0% found this document useful (0 votes)
80 views12 pages

Newton-Cartan Theory and Teleparallel Gravity: The Force of A Formulation

This document discusses how Newtonian gravity and general relativity can both be formulated in ways that alter their typical conceptualizations. Newtonian gravity is commonly viewed as a force theory, but it can be written geometrically as Newton-Cartan theory. Likewise, general relativity is usually viewed geometrically, but it can be formulated as teleparallel gravity in a way that resembles a force theory. This raises questions about theoretical underdetermination. However, the document argues that these alternative formulations do not represent worrying cases of underdetermination, as they can be interpreted as postulating the same spacetime ontology when viewed in the context of our full set of physical theories and experimental data. The geometrical nature of a gravitational theory depends on non-gravit

Uploaded by

Inês Almeida
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
80 views12 pages

Newton-Cartan Theory and Teleparallel Gravity: The Force of A Formulation

This document discusses how Newtonian gravity and general relativity can both be formulated in ways that alter their typical conceptualizations. Newtonian gravity is commonly viewed as a force theory, but it can be written geometrically as Newton-Cartan theory. Likewise, general relativity is usually viewed geometrically, but it can be formulated as teleparallel gravity in a way that resembles a force theory. This raises questions about theoretical underdetermination. However, the document argues that these alternative formulations do not represent worrying cases of underdetermination, as they can be interpreted as postulating the same spacetime ontology when viewed in the context of our full set of physical theories and experimental data. The geometrical nature of a gravitational theory depends on non-gravit

Uploaded by

Inês Almeida
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 12

Studies in History and Philosophy of Modern Physics 42 (2011) 264–275

Contents lists available at SciVerse ScienceDirect

Studies in History and Philosophy


of Modern Physics
journal homepage: www.elsevier.com/locate/shpsb

Newton–Cartan theory and teleparallel gravity: The force of a formulation


Eleanor Knox
Philosophy Department, King’s College London, The Strand, London WC2R 2LS, United Kingdom

a r t i c l e i n f o a b s t r a c t

Article history: It is well-known that Newtonian gravity, commonly held to describe a gravitational force, can be recast
Received 10 May 2010 in a form that incorporates gravity into the geometry of the theory: Newton–Cartan theory. It is less
Received in revised form well-known that general relativity, a geometrical theory of gravity, can be reformulated in such a way
20 September 2011
that it resembles a force theory of gravity; teleparallel gravity does just this. This raises questions. One
Accepted 28 September 2011
Available online 18 November 2011
of these concerns theoretical underdetermination. I argue that these theories do not, in fact, represent
cases of worrying underdetermination. On close examination, the alternative formulations are best
Keywords: interpreted as postulating the same spacetime ontology. In accepting this, we see that the ontological
Philosophy of spacetime commitments of these theories cannot be directly deduced from their mathematical form. The
General relativity
spacetime geometry involved in a gravitational theory is not a straightforward consequence of anything
Newton–Cartan theory
internal to that theory as a theory of gravity. Rather, it essentially relies on the rest of nature (the non-
Teleparallel gravity
gravitational interactions) conspiring to choose the appropriate set of inertial frames.
& 2011 Published by Elsevier Ltd.

When citing this paper, please use the full journal title Studies in History and Philosophy of Modern Physics

0. Introduction physicists have long known that general relativity’s uniqueness


does not lie in its mathematical format alone: Newtonian gravity
A popular account of the development of gravitational theories can also be written in the language of differential geometry.
might go something like this: Newtonian theory casts gravity as a Moreover, it may be reformulated in this language in such a way
force. That is, gravity causes objects to deviate from inertial that Newtonian gravity, as in GR, appears to be a manifestation
trajectories. Newtonian gravity is not a consequence of the of geometrical spacetime structure. This account is known as
geometry of space and time; forces and fields propagate in fixed Newton–Cartan theory (NCT). All this is familiar, and has been
Euclidean space. With the advent of general relativity, however, examined in depth in the literature. What is less well-known, at
we realised that gravity was best seen as a manifestation of least among philosophers, is that general relativity has been given
spacetime geometry, and the force that was gravity faded from an analogous makeover, but in the opposite direction. Teleparallel
physics. Instead, it was suggested, massive bodies move towards gravity (TPG) reproduces the empirical content of GR, but in a
each other because spacetime itself is curved by their presence. format that more closely resembles the gauge theories of the
Thus the effects of gravity are not, in fact, the effects of a force. standard model than gravity does. The surface form of teleparallel
Bodies freely falling in a gravitational field are held to be force- gravity has lead some proponents to claim that it presents a force
free. theory of gravity.1
If the account were somewhat more sophisticated, it might We thus find ourselves in a situation altogether less simple
mention that the conceptual move was prompted by a move to a than a quick flick through the textbooks might suggest. If a
very different mathematical form for the theory. While New- Newtonian universe admits of geometrical gravity, and a general
tonian gravity is written in the language of forces (or, on a more relativistic universe allows gravity as a force, then we face several
sophisticated formulation, potentials and fields), general relativity challenges to our standard conception of physics. One obvious
is written in the language of differential geometry.
For textbook purposes, this might be a reasonable, if simpli-
fied, summary. However, from the perspective of the philosopher 1
The suggestion that TPG is a force theory has been made explicitly by a
of science, the situation is far more complicated. Philosophers and group of researchers based in Sao Paulo (de Andrade & Pereira, 1997, 1999;
de Andrade, Guillen, & Pereira, 2000; Aldrovandi, Pereira, & Vu, 2004a Aldrovandi,
Pereira, & Vu, 2004b; Arcos & Pereira, 2005), but is also suggested by the common
claim elsewhere that TPG is set in ‘Weitzenböck spacetime’. The sense in which
E-mail address: [email protected] this latter claim implies that gravity is a force is explored in Section 1.

1355-2198/$ - see front matter & 2011 Published by Elsevier Ltd.


doi:10.1016/j.shpsb.2011.09.003
E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275 265

worry is that we have here physical examples of the conventionalist will argue that the mathematical accounts under consideration
thesis: it seems we must accept that the geometry of spacetime is posit the same ontology, it is natural to see them as reformula-
underdetermined by data or else accept that it is not an objective tions of a single theory.3
feature of the world. I argue here that such a conclusion is not
warranted; the full structure of our complete set of physical theories
and the data they entail is enough to choose between geometries. 1. Newton–Cartan theory
This is because the concept of an inertial frame is both more central,
and more robust, than the literature typically gives it credit for. There is no lack of philosophical literature examining Newton–
However, the existence of formulations of gravitational the- Cartan theory, and the majority of it dates back 30–50 years. Why,
ories with different geometrical structure poses a challenge to then, revisit such a well-worn topic? For one thing, the advent of
popular ideas about the source of general relativity’s (and New- teleparallel gravity makes possible some interesting comparisons
ton–Cartan theory’s) geometrical nature. In light of alternative between Newtonian theories and GR that cast the equivalence
formulations, we must conclude that a theory’s geometrical between Newton–Cartan theory and Newtonian gravity in a new
nature is not a straightforward consequence of any formal aspects light. For another, I argue that emphasising inertial structure
of the theory qua theory of gravitation. Rather, it essentially relies leads to an argument against theoretical underdetermination.
on the rest of nature conspiring to choose the appropriate set of This argument has occasionally been hinted at, but not, to my
inertial frames. That is, the non-gravitational interactions must knowledge, made explicit. I hold that, despite what the mathe-
confirm that the straightest lines, the geodesics, of our gravita- matical form of the theory seems to suggest, in an empirical
tional theory, are indeed the straightest trajectories in spacetime situation that would confirm Newton–Cartan theory, Newtonian
itself. In GR, this amounts to the other forces obeying the strong gravity should be interpreted as postulating the same set of
equivalence principle, and taking their simplest form in free-fall inertial frames as its more ‘geometrical’ relative.4
frames. While this apparent fact is encoded by the minimal coupling In this discussion it is worth being clear on exactly which form
principle, it is, in a sense, external to the barest mathematical form of Newtonian gravity is under discussion. By NG here I mean the
of the theory. later, field theoretic formulation of the theory developed by
This paper divides naturally into two parts, the first on Newton– Laplace and Poisson. Moreover, I’ll assume (as a jumping off
Cartan theory and the second on teleparallel gravity. In both the point, although I’ll question this later) that this theory is set in
Newtonian and the relativistic cases we are presented with pairs of neo-Newtonian spacetime, rather than Newton’s absolute space.
theories, one of which incorporates gravitational effects into the Although this isn’t the original Newtonian force theory, I take it in
geometrical structure of the theory via the connection, and the other this context that it is still appropriate to refer to the gravitational
of which apparently postulates a division between the geometrical force, in as much as forces remain well-defined in the field
connection and the gravitational field. In both cases, I shall argue theoretic version, even if they are not necessarily fundamental.
that there are grounds for not taking the gravity/inertia split I will begin by going over the details of NCT, making explicit
seriously. Even the theory that appears ‘non-geometrical’ can be the precise form of the theory that is under consideration. I will
interpreted as postulating the freely falling frames as inertial then examine the case for underdetermination, looking particu-
frames; there is, in a sense, an effective geometry that is not larly at whether NCT and NG should be regarded as giving the
reflected in the theory’s explicit mathematical form. In the case of same empirical results, and at how NG should be interpreted. I
Newtonian gravity, there is also a second option: it might be the will conclude that, if they are taken to be empirically equivalent,
case that non-gravitational phenomena pick out inertial frames and they postulate the same spacetime structure, and, if we insist that
thus impose a gravity/inertia split. However, in the teleparallel case, they represent different spacetime structures, then we should
there is no such option: the only candidates for inertial frames regard the two theories as empirically inequivalent. We therefore do
within the theory are those of general relativity. not have a case of ontological underdetermination. Finally, I will
Before beginning, it is worth making a few comments about draw some parallels between the case here and the situation in GR.
the relationship between theory equivalence and underdetermi-
nation. In a standard account, underdetermination worries come 1.1. The theory
about in a relatively straightforward way. It is proposed that it is
possible to have two distinct theories that make identical empiri- With the hindsight provided by general relativity, it is easy to
cal predictions. Assuming (quite reasonably), that different the- ask the following question of Newton’s gravitational theory: why,
ories entail different realist commitments, such a possibility is given that inertial and gravitational mass are equivalent in NG,
taken to undermine scientific realism. However, discussion of this just as they are in GR, should we not make the same conceptual
kind of theoretical underdetermination sometimes obscures the identification in the one as we do in the other? Why not identify
heart of the issue, because the emphasis on the existence of gravitational and inertial structure in Newtonian dynamics? The
distinct theories suggests the need for some criteria of theory answer is, of course, that, at least when considering NG in
equivalence.2 This debate is tempting in the current context; the isolation from other theories, it is perfectly possible to make this
question of whether teleparallel gravity and general relativity identification and thus cast NG in geometrical terms; the result of
constitute distinct theories, or reformulations of the same theory, this reformulation is Newton–Cartan theory.
seems to raise itself immediately. However, the worry for the
realist is that ontology might be underdetermined, and it is
3
possible to have ontological underdetermination even within a While this answers the question of theoretical equivalence in this specific
case, it’s not obvious that one can simply assert that two formulations are
single theory; for example if the theory admits of more than one
theoretically equivalent if and only if they posit identical ontology; occasionally
interpretation. I will focus here on the ontological worry; do these philosophers speak of different formulations of the same theory even when they
theories/formulations posit the same spacetime structure? Hap- prima facie possess different ontologies. Jones (1991) sees force and field
pily, in answering this question in the affirmative, I will also formulations of Newtonian gravity as examples of this.
4
answer the question about theoretical equivalence. Given that I The word ‘geometrical’ in this context, is, of course, somewhat ambiguous.
Newtonian gravity, at least when presented in its neo-Newtonian form, is a
geometrical theory in as much as it describes the geometry of spacetime, but is not
a geometrical theory of gravity, because gravity is not reduced to spacetime
2
See, for example, Glymour (1970) and Quine (1975). geometry.
266 E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275

a
To convince ourselves that it is possible to introduce a connec- For a particle with some four-velocity x , the geodesic equation
tion that will ‘geometrise’ Newtonian gravity, simply look at the has the familiar form
Newtonian equation of motion for a particle in free-fall in some
gravitational potential f xa ra xb ¼ 0: ð12Þ
2
d xj @f Other curvature constraints may be introduced by raising indices
2
þ ¼0 ðj ¼ 1; 2,3Þ ð1Þ with the spatial metric hab.5 We now add the constraint that the
dt @xj
connection is ‘curl-free’, needed to ensure that the theory pro-
We may view this path as a geodesic with affine parameter l, vides the appropriate c-1 limit for GR (see Malament, 1986)
which we may take to represent the time read by a clock moving
along the geodesic. In our framework, time is absolute, and thus l R½a½bc d ¼ 0, ð13Þ
will be a linear function of absolute time: l ¼ at þ b. This gives us
where ½. . . represents antisymmetrization. With conditions
2
d t (9)–(13) in place, we have the minimal version of NCT. This posits
¼ 0, ð2Þ
dl
2 a spacetime with a flat spatial metric and an orthogonal universal
time function which will be read by any clock traversing the
2  2
d xj @f dt spacetime. As with standard NG, this gives us a spacetime foliated
þ ¼ 0: ð3Þ
dl
2 @xj dl into 3-D Euclidean spaces coordinatised by an absolute time.
However, unlike in NG, the connection here, although spatially
By comparison recall the geodesic equation flat, possesses curvature along timelike paths. This curvature is
2 b g affected by mass distribution, and explains accelerations of bodies
d xa dx dx
2
þ Gabg ¼0 ð4Þ relative to one another in the presence of mass.
dl dl dl
This form of NCT, which Bain (2004) calls ‘‘weak’’ Newton–
this gives a connection with coefficients Cartan theory, is the c-1 limit of GR.6 However, the constraints
@f on the connection given by Eqs. (9)–(11) and (13) do not
Gj 00 ¼ : ð5Þ
sufficiently restrict the class of connections to provide either an
@xj
absolute standard of rotation or an absolute standard of accelera-
All other connection coefficients vanish. Inserting these coeffi-
tion. Given that global accelerations are unobservable in Newtonian
cients into the Riemann tensor is straightforward; curvature is
gravity, the fact that the theory does not give an absolute standard
given by
of acceleration is, if anything, an advantage. However, the failure to
@2 f provide a rotation standard is more serious, and prevents ‘‘weak’’
Rj 0k0 ¼ Rj 00k ¼ ð6Þ
@xj @xk NCT from being equivalent to Newtonian gravity, and, for that
with all other Rabgd vanishing. It is therefore clear that this connec- matter, from being a well-defined physical theory.
tion, unlike the usual affine connection in standard neo-Newtonian In order to see that weak NCT will not be empirically equivalent
spacetime, is curved wherever the gravitational field has non-zero to standard Newtonian gravity, consider the fact that, while absolute
gradient. Moreover, by contracting to the Ricci tensor, we get a linear accelerations are not observable in NG, absolute rotations are.
reformulation of the Poisson equation in geometrical terms The water in Newton’s bucket is predicted to be just as concave in
empty space as it is in our own world. Leaving aside the substan-
r2 f ¼ 4pr ð7Þ tivalist/relationist debate concerning whether this is a correct
becomes prediction of the theory,7 let us simply note that, in order for NCT
to be empirically equivalent to NG, it too must introduce an absolute
R00 ¼ 4pr, ð8Þ standard of rotation. In the weak form of the theory developed
where r is the usual mass density function. above, there are too many degrees of freedom in the connection for
The connection, Riemann and Ricci tensors above encode all it to distinguish between straight, or non-rotating, trajectories, and
the content of NG; this rather swift and easy process casts twisted, or rotating trajectories; the class of allowed connections is
Newtonian gravity in an apparently geometrical form. However, simply too large. In order to solve this problem, we may introduce
aside from an awareness that a connection whose curvature is the following constraint on the curvature of the Newton–Cartan
influenced by mass has been introduced, we know relatively little connection8:
about the theory’s geometrical structure. Let us recast the theory
Rabcd ¼ 0: ð14Þ
into the language of differential geometry.
ab
Newton–Cartan theory introduces a classical spacetime /M,h ,
t a , ra , rS, where M is a smooth four-dimensional differentiable 5
The covariant equivalent, hab does not exist in these theories, and we
manifold, hab is a Euclidean spatial metric (given by a tensor field therefore have no means of lowering indices. The index-raising procedure given
of signature (0,1,1,1)), ta is a temporal metric, with signature by the spatial metric is thus not quite the same as rasing indices using a full
spacetime metric like gab. The above procedure is best viewed a prescription for
(1,0,0,0) and ra is a derivative operator associated with the
creating tensors whose contravariant indices are purely spatial.
connection introduced above. r takes its usual significance as 6
That this is the case has been demonstrated by Malament (1986) and Ehlers
the mass density function. The temporal metric is stipulated to be (1991, 1997).
7
orthogonal to the spatial metric Barbour–Bertotti theory (Barbour & Bertotti, 1977, 1982) predicts no
concavity, and is in this sense not empirically equivalent to standard Newtonian
ab
h tb ¼ 0 ð9Þ gravity.
8
The rotation standard is not the only way to bring NCT empirically into line
and the connection is constrained to be compatible with both with NG. The symmetry group of NCT may be radically restricted by the
spatial and temporal metrics introduction of ‘island universe’ boundary conditions; that is, by the assumption
that spacetime is asymptotically flat. If we impose the same boundary conditions
rc hab ¼ ra t b ¼ 0: ð10Þ on standard Newtonian gravity, then both theories have as their symmetry group
the Galilean group; both rotations and accelerations are absolute. However, this
In more general form, the Poisson equation (8) becomes assumption is unrealistic and introduces an unobservable absolute standard of
acceleration. The variations of NCT and NG that result from this addition are
Rab ¼ 4prt a t b : ð11Þ therefore unattractive compared to the types considered here.
E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275 267

This condition was first introduced by Trautman (1965). Bain refers a gravitational field
to it as the rotation standard, and calls the theory that results from
adding it ‘‘strong’’ Newton–Cartan theory. A connection obeying this G0bac ¼ Gbac þ had rd ft b tc : ð15Þ
constraint can differentiate between ‘twisted’ and ‘non-twisted’
world lines—that is, the connection picks out some particular class A standard reading holds that, from the perspective of NG,
of reference frames as non-rotating. there is a unique correct way to effect the split described above;
This rotation standard is essential if NCT is to be a well-defined at any spacetime point, the value of the gravitational field is
physical theory. To see this, note that the restriction imposed on specified by the theory (given initial conditions). From the
the connection by the rotation standard is just that, a restriction. perspective of NCT, on the other hand, there is no preferred way
As such, the connections allowed by strong NCT are a subset of to make the split. This reflects the fact, noted earlier, that NG
those allowed by weak NCT. Now consider a situation in which holds to an absolute, global standard of linear acceleration,
the entire universe is put into a rotating state. Strong NCT, which, whereas NCT does not; acceleration is defined locally with respect
like NG, distinguishes such motion from non-accelerating motion, to inertial frames. As a result, NG has a way of specifying gravita-
will predict a divergence of the inertial paths of particles. What tional acceleration relative to the background inertial frames. On the
ab
will weak NCT predict? The connection that produces the diver- one hand, in NG, we have models /M,h ,t a , ra , f, rS, where ra is a
gence is a solution of the equations of weak NCT, but not the only flat affine connection, and f is the gravitational potential. On the
ab
one; weak NCT also allows for connections that make global rotations other hand, in NCT, we have models /M,h ,t a , ra , rS, where ra is a
unobservable. As such, weak NCT does not always provide determi- dynamical connection with curvature. The ontologies of the two
nate solutions for a given state of motion of the universe.9 Indeed, it is theories appear to differ both in terms of the nature of the spacetime
not clear that the state of motion of the universe can even be structure they posit (absolute and dynamical respectively) and in
specified, given that motion is defined relative to affine structure, and whether they posit the existence of the gravitational potential.
this affine structure is itself underdetermined. Despite being a limit of However, careful consideration of the interpretation of New-
general relativity, weak NCT is simply not a well-defined physical tonian gravity gives us reason to resist this conclusion. It is
theory, let alone an empirically equivalent competitor to NG. generally accepted that the Newtonian gravitational potential f
is a gauge quantity; the equations of motion are entirely unaf-
1.2. Underdetermination fected by the addition of a constant component to f. As a result, it
is usually asserted that the real fundamental gravitational entity
In order to determine whether NCT and NG constitute a case of in NG is the field rf. However, this entity is also subject to a
underdetermination, we must ask two questions. First, are they gauge freedom. Accelerative boosts are symmetries of the NG
empirically equivalent? Second, do they diverge in their ontolo- equations of motion, and uniform gravitational fields are unob-
gical commitments? I will argue that if the answer to the first servable; in a universe in which all bodies are subject to gravity,
question is yes, then the answer to the second is no, and vice there is no unique physically motivated way to make the gravity/
versa. As a result, no underdetermination obtains. Let us see how inertia split.11 In such a situation, both the flat affine connection,
this works. and the gravitational field, are properly thought of as gauge
First, we should note that weak NCT, as noted above, is not quantities; their exact value makes no difference to the equations
empirically equivalent to NG because it fails to make determinate of motion; only their sum is empirically significant. It has been
predictions in cases (a rotating bucket in an empty universe), in frequently noted (for example, by Friedman, 1983) that a great
which NG makes determinate predictions. Therefore, in evaluat- strength of NCT is that it replaces two gauge quantities with one
ing empirical equivalence, we must consider strong NCT. Once we that is non-gauge. Given that we accept the gauge argument for
move to the case of strong NCT, we do seem to have empirical the potential, why not also accept it for the field? On its most
equivalence, at least insofar as gravitational phenomena are perspicacious reading, NG should never have accepted that
concerned. Strong NCT and NG make identical predictions for different choices of the inertia/field split corresponded to differ-
the behaviour of massive particles under gravity. To see this, we ent possible worlds; the precise value of the gravitational field
may note that not only may we derive the NCT Poisson equation has no physical significance.12
(11), from the standard Newtonian Poisson equation (7), but, with However, it is worth pausing at this stage to consider the
the aid of the rotation standard, we may also derive (7) from implications of this suggestion. When we assert that the gravita-
(11).10 However, it is not clear what we should say about their tional potential is pure gauge and has no physical import, we
predictions for non-gravitational phenomena. Suppose we dis- assert that it is a piece of surplus structure, a mathematical
cover massless particles in a Newtonian universe. How do we artefact that does not reflect physical structure. If we take the
expect these to behave? In the absence of an interpretation of NCT same approach to the gravity/inertia split, and hence to both the
and NG, it is far from clear. If we insist that the connections in gravitational field and NG’s flat affine connection, then we assert
each theory represent inertial structure, then it seems we should the gravity/inertia division is a piece of surplus structure; only the
expect each theory to make different predictions for the trajec- sum of the two pieces represents real physical structure. How-
tories of free, massless particles; empirical equivalence is broken. ever, because the physical structure in this case is spacetime
However, I will argue in what follows that there is also another structure, some may find this conclusion less comfortable than in
path: we can refrain from insisting that the connections in both the case of the gravitational potential. Some discussion of the
theories must represent inertial structure.
Let us turn now to the reason that NCT and NG are generally 11
Some might claim that this is not strictly true. If we impose island universe
held to postulate different spacetime structures. While NCT has a boundary conditions, we can impose a standard of acceleration. However, this is a
single connection, NG divides this into an inertial connection and strong physical assumption and cannot be applied in any realistic model of
Newtonian cosmology. We could also use exactly the same conditions to impose a
standard of rest, which is not an argument that most physicists would take
9
Christian (2001) points out that weak NCT does not possess a classical seriously. Moreover, the fact that grounds can be found for imposing a particular
Lagrangian density, a Hamiltonian density or an unambiguous phase space. The gauge does not eliminate the gauge freedom of the entity in question. For a full
point above is connected to these facts, particularly to the ambiguity of the phase survey of the problems caused by insisting on a unique split between gravity and
space. inertia, see Norton (1992), Malament (1995) and Norton (1995).
10 12
For a proof of this, see Bain (2004, pp. 366–372). Malament (1995) makes the case for this.
268 E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275

circumstances under which mathematically geometrical structure connection does not encode the behaviour of non-gravitational
comes to represent spacetime structure is therefore called for. phenomena. In particular, light rays follow the geodesics of a flat
The question here concerns the claim some particular connec- connection, and the laws of electro-magnetism and the other
tion has to represent the geometry of spacetime; what makes it fundamental forces take their simplest form in reference frames
that some particular connection is the right one for some physical that are not the free-fall frames. In such a universe, NCT might give a
situation? It is well-known that, in the context of spacetime useful and concise geometrical formulation of our gravitational
theories, a connection’s link to physical structure comes about theory, but we would not associate its content with space or time.
via it’s capacity for representing the structure of inertial frames; Instead, non-gravitational phenomena would point to a division
this is why I refer to the ‘inertial connection’ above. However, between gravity and inertia. In such a case, the gravitational field
standard discussions of inertial structure blur two definitions of an ceases to be a gauge quantity; different choices of gravity/inertia
inertial frame in such a way as to preempt questions about the split correspond to different non-gravitational phenomena.
representational capacities of geometrical objects. It is common to What may we conclude from our two hypothetical universes
see inertial frames defined as those reference frames13 in which the concerning underdetermination in Newtonian theories? We seem to
components of the connection vanish, and which parallel transport have two choices. The first of these, and the one that I prefer, is to
their own coordinate axes according to the standard defined by the hold that NCT and NG need not be automatically interpreted as
relevant connection. This is all well and good, but it presupposes postulating different spacetime structure. Rather, the bare theories
that we know which connection accurately represents spacetime simply fail to return a single interpretation when taken in isolation.
structure; this mathematical characterisation of an inertial frame In order to determine whether their connections in fact represent
cannot help us to determine which connection is the correct one. spacetime structure, we must look outside Newtonian gravity, to our
Happily, there is a more physical characterisation of an inertial whole physical theory. If gravity in fact turns out to be universal, and
frame available. Inertial frames are those reference frames in which the non-gravitational laws take their simplest form with respect to
force-free bodies move with constant velocities, and which are the freely falling frames, then inertial structure is well-represented
indistinguishable according to the dynamics. This second criterion by the Newton–Cartan connection. However, this is exactly the kind
is a strong one; in order for a class of reference frames to count as of situation in Newtonian gravity in which the inertia/gravity split is
inertial, dynamical laws must take the same form in each frame. not well-defined, and should not be taken seriously. As a result, a
Moreover, inertial frames must be universal; if the theory under correct reading of Newtonian gravity, even without the insight given
consideration allows for multiple types of interaction, each of these by Newton–Cartan theory, should result in the conclusion that the
must pick out the same class of inertial frames. inertial structure of the theory is represented not by the connection
If we ask what makes some connection the right one for a alone, but by the right hand side of Eq. (15). As a result, there is no
theory, the one that accurately represents the theory’s spacetime ontological divergence, and no underdetermination. Likewise, if non-
structure, it is the second, physical, characterisation of inertial gravitational phenomena did specify a gravity/inertia split, NCT
frame that is relevant. A connection only represents spacetime would remain a correct theory of gravity, but lose its claim to a
structure if the class of frames associated with it bears the right spatiotemporal interpretation. If we accept this option, we see that
relations to the rest of the theory. If we rely on the mathematical the spatiotemporal status of the objects of a theory is not merely a
characterisation of inertial frames alone, it gives the impression matter of the theory’s mathematical form, but rather a subtle matter
that we may select a connection at random, and simply assert it to of the interplay of those objects with our total physical theory.
represent the inertial structure of the theory. In reality, inertial The second, to my mind less attractive, option is to insist that
structure is constrained by the rest of the theory’s dynamics. both theories can be interpreted in isolation from a complete account
Turning back to the question of whether to extend the gauge of physics. If we take this option, we insist that it is part and parcel of
argument to the gravity/inertia split in Newtonian gravity, we see Newtonian gravity that it postulates flat inertial structure, and that
that the question is really one of determining the correct class of Newton–Cartan theory postulates curved structure. However, if one
inertial frames, and hence the correct connection, for the theory. is genuinely committed to the idea that these connections must
However, answering this question in the context of Newtonian represent a piece of spacetime structure, then one should expect
mechanics is not straightforward, because Newtonian physics is observable manifestations of the resulting inertial structure.14 In this
not fully defined: nothing internal to the Newtonian picture tells case, we should not see the two theories as empirically equivalent:
us whether all bodies have mass or how the non-gravitational they give different predictions for the behaviour of massless bodies,
interactions will transform. Let us therefore consider two New- and different predictions for the correct form for the non-gravita-
tonian universes, and see what follows in each. tional interactions. Even if there turn out to be no massless bodies,
In the first, Universe A, gravity is universal. All phenomena pick the constraints placed by a given inertial structure on the form
out the freefall frames as the inertial frames: light rays follow of non-gravitational theories is far from trivial. Thus, even on this
geodesics of the NCT connection, and the non-gravitational laws reading, we do not have a case of underdetermination. However, it
look simplest in freely falling frames. In such a universe, it would be should be noted that the second reading might quite well commit us
natural to see the Newton–Cartan connection as representing the to the idea that Newtonian gravity is disproved by a universe in
inertial structure of spacetime; the theory would be geometrical in a which all bodies have mass, assuming that the NCT inertial structure
sense that went beyond its mathematical form (although we might is the most natural in such a case. Given that this may well have
well see that form as particularly appropriate). However, such a been exactly the universe that Newton himself envisioned, this
universe is in no way at odds with Newtonian gravity. Indeed, if makes the position decidedly odd.
Newton’s light corpuscles possessed mass, it may well be precisely
the universe most naturally posited by the original theory.
1.3. Comparing NCT and GR
In a second universe, Universe B, both NG and NCT give a full
account of all gravitational phenomena, but the Newton–Cartan
Of course, we do not live in a Newtonian universe, geometrical
gravitation or no. The above debate is primarily interesting in the
light of the kind of universe we do think we live in, namely, one
13
The formal definition of a frame of reference requires some subtlety, but in
this context it is helpful to think of a reference frame as a class of coordinate
14
systems related by rotations and translations. Earman & Friedman (1973) also make this point.
E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275 269

described by general relativity. We might therefore wish to ask Although ‘teleparallel theory’ can be used to refer to a family of
how a fully geometrical Newton–Cartan theory (in our second theories using the Weitzenböck connection, the variant of interest
kind of universe above) compares to general relativity; does the here recreates the results of general relativity exactly.15 However,
latter still provide a more geometrical account of gravity? it models gravity in a way that at first glance resembles a theory
There is certainly a sense in which GR is the more naturally like electromagnetism much more closely than GR does. This
geometrical theory, in as much as a GR universe is more simply resemblance, and the fact that the formal geometrical structure of
described in geometrical terms than a Newtonian one. Where NCT the theory is such that gravity does not lead to motion along
must postulate a time metric, a spatial metric, and a connection as geodesics of the connection, makes it tempting to think of TPG as
basic geometrical structures, general relativity is so formulated a ‘force’ theory of gravity. This claim is made explicitly by a group
that all relevant structure follows directly from the spacetime working on teleparallel gravity in Sao Paulo (de Andrade &
metric. NCT spacetime is a less coherent and cohesive entity than Pereira, 1997, 1999; de Andrade et al., 2000; Aldrovandi et al.,
that of GR. However, it is possible to distinguish between the 2004a, 2004b; Arcos & Pereira, 2005), but is also implicit in the
broad geometric content of a theory and the extent to which it common claim that teleparallel gravity involves a ‘Weitzenböck
geometrizes gravity; in terms of the extent to which each theory spacetime’.16 Doubtless this phrase is not intended, by most of its
describes gravity as a manifestation of spacetime geometry, there users, to carry a heavy interpretational burden, but it is sugges-
does not appear to be any deep difference between GR and NCT. tive; it implies that the spacetime, and hence inertial structure, of
In each case, gravitational phenomena arise as a result of the the theory is not that of general relativity, and that particles
structure of spacetime, albeit a somewhat impoverished structure moving as predicted by GR are thus accelerating. There will be
in the NCT case. more discussion of the inertial structure of teleparallel gravity in
On the other hand, if we compare general relativity to NCT in Section 1.2.
the absence of any assumptions about the non-gravitational This half of the paper will question this interpretation of TPG.
forces, there is a clear difference. General relativity comes with In fact, TPG is to GR what Newtonian gravity is to Newton–Cartan
a spatiotemporal interpretation as part of the package; the strong theory, and my suggestions here will proceed along the same
equivalence principle is generally presented as part and parcel of lines as those in the previous section. Therefore, I will argue that
the theory. But we might formulate NCT in just such a way, if we TPG and GR do not constitute a case of underdetermination; they
wished; simply exchange the Levi-Civita connection for the in fact postulate the same spacetime structure. If anything, the
Newton–Cartan connection in our minimal coupling rule. In a case for this is more clear cut, because the geometrical structure
universe where such a rule applied, one would be hard-pushed to of these theories, and its interplay with the rest of physics, is
cite a deep sense in which gravity was less a matter of spacetime better defined than it was in the Newtonian case.
geometry than it is in GR. Teleparallel gravity has not been much discussed in the philoso-
Before closing this issue, one more point deserves considera- phical literature,17 and considerably more stage setting will therefore
tion. In a 1973 paper on Newton–Cartan theory John Earman and be necessary than was the case in the Newtonian discussion. I will
Michael Friedman argue that the geometrization of gravity start by introducing the formalism of the theory and examining its
achieved by GR is both ‘more effective’ and ‘a geometrization in motivation. I will then argue that, on close examination, ontological
a very different sense’ from that of NCT. Their basis for this is that differences between the two theories are illusory. In fact, this new
pair of formulations turns out to have much in common with our
If we demand that the affine connection of a relativistic two formulations of Newtonian gravity, not least in that an emphasis
spacetime be symmetric and compatible with the spacetime on inertial structure, and not mathematical form, seems crucial.
R
metric, there is only one such connection r , and therefore, no
possibility of splitting rR into an inertial and a gravitational 2.1. The theory
1
part as there is with the Newtonian connection r (Earman &
Friedman, 1973, p. 355). Our discussion here will be concerned with a particular,
modern variant of a group of theories that have taken advantage
As a result of this, they claim, the very notion of a gravitational of the possibility of a connection with torsion. As a result, it’s
force is incoherent in GR; there is no method, even an arbitrary worth some brief mention of the history of the subject, and the
one, to divide the gravitational force from inertial effects. We contrast between the theory under discussion here, and historical
shall see in the next section that technically, the above is correct. theories of the same name. Teleparallel gravity has its roots in
If the affine connection is symmetric, the split is impossible. work simultaneously developed by Einstein and Cartan (1979),
However, if we allow a non-symmetric spacetime connection, concerning the possibility of establishing a theory similar to general
we can indeed effect a division of the Levi-Civita connection in a relativity, but with the feature of absolute parallelism—the possibi-
way analogous to standard Newtonian gravity. lity determining the angles between distant vectors. This Fernpar-
allelismus theory introduced a curvature free connection with torsion,
and an associated tetrad field in the hope that the extra degrees of
2. Teleparallel gravity freedom associated with these might be used to encode electro-
magnetism, and hence produce a unified field theory. The details of
The elegance of the reduction of gravity to geometry effected by this project, and the reasons for its abandonment, are detailed in an
general relativity (GR) is deeply seductive. But aesthetics, though excellent article by Sauer (2006).
important, are not everything in physics. Theoretical physicists have The idea of using a connection with torsion in a gravitational
long been aware that it is possible to create theories similar to GR in theory did not die with Einstein–Cartan theory. Following the
their empirical consequences, but which replace the simple model of Yang–Mills gauge theories, Poincaré gauge theory was
beauty of Einstein’s original theory with an often messier, but developed in early 1960s in the hope of providing a gauge theory
perhaps more useful, gauge theory. Of interest here is teleparallel
gravity (TPG), a particular variant of gravitational gauge theories. 15
In this paper, the abbreviation TPG always refers to this empirically
In this theory, the Levi-Civita connection of GR, which has equivalent theory, rather than any other.
curvature but no torsion, is replaced with a Weitzenböck connec- 16
See, for example, Blagojevic (2001, p. 69) or Ortin (2004, p. 11).
tion, with torsion but no curvature. 17
Lyre & Eynck (2003) is an exception.
270 E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275

of both gravitation and other fundamental forces.18 Poincaré second notion of geodesic centres around the connection: a
gauge theories allow connections with both torsion and curvature geodesic is a path that parallel transports its own tangent vector.
on the tangent bundle in order to produce gauge theories of the This notion is entirely affine; it would allow us to define geodesics
Poincaré group. Their aim is to incorporate the spin of matter in a space with no metric. In fact, it is the notion of geodesic we
fields into gravitational dynamics; torsion is associated with spin use in Newton–Cartan theory, where no spacetime metric is
much as curvature is associated with energy-momentum in GR. defined. The motivation for the symmetry condition is this: The
General relativity may be thought of as that variant of Poincaré symmetry condition ensures that the two notions of geodesic coin-
gauge theory in which the torsion is set to zero. Likewise, various cide.20 In general, even for a metric compatible connection, there
theories that go under the name ‘teleparallel gravity’ may be will be two sets of geodesics, one associated with the metric, and
thought of as variants in which curvature is set to zero.19 another associated with the connection.21 This will prove impor-
In this context, the theory under discussion here has a rather tant when looking at teleparallel gravity.
odd fit. As mentioned, TPG is a version of teleparallel theory that In addition to the notion of geodesic, the concepts of torsion
is engineered to be empirically equivalent to general relativity. As and curvature will play an important part in what follows, and are
such, spin plays no role in the theory; torsion cannot be inter- worth a closer look. As a result of its central position in general
preted as it is in Poincaré gauge theory. Moreover, the extra relativity, most of us are familiar with the ideas underlying
degrees of freedom conferred by the introduction of the tetrad curvature. Because it can be conveniently realised on two-dimen-
field are not used to represent additional forces, as in Einstein– sional surfaces, curvature appears intuitively familiar. This can
Cartan theory, but are gauge freedoms. These disanalogies mean sometimes obscure the fact that the connection between pictorial
that it will be most instructive to interpret TPG primarily in geometry and more formal geometry can be subtle, because more
isolation, rather than in comparison to these related theories. than one notion of curvature is at play. Nonetheless, two-dimen-
sional surfaces are helpful in seeing that curvature is a measure of
the failure of a vector to return to itself when parallel transported
2.1.1. Curvature, torsion and geodesics
around a closed loop. Put like this, curvature is obviously a property
Teleparallel theory tears apart several geometrical notions that
of the connection, and we can write the Riemann curvature tensor in
we have, in GR, become used to associating. It also introduces the
terms of the connection
relatively unfamiliar notion of torsion. It will therefore be useful,
r r
before looking at teleparallel theory, to go over some GR territory Rrsmn ¼ @m Grns @n Grms þ G ml Glns G nl Gl ms : ð18Þ
and reexamine some geometrical notions.
The most fundamental field in GR is the metric field, which Obviously, curvature is not the only formal property a con-
encodes spatiotemporal distances. However, much of the formal nection may possess. Another is torsion. Torsion is less easy to
apparatus of GR uses the Levi-Civita connection, which can be visualize; it corresponds, not to the failure of a single vector to
derived from the metric only if one stipulates that the connection come back to itself when transported around a loop, but to the
must be metric compatible; i.e. that the metric be covariantly failure of two vectors to form a parallelogram when parallel
constant with respect to the connection transported along one another. Taking two infinitesimal vectors
a a
in the tangent space, wa and z , first parallel transport wa along z ,
rr g mn ¼ 0: ð16Þ a
and then transport z along wa . In space with no torsion, the result
of these two processes will be the same; a parallelogram is
This ensures that the angle between two vectors remains the
formed. However, if the connection has torsion, the parallelogram
same under parallel transport. However, this is not enough to
will not close, as shown in Fig. 1. The non-closure is proportional
specify the connection uniquely. In addition, it is necessary to
to torsion.
specify that the connection must be symmetric in its lower
A more formal definition of torsion can be given in terms of in
indices
terms of the Lie bracket of two vector fields and the covariant
Grmn ¼ Grnm : ð17Þ derivative of one vector field along another. If X and Y are vector
fields, torsion is given by
It is this second condition that will be replaced in teleparallel
theory. However, before moving on to the possibility of dropping TðX,YÞ ¼ rX YrY X½X,Y: ð19Þ
the condition, it is worthwhile to look at its original motivation. In co-ordinate notation, the torsion tensor is given by
After all, on the surface, the symmetry condition appears to have
nothing to do with the metric. T l mn ¼ Gl mn Gl nm : ð20Þ
In order to see why the symmetry condition is introduced, we As such, torsion is a measure of the antisymmetric part of a
may turn to the concept of a geodesic. Clearly, this idea is connection. Thus, general relativity’s demand for a symmetric
fundamental to general relativity and its geometrical nature; connection amounts to a demand that the torsion tensor vanish.
the idea that freely falling particles follow geodesics of the metric Teleparallel gravity, on the other hand, postulates a metric compa-
is an essential component of GR. However, it is important to bear tible connection with zero curvature, but non-zero torsion.
in mind that there are two notions of geodesic at play in The above account makes curvature and torsion, and the
differential geometry. The first is a metrical notion: geodesics connections postulated by GR and TPG, sound pleasantly symme-
are paths of extremal length; they represent local maxima and trical. However, there are some important asymmetries. As was
minima of the interval, ds. There is no need to have defined a
connection in order to introduce this notion of geodesic. The
20
The symmetry condition is sufficient, but not necessary, for the two classes
of geodesic to coincide; the geodesics associated with a given connection depend
18
Poincaré gauge theory was first proposed by Kibble (1961) and Sciama only on its symmetric part. Where torsion is not completely antisymmetric, its
(1962). For a good textbook overview of Poincaré gauge theories, see Blagojevic symmetric part will contribute to geodesic structure and cause deviation from
(2001, Chap. 3). For some discussions that focus on the role of torsion in these metric geodesics. However, the geodesics of a connection with torsion will
theories, see Hehl (1984) or Hammond (2002). coincide with metric geodesics if the torsion is completely antisymmetric.
19 21
It is natural to think of these as ‘limiting cases’ of Poincaré gauge theories, In the literature where torsion is considered, the straight lines associated
but it is worth noting that, unlike some other ‘limiting cases’ of theories, GR and with the connection are often known as autoparallels, while the term geodesic is
TPG are not derived by considering the approach to a particular limit, but rather reserved for the extremal paths determined by the metric. However, to retain
by simply setting curvature/torsion to zero. continuity with more familiar GR literature, I’ll refer to both as geodesics here.
E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275 271

absolute parallelism

ζ rn ha m ¼ @n ha m Gy mn ha y ¼ 0: ð23Þ

χ Teleparallel gravity is an Abelian gauge theory of the transla-


tion group. The gauge potential Ba m is the non-trivial part of the
tetrad field25
χ
a
h m ¼ @m xa þBa m : ð24Þ

ζ
 The field strength may be derived from the potential in the usual
way, and turns out to be simply the torsion of the connection
Fig. 1. Torsion. written in the tetrad basis
a
F a mn ¼ @m Ba n @n Ba m ¼ h r T rmn : ð25Þ
the case for geodesics, there is more than one concept of curvature. The action is given by the following integral:
This comes about because curvature may be defined without making Z b pffiffiffiffiffiffiffiffiffi
use of the connection. Curvature may be defined as the relative S¼ ½m u2 þ mBa m ua um  ds, ð26Þ
acceleration of neighbouring geodesics. If we use the metrical notion a
a b
of geodesic, this does not rely on the connection. Torsion, on the where ds ¼ ðZab dx dx Þ1=2 is the Minkowski tangent space invar-
other hand, has no metrical interpretation; it is purely a property of iant interval, and ua is the particle four-velocity. Use of the Euler–
the connection. Lagrange equations leads to a force equation analogous to the
Another important asymmetry concerns the uniqueness of the Lorentz force law
connection. While there is a unique symmetric metric-compatible
dua
connection for a given metric, demanding metric compatibility and ð@m uxa þ Ba mu Þ ¼ F a mr ua ur : ð27Þ
ds
zero curvature fails to uniquely specify a connection. Teleparallel
gravity therefore involves an equivalence class of connections. This can be reexpressed via Eqs. (24) and (25) to give
dum
Gymn uy un ¼ T ymn uy un , ð28Þ
ds
2.1.2. Summary of the formalism
Armed with the above geometrical concepts, we are ready to which is the equation of motion for teleparallel gravity. This
examine teleparallel theory. As noted, TPG introduces a connec- predicts deviation away from the geodesics of the Weitzenböck
tion with torsion, but zero curvature.22 This is called the Weit- connection, and these deviations depend on torsion, and thus on
zenböck connection. Of course, it is hardly surprising that such a the field strength of the theory.
connection can be defined. What is far more surprising is that the
introduction of the Weitzenböck connection can lead to a theory 2.1.3. Motivation
of gravity that reproduces the results of general relativity. In order It is worth pausing here to ask why one might prefer the
to understand how this comes about, it is necessary to move into messier teleparallel theory to the more elegant general relativistic
tetrad notation, which requires further explanation.23 formulation. TPG aims to reproduce the results of GR exactly, and
In standard GR, we generally work within the constraints of a thus manifestly fails to have any empirical benefits. The perceived
coordinate basis for our tangent space. However, in curved spaces, benefits then, must be theoretical, and I shall try to outline some
these coordinates won’t generally be orthogonal. Nonetheless, possibilities here.
locally, we can always assign an orthonormal basis for the tangent A first possibility concerns the motivation for all gauge theories
space. A tetrad field is a set of four vector fields which at each point of gravity, including that first suggested by Einstein and Cartan:
in the space provide an orthonormal basis for the tangent space at unification. The overwhelming success of the gauge heuristic gives
that point. Writing the components of a tetrad field in a coordinate good reason to believe that the way forwards in the unification
basis gives us a means of transforming between orthonormal and programme lies in a fully fledged gauge theory of gravity. However,
coordinate bases: if V is a vector in the tangent space, and roman unlike Einstein’s version of a teleparallel theory, which was intended
and greek indices represent components of vectors in orthonormal to unify gravity and electromagnetism, modern teleparallel theory
a simply reproduces the results of general relativity, and does not
and coordinate bases respectively, then the tetrad hm gives us
a
pretend to posit any new results connecting gravity to other forces.
V a ¼ hm V m : ð21Þ In this case, then, the unificationist motivation boils down to a
Given a non-trivial tetrad field, the Weitzenböck connection24 vague conviction that nature has written all her laws in the language
is given by of gauge theories; it is a methodological unification and not an
ontological one.
Grmn ¼ ha r @n ha m : ð22Þ A second possible motivation concerns the energy of the
gravitational field. In standard GR, this is represented by the
This means that the Weitzenböck covariant derivative of the energy–momentum pseudo-tensor of the gravitational field: t l .
r
tetrad field vanishes identically, which is the feature that ensures In teleparallel gravity, it is now more naturally represented by the
r
gauge current, ja , which is given in Eq. (36). de Andrade et al.
22
This connection is, in fact, metric compatible. However, the metric is not (2000) consider this a major advantage of the theory. They make
explicitly introduced in the TPG formalism. much of the fact that this quantity is invariant under local
23
The account below will be, of necessity, only a sketch. For a fuller account of translations of the tangent space coordinates, and transforms
the precise theory described here, look at e.g. de Andrade et al. (2000). For an
covariantly under global tangent space Lorentz transformations.
introduction to the tetrad formalism, see Carroll (2003, Chap. 10) or Wald (1984,
pp. 50–53). Jensen (2005) provides a helpful introduction to differential geometry This means it is both a well-behaved gauge invariant quantity of
with torsion. Blagojevic (2001, Chap. 3) gives an introductory overview of Poincaré the theory (remembering that the gauge group of teleparallel
gauge theories, including teleparallel theories.
24
Henceforth the Weitzenböck connection will be represented by G, and the
25
Levi-Civita connection by G . I will also work in units where c ¼1. Note that this division depends on our choice of coordinates.
˚
272 E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275

theory is the translation group), and what they call a ‘‘spacetime relationship between the Newton–Cartan and Newtonian connec-
r
tensor’’.26 This contrasts with the pseudo-tensorial nature of t l .27 tions, is obvious. In fact, the above seems to represent a gravity/
Moreover, although both the gauge current and the energy– inertia split in much the same way that Eq. (15) does. This
momentum pseudo-tensor are conserved quantities, this fact relation is also essential in establishing the empirical equivalence
can be expressed with a covariant derivative for the gauge of TPG and GR, because it acts as a translation dictionary between
current, but cannot be expressed covariantly for the pseudo- the equations off teleparallel gravity and those of standard general
tensor. This is perhaps an advantage, but hardly reason enough to relativity. Given this translation, the Lagrangian above turns out to
abandon GR in favour of teleparallel theory, especially when we be identical, up to a divergence, to the Einstein–Hilbert Lagrangian
note that ja r is only covariant under global, and not local, Lorentz in standard GR.29
transformations.
A final motivation for the theory is a claim, put forward in h pffiffiffiffiffiffiffi ˚
L¼ g R : ð33Þ
two recent papers by Aldrovandi et al. (2004a, 2004b) that 16pG
teleparallel gravity may be formulated in such a way that it The vacuum field equation can likewise either be translated
remains valid in the face of violation of the weak equivalence into the language of teleparallel gravity or alternatively obtained
principle (WEP). That is, teleparallel theory could accommodate by performing variations with respect to the gauge potential. It
the discovery that, for some matter, gravitational and inertial turns out to be
mass are not identical, and that all bodies do not therefore fall at
sr r
the same rate regardless of their constitution. Of course, we have @s ðhSa Þ4pGðhja Þ ¼ 0, ð34Þ
no current reason to think that WEP is violated, and one might r
where ja is the gravitational gauge current 30
well think that, if TPG loses GR’s ability to explain the validity of
the WEP, this is a disadvantage of the theory. We will return to r @L c4 l m r
hja  a ¼ hh S nr T nl þ ha L: ð36Þ
this issue, and its consequences, in the final section of this paper. @h r 4pG a m

2.2. Underdetermination This quantity is conserved as a result of the field equations

Dr jar ¼ 0: ð37Þ
Teleparallel gravity and general relativity appear to be candi-
dates for underdetermination. Is this really the case? As in the The equivalence of the Lagrangians is enough to establish
Newtonian discussion, this rests on whether they are really empirical equivalence, but we may, at this stage, begin to have some
empirically equivalent, and whether they propose different ontol- doubts about TPG’s status as an original and coherent theory. In the
ogies. However, the issues here are slightly more complicated above equation, Dr is the so-called teleparallel covariant derivative,
than in the Newtonian case, and, while I’ll argue that we don’t in not the Weitzenböck covariant derivative. The teleparallel covariant
fact have a case of underdetermination, I will also argue that one derivative is simply the Levi-Civita covariant derivative reexpressed
of the options available in the Newtonian case, insisting that the in terms of the Weitzenböck connection via Eq. (32)
two theories represented different inertial structures and pro- r m
Dr jar  @r jar þ ðG lr K lr Þja l : ð38Þ
vided different predictions, is not available here.
First, we need to understand the results that lead to the claim Thus the gravitational gauge current is not conserved with
that TPG is empirically indistinguishable from GR. The Lagrangian respect to the connection most natural to the theory, but rather
for a pure gravitational field may be written in the following with respect to the Levi-Civita connection. The importance of the
notation28: teleparallel covariant derivative does not stop there. The minimal
h coupling principle applies in TPG just as it does in GR. In order to
LG ¼ Srmn T rmn , ð29Þ
16pG maintain empirical equivalence, however, ordinary derivatives
a are converted not into the Weitzenböck covariant derivative, but
where h ¼ detðh m Þ, and
into the teleparallel covariant derivative.31 We shall see shortly
1 that this has important consequences for the inertial frames
Srmn ¼ ½K mnr g rn T sm s þ g rm T sn s  ð30Þ
2 picked out by TPG.
with K mnr being the contortion tensor So far we have not mentioned the metric. In teleparallel
1 gravity, this is quite possible; it does not appear in the formalism
K rmn ¼ ½T r þ T n rm T mrn : ð31Þ of the theory. Nonetheless, it is worth noticing that it has been
2 mn
hiding in the shadows all along, closely tied to the tetrad field
How does this lead to empirical equivalence? It turns out that
a a
the contortion tensor defined above is simply the difference g mn ¼ Zab h m h n , ð39Þ
between the Weitzenböck connection and the Levi Civita
connection
29
Although this certainly guarantees the empirical equivalence of the two
˚r theories locally, we might have some global worries. Certain topologies do not
Gmn ¼ Grmn K rmn : ð32Þ admit of a global tetrad field. Nonetheless, the Einstein equations as standardly
written are solvable for these manifolds. As a result, it appears that there must be
Given that K rmn is built up from the torsion, which is the field some GR solutions that cannot be instantiated in TPG. Of course, this also applies
strength of the theory, the parallel with Eq. (15), which gave the to any formulation of GR using tetrads, not just TPG. Furthermore, Bob Geroch
(1970) has argued that it is reasonable to restrict the GR space of solutions to
those that admit of spinor fields. Many thanks to an anonymous referee for
26
This term is just another way of expressing the covariance under global bringing this paper to my attention.
Lorentz transformations. 30
ja r may be compared to the standard GR energy–momentum pseudo-
27
This quantity does not transform like a tensor, and vanishes in freely falling tensor of the gravitational field, which can be expressed in our notation as:
frames.
28
I use the vacuum Lagrangian and field equations here for ease of exposition. r c4 h m
htl ¼ G S nr þ drl LG : ð35Þ
The presence of a source field adds an extra term to the Lagrangian, and results in 16pG nl m
the presence of the energy–momentum tensor on the right-hand side of the field
31
equations. See de Andrade & Pereira (1999).
E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275 273

where Zab is the Minkowski metric. In fact, g mn is still used to raise such frames exist for a non-symmetric connection; the compo-
and lower indices, just as it is in GR. nents of the Weitzenböck connection cannot be made to vanish
One might therefore have doubts that teleparallel gravity by an appropriate choice of reference frame.32 It seems that the
really postulates a different ontology; the old entities from GR geodesics of the Weitzenböck connection simply don’t have a full
appear to be waiting in the wings. I suggested in the Newtonian inertial structure associated with them; although we can define a
case that, under circumstances in which non-gravitational phe- class of frames that parallel transport their own axes relative to
nomena failed to determine the gravity/inertia split, we should the Weitzenböck connection, these frames can’t be the ones in
regard Newtonian mechanics as postulating the same spacetime which the laws of physics take a uniform form by virtue of the
structure as Newton–Cartan theory. Eq. (33), like Eq. (15), postulates connection vanishing.
a split between gravity and inertia that constitutes the purported Is TPG therefore a theory without inertial structure, and
ontological difference between TPG and GR. Might we have reason therefore without full spacetime structure? Perhaps not. As noted
here, as in the Newtonian case, not to take this split seriously? Might in Section 1.1, the minimal coupling prescription in TPG uses the
TPG not really be a force theory? teleparallel covariant derivative (which is simply a recasting of
Let us briefly review some reasons one might view TPG as a the Levi-Civita covariant derivative), and not the Weitzenböck
force theory, and thus as ontologically divergent from GR. A covariant derivative. As a result, all other fundamental forces take
reasonable place to begin is with the common claim that TPG their simplest form relative to the inertial frames picked out by
involves a Weitzenböck spacetime. Unlike the straightforward the Levi-Civita connection. Moreover, freely falling bodies still
mathematical fact that the theory involves a Weitzenböck con- follow geodesics of the metric, which, along with the Levi-Civita
nection, this claim carries some metaphysical weight; it makes a connection, is still present in the theory, albeit in disguised form.
statement about the nature of the spacetime in which the theory It therefore seems most sensible to conclude that the spacetime
is set. It implies that the Weitzenböck connection is a ‘spacetime’ of TPG is in fact precisely the same as the one in GR; only the
connection, in contrast to, say, the connections of electromagnetic mathematical form is different.
fibre bundles. In Newtonian gravity, special relativity, and general This conclusion becomes still more compelling when we note
relativity, spacetime connections are those that determine geo- that neither the Weitzenböck connection nor the tetrad field are
desics and the structure of inertial frames. Indeed, this seems uniquely determined by the theory. The tetrad field is defined only
constitutive of what it is to be a spacetime connection. So the up to a local gauge transformation,33 and this gauge freedom passes
claim that TPG involves a Weitzenböck spacetime implies that the on to the connection defined in terms of the tetrads. As a result, the
Weitzenböck connection represents the inertial structure of the gravity/inertia split expressed in Eq. (32) is just as much a gauge
theory. This in turn implies that particles moving in a gravita- matter as it was in the Newtonian case. In fact, the situation is
tional field are subject to accelerations; the geodesics of the worse. Because the tetrad field is subject to a local gauge freedom,
Weitzenböck connection differ from those of the Levi-Civita no amount of information about the tetrad field on, say, a spacelike
connection, and, in a theory empirically equivalent to GR, parti- hypersurface will determine the value of the tetrad field in other
cles moving under gravity alone follow Levi-Civita geodesics. This regions of spacetime. As a result, it is tempting to think that the
grounds the claim that gravity is a force in TPG: if the Weit- metric, and the Levi-Civita connection, should be taken as ontolo-
zenböck connection represents geodesic and inertial structure, gically prior to the tetrad field and Weitzenböck connection.
then gravity causes non-inertial motion. Both TPG and GR appear to take the metric to be fundamental.
This view appears to be supported by the form of the Looked at another way, we can note that both theories admit of
equations listed above. Eq. (27) resembles the Lorentz force law, both the tetrad field and the metric; tetrad formulations of GR
while Eq. (28) appears to be an equation of motion involving a have various uses. Moreover, rods and clocks survey the self-same
force. Some further support might be garnered from the gauge metric in both theories. The only difference is the way in which
form of the theory; Poincaré gauge theories are modelled on this comes about—either via the Weitzenböck connection or the
Yang–Mills gauge theories, which describe forces. Levi-Civita connection. It seems that both theories posit the ‘same’
However, when we dig a little deeper, there is reason to spacetime; if the connections in the two theories are thought of as
question this force interpretation of TPG. The analogy with modelling properties of this spacetime, they should perhaps be seen
Yang–Mills gauge theories is somewhat misleading; TPG proposes as alternative representations of the same properties. The closer we
a Minkowskian frame bundle on the base space R4 . The gauge look, the less there appears to be any underdetermination at all.
group of the bundle is the translation group, and the gauge However, this does present us with a puzzle. Torsion and
potential is the non-trivial part of the tetrad field. This bundle curvature represent very different geometrical properties. How
has markedly different features from the fibre bundles of Yang- then, can TPG and GR be interpreted identically? This query
Mills theories. For one this bundle possesses a soldering form becomes more pressing when we note that in Poincaré gauge
relating the fibre spaces to the base manifold. For another, the theories, the torsion and curvature represent translational and
algebra of constraints on the TPG bundle fails to form a Lie Lorentz gauge field strengths respectively. These then couple to
algebra. One should therefore be wary of simplistic comparisons different fields, and have markedly different effects. While the
with standard model force theories. curvature of the connection plays its traditional role in General
But the problems for the force interpretation of TPG really Relativity; coupling to all matter fields, torsion, at least at the
arise when we consider the inertial structure of the theory in macroscopic level, couples only to fields with spin. Prima facie, then,
more detail. It is essential to ask relative to which inertial frames
TPG represents gravity as a force. The answer might seem simple:
relative to those frames defined with respect to the Weitzenböck 32
The standard way of seeing this is to note that the antisymmetric part of
connection. But which frames are these? In general relativity, the the connection forms a tensor, and therefore its components cannot be made to
inertial frames are those with respect to which the Levi-Civita vanish by a coordinate transformation. In fact, this story turns out to be something
connection coefficients vanish. In these frames, the laws of of a simplification; it is now recognised that one can define ‘normal frames’ for a
physics are returned to their special relativistic form because of non-symmetric connection, but these frames don’t correspond to coordinates in
the right way to represent extended inertial structure. See Hartley (1995) and Iliev
minimal coupling: non-gravitational interactions couple to grav- (2006) for details.
ity only via the connection and when the connection coefficients 33
Note that this gauge freedom exists in addition to the translational gauge
vanish, local gravitational effects vanish as well. However, no freedom that standardly defines the gauge group of the theory.
274 E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275

we would expect very different theories to be produced by postulat- geodesic equation of GR. However, the motions of particles are
ing connections with only torsion and curvature respectively. still well-defined even for mg ami .
However, perhaps the analogy with Poincaré gauge theory is What are we to make of this? First we should note that what is
misleading. The equations of TPG are set up in such a way that the being proposed is quite different from my suggestion for Universe B,
torsion of the connection plays a very different role from that in and much stronger. I proposed violation of a Newtonian version of
Poincaré gauge theories. In particular, despite being broadly the strong equivalence principle; massless phenomena were to
associated with spin in the non-teleparallel literature,34 torsion select the geodesics of a flat spacetime. However, in this case, we
has no such association in teleparallel theory; given the equa- have violation of the weak equivalence principle. It is not exactly
tions, the effects of the Weitzenböck connection on matter must clear what massless bodies do according to the above prescription,
in every way match the effects of the Levi-Civita connection. That but it seems plausible to hold that these bodies, at least, still follow
identical physical effects can be modelled either by a connection geodesics of the Levi-Civita connection. It is the behaviour of massive
with curvature or one with torsion is truly remarkable, but not bodies that changes. However, the change does not help us interpret
because there is any necessary link between torsion and parti- TPG as a force theory on flat spacetime. While WEP violation
cular particles or effects. The fact that the role of torsion in similar certainly threatens the claim of the Levi-Civita connection to
theories can be so different teaches us that interpreting a theory represent inertial structure, it does nothing to award the Weit-
involves more than just looking at its mathematical form. In this zenböck connection inertial significance. A theory like the above
case, we can only understand the theory by thinking hard about appears simply to be a theory without inertial structure at all, and
the inertial structure it posits. Despite appearing to possess very hence without any proper notion of force. It may be that such
different mathematically geometrical meanings, given the right laws, notions must go by the wayside in some eventual theory, but it
two theories involving connections with torsion and curvature can, scarcely seems that TPG is doing any work in showing us this. Given
under very special circumstances, have the same physical geometry. our ability to translate teleparallel quantities to GR ones, we can
always back-engineer the above equations into GR form.
Thus, when it comes to postulating the a flat spacetime, the
2.3. Alternative inertial structure?
analogy between the general relativistic case and the Newtonian
one breaks down. We don’t have underdetermination in the TPG/
The above discussion closes with a stronger conclusion than in
GR case, but the sheer complexity of the geometrical form of GR,
the Newtonian case. What has happened to our two universes and
and the subtlety of the notion of inertial frame within it, restrict
our second interpretative option? If Eq. (32) represents a gravity/
our options. The only coherent spacetime to be found in these
inertia split, can’t we imagine a situation in which non-gravita-
theories is the curved spacetime of GR.
tional phenomena suggested that teleparallel gravity represented
the more natural spacetime theory? Is there no GR analog of the
Universe B mentioned in Section 1.2? 3. Conclusions
The difference between Universe A and Universe B lays in the
satisfaction of the equivalence principle. Recall that in GR the strong Two morals may be drawn from this discussion. The first is
equivalence principle has two components: minimal coupling and that examples of theoretical underdetermination are harder to
universality of gravitation. Minimal coupling is maintained in TPG come by than one might think, especially if one takes a relatively
with respect to the Levi-Civita connection, so there is no reason to liberal attitude to the metaphysical commitments of a theory.
think that TPG might correspond to a universe in which minimal Genuine potential examples of underdetermination are scarce; I
coupling is incorrect. In particular, we can’t even consider modifying have examined two rare examples and found them lacking.
the theory to a form with minimal coupling with respect to the Significantly, the illusion of underdetermination came about as
Weitzenböck connection, because there are no inertial frames in a result of what John Stachel has called ‘‘the fetishism of
which this connection vanishes, and therefore applying the comma mathematics’’.35 It is only when we take the mathematical form
goes to semi-colon rule with respect to the Weitzenböck connection of the theories too seriously that they appear to diverge. Careful
would not give back the pre-GR forms of non-gravitational interac- consideration of inertial structure revealed that geometrical form
tions in any frame whatsoever. However, there has been a suggestion does not always determine a theory’s spatiotemporal commit-
that teleparallel gravity might be compatible with a universe in ments. This is the second moral. We see that geometrical form is
which gravity is not universal, namely one in which gravitational and not a sufficient condition for representing spacetime structure.
inertial mass were not equivalent. Inertial considerations play an important role in the process by
Aldrovandi et al. (2004a, 2004b), derive an equation analogous which mathematical structure comes by its spatiotemporal cre-
to the Lorentz force equation which can accommodate different dentials. Moreover, the requirements on our total theory for a
values for gravitational charge mg, and inertial mass mi. They particular inertial structure to be represented are stringent.
reproduce the action for TPG, but without assuming the weak The consequences of ignoring the above are apparent in the
equivalence principle teleparallel literature. GR’s geometrization of gravity is a very
Z b deep and subtle matter; we should be wary of claims that a
m
S¼ ½mi dsmg Ba m ua dx : ð40Þ theory either reverses or extends that geometrization.
a

The force equation is now


  Acknowledgements
mg a dua mg a
@m þ B m ¼ F ua ur , ð41Þ
mi ds mi mr I am very grateful to Harvey Brown, who originally directed
where F amr is the field strength defined in Eq. (7). When mg ¼mi, my attention towards teleparallel gravity, and gave helpful feed-
this equation reduces to our original equation of motion, and, if back on a number of drafts. This paper was also greatly influenced
we substitute appropriately, can be shown to be identical to the by discussions with and feedback from David Wallace. I also owe
thanks to Erik Curiel, Malcolm Forster, Dennis Lehmkuhl, Oliver
34
Hehl (1984) justifies this association by suggesting that the spins of
35
particles act like tiny gyroscopes which detect the translational field. Stachel (2005, p. 17).
E. Knox / Studies in History and Philosophy of Modern Physics 42 (2011) 264–275 275

Pooley, and 2 anonymous referees for helpful suggestions, as well Geroch, R. (1970). Spinor structure of space-times in general relativity. II. Journal of
as to audiences in Oxford, Montreal, and St. Andrews. Mathematical Physics, 11, 343.
Glymour, C. (1970). Theoretical realism and theoretical equivalence. In: PSA:
Proceedings of the biennial meeting of the philosophy of science association (pp.
References 275–288).
Hammond, R. (2002). Torsion gravity. Reports on Progress in Physics, 65, 599–649.
Hartley, D. (1995). Normal frames for non-riemannian connections. Classical and
Aldrovandi, R., Pereira, J., & Vu, K. (2004a). Gravitation without the equivalence Quantum Gravity, 12, L103.
principle. General Relativity and Gravitation, 36, 101–110. Hehl, F. W. (1984). On the kinematics of the torsion of space-time. Foundations of
Aldrovandi, R., Pereira, J., & Vu, K. (2004b). Selected topics in teleparallel gravity. Physics, 15, 451–471.
Brazilian Journal of Physics, 34, 1374–1380. Iliev, B. (2006). Handbook of normal frames and coordinates. Basel: Birkhauser.
de Andrade, V., Guillen, L., & Pereira, J. (2000). Teleparallel gravity: An overview. Jensen, S., 2005. General relativity with torsion: Extending Wald’s chapter on
ArXiv:gr-qc/0011087. curvature. /www.slimy.com/  steuard/teaching/tutorials/GRtorsion.pdfS.
de Andrade, V., & Pereira, J. (1997). Gravitational Lorentz force and the description Jones, R. (1991). Realism about what?. Philosophy of Science, 58, 185–202.
of the gravitational interaction. Physical Review D, 56, 4689–4695. Kibble, T. W. B. (1961). Lorentz invariance and the gravitational field. Journal of
de Andrade, V. C., & Pereira, J. G. (1999). Torsion and the electromagnetic field. Mathematical Physics, 2, 212.
International Journal of Modern Physics, 8, 141. Lyre, H., & Eynck, T. (2003). Curve it, gauge it, or leave it? Practical under-
Arcos, H. I., & Pereira, J. G., 2005. Torsion gravity: A reappraisal. ArXiv:gr-qc/ determination in gravitational theories. Journal for General Philosophy of
0501017. Science, 34, 277–303.
Bain, J. (2004). Theories of Newtonian gravity and empirical indistinguishability. Malament, D. (1986). Newtonian gravity, limits, and the geometry of space. In: R.
Studies in History and Philosophy of Modern Physics, 35, 345–376. G. Colodny (Ed.), From quarks to quasars: Philosophical problems of modern
Barbour, J., & Bertotti, B. (1977). Gravity and inertia in a Machian framework. Il physics. University of Pittsburg Press.
Nuovo Cimento B, 38, 1–27. Malament, D. (1995). Is Newtonian cosmology really inconsistent?. Philosophy of
Barbour, J., & Bertotti, B. (1982). Mach’s principle and the structure of dynamical Science, 62, 489–510.
theories. Proceedings of the Royal Society of London. Series A, Mathematical and Norton, J. (1992). A paradox in Newtonian gravitation theory. In: PSA: Proceedings
Physical Sciences, 382, 295–306. of the biennial meeting of the philosophy of science association (pp. 412–420).
Blagojevic, M. (2001). Gravitation and gauge symmetries. Bristol: Taylor and Francis. Norton, J. (1995). The force of Newtonian cosmology: Acceleration is relative.
Carroll, S. (2003). Spacetime and geometry: An introduction to general relativity. San Philosophy of Science, 62, 511–522.
Franscisco: Benjamin Cummings. Ortin, T. (2004). Gravity and strings. Cambridge University Press.
Christian, J. (2001). Why the quantum must yield to gravity. In: C. Callendar, & Quine, W. (1975). On empirically equivalent systems of the world. Erkenntnis, 9,
N. Huggett (Eds.), Physics meets philosophy at the Planck scale. Cambridge 313–328.
University Press. Sauer, T. (2006). Field equations in teleparallel space-time: Einstein’s Fernpar-
Earman, J., & Friedman, M. (1973). The meaning and status of Newton’s law of allelismus approach toward unified field theory. Historia Mathematica, 33, 399.
inertia and the nature of gravitational forces. Philosophy of Science, 40, Sciama, D. W. (1962). On the analogy between charge and spin in general
329–359. relativity. In: D. W. Sciama (Ed.), Festschrift for infeld (p. 415). Oxford:
Ehlers, J. (1991). The Newtonian limit of general relativity. In: G. Ferrarese (Ed.), Pergamon.
Classical mechanics and relativity: Relationship and consistency monographs and Stachel, J. (2005). Development of the concepts of space, time and space-time from
textbooks in physical science. Naples: Bibliopolis. Newton to Einstein. In: 100 years of relativity: Space-time structure: Einstein and
Ehlers, J. (1997). Examples of Newtonian limits of relativistic spacetimes. Classical beyond (p. 3). Singapore: World Scientific.
and Quantum Gravity, 14, A119–A126. Trautman, A. (1965). Foundations and current problems of general relativity. In:
Einstein, A., & Cartan, E. (1979). In: R. Debever (Ed.), Letters on absolute parallelism, S. Deser, & K. W. Ford (Eds.), Lectures on general relativity. New Jersey:
1929–1932. Princeton University Press. Englewood Cliffs, Prentice-Hall.
Friedman, M. (1983). Foundations of space time theories. Princeton University Press. Wald, G. (1984). General relativity. University of Chicago Press.

You might also like