Energymomentum Complex in General

Relativity and Gauge Theory

David Hestenes1
Department of Physics and Astronomy
Arizona State University, Tempe, Arizona 85287-1504
Alternative versions of the energymomentum complex in general relativity
are given compact new formulations with spacetime algebra. A new unitary
form for Einstein’s equation greatly simplifies the derivation and analysis
of gravitational superpotentials. Interpretation of Einstein’s equations as a
gauge field theory on flat spacetime is shown to resolve ambiguities in ener-
gymomentum conservation laws and reveal intriguing new relations between
superpotential, gauge connection and spin angular momentum with rich new
possibilities for physical interpretation.

I. Introduction
General relativity (GR) has recently been reformulated as a gauge theory in
terms of spacetime algebra STA [1, 2, 3]. This brings new mathematical tools
to the formulation and analysis of physical problems in GR as well as new ideas
for refining and extending the theory. Those tools and ideas are applied in this
article to long-standing problems about energymomentum localization in GR.
Though many different mathematical formalisms for GR have been proposed,
the “covariant tensor formalism” originally introduced by Einstein is still the
clear favorite in the literature. However, since the seminal paper on gauge theory
by Weyl [6], it has been clear that orthonormal frames called tetrads, or vier-
beins, are essential for incorporating the Dirac equation into GR. This involves
an elaborate “tetrad formalism” that considerably complicates the classical ten-
sor formalism, so it has remained on the fringes of mainstream GR. The “STA
gauge theory formalism” employed in this paper can be regarded as a marriage
of tensors and tetrads that enhances both with new algebraic methods, though
it is recognized as a device to make equations coordinate-free [8]. While the
classical features of GR are incorporated without change into the formalism,
advantages of the new algebraic methods are illustrated in an analysis of ener-
gymomentum tensors. Thus, we have one more way to do GR. However, gauge
theory provides added value to GR with a new way to incorporate a flat space
background and thus interpret GR as a gravitational field theory on Minkowski
spacetime without altering its mathematical content. Indeed, it unmasks the
tetrad formalism as an indirect way to smuggle flat space into GR. The main
purpose of this paper is to show how STA gauge theory provides new insight
into the problem of defining a suitable energymomentum tensor for gravity.
1 electronic mail:[email protected]

1
 To set the stage, let us review some highlights in the troubled history of
energy-momentum conservation in GR. The core of GR is Einstein’s famous
gravitational field equation

Gµν ≡ Rµν − 12 gµν R = κTµν , (1)

where κ = 8π in natural units, and the meaning of the terms is presumed to

be known by the reader. Shortly after establishing this equation in his seminal
paper of 1915, Einstein set forth a general argument for decomposing it as
follows [7]:

gGµα = κ( E Tαµ − g E tµα ) = κgTαµ , (2)

1
where g ≡ |det (gµν )| 2 . Let’s call this the Einstein split of Einstein’s tensor
density gGµν . Einstein interpreted g E tµα as the gravitational energymomentum
density and defined it so that

∂µ E Tαµ = 0, (3)

where ∂µ = ∂/∂xµ . It follows that

∂µ [g(Tαµ + E tµα )] = 0 , (4)

which Einstein interpreted as a general energymomentum conservation law.

Problems with Einstein’s split began to surface immediately [10], beginning
with the fact that it is not covariant because E tµα is a pseudotensor rather a
tensor. Thus, E tµα has the perplexing property of vanishing in the absence of
gravitation (flat space) for retangular coordinates but not for polar coordinates.
And in curved space, it is always possible to select coordinates so that E tµα = 0
at any given point.
Furthermore, over subsequent years, many authors also noted that the pre-
cise functional form of Einstein’s E tµα is not determined by a split satisfying (3),
which is all one needs to get the “conservation law” (4). Thus, for any pseu-
dotensor of the form Xαµν = −Xανµ , we have the identity ∂µ ∂ν Xαµν = 0. Hence,
we can define a new split of Gµα into primed quantities T
µα = E Tαµ + ∂ν Xαµν and
gt
µα ≡ g E tµα + ∂ν Xαµν replacing the unprimed quantities in (3) and (4). Other
authors obtained somewhat different splits with quantities of different density.
Most noteworthy of these is the “Landau-Lifshitz split,” which has the form
[11]:

g 2 Gµν = κ( L T µν − g 2 L tµν ) = κg 2 T µν . (5)

This split yields a symmetric gravitational energymomentum pseudotensor L tµν .

The most thorough study of gravitational energymomentum laws was carried
out by Møller [12], who refered to the various alternatives as the energymomen-
tum complex. He argued that components of the metric tensor gµν may not
be the best choice for gravitational field variables; instead, components of an

2
(orthonormal) tetrad field are more suitable, because they are essential for in-
corporating the Dirac equation into the theory. This enabled him to define a
split of the form (2) with a new expression for tµα that is a true tensor. Let’s call
it the Møller split. After carefully evaluating its properties and explaining how
it resolves the problematic aspects of alternatives, he declared himself satisfied
with the result. His result has scarcely been noticed in the literature, but we
shall see that there is much to recommend it when formulated in STA gauge
theory.
Møller noted, however, that his gravitational energymomentum tensor, though
covariant, is still not unique, because its values can be changed by local Lorentz
rotations (gauge transformations) of the tetrad field. He investigated ways to
make it unique with supplementary conditions, but wisely did not insist on any
of them. Instead he concluded that energy in the gravitational field is not local-
izable, that is, no exact physical meaning can ascribed to “energy content in a
finite part of space.” Einstein came to the same conclusion about his pseudoten-
sor, which he maintained throughout his life as the best available descriptor of
gravitational energymomentum.
The view that gravitational energymomentum and angular momentum con-
servation laws are well-defined only asymptotically seems to be the general con-
sensus of workers in GR today [8, 9], though they recognize that this entails
problematic ambiguities in the detection of gravitational waves. Still, there are
dissenters who argue that the ambiguities can be removed by regarding GR
as field theory on (flat) Minkowski spacetime. In particular, GR has been de-
veloped as the field theory of a spin-2 particle, the graviton [13, 14, 15]. It
is noteworthy that generating a self-consistent gravitational energymomentum
tensor is a key ingredient of that approach.
On quite a different note, Rosen observed that physicists working with GR
invariably have a flat space background in mind, so he elevated the existence of a
flat background to an axiom in his Bimetric general relativity [16]. Later it was
noted that introducing a flat background metric is mathematically equivalent
to the tetrad formulation of GR [17], though it appears to be quite different
conceptually. Indeed, Rosen was most interested in using his Bimetric theory
to generate alternatives to Einstein’s equation (1), especially to eliminate Black
Hole singularities.
Introducing a Minkowski background into GR does not alter its mathemat-
ical content, as is evident in the gauge theory formulation below. Moreover,
it is essential for energymomentum conservation as a consequence of Poincaré
invariance in special relativity, with its unparalleled success in relativistic quan-
tum theory. Indeed, Babak and Grishchuk argue that “the Minkowiski metric
is not an artificially imposed “prior geometry” but a reflection of experimental
facts,” and they employ it to derive a unique gravitational energymomentum
tensor [18] without the deficiences of standard pseudotensors. Though differing
in detail, the present approach supports their general conclusions.
In preparation for the main event, Section II reviews the necessary basics
of spacetime algebra, while Sections III and IV summarize essentials of gauge
theory gravity and its relation to general relativity. In Section IV Einstein’s

3
gravitational field tensor is given a simple new “unitary” formulation directly
in terms of the curvature tensor without the usual reference to the Ricci tensor.
This greatly simplifies the formulation and analysis of the classic energymo-
mentum complexes, as demonstrated in Section V. Then, the main event in
Section VI presents a stunning simplification in the formulation and analysis of
the energymomentum complex that opens up many new possibilities. This new
canonical energymomentum complex passes an initial test for physical relevance
in Section VII.

II. Spacetime Algebra Basics

This section summarizes basic STA definitions and algebraic identities to be
used in this paper. Further results will be introduced as needed. More details
and broader discussion are given in the references [1, 2, 19, 20].
For physicists familiar with the Dirac matrix algebra, the quickest approach
to the STA is by reinterpreting the Dirac matrices a righthanded orthonormal
basis {γµ ; µ = 0, 1, 2, 3} for a 4D real Minkowski vector space V 4 with signature
specified by the rules:

γ02 = 1 and γ12 = γ22 = γ32 = −1 . (6)

Note that the scalar 1 in these equations would be replaced by the identity
matrix if the γµ were Dirac matrices. Thus, (6) is no mere shorthand for matrix
equations but a defining relation of vectors to scalars that encodes spacetime
signature in algebraic form.
The frame {γµ } generates an associative geometric algebra that is isomorphic
to the Dirac algebra. The product γµ γν of two vectors is called the geometric
product. The usual inner product of vectors is defined by

γµ · γν ≡ 12 (γµ γν + γν γµ ) = ηµ δµν , (7)

where ηµ = γµ2 is the signature indicator. The outer product

γµ ∧ γν ≡ 12 (γµ γν − γν γµ ) = −γν ∧ γµ , (8)

defines a new entity called a bivector (or 2-vector), which can be interpreted as
a directed plane segment representing the plane containing the two vectors.
STA is the real geometric algebra G4 = G(V 4 ) generated by V 4 with real
scalars. A full basis for the algebra is given by the set:
1 {γµ } {γµ ∧ γν } {γµ i} i
1 scalar 4 vectors 6 bivectors 4 trivectors 1 pseudoscalar
grade 0 grade 1 grade 2 grade 3 grade 4
where the righthanded unit pseudoscalar

i ≡ γ0 γ1 γ2 γ3 (9)

4
squares to −1, anticommutes with all odd grade elements and commutes with
even grade elements. Thus, G4 is a graded linear space of dimension 1 + 4 + 6 +
4 + 1 = 23 = 16, and any element M can be expressed as a linear combination
of the 16 basis elements.
A generic element of the STA is called a multivector. Any multivector M
can be expanded into the in the graded forms

4
M = α + a + F + bi + βi = M k , (10)
k=0

where α = M 0 and β are scalars, a = M 1 and b are vectors, and F = M 2

is a bivector, while bi = M 3 is a trivector (or pseudovector) and βi = M 4 is
a pseudoscalar. It is often convenient to drop the subscript on the scalar part
so M  = M 0 . The scalar part behaves like the “trace” in matrix algebra; for
example, M N  = N M  for arbitrary multivectors M and N . A multivector
is said to be even if the grades of its nonvanishing components are all even. The
even multivectors compose the even subalgebra of G4 , which is, of course, closed
under the geometric product.
Coordinate-free computations are facilitated by various definitions. The op-
eration of reversion reverses the order in a product of vectors, so for vectors
a, b, c it is defined by

(abc)= cba . (11)

 is
It follows for any multivector M in the graded form (10) that the reverse M
given by
 = α + a − F − bi + βi .
M (12)

Computations are also facilitated by defining various products in terms of the

fundamental geometric product.
The inner and outer products of vectors defined by (7) and (8) can be gen-
eralized to arbitrary multivectors as follows. For A = Ar and B = Bs of
grades r, s ≥ 0, inner and outer products can be defined by

A · B ≡ AB|r−s| , A ∧ B ≡ ABr+s . (13)

This enables a remarkably simple realization for the classical mathematical con-
cept of duality. The dual of any multivector is obtained by multiplying it by the
pseudoscalar i (or some scalar multiple thereof). Thus, in (10) the trivector bi
is the dual of the vector b. Inner and outer products are related by the duality
identities

a · (Bi) = (a ∧ B)i , a ∧ (Bi) = (a · B)i . (14)

Coordinate-free manipulations are facilitated by a system of identities in-

volving inner and outer products [4, 5]. These identities generalize to arbitrary

5
dimensions the well-known identities for dot and cross products in ordinary
3D vector algebra. We will need the following identities. For a vector a, the
geometric product is related to inner and outer products by
aB = a · B + a ∧ B . (15)
For vectors a, b, c the most commonly used identity is
a · (b ∧ c) = (a · b)c − (a · c)b = a · b c − a · c b , (16)
where parentheses have been dropped with the understanding that inner prod-
ucts have precedence in ambiguous expressions. The expansion (16) is a special
case of
a · (b ∧ B) = a · b B − b ∧ (a · B) (17)
for grade(B) ≥ 2, and we have a related identity
a · (b · B) = (a ∧ b) · B . (18)
Inner and outer products are very helpful for defining and manipulating
relations among vectors. For example, the entire theory of determinants can be
efficiently developed along the following lines [5]. Let {aj ; j = 1, 2, ..., m} and
{bk ; k = 1, 2, ..., m} be sets of m vectors referred to “row vectors” and “column
vectors” for convenience. The m × m determinant of the matrix {ajk ≡ aj · bk }
is defined by
det (ajk ) = det (aj · ak ) = (am ∧ ... ∧ a2 ∧ a1 ) · (b1 ∧ b2 ∧ . . . ∧ bm ) . (19)
Note how the outer product gives the antisymmetry of the determinant under-
change of any pair of rows or columns, and the inner product gives symmetry
under interchange of all rows with all columns. The standard Laplace expan-
sion for a determinant is an application of the rules (18) and (17); thus, the
expansion with respect to the first row of the determinant specified by vector
a1 is
det (ajk ) = (am ∧ . . . ∧ a2 ) · [a1 · (b1 ∧ b2 ∧ . . . ∧ bm )]
= a11 (am ∧ ... ∧ a2 ) · (b2 ∧ . . . ∧ bm ) (20)
− a12 (am ∧ . . . ∧ a2 ) · (b1 ∧ b3 ∧ . . . ∧ bm ) + a13 . . . .
Now, let {gµ } be a righthanded frame of vectors spanning V 4 , so its pseu-
doscalar is an oriented 4-volume element proportional to the unit pseudoscalar
i, that is
g0 ∧ g1 ∧ g2 ∧ g3 = g i . (21)
In terms of the metric tensor for the frame gµν ≡ gµ · gν , the magnitude g of
the volume is given by the determinant
−g 2 = (g0 ∧ g1 ∧ g2 ∧ g3 )2 = det (gµν )
(22)
= (g3 ∧ g2 ∧ g1 ∧ g0 ) · (g0 ∧ g1 ∧ g2 ∧ g3 ) .

6
This can be evaluated in terms of the gµν by the Laplace expansion (20).
Algebraic manipulations with the frame {gµ } are greatly facilitated by in-
troducing its reciprocal frame {g µ } defined by the conditions

g µ · gν = γ µ · γν = δνµ . (23)

These equations have the explicit solution

∨
g µ = (−1)µ (g0 ∧ . . . ∧ g µ ∧ . . . ∧ g3 )(g0 ∧ g1 ∧ g2 ∧ g3 )−1
∨
(24)
= (−1)µ+1 (g0 ∧ . . . ∧ g µ ∧ . . . ∧ g3 ) · (g0 ∧ g1 ∧ g2 ∧ g3 )g −2 ,
∨
where g µ indicates that gµ is omitted from the product. With the definition
g µν ≡ g µ · g ν for the reciprocal metric tensor, we can adopt the usual tensor
algebra conventions for raising and lowering tensor indices and summing over
repeated upper and lower index pairs.
Besides inner and outer products, we need the commutator product

A × B ≡ 12 (AB − BA). (25)

This is especially useful when A is a bivector. Then we have the identity

AB = A · B + A × B + A ∧ B, (26)

which should be compared with (15). It is important to note that for bivector
A the commutator product is “grade-preserving” and acts as a “derivation” on
geometric products; that is, for any multivectors M and N , we have

A × M k = A × M k (27)

and

A × (M N ) = (A × M )N + M (A × N ). (28)

A considerable advantage of STA is enabling a coordinate-free treatment of

Lorentz transformations without matrices [19]. Lorentz transformations contin-
uously connected to the identity are called Lorentz rotations. A Lorentz rotation
of any multivector M has explicit algebraic representation in the canonical form

M → M
= L M L , (29)

where L is even multivector called a rotor, which is subject to the normalization

condition

LL = 1 . (30)

It is obvious from (29) that representation of a Lorentz rotation by a rotor is

double-valued, though the ambiguity in sign is seldom an issue.

7
III. Gauge Gravity with Coordinate Frames
Our flat space model of spacetime M4 = {x} represents each spacetime point
as a vector x in the Minkowski vector space of special relativity [19]. Let x =
x(x0 , x1 , x2 , x3 ) be a parametrization of the points, in some spacetime region, by
an arbitrary set of coordinates {xµ }. Partial derivatives then give a coordinate
frame of tangent vectors to the coordinate curves:
∂x
eµ ≡ ∂µ x = . (31)
∂xµ
The inverse of x = x(x0 , x1 , x2 , x3 ) is a set of coordinate functions xµ = xµ (x)
determining a coordinate frame of gradient vectors

eµ = xµ (32)

reciprocal to the frame {eµ }, as expressed by

eµ · eν = eµ · xν = ∂µ xν = δµν . (33)

In these terms, the gradient operator  = ∂x , better known as the vector

derivative by the spacetime point x [19], can be defined by the operator equation
 = eµ ∂µ . (34)

Flat spacetime is characterized by the existence of preferred coordinate sys-

tems called inertial coordinates, for which the coordinate frame is an orthonor-
mal set of constant vectors γµ = ∂µ x, so the coordinate functions are given by
the simple functions

x = xµ γµ and xµ = x · γ µ . (35)

Inertial systems are related by Poincaré transformations of the general form

x → x
= LxL + c , (36)

where L is a constant rotor. The corresponding inertial frames are then related
by the Lorentz rotation:

γµ
= Lγµ L. (37)

Gauge theory gravity (GTG) is based on two general principles: Position

Gauge Invariance and Local Rotation Covariance, which can be regarded as
refinements of Einstein’s principles of general relativity and equivalence [?].
The first principle is satisfied by introducing gravity as a gauge tensor field
h(a) = h(a; x), which is a position-dependent linear function on vector fields,
and the dependence on x is usually left implicit. In particular, the gauge tensor
h maps the coordinate frame eµ into a corresponding gravity frame

g µ = h(eµ ) = h(xµ ) . (38)

8
Its reciprocal frame is then

gµ = h −1 (eµ ) , (39)

where the linear operator h is the transpose of h.

Either frame {g µ } or {gµ } can be regarded as representing the gravitational
field. Indeed, the inner products

gµν = gµ · gν = h −1 (eµ ) · h −1 (eν ) = eµ · [h −1 h −1 (eν )] . (40)

are precisely equivalent to the components of the metric tensor in GR. Whence
the displacement differential dx = dxµ eµ generates the familiar line element

dτ 2 = gµν dxµ dxν (41)

in the tensor formulation of GR.

For an inertial frame the pseudoscalar is determined solely by the gauge
tensor, thus

−g 0 ∧ g 1 ∧ g 2 ∧ g 3 = h(γ0 ) ∧ h(γ1 ) ∧ h(γ2 ) ∧ h(γ3 ) = det (h) i . (42)

Writing h ≡ det (h), we find from (22) that the squared volume of an arbitrary
frame is given by

det (gµν ) = det (gµ · gν ) = h−2 det (eµ · eν ) . (43)

Thus, the ratio h−1 of the volume elements for the two metric tensors is a
gravitational invariant independent of the coordinate system. This quantity
plays a prominent role in Rosen’s Bimetric Theory [16].
The second GTG principle is satisfied by introducing a “gauge covariant
derivative” or coderivative, with components Dµ defined, for operating on a
multivector field M = M (x), by

Dµ M = ∂µ M + ωµ × M , (44)

where the connexion for the coderivative

ωµ = ω(gµ ) (45)

is a bivector-valued tensor evaluated on the frame {gµ }. It follows from the

bivector properties (27) and (28) that the coderivative Dµ has the same algebraic
properties as the operator ∂µ . Note that (27) implies that

Dµ M  = ∂µ M  . (46)

In other words, Dµ = ∂µ when operating on scalar-valued functions.

To maintain form invariance of the coderivative under local Lorentz rotations
of the form (29) generated by a position dependent rotor L = L(x), that is, to
ensure the transformation

Dµ M → L(Dµ M )L = Dµ
M
= ∂µ M
+ ωµ
× M
, (47)

9
the connexion must obey the transformation law

ωµ = → ωµ
= Lωµ
L − 2(∂µ L)L . (48)

A coordinate-free vector coderivative can be defined by the operator equation

D = g µ Dµ , (49)

from which the directional coderivatives Dµ = gµ · D can be recovered.

For the gravity frame {gµ } the coderivative (44) takes the form

Dµ gν = ∂µ gν + ωµ · gν . (50)

Note that under coordinate transformations the connexion ωµ is a (covariant)

tensor, whereas Dµ gν is a pseudotensor. Explicit transformation laws induced
by a change of coordinates from xµ to x
µ follow from the chain rule for partial
differentiation:
∂xµ
∂µ = aνµ ∂ν
with aνµ = . (51)
∂x
µ
Hence, from (31) and (39) we get

gµ = aνµ gν
and ωµ = ω(aνµ gν
) = aνµ ων
. (52)

Whence, from (50) we obtain the pseudotensor transformation law

β

α
β

Dµ gν = aα
µ aν Dµ gν + aµ (∂α aν )gβ . (53)

To relate (50) to the standard covariant derivative in GR, we write

Dµ gν = Lα
µν gα , (54)

where the Lα µν = g · (Dµ gν ) are the usual coefficients of connexion. And dif-
α

ferentiating (23) we get

Dµ g α = −Lα ν
µν g . (55)

From (44) and (46) it follows that

∂µ gαβ = Dµ gαβ = (Dµ gα ) · gβ + gα · (Dµ gβ ) = Lνµα gνβ + gαν Lνµβ , (56)

which is the usual condition for “metric compatibility” of the connexion.

To make the connexion purely “metric based,” we need to impose the con-
dition for vanishing torsion:

D ∧ g µ = D ∧ Dxµ = 0 . (57)

It follows that

D ∧ g µ = g α ∧ Dα g µ = g β ∧ g α Lµαβ = 0 (58)

10
implies Lµαβ = Lµβα , or equivalently,

Dα gβ = Dβ gα . (59)

This symmetry condition suffices to solve the equations (54) to get

Lα 1 αβ
µν = 2 g (∂µ gνβ + ∂ν gµβ − ∂β gµν ) , (60)

which will be recognized as the Christoffel form for the connexion in GR. This
completes our demonstration of equivalence between GTG and GR.
To facilitate subsequent computations, we record a number of useful identi-
ties that follow from the above relations. From (50) we get

2ωµ = g ν ∧ ∂µ gν − g ν ∧ Dµ gν , (61)

which can be put in several alternative forms by differentiating g ν ∧ g α gνα = 0

to get

g ν ∧ ∂µ gν = gν ∧ ∂µ g ν = g ν ∧ g α gν · ∂µ gα , (62)

and using (57) to get

g ν ∧ Dµ gν = D ∧ gµ = (Dgνµ ) ∧ g ν . (63)

Contraction of (54) gives

g ν · (Dµ gν ) = Lνµν = 12 g αβ ∂µ gαβ = ∂µ ln g = D · gν . (64)

Whence

D · (g −1 gν ) = 0 . (65)

In a similar way we obtain

D · g µ = −g αβ Lµαβ = ∂ν (g g µν ) . (66)

Contraction of (50) gives

Dµ g µ = −Lµµµ g ν = −D ln g = ∂µ g µ + ωµ · g µ . (67)

Whence

Dµ (g g µ ) = 0 , (68)

and

g µ · ωµ = ∂µ g µ + D ln g = ∂µ (g g µ ) . (69)

Experts will recognize that the frame vectors g µ are equivalent to the usual
tetrad variables hµ(α) , as expressed by

hµ(α) = g µ ·γ(α) , (70)

11
where the lower index is placed in parenthesis to indicate that it is not (neces-
sarily) related to the coordinates. Thus, the STA formalism developed here can
be regarded as just another way — perhaps a more elegant, transparent and
efficient way — to treat GR in terms of tetrad variables. However, gauge theory
offers a deeper theoretical advantage, which is manifested by inserting (38) into
(70) to get

hµ(α) = h(eµ ) · γ(α) = eµ · h(γ(α) ) = h(γ(α) ) · xµ , (71)

This provides an explicit separation of tetrad dependence on arbitrary coordi-

nates xµ from its dependence on the gravitational gauge tensor h. It is the
critical difference between GTG and standard tetrad theory of GR.

IV. Curvature and Field Equations

The commutator of coderivatives defined by (44) gives us

[Dµ , Dν ]M = R(gµ ∧ gν ) × M , (72)

where

R(gµ ∧ gν ) = ∂µ ων − ∂ν ωµ + ωµ × ων (73)

is the curvature tensor expressed as a bivector-valued function of a bivector

variable. Contraction of the curvature tensor gives the Ricci tensor

R(gν ) = g µ · R(gµ ∧ gν ) , (74)

and a second contraction gives the scalar curvature

R = g ν · R(gν ) = (g ν ∧ g µ ) · R(gµ ∧ gν ) . (75)

Einstein’s gravitational field equation can thus be written in the form

G(gν ) ≡ R(gν ) − 12 gν R = κT (gν ) . (76)

Of course, decomposing this into tensor components Gµν = gµ · G(gν ) and Tµν =
gµ · T (gν ) yields Einstein’s equation in its standard tensor component form (1).
In this paper we are interested only in equations for the gravity field variables
gµ , so we write (72) in the form

[Dµ , Dν ]g α = ωµν · g α = Rµνβ

α
gβ , (77)

where

ωµν ≡ R(gµ ∧ gν ) = Dµ ων − Dν ωµ − ωµ × ων , (78)

and, to make contact with the standard tensor formalism, we use (54) to write
α
Rµνβ = ωµν · (gβ ∧ g α ) = ∂µ Lα
νβ − ∂ν Lµβ + Lνβ Lµλ − Lµβ Lνλ .
α λ α λ α
(79)

12
We will also make use of the curvature bivector expressed directly in terms of
the coderivative:

ωµν = − 12 gα ∧ [Dµ , Dν ]g α = − 12 g α ∧ [Dµ , Dν ]gα . (80)

At last we are prepared to formulate the central new contribution of this

paper. STA enables us to express Einstein’s tensor Gβ = G(g β ) in the new
unitary form:

Gβ = 12 (g β ∧ g µ ∧ g ν ) · ωµν = 12 (g β ∧ ωµν ) · (g µ ∧ g ν ) . (81)

The last equality is a consequence of the well-known symmetry of the curvature

tensor,

(gα ∧ gβ ) · R(gµ ∧ gν ) = (gµ ∧ gν ) · R(gα ∧ gβ ) . (82)

Using (17) and (18) to expand the inner product in (81), we get

Gβ = g µ · ωµν g νβ − 12 g β (g ν ∧ g µ ) · ωµν , (83)

which will be recognized as equivalent to the standard form (76) for the Einstein
tensor. Thus, expansion of the inner product has split the “unitary” Einstein
tensor into two parts. Let us refer to this as the Ricci split of the Einstein
tensor. We shall see that there are alternative splits that may be more physically
significant.
As is well-known, the Bianchi identity implies that the codivergence of the
Einstein tensor vanishes, that is

G̀(D̀) = 12 (g µ ∧ g ν ∧ D) · ωµν = 12 (g β ∧ g µ ∧ g ν ) · (Dβ ωµν ) = 0 , (84)

where the implication D ∧ (g µ ∧ g ν ) = 0 from (57) has been used. From the
antisymmetry of the outer products, it follows that the Bianchi identity can be
written in the instructive form

Dα ωµν + Dµ ωνα + Dν ωαµ = 0 . (85)

With the help of (68) we can write the codivergence of Einstein’s tensor in the
alternative form

Dµ (g Gµ ) = g G̀(D̀) + G[Dµ (g g µ )] = 0 . (86)

V. Energymomentum Splits in General Relativity

To facilitate comparison with standard GR, we first study the energymomentum
tensor in component form. According to (81), the components of Einstein’s
tensor are given by

Gβα = Gβ · gα = 12 (g β ∧ g µ ∧ g ν ) · (ωµν ∧ gα ) = g β · Gα , (87)

13
which shows that it is a symmetrical tensor. With the help of (83), contraction
retrieves the scalar curvature:

α = G · gα = (g ∧ g ) · ωµν = −R .
Gα α µ ν
(88)

Adopting gR as Lagrangian density, Hilbert retrieved Einstein’s tensor by a

variation with respect to the metric tensor. Einstein put this Lagrangian in an
alternative form which we can obtain by inserting (80) to get

gR = g(g ν ∧ g µ ) · [g λ ∧ Dν Dµ gλ ]
= g(g ν ∧ g µ ) · [Dν (g λ ∧ Dµ gλ ) − (Dν g λ ) ∧ (Dµ gλ )] (89)
= ∂ν W ν + L ,

where W ν ≡ g(g ν ∧ g µ ) · (g λ ∧ Dµ gλ ) and

L = g(g ν ∧ g µ ) · [−(Dν g λ ) ∧ (Dµ gλ )]

= g(g ν ∧ g µ ) · (g β ∧ gα )Lλνβ Lα
µλ (90)
= gg νβ
(Lλµβ Lµνλ − Lλνβ Lµµλ ) .

The last line in (89) follows from

g(g ν ∧ g µ ) · [Dν (g λ ∧ Dµ gλ )] = Dν W ν = ∂ν W ν (91)

by using the identity

Dν [g(g ν ∧ g µ )] = g(D ∧ g µ ) = 0 , (92)

which is an easy consequence of (57) and (68).

Einstein observed that ∂ν W ν does not contribute to variations on the bound-
ary, so he derived his energymomentum tensor by varying his “effective La-
grangian” L with respect to the metric tensor [7]. The calculation is rather
messy because it requires expressing the Lµνλ in (90) in terms of the metric
tensor. An easier and more transparent calculation is given below.
Møller got a different result by applying Einstein’s approach to his tetrad
formulation of the energymomentum complex [12]. We can obtain his result by
inserting (78) into (88) to get

gR = g(g ν ∧ g µ ) · [2Dµ ων + ων × ωµ ]
ν (93)
= ∂ν Ŵ + L̂ ,
ν
where the identity (92) is again used to identify Ŵ ≡ 2g(g ν ∧ g µ ) · ων and
Møller’s “effective Lagrangian” has the various forms

L̂ = g(g ν ∧ g µ ) · [ων × ωµ ] = g(g ν ∧ g µ ) ων ωµ 

(94)
= g[(g ν ∧ g µ ) × ων ] · ωµ .

14
 Again, rather that calculating the variation of this Lagrangian with respect
to tetrad components as Møller did, we derive the Møller split directly from the
Einstein tensor below.
We are now prepared to derive and elucidate the Einstein split (2) of the
gGβα . The derivation begins by inserting (80) into (81) to get Einstein’s tensor
in the form

Gβ = 12 (g β ∧ g µ ∧ g ν ) · [g λ ∧ Dν Dµ gλ ] . (95)

The split is then easily then immediately given by the following:

gGβα = 12 g(g β ∧ g µ ∧ g ν ) · [gα ∧ g λ ∧ Dν Dµ gλ ]

= 12 g(g β ∧ g µ ∧ g ν ) · [Dν (gα ∧ g λ ∧ Dµ gλ ) − Dν (gα ∧ g λ ) ∧ (Dµ gλ )] (96)
= κ[∂ν F Uαβν − g E tβα ] .

where F Uαβν is Freud’s superpotential [21], explicitly defined by the determinant

g β
F Uαβν = (g ∧ g µ ∧ g ν ) · (gα ∧ g λ ∧ Dλ gµ )
2κ
 ν 
 δα δαµ δαβ  (97)
g  νλ µλ βλ 
= g g g .
2κ  ν
Lµλ Lµλ Lβµλ 
µ

The last equality in (96) is a consequence of

g β
∂ν F Uαβν = Dν F Uαβν = (g ∧ g µ ∧ g ν ) · Dν (gα ∧ g λ ∧ Dµ gλ ) , (98)
2κ
which follows from the identity

Dν [g(g β ∧ g µ ∧ g ν )] = 0 . (99)

Identifying Einstein’s gravitational energymomentum density in (96) with

g β
g E tβα = (g ∧ g µ ∧ g ν ) · [Dν (gα ∧ g λ ) ∧ (Dµ gλ )] , (100)
2κ
we can put Einstein’s split (2) in the more explicit form

E Tαβ = ∂ν F Uαβν = gTαβ + g E tβα , (101)

from which the total energymomentum conservation law

∂β E Tαβ = ∂β ∂ν F Uαβν = −∂β ∂ν F Uανβ = 0 (102)

follows as an identity.
Alternative forms for Einstein’s tensor E tβα are most easily derived by using
the surprising identity

Dα [g(g β ∧ g ν )] = g(g β ∧ g ν ∧ g µ ) · (Dα gµ ) , (103)

15
which is readily verified after it has been discovered. Now equation (100) can
be expanded as follows:
2κg E tβα = g(g β ∧ g µ ∧ g ν ) · [(Dν gα ) ∧ g λ ∧ (Dµ gλ ) + gα ∧ (Dν g λ ) ∧ (Dµ gλ )]
= Dα [g(g β ∧ g µ )] · (g λ ∧ Dµ gλ ) + g[(g β ∧ g µ ∧ g ν ) · gα ] · [(Dν g λ ) ∧ (Dµ gλ )]
= Dα [g(g β ∧ g µ ) · (g λ ∧ gσ )]Lσµλ + δαβ g(g µ ∧ g ν ) · [(Dν g λ ) ∧ (Dµ gλ )] .

Identifying Einstein’s Lagrangian (90) in the last term, we can put this result
into the form
2κg E tβα = ∂α [g(g β ∧ g µ ) · (g λ ∧ gσ )]Lσµλ + δαβ L
(104)
= Lβµλ ∂α (gg µλ ) − Lµµλ ∂α (gg βλ ) + δαβ L ,
This expression for Einstein’s gravitational energymomentum tensor was first
derived by Møller [12].
Møller also derived the following compact expression for Freud’s superpo-
tential:
κ E Uαβν = 12 g −1 gασ ∂λ [g 2 (g νσ g βλ − g βσ g νλ )]
(105)
= 12 g −1 gασ ∂λ [g 2 (g β ∧ g ν ) · (g σ ∧ g λ )] .
Here, it is an immediate consequence of using identity (103) in (97).
As Møller observed, an obvious variation of this result is a superpotential of
the form
κ L U βνσ = 12 g −1 ∂λ [g 2 (g β ∧ g ν ) · (g σ ∧ g λ )]
(106)
= 12 g(g β ∧ g µ ∧ g ν ) · (g σ ∧ g λ ∧ Dλ gµ )
This provides an elegant superpotential for the Landau-Lifshitz split (5), where
the total energymomentum tensor can now be put in the form:

L T βσ = ∂ν (g L U βνσ ) = ∂ν ∂λ (ϕβνσλ ) = L T σβ , (107)

where ϕβνσλ = g 2 (g β ∧ g ν ) · (g σ ∧ g λ ) and the symmetry is obvious. However,

we shall not follow up this lead, because the alternatives appear to be more
promising.
A major consideration in Einstein’s energymomentum analysis was that his
effective Lagrangian (90) and hence his gravitational energymomentum tensor
(100) do not involve derivatives of the metric tensor of higher order than the
first. This was reinforced by the simplicity of his decomposition (89) for the
scalar curvature density. However, a few years after Einstein’s death, Møller
discovered that the alternative Lagrangian (94) has the same advantages as
Einstein’s Lagrangian and even more to recommend it, as we see next.
We derive the Møller split of Einstein’s tensor by inserting (78) into (81) to
get
Gβ = 12 (g β ∧ g µ ∧ g ν ) · [2Dµ ων − ωµ × ων ] . (108)

16
The split is then immediately given by the following:

gGβα = g(g β ∧ g µ ∧ g ν ) · [(Dµ ων + 12 ων × ωµ ) ∧ gα ] = κ[∂µ M Uαβµ − g M tβα ] ,

(109)

where M Uαβµ is dubbed Møller’s superpotential [12] and explicitly defined by

g β
M Uαβµ = (g ∧ g µ ∧ g ν ) · (ων ∧ gα ) (110)
κ
Møller’s own expression for his superpotential is considerably more complicated
than this, in part because he did not identify the bivector ων as its crucial
ingredient, and, of course, the felicities of STA were not available to him.
As in Einstein’s split, an explicit expression for the total energymomentum
tensor is derived from Møller’s superpotential; thus,
g β
∂µ M Uαβµ = Dµ M Uαβµ = (g ∧ g µ ∧ g ν ) · [Dµ (ων ∧ gα )] . (111)
κ
Møller’s gravitational energymomentum density in (109) can accordingly be
identified with
g β
β
M tα = (g ∧ g µ ∧ g ν ) · [ 12 (ωµ × ων ) ∧ gα + ων ∧ (Dµ gα )]
κ (112)
= M Uαβ + M Uσβµ (g σ · ∂µ gα ),

where
g β
M Uαβ = (g ∧ g µ ∧ g ν ) · [ 12 (ωµ × ων ) ∧ gα + ων ∧ (ωµ · gα )] . (113)
κ
This expression can be expanded into many terms in agreement with Møller’s
result [12]. However, that will not be done here, as the assorted terms are not
physically informative. The result of greatest physical interest here is the simple
expression (110) for Møller’s superpotential, as its divergence gives Møller’s total
energymomentum tensor according to (109).
The advantages of Møller’s split over Einstein’s split should now be evident.
We have already mentioned Møller’s argument that tetrads are more fundamen-
tal than metric tensors, because they are essential for introducing spinors and
quantum mechanics into GR. Though he never made use of that fact, he did
demonstrate that tetrads enable him to create a new kind of superpotential.
Møller emphasized that his superpotential (110) is a covariant tensor, whereas
Freud’s superpotential (97) is a pseudotensor. Accordingly, he was able to show
that his energymomentum tensor had certain advantages over Einstein’s. His
analysis will not be reviewed here, because GTG introduces a new element that
greatly simplifies and clarifies the theory. That is the subject of the next section.
As Møller discovered by direct calculation, Møller and Einstein splits are
equivalent for some solutions of Einstein’s equation such as Schwarzschild’s. A
general condition for such equivalence is vanishing of the first term on the right
side of (61), and (62) shows that this can be reduced to the equivalent condition
gν · ∂µ gα = gα · ∂µ gν .

17
VI. Canonical Energymomentum Split and
Inertial Frames
In the preceding section the energymomentum complex was analyzed in terms
of gravity frames {gµ } to make direct contact with the literature in general
relativity and introduce simplifications afforded by the STA formalism. With
that background we can appreciate the ultimate simplification in the following
version of an energymomentum split.
The new split is obtained by using (99) to put (108) and Einstein’s equation
in the form

gG(g β ) = κDµ U βµ − 12 g(g β ∧ g µ ∧ g ν ) · (ωµ × ων ) = κgT (g β ) , (114)

where U βµ is the canonical energymomentum superpotential defined by

κU βµ ≡ g(g β ∧ g µ ∧ g ν ) · ων , (115)

Next we use (44) to get

Dµ U βµ = ∂µ U βµ + ωµ × U βµ , (116)

and thus put (114) in the form

κ−1 gG(g β ) = ∂µ U βµ − tβ = T β , (117)

where

tβ ≡ U βµ × ωµ + 12 κ−1 g(g β ∧ g µ ∧ g ν ) · (ωµ × ων ) (118)

is identified as the energymomentum tensor for the graviational field and T β ≡

gT (g β ). Consequently, the total energymomentum tensor (strictly speaking a
“tensor density”) is defined by

T β ≡ ∂µ U βµ = tβ + T β , (119)

and we have the energymomentum conservation law:

∂β T β = ∂β (tβ + T β ) = 0 . (120)

We refer to (119) with the associated definitions (115) and (118) as the canonical
energymomentum split. It is essentially equivalent to the Møller split discussed
in the previous section. However, its formulation is now so simple and its
derivation is so straightforward as to be almost trivial. Consequently, we can
expect its physical implications to be more transparent and easier to analyze.
The most striking and perhaps the most profound feature of the canonical
split is the simple linear relation (115) between the the superpotential, which de-
termines energymomentum density, and the connexion, which determines grav-
itational force. It is a kind of duality between a vector-valued function of a

18
bivector variable κU βµ and a bivector-valued function of a vector variable ωσ .
The duality relation can be inverted by expanding the right side of (115) to get

κU µν = g(g µ ∧ g ν ∧ g σ ) · ωσ = 12 κ(g µ ∧ g ν ) · I + g(g µ ∧ g ν ) · ωσ γ σ , (121)

where

I ≡ 12 (gν ∧ gµ ) · U µν = gν (gµ · U µν ) = 2κ−1 gg σ · ωσ (122)

is a “frame-independent” vector field. Later we shall be interested in the related

frame-independent trivector field

iH ≡ gν ∧ gµ ∧ U µν = 2κ−1 gg σ ∧ ωσ = −Hi , (123)

which has been expressed as the dual of a vector field H. Solving (121) for the
connexion bivector we obtain

κ−1 gωσ = 12 (gσ ∧ gν ∧ gµ ) · U µν + 12 I ∧ gσ . (124)

This surprising dual equivalence of the connexion ωσ to the superpotential κU βµ

can be regarded as a hitherto unsuspected relation of gravitational force to
energymomentum. Accordingly, it can be expected to provide new physical
insights, which this paper begins to explore.
Physically, the most important question about the canonical split is the
uniqueness of the energymomentum tensor defined by (119). In other words,
to what degree is energymomentum localizable? We have already noted that
the connexion ωµ = ω(gµ ) is invariant under the position gauge transformations.
Hence, the superpotential and all other elements of the canonical split are gauge
invariant. This completely solves Einstein’s problem of defining a coordinate in-
dependent energymomentum tensor. However, in contrast to the Ricci split, the
canonical split is not rotation gauge covariant. According to (48), the connexion
transforms as a “pseudotensor” under local rotations. Since the superpotential
U µν is directly proportional to ωµ and the energymomentum tensor is a gradient
thereof, these quantities depend on the choice of the local rotation gauge, which
is not uniquely defined by the theory. Similar considerations led Møller [12] to
conclude that gravitational energymomentum is not localizable. However, his
conclusion overlooks a crucial point: the duality of gravitational force to local
energymomentum density. To make the point explicit, consider the geodesic
equation for a particle with invariant velocity v = h −1 (ẋ):

v · Dv = v̇ + ω(v) · v = 0. (125)

For a given gauge, the term −ω(v) · v can be regarded as the gravitational force
on the particle. This force changes with a change in gauge, but the superpoten-
tial and energymomentum tensor change in the same way. Therefore, the local
relation of force to energymomentum is preserved. In other words, with respect
to local energymomentum exchange, all gauges are equivalent. In this sense,
energymomentum is localizable whatever gauge is chosen.

19
 So far, we have not fully exploited the fact that GTG augments GR with
the explicit assumption of a flat spacetime background. As we have already
noted, that entails the existence of distinguished coordinate systems and frames
called inertial coordinates and inertial frames. In others words, for the purpose
of spacetime measurement we can set up a distinguished class of ideal observers
called inertial observers [19, 1]. In contrast to the spacetime history of an inertial
observer, the history of a real observer is “curved” by the ambient gravitational
field. However, with suitable physical measurements and theoretical calculations
the effect of the gravitational field can be ascertained, so an inertial frame can be
determined indirectly. In this respect, GTG is similar to Newtonian mechanics,
where an inertial frame cannot be defined physically without identifying the
ambient forces.
The remainder of this section is devoted to formulation and analysis of con-
servation laws from the “viewpoint of the mythical inertial observer.” It demon-
strates how inertial frames simplify and clarify physical interpretation, especially
for angular momentum.
We begin by expressing the connexion as an explicit function of the γµ :

ωµ = ω(gµ ) = ω(h −1 (γµ )) ≡ Ω(γµ ) ≡ Ωµ . (126)

Note that, according to (42), the metric density g for inertial coordinates is
entirely determined by gauge tensor and given explicitly by g −1 = h = det (h).
Next we introduce an inertial energymomentum superpotential defined by

κUβµ = κU(γ β ∧ γ µ ) ≡ (γ β ∧ γ µ ∧ γ ν ) · Ω ν , (127)

where the “inertial connexion”

Ω ν ≡ g h(Ων ) (128)

results from transforming the connexion to the “inertial system.” It follows that

U βµ = h(Uβµ ) . (129)

This can be proved by using the definition of h as the adjoint of h as follows:

Uαβµ ≡ U βµ · gα = g(g β ∧ g µ ∧ g ν ) · (Ων ∧ gα )

= g[h(γ β ∧ γ µ ∧ γ ν )] · [Ων ∧ h −1 (γα )]
(130)
= g(γ β ∧ γ µ ∧ γ ν ) · h[Ων ∧ h −1 (γα )]
= (γ β ∧ γ µ ∧ γ ν ) · Ω ν · γα = Uβµ · γα .

Hence

U βµ = U βµ · gα g α = h(Uβµ ) · γα h(γ α ) = h(Uβµ ) (131)

as advertised.

20
 In strict analogy to (119), from the inertial superpotential Uβµ we obtain an
inertial energymomentum tensor
T β ≡ ∂µ Uβµ . (132)
The relation between the canonical and inertial energymomentum tensors is
obtained by differentiating (129), which gives
∂µ U βµ = h(∂µ Uβµ ) + Uαβµ ∂µ g α . (133)
Accordingly, the canonical and inertial conservation laws are related by
∂µ T β = h[ ∂µ T β ] = 0 . (134)
Of course, the entire canonical conservation law (119) can be recast as “inertial
conservation law” T β = tβ + T β , but we will not bother with explicit definitions
for the “inertial tensors” tβ and T β as we shall not need them. Rather, we
devote the rest of this section to analysing properties of T β . Also, we continue
the practice introduced in the last two paragraphs of denoting inertial quantities
with underbars to distinguish them from their canonical counterparts, though
it is an abuse of our underbar notation for linear operators.
The energymomentum tensor T µ is not generally symmetric. Its transpose
T µ is defined by the condition γ µ · T µ = T µ · γ ν , so from (127) and (129) it
has the specific form
κT µ
=  · [γ σ · (Ω σ ∧ γ µ )] =  · [Ω µ + (γ σ · Ω σ ) ∧ γ µ )] . (135)

Thus T µ can be calculated directly from the connexion without bothering with
duality. The antisymmetry Uµν = −Uνµ of the superpotential immediately
implies the total energymomentum conservation law in the two equivalent forms
∂µ T µ = 0 ⇐⇒ ·T µ
= 0. (136)
Of course, this ignores the possibility of source singularities, which will be
treated separately.
We start by noting that the trace of the energymomentum tensor is
Tr T µ ≡ γµ · T µ = ∂ν (γµ · Uµν ) =  · I , (137)
where, in analogy with (122),
I ≡ 12 (γν ∧ γµ ) · Uµν = γν (γµ · Uµν ) = 2κ−1 γ σ · Ω σ . (138)
Anticipating a connection to gravitational mass, let us tag the vector field I
with the suggestive name inertial vector.
Further physical insight comes from considering angular momentum conser-
vation. To get an angular momentum conservation law, we define a generalized
orbital angular momentum tensor T µ ∧ x at each spacetime point x = xµ γµ .
Noting that
∂µ (T µ ∧ x) = T µ ∧ γµ = ∂ν (Uµν ∧ γµ ),

21
we identify S µ ≡ Uνµ ∧ γν as an intrinsic spin angular momentum tensor and
define the total angular momentum tensor

Mµ = ∂ν (Uµν ∧ x) = T µ ∧ x + S µ . (139)

Thus we have an angular momentum superpotential Uµν ∧x, from which angular
momentum conservation follows:

∂µ Mµ = 0 . (140)

It is noteworthy that this simple form for the angular momentum conservation
law is unique to the inertial complex, though there an equivalent, but more
complicated, gauge covariant version that we need not explicate.
Of course, spin appears in (139) because T µ is not a symmetric tensor.
Indeed

(∂ν S ν ) · γ µ = ∂ν (Uσν ∧ γσ ) · γ µ = (T σ ∧ γσ ) · γ µ = T µ − T µ
(141)

shows that spin divergence determines the skew symmetric part of T µ . When
∂ν S ν = 0, spin and orbital angular momentum are separately conserved. The
existence of such a general spin conservation law has hitherto gone unnoticed.
Of course, it is a “spin-off” of the superpotential. Indeed, the spin tensor is
completely equivalent to the superpotential, and its interpretation as angular
momentum may make it physically more perspicuous, as we see next.
To clarify the structure of the spin tensor, consider the expansion

κS µ = κUνµ ∧ γν = [(γ ν ∧ γ µ ∧ γ σ ) · Ω σ ] ∧ γν
= [γ ν (γ µ ∧ γ σ ) · Ω σ − γ µ (γ ν ∧ γ σ ) · Ω σ + γ σ (γ ν ∧ γ µ ) · Ω σ ] ∧ γν
= −γ µ ∧ (γ σ · Ω σ ) + γ σ ∧ (γ µ · Ω σ ) (142)
= γ σ · (γ µ ∧ Ω σ ) − γ µ · (γ σ ∧ Ω σ )
= Ω µ + (γ σ · Ω σ ) ∧ γ µ − γ µ · (γ σ ∧ Ω σ ).

Therefore, the spin tensor is directly related to the superpotential and the con-
nexion by

S µ = Uνµ ∧ γν = κ−1 (γ σ ∧ γ µ ) × Ω σ = κ−1 Ω µ + 12 [I ∧ γ µ + (iH) · γ µ ] , (143)

where the “frame invariant” vectors are given by

I = −γµ · S µ and iH = −γµ ∧ S µ . (144)

This equivalence of spin to superpotential suggests that gravitational energy-

momentum is fundamentally rotational. Without taking a definitive stand on
this issue, we can formally replace the superpotential by the spin tensor as the
generator of the energymomentumn complex. Thus, using (142) in (135) we
obtain

T µ
=  · [S µ − 12 (γν ∧ S ν )·γ µ ] . (145)

22
We could solve (143) for Uνµ as a function of S µ , but that is unnecessary, as we
have already replaced the superpotential by the spin tensor in the significant
physical quantities.
Now let us turn to the formulation of integral conservation laws for inertial
systems. Incorporation of integration theory with differential forms into geo-
metric calculus has been expounded elsewhere [23]. Here we use only a fragment
of the theory for non-null manifolds.
The differential on an m-dimensional submanifold of spacetime is a simple
m-vector dm x tangent to the manifold at the point x, so it can be resolved
into its magnitude | dm x | and its direction, represented by a unit m-vector field
Im = Im (x) on the manifold; thus,

dm x = Im | dm x | . (146)

For a coordinate system on the manifold, we can write

| dm x | = | det gµν |1/2 dx1 dx2 . . . dxm , (147)

a familiar expression for the “volume element” in a “multiple integral.”

Now let M µ = M (γ µ ) be any multivector-valued tensor of rank 1 on space-
time. For any m-dimensional submanifold of spacetime, the generalized Gauss’
theorem (a version of Stokes’ Theorem) can be expressed in the form
 
M̀ (∂) | d x | = M (n−1 ) | dm−1 x | .
` m
(148)

where
−1
∂ = Im Im ·  (149)

is the projection of  into the tangent space at each interior point, and at each
boundary point x the unit outward normal n = n(x) is determined by

Im = Im−1 n . (150)

This theorem applies to manifolds of any signature. The effect of signature in

the theorem is incorporated in the n−1 , which becomes n−1 = n if n2 = 1 or
n−1 = −n if n2 = −1.
The constant timelike vector γ0 determines a 1-parameter family of space-
time hyperplanes S(t) satisfying the equation

x · γ0 = t ; (151)

In other words, S(t) is a surface of simultaneous t. Let V(t) be a convex 3-

dimensional region (submanifold) in S(t) which sweeps out a 4-dimensional re-
gion M in the time interval t1 ≤ t ≤ t2 . In this interval the 2-dimensional
boundary ∂V(t) sweeps out a 3-dimensional wall W, so M is bounded by
∂M = V(t1 ) + V(t2 ) + W.

23
 ` ) = 0 over
Now, integrating the differential conservation law ∂µ T µ = T̀ (
the region M, we obtain the integral conservation law
 t2 
P (t2 ) − P (t1 ) = dt T (n) | d2 x | . (152)
t1 ∂V(t)

where the energymomentum vector P (t) = P [V(t)] for the region V(t) is defined
by

P (t) = T (γ 0 ) | d3 x | . (153)
V(t)

` we can
Introducing the superpotential with T (γ 0 ) = ∂ν U(γ 0 ∧ γ ν ) = Ù(∇),
apply Gauss’ theorem again to express the momentum as a boundary integral:

P (t) = U(n) | d2 x | , (154)
∂V(t)

where n = nγ0 = γ0 n−1 is the 3D outward normal. Note that the operator
∇ = γ 0 ∧  is a projection of the 4D vector derivative onto the 3D manifold
V(t), as required for applicability of Gauss’ theorem to a submanifold.
The integral angular momentum conservation laws are derived in exactly
the same way, simply by replacing the energymomentum tensor and its super-
potential by the angular momentum tensor and its superpotential, so the steps
need not be spelled out. The differential conservation laws for charge and other
quantities can likewise be converted to integral conservation laws by analogous
application of Gauss’ theorem.
It should be noted that the total energymomentum vector P defined by (153)
is in perfect accord with special relativity. In particular, a rigid displacement
of the whole system contained in the region M, as defined by the Poincaré
transformation (36), induces the Lorentz rotation

P → P
= LP L . (155)

This, of course, is the most general transformation that leaves the flat space
background invariant. Thus, the Poincaré group is the kinematic group of the
theory.
Recall from Section III that gauge theory gravity requires that the equations
of physics be invariant under the Position Gauge Group (PGG) and covariant
under the Local Rotation Gauge Group (LRGG). Factoring out the kinematic
Poincaré group (PG), we can identify

DGG ≡ P GG ⊗ LRGG/P G . (156)

as the dynamical gauge group of the theory.

Finally, we note that the existence of position gauge invariant conservation
laws is a consequence of the homogeneity and isotropy of the assumed flat space-
time background. Rosen [16] has emphasized that spaces of constant curvature

24
are also homogeneous and isotropic, so our analysis of conservation laws readily
generalizes to those cases. This eliminates the objection to GTG that it is does
not allow a closed topology for the universe. That is a subject to be addressed
at another time.

VII. Assessing the Energymomentum Complex

The deciding factor in any formulation of the energymomentum complex is
physical significance. To assess the inertial energymomentum complex of the
preceding section, we examine its implications for a sample of significant gauge
tensor solutions to Einstein’s equation. For any given gauge tensor, the con-
nexion Ωµ can be calculated from (61) with (62) and (63), which for an inertial
frame can be put in the form

2Ωµ = h[γ ν ∧ γ σ gν · ∂µ gσ −  ∧ (gµν γ ν )] , (157)

where g ν ∧ g α = h(γ ν ∧ γ α ) has been used. Therefrom, the superpotential U µν

and all other features of the inertial energymomentum complex can be obtained
by direct calculation.
A good test bed is the class of Kerr-Schild solutions, which Doran and
Lasenby [22] have thoroughly studied with STA techniques. This class in-
cludes the important spherically-symmetric Schwarzschild, Reissner-Nordstrom
and Viadya solutions as well as the axially-symmetric Kerr solution. Doran and
Lasenby show that the Einstein tensor itself is a suitable energymomentum ten-
sor for this class of solutions. That might be just a lucky accident, because their
choice happens to be identical to our transposed energymomentum tensor (135)
for this class. We can conclude immediately that their analysis of Kerr-Schild
field properties is fully consistent with our characterization of the inertial en-
ergymomentum complex, though it does not validate the complex conclusively.
Nevertheless, it is instructive to reconsider the Kerr-Schild energymomentum
complex from the ground up. In the following analysis, we adhere closely to the
notation in [22], which can be consulted for more details.
For an inertial frame, a Kerr-Schild gauge tensor has the general form

gν = h −1 (γν ) = γν + lν l, with l2 = 0 (158)

and lν = l · γν . Consequently, the metric tensor has the form gσν = ησν − 2lσ lν ,
and we easily calculate that

gν · ∂µ gσ = gσ · ∂µ gν . (159)

[Recall from the end of Section V that condition (159) implies equivalence of
Møller and Einstein energymomentum complexes. Therefore, our analysis of
Kerr-Schild fields cannot provide grounds to discriminate between them, though
we have seen that Møller’s complex is much to be preferred on theoretical
grounds.]

25
 Continuing our analysis, we find from (158) that

g ν = h(γ ν ) = γ ν − lν l , (160)

whence, for any k-vector K,

h(K) = K − (K · l)l . (161)

In particular, for the unit pseudoscalar we have (i · l)l = (il)l = 0, so

h(i) = i , (162)

and from (42) we conclude that g = h = det(h) = 1. Inserting these details into
(157), we obtain a specific form for the Kerr-Schild connexion:

Ωµ = h[ ∧ (lµ l)] =  ∧ (lµ l) + lµ l ∧ (l · l) . (163)

Following [22], with little loss of generality we restrict our considerations to the
case l ∧ (l · l) = 0 even inside matter, so the connexion (163) reduces to the
simple form

Ωµ =  ∧ (lµ l) . (164)

Moreover, by the same argument Ω µ = g h(Ωµ ) = Ωµ , so we can dispense with

the underbars in the previous section. Equation (164) has the obvious property
1
2 κiH ≡ γ µ ∧ Ωµ =  ∧ (l ∧ l) = 0 . (165)

Hence, from (142) the spin tensor is given by

` ∧ `l ∧ γ µ ) = ∂ [l · (l ∧ γ µ ∧ γ ν )] ,
κS µ = γ σ · (Ωσ ∧ γ µ ) = `l · ( (166)
ν

` to the left as well as to the right.

where the accent indicates differentiation by 
From (145) the energymomentum tensor is therefore given by

T µ ` ∧ `l ∧ γ µ )] = ∂ ∂ [(γ α ∧ l) · (l ∧ γ µ ∧ γ ν )] .
=  · S µ =  · [`l · ( (167)
ν α

To incorporate further details, we need to evaluate the connexion for the specific
special cases.
To describe spherically-symmetric solutions in an inertial system specified
by the timelike vector γ0 , we introduce the “spacetime split”

xγ0 = x · γ0 + x ∧ γ0 = t + rσ r . (168)

where t = x · γ0 = x · γ 0 is coordinate time,

r = |x ∧ γ0 | (169)

26
is the radial coordinate to the spacetime point x, and the radial direction is
described by

σ r = er γ0 with er = ∂r x = −er = −r . (170)

For spherically-symmetric solutions the null vector l can be rescaled to

√
l=e ϕ with e = γ0 − er = (t − r) (171)

and ϕ = ϕ(r). Inserting this into (164) and replacing γµ with a constant vector
a, we get the connexion in the more explicit form

Ω(a) =  ∧ (eϕ e · a) = (ϕ e · a) ∧ e , (172)

where the last equality is a consequence of

 ∧ e = − ∧ er =  ∧ r = 0 . (173)

Before continuing, it is worth reviewing some techniques for evaluating vector

derivatives.
From the definitions (34) and (35), we evaluate basic derivatives

x · a = a · x = a and x = 4. . (174)

Therefrom, other derivatives can be calculated. Thus

(x ∧ a) = (xa − x · a) = 3a . (175)

From er = (x ∧ γ0 )γ0 /r we calculate

e = −er = − 2 =  · e (176)
r
and

e · a = −a · er = 1 [(γ0 ∧ a)γ0 − a · er er ] = 1 (a · e er − a + a · γ0 e) . (177)

r r
We will also need
     
 σ r =  · x ∧ γ0 = 4π δ 3 (x ∧ γ0 )γ0 = γ0  er , (178)
r2 r3 r2
where δ 3 (x ∧ γ0 ) = δ 3 (x) is the usual 3D delta function for the position variable
x = x ∧ γ0 .
Returning now to the connexion (172), we use (177) to complete the differ-
entiation and obtain the general form
ϕ  ϕa
Ω(a) = − ∂r ϕ a · e er − ∧ e. (179)
r r

27
For the Schwarzschild solution ϕ = M/r, so the connexion has the specific form
M
Ω(a) = 2a · e er − a ∧ e , (180)
r2
or, equivalently,
M
Ωµ = 2γµ · e σ r + e ∧ γµ . (181)
r2
Whence we obtain the inertia vector
2M
I = 2κ−1 γ µ · Ωµ = − e. (182)
κ r2
Its divergence is
 
 · I = 2M  · er = M δ 3 (x ∧ γ0 ) . (183)
κ r2
This connection with mass is the motivation for the name given to the vector
field I.
Using (127), from (181) we obtain an explicit form for the superpotential:
2M µ
U µν = (γ ∧ γ ν ) · σ r e , (184)
κ r2
and therefrom the energymomentum tensor:
2M µ   σ r 
T µ = ∂ν U µν = γ · · e = γ µ · γ0 M δ 3 (x ∧ γ0 ) e . (185)
κ r2
Its adjoint is
2M 2M µ   σ r 
T µ
= · σ r γ µ
· e = γ ·e · . (186)
κ r2 κ r2
This is identical to the expression found by Doran and Lasenby [22]. The spin
tensor has the form
2M µ
S µ = γν ∧ U µν = (γ · σ r ) ∧ e , (187)
κ r2
and its divergence is
2M   σ r 
∂µ S µ = γν ∧ U µν = · ∧ e = M δ 3 (x ∧ γ0 )γ0 ∧ e , (188)
κ r2
which is equal to γµ ∧ T µ as required. Note that it vanishes outside the origin,
in keeping with symmetry of the gravitational energymomentum tensor.
The energymomentum integral (153) is easily evaluated from (184). For a
sphere with outward normal n = σ r we obtain
 
2M
P = U (n) | d2 x | = 2
(1 − n)γ0 r2 d2 Ω = M γ0 , (189)
∂V κ r

28
as required for a point particle with mass M at rest.
The Riessner-Nordstrom solution for the gravitational field of a point charge
at rest differs from the Schwarzschild solution only in the “scalar potential,”
which has the form
M q2
ϕ= − ≡ ϕ1 + ϕ2 . (190)
r κ r2
Consequently, the the connexion (179) is expressible as a superposition

Ω(a) = Ω1 (a) + Ω2 (a) , (191)

where Ω1 (a) is the Schwarzschild connection, which we have already considered.

With ∂r ϕ2 = −2ϕ2 /r in (179), we find that the contribution of the electric field
is given by the connection

ϕ2 −q 2
Ω2 (a) = (3a · e er − a) ∧ e = (3a · e σ r + e ∧ a) . (192)
r κ r2
Whence we find that the inertia vector vanishes:
ϕ2
I2 = 2κ−1 γ µ · Ω2 (γµ ) = [(3e · σ r + γ µ · (e ∧ γµ )] = 0 , (193)
κr
and the superpotential has the form
ϕ2
U2µν = κ−1 (γ µ ∧ γ ν ) · Ω2 (γλ )γ λ = [3(γ µ ∧ γ ν ) · σ r e + (γ µ ∧ γ ν ) · e] . (194)
κr
However, it is more convenient to use (135) to get

q2  e · a
κT 2 (a) =  · Ω2 (a) = · e∧
κ r2 (195)
q 2
e ∧ a − 3e · aσ r
= · .
κ r3
This result agrees with [22], which gives its components such as

q 2   −2σ r  q2
κT 0
2 = · = γ0 , (196)
κ r3 4πr2
and shows that it is equivalent to the standard form for the electromagnetic
energymomentum tensor:
µ qσ r
T 2 = − 12 Fγ µ F where F= (197)
4πr2
is the Coulomb field of the point charge.
Following [22], we can investigate the singularity at the origin by considering
the integral of the total energymomentum tensor T (a) = T 1 (a) + T 2 (a) over a

29
sphere centered at the origin; thus
 
2M q2 
|d x|T (a) = |d3 x| ·
3
2
e · aσ r + 2 3
e ∧ a − 3e · aσ r
κr κ r

2M q2 
= − d2 Ω er · e · aσ r + 2 e ∧ a − 3e · aσ r (198)
κ κ r
q2  q2
= M− a · γ0 γ0 + (a ∧ γ0 )γ0 .
κr 3κr
Thus the total energymomentum in the sphere is

q2 
P = |d3 x|T 0 = M − γ0 . (199)
κr
As observed in [22], this tells us that the electromagnetic contribution to the
energy is negative, and it vanishes when the integral is extended to infinity.
Therefore, inclusion of the gravitational field has removed the divergent self-
energy of the classical electromagnetic field!
The gauge tensor for Vaidya’s “shining star” solution has the form [22]
µ(t − r)
h(a) = a + a · e+ e+ where e+ = γ0 + er . (200)
r
This is nearly the same as the the Schwarzschild gauge tensor, so it leads to a
connexion of the same form as (172):
µ a · e 
Ω(a) =  ∧ e+ · a e+ = µ 
+
∧ e+ , (201)
r r
where µ = e+ ∂t µ was used to kill the derivative of µ in the last step. Conse-
quently, it differentiates to a perfect analog of the Schwarzschild form (181):
µ
Ω2 (γµ ) = 2γµ · e+ σ r + e+ ∧ γµ . (202)
r2
We can skip other details that are similar to the Schwarzschild case. However,
one significant difference is that the energymomentum tensor no longer vanishes
outside the source. Indeed, in analogy to (186) the adjoint tensor has the form
2µ 2∂t µ
T µ
= · σ r γ µ · e+ = − 2 γ µ · e+ e+ . (203)
κ r2 κr
This is identical to the result in [22], which notes that it describes a steady loss
of mass from the source of the gravitational field.
For the Kerr solution we adopt the notation of [22] to write the connexion
(164) in the form

Ω(a) = M  ∧ [α(a · n)n] = M [α(a · n) ∧ n + a · n  ∧ (α n)] , (204)

where n = (1 − n)γ0 is a null vector so n = γ0 ∧ n is a unit bivector, and

we anticipate that the source has mass M . The functional forms of α = α(x)

30
and n = n(x) are determined by Einstein’s equation, which [22] shows can be
reduced (outside the source singularity) to the equations

∇n = ∇ · n + ∇ ∧ n = 2(α − iβn) , (205)

∇2 γ = 0 with (∇γ −1 )2 = 1 , (206)

with scalar function β = β(x) incorporated into a formally “complex” funcction

γ ≡ α + iβ, and ∇ = γ0 ∧  = ∂x .
The axially-symmetric solution of Laplace’s equation (190) with the right
boundary conditions is
1
γ = [ρ2 + (z − iL)2 ]− 2 , (207)

where z ≡ x · γ 3 = x · (γ3 γ0 ) and ρ2 = x2 − z 2 with scalar constant L as angular

momentum of the source. This solution has a disk singularity in the z = 0 plane
that is thoroughly studied in [22].
For our computations we need the following definitions and derived relations:

σ γ ≡ ∇γ −1 = γ(x − Liγ3 γ0 ) , (208)

 ∧ (αn) = −∇γ = γ 2 σ γ , (209)

nσ γ = n = −σ γ n , (210)

n ∧  a · n = n(a · n γ0 − a)γ = [n ∧ γ0 a · n + a ∧ n]γ . (211)

Using these results, we can express the Kerr connexion (204) in the explicit form

Ω(a) = M γ[a · n(γσ γ + αγ0 ∧ n) + αn ∧ a] . (212)

As a check, it may be noted that for L = 0 this reduces to the Schwarzschild

connexion (180) with γ = α = r−1 and σ γ = σ r = n.
The spin tensor is given by

κS(a) = γ σ · (Ωσ ∧ a) = M [a · n(γ 2 σ γ + (αγ)γ0 ∧ n) + (iβγ)n ∧ a] , (213)

We are especially interested in the spin flux in the γ0 direction:

κS 0 = M [γ 2 σ γ + (γγ ∗ )γ0 ∧ n] = M [−∇γ + (α2 + β 2 )n] . (214)

Whence the energymomentum flux is

T 0
=  · S 0 = κ−1 M  · [−∇γ + (α2 + β 2 )n] . (215)

31
As defined in (153), the energymomentum vector for a spacelike region V = V(t)
is given by
 
P = |d x|T =
3 0
| d2 x | êu · S 0 (216)
V
 ∂V

= κ−1 M | d2 x | êu · [−∇γ + (α2 + β 2 )n] = M γ0 , (217)

the expected result. Though the integral is independent of the the region V
as long as it contains the singularity, it is most easily evaluated for an oblate
spheroid with outward unit normal êu , which reflects the symmetry of the source.
Details of the integration are given in [22].
The Inertia vector is given by

I = S µ · γµ = 2κ−1 ∂µ (lµ l) = −2κ−1 M (α2 + β 2 )n . (218)

Whence, by a similar integration, we obtain

  
| d x | Tr T =
3  −1
| d x | · I = −2κ M
3
| d2 x | (α2 + β 2 )êu · n = M ,
V V
(219)

the same as the result for the Schwarzschild case.

Since the energymomentum tensor is symmetric, the orbital and spin angular
momenta are separately conserved. They are nevertheless related by

T (a) ∧ x = [ · S(a)] ∧ x = ∂µ [(γ µ · S(a)) ∧ x] + 2S(a) . (220)

Accordingly, the orbital angular momentum for the region V is given by

 
| d3 x | T 0 ∧ x = 2 | d3 x | S 0 = M Liγ3 γ0 , (221)
V V

Again (except, perhaps, for the factor 2) this is exactly what we expect, and
details of the integration are given in [22] (though the present formulation in-
troduces some simplifications).
The present study of the energymomentum complex for Kerr geometry is
completed by the thorough analysis of the Kerr ring singularity given in [22].

VIII. Conclusions
It took almost a quarter century to identify the superpotential in Einstein’s ener-
gymomentum complex [21], and another quarter century to scope out the range
of alternative superpotentials until Møller realized the unique advantages of a
“tetrad superpotential” [12]. And to this day, Møller’s contribution is generally
overlooked. One reason for the long time lag may be obscurity and complexity
of the tensor and tetrad formulations. We saw in Section V that, with a new
unitary formulation of Einstein’s tensor, STA greatly simplifies and clarifies the

32
formulation and analysis of superpotentials. Then we saw that the canonical
split of Einstein’s tensor in Section VI brings even greater simplifications and
surprising new insights into the structure of the “energymomentum complex.”
We have seen that the energymomentum superpotential (??) is manifestly
gauge invariant, and it generates an energymomentum conservation law that is
consistent with the requirements of special relativity. Nevertheless, more work is
needed before we can be sure that this is a satisfactory solution to the longstand-
ing problem of energymomentum conservation in general relativity. Though our
application of the “inertial complex” to Kerr-Schild fields provides significant
confirmation, it is not even sufficient to distinguish the canonical complex from
Einstein’s original energymomentum complex. Further tests on solutions on
Einstein’s equation are needed, especially for gravitational radiation and the
new ideas about spin. The effect of local gauge transformations, especially on
physical interpretion, remains to be studied. The gauge can be fixed in many
ways, such as boundary conditions at sources or globally. It remains to be seen
if some gauges have special physical significance.
Finally, it should be observed that the formulation of the canonical energy-
momentum complex in Section V is sufficiently general to include torsion, so
it is immediately applicable to Dirac theory [3], where it raises new questions
about the relation of mass and spin in sources to the gravitational field.

References
[1] D. Hestenes, “Gauge Theory Gravity with Geometric Calculus,” Am. J.
Phys. ??: ??? (2005).
[2] C. Doran & A. Lasenby, Geometric Algebra for Physicists (Cambridge U
Press, Cambridge, 2003).
[3] A. Lasenby, C. Doran, & S. Gull, “Gravity, gauge theories and geometric
algebra,” Phil. Trans. R. Lond. A 356: 487–582 (1998).
[4] D. Hestenes, Space-Time Algebra, (Gordon & Breach, New York, 1966).
[5] D. Hestenes & G. Sobczyk, CLIFFORD ALGEBRA to GEOMETRIC
CALCULUS, a Unified Language for Mathematics and Physics (Kluwer
Academic, Dordrecht/Boston, 1986).
[6] H. Weyl, “Elektron and Gravitation I.” Zeitshrift für Physik, 56: 330
(1929).
[7] A. Einstein, “Hamilton’s Principle and the General Theory of Relativity,”
Sitzungsber.D. Preuss. Akad. D. Wiss., 1916. English translation in The
Principle of Relativity (Dover, 1923).
[8] J. Goldberg, “Invariant Transformations, Conservation Laws, and Energy-
Momentum.” In A. Held, General Relativity and Gravitation, Vol. I
(Plenum, New York, 1980). pp. 469–489.

33
 [9] J. Winnicour, “Angular Momentum in General Relativity.” In A. Held,
General Relativity and Gravitation, Vol. II (Plenum, New York, 1980). pp.
71–95.
[10] The early history of conservation laws in GR is surveyed by C. Cattani
& M. DeMaria, “Conservation Laws and Gravitational Waves in General
Relativity,” In J. Earman, M. Janssen & J. Norton (Eds.) The Attraction of
Gravitation: New Studies in the History of General Relativity. Birkhäuser,
Boston, 1993).
[11] L. Landau & E. Lifshitz, The Classical Theory of Fields, (Addison-Wesley,
Reading, MA 1971.
[12] C. Møller, “Momentum and Energy in General Relativity and Gravitational
Radiation,” Mat. Fys. Medd. Dan. Vid. Selsk. 34 (no. 3), 1–67 (1964). This
completes a long analysis in several articles beginning in 1958. The analysis
is summarized in: C. Moller, “Survey of Investigations on the Energy-
momentum Complex in General Relativity,” Mat. Fys. Medd. Dan. Vid.
Selsk. 35 (no. 3), 1–14 (1966); also in C. Moller, “The Four-Momentum
of an Insular System in General Relativity,” Nuclear Physics 57: 330–338
(1964).
[13] S. Gupta, “Einstein’s and Other Theories of Gravitation,” Rev. Mod. Phys.
29: 334–336 (1957).
[14] W. Thirring, “An Alternative Approach to the Theory of Gravitation,”
Annals of Physics 16: 96–117 (1961).
[15] R. Feynman, Feynman Lectures on Gravitation (Addison-Wesley, Reading,
1995).
[16] N. Rosen, “A Theory of Gravitation,” Annals of Physics 84 455–473 (1974).
This is representative of many papers by Rosen.
[17] M. Israelit, “A Gauge-Covariant Bimetric Tetrad Theory of Gravitation
and Electromagnetism,” Found. Physics 19 33–55 (1989).
[18] S. Babak & L. Grishchuk, “Energy-momentum tensor for the gravational
field,” Phys. Rev. D 61, 1–18 (1999).
[19] D. Hestenes, “Spacetime physics with geometric algebra,” Am. J. Phys. 71
691–704 (2003).
[20] Much more literature on extensions and applications of Geometric Algebra
and Calculus can be accessed from the websites
<http://modelingnts.la.asu.edu> and
<http://www.mrao.cam.ac.uk/˜clifford/>.
[21] Ph. Freud, “On the Expression of the Total Energy and the Total Momen-
tum of a Material System in the Theory of General Relativity,” Ann. Math.
Princeton 40 417–419 (1939).

34
[22] C. Doran & A. Lasenby, “Integral Equations, Kerr-Schild Fields and Grav-
itational sources.” (arXiv:gr-qc/0404081 v1 19 Apr 2004).

[23] D. Hestenes, “Differential Forms in Geometric Calculus.” In F. Brackx,

R. Delanghe, and H. Serras (eds), Clifford Algebras and their Applications
in Mathematical Physics (Kluwer Academic, Dordrecht/Boston, 1993). pp.
269–285.
[24] C. Misner, K. Thorne & J. Wheeler, Gravitation (Freeman, San Francisco,
1973).

35