature tensor, null or not, that satisfies the Rainich conditions, has a
Maxwell square root, j<k . that is unique up to a duality rotation. In addition, in
the non-null case the Ricci curvature-or the field-determines a structure with
two blades, A and B, at each point in space time.

The Blades Do Not Mesh.

How is the geometric structure at one point in space-time related to that at a
neighboring point? A geometrical description of the electromagnetic field has to
answer this question. It therefore appeared useful to us to raise the issue, what
happens when one moves in the local two dimensional surface, or tangent plane,
of blade A to neighboring points where blade A is tilted at slightly different
orientations. Will one arrive in this way at a well defined two dimensional
surface,
2” = x.~yu, v)? (58)
Or will the pattern of the blades lead to structures like those in a spiral disloca-
Con in a crystal, so that one can get from one point to any other point by mov-
ing about on blades A via a suitably selected route? In this case no surface of the
type (58) will exist. Does the demand that blades fit into a surface of type (58)
lead to Maxwell’s equations? We investigated and found that this demand is too
restrictive to be satisfied by the general solution of the equations of electromag-
netism. Therefore we decided to let Maxwell’s equations speak for themselves.
We expressed the electromagnetic field as the Maxwell root of the Ricci tensor,
where

.fpy= eroltpv
, or f = eY$,
(59)
[ = extremal Maxwell root of R,, ,
inserted into Maxwell’s equations,
f”‘;” = 0,

*f”” ;u - 0,

and learned at first hand what Rainich had already learned before us about
the true geometrical content of electromagnetism.

Four of Maxwell’s Equations as Consequences of the Identities Satisfied by the

Curvature or Stress Energy Tensor.
The eight equations (60) of electrodynamics take the form
0 =fpYGY = ({c”;y + *(“&~/ax’) cos (Y + (*4”y;y - ya(Y/ax’) sin ff,
(61)
0 = *fcyiy = (- pviv - ~Q’aa/a~‘) sin LY + (*[pVzy - yacr/aq cos (y,
 CLASSICAL GEOMETRODYNAMICS 549

or, by simple combination

~~v;y + *Q”aa/ax” = 0, (6%
*fb”;” - ~P”a~/ax” = 0. (63)
We now combine these equations in such a way as to separate the information
they give about t and about CL Multiply (62) by &,, and sum over p. Use the
properties (39) of f and the identity

fap*EP” = 0, (64)
which, being true in a simple Minkowski frame, and being covariant, is true in
any frame. We find the result

&LPY;v = 0, (654
and, by similar reasoning starting from (63),
*&#*(c”;” = 0. (65b)
Only four of the eight Eqs. (65a), (65b) are independent, as may be seen by
going to the simple coordinate system where to1 and *& = -& are the only
nonvanishing components of h. In these coordinates we see that the eight Eqs.
(65a), (65b) are equivalent to the four independent equations
SwtL”;” + *Lp*SC”;” = 0. 6-w
The identity”

and a similar equation for *f in place of r puts Eq. (66) in the form
$f&pS’” + *Eap*SP”);” = 0. (67)
The quantity in parentheses is the Maxwell stress tensor. This quantity is to be
identified with the Ricci curvature, whose trace, R, is zero so that (67) can be
written in the form
[R," - $@,"R];" = 0. 033)
However, this condition is no requirement at all. Bianchi proved that the Ricci
tensor calculated from an ar6itrary metric tensor will satisfy (68) identically. It
l2 The second line of this identity is a special case of the identity

%A=Vr.+ + B.o:r + B8& = *A,m*Ba?~

which is valid for any two antisymmetrical tensors in 4-space.

550 MISNER Aim WHEELER

has long been known that half of Maxwell’s equations are given by these Bianchi
identities. Now what have the other half of Maxwell’s equations to say about the
curvature of space?

Maxwell’s Other Four Equations Demand That a Certain Vector Combination of

the Curvature and Its First Derivative Shall Have Zero Curl.
Multiply (62) by *fb,, and (63) by ,$, , sum over P, use (15), and find the result

%3wY: Y + &9/&“;” + 6~‘~2aa/ax” = 0

or
acu/axa= aS, (69)
where we make the dejkition

%8 = - (*S;BlrY;” + &BP**f”‘:Y)/f2. (70)

This vector expresses itself in terms of the Ricci curvature in the form
‘yB = ( -g)1’21BX~y]RXY’~RRyY/Rr7R”‘. (71)
For proof of (71), recall the expression (41) for the Ricci part (35) of the Rie-
mann curvature tensor in terms of the extremal field, 5; and form the first con-
tracted covariant derivative of t,his tensor:
Er6?, = -&5’;7 - *~yS+?7 - [y’;T~5* - qY8;r~~5r
= >$‘(-g’@R6’ + gy’R”B - gb’RyB + g”‘R”);, (72)
= $$(R@:Y _ RY@:“).

Here we use the fact that every component of the covariant derivative of gy”
vanishes, and employ the Bianci identities, Rari7 = 0, to annul all but two terms
among the eight that arise from the differentiation. Define
Km:= %X-g>““h&4%?’
= f5( --s)“‘[r~~~l (- 4q9t”’- *.ha*t”‘)
= -&@*h + *L&s
= +&( -g)“‘[r&w]+$( -60”RBy + 8oryRBp- +*Rp” + 68’Rrry) (73)
= )/i(-g)l’z[ySw] (S,‘R/ - s;R/).
Finally, form the product
FUSYsET8”;7 = -4T2(*&9P;7 + ta5*E5’;r)
= M( -g)1’2[$pv] (6,‘Rp’ - 6avRR,“)R68’’ (74)
= fG( - g)1’2[c&p]R68’ ‘R/.
 CLASSICAL GEOMETRODYNAMICS 551

The equation

which follows from [ X t = 0 by differentiation, as well asthe identity (64)) were

used in computing (74) from (72) and (73). Now divide (74) by a(&” = >$R,,R”’
to obtain (71), as was to be proven.
The vector ap of (71) has a well defined existence in any Riemannian space
where the Ricci curvature tensor R,, is non-null and differentiable. From such
general Riemannian spacesthe geometry of the Einstein-Maxwell theory is dis-
tinguished by the circumstance that this vector-as shown by Rainich-is not
arbitrary, but is the gradient of the complexion, a, of the electromagneticfield.
Consequently it follows that the curl of cupmust vanish:

cQ;r - a!?;B= cYg,y- cYr,p= 0. (75)

Conversely, when the curl (75) vanishes, then the line integral of the vector LYE
of (71) from some selected point 0 to any arbitrary point 5’ definesa scalar com-
plexion,

(7’3

up to an additive constant, (~0, provided that the region of space under con-
sideration is simply connected, and provided that the line of integration does
not include any point where R,,R”’ vanishes. For a multiply-connected space
it is necessary to replace (75) by the demand that the line integral of “0 around
any closed path shall be an integral multiple of 2a,

! ‘q9 dx’ = 27rn (77)

provided the line of integration does not touch any null points. This condition,
plus the algebraic requirements of Rainich
R,” = 0; Roo 3 0; R,‘RPP = &!(~R,,R”‘), (78)
gives the necessary and su$.cient conditions that a Riemannian geometry shall
reproduce the physics of Einstein and Maxwell, provided that the curvature
R,, is non-null. It may well be that trivial changes in the statement of the theo-
rem will cover the case of null fields, but this point remains to be investigated.

WHY RAINICH-RIEMANNIAN GEOMETRY?

We shall give the name Rainich-Riemannian geometry to any 4-space with
signature - + + + which satisfies (77) and (78). The question poses itself
insistently to find a point of view which will make Rainich-Riemannian geometry
seem a particularly natural kind of geometry to consider. Presumably a varia-
552 MISNER AND WHEELER

tional principle based on an appropriate scalar Lagrange density will prove the
most natural starting point to discuss this question. Whatever the deeper sim-
plicities may be, it is extraordinarily beautiful that the Rainich-Riemannian
geometry (77), (78) of empty curved spacereproduces all the standard machinery
of Maxwell stresses, electromagnetic waves, and generation of gravitational
forces by field energy.
III. CHARGE AS FLUX IN MULTIPLY-CONNECTED SPACE: ELECTROMAG-
NETISM WITHIN THE ARENA OF A PRESCRIBED METRIC; PLAN OF THIS
SECTION; TOPOLOGY; DIFFERENTIAL GEOMETRY; MAXWELL’S EQUA-
TIONS AND CHARGE

A. ELECTROMAGNETISM WITHIN THE ARENA OF A PRESCRIBED METRIC

“Already unified field theory” or geometrodynamics appears to consist en-
tirely of differential equations and algebraic relations on the curvature (Eqs.
75 and 78)-in other words, appears to have an exclusively local character-so
long as space is assumed to be simply connected. As soon as the possibility is
admitted of two or more topologically distinct routes to pass from one point to
another (Fig. 3), then in addition one has to impose a periodicity condition

aa dx’ = 2rrni
ci
for each topologically distinct closed circuit Ci that is free of null points. Only
when this condition is satisfied will the electromagnetic field, f = eiut;, be a single
valued function of position. The condition (79) has a nonlocal character. This
circumstance forces us to ask, what additional features of a nonlocal character
appear when Einstein-Maxwell physics goes on in a multiply-connected space?
To recognizechargeas a nonlocal manifestation of chargefree electrodynamicsin a
multiply-connected space, we will find it helpful to revert from the ideas of the
more familiar geometrodynamics to language of Maxwell theory in a preexisting
spacetime continuum. In keeping closeto the most easily visualized terminology,
we will not deny that electrodynamics is intrinsically nonlinear. A field has an
energy density that curves space. This curvature affects the propagation of the
field. A field of twice the strength therefore propagates differently from the
original field. This nonlinearity will remain hidden behind the Scenesin the
following analysis. Attention will be limited to Maxwell’s equations and their
consequences.However, these consequenceswill remain valid when Maxwell’s
equations for the field are supplemented by Einstein’s equations for the metric.
The concept of multiple connectednessis topological in character and logically
precedesany idea of metric. Section B therefore summarizes the necessary topo-
logical preliminaries, including the concepts of continuity, manifold, boundary,
homology class, Betti number, differentiable manifold, and coordinate patches.
No space topologically inequivalent to an open subset of Euclidean space can
 CLASSICAL GEOMETRODYNAMICS 553

be covered without singularity by a single nonsingular coordinate system. This

circumstance makes the analysis of vectors, tensors, and other quantities by way
of components-in the traditional spirit of tensor analysis-less appropriate than
an intrinsic type of calculus, such as is familiar from vector analysis in S-space.
Section C uses this intrinsic notation of Cartan side by side with the familiar
notation to define the needed concepts of differential geometry in a space not yet
endowed with any metric: vectors, alternating tensors, cross product, curl,
Stokes theorem, the metric-free half of Maxwell’s equations, conservation of flux,
charge, vector potential, de Rham’s theorem, and the issue of electric versus
magnetic charges. The rest of Sec. C traces some of the more important additional
consequences that flow from existence of a metric: duality, divergence, differen-
tial operators, and relations and integral formulas involving these operators.
Section D analyzes electric and magnetic fields on one or two space like surfaces
as initial value data for Maxwell’s equations; the new feature of charge brought
in by multiple connectedness; and the description of charge by way of the theory
of harmonic vector fields.
B. TOPOLOGY

Topology and Point Sets.

Topology13 is the study of a nonquantitative idea of “nearby”. This idea itself
is not ordinarily axiomatized; instead, topology axiomatically defines what is
meant by a neighborhood, an open set, or a closed subset of space. The relation-
ship of these ideas to “nearby” may be suggested as follows: (1) A neighborhood
of a point is a subset containing all points sufficiently near z. (2) An open
subset is one which contains all points sufficiently near any of its points. (3) A
closed subset C is one which contains every point that is arbitrarily close to C.
Let two sets X1 , Xp be equivalent as sets, so that they also contain the same
number of points. Then the two sets are equivalent as topological spaces, that
is homeomorphic, if there is a l-l transformation h of X1 onto X2 , called a homeo-
morphism, under which the open sets of X1 are in l-l correspondence with the
open sets of X2 . The open sets define the notion of “nearby”. Therefore we may
say that a homeomorphism h is a l-l onto correspondence of X1 and X2 which
always makes nearby points in one space correspond to nearby points in the
other, i.e., it is continuous, and so is its inverse.
We have found outselves forced to consider spaces that are topologically more
general than those usually treated in physics. Nevertheless, we shall restrict
ourselves to a very special class of spaces, called manifolds: An n-manifold (27’)
is a topological space which (1) is locally Euclidean of dimension n, (2) is Hausdw,
13 A standard text on point set topology is one by Kelley (24). For a brief development
of the theory see a book by Pontrjagin (26). Statements of definitions and theorems from
point set topology which are most pertinent for a study of manifolds can be found in a book
by Aleksandrov (26).
554 MISNER AND WHEELER

and (3) has a countable basis. (1) A space is locally Euclidean of dimension n if
each point has a neighborhood homeomorphic to Euclidean n-space. (2) A space
is Hausdofl if every two distinct points have disjoint neighborhoods. Effectively
Hausdorff means that no two distinct points are arbitrarily close to each other,
but this intuitive notion may also be made precise in other ways. (3) A space
has a countable basis if there is a countable collection of open sets such that every
open set is a union of open sets in this collection. This condition is imposed so
that a manifold will not be uncomfortably large.

Examples of Manifolds.
The prime examples of manifolds are the Euclidean spaces. Euclidean n-space,
R”, is the space of all n-tuples of real numbers, x = (x1, x2, v . . zn), with the
topology customary in analysis. The open subset of R” defined by
x*x = qxy < 1 WV
is called the n-ball. It is also an n-manifold, and is in fact homeomorphic to R”
by the transformation x -+ x(1 - x.x)-‘. Another familiar example of a mani-
fold is the n-sphere S” which is a subspace of R”+* consisting of all points x
satisfying the condition
TL+1
x.x = c
z=l
(2y = 1. (81)
From S” we may obtain projective n-space P” by identifying antipodal points
on S”; i.e., a point x of P” is an unordered pair (x, -x) of points x of Rn+l satis-
fying X-X = 1. In general relativity, S3 and P3 have been used in cosmological
models (28). Another manifold, the 3-torus T3, is encountered in theoretical
physics in the frequent instances where one imposes periodic boundary condi-
tions. The n-torus T” is a space whose points are families of points in R”, x =
{X + 27rm) m where m ranges over all n-tuples of integers m = (ml , m2 , . . . , m,) .
As a final example we construct a space IV, which illustrates in a simple way some
of the topological possibilities which we are interested in investigating in relation
to the idea of charge. Starting from the 3-sphere S3, we consider only those points
x = (CC’,x2, x3, x4) which satisfy both x.x = 1 and j z4 / < 1 - E. We then iden-
tify each point (x1, x2, x3, 1 - C) with (x1, x2, x3, - 1 + e). This may be called
the pierced sphere, or W1 . By cutting out Ic pairs of antipodal balls (such as
x4 > 1 - E and x4 < 1 + E above) around k pairs of points of X3 (such as x =
(0, 0, 0, &l) above) and making similar identifications of the boundaries, we
construct the k-pierced sphere Wk . (See Fig. 4.)

Product Manifolds.
When 9P is a p-manifold and amp a q-manifold, then we can construct from
them a (p + q)-manifold 3np X YIZ’ whose points are pairs (x, y) where x is any
 CLASSICAL GEOMETRODYNAMICS 555

HOLLOW SPHERE CYLINDER AODED

TOP AND BOTTOM CUT AWAY TO JOIN TOP DOUBLY PIERCED
EDGES OF CUT IDENTIFIED AND BOTTOM SPHERE

(b)
FIG. 4. The pierced spheres. When the polar caps are excluded from a sphere and the
resulting boundaries identified as indicated by the corresponding points A, B, in (a), there
results a space which may be called a pierced sphere, Wr . The identification may be
visualized by pulling the edges together through the center of the sphere as in (b), so that
the resulting manifold is the surface of a ball with a hole drilled through the center. Drilling
k such holes in a sphere, in such a way that the drill holes never intersect, gives the k-pierced
sphere Wk . In (c) a view of WZ is shown. The figures show two dimensional manifolds, while
the WI, in the text are 3-manifolds.

point of %i~’ and y any point of ~2’. Euclidean 2-space R2 is, for example, the
product R1 X RI, and similarly R” = R’ X R’ X . . . X R1 (n factors). We
note that S’ and T’ are homeomorphic,
(z + 27rnJ n -+ (x’, x”) with z’ + ix2 = e”“. WJ
(We write T’ = S’ since we usually do not need to distinguish homeomorphic
manifolds, unless the names of the points, i.e., (X + 2?m}, c T’ us (x1, x”) E S’,
are needed in a construction.) Then T” is a product of n circles, T” = S’ X S1 X
. . . X S’. It may also be seen that WI = S2 X S’.
In the examples above we have not defined the topology, or in other words,
we have not made the choice of open sets, since the appropriate choice is fairly
obvious.
Diferentiable Manifolds.
To discuss differential geometry we need both a manifold, and a concept of
dilerentiable junction on this manifold. Let a criterion be supplied which decides
which functions defined on a manifold m are to be called differentiable. Then
this criterion is called a differentiable structure on 311, and %? is called a difJer-
entiable munijold. The class of differentiable functions is however subject to
certain axioms (d7) :

For every point 5 of Eli there exists an open neighborhood U of x, and n real valued
functions 21(z), z2(5), . . z,(z) defined on U such that:
(a) The transformation x -+ (zrl(z), x*(x), . . * x,(r)) of U into Rn is a homeomorphism
of U onto an open set of Rn, so that every functionf defined in U can be expressed in terms
of the xi ,
f(z) = f(G) 22 ‘.. 4. (83)
556 MISNER AND WHEELER

(b) A function f(z) is differentiable at a point of U if and only if it is defined in an open

neighborhood W of that point, with W contained in U, and f (xl , x2 , . . . x,) has continuous
derivatives of all orders with respect to the zi for values of the xi corresponding to points
in W.

The neighborhood U with the functions xi is called a coordinate patch, and the
set of functions X1(z), X2(2), * . . z,(z) is called a system of local coordinatesin U.
It follows from the definition that a coordinate xi(z) is a differentiable function.
We apply the definition of differentiable function by way of illustration to
some of the manifolds already mentioned. On the n-sphere we may define xi(x),
where x = (~1, 22 , . . . x,+1) to be a differentiable function of the points x in
S”. Then some n of these n + 1 functions will serve as coordinates about any
point in part (a) of the axioms, while the remaining differentiable functions are
given by (b). For the projective n-space P” whose points z are pairs of points,
(x, -x), of S”, we say f(z) is a differentiable function on P” if f((x, -x)) is a
differentiable function of x ES”. For the space WI, the pierced sphere, we con-
sider x1 , x2 and x3 as differentiable functions, as they were on the sphere. The
quantity x4 is to be differentiable only in the region 1x4 1 < 1 - E.This function
is not defined at the rims that have been identified with each other. We need
another function which is defined and which can be called ditIerent,iable across
this region of identification. For this purpose we define f(z) by
j(x) = 1 - E - x4(x) for 0 < x4 5 1 - E,
(84)
j(x) = -1 + t - i(z) for -l+eIx4<0,

and insist that f(z) is differentiable for d # 0. From among the functions
x1 , x2, x3, x4, and j we can chooselocal coordinates about any point.
Although a topological manifold may be given more than one differentiable
structure (29), we shall abbreviate differentiable manifold to manifold in what
follows, and the word topology will include differentiable structure as well as
topology.

C, DIFFERENTIAL GEOMETRY AND THE FORMULATION OF MAXWELL'S EQUA-

TIONS

Diferentiable Functions and Fields

To state physical laws in quantitative form, we are compelled to deal with
vectors and tensors in a curved space.These quantities are not most conveniernly
described in the familiar language of tensor analysis when the space is multiply
connected. A new or intrinsic formulation is required.
Tensor analysis is not adequate. It demands a nonsingular coordinate system
with respect to which one can give the components of vectors and tensors. How-
ever, according to the definition of a differentiable manifold, a single nonsingular
 CLASSICAL GEOMETRODYNAMICS 557

coordinate system is not enough to cover a manifold that is topologically in-

equivalent to an open set in Euclidean space. It is essential for our physical
applications that we be able to distinguish singular tensor fields from nonsingular
ones. A singularity in a field will ordinarily imply a localized source term in the
differential equations. Such a source term will represent a nongeometric charge
or mass, which has not been eliminated from the theory, but merely idealized
to a point charge or a point mass. To investigate the content of pure geometro-
dynamics we therefore exclude all singularities in the fields. Consider for example
the two dimensional surface of the unit sphere. The polar angles, 8, ‘p, are or-
dinarily called coordinates, but they do not cover the surface without singu-
laxity: (1) The metric,
ds’ = de2 + sin2 edv2, (85)
has a component, g”‘+” = sin-’ 0, which goes to infinity at the two poles, 8 = 0
and e = P. (2) A vector field with apparently nonsingular components, vg = 1,
Vq = 0, is not well defined with respect to direction at either pole. This singu-
larity in the vector v = grad e raises a question about calling e a coordinate.
(3) Along the Greenwich meridian the angle cp suffers a discontinuous change
from 0 to 27r, in this region making this quantity also inappropriate for a co-
ord nate. (4) Only in the region 0 < e < ?r, 0 < cp < 2~ do the polar angles
supply a well defined “coordinate patch.”
Another %oordinate patch” is needed. Let a small circular ink pad be cen-
tered at the north pole and let the pad be moved down the Greenwich meridian
to the south pole. We ask for a new coordinate system that is regular everywhere
in the blackened region. Of course the new coordinate system may be regular
over a wider region. For example, consider the point 0 = 7r/2, y, = n/2. Let this
be north pole of a new set of polar angles, 0’, (o’, in terms of which (1) the pre-
vious north pole is 0’ = 7r/2, ‘p’ = 7r/2 (2) the previous Greenwich meridian runs
from 8’ = x/2, ‘p’ = 7r/2 to 8’ = 7r/2, ‘p’ = 3*/2 (3) the previous south pole
lies at 8’ = 7r/2, cp’ = 3~/2. There is evidently a wide region where the old and
new coordinate patches overlap. Moreover, at any fixed point on the sphere at
least one non-singular pair of coordinates 0, cp or 0’, ‘p’, is well behaved. If we do
not know what nonsingular coordinates are, then we have not yet defined the
differentiable structure of the manifold. However, as soon as we have said what
we mean by a differentiable function, then the idea of a diserentiable tensor jield
is well defined: A tensor is differentiable at a point x if its components with re-
spect to a (nonsingular) coordinate system around II: are differentiable functions
at 2.
Intrinsic Representation of Tensors.
In the general differentiable manifold it will be impossible to describe a field,
such as the Maxwell field, by giving its components Fab with respect to a single
558 MISNER AND WHEELER

set of coordinates. If components are to be used at all, they have to be given with
respect to the coordinates of the several distinct coordinate patches. In each
patch there is great freedom of choice about the coordinate system. Consequently
the components, F-0 , of the Maxwell field in the several coordinate systems are
not individually so important as the concept that can be abstracted from them:
the intrinsic value, F, of the Maxwell field. This procedure of abstraction is
familiar from the analysis of Cartesian vectors. To say that the velocity vector
v is known is to say that on demand one can give the components of v in any
nonsingular coordinate system. We propose to generalize this abstract formalism
to the case of curved space and tensors of order higher than the first.
In vector analysis it is convenient to express the connection between a vector
v and its components by the formula
m
v=v e,, (86)
where the components urn depend upon the choice of base vectors e, . “Intrinsic
differential geometry” expresses the same type of relation more explicitly in the
form
v = va grad xa, (87)
and more briefly in the notation,
v = v,dxa, 63)
where the letter d stands for gradient. Modern text books on differential geometry
(SO) even replace the boldface d by lightface d-as we also shall do later-but for
the time being we keep the boldface notation to emphasize the vectorial charac-
ter of the gradient operation.
The gradient operation, applied to any function of position, f = f(P), expresses
itself in intrinsic notation in the familiar form
grad f = ($f/ax*) grad xa
or
df = (#/axa) dx”. (89)
Covariant vectors, such as grad f, or df, and v are linear combinations of the
gradients of the coordinates and are called differential forms of the first order,
or l-forms.
Instead of using the gradients of a particular set of coordinates as base vectors
in a certain coordinate patch, one can use there alternatively as base vectors
any other set of n linearly independent vectors, Ok. Here the superscript or tells
which vector, not which component of one vector. Relative to this set of basic
vectors the components of the vector v are again read out of the expansion
v = v,oa. (90)
 CLASSICAL GEOMETRODYNAMICS 559

The components of grad f or df in such a frame are called its Pfaffian derivatives
f.ol with respect to the o”:
grad 3’ = df = f,=tia (91)
Exterior Diflerential Forms.
To describe the Maxwell field requires one to deal with a higher order tensor,
or 2-form. The simplest 2-form, (Y, is the cross product of two l-forms, u and v,
thus,

in intrinsic notation; or

%v = uy, - V&p (92)

in the familiar language of tensor analysis, or
(I! = >$+dxP A dx = >!&,,(grad ze) X (grad x’)
in a formulation that connects the intrinsic form with the component representa-
tion.
The exterior product operation A (“hook”) generalizes the cross product into
a means for multiplying vectors and alternating tensors in such a way as to ob-
tain tensors that are also alternating. This operation does not apply to symmetric
tensors. The A multiplication is defined by the requirements (1) it is associa-
tive (2) it is distributive (3) exterior multiplication of two vectors is anticom-
mutative; that is, we ask
UAV=-VAU (93)

for any two l-forms or vectors u, v. In particular, u A u = 0.

The exterior product of p vectors, or a linear combination of such vectors,
gives a p-form. Let each vector be expressed in terms of base vectors Ok. The
product of the p vectors, or the linear combination of such products, can be
expressed in the form
a= c aalan...apo @I I\ “OL2 ,, . . . oap
.Z1<CZ!j...<ap
(94)
= (p!)%a,az.. .mpoal A “az A . . . A we’.
Here the coefficients or tensor components in the first sum are well defined. In
terms of these components those that appear in the second sum are defined by
the requirement that they be alternating functions of the indices.
The exterior product dx’ A dz’ = (grad x1) X (grad x2) ,signifies geometrically
an area spanned by base vectors along the x1 and x2 coordinate axes. Similarly
the product dx’ A dx2 A dx3 represents the parallelepipedal volume spanned
by three base vectors.
560 MISNER AND WHEELER

The commutation rule for a p-form a and a q-form b is easily found:

a A b = (-l)PQb A a. (95)
Functions and vectors are included in this scheme of exterior differential forms
as O-forms and l-forms, respectively.

Exterior Diferentiation.
The operator d which produces the l-form dj = grad j from the O-form f can
be extended to a form of higher degree,
a = (P!)-laal...,,dz”’ A . . . A dxap, (96)
by way of the definition
da = (p!)-‘da,,..., A dxal A ... A dx”lp. (97)
In the language of tensor analysis this equation takes the form
(Wa,...m,+, = LZ( - l)pf3a~2...~p+l/dxB1, (98)
where P = 0 or +l according as the indices p1 +. . &,+1 form an even or an odd
permutation of the indices CY~. . . LY~+~. The operation d, the exterior derivative,
is an alternating differentiation which generalizes the familiar operations of
gradient and curl. It is linear:
d(al + aJ = da1 + da2 . (99)
Applied to a product where the first factor is a p-form, it gives the result
d(a A b) = (da) A b + (-1)“a A db. (100)
Applied twice it gives zero
d(da) = 0. (101)
This circumstance was employed in passing from expression (96) for an exterior
differential form a to expression (97) for its exterior derivative, da: the base vec-
tors dx” = grad xol give no contribution when the operation of exterior differen-
tiation is applied to them:
ddx” = curl grad xa = 0. (102)
No such simplicity results in the case where the same exterior form a is expressed
in terms of a more general system of base vectors o*; thus do” = curl tia will
not ordinarily vanish.
Maxwell’s equations deal with two antisymmetrical tensors, fns and *fols ;
or in invariant terminology, with two 2-forms,
f = 3$jaadx” A dg
 c~4ssIcAL GEOMETRODYNAMICS 561

and
*f = >S*fas axa A dxO. (103)
Consider the exterior derivative of f; that is, the 3-form
df = $~(afola/az’)dz’ A dx” A dz’. (104)
Here the coefficient of a typical basic S-form, such as dx’ A dx” A dx”, with
X < p < v, has the value
a.fpv/axx + afdax’ + a.fpX/ax~, (105)
or in a locally Minkowskian frame,
div h 0, k v = 1, 2, 3),
(1’36)
curl e + ah/aT (X = 0; p’, v = space).
These expressions vanish, according to Maxwell, so one half of his equations
take the form
df = 0. (107)
Similarly, in charge free space the other half of Maxwell’s equations take the
form
d*f = 0. (108)
To satisfy Maxwell’s first set of equations it is suficient to express the field f
as the curl of a four potential
f = da;
or
+df,edx” A dz” = d(a,dxY) = da, A dxY
= (aa,/ax*)dx*A dxY = fd(aa,/ax* aa,/axs)dx” A d$
or
faa = aa,/ax” - aa,/a2. (109)
Then the exterior derivative of f vanishes automatically:
df = dda = 0. W)
Integration: Motivation and Method.
From the differential properties of the 2-form f what integral properties can be
deduced? How do these consequences differ (1) when we know merely that
df = 0 and (2) when we know in addition f = da? More generally, how does one
define integrals of forms?
562 MISNER AND WHEELER

A p-form can be integrated over a p-dimensional surface to give a scalar or

coordinate-independent quantity. To combine a local q-form (q # p) with a local
p-dimensional surface element would give a result with local directional prop-
erties; but, there is no invariant way to add together vectors or other directed
quantities at different points in curved space to obtain a well-defined total.
Therefore integrals of p-forms on p-dimensional surfaces alone need be con-
sidered.
The surface of integration will require for its coverage one or more p-dimen-
sional coordinate patches. Break the surface up into curvilinear p-dimensional
parallelepipeds at least equal in number to the number of required coordinate
patches. Consider the standard unit p-dimensional cube, I’, with coordinates h”:
0 5 x” < 1 (Q! = 1,2, * * * p). (111)
Map the points of I’ into %?z,
C:IP + SK (112)
This map C may be thought of as a curvilinear parallelepiped in 3n, but we shall
call it a cube in 3n because we have no metric with which to distinguish “curved”
from “straight”. In terms of coordinates the cube takes the form
Xa = C”(X’, x2, f * * A”). (113)
where the C” are differentiable functions. A great number of different cubes cover
the same set of points in XL For instance
XU = C”(1 - x1, x2, . . * X”) (114a)
and
XU = C(X”, x1, *. * A”) (114b)
are cubes in X different from C, and with the opposite orientation to C, while
XU = CU(l - x’, 1 - x2, x3, . . * X”) (115)
is a cube in x different from C, but with the same orientation. The orientation
of the cube governs the sign of the integral we are evaluating, but other differ-
ences in cubes that cover the same points of 312 will not influence the value of
the integral. If f :I’ -+ 1’ is an orientation preserving homeomorphism of 1’ onto
itself, if in other words it has a positive Jacobian, we will not distinguish the
cube x = C(h’, X2, . . . X”) from the cube 2 = C(f(X’, X2, . . . A”)). We should
therefore speak of an “oriented cube in ST.” Substitute the parametric description
(113) of the surface into the expression (96) for a p-form, a, to get the mapping
C:a + a’, (116)
 CLASSICAL GEOMETRODYNAMICS 563

of the form a on 311into the form a’ on I’:

a’ = (P!)-‘aoll...,,dz”‘(X) A . . . A dP(X)
= (p!)-‘a,,...,,(dx”‘/aXB’)dXB’ A . . . A (&P’/X?p)dXB~J
(117)

Replace the volume element dX’ A . . * A dXp by the volume element of the
standard unit cube, dx’ dx2 . * * dXp and dejine the integral of a over the oriented
cube C as the quantity

The appearance of the Jacobian in the integral gives assurance what the value
of the integral is independent of (a) the choice of coordinates xLI in w and (b)
the choice of the map I’ ---f 311used to represent the oriented cube in 311.
Having defined an integral over one oriented curvilinear p-cube, we have now
to piece t,ogether the entire p-dimensional surface in in as the sum of a finite
number of such oriented p-cubes,
c = Cl + CP + . . . + CN (119)
Instead of concentrating our attention on such surfaces, it turns out to be more
convenient and more significant (see Homology Theory, below) to include as
well a slightly more general object, a p-chain, c, which is simply a formal linear
combination of a finite number of oriented p-cubes:

c = 2i=l 2ci.
It might appear that the only algebraic structure necessary is f signs, so that
if C is the oriented cube defined by the map (113), we may write -C for the
oriented cube defined by (114). However it may happen that the same cube C
appears several times on the boundary of a surface and we therefore wish to
write nC, where n is a positive or negative integer or zero. If the coefficients si
in equation (120) were required to be integers, we would call c an integral chain.
In connection with integration, it is more appropriate to let the coefficients si be
any real numbers, so that we have in Eq. (120) a real p-chain. The integral of a
p-form a over the p-dimensional surface, or p-chain, c, then is defined as the sum
of elementary contributions of the form (118) :

seazesi s a= 5 ,yi a”,

i=l Ci i=l s P
(121)
564 MISNER AND WHEELER

Surfaces or p-Chains, and Their Boundaries: Homology Theory.

Particularly interesting is the relation between a p-chain or surface c and its
boundary, &. This relation is very simple in the caseof a cube, C. Use the symbol
C’+ to stand for the jth upper face of C; that is, for the (p - 1) dimensional cube,
Xa = C”@‘, . . . if-l, 1, p+‘, . . . A”). (122)
Likewise let C’ stand for the lower face,
XU = C”(X’, . . . p, 0, Q+l, . . . A”). (123)
Then define the boundary of the oriented cube, C as the chain

ac = g ( -l)J-‘(ci+ - C’-). (124)

The p-dimensional cube has a (p - 1) dimensional boundary. It might be

thought that this (p - 1) dimensional surface has in turn a (p - 2) dimensional
boundary. However, being closed, the surface evidently has no boundary at all.
One comes to the same conclusion by applying Eq. (124) twice in succession.
The result derived in this way for cubesapplies to volumes and surfaces and more
generahy to any p-chain built up out of cubes.
Let a p-chain be expressedin the form
c = c 2ci. (1%)

Then we define the boundary of the p-chain in terms of the boundaries of the
constituent cubes, C; , in nZ, through the equation
a~ = C siaci. (126)

When the boundary of the p-chain vanishes,

de = 0, (127)
then c is called a cycle or closed p-chain. For example, a closed orientable surface
which is broken up into cubes and represented by a chain c will be represented
by a closed chain. Whether the chain is closed or not, whether ac vanishes or
not, it follows from the formula adCi = 0 and from the Definition (125) that the
boundary of any chain is a closed chain; i.e., has no boundary:
sac = 0. wfu
Table V gives a few illustrations.
The notion of boundary as now defined is adequate for deriving the theorem
of Stokes in a general form, and from this as a special casethe theorem of Gauss
with its electromagnetic applications. A fuller analysis of boundaries and their
 CLASSICAL GEOMETRODYNAMICS 565

TABLE V
EXAMPLES OF ~-CHAINS, BOTH CLOSED (-CYCLES) AND OPEN”

Dimen- Is itc;8c$wl
sionality, Description of p-chain, c Its boundary, ac
P (= cycle)?

open line no 2 end points

closed line in a Euclidean space yes 0
closed line about minor yes 0
circumference of a torus
closed line about major circumference yes 0
upper face of a disc no circle
surface of a sphere in flat a-space yesb 0
surface of a sphere inscribed about. mouth of one worm- yesC 0
hole
2 surface of a torus yes 0
3 volume of a sphere in flat a-space no sphere
3 3.dimensional closed space yes 0

8 There is no need to include in the table a last column to say that the (p - 1) dimen-
sional boundary, i%, of the p-cycle, c, has itself in every case no boundary: a(&) = 0.
byes; and it also bounds a certain volume interior to the sphere.
c Yes; but there is no limited volume interior to the sphere which the surface can be
said to bound.

relationships is given by that branch of algebraic topology which is known as

homology theory14. Some of the ideas of this theory are needed to analyze charge-
free electrodynamics in a multiply-connected space. Consider by way of example
a sphere drawn in 3-space around one mouth of a wormhole-or, in the lower di-
mensional illustration of Fig. 3, consider a circle drawn on the indicated 2-space
around one mouth of the handle. As a consequence of the existence of the worm-
hole, there is no well-de$ned interior to the sphere or the circle. An imaginary
being supposedly locked up on the inside can enter the wormhole, pass freely
through it, emerge from the other mouth, and come around outside and look at
the exterior of his prison. The same topological information can be stated in an-
other form by thinking of the spherical boundary as a bubble. Let the bubble be
shrunk but not be allowed to break. It contracts into the interior of the worm-
hole. Let, it be pushed through the tunnel and out the other side and expanded
about the other wormhole mouth. Let the positive sides of the original sphere
and of the new sphere be defined by the direction of a line of force that goes
through the tunnel and crosses both spheres. Let the two oriented surfaces or
14 Basic theory in extenso is given by Eilenberg and Steenrod (31); the forthcoming
Vol. II is t,o consider practical methods of computation. Presentation of the theory in a
less extensive form, with applications, is made by Seifert and Threlfall (92). For details
and additional intuitive insight see Lefschetz (.?a) and Seifert and Threlfall (94). Rigorous
treatment of cubical singular homology theory is given by Serre (95).
566 MISNER AND WHEELER

2-chains so defined be called cl and cz , or more explicitly, cl@) and cZC2)to indicate
the dimensionality, p = 2, of the chains. Then cl and c2 are said to be homologous
to each other. More generally, when two p-chains, cl’ and ~2’ together form the
boundary of a finite (p + 1) dimensional chain, in the sense
a$+’ = CIP - QP, (129)
then we say that the two p-chains are homologous. In the example of Fig. 3, the
two l-chains or circles about the two mouths together completely bound the
two-dimensional region of the tunnel, or 2-chain, c,‘. In contrast, a circle, csl,
drawn in the 2-space of Fig. 3 in a place where it does not enclose the mouth of
a wormhole will by itself completely bound the 2-space, or 2-chain, G,‘, interior
to it:
ad = c,‘. (130)
If in this case the bounding line c, ’ is considered subject to pressure from one
side, and considered to deform accordingly, it will not have any wormhole to
pass through nor any chance to reexpand into a new circle. Instead, it will col-
lapse to nothingness. More generally, when a p-cycle cp all by itself is a boundary,
cp = acp+’ 2 (131)
it is said to be homologous zero.
Two surfaces-or more generally, two p-chains-that are homologous to each
are said to belong to the same homology class. A homology class is defined to
consist of all the closed p-chains or p-cycles that are homologous to each other.
We denote the homology class containing a p-cycle tip by (CC*}. In space that is
topologically Euclidean all the p-cycles are shrinkable to nothingness; are there-
fore all homologous zero; and therefore all belong to the zero homology class,
{ cgp} = { 0). In a multiply connected space the p-cycles are not in general all
homologous to each other. The number of linearly independent homology classes
of order p,
{CIPJ, IczP), ..‘, {CRpP} (132)
is called the pth Betti number R, of the space (Table VI). The number

x = 2 (-l)PR, (133)
p=o
is called the Euler characteristic of the space.
Consider two surfaces of the same homology class, one of which is identical
with the other except for having a wart or incipient bubble on its surface. Com-
bine the two surfaces with opposite signs. Then they cancel out except at the
bubble. The resultant surface, like a small sphere, is homologous zero. These
considerations generalize readily from surfaces to p-cycles of any order. In count-
 CLASSICAL GEOMETRODYNAMICS 567

TABLE VI
HOMOLOGY CLASSES (cp), {cP), ..* {c,$] AND BETTI NUMBERS, R,, FOR Two SAMPLE
SPACES. A ~-CYCLE IS A ~-DIMENSIONAL REGION, OR ~-CHAIN, WITH
THE SPECIAL PROPERTY THAT IT HAS NO BOUNDARY

Space: The two dimensional surface of a torus imbedded in Euclidean 3-space; more briefly,
a a-torus, !P. Its Betti numbers: Ro = 1; RI = 2; Rz = 1. Euler characteristic x = 0

Homology classes Description of typical g-cycle in this homology class

.-.

(4 A point; all points are homologous

id\ A closed curve that goes around minor circumference’
(C21 A closed curve around major circumferencea
(C”l The a-torus itself

Space: Start with a 3-dimensional closed space-visualizable as the a-surface of a sphere

imbedded in Euclidean 4-space-and modify this space topologically to the extent needed
to install one wormhole with mouths Ml and MS This is the pierced sphere W1 = S2 X S.
Its Betti numbers: RO = 1; RI = 1; Rz = 1; Ra = 1. Euler characteristic x = 0.

Homology classes Description of typical p-cycle in this homology class

A point; all points are homologous

ie”:; Closed line of force that goes once through the wormholeb
Closed surface about the mouth M1c
The 3-space itself.

a A closed curve that can be shrunk to zero can be constructed as the combination of
two such lines drawn in opposite senses about the torus.
b A closed curve that can be shrunk to zero is the linear combination with opposite
senses-that is, the difference-of two such lines. A closed curve that goes around the
mouth, Ml , of the wormhole can be moved away from this location and shrunk to zero.
Three dimensions gives freedom to conduct this operation. It is impossible on the lower
dimensional space of Fig. 3.
c A closed surface that can be shrunk to zero is the linear combination of two such sur-
laces with opposite sign.

ing distinct homology classes, we therefore always omit the class of cycles which
are homologous zero. Normally there exist infinitely many homology classes.
None is more fundamental than another. The situation is reminiscent of the dis-
placements of a crystal lattice-say a simple cubic lattice-which carry it from
one position to an equivalent. position. One can have a displacement, a in the
x-direction; or a displacement a in the y-direction; or a knight’s move of a in the
x-direction simultaneously with 2a in the y-direction; etc. The general allowed
move can be represented as a linear combination of three basic or linearly inde-
pendent moves. However, there is no unique choice of these basic moves. Simi-
larly, by number of distinct homology classes we mean the number of linearly
independent homology classes in the following sense: A set of homology classes
568 MISNER AND WHEELER

(c;‘) is linearly independent if and only if every relation of the form

F s”{Q} = [T SV} = (01,

with real coefficients si, implies that all the si are zero; that is, if and only if no
linear combination of the cycles tip is a boundary. When the homology classes
{c,‘) are linearly independent we shall say that the cycles tip representing
these homology classes are independent cycles.
The basic relationships between chains, cycles and boundaries which we have
been discussing may be summarized in a purely algebraic form. Since we can add
chains together, the set of chains in a manifold 311forms a group. We may also
multiply a p-chain by a real number. Consequently the p-chains form an infinite
dimensional vector space, CP(3n, R), called the group of real p-chains. The chains
cp satisfying 83 = 0 form a subspace 2,(3-n, R) called the group of p-cycles.
Every chain cp which is a boundary, cp = dc’+‘, is a cycle. Consequently the set
of all bounding p-cycles Bp(5n, R) is a vector space which is a subspace of
.Z,(F~, R). The quotient space
HA% R) = +W% R)l&(w RI (134)
is called the p-dimensional homology group of 3~ with real coefficients. The Betti
number R, is the dimension of the vector space Hp(3nr R) and is finite for any
compact manifold.
It may happen that in a closed (compact) n-manifold, every n-chain has a
boundary so that R, = 0. In this case the manifold is called nonorientable; in
the opposite case R, = 1 and the manifold is orientable. This distinction points
up a difference between the idea of boundary expressed by d and the idea of a
boundary as consisting of points on the “edge” of a set. In a manifold every
point has a neighborhood homeomorphic to Euclidean space, so no point can be
considered as being on an “edge” or “boundary” of the manifold. When, how-
ever, we break the manifold up into cubes,

fJll= c” = 2 C<“, (135)

i=4
we may not be able to choose the orientation of the cubes in such a way that
the faces of all the cubes cancel out in pairs when we calculate ac: This is the
situation in a nonorientable manifold. An example is given in Fig. 5 where we
compute the Betti numbers of the projective plane P2.

Stokes Theorem.
We are now in a position to state Stokes theorem concerning the integral of a
p-form a over the p-dimensional boundary acP+’ of a (p + l)-chain c’+l:

(136)
 CLASSICAL GEOMETRODYNAMICS 569

FIG. 5. The projective plane Pz as an example of a nonorientable manifold. The points

of the projective plane are pairs of antipodal points on the a-sphere S2. In the figure we
show the upper hemisphere ca of P, and its equator. Two points on the equator of St, such
as those marked co in the figure, correspond to exactly one point co in P*. Similarly a single
line c1 in Pz appears twice as shown in the figure on the equator of 82. Regarding co, cl, c*,
as chains, i.e., as being broken up into a large finite number of small “cubes”, we see that
a? = 2cr when ca and cl are oriented as indicated by the arrows in the figure. Since c2 has
a boundary 2c1 in the a sense of boundary, the Betti number Rz(Pz) is zero. The curve cl
has no boundary: ac’ = co - co = 0. However c’ is the boundary of the a-chain >ir?, so
every l-cycle bounds and R1(P2) = 0. The point co has no boundary of course, and it is not
the boundary of any l-chain, so there is one class of O-cycles, [CO), different from zero, and
Ro(P2) = 1.
The integral of a form over a chain is defined in terms of the integral in Eq. (118)
over the standard unit cube. Therefore the proof of Stokes theorem reduces to
integration by parts on this cube, I’+‘. The simplest case of Stokes theorem is
that of a O-form or function, f, whose gradient is integrated over a curve c1with
end points x1 and XO:

I df = s,l~,l-z.
f = fhl) - fbo>. (137)

Here we make the convention that the integral of a function over a O-cube, or
point, is the value of the function at that point. For a l-form, such as the mag-
netic field h = hidx’ in 3-space, Stokes theorem reads

This is the formula which usually carries Stokes’ name. Gauss’ theorem is also
included in formula (136) as we may seeby considering a 2-form

*f! = & *e;j &Ji A CIX’. (139)

We have in this case,for a volume c” in S-space,

=- 1 *eij dxi A dx’ = *e.

2 saca sa03
570 MISNER AND WHEELER

This equation takes the usual form of Gauss’ theorem in flat space when we write
el for *e23, etc.

Applications of Xtolces Theorem: Electric Charge, Magnetic Charge, and the Vec-
tor Potential.
Two particular cases of Stokes theorem are of principal interest:
(1) If cp+l is a closed chain, i.e., a cycle, and a = a’ is a p-form, then Eq. (136)
reads :

sCP+l
clap = 0 if

(2) If dap = 0 we call a’ a closed form. In this case Stokes theorem gives:
&pi1 = 0. (141)

s aCp+r
a’ = 0 if da” = 0. (142)

This second case may be put in the form of a conservation law: Let c, and cb be
homologous p-cycles, so CZ,- ca = acPtl. Then Eq. (142) becomes

I,p+l a’ = I, 8’ - I, 8’ = 0. (143)

In other words, the integral of a closed form over a closed surface depends only upon
the homology class of the surface. The value of this integral is called the period of
a on ( c,) , the common homology class of c, and cb . According to Maxwell’s equa-
tions, (104) and (108), both f and *f are closed forms. To the periods of these
tensors we give names15, to be justified in the next paragraph:

Jc2f = 47rp* = 47r (magnetic charge of (c”)) (144a)

*f = 47rp* = 47~ (electric charge of {c”]) (144b)

s c2
When the electromagnetic field is derived from a vector potential a = a,,dx”,
then there is zero net flux through every surface c2 that is closed (ac’ = 0):

18f = 12da = 6,.a = 0. (145)

16 The superscript * indicates that we are measuring both kinds of charge in the purely
geometrical units of length. As we pass from the geometrical measure of field strength,
f (cm-r) to the conventional measure of field, F (gauss or g1/2/cm1’2 see) by multiplication
with the factor ce/G”e(g I/* cmr/2/sec), so the same factor translates from the geometrical
measure of charge, q* (cm), to the conventional measure of charge in electrostatic units
(glln cm3/t/sec) :

q(esu) = (c*/G’@)q* (cm).

 CLASSICAL GEOMETRODYNAMICS 571

In other words, the existence of a vector potential implies that there is no magnetic
charge.
In reasonable examples the homology class of the surface c2 includes (a) similar
appearing surfaces of larger or smaller size at the same time, that is, in the same
space like surface and (b) surfaces like these at earlier and later times. Both f and
*f are closed. Then Eq. (143) says that the flux through all of these surfaces is
the same-independent of the size of the surface (Gauss’ law) and independent
of time (law of conservation of charge). Therefore charge as defined in Eqs. (144)
has the properties that one is accustomed to demand. The four dimensional
character of the analysis has combined into a single law two laws normally re-
garded as separate. In brief, we have shown from Maxwell’s equations that
charge-regarded as lines of force trapped in the topology-stays constant with
time regardless of changes in the details of the electromagnetic field, and regard-
less of changes in space. No mention of the metric even entered the proof of the
conservation theorem.
To see in a little more detail the reasonableness of the identification (144),
consider flat space with the usual time and space coordinates. Then the forms f
and *f are related to electric and magnetic fields in the following way:
f = -dt // (ei dzi) + (*h)ijdzi A dzf, (146a)
*f = dt A (hi dxi) + (*e)ijdxi // hi. (146b)
Here e; and hi are the components of the usual electric and magnetic fields, and
(*h),, = h, , (*e),, = e, , etc. When c2 is a closed surface, say a sphere, in the
T = const. hyperplane, and when f is the electromagnetic field produced by a par-
ticle inside this sphere, then the magnetic pole strength p* and the electric charge
q* of this particle are given by

47~p* = l2 (*h)ijdzi A dx3 = 1 f, (147a)

c2

4rq* = 1. (*e);j do” A ds’ = I2 *f, (147b)

The first equalities in these equations are familiar expressions of Gauss’ law in
flat space; the second equalities hold because the dT terms in Eqs. (146) eon-
tribute nothing to an integral over a surface such as c2 on which T is constant.
In other words, the vector dT = grad T is orthogonal to any surface on which
T is constant, and therefore the projection of dT on that surface is zero. In this
way we see that Eqs. (144) are direct generalizations of familiar equations to a
curved space-time.
Examples of Spaces Permitting Charge.
To fill the gap between the very general language of the conservation laws in
the last section and the simple intuitive picture of charge in Fig. 3 as lines of
572 MISNER AND WHEELER

force trapped in the topology, it is appropriate to present here two examples of

spaces permitting charge; that is, spaces endowed with one or more wormholes:
Rz 2 1. These examples are the manifolds R X Wk and R X T3 which were de-
fmed earlier in part B of this section (Examples of Manifolds, Product of Mani-
folds). To define the differentiable structure of W,+ (see Differentiable Manifolds)
we introduced on WI a functionf(z); in Wk weuse k suchfunctionsf&), i = 1,
2, “.) k. Let c?(r) be the sphere in Wk defined by f&r) = r. In the two dimen-
sional analog of Wk shown in Fig. 4, the sphere is doubly pierced (k = 2). The
typical surface cl(r) is a circle that runs around the mouth of one of the tubes
that connects opposite faces of the sphere. The typical surface Q(T) runs around
the mouth of the other tube. In the product manifold R X Wk , of a real line,
- 00 < T < 00, with Wk , there is on each hypersurface T = const. a copy of
c:(r) which we will call t?(r, T). Moreover, the pth Betti number R,(R X W,LJ
of the product manifold is the same as the pth Betti number Rp(Wk) of Wk itself.
Now consider on R X Wk a 2-form f or *f. Then the flux

or

apparently depends on r and T. To see that p;*(r, T) or Q~*(T, T) in fact depends

only on i we need (a) to require that df = 0 or d*f = 0 and (b) to see that all
the surfaces c?(r, T) for fixed i belong to the same homology class. The surface
c?(O, 0) is a 2-sphere. As we let r increase this sphere sweeps out a 3-surface
ci3(R) 0 < r 6 R which is topologically identical to the volume 1 < x2 + y2 +
z? < 2 in Euclidean S-space. From this mental image of the S-space cc3(R) in the
familiar Euclidean 3-space R3, we see the boundary of c:(R):
k;(R) = c:(R, 0) - c:(O, 0). (149)
We conclude that c?(r, 0) is homologous to c?(O, 0). No metric notions have en-
tered yet. Therefore we can let c?(O, T) sweep out a 3-surface c?(T) just as we
did above for r, and find
d(T) = ci2(0, T) - ci2(0, 0) (150)
so that ci”(O, T) and ~‘(0, 0) are homologous. A similar construction shows that
ci2(r, T) is homologous to ci’(r, 0). More generally, we conclude that all the sur-
faces ct(r, T) for the same i are homologous. Moreover, the equations df = 0
and d*f = 0 say that no flux gets lost between homologous surfaces. Therefore
the charges pi*(r, T) and qi*(r, T) are independent of r and T.
 CLASSICAL GEOMETRODYNAMICS 573

The preceding discussion made no use of any metric on R X Wk , and is there-

fore valid no matter what metric we introduce. For example, at one value of T
the space-like hypersurface T = const may have the appearance of the doubly-
pierced sphere of Fig. 4(c). At a later T the two ends of one of the wormholes
may move together preparatory to annihilation as in Fig. 3. But as long as the
topology does not actually change, the lines of force remain trapped and the flux
47rqi* through the closed surface c?(r, T) is unaltered by disturbances of the
metric.
In this analysis of the wormhole space-time R X Wk , we assumed the existence
of a closed form *f or f for which the charges and pole strengths in Eqs. (148)
do not all vanish. In the following section on de Rham’s theorem we shall see
that many such forms do exist. In a simpler example, R X T3, we can display
them explicitly.
On the torus, T3, functions s”(i = 1, 2, 3) cannot be defined continuously
everywhere except to within an additive multiple of 2~. However, the gradients
dxi are differentiable vector fields. In the simpler case of a 2-torus one of these
gradients may be a field of unit vectors (red!) that circle about the minor cir-
cumference of the torus. The other field of unit vectors (green!) may be taken
to be parallel to the major circumference. Similarly vector fields are definable in
the space R X T3. From them one can construct a form,
*f = dzr2 A dx3,
which is obviously closed (d*f = 0). It represents a uniform electric field in the
x1 direction [see Eq. (146b)]. Integrating this 2-form over the surface cr2(x1, T)
defined by T = const, x1 = const, we get the value

47rq1* = dx2 A dx3 = (2x)‘. (151)

s el*(ZlJ)
Thus we verify directly that in this case ql* is independent of x1 and T. We might
hesitate to call ql* a charge since we cannot easily imagine any metric that makes
all the lines of force appear to originate in a small region of space. However ql*
clearly measures the total flux of the electric field around the 3-torus through
the closed surface c12. Similarly, there are two other independent directions on
the 3. torus, in each of which the flux can be given what value one pleases. There
this space is endowed with three arbitrary constants of charge.
De Rham’s Theorem Proves the Existence of %Forms Endowed with Charge.
To have given an example of a charge-like solution is far from having proven
that such solutions are always possible in a four space endowed with wormholes;
that is, in a space with Betti number Rz >/ 1. The existence of such solutions is
established by the jirst theorem of de Rham (36) : If ci* are R, independent p-cycles
of a manifold 311,and &i any R, real numbers, there exists a p-form a’ which is
574 MISNER AND WHEELER

closed (dap = 0) and differentiable throughout sx and which satisfies

scp
a’ = Qi (i = 1, 2, . * * , R,).

To translate into physical terms, we note that a = f2 or *f” is the form that repre-
sents the electromagnetic field or its dual. The Betti number Rz is the number
of wormholes, and Q, = 4?rpi* or 47rqi* is the magnetic or electric flux through
the ith wormhole. More precisely, &a is the flux through surfaces of the ith inde-
pendent homology class, or in mathematical terms, the so-called period on this
class. We have to recall again the arbitrariness in the choice of the R2 independent
or basic homology classes of order 2, linear combinations of which give all homol-
ogy classes of order 2. There are situations where it is reasonably clear that two
wormholes are well separated from each other, so that it might be particularly
appropriate for Q1 to designate the flux through one, and Q2 the flux through the
other. However, it is also possible to give instances where one wormhole is im-
bedded inside the other and where Q’1 = &I + QZ and Q: = Q1 - Q2, or some
other combinations, are just as natural as &I and Q2 as measures of charge. Thus
de Rham’s theorem says that there exists a solution f of the Maxwell equation,
df = 0 or a solution, *f, of the Maxwell equations, d*f = 0, which has arbitrarily
assignable real magnetic or electric charges.
Now demand that the magnetic charge of every wormhole vanish; in other
words, demand that the period of f on every homology class {c2) shall be zero:

f = 0 for all c2. (153)

s

Then the second theorem of de Rham guarantees the existence of a vector poten-
tial, a:
When fP is a p-form on 311,which is closed (df” = 0) and all of whose periods
are zero, then fp is derived from a differentiable (p - 1)-form:
fP = daP-‘. (154)
In a case such as this, where there exists a potential, so that the one form, f,
can be written as the curl of another form, a, then the form f is said to be exact.
To be exact is a more demanding requirement on a form than merely to be closed
(Table VII).
ConceptsBased on the Metric: Duality and Divergence: Completion of Maxwell’s
Equations.
We have just derived Gauss’ theorem and the law of conservation of charge
from Maxwell’s eight equations, df = 0 and d*f = 0, without making any appeal
whatsoever to the notions of metric or length. The burden of the proof lay on
 CLASSICAL GEOMETRODYNAMICS 575

TABLE VII
SIMILARITIES BETWEEN THE CONCEPTS OF CYCLES OR CLOSED CHAINS
(SUCH AS CLOSED SURFACES) AND CLOSED FORMP

Surfaces

c General p-chain; a surface for p = 2.

ac = 0 Means that p-chain is closed; c is then called a p-cycle; signifies for
p = 2 a closed surface.
p = +y+l Means that cp is the boundary of c p+i; it then follows that a@ = 0,
in other words, 82 = 0. However, from a@ = 0 it does not necessar-
ily follow that there exists any cp+l of which cp is the boundary.
Stokes’ theorem: Given one chain cp that is a boundary: cp = &pi1 (whence a@ = 0
and cp = a cycle); then on this cycle the period f vanishes for
s cp
every p-form f that is closed (df = 0).
de Rham’s theorem: Given one chain cp which is closed (cp = a cycle; a@ = 0) ; and given

that the period f vanishes for every p-form f that is closed; then
s cp
c* is itself a boundary: cr = a&‘.

f General p-form; a vector for p = 1; alternating tensor for p = 2.

df = 0 Then f is called a closed form. The Maxwell field tensor is described
by a closed form.
f=da Then f is said to be an ezaet form, or f is said to bound a (the four
potential, for example). It then follows from the relation d2 = 0
that f is automatically closed. However, from the closure of a
form f it does not necessarily follow that there exists any (p - 1)
form a of which f is the curl or boundary.
Stokes’ theorem: Given one p-form f that is exact: f = U-i (whence df = 0); then

the period f vanishes on every p-cycle (a@ = 0).

s2
de Rham’s theorem: Given one p-form f that is closed (df = 0) and given that the period

f vanishes on every p-cycle (acp = 0), then f is exact: f = dae-1.

s2

* The concept of chain as defined earlier and used here means a linear combination of a
finite number of cubes-a form of definition important for open or non-compact manifolds.

topology. We avoided any appeal to Riemannian geometry by treating the Max-

well field
f = >4jNydx’ A dx’ (155a)
and its dual
*f = s*f,,ydx” A dx” (155b)
576 MISNER AND WHEELER

as two completely independent objects. We extract extra content from Maxwell’s

equations, and impose additional restrictions on the field f, when we note the
one-to-one relation between one field and the other that is established by the
metric. We shall therefore now define duality and divergence, and formulate the
remaining content of Maxwell’s equations.
The connection (11) between the electromagnetic tensor f and its dual *f is
generalizable to a connection between any p-form
a = (p!)-la,, . . .+,dzLI1 A . . . A dP’ (156)
and its dual, the (n - p)-form

Here n is the dimensionality of the manifold and p and (n - p) the orders of

the two forms, g is the determinant of the metric tensor guB, and the quantities
with upper indices are the contravariant components of the original tensor:
a~.*‘cTp_
a - gU”’ . * * gapp’a,,...pp. (158)
Applied to a special O-form, the constant function 1, the duality gives the volume
element
*l = j g (“‘dzl rl . . . A dz”. (159)
The following general formulas are most readily proven by applying the defini-
tion of the duality operation in an orthonormal reference system:
* *BP = (- l)np+p+sap (160)
(double duality)
a’ A *bP = b A *a = (~!)-‘a~“““~b,,...~~*l (161)
(relation between exterior product and scalar product)
(*a) A *(*b) = (-1)“a A *b (162)
(scalar product of dual forms expressed in terms of scalar product of forms
themselves). In these equations s stands for the signature of the metric: s = 0
on space-like hypersurfaces, where the metric is of the form + + + ; s = 1 in
four space with the metric - + + + ; and more generally, s is the maximum
number of orthogonal vectors with negative norm.
The p-form and its dual are distinguished in the following way, that the in-
tegral

s =P
a (163)
 CLASSICAL GEOMETROWi’NA’MICS 577

may be said to be the integral of the tangential component of a over a surface or

p-chain cp, ,whereas the quantity

s#9--P *a

represents the integral of the normal component of a over an (n - p)-chain

(164)

c n--Z,.
Take the electric field e, a l-form in 3-space; form its dual, a 2-form; take its
exterior derivative and secure a S-form; and finally take its dual and obtain a
O-form or scalar. This scalar represents the divergence of the electric field, as
seen most quickly by writing out the indicated sequence of operations in flat
space :
e = ekdxk ;
*e = eldxz A dx3 + two similar terms;
d*e = (3el/dz1)dz1 A dx’ A dx3 + two similar terms W-9
= (div e)dx’ A dx2 A dx3;
*d*e = div e.
Generalizing to a curved space and a p-form of any order, the theory of forms
defines a general divergence or co-cliflerential operation 6 which takes a p-form a
into a (p - 1) form,
bp-’ = gap G (- l)np+n+s+l*d*ap, (166)
where the f sign is introduced to simplify the statement of Green’s theorem.
For a (p - 1) form a and a p-form b this theorem connects the n-fold integral
of two scalars over all or part of the manifold m with the (n - 1) fold integral
of a normal vector, a*b = a A *b, over the boundary of this region:

sam
a*b = jrn d(a*b)

The negative d’Alembertian operator

= lrn b*da - Irn 6b*a. (167)

A = 6d + d6( = -‘< 02” or -“o” for a scalar) W33)

carries a p-form into a p-form. In particular it carries the scalar, ‘p, into the
scalar
Acp = 6dp = -div grad C,D= -Cp”;$ . (169)
The properties of the differential operators d, 6, and A are summarized in Table
VIII.
578 MISNER AND WHEELER

TABLE VIII
PROPERTIES OF &E DUALITY OPERATION*, THE EXTERIOR DERIVATIVE d, AND THE
DIVERGENCE OPERATION 6 IN SPACE TIME AND ON A SPACE-LIKE HYPERSURFACE.
THE NOTATION uw STANDS FOR THE EXTERIOR PRODUCT wI*v = vA*u

Property 4.manibld d-manifold

--.
Sigdature of metric -+++ +++
Dual of dual of p-form * * = (-)“+I **=l
Divergence of p-form 6 = +d* 6 = *d* (-)”
Second order differential operator “- 0” = A _= &j + &j (‘- V2” = A s f,d + d&

General formulas, n-manifold YP, any metric

*d = 6* (-l)p+‘; *6 = d* (-1)~
*A = A*; dA = Ad; ISA = A6
Stokes’ theorem

Integral formulas for case @JP = 0

S is adjoint of d (a = aP-1; b = bp)

A is self adjoint (a = ap; b = br)

For n = 4m - I,*d is self ad-

b*(*da) = (*db)*a (a = azm-1; b = +-I)
joint on (2m - 1) forms s gpm-1 s mm4”-1

In terms of the divergence operation 6, the entire content of Maxwell’s equa-

tions can now be summarized in the form
df = o= 6f. (170)
D. CHARACTERIZATION OF SOLUTIONS OF MAXWELL'S EQUATIONS BY INITIAL
VALUE DATA:CHARGE AND WAVE NUMBER
Alternative FormulaLions of the Initial Value Problem: Summary of Analysis.
On a space like surface endowed with a metric it is well known to be enough
to specify arbitrary continuous electric and magnetic fields-satisfying div e = 0,
div h = O-in order to be able to continue the fields to the future and the past
by way of Maxwell’s equation8. This type of initial value data is the natural
analogue of the x and 2, or x and p, needed to predict the future of a classical
particle. A different choice of data is made in the Lagrange formulation. Applied
to a particle, it demands that x be known at the initial and final times, with no
16 This is a familiar application of the standard Cauchy-Kowaleski theory (R’).
 CLASSICAL GEOMETRODYNAMICS 579

information needed about either 2 or p. Applied to the Maxwell field, the La-
grange formulation asks for example for the magnetic field-or the electric field,
but not both-on the two space like hypersurfaces T = T, and T = T, . This
“two surface” type of boundary condition is appropriate when one is interested
in the classical action of the field integrated over the space-time interval between
the two surfaces, or when one wants to know Feynman’s quantum propagator:
the probability amplitude to passfrom configuration C1on u1to configuration CZ
on u2 . When the two cotimes have a finite separation one knows how to analyze
this classical boundary value problem in flat space and in some types of curved
space (seebelow) but not in the general type of curved space.However, when the
two surfaces in question are so close together that the boundary value data can
be considered to define the magnetic field and its first cotime derivative, then
the two surface problem reduces to the one surface initial value problem. To it
we shall henceforth limit our attention in either form A or form B:
A: Give h and ah/aT on a;.and give the RPelectric charges or worm- (171A)
hole fluxes Q<*once and for all.
B:Givehandeona. (171B)
We shall first analyze the relation between a form in 4-spaceand its projection
or trace on a 3-space; then the connection between the metric on the hyper-
surface and the metric of 4-space; and then Maxwell’s equations expressed in
terms of forms on the hypersurface. We show that specifications (171A) and
(171B) are equally appropriate to determine uniquely the future evolution of the
system. Finally we treat briefly the specification of the initial value data, not
directly in a spacelike form as in (171A, B), but by way of a natural generaliza-
tion of Fourier analysis to curved space.
The Concept of Projection or Trace.
A hypersurface u in the manifold %i!can be defined by an equation of the form
p(x) = 0, where cpis a function with dq # 0 at u. We may, if we wish, choose
cpas one of our coordinates and call it T or x0. We now consider an operation
which enables us to obtain from any covariant tensor on the 4-manifold ‘SK,a
corresponding tensor on the 3-manifold u. In the special casewhere &c) is any
scalar function on SE, its projection or trace on u is the function q”(x) defined for
x on u by the equation:
cp”(x) = p(x), 2 on u. (172)
Representing cpin terms of coordinates as (o(T, x1, x2, x3), where u is given by
x”(x) = T(z) = To , z on u, (173)
we can rewrite Eq. (172) in the particularly simple form
cp”(x’, x2, x3) = q(To , x1, x2, x3). (174)
580 MISNER AND WHEELER

Similarly, for a p-form

a = (p!)-lu,,...,,(xo, x1, x2, z3)dz” A . . . A dP, (175)
we obtain the trace a” of a on u by substituting in Eq. (175) the values $0 = To
dr” = 0. Likewise, the quadratic form
ds2 = g,,dz”dz’, (p, Y = 0, 1, 2, 3) (176)
on u reduces to
d12 3 gii”dxidx’ = (ds’) = gijdxidxi (4 2. = 1, 2, 3), (177)
which displays the components giy of the metric we use on the 3-manifold u.
The exterior derivative d” on u involves no differentiations with respect to x0:“,
but the identity
d”a” = (da)” (178)
assures us that it will cause no confusion to write d for d”.
Now we may define, relative to the space like hgpersurface u, the dual magnetic
field, *h, and the dual electric field, *e, by the equations
*h = f”, (179a)
*e = (*f)“. (179b)
We have written *h and *e here so that e and h will be vectors or l-forms of the
familiar kind on u. The duality to be used here is based on the metric g;r of u,
and so could be written *o . This allows us to define
h = *$ (180a)
e = *,(*f)“, (Bob)
consistent with Eqs. (179) because of the identity *r*C = 1. We shall never apply
*- to a form on S& so the meaning wiI1 be clear from the context if we drop the
subscription on *a as in Eqs. (179). From Maxwell’s equations on 311,df = 0 =
d*f, and the identity (178) we obtain the trace of Maxwell’s equations on U, df” =
0 = d(+f)“. In terms of the definitions of (180) and of Table VIII, the trace of
Maxwell’s equations on a space l&e surface takes the form
S,h = 0 = 6-e. (181)
We will normally omit the subscript u on 6 because it will be clear from the con-
text that we are dealing with a space like surface. We use subscripts on 6, and
3 because they are not simply the traces on u of the corresponding operations
on em; in other words, no identity holds for 6 or * like (178) for d; for instance,
*#’ is a l-form, whereas (*f)” is a 2-form.
 CLASSICAL GEOMETRODYNAMICS 581

Maxwell’s Equation in S-Dimensional Form.

To reconstruct the electromagnetic field f from the 3-dimensional fields e and
h it simplifies the analysis to make a more careful choice of a cotime coordinate
T than the one that was demanded in the previous section. Let T be the cotime
measured along a geodesic which starts out normal to a space like hypersurface
uo . Then go is given by T(x) = 0 and each uT , defined by T(x) = const, is again
a space like hypersurface. The quadratic form ds’ can be written

ds2 = - (dT)2 + d12, w32)

where d12 is the positive definite quadratic form defined by the metric on uT .
Our construction of geodesically parallel surfaces uT may carry us into 4-space
for only a small interval of T (see ref. SS), but in the 3-spaces uT we nevertheless
have a well-defined global topology. We know from the definitions of e and h
that at CT,f and *f must have the form
f = *b - dT A E (183a)
*f = *,,e + dT A K, (183b)
where E and K are l-forms which remain to be determined. When we use T, and
any coordinates xi(i = 1, 2, 3) on (~0, as coordinates x’ in 4-space (cc = 0, 1, 2, 3)
with x0 = T, the metric components have the property go0 = - 1, goi = 0. Then
from Eqs. (157) and (183) we compute *f and *(*f) = -f. In this way we dis-
cover that e = e and K = h, that is
f = *b - dT A e (184a)
*f = *,e + dT A h (184b)

Since d and d” are related by the formula

d=d”+dTh&,

we can use this expression in Maxwell’s equations, df = 0 = d*f, together with

Eqs. (184) and (100) to find

d”(*b) + dT A -&, *,h + due = 0, (186a)

d”(*,e) + dT A & *,,e - d”h = 0.

Here a/aT is the derivative along the geodesic normal to u. The terms in dT in
Eqs. (186) must vanish independently of the other terms, so we may rewrite
Eqs. (186) as 3-dimensional equations (Table IX), treating T as a parameter
582 MISNER AND WHEELER

TABLE IX
THE TRACE OF MAXWELL’S EQUATIONS ON A SPACE LIKE SURFACE

“Electric equations” “Magnetic equations”

Intrinsic J-vector notation Intrinsic 3-vector notation on
notation on flat space notation flat space

Equations in bf = 0 div e = 0 and df,= 0 div h = 0 and

4-space or d*f = 0 ae/aT = curl h (or 8*f = 0) dh/aT = - curl e

Trace on (r d(*f)’ = 0 d(f”) = 0

Define Define
e = *,(*f)’ h = *,(f’)
Then Then
d(*,e) = 0 d(*,h) = 0
or he = 0 dive = 0 or &h = 0 div h = 0

and omitting the u’s:

tih = 0, a(*h)/dT = -de, (187a)
iie = 0, d(*e)/aT = dh. (187b)
Here we have written Sh = 0 in place of the d*h = 0 of Eq. (186a) to agree with
the customary three dimensional forms of Maxwell’s equations.
Formulation B of the Initial Value Problem.
Equations (187) are now in the form to which the Cauchy-Kowalewski theo-
rem applies. More explicitly, the equations

a(*h)/t?T = -d*(*e), (188a)

a(*e)/aT = d*(*h),

are six equations, solved for the time derivatives, for six unknown functions, the
components of *e and *h. They determine *e and *h uniquely when the initial
values of *e and *h on the hypersurface a0 are given. When in addition these
initial values satisfy
d*h = 0 = d*e uw
on a0, then these same requirements of zero divergence will also be satisfied on
Q=as a consequenceof Eqs. (188) :

a(d*h)/aT = 0 = e(d*e)/aT. ow
The proof depends on the facts that d = d” commutes with d/CiT, and that
d2 = 0. De Rham’s theorem assuresus that we can find initial values *h and *e
 CLASSICAL GEOMETRODYNAMICS 583

on u. which (1) satisfy the divergence requirements (189) and (2) have any de-
sired values for the magnetic and electric charges

4rpi* = Ii2 *h (191a)

and

on Rz independent wormholes in co . Granted such initial value data, we are

assured” of the existence for some finite time beyond u. of a unique solution of
the source-free Maxwell equations that displays constantly the specified charges.
Formulation A of the Initial Value Problem.
So much for the case where e and h are specified on u; now for the alternative
case where (1) we know the electric charges pi* once and for all, and where in
addition (2) we know on u the quantities *h and &h/aT, as the limit of infor-
mation about the magnetic field on two infinitesimally close space like surfaces-
information of the kind suited to the Lagrangian formulation of quantum me-
chanics.
The initial value data must satisfy the divergence conditions,
d*h = 0 (192)
and
d(&h/aT) = 0 (193)

and the requirement that the magnetic fluxes-if any-through all wormholes
shall stay constant in time:

sci2
(&h/dT) = 0. (194

This granted, we find the electric field e on the space like surface uniquely from
the three pieces of information
Se = 0, (195)
de = - a*h/aT,

17 The Cauchy-Kowalewski theorem

s
Ci2
e = 4*qi*p

requires analyticity, while the de Rham

(197)

theorem
supplies only differentiable initial values e and h. To remedy this fault in the demonstra-
tion one may employ the existence theorem of Four&s-Bruhat (39).
584 MISNER AND WHEELER

TABLE X
CLASSIFICATION OF REAL NONSINGULAR VECTOR FIELDS IN A COMPACT ~-SPACE
ENDOWED WITH POSITIVE DEFINITE METRIC; THE TRANSVERSE WAVES ARE
SAID To HAVE RIGHT- OR LEFT-HANDED POLARIZATION ACCORDING AS

THE SCALAR WAVE NUMBER. k. Is POSITIVE OR NEGATIVE

Type of vector field Longitudinal TIUWWX Coulomb (harmonic)

Symbol L X c
Definition AL = 2L *dX = kX dC = 0 = N.!
dL = 0 6X =o
Ikl >O

Association with ei- L1 Lz ..’ x, x1 . . . c, Cb . . ’ CR2

genvalues or sur- Kp 42 . . . k, kz . . 8, sb ‘.. &Q

faces

Consequence A(6L) = tc2(6L) = k2X

AX AC = 0
Potential scalar (0 Z-form Y none
Potential from field -
P =K -%L Y = *X = k-‘dX
Field from potential L = tcldp X = k-‘&Y = aY -
Potential equations Ap = .k+o *SY = kY
dY =0
Potential simplifies Yes No -
analysis?

Standardization L*L = 1 x*x = 1

J s

-r--

s L,*Ln = 6,%
s
L,*X, = 0
s
L,*C* = 0

Scalar pro- x
X,*L, = 0 x,*x, = sm, x,*cb = 0
duct with s s s

b I C,*L, = 0
s
c,*x, = 0
s C,*C, + 8ir/Romm

C,*Q i 4*
s
‘2 (fl)c~ (flhha+)(b+)
(*)a
(f)b

-for example, by the methods of Fourier analysis described in the following

section (Table X). Knowing both e and h, we are back at an initial value problem
that has already presented itself, and that possessesa unique solution for some
finite time beyond uo.
 CLASSICAL GEOMETRODYNAMICS 585

Generulizution of Fourier Analysis to Curved Space.

When one is dealing with electrodynamics in curved space it is most natural
to specify initial value data as a function of position on U, as in (171A, B). An
ordinary Fourier analysis is completely out of the question. However, one can
give a generalization of Fourier analysis to curved and even multiply-connected
space which is of use in considering special situations where the metric is con-
stant, or nearly constant, with respect to a suitably chosen cotime coordinate,
T. It permits the initial value data to be specified in terms of Fourier coefficients.
On the space-like hypersurface r consider initial value data such as the elec-
tric field e, the magnetic field h, or more generally any vector field or l-form, v.
According to the analysis of Hodge (SO), we can decompose v uniquely into the
sum of three parts, characterized as in Table X by the names longitudinal, trans-
verse, and harmonic-or, as we may say, Coulomb. We will be limiting attention
to 3-space. Therefore it will be permissible as in Table X to omit the superscripts
(r that evidences the space-like character of the operations of curl, d”, and di-
vergence, 5”. For simplicity of analysis we shall assumethat the space is closed.
A longitudinal field has zero curl, but a nonconserved flux because it has a
nonzero divergence. Electric or magnetic fields of this type would correspond
to distributions of “real” charge or magnetic poles. Therefore we exclude such
fields from attention.
A transverse field has zero divergence and zero flux through any wormhole,
but its curl is nonzero. Such a field can be decomposed into proper modesX, ,
each endowed with a characteristic scalar wave number k,(#O) of its own:
*dX,( = curl X,) = kX, . (198)
Modes with positive k we describe as right-handedly polarized; conversely for
modes with negative k.
A harmonic or Coulomb field has zero divergence and zero curl, but a non-
vanishing flux through at least one wormhole. Such a vector field can be decom-
posed uniquely into a linear combination of characteristic harmonic vector Jields,
C,(a = 1,2, . * +, R2), such that C, attributes a unit charge to the ath wormhole,
and contributes no flux at all to any other wormholes-that is, to any other of
Rz basic linearly independent homology classes.
The general vector field in curved multiply-connected 3-space can be Fourier-
analyzed in the form

The orthogonality relations and the normalization conventions summarized in

Table X permit one to find all the coefficients in this expansion by relations of
which the following is typical:
(200)
586 MISNER AND WHEELER

Without attempting to derive here the general expansion (199), we can note
some of the underlying principles. (1) The inner product of two vector fields,

(a, b) = [ a*b, (201)

is positive definite, so that

(a, a> > 0 (202)

unless a = 0. (2) It follows that the set of all vector fields or l-forms of finite
norm generate a Hilbert space. (3) The operators A and *d are self adjoint when
applied to vectors in S-space (Table IX) and therefore have eigenvalues and
complete sets of eigenfunctions. (4) From the property
(v, Av) = (6v, 6v) + (dv, dv) > 0 (203)
it follows that the eigenvalues of A are non-negative numbers, lc’. (5) Given any
nonzero solution of the equation
Av = Ii’v (204)
with k > 0, one can decompose it into two parts,
v=a+b,
a = Icm2d&v7 (205)
b = K26dv,

both of which satisfy the same space-like version (204) of the wave equation, and
one of which is longitudinal; the other, transverse:
da = 0, (206)
6b = 0. (207)
(6) For those eigenvalues A for which there exists a nonzero transverse field b,
there can be constructed out of this field a curl,

c = k-‘*db( = k-’ curl b), ( 208)

such that
k-‘*dc( = k-’ curl c) = Ic-‘Ab = b. (209)
From b and c we can build right-handedly and left-handedly polarized fields

X, = b + c,
(210)
k-‘*dX+( = k-l curl X,) = X+ ,
 CLASSICAL GEOMETRODYNAMICS 587

and
X-=b-c
(211)
- k-‘*dX_( = -k-l curl X_) = X. ,
not both of which can vanish. As example, consider in flat space the transverse
field or l-form,
b = dx2 sin kxl,
c = k-l curl b = k-‘*db = dx3 cos kx’,
from which we build the circularly polarized fields
X* = dx2 sin kx’ f dx3 cos lcx’. (212)
Relation of Harmonic Field to Familiar Pattern of Electric Lines of Force.
The harmonic or Coulomb fields are unique to multiply-connected space, with
Betti number R2 3 1. A typical one of these fields has a pattern of lines of force
like that shown in Fig. 3. Consider the case where the space has an enormous
radius of curvature, so that it is nearly flat, except in the immediate vicinity of
the + and - mouths of typical wormholes, a, b, with exceedingly small radii
R, , Rb , * . . . Let the length “underground” of the connection between matching
wormhole mouths be very short, and also of the order R, , Rb , . . * . Let the
fluxes hhf.&*, hqb*, ’ ’ ’ go through the respective wormholes. In this very special
model the field almost everywhere looks like that of Rz pairs of equal and oppo-
site charges :
e = Pa*& + qb*cb + . * ’
(213)
ya*(ro+/raC3 - r,-/ra-“) +
The electrostatic energy of the fields,

e*e/8r, (214)
s
consists of (1) self energy terms of the order
qa*2/Re+ qb*2/Rb+ . . .
(215)
and (2) interaction terms of the form
qa*&b*(ra+b+-’- r,+b--l - ‘t-,-b+-’ + ?-,-b--l )+ - - - . (216)
From this circumstance one can immediately deduce the integrated value of the
products of individual harmonic forms,

co*cb , (217)
s
588 MISNER AND WHEELER

as indicated in Table X. One sees that these harmonic forms give one all the ma-
chinery needed to describe classical charge in a completely consistent and di-
vergence free manner.
Fourier Coeficients as Initial Value Data; Dynamic Equations for Their Rate of
Change.
To specify the initial values of e and h on a space like surface u it is enough,
we conclude, to know the Fourier coefficients in the following expansions:

e = 2 em% + 2 qa*Ca ,
m a=1
(218)
h=gh,L.
m
We have omitted magnetic pole terms from the second expression as without
interest.
The evolution of the electromagnetic field with time proceeds most simply
when the metric does not change with increase of the geodesically normal cotime
coordinate T. Then the dynamical equations (188) reduce to an immediately
integrable form,
dqi*/dT = 0; qi* = const
de,/dT = i&h,,,; (219)
dh,/dT = -kmen;
e, = (M)(e, + ih,& exp (-ik,T) + c.c.;
(220)
h, = (+Q(e, + ih& exp (-i&T) + C.C.
Here the generalizations of the familiar normal mode decomposition of the field
are simple and obvious.
No such simplicity reigns when the metric changes with time in a complicated
way. Then the normal modes themselves alter. Consequently the dynamical
equations (219) acquire new terms,
de,ldT = k,b, + c Iken + c L,,qa*,
a
(221)
dh,ldT = -i&e, + c K,,h, ,
which might loosely be considered to represent (1) the emission of energy into
the transverse field, or absorption of energy from the transverse field, by the
Coulomb fields due to the motion of the charges or other changes in the metric
(2) the pumping or scattering of energy into one transverse mode out of another
mode or out of the metric itself, as a result of changes in the metric. It is true
that one can define the local density of electromagnetic field energy in an un-
 CLASSICAL GEOMETRODYNAMICS 589

ambiguous way. However, the sameis not true of the integrated field energy (40).
There is no common set of space and time coordinates with respect to which one
could even hope to refer the components of a total energy-momentum four vec-
tor. Moreover, the local conservation’ laws, integrated ‘over a closed space such
as we are considering here, produce nothing of interest, only the trivial identity
0 = 0. Consequently it is necessary to use with caution the idea of energy trans-
fer from one mode to another. This caution in no way depreciates either the
dynamical equations (221) or what they have to say about the most interesting
coupling between modes of the field and modifications of the metric.
IV. CHARGE AND MASS AS ASPECTS OF GEOMETRY

We have just, finished examining the response of an electromagnetic field to

curvature and multiple connectednessin the metric, without asking-or having
to ask-about the influence of this field on the metric. However, an adequate
analysis of classical charge and mass (in the senseof Table I) demands that both
sides of the interaction be examined. For this reason (1) we consider a space
like surface u and ask what initial value data has to be given on this surface-
and what, conditions this data must satisfy-to specify uniquely the future
evolution of both the electromagnetic field and the metric (2) we examine the
properties of the Schwarzschild and Reissner-Nordstrom solutions as solutions of
these equations that manifest mass, and mass together with charge (3) we note
an exact solution of the initial value conditions that generalizes the Reissner-
Nordstrom solution to a space endowed with many charges and masses,all mo-
mentarily at rest at the moment of observation and (4) we remark that such
wormhole solutions, together with geons, furnish two techniques of building dis-
turbances in empty space, out of which one can make rich combinations.

THE INITIAL VALUE CONDITIONS AND THE EXISTENCE OF SOLUTIONS TO THL

MAXWELL-EINSTEIN EQUATIONS
Consider a space-like surface u. On it let the trace of the metric tensor, dl’, take
on prescribed values as a function of position,
dl” = g$dzidxk. (22%
Also let e and h have prescribed values on u. Then there will in general exist no
acceptable solution either of the Maxwell equations or of the Einstein equations.
The equations of Maxwell will have a well defined solution if and only if the
initial value conditions (189) are satisfied :
S,e = 0 = S,h. (223)
Similarly, the field equations of Einstein will have a well-defined solution if and
only if certain initial value requirements are satisfied (Eqs. (224) and (225) be-
590 MISNER AND WHEELER

low). Grunted that these conditions as well as Eqs. (223) are fulfilled, then a
solution of the combined Einstein-Maxwell equations with the prescribed initial
values will exist for some finite time.
The initial value problem for general relativity has been studied extensively
by Lichnerowicz (41), and an existence theorem which doesnot require analytic-
ity has been given by Four&s (42). In general relativity, the four of Einstein’s
equations involving RIP - >@,‘R provide the initial value requirements analogous
to Eqs. (.%ZY)in electromagnetic theory-requirements which connect the time
rate of change of curvature on u with the Poynting flux and Maxwell energy
density:
(Pii - &jP);j = -2(gc3’)“z[ijk]eih” = -2[*(e A h)]i (224)
p2 - pijpij + Rc3’ = 2(eaei+ hihi) = 2*(e*e + h*h). (225)
Here we follow the notation of Mme. Four&s (43, who has demonstrated the
existence of solutions of Eqs. (224) and (225) in the case where there is no elec-
tromagnetic field. We let the surface CTbe given by x0 = T = 0, and write go0=
- V’, goi = Vvi(i = 1, 2, 3). Then gij = gijC3)provides the metric d12on c which
is used to raise indices and define covariant derivatives in Eqs. (224) through
(228). With q2 = viqi, we define
hi = >~(g’“‘)““[ijk]fjk , (226a)
ei = (1 + ~2)-1’2(Vi-lj~o- fij?jj), (22613)
P;j = >s(l + q2)-1’z[V-1dgij/aXo - (,l$/dXj + Tja/dxi) V - (q;;j + qj;i)], (227)
and
p = pii (228)
Thus Pij is essentially the time derivative of the metric, while RC3’is the scalar
curvature of the metric gij@)on u. The formulas (226) translate Eqs. (180) into
components relative to the coordinates chosen here.

THE SCHWARZSCHILD AND REISSNER-NORDSTROM SOLUTIONS

The Schwarzschild metric furnishes the most familiar example of a solution
of the field equations and therefore a solution of the inkial value requirements
(224) and (225). In this example the electric and magnetic fields vanish,
e=O=h. (229)
The metric
ds2 = - (1 - 2m*/r)dT2 + (1 - 2m*/r)-‘dr’ + r’(dO* + sin’ 0 &‘j (230)
satisfies the field equations R,, = 0. In contrast, the Ricci curvature tensor for
 CLASSICAL GEOMETRODYNAMICS 591

the metric on the space like surface T = const has the nonzero value
- 2m*p
GM =
Ri m*/r” (231)
( m*/r” )
but still a zero trace,
R@’ = 0, (232)
Also the quantities P;k vanish in virtue of the constancy of the metric. Thus the
initial value requirements (224) and (225) are fully satisfied by the Schwarzschild
solution.
The Schwarzschild metric appears at first sight to have a singularity at the
point r = 2m*. However the Ricci curvature tensor (231) is perfectly well be-
haved at this point, as is also the full Riemann curvature tensor in the appro-
priate mixed covariant-contravariant representat,ion. To bring part of this regu-
larity into evidence, we make a coordinate transformation of a type suggested
before by more than one writer (44)‘“.
r = [l + (m*l2p)l”p. (233)
Then the metric takes the form:

ds2 = -(; ;“m$;) dT” + (1 + m*/2p)4(dp” + p2 d!?). (234)

Here
dL?’ = de2 + sin’8 dp” (235)
is the metric of a unit sphere. This result can be interpreted as follows: In making
the transformation (233) we have redefined the differentiable structure of the
I* The regularization in all of these cases removes only the singularity in the space part
of the metric, not that in the time part, and therefore falls short in every instance of the
regulariz&ion described in the text. Einstein and Rosen, supposing that they would not
be able to eliminate the singularity in the space part of the metric by a coordinate trans-
formation alone, also changed the sign of the constant’ of gravitation. Of course this step
gives a negative value for the electromagnetic contribution to the mass, and for the energy
of electromagnetic fields in general, in contradiction with experience. Einstein and Rosen
pictured their regularizing transformation in terms of a space nearly flat except near par-
ticles, and a mirror space, close to and parallel to the first space, with a bridge across at the
Schwarzschild singularity. On this picture the charge or charges in the primary space need
not add up to zero. In contrast, the analysis in terms of harmonic forms (1) imposes no such
mirror property on space (2) treats all points of space as on the same footing (3) brings in
the idea of a real cont,inuation of electric lines of force through a wormhole (4) normally
contemplates a closed space, with no excess unbalanced lines of force that have to end at
infinity.
592 MISNER AND WHEELER

manifold we are considering, so that now p, rather than r, is to be called a diferen-

tiable function. As a result, the metric d12 on the space-like surface T = 0 is
now nonsingular at the Schwarzschild radius. The space-time metric ds2 remains
singular, however, since g T T vanishes at the Schwarschild radius, i.e., gTT --f CC.
We do not know how to eliminate this singularity explicitly for all time, but ac-
cording to the initial value theory of Lichnerowicz and Four&, we can remove it
for some undetermined, but nonzero, length of time. This we see by noting that
the initial value conditions (224) and (225) leave gTT = -l’* completely arbi-
trary on the initial surface. We take as initial conditions therefore, the following:
ds= = -V2dT2 + d12, (236)
d12 = (1 + m*/2p)4(dp2 + p2 dQ*), (237)
eg,,/‘aT = 0. (238)
We may, for instance, choose V = 1. Since V need only be differentiable, not
analytic, we may also choose V = V seha.arrehiidas in (234) when p is a small
distance Eaway from the singular surface p = m*/2, but fill in over the previously
singular surface with differentiable, nonzero, values of V. With gTT specified on
the initial surface T = 0 in a manner such as this, the solution that results from
integrating the Einstein equations is at least asymptotically stable: It will exist,
and coincide (45) with the Schwarzschild solution for a cotime T which is ar-
bitrarily large as p + w. Furthermore, for any fixed po > m*/2 a solution of
this type may be found which exists for all p > p0 for at least any prescribed
length of time, say 101” years.
These same arguments apply equally well when the center of attraction is
endowed with charge in addition to mass. We start with the Reissner-Nordstrom
(46) solution for this problem. It differs from the Schwarzschild solution (230)
only in the replacement of the factor (1 - 2m*/r) everywhere by
(1 - 2m*/r + q*‘/r2).
The dual of the electric field in this case is
*e = q* sin 8d6’ A c&. (239)
Then we write
r = p[( 1 + m*/2p)’ - q*‘/4p2] (240)
a,nd find for the metric
ds* = - VR.v2dT2+ dl’, (241)
where

VRN*= 1 - m*$q “>“/[(l + g>’ - ($)3

(
 CLASSICAL GEOMETRODYNAMICS 593

and
dt = [( 1 + m*/2p)’ - (4*/2p)‘] (dp’ + p2 d@. GW
The singularity of the space-time metric due to the vanishing of VRN when
2p = (m*” - q*2)“2, may again be removed by modifying gTT = - V2 on an
initial surface near this critical value of p. To make possible this regularixation
we must demand that the mass exceed the minimum limit that is associated by
general relativity with the charge in question:
m*(cm) > q*(cm)
or
m(g) > G-l”g = (3.88 X 103g/esu)q(esu). (244)
This condition is quite incompatible with the properties of the “dressed” par-
ticles studied by experiment. This circumstance stresses again that we are
dealing here exclusively with a description of classical charge and mass that has
nothing whatsoever directly to do with the particles found in nature.
Thus far we have shown how the singular surface in the Schwarzschild metric
may be removed, but there apparently remains in Eqs. (237) and (243) a singu-
larity as p --f 0. We shall now see that we cannot call this a singularity unless
we are willing to say that the metric is also singular as p -+ + 00. The transforma-
tion
J (m*2 - q*y
(m*2 Pp*2)l!2 - 2/) G’W

maps the entire manifold, p > 0, onto itself in a differentiable, l-l, metric pre-
serving way. Therefore the vicinity of p = 0 is in every meaningful way equiva-
lent to the vicinity of p = ~0. The initial surface T = 0 is a complete Riemannian
manifold, that is, every geodesic can be continued to infinite length. One might
nevertheless wish a stronger property to specify the global meaning of “non-
singular”. One might insist that the manifold be either closed or that there be
only one region (either p + 00 or p + 0) where the space is asymptotically flat.
An example of this type may be constructed from the charge-free Schwarzsehild
solution (237). We complete the definition of the mapping J of Eq. (246) by
defining

TJ T, (246)

The mapping J then has no fixed points, and we may identify the point x with
594 MISNER AND WHEELER

the point Jx to obtain a new manifold with only one region, p + + m, which is
asymptotically flat. The surface T = 0 in this manifold is topologically equiva-
lent to projective 3-space P3 (seeSec. III B) with one point removed, “the point
at CC“. This procedure cannot be applied to the Reissner-Nordstrom solution,
for although the metric is invariant under J, the electromagnetic field changes
sign and is therefore not consistent with the identifications. The time dependence
of this nonsingular modification of the charge-free Schwarzschild solution has
been investigated by Dubman (47) to first order in 1”. He finds’that the area
of the critical sphere begins to decrease.

MANY CHARGES AND MASSES

The Reissner-Nordstrom initial conditions (239) and (241-243) can be gen-

eralized to the case of N wormhole mouths endowed with charge and mass in
the following way: Set
di’ = - V2dT2 + (x2 - +*)*(dx* + d.y2+ dz’), (249)

dx’ A dxk,

ag,,v/dT = 0 (251)
at T = 0, let V be any nonsingular, positive definite function, and let each x,
C#J,
satisfy the flat space Laplace equation

(a*/a~* + a2/av2 + a*/a& = 0. (252)

These initial values make the initial surface a complete Riemannian manifold
when we choose
x = 1 + c wz/r,,+ = F Paira, (253)
a

with r, = ( r - r, 1and ct= 3 ( Pa 1, a = 1, 2, . . * N. The points ro = (x, , YJ~, 2,)

are of course excluded from the manifold, The restriction (Y, 2 1&, ( arises from
the fact that the metric would be singular if (x” - 4”) = 0 at any point. The
completenessrequirement also excludesnegative massesas well as multipole terms
in either x or 4. The cy’sand p’s are related to massand charge respectively, but
do not give the massesand charges directlylg. In the special case where there is
no charge, + = 0, one obtains a class of problems which has been studied by
Lichnerowicz (49). As long as only instantaneously static solutions of the initial
value conditions are desired, a similar (x2 - +“)” factor may be used to modify
I9 See Lindquist and Wheeler (48) for the relationship between m,,* and the 01. , and for
a study of the time dependence of some solutions of this type.
 CLASSICAL GEOMETRODYNAMICS 595

any three dimensional metric in place of the dx2 + dy’ + dz2of Eq. (249). The
initial value conditions then reduce to the form
& + S& = 0 (254)
plus a similar equation for C#J
that replaces Eq. (252). Here --a and R are the
Laplacian and scalar curvature of the unmodified metric.

RELATION TO GEONS
Electromagnetic geons (50) are objects built out of electromagnetic radiation
and held together by mutual gravitational attraction. The metric is great,ly
altered in the region occupied by the geon but the topology of space is still
isomorphic to the normal Euclidean topology. Shortwavelength radiation sent
directly at the geon comes out again whether it experiences a small or a large
alteration in direction in the encounter. In contrast, shortwavelength radiation
directed at one mouth of a wormhole will come out the other mouth. Despite this
difference in properties, the two objects curve space in the same way and are
indistinguishable, mass for mass, as regards their l/r2 gravitational attraction.
These two techniques for constructing mass-like solutions of the equations of
geometrodynamics need not be employed separately. One can envisage an object
endowed both with circulating radiation and with wormhole mouths, drawn
together and moving about in a limited region of space as a consequenceof mu-
tual gravitational attraction. In addition, a sufficient number of wormhole
mouths of the same sign moving on nearly identical orbits will simulate a cur-
rent. They can create magnetic fields strong enough to have a substantial or
even a dominating effect on the structure. It follows that the variety of objects
that can be built out of curved empty space is exceedingly rich and very far
from having been explored or even surveyed.

EQUATIONS OF MOTION
Masses and charges are not objects distinct from the fields in the theory we
are presenting here-not even to the extent of being singularities in the fields.
It is therefore obvious that the field equations determine the motions of the
massesand charges. We must, however, insist that these motions correspond in
the appropriate limit to the Newtonian and Lorentz force laws, for these laws
expressmost basic and well tested properties of idealized classicalpoint particles.
In geometrodynamics, mass and charge are not idealized as properties of poinl
particles, they are rather aspects of the geometrical structure of space. To dis-
cussequations of motion it will be necessary to view space-time with less resolv-
ing power than heretofore, and collapse the entire structure of a mass or charge
down to a point whose motion we may then compare with the laws of Newton
and Lorentz. For instance, an entire region about one mouth of a wormhole, as
596 MISNER AND WHEELER

in Fig. (3), may be called one point xl , and a similar region about the other
mouth may be called x2 . The tube connecting these two region we simply leave
out altogether. A similar idealization may be made for a geon, and likewise for
an object that is built out of radiation and wormholes. In this way the manifold
of Fig. (3) is mapped in a singular way onto a topologically Euclidean space, and
the metric which is carried over will have two singularities, at r1 and .r2 . Simi-
larly for many massesand charges: by collapsing each to a point we ignore all
details of inner structure and passto a limit where it becomesappropriate to com-
pare motions with those predicted by the force laws of Newton and Lorentz.
The problem so defined is the problem so carefully studied by Einstein, Infeld,
and Hoffman. Those and other investigators derive the mechanical equations of
motion of singularities” from the field equations. Consequently we can conclude
that the objects discussedhere-charges and massesbuilt out of curved empty
space and not,hing more-satisfy in the appropriate limit the equations of New-
ton and Lorentz.

No UNITS BUT LENGTH IN PURELY CLASSICAL PHYSICS

The purely geometrical character of classical physics shows itself in the cir-
cumstance that space curvatures are measured in cm-‘, electromagnetic fields
in cm-‘, charges in cm, and massesin cm. There is no place for any units other
than length. A parable may be pardoned, of a kingdom where distances to the
north were sacred and measured in miles, while those to the east and west and up
and down were measured in feet. A special education was needed to calculate
diagonal distances from coordinate readings until the discovery was made that a
single constant of nature sufficed for the theory of the calculation. Thereaft#er
much attention was devoted to “explaining” how nature happened to be en-
dowed with a natural “slope” of 5280 feet per mile. The parable perhaps makes
one charitable towards similar attempts to “explain” why the speed of light
should be 3 X lOi cm/set. That Roltzmann’s constant k: is only a conversion
factor between two chance units of energy is of course a familiar idea. It is less
familiar that grams and centimeters are two equivalent units for length; that
the Schwarzschild radius of an object, r&w = Gm/c’ = m*(cm), is a purely geo-
metrical way to characterize its inertia. We have seen that classical electromag-
netism likewise requires nothing but units of length for its simplest expression.
The field is only a manifestation of curved empty space. Classical physics in the
senseof Table I reduces to pure geometry.
2.Einstein, Infeld, and Hoffman (51) deal with uncharged singularities of finite mass in
slow motion; Infeld and Schild (52) treat uncharged singularities of infinitesimal mass
moving at arbitrary velocities; and D. M. Chase (53) analyzes charged singularities of
infinitesimal mass moving at arbitrary velocities.
 CLASSICAL GEOMETRODYNAMICS 597

V. PROBLEMS AND PROSPECTS OF GEOMETRODYNAMICS

Certain gaps remain to be filled in the logical structure of geometrodynamics

and almost everything remains tlo be done to exploit the richness of this subject

CASE OF NULL FIELDS

The algebraic relations (4), (5) of already unified field theory were developed
quite generally, but the differential equations (7), (8) were derived only on the
assumption that the Ricci curvature tensor is non-null. For this simplifying as-
sumption to fail, for the tensors R,, (and FPy) to be null, it is necessary that the
two electromagnetic field invariants should simultaneously vanish:
e.h = 0; h2 - e2 = 0. (256)
At a given moment of time this condition will ordinarily be fulfilled only on cer-
tain isolated lines in space. As time advances these lines will trace out surfaces
in space time,
x8 = x0(& r]). (257)
On such surfaces Eq. (7) gives no well defined value for the vector olg, the gradi-
ent of the complexion of the electromagnetic field. Is (Yitself well defined? Can
the equations of already unified theory be formulated in such a way as to hold
right across such a surface? Does the presence of such surfaces impose any addi-
tional topological or periodicity requirements on the Ricci curvature tensor?
Do any special problems arise when the invariants (256) vanish, not merely on
surfaces in space time, but throughout regions of greater dimensionality? These
questions all obviously hang together.

WHY ONLY ONE KIND OF CHARGE?

A second group of questions concerns charge. The interpretation of charge
in terms of lines of force trapped in the topology allows all charges to be as well
purely magnetic as purely electric. However, the difference between the two
possibilities is well known to be only one of names. The duality transformation
f’ = *f; *f’ = * *f = -f
or
h’ = e; e’ = -h (258)
renames magnetic charges so that they are all electric, in conformity with the
usual convention. The charges associated with all wormholes-or homology
classesin dimension two-can again be renamed as purely electric when in the
original frame of reference each is a mixture of an electric charge ei and a mag-
 598 ~V~SNER AND WIIIEEL~~:R

netic pole pi , provided that t.he ratio of the two has for each wormhole the same
value
ei/pi = cos P/sin 0. (259)
Then the duality rotation
f’ = e*+ (260)
accomplishes the renaming in accordance with tradition. However, this renam-
ing of all charges as electric is only possible when the ratio (259) of the two
kinds of charges is identical for all the homology classesor wormholes in the
original duality reference system. Can this ratio condition be restated in other
terms?
When the regions over which space is appreciably curved are small compared
to the distances between different wormhole mouths, then there is an approxi-
mate localizability of the typical charge. We are invited to look at it in a local
Lorentz frame where it appears momentarily to be at rest. In this frame the
field close to the charge is practically purely electric. Consequently there will be
a surface, either around the mouth of the wormhole, or deep down in its throat,
where the complexion, (Y, is either everywhere zero or equal to some integral
multiple of 27r. Moreover, we expect the surface (Y = 2m to remain tied to the
neck of the wormhole in the sameway as time advances; similarly for other worm-
holes, i. Insofar as the characteristic complexion, achar= (Y(~),for each is well de-
fined, the condition that all charges be electric therefore says

(9 ti) = positive or negative

ff -a 2* . (261)
( integer or zero ) *
This condition reminds one of the periodicity requirement,

CY,,dx” = 2n’integer (262)

f
(Eq. 77) imposed by already unified field theory upon the Ricci curvature tensor.
Nevertheless, we have seenno way to derive (261) from (262). Therefore we are
not clear whether the requirement that all charges be electric is a part of the
theory as it exists, or whether it has to be added to the theory.
For all magnetic poles to be zerozl it is enough, according to Eq. (145), that
the field should be derivable from a 4-vector potential:
f = da (263)
or
spy= aaJax" - aa,/axv. (53344)
21 See Malkus (64)for the convincing experimental evidence against the existence of free
magnetic poles.
 CLASSICAL GEoMETRoDYNAMIcS 599

This is an assumption that goes beyond Maxwell’s equations, for in Sec. III-D
we saw that there exist solutions of Maxwell’s equations which display electric
and magnetic charge side by side. There, however, electromagnetism was con-
sidered within the arena of a prescribed metric. Only in this framework of ideas
is it clear that the exclusion of magnetic poles-or the existence of a 4-potential-
is a demand independent of, and supplemental to, the Maxwell equations them-
selves.
When we turn to the full coupled Einstein-Maxwell equabions, and note that
the Ricci curvature R,, is completely determined by the field, then it is not clear
that the nullity of magnetic poles, or the existence of a 4-potential, has to be
added as a supplementary condition. Moreover, no example is known of a solu-
tion of the coupled Einstein-Maxwell equations in which the magnetic poles can-
not all be eliminated by a duality rotation. It is obviously an important issue
of principle to decide whether the existence of a 4-potential really has to be added
to Eqs. (41, (5), (7), and (8) of already unified field theory. Can it be derived
from those equations?
Another issue presents itself: Can Eqs. (4), (5), (7), and (8) of geometro-
dynamics, with or without a possible supplementary condition about a ~-PO-
tential, all be derived from a single variational principle? Some work has been
done that is relevant to this issue (55) but the problem itself appears never even
to have been formulated in the literature.
Now we turn from the question of the best formulation of the field equations
to the problem how best to deal with the initial value requirements. For the
case of pure electromagnetism one has long ago learned to satisfy automatically
the requirement div h = -6h = 0 on the initial hypersurface by introducing as
primary data, not h itself, but a S-potential a that generates an acceptable h.
In the case of the coupled equations of geometrodynamics, we meet nonlinear
initial value requirements (224) and (225) on the measures, P;j, of the time
derivative of the metric. Does there exist any kind of freely choosable superpo-
tential, analogous to a: which will generate a tensor P;i which in turn will auto-
matically satisfy (224) and (225)? If so, the properties of such a superpotential
should reveal much about the truly independent variables of geometrodynamics.
To clarify this point is essential for the understanding of already unified theory,
for its most efficient application, and for illumination on what it means to quan-
tize it.
Still another question of principle raises itself. In electromagnetism we can
make a linear combination of two solutions to obtain a third. Geometrodynamics
is of course nonlinear. Does there nevertheless exist a method to combine two
solutions to obtain a third? To search for such a combinatorial scheme would
seem to demand an investigation of the continuously infinite dimensional space
of field histories. On raising this issue, we were kindly advised by our colleague,
600 MISNER AND WHEELER

Professor V. Bargmann, that the methods of Lie (56) should suffice to obtain a
definitive answer to our question.
To the questions of principle that we have raised, about null fields, about the
existence of a 4-potential, about a possible superpotential, and about the combi-
natorial properties of the infinite dimensional space of field histories, there should
be added many issues about the consequences of the classical theory, on which
we shall only touch”. (1) How wide a variety of wormholes and multiple con-
nectedness is topologically conceivable. (2) When the deterministic evolution of
the metric with time leads at a certain moment to fission or coalescence of worm-
hole mouths or to any other change in topology, what new phenomena occur?
(3) What can one do to construct closed mathematical expressions for the metrics
of spaces that show as much as possible of the richness of geometrodynamics-
gravitational waves, wormholes, shocks, trapped radiation, and combinations
of all these features?
One’s attention is inevitably drawn beyond these fascinating questions to the
still deeper issue, what is the nature of quantum geometrodynamicsz3 and what
ideas have to be added to quantum geometrodynamics for the description of
nature?
ACKNOWLEDGMENTS
We wish to express our appreciation to Valentine Bargmann, Peter Bergmann, Hugh
Everett, and Markus Fiers for discussions on the physical side, and to Pierre Connor, J.
Geheniau, J. A. Schouten, Donald Spencer, Hale Trotter, and many other colleagues for
illumination on the mathematics of geometrodynamics. We are grateful to Professor S. R.
de Groot and the other members of the Lorentz Institute for the hospitality shown us during
our stay in Leiden.

RECEIVED: July 11, 1957

REFERENCES
1. C. W. MISNER, Revs. Modern Phys. 29, 497 (1957).
2. J. A. WHEELER, Annals of Physics 2, 604 (1957).
3. G. Y. RAINICH, Trans. Am. Math. Sot. 27, 106 (1925).
4. G. Y. RAINICH, “The Mathematics of Relativity.” Wiley, New York, 1950.
6. J. L. SYNGE, “Principal Null-Directions Defined in Space Time by an Electromagnetic
Field,” No. 1 in the University of Toronto Studies, Applied Mathematics Series.
Univ. of Toronto Press, Toronto, 1935.
6. J. L. SYNGE, “Relativity: The Special Theory,” p. 326. North Holland Publishing
Company, Amsterdam, 1956.
7. W. B. BONNOR, Proc. Phys. Sot. A67, 225 (1954).
8. LOUIS MARIOT, thesis, “Le champ dlectromagnetique pur en relativite generale,” Chap-
ter IV. University of Paris, mimeograph, privately reproduced, April 6, 1957.

24 See Power and Wheeler (67) for a table of the analogies between geometrodynamics
and hydrodynamics, and the problems that are suggested by this analogy.
23 In this connection see papers by Everett, Misner, and Wheeler (68).
 CLASSICAL GEOMETRODYNAMICS 601

9. A. EINSTEIN, “The Meaning of Relativity,” 3rd ed., p. 107. Princeton Univ. Press,
Princeton, 1950.
10. BERNHARD RIEMANN (1826-1866), Habilitationsvorlesung of June 10, 1854 on entry into
the philosophical faculty of the University of Gottingen, “Ueber die Hypothesen
welche der Geometrie zu Grunde liegen,” in B. Riemann, “Gesammelte Mathe-
matische Werke,” (H. Weber, ed.), 2nd ed. Reprinted by Dover Publications, Inc.,
New York, 1953.
11. CLIFFORD, Nature 8, 14 (1873).
12. H. WEYL, “Raum Zeit Materie,” 4th ed., Section 34. Springer, Berlin, 1921; quotation
from Weyl’s “Philosophy of Mathematics and Natural Science” (translated by Olaf
Helmer), p. 91. Princeton Univ. Press, Princeton, New Jersey, 1949.
13. A. EINSTEIN AND N. ROSEN, Phys. Rev. 48, 73 (1935).
14. J. A. WHEELER, Phys. Rev. 97, 511 (1955).
16. K. K. DARROW, Physics Today, 9, No. 8, 24 (1956).
16. See, for example, F. DYSON, “Advanced Quantum Mechanics,” mimeographed notes,
pp. 120, 167. Cornell University, Ithaca, 1954.
17. L. INFELD, Acta Phys. Polonica 10, 284 (1950).
J. CALLAWAY, Phys. Rev. 92, 1567 (1953).
18. First quote from W. K. CLIFFORD, “Mathematical Papers” (R. Tucker, ed.), p. 21.
Macmillan, London, 1882.
second and third quotes from W. K. CLIFFORD, “Lectures and Essays,” (L. Stephen and
F. Pollock, eds.), Vol. I, pp. 244 and 322, respectively. Macmillan, London, 1879.
19. F. CAJORI, revision of Motte’s English translation of I. Newton’s “Mathematical
Principles of Natural Philosophy.” Univ. of California Press, Berkeley, 1934.
20. M. FIERZ, “Ueber den Ursprung und die Bedeutung der Lehre Isaac Newton’s vom abso-
luten Raum,” Gesnerzks, 11, 62 (1954).
Il. F. PATRIZZI, “Nova de Universis Philosophia . ,” Part IV, “Pancosmia,” Book I,
“de Spacio Physico”. Meiettus, Venice, 1593.
22. “The Taittiriya Upanishad with commentaries” (translated by A. Mahadeva Sastri),
p. 293 and pp. 305-307. Mysore, 1903. See also the forthcoming Vol. III of “The
Upanishads” (translated by Swami Nikhilananda). Harper, New York.
23. J. A. SCHOUTEN, “Ricci-Calculus: An Introduction to Tensor Calculus and Its Geo-
metrical Applications,” 2nd ed., pp. 36 and 46. Springer, Berlin, 1954. See also foot-
note 4 of present paper.
2.4. JOHN I,. KELLEY, “General Topology,” Van Nostrand, New York, 1955.
26. L. PONTRJAGIN, “Topological Groups,” Chapter II. Princeton Univ. Press, Princeton,
1946.
26. P. S. ALEKSANDROV, “Combinatorial Topology,” Vol. I, Chapter I. Graylock Press,
Rochester, 1956.
27. GEORCES DE RHAM, “Varietes Differentiables,” p. 1. Hermann, Paris, 1955.
28. C. MILLER, “The Theory of Relativity,” p. 361. Oxford Univ. Press, London and New
York, 1952;
L. LANDAU AND E. LIFSHITZ, “The Classical Theory of Fields,” p. 336. Addison-Wesley
Press, Cambridge, 1951.
29. JOHN MILNOR, Ann. Math. 64, 399 (1956).
SO. E. CARTAN, “LeCons sur la geometric des espaces de Riemann.” Gauthier-Villars,
Paris, 1946.
C. CHEVALLEY, “Theory of Lie Groups I,” Chapter 3. Princeton Univ. Press, Prinee-
ton, 1946.
602 MISNER AND WHEELER

W. V. D. HODGE, “The Theory and Application of Harmonic Integrals.” Cambridge

Univ. Press, London and New York, 1952.
A. LICHNEROWICZ, “Algebre et analyse lineaires,” 2nd ed., Part 2, Chapter 1. Masson,
Paris, 1956.
G. DE RHAM, “Varietes Differentiables.” Hermann, Paris, 1955.
31. S.EILENBERGANDN.STEENROD, “Foundations of Algebraic Topology,” Vol. I. Prince-
ton Univ. Press, Princeton, 1952.
32. H. SEIFERTAND W. THRELFALL, “Variationsrechnung im Grossen,” especially pp. 16-21.
Reprinted by Chelsea, New York, 1951.
3. S. LEFSCHETZ, “Introduction to Topology.” Princeton Univ. Press, Princeton, 1949.
34. H. SEIFERT AND W. THRELFALL, “Lehrbuch der Topologie,” reprinted! by Chelsea,
New York, 1947.
65. J. P. SERRE, Ann. Math. 64, 439-441 (1951).
$6. W. V. D. HODGE, “The Theory and Applications of Harmonic Integrals,” pp. 88 and
100. Cambridge Univ. Press, London and New York, 1952.
G. DE RHAM, “Varietes differentiables: formes, courants, formes harmoniques,” Chap-
ter 4. Hermann, Paris. 1955.
97. See E. GOURSAT, “Cours d’analyse mathematique,” 2nd ed., Vol. 2, pp. 632-637. Gauthier-
Villars, Paris, 1911. See also M. RIESZ, Acta Math. 81, l-223 (1949).
G. F. DUFF, Can. J. Math. 6, 57 (1953).
98. A. RAYCHAUDHURI, Phys. Rev. 104,544 (1956); Phys. Rev. 106,172 (1957).
A. KOMAR, Phys. Rev. 104, 544 (1956).
69. Y. FOURES-BRUHAT, Acta Math. 88, 141 (1952) ; see also DUFF, ref. S7.
40. L. LANDAU AND E. LIFSHITZ, “The Classical Theory of Fields,” translated by M.
Hamermesh, p. 320. Addison-Wesley, Cambridge, 1951.
41. A. LICHNEROWICZ, “Prohlbmes Globaux en Mdchanique Relativiste.” Hermann, Paris,
1939; J. Math. pure appl. (9) 23, 37 (1944); Helo. Phys. Acta, Suppl. IV, 176 (1956).
.@. Y. FOURER-BRUHAT, Acta Math. 88, 141 (1952).
43. Y. FOURES-BRUHAT, J. Rational Mech. Anal. 4, 951 (1956).
44. G. LEMAITRE, Ann. sot. sci. Brzcxelles 63A, 51 (1933).
A. EINSTEIN AND N. ROSEN, Phys. Rev. 48, 73 (1935).
J. L. SYNGE, Proc. Roy. Irish Acad. DA, 83 (1950).
I,. LANDAU AND E. LIFSCHITZ, “The Classical Theory of Fields,” p. 310. Addison-Wesleyr
Press, Cambridge, 1951.
45. K. STELLMACHER, Math. Bnn. 116, 136, 740 (1938).
46. K. REISSNER, Ann. Phys. (4) 59, 106 (1916).
L. NORDSTROM, Proc. Amsterdam Acad. 20, 1238 (1918).
47. MORTON R. DUBMAN, senior thesis submitted to Princeton University in partial ful-
fillment of the requirements for the A.B. degree, 1957 (unpublished).
48. R. W. LINDQUIST ANY J. A. WHEELER, Revs. Modern Phys. 29, 432 (1957).
49. A. LICHNEROWICZ, J. Math. pure appl. (9) 23, 37 (1944).
60. J. A. WHEELER, Phys. Rev. 97, 511 (1955).
E. POWER AND J. A. WI-IEELER, Revs. Modern Phys. 29, 480 (1957).
51. A. EINSTEIN, L. INFELD, AND B. HOFFMAN, Ann. Math. 39, 65 (1938).
62. L. INFELD AND A. SCHILD, Revs. Modern Phys. 21, 408 (1949).
63. D. M. CHASE, Phys. Rev. 96, 243 (1954). S ee also L. INFELD, Revs. Modern Phys. 29, 398
(1957).
64. WILLEM V. R. MALICUS, Phys. Rev. 81, 315 (1951).
66. W. PAULI, Physik 2. 20, 457 (1919).
 CLASSICAL GEOMETRODYNAMICS 603

H. WEYL, Math-Seitz. 2, 384 (1918); Ann. Phys. (4) 69, 101 (1919); Physik Z. 22, 473
(1921).
R. BACH, Math. Z. 9, 110 (1921).
C. LANCZOS, Phys. Rev. 39, 716 (1932) and 61. 713 (1942); Ann. Math. 39, 842 (1938);
Revs. Modern Phys. 21, 497 (1949).
56. S. LIE AND F. ENGEL, “Theorie der Transformationsgruppen,” 3 ~01s. Teubner, Leipzig,
1888-1893.
s. LIE, “Vorlesungen iiber continuierliche Gruppen mit geometrischen und anderen
Anwendungen” (G. Scheffers, ed.). Teubner, Leipzig, 1893.
67. E. POWER AND J. A. WHEELER, Revs. Modern Phys. 29,480 (1957).
68. H. EVERETT, Revs. Modern Phys. 29, 454 (1957).
C. MISNER, Revs. Modern Phys. 29, 497 (1957).
J. A. WHEELER, Revs. Modern Phys. 29, 463 (1957); Annals of Physics 2, 604 (1957).