1
July 20, 2001; rev. August 16, 2001
General Covariance, Gauge Theories and the Kretschmann
Objection.
Submitted to Symmetry in Physics: New Reflections,
Katherine Brading and Elena Castellani ( eds), in preparation.
John D. Norton 1
Department of History and Philosophy of Science
University of Pittsburgh, Pittsburgh PA USA 15260
jdnorton@pitt.edu
How can we reconcile two claims that are now both widely accepted:
Kretschmann's claim that a requirement of general covariance is physically
vacuous and the standard view that the general covariance of general relativity
expresses the physically important diffeomorphism gauge freedom of general
relativity? I urge that both claims can be held without contradiction if we attend
to the context in which each is made.
1 I thank Carlo Rovelli, John Earman, Elena Castellani and Chris Martin for their discussion
and for forcing me to think this through. I am also grateful for discussion by the participants
in the "International Workshop: General Covariance and Quantum?: Where Do We Stand,"
Department of Physics, University of Parma, June 21-23, 2001, organized by Massimo Pauri.


2
1. Introduction
Two views...
When Einstein formulated his general theory of relativity, he presented it as the
culmination of his search for a generally covariant theory. That this was the signal
achievement of the theory rapidly became the orthodox conception. A dissident view,
however, tracing back at least to objections raised by Erich Kretschmann in 1917, holds that
there is no physical content in Einstein's demand for general covariance. That dissident view
has grown into the mainstream. Many accounts of general relativity no longer even mention
a principle or requirement of general covariance.
What is unsettling for this shift in opinion is the newer characterization of general
relativity as a gauge theory of gravitation, with general covariance expressing a gauge
freedom. The recognition of this gauge freedom has proven central to the physical
interpretation of the theory. That freedom precludes certain otherwise natural sorts of
background spacetimes; it complicates identification of the theory's observables, since they
must be gauge invariant; and it is now recognized as presenting special problems for the
project of quantizing of gravitation.
...That We Need not Choose Between
It would seem unavoidable that we can choose at most one of these two views: the
vacuity of a requirement of general covariance or the central importance of general
covariance as a gauge freedom of general relativity. I will urge here that this is not so; we
may choose both, once we recognize the differing contexts in which they arise. Kretschmann's
claim of vacuity arises when we have some body of physical fact to represent and we are
given free reign in devising the formalism that will capture it. He urges, correctly I believe,
that we will always succeed in finding a generally covariant formulation. Now take a
different context. The theory—general relativity—is fixed both in its formalism and physical


3
interpretation. Each formal property of the theory will have some meaning. That holds for its
general covariance which turns out to express an important gauge freedom.
To Come
In Section 4 I will lay out this reconciliation in greater detail. As preparation, in
Sections 2 and 3, I will briefly review the two viewpoints. Finally in Section 5 I will relate the
reconciliation to the fertile "gauge principle" used in recent particle physics. An Appendix
discusses the difficulty of making good on Kretschmann's claim that generally covariant
reformulations are possible for any spacetime theory.
2. Einstein and Kretschmann's Objection
Einstein...
In November 1915 an exhausted and exhilarated Einstein presented the
gravitational field equations of his general theory of relativity to the Prussian Academy of
Science. These equations were generally covariant; they retained their form under arbitrary
transformation of the spacetime coordinate system. This event marked the end of a seven
year quest, with the final three years of greatest intensity, as Einstein struggled to see that a
generally covariant theory was physically admissible. 2
Einstein had several bases for general covariance. He believed that the general
covariance of his theory embodied an extension of the principle of relativity to acceleration.
This conclusion seemed automatic to Einstein, just as the Lorentz covariance of his 1905
formulation of special relativity expressed its satisfaction of the principle of relativity of
2 Over the last two decades there has been extensive historical work on this episode. Earlier
works include Stachel (1980) and Norton (1984); the definitive work will be Renn et al. (in
preparation).


4
inertial motion. 3 He also advanced what we now call the "point-coincidence" argument. The
physical content of a theory is exhausted by a catalog of coincidences, such as the coincidence
of a pointer with a scale, or, if the world consisted of nothing but particles in motion, the
meetings of their worldlines. These coincidences are preserved under arbitrary coordinate
transformations; all we do in the transformations is relabel the spacetime coordinates
assigned to each coincidence. Therefore a physical theory should be generally covariant. Any
less covariance restricts our freedom to relabel the spacetime coordinates of the coincidences
and that restriction can be based in no physical fact.
...and Kretschmann
Shortly after, Erich Kretschmann (1917) announced that Einstein had profoundly
mistaken the character of his achievement. In demanding general covariance, Kretschmann
asserted, Einstein had placed no constraint on the physical content of his theory. He had
merely challenged his mathematical ingenuity. For, Kretschmann urged, any spacetime
theory could be given a generally covariant formulation as long as we are prepared to put
sufficient energy into the task of reformulating it. In arriving at general relativity, Einstein
had used the "absolute differential calculus" of Ricci and Levi-Civita (now called "tensor
calculus.") Kretschmann pointed to this calculus as a tool that made the task of finding
generally covariant formulations of theories tractable. 4
Kretschmann's argument was slightly more subtle than the above remarks.
Kretschmann actually embraced Einstein's point-coincidence argument and turned it to his
own ends. In his objection, he agreed that the physical content of spacetime theories is
3 The analogy proved difficult to sustain and has been the subject of extensive debate. See
Norton (1993).
4 For further discussion of Kretschmann's objection, Einstein's response and of the still active
debate that follows, see Norton (1993) and Rynasiewicz (1999)


5
exhausted by the catalog of spacetime coincidences; this is no peculiarity of general
relativity. For this very reason all spacetime theories can be given generally covariant
formulations. 5
Kretschmann's objection doe s seem sustainable. For example, using Ricci and Levi
Civita's methods it is quite easy to give special relativity a generally covariant formulation.
In its standard Lorentz covariant formulation, using the standard spacetime coordinates (t,
x, y, z), special relativity is the theory of a Minkowski spacetime whose geometry is given by
the invariant line element
ds2 = c 2dt2 - dx2 -dy2 - dz2 (1)
Free fall trajectories (and other "straights" of the geometry) are given by
d2x/dt2 = d2y/dt2 = d2z/dt2 = 0 (2)
We introduce arbitrary spacetime coordinates x i, for i = 0,...,3 and the invariant line element
becomes
ds2 = gik dxi dxk (3a)
where the matrix of coefficients g ik is subject to a field equation
Riklm = 0 (3a)
with Riklm the Riemann-Christoffel curvature tensor. The free falls are now governed by
d2xi/ds2 + {kim} dxk/ds dxm/ds = 0 (4)
where {kim} are the Christoffel symbols of the second kind.
5 Rhetorically, Kretschmann's argument was brilliant. To deny it, Einstein may need to deny
his own point-coincidence argument. However a persistent ambiguity remains in Einstein's
original argument. Just what is a point-coincidence? Einstein gives no general definition. He
gives only a list of illustrations and many pitfalls await those who want to make the
argument more precise. For example, see Howard (1999).


6
Examples such as this suggest that Kretschmann was right to urge that generally
covariant reformulations are possible for all spacetime theories. While the suggestion is
plausible it is certainly not proven by the examples and any final decision must await
clarification of some ambiguities. See Appendix 1: Is a Generally Covariant Reformulation
Always Possible? for further discussion.
3. The Gauge Freedom of General Relativity
Active General Covariance
Einstein spoke of general covariance as the invariance of form of a theory's equations
when the spacetime coordinates are transformed. It is usually coupled with a so-called
"passive" reading of general covariance: if we have some system of fields, we can change our
spacetime coordinate system as we please and the new descriptions of the fields in the new
coordinate systems will still solve the theory's equations. Einstein's form invariance of the
theory's equations also licenses a second version, the so-called "active" general covariance. It
involves no transformation of the spacetime coordinate system. Rather active general
covariance licenses the generation of many new solutions of the equations of the theory in the
same coordinate system once one solution has been given.
For example, assume the equations of some generally covariant theory admit a scalar
field φ(xi) as a solution. Then general covariance allows us to generate arbitrarily many more
solutions by, metaphorically speaking, spreading the scalar field differently over the
spacetime manifold of events. We need a smooth mapping on the events—a
diffeomorphism—to effect the redistribution. For example, assume we have such a map that
sends the event at coordinate x i to the event at coordinate x' i in the same coordinate system.
Such a map might be a uniform doubling, so that x i is mapped to x' i = 2. xi. To define the
redistributed field φ', we assign to the event at x' i the value of the original field φ at the event


7
with coordinate x i.6 If the field is not a scalar field, the transformation rule is slightly more
complicated. For further details of the scalar case, see Appendix 2: From Passive to Active
Covariance.
Why it is a Gauge Freedom
The fields φ(xi) and φ'(xi) are mathematically distinct. But do they represent
physically distinct fields? The standard view is to assume that they do not, so that they are
related by a gauge transformation, that is, one that relates mathematically distinct
representations of the same physical reality. That this is so cannot be decided purely by the
mathematics. It is a matter of physics and must be settled by physical argumentation.
A vivid way to lay out the physical arguments is through Einstein's "hole
argument."7 The transformation on the manifold of events can be set up so that it is the
identity everywhere outside some nominated neighborhood of spacetime ("the hole") and
comes smoothly to differ within. We now use the transformation to duplicate
diffeomorphically all the fields of some generally covariant theory. Do the new fields
represent the same physical reality as the old? It would be very odd if they did not. Both
systems of fields agree completely in all invariants; they are just spread differently on the
manifold. Since observables are given by invariants, they agree in everything observable.
Moreover, the two systems of field will agree everywhere outside the hole, but they differ
only within. This means that, in a generally covariant theory, fixing all fields outside this
neighbor fails to fix the fields within. This is a violation of determinism. In short, if we
assume the two systems of fields differ in some physical way we must insist upon a difference
6 To visualize this redistribution in the two dimensional case, imagine that the original field
is represented by numbers written on a flat rubber membrane. If we now uniformly stretch
the rubber membrane so it doubles in size, we have the new field.
7 See Earman and Norton (1987), Norton (1999).


8
that transcends both observation and the determining power of the theory. The ready
solution is that these differences are purely ones of mathematical representation and that
the two systems of fields represent the same physical reality.
Its Physical Consequences
Accepting that this gauge freedom has important consequences for the physical
interpretation of a theory such as general relativity. 8 The theory is developed by positing a
manifold of spacetime events which is then endowed with metric properties by means of a
metric tensor field g ik. The natural default is to take the manifold of events as supplying
some kind of independent background spacetime in which physical processes can unfold. The
gauge freedom makes it very difficult to retain this view. For, when we apply a
diffeomorphism to the field and spread the metrical properties differently over events, the
transformation is purely gauge and we end up changing nothing physical. So now the same
events are endowed with different properties, yet nothing physical has changed. The simplest
and perhaps only way to make sense of this is to give up the idea of an independent existence
of the events of the manifold. In so far as we can associate an event of the manifold with real
events in the world, that association must change in concert with our redistribution of the
metrical field over the manifold.
Our notion of what is observable is affected by similar considerations. What is
observable is a subset of the physically real and that in turn is expressed by the invariants of
a theory. Might an observable result consist of the assertion that an invariant of some field
has such and such a value at some event of the manifold? No. The invariance must also
include invariance under the gauge transformation and the assertion would fail to be
invariant under the gauge transformation. In redistributing the fields, the transformation
8 For further discussion of these and related issues and their import for the quantization of
gravity see Rovelli (1997).


9
might relocate that invariant with that value at quite another event of the manifold. If some
result is eradicated by a gauge transformation, it cannot have been a result expressing
physical fact since the gauge transformation alters nothing physical. We must resort to more
refined ways of representing observables. For example, they may be expressed by an
assertion that two invariants are equal. The event at which the equality resided may vary
under gauge transformation; but the transformation will preserve the equality asserted.
4. Reconciliation
The Context in which Kretschmann's Objection Succeeds
Kretschmann's objection s ucceeds because he allows us every freedom in
reformulating and reinterpreting terms within a theory. Thus we easily transformed special
relativity from its Lorentz covariant formulation (1), (2) to a generally covariant formulation
(3a), (3b), (4). In doing so, we introduced new variables not originally present. The are the
coefficients of the metric tensor g ik and the Christoffel symbols { kim}.
With this amount of freedom, it is plausible that we can arrive at formulations of any
theory that have any designated formal property. 9 Imagine, for example, that we wanted a
9 I am distinguishing the formalism of the theory (and its formal properties) from its
interpretation. The formalism of a theory would be the actual words used, if the theory
consisted of an English language description, independently of their meanings. Formal
properties would include such things as the choice of English and the number of words. More
commonly physical theories use mathematical structures in place of words. These structures
can be considered quite independently of what we take them to represent in the world. The
properties we then consider are the purely formal properties. A real valued field on some
manifold is just a mathematical structure until we specify what it may represent in the
world. That specification is the job of the interpretation. See next footnote.


10
formulation of Newtonian particle mechanics in which the string of symbols " E=mc2"
appears. (This is a purely formal property since we place no conditions on what the string
might mean.) Here is one way we can generate it. We take the usual expression for the
kinetic energy K of a particle of mass m moving at velocity v, K=(1/2) mv2. We introduce a
new quantity E, defined by E = 2K, and also a new label "c" for velocity v, so that c=v. Once
we substitute these new variables into the expression for kinetic energy, our reformulated
theory contains the string " E=mc2".
The physical vacuity arises because we are demanding the formal property of general
covariance (or some other formal property) without placing further restrictions that would
preclude it always being achievable. The vacuity would persist even if we demanded a fixed
physical content; we must simply be careful not to alter our initial physical content as we
adjust its formal clothing. In the case of the discovery of general relativity, Einstein did not
keep the physical content fixed. It became fully fixed only after he found a generally
covariant formulation that satisfied a number of restrictive physical limitation.
The Context in Which the Diffeomorphism Gauge Freedom has Physical Content
Matters are quite different if we fix the formalism of the theory and its
interpretation. So we might be given general relativity in its standard interpretation. 10 If a
theory has any content at all, we must be able to ascribe some physical meaning to its
assertions. A fortiori there must some physical meaning in the general covariance of general
10 By "interpretation" I just mean the rules that tell us how to connect the various terms or
mathematical structures of the theory with things in the physical world. These rules can
vary from formulation to formulations and theory to theory. So, in ordinary formulations of
special relativity, "c" refers to the speed of light. In thermodynamics "c" would refer to
specific heat.


11
relativity. It may be trivial or it may not. 11 Consulting the theory, as we did in Section 2
above reveals that the content is not trivial.
Things are just the same in our toy example of forcing the string " E=mc2" into a
formulation of Newtonian particle mechanics. Let us fix the formulation to be the doctored
one above. We had forced the string " E=mc2" into it. But now that we have done it, the string
uses symbols that have a meaning and, when we decode what it says about them, we
discover that the string expresses something physical, the original statement that kinetic
energy is half mass x (velocity) 2. Mimicking Kretschmann, we would insist that, given
Newtonian particle mechanics or any other theory, some reformulation with the string is
assuredly possible; so the demand for it places no restriction on the physically possible. But,
once we have the reformulation, that string will express something.
The analogous circumstance arises in the generally covariant reformulation of special
relativity. The existence of the reformulation is assured. Once we have it, its general
covariance does express something. In this case, it is a gauge freedom of the geometric
structure just like that of general relativity. The Lorentz covariant formulation of (1) and (2)
admits preferred coordinate systems. In effect, some of the physical content of the theory is
encoded in them. They specify, for example, which are the inertial motions; a body moves
inertially only if there is a coordinate system in which its spatial coordinates do not change
with the time coordinate. In the transition to the generally covariant formulation, this
11 Indeed the assertion may prove to be a logical truth, that is, it would be true by the
definition of the terms it invokes or it may amount to the definition of term.While their truth
is assured, such assertions need not be trivial. For example in a formulation of special
relativity we may assert that that coefficients of the metric tensor are linear functions of the
coordinates. This turns out to place no physical restriction on the theory; it merely restricts
us to particular coordinate systems. It is what is known as a coordinate condition that
defines the restricted class of coordinate systems in which the formulation holds.


12
content is stripped out of the coordinate systems. We can no longer use constancy of spatial
coordinates to discern which points move inertially. This content is relocated in the
Christoffel symbols, which, via equation (4) determine whether a particular motion is
inertial. The general covariance of (3a), (3b) and (4) leave a gauge freedom in how the metric
gik and the Christoffel symbols { kim} may be spread over some coordinate system. In one
coordinate system, they may be spread in many mathematically distinct but physically
equivalent ways.
To summarize
There is no restriction on physical content in saying that there exists a formulation of the
theory that has some formal property (general covariance, the presence of the string of
symbols " E=mc2", etc.) But once we fix a particular formulation and interpretation, that very
same formal property will express something physical, although there is no assurance that it
will be something interesting.
5. Gauge Theories in Particle Physics
This summary generates a new puzzle. One of the most fertile strategies in recent
decades in particle physics has been to extend the gauge symmetries of non-interacting
particles and thereby infer to new gauge fields that mediate the interaction between the
particles. Most simply, the electromagnetic field can be generated as the gauge field that
mediates interactions of electrons. This power has earned the strategy the label of the "gauge
principle." How can this strategy succeed if Kretschmann is right and there is no physical
content in our being able to arrive at a reformulation of expanded covariance? In the particle
context, this corresponds to a reformulation of expanded gauge freedom. So why doesn't
Kretschmann's objection also tell us that the strategy of the gauge principle is physically
vacuous?


13
The solution lies in the essential antecedent condition of Kretschmann's objection.
The physical vacuity arises since there are no restrictions placed how we might reformulate a
theory in seeking generally covariance. It has long been recognized that the assured
achievement of general covariance can be blocked by some sort of additional restriction on
how the reformulation may be achieved. Many additional conditions have been suggested,
including demands for simplicity and restrictions on which extra variables may introduced.
(For a survey, see Norton, 1993, Section 5; Norton, 1995, Section 4.) The analogous solution
is what gives the gauge principle its content. In generating gauge fields, we are most
definitely not at liberty to expand the gauge freedom of some non-interacting particle field in
any way we please. There is a quite precise recipe that must be followed: we must promote a
global symmetry of the original particle field to a local symmetry, using the exemplar of the
electron and the Maxwell field, and the new field arises from the connection introduced to
preserve gauge equivalence. 12
There is considerably more that should be said about the details of the recipe and the
way in which new physical content arises. The recipe is standardly presented as merely
expanding the gauge freedom of the non-interacting particles, which should mean that the
realm of physical possibility is unaltered; we merely have more gauge equivalent
representations of the same physical situations. So how can physically new particle fields
12 The transition from special relativity in (1) and (2) to the generally covariant formulation
(3a), (3b) and (4) can be extended by one step. We replace the flatness condition (3a) by a
weaker condition, a natural relaxation, R ik = glmRilmk = 0. The result is general relativity in
the source free case. Arbitrary, source free gravitational fields now appear in the generalized
connection { kim}. We have what amounts to the earliest example of the use of the gauge
recipe to generate new fields. The analogy to more traditional examples in particle physics is
obvious.


14
emerge? This question is currently under detailed and profitable scrutiny. See Martin (2000),
(manuscript) and contributions to this volume.
Appendix 1: Is a Generally Covariant Reformulation Always
Possible?
As Earman (manuscript, Section 3) has pointed out, it is not entirely clear whether a
generally covariant reformation is always possible for any spacetime theory. The problem lies
in ambiguities in the question. Just what counts as "any" spacetime theory? Just what are
we expecting from a generally covariant reformulation? Let me rehearse some of the
difficulties and suggest that for most reasonable answers to these questions generally
covariant reformulation will be possible though not necessarily pretty.
The Substitution Trick...
Let us imagine that we are given a spacetime theory in a formulation of restricted
covariance. It is given in just one spacetime coordinate systems X i. Let us imagine that the
laws of the theory happen to be given by n equations in the 2n quantities A k, Bk
Ak(X i) = Bk(X i) (5)
where k = 1, ..., n and the A k and Bk are functions of the coordinates as indicated. Consider
an arbitrary coordinate system x i to which we transform by means of the transformation law
xi = xi(X m) (6)
We can replace the n equations (5) by equations that hold in the arbitrary coordinate system
by the simple expedient of inverting the transformation of (6) to recover the expression for
the X m as a function of the x i, that is X m = X m(xi). Substituting these expressions for X m into
(5), we recover a version of (5) that holds in the arbitrary coordinate system
Ak(X i(xm)) = Bk(X i(xm)) (5a)


15
We seemed to have achieved a generally covariant reformulation of (5) by the most direct
application of the intuition that coordinate systems are merely labels and we can relabel
spacetime events as we please.
...Yields Geometric Objects
While (5a) is generally covariant, we may not be happy with the form of the general
covariance achieved—one of the ambiguities mentioned above. We might, as Earman
(manuscript, Section 3) suggests, want to demand that (5a) be expressed in terms of
geometric object fields. The standard definition of a geometric object field is that it is an n
tuple valued field of components on the manifold, with one field for each coordinate system,
and that the transformation rule that associates the components of different coordinate
systems have the usual group properties.
While this definition may appear demanding, it turns out to be sufficiently
permissive to characterize each side of (5a) as a geometric object field. For example, in each
coordinate system x m, the geometric object field A has components A k(X i(xm)), which I now
write as Ak(xm). The transformation rule between the components is induced by the rule for
coordinate transformations. That is, under the transformation x m to yr(xm), Ak(xm)
transforms to A k(xm(yr)), where xm(yr) is the inverse of the coordinate transformation. With
this definition of the transformation law for A k, the components will inherit as much group
structure as the coordinate transformations themselves have; that is, it will be as much of a
geometric object field as we can demand. 13 For example, assume the transformations of
13 Why the hedged "as much group structure as the coordinate transformations themselves
have"? These general coordinate transformations may not have all the group properties if the
domains and ranges of the transformations do not match up well. Assume T 1 maps
coordinates x i on a neighborhood A to coordinates y i on a neighborhood B that is a proper
subset of A and T 2 maps coordinates y i on B to coordinates z i on A. Then the composition


16
coordinate systems z p to yr and yr to xm conform to transitivity. Then this same transitivity
will be inherited by A. We will have A k(xm(zp)) = A k(xm(yr(zp))) since the transitivity of the
coordinate transformation yields x m(zp) = xm(yr(zp)).
But areThey the Geometric Objects We Expect?
While the components A k turn out to be geometric object fields, they are probably not
the ones we expected. In brief, the reason is that the transformation rule induced by the
substitution trick does not allow any mixing of the components. That precludes it yielding
vectors or tensor or like structures; it turns everything into scalar fields. To see how odd this
is, take a very simple case. Imagine that we have special relativity restricted to just one
coordinate system X i. Our law might be the law governing the motion of a body of unit mass,
Fi = A i, where Fi is the four force and A i the four acceleration. Under a Lorentz
transformation
Y0 = γ(X 0 – vX 1) Y1 = γ(X 1 – vX 0) Y3 = X 3 Y4 = X 4
with velocity v in the X 1 direction, c=1 and γ = (1-v2)-1/2.The usual Lorentz transformation for
the components A i of the four acceleration would be
A' 0 = γ(A 0 – vA1) A' 1 = γ(A 1 – vA0) A' 3 = A 3 A' 4 = A 4 (6)
Note that the transformed A' 0 and A' 1 are linear sums of terms in A 0 and A1. For this same
transformation, the substitution trick merely gives us
A' i = A i(X m(Yr)) (6a)
That is, A' 0 is a function of A 0 only and A' 1 is a function of A 1 only.
This oddity becomes a disaste r if we apply the substitution rule in a natural way.
Instead of starting with Ai in one fixed coordinate system X i, we might start will the full set
T2T1 cannot coincide with the direct transformation of x i to zi since the composition has lost
that part of the transformation outside B.


17
of all components of A i in all coordinate systems related by a Lorentz transformation to X i. If
we now try and make this bigger object generally covariant by the substitution trick, we will
end up with two incompatible transformation laws for the transformation X i to Yi when we
try to transform the components A i—the law (6) and law (6a). We no longer have a geometric
object field since we no longer have a unique transformation law for the components.
The escape from this last problem is to separate the two transformation groups. We
consider A i in coordinate system X i and A' i in coordinate system Y i separately and convert
them into distinct geometric object fields by the substitution trick. As geometric object fields
they have become, in effect, scalar fields. The Lorentz transformation then reappears as a
transformation between these geometric objects.
The Coordinates as Scalars Trick
If this is our final goal, then another general trick for generating generally covariant
reformulations could have gotten us there much faster. We return to A k(X i) of (5). We can
conceive the X i as scalar fields on the manifold—that is really all they are. 14 Scalar fields are
geometric object fields already. The A i are functions of X i, that is, functions of scalar fields.
Therefore they are also geometric objects. So we can conceive of the entire structure A k(X i) as
a geometric object field. We have gotten general covariance on the cheap. We cannot avoid a
cost elsewhere in the theory however. Our reformulation is overloaded with structure, one
geometric object field for each of what was originally a component. There is clearly far more
mathematical structure present than has physical significance. So the theory will need a
careful system for discerning just which parts of all this structure has physical significance.
14 Ask, what is the X 0 coordinate in coordinate system X i of some event p? The answer will be
the same number if we ask it from any other coordinate system y i as long as we are careful to
ask it of the original coordinate system X i. That is, each coordinate can be treated as a scalar
field.


18
Temptations Resisted
These devices for inducing general covarianc e are clumsy but they do fall within the
few rules discussed. We might be tempted to demand that we only admit generally covariant
formulations if their various parts fall together into nice compact geometric objects. But what
basis do we have for demanding this? Are we to preclude the possibility that the theory we
started with is just a complicated mess that can only admit an even more complicated mess
when given generally covariant reformulations. (Newtonian theory has been accused of this!)
And if we are to demand only nice and elegant reformulations, just how do we define "nice
and elegant"?
My conclusion is that generally covariant reformulations are possible under the few
rules discussed and that efforts to impose further rules to block the more clumsy ones will
cause more trouble than they are worth elsewhere.
Appendix 2: From Passive to Active Covariance
As above, assume the equations of some generally covariant theory admit a scalar
field φ(xi) as a solution. We can transform to a new coordinate system by merely relabeling
the events of spacetime; x i is relabeled x' i, where the x'i are smooth functions of the x i. The
field φ(xi) transforms to field φ'(x'i) by the simple rule φ'(x'i) = φ(xi). Since the equations of the
theory hold in the new coordinate system, the new field φ'(x'i) will still be a solution. The two
fields φ(xi) and φ'(x'i) are just representations of the same physical field in different
spacetime coordinate systems.
This is the passive view of general covariance. It can be readily t ransmogrified into
an active view, a transition that Einstein had already undertaken with his 1914 statements
of the "hole argument". What makes φ'(x'i) a solution of the theory under discussion is
nothing special about the coordinate system x' i. It is merely the particular function that φ'
happens to be. It is a function that happens to satisfy the equations of the theory. We could


19
take that very same function and use it in the original coordinate system, x i. That is, we
could form a new field φ'(xi). Since this new field uses the very same function, it retains every
property except the mention of the primed coordinate system x' i. Thus it is also a solution of
the equations of the theory.
In short, the passive general covariance of the theory has delivered us two fields,
φ(xi) and φ'(xi). They are not merely two representations of the same field in different
coordinate systems. They are defined in the same coordinate system and are mathematically
distinct fields, in so far as their values at given events will (in general) be different. Active
general covariance allows the generation of the field φ'(xi) from φ(xi) by the transformation xi
to x'i.
References
Earman, John (manuscript) "Once More General Covariance."
Earman, John and Norton, John D. (1987): "What Price Spacetime Substantivalism? The
Hole Argument," British Journal for the Philosophy of Science. 38, pp. 515-25.
Howard, Don (1999) "Point Coincidences and Pointer Coincidences: Einstein on the Invariant
Content of Space-Time Theories," in H. Goenner, J. Renn, J. Ritter, T. Sauer (eds.)
The Expanding Worlds of General Relativity (Einstein Studies volume 7) pp. 463
500.
Kretschmann, Erich (1917): "Über den physikalischen Sinn der Relativitätspostulat, A
Einsteins neue und seine ursprünglische Relativitätstheorie," Annalen der Physik,
53, 575-614.
Martin, Christopher (2000) "The Gauge Argument," Talk at the Philosophy of Science
Association Biennial Meeting, Vancouver, 2-5, November, 2000.
Martin, Christopher (manuscript) Dissertation, Department of History and Philosophy of
Science, University of Pittsburgh.


20
Norton, John D. (1984) "How Einstein found his Field Equations: 1912-1915," Historical
Studies in the Physical Sciences, 14, 253-316; reprinted in Don Howard and John
Stachel (eds.) Einstein and the History of General Relativity: Einstein Studies, Vol.
1 Boston: Birkhäuser, 1989, pp.101-159.
Norton, John D. (1993), "General Covariance and the Foundations of General Relativity:
Eight Decades of Dispute," Reports on Progress in Physics, 56, pp. 791-858.
Norton, John D. (1995) "Did Einstein Stumble? the Debate Over General Covariance,"
Erkenntnis 42, pp. 223-45.
Norton, John D. (1999) "The Hole Argument," Stanford Encyclopedia of Philosophy
http://plato.stanford.edu/entries/spacetime-holearg/
Renn, Jürgen; Sauer, Tilman; Janssen, Michel; Norton, John D. and Stachel John (in
preparation) General Relativity in the Making; Einstein's Zurich Notebook.
Rovelli, Carlo (1997) "Halfway through the Woods: Contemporary Research on Space and
Time," pp. 180-223 in J. Earman and J. D. Norton (eds.) The Cosmos of Science:
Essays of Exploration. Pittsburgh: University of Pittsburgh Press.
Rynasiewicz, Robert (1999) "Kretschmann's Analysis of General Covariance and Relativity
Principles," in H. Goenner, J. Renn, J. Ritter, T. Sauer (eds.) The Expanding
Worlds of General Relativity (Einstein Studies volume 7) pp. 431-462.
Stachel, John (1980): "Einstein's Search for General Covariance," paper read at the Ninth
International Conference on General Relativity and Gravitation, Jena; printed in
Don Howard and John Stachel (eds.) Einstein and the History of General Relativity:
Einstein Studies, Vol. 1 (Boston: Birkhäuser, 1989) pp.63-100.