Reopening the Hole Argument∗

Klaas Landsman
Institute for Mathematics, Astrophysics, and Particle Physics (IMAPP),
Department of Mathematics, Radboud University, Nijmegen, The Netherlands
Email: [email protected]
Abstract
The aim of this expository paper is to argue that although Weatherall (2018)
and Halvorson & Manchak (2022) claim to ‘close the Hole Argument’, its
arXiv:2206.04943v1 [gr-qc] 10 Jun 2022

philosophical thrust may be resurrected by rephrasing the argument in terms

of the theorem of Choquet-Bruhat & Geroch (1969) on the existence and
uniqueness of maximal globally hyperbolic solutions to the Einstein field equa-
tions for suitably posed initial data. This not only avoids the controversial
pointwise identification of manifolds underlying non-isometric spacetimes, but
also suggests a precise notion of determinism intrinsic to gr (as an initial value
problem), which may be compared e.g. with the one proposed by Butterfield.
We also discuss generally covariant special relativity in this context, and fi-
nally muse about possible implications for the general philosophy of science.

1 Introduction
Initially, the Hole Argument (Lochbetrachtung) was an episode in Einstein’s strug-
gle between 1913–1915 to find the gravitational field equations of general relativity.
At a time when he was already unable to find generally covariant equations for the
gravitational field (i.e. the metric) that had the correct Newtonian limit and satisfied
energy-momentum conservation, the Hole Argument confirmed him in at least tem-
porarily giving up the idea of general covariance (which he later recovered without
ever mentioning the Hole Argument again). More generally, Einstein’s invention of
the argument formed part of his analysis of the interplay between general relativ-
ity (of motion), general covariance (of equations under coordinate transformations),
and determinism. The more recent emphasis on substantivalism versus relational-
ism (Earman & Norton, 1987) is not Einstein’s, but since for him this opposition
was closely related to the problem of absolute versus relative motion and hence to
general relativity (Earman, 1989), he would certainly have been interested in it.1
∗
If anything, this paper is a tribute to the weekly Cambridge–LSE Philosophy of Physics
Bootcamp, in which the hole argument has often been discussed. I am especially indebted to the
organizers of this seminar, Jeremy Butterfield and Bryan Roberts, as well as to Henrique Gomes,
Hans Halvorson, Joanna Luc, and JB Manchak, for comments on earlier drafts. I also wish to thank
Michel Janssen for historical comments, some of which have found a way to the Introduction.
1
See Janssen & Renn (2022) for the final reconstruction of Einstein’s struggle, with §4.1 devoted
to the Hole Argument. The earliest known reference to the Hole Argument is in a memo by
Einstein’s friend and colleague Besso dated August 1913, provided this dating is correct (Janssen,
2007). Einstein subsequently presented his argument four times in print; I just cite Einstein (1914)
In modernized form (using a global perspective and replacing Einstein’s coordinate
transformations by diffeomorphisms), his reasoning was essentially as follows:2

• Let (M, g) be a spacetime.3 The transformation behaviour of the Einstein

tensor Ein(g) under diffeomorphisms ψ of the underlying manifold M is

ψ ∗ (Ein(g)) = Ein(ψ ∗ g). (1)

Similarly, for any healthy energy-momentum tensor T (g, F ) constructed from

the metric g and the matter fields F that matter we should have

ψ ∗ (T (g, F )) = T (ψ ∗ g, ψ ∗F ). (2)

Consequently, if g satisfies the Einstein equations Ein(g) = 8π T (g, ϕ), then

ψ ∗ g satisfies these equations for the transformed matter fields ψ ∗ F .

• Now consider an open connected vacuum region H in spacetime possibly sur-

rounded by matter (i.e. F = 0 in H); H is referred to as a “hole”, whence
the name of the argument.4 Furthermore, find a diffeomorphism ψ that is
nontrivial inside H and equals the identity outside H, so that in particular,

T (ψ ∗ g, ψ ∗F ) = T (ψ ∗ g, F ) = T (g, F ), (3)

both outside H (where ψ is the identity) and inside H (where T (g, F ) = 0).

• It follows from the previous points that if g satisfies the Einstein equations
for some energy-momentum tensor T , then so does ψ ∗ g. Hence the space-
times (M, g) and (M, ψ ∗ g) satisfy the Einstein equations for the same matter
distribution and are identical outside H. But they differ inside the hole.

Einstein saw this as a proof that the matter distribution fails to determine the
metric uniquely, and regarded this as such a severe challenge to determinism that,
supported by the other problems he had at the time, he retracted general covariance.
as the paper containing his final version. See Stachel (2014), Norton (2019), and Pooley (2022),
and references therein for reviews of the Hole Argument in both a historical and a modern context.
2
We write the Einstein tensor as Ein(g), where its dependence on the metric g is explicitly
denoted; in coordinates we have Ein(g)µν = Gµν = Rµν − 12 gµν R.
3
A spacetime is a smooth four-dimensional connected Lorentzian manifold with time orientation
(this nomenclature of course hides philosophical issues to be discussed later in this paper). More
generally, my notations and conventions follow Landsman (2021) and are standard, e.g. spacetime
indices are Greek whereas spatial ones are Latin, the metric has signature − + ++, etc.
4
Einstein’s arrangement looks unnatural compared to Hilbert’s (1917) reformulation as an
initial-value problem in the pde sense, see below, but Einstein was probably inspired by Mach’s
principle, where “the fixed stars at infinity” determine the local inertia of matter; see Maudlin
(1990), Hofer (1994), and Stachel (2014). There is another argument that actually favours Ein-
stein’s curious setting for the Hole Argument: the smaller the hole, i.e. the larger the complement
of the hole, the greater the challenge to determinism, for if even things almost everywhere except
in a tiny hole fail to determine things inside that hole, then we should really worry (Butterfield,
1989). This pull admittedly gets lost in the initial-value formulation of the argument below. See
Muller (1995) for the explicit construction of a hole diffeomorphism (the only one I am aware of).

2
 From a modern point of view the energy-momentum tensor is a red herring in
the argument,5 which may just as well be carried out in vacuo, as will be done from
now on; this also strengthens my subsequent reformulation of the argument, since
the theorem on which this is based is less well developed in the presence of matter.
Earman & Norton (1987) famously revived the Hole Argument. Streamlining:
1. Although (M, g) and (M, ψ ∗ g) are different spacetimes (unless of course ψ is
an isometry of (M, g), i.e. ψ ∗ g = g), physicists—usually tacitly—circumvent
this alleged lack of determinism of gr by simply “identifying” the two, i.e. by
claiming that (M, g) and (M, ψ ∗ g) represent “the same physical situation”.
2. In this practice they are encouraged by the trivial observation that (M, g) and
(M, ψ ∗ g) are isometric; indeed, the pertinent isometry is none other than ψ.6
3. However —and this is their key point—this spells doom for spacetime substan-
tivalists (like Newton), who (allegedly) should be worried that if in order to
save determinism, x ∈ M, carrying the metric ψ ∗ g(x), must be identified with
ψ(x) ∈ M, carrying the same metric, then points have lost their “this-ness”:
they cannot be identified as such, but only as carriers of metric information.
Apart from its primary bearing on spacetime substantivalism and determinism, the
Hole Argument also has implications for the general philosophy of science (see §4).
But all of this is pointless if the argument is void, as claimed by Weatherall (2018)
and his followers (Fletcher, 2020; Bradley & Weatherall, 2021; Halvorson & Man-
chak, 2022), against critics including e.g. Arledge & Rynasiewicz (2019), Roberts
(2020), Pooley & Read (2021), and Gomes (2021). My goal is not to decide who is
right or wrong, but, in the light of the indisputable fact that all arguments so far
have turned out to be controversial, to present a version of the Hole Argument that
should be uncontroversial, while leading to exactly the same philosophical issues.
In order to see some of the advantages of the version of the Hole Argument
I prefer (i.e. based on Theorem 2 below), it may be useful to briefly discuss the
main point raised by Weatherall (2018) and subsequently by Halvorson & Manchak
(2022). The main issue seems to be whether it is a valid move in gr to compare—or
even put—two different metrics g and g ′ on the same manifold M (for example in
the Hole Argument above one takes g ′ = ψ ∗ g). Since this involves comparing g(x)
with g ′ (x), the objection would really be against identifying x ∈ M seen as a point
in the spacetime (M, g) with x ∈ M seen as a point in a different spacetime (M, g ′ ):
5
Continuing footnote 4: Janssen (2007), footnote 98, notes that Einstein formulated his re-
quirement that the matter distribution fully determines the metric only in 1917; in 1913 Einstein
still thought of Mach’s principle in the light of the relativity of inertia. Furthermore, Einstein
(1914) explicitly introduced the final version of the hole argument in terms of a conflict between
general covariance and the “law of causality” (“Kausalgesetz ”), which was contemporary parlance
for determinism. In sum, it seems safe to say, with Janssen (2007), that the ‘worries about deter-
minism and causality that are behind Einstein’s hole argument have strong Machian overtones.’
See Norton (1993) for Einstein’s general struggle with general covariance, and its aftermath.
6 ψ
We say that (M ′ , g ′ ) → (M, g), where ψ : M ′ → M is a diffeomorphism, is an isometry iff
g = ψ ∗ g (in particular, following e.g. Hawking & Ellis, 1973, we always take an isometry to be a
′

diffeomorphism). Now simply take M ′ = M and g ′ = ψ ∗ g. See Weatherall (2018) and Halvorson
& Manchak (2022) for the meaning of this for gr and for what the alternatives could (not) be.

3
 The basic principles of general relativity—as encompassed in the term ‘the
principle of general covariance’ (and also ‘principle of equivalence’)—tell us
that there is no natural way to identify the points of one space-time with
corresponding spacetime points of another. (Penrose, 1996, p. 591)7

There are also philosophical arguments against such trans-world identifications, see
e.g. Lewis (1986) and, in connection with the Hole Argument, Butterfield (1988,
1989). Such arguments may be convincing, but remarkably, Weatherall (2018)
mainly appeals to mathematical practice, according to which two Lorentzian mani-
folds (M, g) and (M ′ , g ′) may only be “compared” using isometries (or so he claims).
In particular, if M ′ = M and g are given, then putting a metric ψ ∗ g on M next
to g amounts to the simultaneous use of a (hole) diffeomorphism ψ : M → M, which
is used to pull back the metric, and the identity map idX : M → M, which is used to
compare ψ ∗ g and g and in particular express the key to the Hole Argument, namely

g ′ (x) ≡ ψ ∗ g(x) 6= g(x),

unless, of course, ψ is an isometry, in which case the Hole Argument would be void.8
Such double talk, then, is supposed to be at best confused and at worst illegal.
Wheatherall’s argument is categorical: he decides to only reason in the category
Lor that has Lorentzian manifolds as objects and isometries as maps (or at least as
isomorphisms).9 Or, using a logical language, one uses model theory (and only that),
where Lorentzian manifolds are models and isometries are isomorphisms thereof.
The specific nature of the individual objects or models cannot be used.
But this is not the only valid way of doing mathematics. Much of differential
geometry could be thrown away if one is not allowed to compute the pullback of
a metric under a diffeomorphism; even the very definition of an isometry rests on
the ability to say whether or not ψ ∗ g(x) equals g(x), at each x ∈ M. The usual
definition of the action of a diffeomorphism on a tensor (field) would be a “category
mistake”, and with it, the Lie derivative (see also Gomes, 2021, §2.4). All of the
coordinate-based definitions of tensors used in the past by Einstein (and even by
mathematicians like Ricci and Levi-Civita) would have to go, and so it seems no
accident that not only Earman & Norton’s but also Einstein’s original version of
the Hole Argument is deemed suspicious. Even existence proofs of a Riemannnian
metric on a (paracompact) manifold (typically through an explicit construction using
partitions of unity) would be suspicious in a purely categorical framework. Et cetera.
7
Taken from the penultimate version of Gomes (2021); omitted, alas, from the final one.
8
The emphasis Halvorson & Manchak (2022) put in this context on their otherwise highly
valuable Theorem 1 (see footnote 19) seems like flogging a dead horse. This theorem implies that
a hole diffeomorphism of the kind envisaged by Einstein (1914) and Earman& Norton (1987), and
explicitly constructed by Muller (1995), cannot be an isometry (which, or so it is suggested, would
be the only remaining hope for the Hole Argument to work, accepting Wheatherall’s critique).
But if it were, then ψ ∗ g = g all across M and the dilemma of having both (M, g) and (M, ψ ∗ g)
as models with the same matter distribution or other initial data simply would not arise: both
(naive) determinism and substantivalism would be safe in gr: the Hole Argument would be a dud.
9
In view of Theorem 2 below, within such reasoning one should optimally work in the category
ST of spacetimes (see footnote 3), whose isomorphism are isometries preserving time orientation.

4
 However, even accepting the claim that the only valid comparison maps in
Lorentzian geometry are isometries, the Hole Argument may survive, for it actu-
ally uses neither idM seen as a map from (M, ψ ∗ g) to (M, g), nor ψ seen as a map
from (M, g) to (M, g), both of which indeed fail to be isometries (unless of course ψ
from (M, g) to (M, g) is an isometry). Instead, as is clear from the main text, the
Hole Argument relies on ψ seen as a map from (M, ψ ∗ g) to (M, g), which is surely
an isometry, cf. footnote 6. The complaint that (M, ψ ∗ g) should never have been
introduced in the first place seems feeble to me, since any spacetime (M ′ , g ′ ) is an
object in the category Lor, whatever its construction; given (M, g), in making the
identifications M ′ = M and g ′ = ψ ∗ g one might see the (time orientation preserv-
ing) diffeomorphism ψ of M as the proverbial Wittgensteinian ladder, to be thrown
away after it has served its purpose. It should not matter where (M ′ , g ′) = (M, ψ ∗ g)
comes from: one cannot eat one’s cake and have it by on the one hand insisting on
the use of categories like Lor but on the other hand banning all kinds of objects
from them—in fact any object is of the said kind and nothing would be left.
In any case, this discussion, even if it turns out to be misguided, is only included
here to show that the nullification arguments in Weatherall (2018) and Halvorson
& Manchak (2022) are controversial, which seems hardly the case for the Choquet-
Bruhat–Geroch theorem reviewed in §2, and which, I claim, leads to similar con-
clusions as the original Hole Argument(s). If Weatherall c.s. turn out to be correct
after all, then I would see their principles as a consequence of the Hole Argument.
In some sense, discussed e.g. by Stachel (2014), my preferred version of the Hole
Argument in §2 goes back to Hilbert (1917), who gave the first analysis of gr from
a pde point of view.10 The initial-value problem of Einstein’s equations is very in-
volved. But due to the efforts of especially the “French school”, consisting (in direct
lineage of doctoral descent) of Darmois, Lichnerowicz, and Choquet-Bruhat (whose
early papers carry the name Fourès-Bruhat), the abstract situation is well under-
stood now, at least in vacuo and for initial data given on a spacelike hypersurface.11
The culmination of the pde theory is a theorem due to Choquet-Bruhat & Ge-
roch (1969), recalled in §2. This theorem seems the least vulnerable version of
the Hole Argument, in that it has exactly the same philosophical implications and
seems uncontroversial. It also yields the appropriate notion of determinism; namely
existence and uniqueness up to isometry of the very specific geometric intial-value
problem posed by the Einstein equations (which on the one hand is unique to gr but
on the other hand is close to what one expects in classical mathematical physics).
This will be compared with the influential definition of determinism proposed by
Butterfield (1987, 1988, 1989). In §3 I review the Hole Argument for generally co-
variant special relativity, which leads to very similar questions as the one for gr.
These questions will be discussed in the (provisional and speculative) final section.
10
Hilbert addresses the indeterminism of Einstein’s equations, and also refers to Einstein (1914)
in connection with this problem, but does not explicitly relate his analysis to the Lochbetrachtung.
11
See Stachel (1992) and Choquet-Bruhat (2014) for some history, summarized in Landsman
(2021), §1.9. It is in fact more popular nowadays to give initial data for the Einstein equations on
a null hypersurface (Penrose, 1963). See e.g. Klainerman & Nicolò (2003) for a detailed treatment.

5
2 The Choquet-Bruhat–Geroch theorem
The initial-value approach to gr is based on pde-theory and the following ideology:
• All valid assumptions in gr are assumptions about initial data (Σ̃, g̃, k̃).
Such an initial data triple, assumed smooth, is obtained by equipping some 3d
Riemannian manifold (Σ̃, g̃) with a second symmetric tensor k̃ ∈ X(2,0) (Σ̃), i.e. of
the same “kind” as the 3-metric g̃, such that (Σ̃, g̃, k̃) satisfies the vacuum constraints
R̃ − Tr (k̃ 2 ) + Tr (k̃)2 = 0; ˜ j k̃ j − ∇
∇ ˜ i Tr (k̃) = 0. (4)
i

Here R̃ is the Ricci scalar on Σ̃ for the Riemannian metric g̃ and likewise ∇˜ is the
˜
unique Levi-Civita (i.e. metric) connection on Σ̃ determined by g̃ (so that ∇g̃ = 0).
• All valid questions in gr are questions about “the” mghd (M, g, ι) thereof.
Among these questions, the one relevant to the Hole Argument concerns the unique-
ness of (M, g, ι), whence the scare quotes around ‘the’. Roughly speaking, a mghd
(for maximal globally hyperbolic development) of (Σ̃, g̃, k̃) is a maximal spacetime
(M, g) “generated” by these initial data via the Einstein equations, in that
ι : Σ̃ ֒→ M
injects Σ̃ into M as a “time slice” on which the 4-metric g induces the given 3-metric
g̃ and extrinsic curvature k̃. In more detail,12 A Cauchy development or globally hy-
perbolic development of given initial data (Σ̃, g̃, k̃) satisfying the constraints (4) is a
triple (M, g, ι), where (M, g) is a spacetime that solves the vacuum Einstein equa-
tions Rµν = 0 and ι is an injection making ι(Σ̃) a spacelike Cauchy (hyper)surface
in M such that g induces these initial data on ι(Σ̃) ∼ = Σ̃, i.e. g̃ = ι∗ g is the metric
and k̃ is the extrinsic curvature of Σ̃, induced by the embedding ι and the 4-metric
g.13 It follows that (M, g) is globally hyperbolic, since it has a Cauchy surface.14
This formulation of the (spatial) initial-value problem for the (vacuum) Einstein
equations was an achievement by itself. In particular, it cleverly circumvents the
vicious circle one ends in by trying to find initial data for an already given spacetime
(solving the Einstein equations); for it is part of the problem to find the latter from
the given initial data, and hence one cannot give say dg/dt at t = 0 as initial data.
However, the main achievement concerns the existence and uniqueness of (M, g, ι),
which depends on a suitable notion of maximality (as in the far simpler case of odes,
where in order to guarantee uniqueness the time interval on which the solution is
defined should be maximal). This notion is also non-trivial, and tied to gr. Namely:
12
See also the references in footnote 19, or Landsman (2021), §7.6. Tildes adorn 3d objects.
13
Let N be the unique (necessarily timelike) future-directed normal vector field on ι(Σ̃) such
that gx (Nx , Nx ) = −1. Then k̃(X, Y ) = −g(∇X N, Y ) defines the extrinsic curvature of ι(Σ̃).
14
This procedure by no means excludes the study of non-globally hyperbolic spacetimes in gr,
which in this approach emerge as possible extensions of globally hyperbolic spacetimes (which is
possible even if they are maximal in the above, i.e. globally hyperbolic sense). This is closely
connected to strong cosmic censorship (Penrose, 1979), which in turn is related to a kind of
indeterminism in gr that is outside the scope of the Hole Argument and may occur even if we all
agree that the mghd of given initial data is essentially unique. See e.g. Earman (1995), Doboszewski
(2017, 2020), Smeenk & Wüthrich (2021), or Landsman (2021), Chapter 10, and references therein.

6
 • A maximal Cauchy development or maximal globally hyperbolic development,15
acronym mghd, of given smooth initial data (Σ̃, g̃, k̃), satisfying the constraints
(4), is a Cauchy development (M, g, ι) with the property that for any other
Cauchy development = globally hyperbolic development (M ′ , g ′, ι′ ) of these
same data there exists an embedding ψ : M ′ → M that preserves time orien-
tation, metric, and Cauchy surface as defined by ι, i.e., one has

ψ∗g = g′; ψ ◦ ι′ = ι. (5)

The Hole Argument à la Hilbert (1917) then follows from the straightforward ob-
servation that if (M, g, ι) is a mghd of the initial data (Σ̃, g̃, k̃) and ψ : M ′ → M is
a diffeomorphism (pace Weatherall c.s.!), then the triple (M ′ , g ′, ι′ ), where g ′ and ι′
are defined by (5), i.e. g ′ = ψ ∗ g and ι′ = ψ −1 ◦ ι, with time orientation induced by
ψ,16 is a mghd of the initial data (g̃ ′ , k̃ ′ ) induced on Σ̃ via ι′ and g ′ . In particular:17

Proposition 1. Given some mghd (M, g, ι) of the initial data (Σ̃, g̃, k̃), let U be a
neighbourhood of ι(Σ̃) in M. Take a (time orientation preserving) diffeomorphism
ψ of M that is the identity on U, so that in particular ι′ = ι and (g̃ ′ , k̃ ′ ) = (g̃, k̃).
Then the “Hilbert-triple” (M, ψ ∗ g, ι) is a mghd of the same initial data (Σ̃, g̃, k̃).

This is a decent version of the Hole Argument.18 But since it starts from a diffeo-
morphism ψ of M that only becomes an isometry from (M, ψ ∗ g) to (M, g) “with
hindsight”, it may be vulnerable to the reasoning in Weatherall (2018), Fletcher
(2020), Halvorson & Manchak (2022), etc. My claim is that this is not the case
for the highly nontrivial converse of the reasoning preceding Proposition 1, which
nonetheless poses the same philosophical problems as the original Hole Argument(s).
This converse is the celebrated theorem of Choquet-Bruhat & Geroch (1969):19
15
It might be thought that isometries enter surreptitiously via this definition of maximality, but
this is not the case. The appearance of isometries is a consequence of a local version of Theorem 2:
Any two Cauchy developments (M, g, ι) and (M ′ , g ′ , ι′ ) of the same (smooth) initial data are locally
isometric, in that ι(Σ̃) and ι′ (Σ̃) have open neighbourhoods U and U ′ in M and M ′ , respectively,
such that (U, g) and (U ′ , g ′ ) are isometric through a diffeomorphism ψ : U ′ → U satisfying (5).
See Choquet-Bruhat (2009), Theorem VI.8.4, or Ringström (2009), Theorem 14.3.
16
Defining time orientation by (the equivalence class of) a global timelike vector field T on M ,
so that some causal vector X is future-directed iff g(X, T ) < 0, this means that T ′ = ψ∗−1 T .
17
This construction also works if U = J − (ι(Σ̃)), cf. Curiel (2018) and Pooley (2022).
18
Continuing footnote 4, it is superior to Einstein’s and Earman & Norton’s formulation in that
it has shaken off any implicit reference to Mach’s principle and is closer to the usual initial value
problem for hyperbolic pdes (with a special gr twist though). But it may be weaker as a challenge
to determinism in that the open set on which initial data are given can be made arbitrarily thin.
19
The original source is Choquet-Bruhat & Geroch (1969), who merely sketched a proof (based
on Zorn’s lemma, which they even had to use twice). Even the 800-page textbook by Choquet-
Bruhat (2009) does not contain a proof of the theorem (which is Theorem XII.12.2); the treatment
in Hawking & Ellis (1973), §7.6, is slightly more detailed but far from complete, too. Ringström
(2009) is a book-length exposition, but ironically the proof of Theorem 16.6 is wrong; it is corrected
in Ringström (2013), §23. A constructive proof was given by Sbierski (2016), which is streamlined
and summarized in Landsman (2021), §7.6. Though never mentioned in statements of the theorem,
the isometry ψ is unique. This can be shown by Proposition 3.62 in O’Neill (1983) or the equivalent

7
Theorem 2. For each initial data triple (Σ̃, g̃, k̃) satisfying the constraints (4) there
exists a mghd (M, g, ι). This is unique up to (unique) time-orientation-preserving
isometries fixing the Cauchy surface, i.e. for any other mghd (M ′ , g ′, ι′ ) there exists
an isometry ψ : M ′ → M that preserves time orientation and satisfies ψ ◦ ι′ = ι.

All reference to diffeomorphisms that are not (yet) isometries has gone! And yet
in a sense, this theorem is the Hole Argument, for it forces us to choose between:

1. Determinism, in the precise version that the Einstein equations for given ini-
tial data have a unique solution in the sense that we agree that triples (M, g, ι)
and (M ′ , g ′, ι′ ) as in the statement of the theorem are seen as different repre-
sentatives of the same physical situation (i.e., are “physically identified”).

2. Space-time substantivalism in the sense of denial of Leibniz equivalence. This

denial says that triples (M, g, ι) and (M ′ , g ′ , ι′ ) represent “distinct states of af-
fairs” (although they are observationally indistinguishable). This choice saves
the “this-ness” of points at the cost of accepting some invisible indeterminism.

Or, at least, this is the dilemma Earman & Norton (1987), or even Einstein (1914),
left us with on the basis of their own versions of the argument. Most philosophical
discussions of this dilemma, including more precise formulations thereof (e.g. But-
terfield, 1989; Pooley, 2022) or dismissals (e.g. Curiel, 2018), remain relevant if we
replace the controversial earlier versions of the Hole Argument by Theorem 2. This
is the sense in which the Hole Argument remains alive; which is all I wish to argue.
If we opt for determinism, the specific version thereof in gr that seems enforced
by Theorem 2 is that we must “physically identify” all maximal globally hyper-
bolic spacetimes (M, g, ι) with Cauchy surface ι(Σ̃) that carry fixed (and a priori
“timeless”) initial data (Σ̃, g̃, k̃). Theorem 2 states that all putatively different pos-
sibilities are isometric, and hence isometries (preserving ι) play the role of gauge
symmetries.20 This may be unsurprising, since the isometries in Theorem 2 are a
shadow of the diffeomorphism invariance of the Einstein equations. But it is also
somewhat surprising, since the isometries of a fixed spacetime (M, g) are not given
by freely specifiable functions on M, as in the case of gauge theories.21 See also §3.
It may be interesting to compare the notion of determinism in gr provided for
free by Theorem 2 to some others that have been used in the literature on the Hole
Argument. To facilitate this, here is a somewhat awkward weakening of Theorem 2:
argument in footnote 639 of Landsman (2021), to the effect that an isometry ψ is determined at
least locally (i.e. in a convex nbhd of x) by its tangent map ψx′ at some fixed x ∈ M ′ . Take x ∈ ι′ (Σ̃).
Since ψ in Theorem 2 is fixed all along ι′ (Σ̃) by the second condition in (5) and since it also fixes
the (future-directed) normal Nx to ι′ (Σ̃) by the first condition in (5), it is determined locally.
Theorem 1 in Halvorson & Manchak (2022) then applies, which is a rigidity theorem for isometries
going back at least to Geroch (1969), Appendix A (as Halvorson & Manchak acknowledge).
20
See Gomes (2021) for a detailed analysis of the relationship between gauge symmetries in
gauge theories and diffeomorphisms in gr, including a discussion of the Hole Argument.
21
If dim(M ) = n, then for any semi-Riemannian metric g the isometry group of (M, g) is at most
1
2 n(n + 1)-dimensional. See O’Neill (1983), Lemma 9.28; Kobayashi & Nomizu (1963), Theorem
VI.3.3 does the Riemannian case. Thus the Poincaré-group in n = 4 has maximal dimension 10.

8
Corollary 3. If two globally hyperbolic spacetimes (M, g) and (M ′ , g ′) contain Cauchy
surfaces Σ̃ ⊂ M and Σ̃′ ⊂ M ′ , respectively, which carry initial data (Σ̃, g̃, k̃) and
(Σ̃′ , g̃ ′ , k̃ ′ ) induced by the 4-metrics g and g ′ on M and M ′ , respectively, where both
(M, g) and (M ′ , g ′) are maximal for these initial data, and there is a 3-diffeomor-
phism α : Σ̃ → Σ̃′ such that g̃ = α∗ g̃ ′ and k̃ = α∗ k̃ ′ , then there exists an isometry
ψ : M ′ → M that preserves time orientation and reduces to α on Σ̃.
Recall that an isometry is always a diffeomorphism. This corollary is weaker than
Theorem 2, for it lacks the existence claim of (M, g) and (M ′ , g ′), which are now
taken as given. We mention this corollary because it relates to an influential notion
Dm2 of determinism introduced in this context by Butterfield (1987, 1988, 1989):
A theory with models hM, Oi i is S-deterministic, where S is a kind of region
that occurs in manifolds of the kind occurring in the models, iff: given any
two models hM, Oi i and hM ′ , Oi′ i containing regions S and S ′ of kind S, re-
spectively, and any diffeomorphism α from S onto S ′ : if α∗ (Oi′ ) = α(Oi ) on
α(S) = S ′ , then: there is an isomorphism β from M onto M ′ that sends S to
S ′ , i.e. β ∗ Oi′ = Oi throughout M ′ and β(S) = S ′ .
(Butterfield, 1987, p. 29; 1989, p. 9).22

To cover gr, this definition should be amended by: firstly adding the extrinsic cur-
vature to the initial data induced on S and S ′ ; secondly specializing to globally
hyperbolic solutions to the vacuum Einstein equations; and finally, adding a maxi-
mality condition on M and M ′ , as in Corollary 3. In that case, gr is deterministic
by Theorem 2. But this theorem does more: whereas Dm2-like definitions assume
the existence of the spacetimes in question, Theorem 2 includes an existence proof.23
This closes our discussion of the Choquet-Bruhat–Geroch theorem, at least for
the moment. We now turn to a version of this theorem for special relativity.24
22
Butterfield (1987, 1989) contrasts Dm2 with a Laplacian kind of definition of determinism
Dm1 he attributes to Montague and Earman: ‘A theory with models hM, Oi i is S-deterministic,
where S is a kind of region that occurs in manifolds of the kind occurring in the models, iff:
given any two models hM, Oi i and hM ′ , Oi′ i and any diffeomorphism β from M onto M ′ , and any
region S of M of kind S: if β(S) is of kind S and also β ∗ Oi′ = Oi on β(S), then: β ∗ Oi′ = Oi
throughout M .’ If we correct this similarly to Dm2, Butterfield’s point still stands: the Hole
Argument (in any version) shows that gr violates Dm1. See also Pooley (2022) for a detailed
analysis of similar definitions. Pooley’s version of Dm2 is a bit more general and also applies to
gr: ‘Theory T is deterministic just in case, for any worlds W and W ′ that are possible according
to T , if the past of W up to some timeslice in W is qualitatively identical to the past of W up to
some timeslice in W ′ , then W and W ′ are qualitatively identical.’ Apart from my complaint that
also this definition assumes the existence of W and W ′ (instead of proving it), a definition like this
requires a sub-definition of what is meant by ‘qualitative’, which Theorem 2 also takes care of.
23
Similarly, an amended version of Property R introduced by Halvorson & Manchak (2022)
would state that if (M, g) and (M ′ , g ′ ) are two maximal globally hyperbolic spacetimes solving the
vacuum Einstein equations, and if two time-orientation preserving isometries ψ : M ′ → M and
ϕ : M ′ → M coincide on the causal pasts J − (Σ̃) of some Cauchy surface Σ̃ ⊂ M , then ψ = ϕ
altogether. This definition is compatible with gr because of the uniqueness of the isomorphism ψ
in Theorem 2 (see footnote 19). But like Butterfield’s definition Dm2, Halvorson & Manchak’s
Property R assumes the existence of models and then makes some uniqueness claim about it.
24
The following section was inspired by correspondence with Henrique Gomes and Hans Halvor-
son, who specifically proposed to look at special relativity in this context.

9
3 The case of generally covariant special relativity
It may be instructive to compare Theorem 2 with the corresponding situation in
special relativity, perhaps unusually seen as a generally covariant field theory à la
vacuum gr, but this time with field equation Rρσµν = 0 instead of Rµν = 0. The
initial value problem may then be posed in almost the same way as in gr;25 the
only difference is that the initial data triple (Σ̃, g̃, k̃) now satisfes the constraints
R̃ijkl − k̃il k̃jk + k̃ik k̃jl = 0; ˜ i k̃jk − ∇
∇ ˜ j k̃ik = 0. (6)
The constraints (6) are stronger than (4), which follows from (6) by contracting
with g̃ ik g̃ jl and g̃ ik , respectively. The reason is that in gr one merely asks for an
embedding of the initial data in a Ricci-flat Lorentzian manifold (M, g), whereas in
special relativity (M, g) is (locally) flat altogether, i.e. Rρσµν = 0. To avoid irrelevant
global topological issues (interesting as these might be in a different context), we
assume that Σ̃ is diffeomorphic to R3 . By the splitting theorem of Geroch (1970) as
improved by Bernal and Sánchez (2003), global hyperbolicity of (M, g) then gives
M∼
= R × Σ̃ = R4 (7)
diffeomorphically. Without loss of generality we may actually take M = R4 , so
that,26 even without using (6) yet, (M, g) must be Minkowski spacetime (R4 , η) ≡ M.
As a solution to the Einstein equations, M is not only globally hyperbolic, but
also maximal.27 Using (6), the Minkowskian version of the so-called fundamental
theorem for hypersurfaces (Kobayashi & Nomizu, 1969, Theorem VII.7.2) yields:28
Theorem 4. For each initial data triple (R3 , g̃, k̃) satisfying the constraints (6)
there exists an isometric embedding ι : R3 → R4 carrying the Minkowski metric,
whose extrinsic curvature is the given tensor k̃. This embedding is unique up to
Poincaré transformations preserving time-orientation, in the sense that for any other
map ι′ : R3 → R4 with the same properties as ι there exists a (unique) Poincaré
transformation ψ : R4 → R4 that preserves time-orientation and satisfies ψ ◦ ι′ = ι.
25
This upsets the idea that special relativity uses only linear subspaces of space-time as hy-
persurfaces of simultaneity whereas general relativity uses general curved surfaces, but already
Schwinger (1948) employed arbitrary initial data surfaces in relativistic quantum field theory.
26
This follows from the fundamental theorem of (semi) Riemannian geometry, which states that
(M, g) is locally flat iff its Riemann tensor vanishes. See e.g. Landsman (2021), Theorem 4.1.
27
Maximality of Minkowski spacetime follows from its inextendibility; see e.g. Corollary 13.37
in O’Neill (1983) for the smooth case and Sbierski (2018) for inextendibility even in C 0 .
28
See also Landsman (2021), Theorem 4.18, for the fundamental theorem for hypersurfaces. This
theorem is concerned with embeddings of curved surfaces with prescribed second fundamental form
into Euclidean space and goes back to the nineteenth century. The proof of the Minkowskian case
is practically the same, up to some sign changes: in the Euclidean case the first constraint in (6)
is R̃ijkl + k̃il k̃jk − k̃ik k̃jl , the sign changes going back to the different signs in the Gauss–Codazzi
equations in Euclidean and Lorentzian signature, see e.g. eqs. (4.147) - (4.148) in §4.7 in Landsman
(2021). These sign changes do affect the outcome. For example, Hilbert (1901) proved that it is
impossible to isometrically embed two-dimensional hyperbolic space (H 2 , gH ) in R3 with Euclidean
metric. But hyperbolic space can be isometrically embedded in R3 with Minkowski metric, cf. e.g.
Landsman (2021), §4.4. Hence given (H 2 , gH ), a symmetric tensor k̃ such that (gH , k̃) satisfy the
Euclidean constraint do not exist, but such a k̃ can be found satisfying the Minkowski constraints.

10
 There is a clear conceptual analogy between Theorems 2 and 4, except that the
former refers to the initial-value problem in general relativity, whilst the latter states
the situation in special relativity (albeit in a somewhat unusual way). In particular,
the role of isometries in the general theory is now played by Poincaré transformations
(i.e. isometries of the Minkowski metric), as was to be expected. And yet, whereas
most physicists would be happy to regard isometries in general relativity as gauge
symmetries akin to coordinate transformations, few if any would regard Poincaré
transformations as physically inert. But in Theorem 4, they are. In a more general
context, this is explained by Gomes (2021), partly reflecting on Belot (2018):
But some familiar symmetries of the whole Universe, such as velocity boosts
in classical or relativistic mechanics (Galilean or Lorentz transformations),
have a direct empirical significance when applied solely to subsystems. Thus
Galileo’s famous thought-experiment about the ship—that a process involving
some set of relevant physical quantities in the cabin below decks proceeds in
exactly the same way whether or not the ship is moving uniformly relative
to the shore—shows that subsystem boosts have a direct, albeit relational,
empirical significance. For though the inertial state of motion of the ship is
undetectable to experimenters confined to the cabin, yet the entire system,
composed of ship and sea registers the difference between two such motions,
namely in the different relative velocities of the ship to the water.
(Gomes 2021, p. 150)
In other words, in thinking about Poincaré transformations as bringing physical
change, as for example in boosts of Galieli’s ship or Einstein’s train, we apply such
transformations to subsystems of the universe. But Theorem 4 concerns the action
of Poincaré transformations on space-time as a whole. See also Wallace (2021).
This brings us back to the substantivalism versus relationalism debate (Earman,
1989; Pooley, 2013); and indeed I see little difference between general and special
relativity in this context. The difference between the former and the latter is merely
the one between Theorems 2 and 4, respectively, which are good starting points for
this debate. Whatever differences there are seem technical rather than conceptual to
me, like the underlying difference between the field equations Rµν = 0 and Rρσµν = 0.
Finally, to some extent Theorems 2 and 4 are reminiscent of the spontaneous
breakdown of gauge symmetry through the Higgs mechanism.29 Here, in order
to settle into a minimum of the Higgs potential, the Higgs field ϕ must “choose”
a point ϕc on a circle as its “frozen” vacuum value. The global U(1) symmetry
involved in this choice is a finite (in fact one) dimensional shadow of the original
infinite-dimensional gauge symmetry of the theory. Different choices of ϕc yield
phenomenologically indistinguishable worlds and the analogy is between moving the
vacuum value ϕc around on a circle and moving a spacetime (M, g) around in its orbit
under its isometry group.30 Note that also here we are talking about symmetries
of the universe as a whole, which is what makes them unobservable; the situation
changes completely if different domains in the universe have different values of ϕc .
29
See Struyve 2011, Landsman (2017), §10.10, or any book on the Standard Model.
30
This analogy is admittedly weak, since Theorem 2 involves both the embedding maps ι and the
possibility that isometries move a given spacetime (M, g) to one (M ′ , g ′ ) with a different underlying
(but diffeomorphic) manifold M , neither of which have a counterpart in the Higgs mechanism.

11
4 Confronting the Hole Argument
Despite their denial of the Hole Argument, Weatherall (2018) and Halvorson &
Manchak (2022) make some of the most useful comments towards its resolution:
Mathematical models of a physical theory are only defined up to isomorphism,
where the standard of isomorphism is given by the mathematical theory of
whatever mathematical objects the theory takes as its models. One conse-
quence of this view is that isomorphic mathematical models in physics should
be taken to have the same representational capacities. By this I mean that if
a particular mathematical model may be used to represent a given physical
situation, then any isomorphic model may be used to represent that situation
equally well. Note that this does not commit me to the view that equivalence
classes of isomorphic models are somehow in one-to-one correspondence with
distinct physical situations. But it does imply that if two isomorphic models
may be used to represent two distinct physical situations, then each of those
models individually may be used to represent both situations.
(Weatherall, 2018, pp. 331–332)
Why is it, then, that there has been, and will surely continue to be, a feeling
that there is some remaining open question about whether general relativity
is fully deterministic? Our conjecture is that the worry here arises from the
fact that general relativity, just like any other theory of contemporary math-
ematical physics, allows its user a degree of representational freedom, and
consequently displays a kind of trivial semantic indeterminism: how things
are represented at one time does not constrain how things must be represented
at later times. (Halvorson & Manchak, 2022, p. 19)

These comments could just as well have been made about Theorem 2, which by
itself already makes it worth delving into the idea of “representational freedom”.31
In particular, the Hole Argument (if it is correct) and Theorem 2 give us a choice
between two positions in the philosophy of mathematics that are traditionally seen as
opposites, namely a Hilbert-style structuralism and a Frege-style abstractionism.32
31
See also Belot (2018), Fletcher (2020), Gomes (2021), Luc (2022), and Pooley (2022).
32
See e.g. Hallett (2010), Ebert & Rossberg (2016), Mancosu (2016), Blanchette (2018), Hellman
& Shapiro (2019), and Reck & Schiemer (2020). Historically, Frege’s abstractionism served his
higher goal of logicism, but the former stands on its own and can be separated from the latter. It
may be objected that the heart of the Frege–Hilbert opposition does not lie in abstractionism versus
structuralism but in differences about the nature of mathematical axioms, definitions, elucidations,
and existence, and in particular about Frege’s insistence that every mathematical concept (such
as “point” or “line”) be defined on its own through reference, against Hilbert’s revolutionary
idea of implicit and “holistic” definition of concepts through an entire axiom system in which
they occur. But these issues are closely related. For example, Hilbert’s contextual and relational
way of defining concepts naturally implies that whatever makes them concrete is given only up to
isomorphism. Abstractionism of the kind considered here arguably goes back to Aristotle, since the
kind of equivalence relation lying at the basis of Frege’s abstraction principle is typically obtained
by Aristotle’s procedure of abstraction by deletion (Mendell, 2019). For example, a mathematician
sees a bronze sphere as a sphere, deleting its bronzeness. Also in so far as Hilbert famously claimed
that mathematical objects exist as soon as the axioms through which they are implicitly defined
are consistent (leaving their precise manner of existence in the dark, like Plato), the Frege–Hilbert
opposition has its roots in the Aristotle–Plato one (Bostock, 2009).

12
 • Structuralism: spacetimes (with fixed initial data) are mathematical structures
which by their very nature can only be studied up to isomorphism. Since
isometry is the pertinent notion of isomorphism, the identitifcation of isometric
spacetimes called for by the Hole Argument or Theorem 2 was to be expected.
• Abstractionism: the relevant mathematical object is the equivalence class of
all spacetimes (with fixed initial data) up to isometry. Quoting Wilson (2010):
Appeals to equivalence classes will seem quite natural if one regards the
novel elements as formed by conceptual abstraction in a traditional philo-
sophical mode: one first surveys a range of concrete objects and then
abstracts their salient commonalities. (Wilson, 2010, p. 395)

In the case at hand, the ‘salient commonalities’ seem to be the property that
all members of a given equivalence class satisfy the vacuum Einstein equations
with identical initial data. In the spirit of the abstractionist programme, this
commonality may be expressed by the function f from the class of all triples
(M, g, ι) to the class of all triples (Σ̃, g̃, k̃) that maps (M, g, ι) to the initial
data it induces on ι(Σ) ⊂ M, where it is assumed that each (M, g, ι) is a
maximal globally hyperbolic space-times with given Cauchy surface ι(Σ̃).
These two options are put in perspective by the opening quote of Benaceraff (1965):
The attention of the mathematician focuses primarily upon mathematical
structure, and his intellectual delight arises (in part) from seeing that a given
theory exhibits such and such a structure, from seeing how one structure is
“modelled” in another, or in exhibiting some new structure and showing how
it relates to previously studied ones . . . But . . . the mathematician is satisfied
so long as he has some “entities” or “objects” (or “sets” or “numbers” or
“functions” or “spaces” or “points”) to work with, and he does not inquire
into their inner character or ontological status.
The philosophical logician, on the other hand, is more sensitive to matters of
ontology and will be especially interested in the kind or kinds of entities there
are actually . . . He will not be satisfied with being told merely that such and
such entities exhibit such and such a mathematical structure. He will wish
to inquire more deeply into what these entities are, how they relate to other
entities . . . Also he will wish to ask whether the entity dealt with is sui generis
or whether it is in some sense reducible to (or constructible in terms of) other,
perhaps more fundamental entities.
—R.M. Martin, Intension and Decision
Against abstractionism (both in the context of the Hole Argument and in Frege’s
original application to the definition of Number), one may claim extravagance by
noting that an equivalence class [x] with respect to any equivalence relation ∼ on
some given set X is typically huge;33 no theoretical or mathematical physicist ever
works with such equivalence classes of spacetimes, or even a tiny fraction of it.34
33
Recall that an equivalence class [x] ⊂ X consists of all y ∈ X such that y ∼ x
34
See Gomes (2021) for an analysis of physical practice, which in the context of gauge theories
and gr amounts to the smart choice of cross-sections of the canonical projection from X to X/ ∼.

13
In practice, one picks some representative (M, g, ι), from which one may switch to an
equivalent triple (M ′ , g ′, ι′ ) now and then, but one never uses the entire equivalence
class. And yet it is, strictly speaking, the entire equivalence class that Frege would
invoke in order to obtain a proper definition or reference of the word “spacetime”
(provided the analogy with his definition of natural numbers is valid). See also
Benaceraff (1965). To resolve this, one might try to work with the single object
(Σ̃, g̃, k̃), i.e. the initial data that give rise to all of these isometric spacetimes, but
no one does this either; all actual work in gr is done in terms of just a few of the
triples (M, g, ι), whose choice (within its isometry class) is made for convenience.
If instead we go for a structuralist resolution, seemingly incompatible philosoph-
ical points of view remain possible. For example, fixing a triple (M, g, ι) within its
equivalence class, Newton may be perfectly right in thinking of elements x ∈ M as
points in spacetime, which carry a metric g(x) as a secondary quality. But Gelfand
may be equally justified in regarding points of M as nonzero multiplicative linear
functionals C ∞ (M) → R, which by definition carry fields as a primary quality.35
Moreover, within mathematical structuralism, the Hole Argument seems com-
patible with both structural realism (Ladyman, 2020) and empiricist structuralism
(van Fraassen, 2008); in the former, the structures in question are so to speak parts
of reality whereas in the latter they model empirical phenomena. Let me quote:
1. Science represents the empirical phenomena as embeddable in certain
abstract structures (theoretical models).
2. Those abstract structures are describable only up to structural isomor-
phism.
(. . . ) How can we answer the question of how a theory or model relates
to the phenomena by pointing to a relation between theoretical and data
models, both of them abstract entities? The answer has to be that the data
model represent the phenomena; but why does that not just push the problem
[namely: what is the relation between the data and the phenomena it models]
one step back? The short answer is this:

construction of a data model is precisely the selective relevant de-

piction of the phenomena by the user of the theory required for the
possibility of representation of the phenomenon.
(van Fraassen, 2008, pp. 238, 253)

This last comment seems to describe the practice of physicists and mathematicians
working in gr: some user of the theory chooses a member (M, g, ι) of its equivalence
class, whilst some other user (or even the same one) may pick another member.36
In conclusion, empiricist structuralism seems to have strong cards in confronting
the Hole Argument (in both its original versions or rephrased as Theorem 2): it does
not suffer from the abstractionist extravaganza and warrants scientific practice.
35
Gelfand then uses an isometric triple (M ′ , g ′ , ι′ ), where M ′ consists of the said functionals.
36
Van Fraassen’s emphasis on the user also explains why say Kerr spacetime, even with fixed
parameters m and a, can be used to describe different black holes, despite the mathematical identity
of the two models. Indeed, one user models the phenomena produced by one black hole, whilst
another user uses (!) the “spacetime” in question to model the phenomena produced by another.

14
References
[1] Arledge, C., Rynasiewicz, R. (2019). On some recent attempted non-metaphysical
dissolutions of the hole dilemma. http://philsci-archive.pitt.edu/16343/.

[2] Belot, G. (2018). Fifty million Elvis fans can’t be wrong. Nous 52, 946–981.

[3] Benaceraff, P. (1965). What numbers could not be. Philosophical Review 74, 47–73.

[4] Bernal, A. N., Sánchez, M. (2003). On smooth Cauchy hypersurfaces and Geroch’s
splitting theorem. Communications in Mathematical Physics 243, 461–470.

[5] Blanchette, P. (2018). The Frege–Hilbert controversy. Stanford

Encyclopedia of Philosophy (Fall 2018 Edition), ed. Zalta, E.N.
https://plato.stanford.edu/archives/fall2018/entries/frege-hilbert/.

[6] Bostock, D. (2009). Philosophy of Mathematics (Wiley).

[7] Bradley, C., Weatherall, J.O. (2021) Mathematical responses to the Hole Argument:
Then and now. http://philsci-archive.pitt.edu/19703/.

[8] Butterfield, J. (1987). Substantivalism and determinism. International Studies in the

Philosophy of Science 2, 10–32.

[9] Butterfield, J. (1988). Albert Einstein meets David Lewis. PSA 1988, 65–81.

[10] Butterfield, J. (1989). The hole truth. British Journal for the Philosophy of Science
40, 1–28.

[11] Choquet-Bruhat, Y. (2009). General Relativity and the Einstein Equations (Oxford
University Press).

[12] Choquet-Bruhat, Y. (2014). Beginnings of the Cauchy problem.

https://arxiv.org/abs/1410.3490.

[13] Choquet-Bruhat, Y., Geroch, R. (1969). Global aspects of the Cauchy problem in
general relativity. Communications in Mathematical Physics 14, 329–335.

[14] Curiel, E. (2018). On the existence of spacetime structure. British Journal for the
Philosophy of Science 69, 447–483.

[15] Doboszewski, J. (2017). Non-uniquely extendible maximal globally hyperbolic space-

times in classical general relativity: A philosophical survey. European Studies in Phi-
losophy of Science 6, 193– 212.

[16] Doboszewski, J. (2020). Epistemic holes and determinism in classical general relativ-
ity. British Journal for the Philosophy of Science 71, 1093–1111.

[17] Earman, J. (1989). World Enough and Space-Time: Absolute versus Relational The-
ories of Space and Time (The MIT Press).

[18] Earman, J. (1995). Bangs, Crunches, Whimpers, and Shrieks: Singularities and
Acausalities in Relativistic Spacetimes (Oxford University Press).

15
[19] Earman, J., Norton, J.D. (1987). What price substantivalism? The hole story. British
Journal for the Philosophy of Science 9, 251–278.

[20] Ebert, P., Rossberg, M., eds. (2016). Abstractionism: Essays in Philosophy of Math-
ematics (Oxford University Press).

[21] Einstein, A. (1914). Die formale Grundlage der allgemeinen Relativitätstheorie.

Sitzungsberichte der Königlich Preußischen Akademie der Wissenschaften (Berlin)
1030–1085. https://einsteinpapers.press.princeton.edu/vol6-doc/100.

[22] Fletcher, S.C. (2020). On representational capacities, with an application to general

relativity. Foundations of Physics 50, 228–249.

[23] Geroch, R. (1969). Limits of spacetimes. Communications in Mathematical Physics

13, 180–193.

[24] Geroch, R. (1970). Domain of dependence. Journal of Mathematical Physics 11, 437–
449.

[25] Gomes, H. (2021). Why gauge? Conceptual Aspects of Gauge Theories. PhD Thesis,
University of Cambridge. https://arxiv.org/abs/2203.05339.

[26] Hallett, M. (2010). Frege and Hilbert. Cambridge Companion to Frege, eds. Potter,
M., Ricketts, T., pp. 413–464 (Cambridge University Press).

[27] Halvorson, H., Manchak, J.B. (2022). Closing the Hole Argument. British Journal for
the Philosophy of Science, in press. http://philsci-archive.pitt.edu/19790/.

[28] Hawking, S.W., Ellis, G.F.R. (1973). The Large Scale Structure of Space-Time (Cam-
bridge University Press).

[29] Hellman, G., Shapiro, S. (2019). Mathematical Structuralism (Cambridge University

Press).

[30] Hoefer, C. (1994). Einstein’s struggle for a Machian gravitation theory. Studies in
History and Philosophy of Science 25, 287–335.

[31] Hilbert, D. (1917). Die Grundlagen der Physik (Zweite Mitteilung). Nachrichten
von der Königlichen Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-
Physikalische Klasse, 53–76.

[32] Janssen, M. (2007). What did Einstein know and when did He know it? A Besso
memo dated August 1913. The Genesis of General Relativity, Volume 2, ed. Renn,
J., pp. 787–837 (Springer).

[33] Janssen, M., Renn, J. (2022). How Einstein Found His Field Equations: Sources and
Interpretation (Springer).

[34] Klainerman, S., Nicolò, F. (2003). The Evolution Problem in General Relativity
(Birkhäuser).

[35] Kobayashi, S., Nomizu, K. (1963). Foundations of Differential Geometry. I. (Wiley).

[36] Kobayashi, S., Nomizu, K. (1969). Foundations of Differential Geometry. II. (Wiley).

16
[37] Ladyman, J. (2020). Structural Realism. Stanford Encyclo-
pedia of Philosophy (Winter 2020 Edition), ed. Zalta, E.N.
https://plato.stanford.edu/archives/win2020/entries/structural-realism/.

[38] Landsman, K. (2017). Foundations of Quantum Theory: From Clas-

sical Concepts to Operator Algebras (Springer). Open Access from
https://link.springer.com/book/10.1007/978-3-319-51777-3.

[39] Landsman, K. (2021). Foundations of General Relativity: From Ein-

stein to Black Holes (Radboud University Press). Open Access from
https://radbouduniversitypress.nl/site/books/m/10.54195/EFVF4478/.

[40] Lewis, D. (1986). On the Plurality of Worlds (Blackwell).

[41] Luc, J. (2022). Arguments from scientific practice in the debate about the physical
equivalence of symmetry-related models. Synthese 200, article number 72, in press.
https://doi.org/10.1007/s11229-022-03618-w.

[42] Mancosu, P. (2016). Abstraction and Infinity (Oxford University Press).

[43] Maudlin, T. (1990). Substances and spacetime: What Aristotle would have said to
Einstein. Studies in History and Philosophy of Science 21, 531–561.

[44] Mendell, H. (2019). Aristotle and Mathematics. Stanford Encyclopedia of Philoso-

phy (Fall 2019 Edition), ed. Zalta, E.N. https://plato.stanford.edu/archives/
fall2019/entries/aristotle-mathematics.

[45] Muller, F.A. (1995). Fixing a hole. Foundations of Physics Letters 8, 549–562.

[46] Norton, J.D. (1993). General covariance and the foundations of general relativity:
Eight decades of dispute. Reports on Progress in Physics 56, 791–858.

[47] Norton, J.D. (2019). The Hole Argument. Stanford Encyclo-

pedia of Philosophy (Summer 2019 Edition), ed. Zalta, E.N.
https://plato.stanford.edu/archives/sum2019/entries/spacetime-holearg/.

[48] O’Neill, B. (1983). Semi-Riemannian Geometry (Academic Press).

[49] Penrose, R. (1963). Null hypersurface initial data for classical fields of arbitrary spin
and for general relativity. Aerospace Research Laboratories 63–65. Reprinted in Gen-
eral Relativity and Gravitation 12, 225–264 (1980).

[50] Penrose, R. (1979). Singularities and time-asymmetry. General Relativity: An Ein-

stein Centenary Survey, eds. Hawking, S.W., Israel, W., pp. 581–638 (Cambridge
University Press).

[51] Penrose, R. (1996). On gravity’s role in quantum state reduction. General Relativity
& Gravitation 28, 581–600.

[52] Pooley, O. (2013). Substantivalist and relationalist approaches to spacetime. Oxford

Handbook of Philosophy of Physics, ed. R. Batterman, chapter 16, (Oxford University
Press). http://philsci-archive.pitt.edu/9055/.

17
[53] Pooley, O. (2022). The hole argument. Routledge Companion to Philoso-
phy of Physics, eds. Knox, E., Wilson, A., chapter 10 (Taylor & Francis).
http://philsci-archive.pitt.edu/18142/.

[54] Pooley, O., Read, J. (2021). On the mathematics and metaphysics of the
Hole Argument. British Journal for the Philosophy of Science, in press.
http://philsci-archive.pitt.edu/19774/.

[55] Reck, E., Schiemer, G. (2020). Structuralism in the philosophy of mathemat-

ics. Stanford Encyclopedia of Philosophy (Spring 2020 Edition), ed. Zalta, E.N.
https:plato.stanford.edu/archives/spr2020/entries/structuralism-mathematics/.

[56] Ringström, H. (2009). The Cauchy Problem in General Relativity (EMS).

[57] Ringström, H. (2013). On the Topology and Future Stability of the Universe (EMS).

[58] Roberts, B.W. (2020). Regarding ‘Leibniz equivalence’. Foundations of Physics 50,
250–269.

[59] Sbierski, J. (2016). On the existence of a maximal Cauchy development for the Ein-
stein equations: A dezornification. Annales Henri Poincaré 17, 301–329.

[60] Sbierski, J. (2018). The C 0 -inextendibility of the Schwarzschild spacetime and the
spacelike diameter in Lorentzian geometry. Journal of Differential Geometry 108,
319–378.

[61] Schwinger, J. (1948). Quantum electrodynamics. I. A covariant formulation. Physical

Review 74, 1439–1461.

[62] Smeenk, C., Wüthrich, C. (2021). Determinism and general relativity. Philosophy of
Science 88, 638–664.

[63] Stachel, J. (1992). The Cauchy problem in general relativity–The early years. Studies
in the History of General Relativity, eds. Eisenstaedt, J., Kox, A.J., pp. 407–418
(Birkhäuser).

[64] Stachel, J. (2014). The hole argument and some physical and
philosophical implications. Living Reviews in Relativity 17, 1-66.
https://link.springer.com/article/10.12942/lrr-2014-1.

[65] Struyve, W. (2011). Gauge invariant accounts of the Higgs mechanism. Studies in
History and Philosophy of Modern Physics 42, 226–236.

[66] Van Fraassen, B.C. (2008). Scientific Representation (Oxford University Press).

[67] Wallace, D. (2021). Isolated systems and their symmetries, part II: Local and global
symmetries of field theories. http://philsci-archive.pitt.edu/19729/.

[68] Weatherall, J.O. (2018). Regarding the ‘hole argument’. British Journal for the
Philosophy of Science 69, 329–350.

[69] Wilson, M. (2010). Frege’s mathematical setting. Cambridge Companion to Frege,

eds. Potter, M., Ricketts, T., pp. 379–412 (Cambridge University Press).

18