Professional Documents
Culture Documents
Gordon Belot
1
3
Great Clarendon Street, Oxford ox2 6dp
Oxford University Press is a department of the University of Oxford.
It furthers the University's objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
q Gordon Belot 2011
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2011
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data available
Typeset by SPI Publisher Services, Pondicherry, India
Printed in Great Britain
on acid-free paper by
MPG Books Group, Bodmin and King’s Lynn
ISBN: 978–0–19–959532–7
1 3 5 7 9 10 8 6 4 2
To my parents, with love
This page intentionally left blank
Preface
Mark Wilson, and Bas van Fraassen. Very special thanks to the following,
who voluntarily read great chunks of the manuscript and provided many
invaluable suggestions: Anonymous, Caro Brighouse, Jeremy Butterfield,
Craig Callender, Adam Caulton, Laura Ruetsche, Brad Skow, and Ed
Slowik. Most special thanks of all for my fellow inhabitants of Le Vieux
Shack: Grisbi, Stargell, and Honus for their forceful and enigmatic leadership;
and Laura for many, many things.
Contents
Introduction 1
IV Primitivist Approaches 78
1. Introduction 78
2. A First Primitivist Approach 83
3. Geometric Facts for Substantivalists 86
4. Another Primitivist Strategy 90
5. Beyond Metricity 95
6. Summation 100
x contents
Conclusion 134
Appendix A. Simplicity and Ontology 139
Appendix B. Limits of Sequences of Metric Spaces 150
Appendix C. Some Background to the Absolute-Relational Debate 157
Appendix D. Leibniz and Modal Relationalism 173
Appendix E. More on Congruence and Superposability 186
References 199
Index 213
Introduction
1
Bayle describes a sceptical argument against the existence of space in Remark G of the
article on Leucippus in the Historical and Critical Dictionary. The arguments against material
extension in the first part of Remark G of the article on Zeno of Elea would appear to apply
equally against any sort of extension.
2 introduction
between bodies are derivative on the relations between the parts of space
that they occupy. Relationalists are realists about spatial structure who deny
this. Among archetypical relationalists, we might number Aristotle,
Descartes, and Leibniz (although there is some question in some people’s
minds about Descartes).
The above characterization of relationalism is a little involved. Why not
simply take the characteristic difference between substantivalists and rela-
tionalists to be that the former but not the latter include space in their
ontology (or in their fundamental ontology)? Because that would threaten
to obscure the distinction between relationalists and anti-realists about
space. Realists about space can be either relationalists or substantivalists.2
And anti-realists who recognize a sense in which reality can be said to be
spatially organized can adopt either a relationalist or a substantivalist stance
about that sense.3 Consider the case of Leibniz, who is usually treated as
first among equals among relationalists. Leibniz was of course an anti-
realist as well a relationalist. Speaking as an anti-realist, Leibniz denied that,
fundamentally speaking, anything bears a spatial relation to anything
else—and in this mode he would presumably deny sentences asserting
that space is three-dimensional or that it is infinite. But Leibniz was also
willing to speak about the phenomenal world. And in that mode he spoke
as a realist and a relationalist—and was happy to speak of space as a sort of
thing (see the epigraph to this chapter) and would have been willing to
affirm that space is three-dimensional and infinite (see Appendix D
below). So it seems unnatural to say that Leibniz qua relationalist excluded
space from his ontology—he spoke of it as a thing and was willing to
attribute features to it. And while Leibniz did exclude space from his
fundamental ontology, this was because he was an anti-realist, not because
he was a relationalist.
Relationalists and substantivalists can agree (as Leibniz and Newton
did) that space is a thing of some sort and that it has some given geometric
structure. Their disagreement concerns the nature of the existence
of space. This disagreement is reflected in a disagreement over the
truth conditions appropriate for claims about the geometric structure of
space.
2
Aristotle was a realist and a relationalist, Newton a realist and a substantivalist.
3
See e.g. the discussion on pp. 513 f. of Foster, “In Defence of Phenomenalistic Idealism.”
introduction 3
4
See e.g. bk. iv of Aristotle’s Physics or §§ii.16 ff. of Descartes’s Principles of Philosophy.
5
The exceptions being those philosophers who hold that the only metaphysically possible
worlds are the physically possible ones while also denying that worlds of the sort described are
physically possible. They are invited to substitute examples that they like better, if they take
substantivalism and relationalism to be worth discussing at all.
4 introduction
6
For further discussion, see Appendix D below.
introduction 5
Appendix D is more detailed, and argues in favour of the standard ( but not
universally accepted) view that Leibniz was a modal relationalist.
In the course of the discussion of Chapter IV, a certain technical
question arises: In what sort of metric spaces does knowing the distance
relations between a set of points determine the embedding of those points
in the space up to a symmetry of the space? This question is briefly
discussed in Appendix E.
I
Possible Structures of Space
1 Introduction
The goals of this preliminary chapter are primarily of an expository nature:
to get on the table for later use some mathematical notions and facts that
provide a framework for talking about the structure of space at various
possible worlds.
In order to avoid complete tedium, the discussion is structured around
the question: Which mathematical structures should we think of as repre-
senting the spatial structure of metaphysically possible worlds? For short:
What are the possible structures of space?
The question will strike many readers as a silly one. Some will think it
has a determinate answer—but that we have little hope of finding it.
Others will doubt that the question is anywhere near sharp enough to
admit of a determinate answer.
The question is a little silly. But I don’t think it is completely pointless.
Most of us have some intuitions about this question—so our concept of
spatial geometry is not a completely formless one. Here I will in effect be
developing one way of assigning content to this concept. I proceed partly
by exploring our ordinary concept, partly via stipulation.1
I take it to be obvious that Euclidean three-space corresponds to a
possible structure of space. And I think it reasonable to assume the class
1
So often ‘argue,’ ‘argument,’ etc. have invisible scare quotes in this and the next two
sections.
possible structures of space 9
2
Bricker argues for a view like this in “Plenitude of Possible Structures.”
3
See Shoemaker, “Causal and Metaphysical Necessity” and Bird, Nature’s Metaphysics, chs. 3
and 8. For critical discussion, see Fine, “The Varieties of Necessity” and Sidelle, “On the
Metaphysical Contingency of Laws of Nature.”
4
This seems to be the position of Leeds in “Physical and Metaphysical Necessity” and of
Maudlin in the Epilogue to The Metaphysics within Physics.
5
In “Plenitude,” Bricker interprets Adams as expressing a view like this in §III of his
“Presumption and the Necessary Existence of God.” There is room to question this inter-
pretation—see Adams, op. cit., pp. 30 f. and Leibniz, p. 213.
6
On the Plurality of Worlds, §§1.8 and 2.2. One gets “too many” possible spatial geometries
if one can associate with each cardinal number a distinct structure representing a possible
spatial geometry—e.g., if there is such a structure of cardinality Œ for each cardinal number Œ.
7
For critical discussion see Nolan, “Recombination Unbound.” For Lewis’s reaction see
“Tensing the Copula,” p. 8.
10 possible structures of space
2 Distance
Let us turn, then, to the project of considering generalizations of Euclidean
geometry. Of course, Euclidean geometry can be characterized in many
ways. And each variant characterization will support a plethora of gener-
alizations.
Where to begin? Consider the most famous sort of axiomatization of
Euclidean geometry, the quantifiers of which range over points and the
non-logical vocabulary of which consists of non-quantitative relations like
betweenness and congruence.9 We can generalize this approach by con-
sidering the result of tinkering with the primitives or the axioms in various
ways—and in this way we will easily reach generalizations that include the
8
For recent influential defences of this view of the relation between geometry and physics,
see Friedman, Dynamics of Reason and Brown, Physical Relativity.
9
A variety of approaches of this kind are discussed in pt. i of Henkin et al. (eds.), The
Axiomatic Method.
possible structures of space 11
0 1 1 1 1
1 0 a212 a213 a214
1 a212 0 a223 a224
1 a213 a223 0 a234
1 a214 a224 a234 0
Metric Spaces
Let us now begin the task of considering generalizations of the distance
structure of Euclidean geometry.
Once we have made a decision to encode facts about distance in a
real-valued function on pairs of points, Axioms I–III would appear to be
unavoidable: it is difficult to imagine calling anything a notion of distance
which did not satisfy them.13
The remaining axioms do not have this status: it is not difficult to think
of examples in which they fail, but which are nonetheless plausible
examples of spatial geometries. Axiom IV is violated, e.g., if the points
of space are discrete atoms. Axiom V can be violated, e.g., when space is
finite in extent. Axiom VI is violated if space is one-dimensional. Axiom
VII is violated if space has gaps of a certain kind—e.g., if space is all ordered
pairs of rational numbers (with the usual Euclidean notion of distance).
Axiom VIII is violated, e.g., in the classical non-Euclidean plane geometries.
So the class of structures satisfying just Axioms I–III for Euclidean
geometry provides an interesting upper bound for the class of structures
we are looking for. Let us codify this with a definition: a metric space (X, d )
is a set X together with a function d that assigns non-negative real numbers
to pairs of points of X in such a way that for any x, y, z 2 X the following
conditions are satisfied: (i) d(x, y) ¼ 0 if and only if x ¼ y; (ii) d(x, y) ¼ d(y, x);
(iii) d(x, z) # d(x, y) þ d(y, z). We call d a distance function on X.
The notion of a metric space abstracts just a few of the most essential-
looking features of Euclidean space. The resulting family of metric spaces is
very capacious. But the notion of distance built into the definition is strong
enough to support a number of familiar and important mathematical
notions. We will rely on a few of these in considering various generaliza-
tions of Euclidean geometry in the next section.
A map f : X ! Y between metric spaces (X, d ) and (X * , d * ) is distance-
preserving if d(x1 , x2 ) ¼ d* ( f (x1 ), f (x2 ) ) for all x1 , x2 2 X. A distance-
preserving map is an isometry if it is also bijection (i.e., every y 2 Y is the
13
Later we will consider alternatives to the strategy of taking distances to be encoded via
an assignment of real numbers to pairs of points (see pp. 28 ff.). But if this strategy is pursued,
then Axioms I and II seem unassailable. One could conceivably entertain (but not sustain)
scepticism about Axiom III, worrying that it represents an artificial constraint on the notion of
distance; for something along these lines, see Blumenthal, “Distance Geometries,” §4. But
most philosophers seem happy to take Axiom III to be somehow constitutive of our concept
of distance. See e.g. Bricker, “The Fabric of Space,” p. 382; Forrest, “Is Space-Time Discrete
or Continuous?,” p. 329; Maudlin, Metaphysics within Physics, §3.2.
14 possible structures of space
Elliptic geometry is locally like spherical geometry (i.e., for every point,
there is a ball around that point isometric to a ball in spherical geom-
etry). But the elliptic plane has a global structure different from that of
spherical geometry, and in this way it manages to avoid the awkward
features of spherical geometry noted above. In the standard case, we
begin with a sphere of unit radius and in the resulting elliptic space lines
are closed curves of length . One can think of the elliptic plane as
constructed out of spherical geometry by identifying diametrically
opposite points. It can also be thought of as follows: points in
the elliptic plane are represented by lines through the origin in R3 ;
the distance between two points in the elliptic plane is given by the
smaller of the angles formed by the corresponding lines in R3 ; a line in
the elliptic plane is represented by all of the lines through the origin
in R3 that lie on some plane passing through the origin.
A traditional topic in philosophy of space and time concerns status and
implications of the axiom of free mobility (this says, roughly speaking, that
bodies may be transported without altering their shape). One can show
that within a very large family of metric spaces, the only ones satisfying a
certain strong version of the axiom of free mobility are the Euclidean,
hyperbolic, and spherical spaces of various dimensions.19 For this reason
and others, it is helpful to have a label that applies to the Euclidean,
hyperbolic, and spherical spaces (but not to the elliptic spaces): I will
speak of these three as the elementary geometries. Elliptic geometry will
prove to be an important source of examples in later chapters.
22
A subset U X of a metric space is open if for every x 2 U, there is some open ball
around x in X that is contained in U. A set A X is closed if is the complement of some open
set. In the present context, a set is connected if and only if every pair of points in it can be
connected by a curve in the set.
23
These are the possible topologies for complete connected flat Riemannian two-mani-
folds (each topology admits a small family of isometry classes of flat Riemannian metrics); see
Wolf, Spaces of Constant Curvature, §2.5. Our flat spaces are path metric spaces because of
condition (iii) in their definition; and it follows that they are Riemannian manifolds because
by assumption they are locally isometric to a Riemannian manifold; see Burago et al., A Course
in Metric Geometry, pp. 38 and 143.
18 possible structures of space
28
As in the flat case, all of our spaces of constant curvature are Riemannian manifolds. A
Riemannian manifold has constant curvature k if at every point the sectional curvature of every
tangent plane at that point is k. The sectional curvature of a tangent plane is defined as follows.
Let x be a point in a Riemannian manifold M and let — be a tangent plane at x (i.e., a two-
dimensional linear subspace of the tangent space at x). Let Br,—(x) be the subset of Br(x)
consisting of those points that can reached by geodesics from x of length r or less and whose
tangent vectors at x lie in —. For sufficiently small r > 0, Br,— (x) is a well-behaved surface
sitting inside M. For small > 0, let L() be the circumference of the circle in this surface with
centre x and radius . The quantity K(—): ¼ lim!0 33 (2r L() ) is the sectional curvature of
the tangent plane — at x.
29
In fact, this is true in any even dimension. The story is more complicated in odd
dimensions. See pt. iii of Wolf, Spaces of Constant Curvature.
30
For the compact, oriented case, see e.g. ch. B of Benedetti and Petronio, Lectures on
Hyperbolic Geometry.
31
For complaints about the ugliness of this route, see Gromov, Metric Structures for
Riemannian and Non-Riemannian Spaces, p. xvi.
32
i.e., there is a path ª : [a, b] ! Rn with ª(a) ¼ x and ª(b) ¼ y such ª(t) 2 X for each
t 2 [a, b].
20 possible structures of space
33
i.e., X is an embedded Riemannian submanifold of Rn. That all Riemannian manifolds
arise in this way follows from the Nash embedding theorem.
34
In a general Riemannian manifold, one finds variation in the sectional curvature of
tangent planes even if one restricts attention to tangent planes at a single point.
35
In a flat space, if we look at a small scale, all directions look the same. This is not true for
a torus embedded in R3. The flat structure on the torus arises from an embedding in R4.
possible structures of space 21
36
For these and other results, see Burago et al., Course, §5.1; and Petersen, Riemannian
Geometry, ch. 5, esp. §§5.10.1 and 5.10.3. Grove, Riemannian Geometry offers a treatment of
Riemannian manifolds as metric spaces.
37
For a compact overview, see Berger, Riemannian Geometry During the Second Half of the
Twentieth Century, §TOP.9.
38
For details, see e.g. §5.5 of Folland, Real Analysis.
22 possible structures of space
Remark 3.2. Recall that for a, b 2 R with a < b, the closed interval is [a, b] : ¼ {x
2 R : a # x # b} and that a path in a metric space X is a continuous function of
the form ª : [a, b] ! X. A partition of [a, b] is a finite set Y ¼ {y0 , . . . , yn } with
yi < yiþ1 , y0 ¼ a, and yn ¼ b. Define for any partition Y:
P P
i¼n
(Y ) ¼ d((yi1 ), (yi ) ):
i¼1
Then we take L(ª), the length of ª, to be the least upper bound on (Y ) as
Y varies over all partitions of [a, b]. &
39
For details, see Lang, Fundamentals of Differential Geometry or Abraham et al., Manifolds,
Tensor Analysis, and Applications.
40
For each n $ 2 there are at least continuum-many diffeomorphism classes of differenti-
able manifolds; Eichhorn, Global Analysis on Open Manifolds, proposition 5.0.3. Any differ-
entiable manifold admits continuum-many isometry classes of Riemannian metrics.
41
Every Hilbert space is an infinite-dimensional Riemannian manifold; see Abraham et al.,
Manifolds, p. 353. And for every cardinal number Œ, there is a Hilbert space that has Œ dimensions,
in the sense that its Hilbert space bases have cardinality Œ; see e.g. Folland, Real Analysis, p. 169.
possible structures of space 23
As we will see below, not every metric space is a path metric space.
Every Riemannian (Hilbert) manifold is a path metric space—but there
are many non-Riemannian path metric spaces.
Example 3.1 (A Non-standard Metric on the Plane.). The taxi-cab metric on
R2 is given by:
The standard Euclidean metric and the taxi-cab metric disagree systemat-
ically about distances. This disagreement is reflected, e.g., in a divergence
over the shape of balls and spheres: these are round in the Euclidean case
and diamond-shaped in the taxi-cab case. But this makes no difference to
the notions of convergence for these spaces (or, therefore, of continuity): a
sequence of points in R2 that converges according to one of these notions
of distance if and only if it converges according to the other. It is not hard
to see that the plane equipped with the taxi-cab metric is not a Riemann-
ian manifold.42 &
Example 3.2 (Axes of Evil). Begin with the plane equipped with the taxi-
cab metric. Delete all points except those lying on the x- and y-axes, but
leave the distance relations between the remaining points as in the original
space. Using an obvious convention according to which the name of a
point tells us which axis it lies on and the value of the relevant coordinate
on that axis, the metric on this space is given by the conditions:
d(x1 , x2 ) ¼ jx1 x2 j, d(y1 , y2 ) ¼ jy1 y2 j, and d(x, y) ¼ jxj þ jyj. This
space is not a Riemannian manifold (it doesn’t even have the local
structure of Rn at the point where the two axes meet). &
Example 3.3 (Metric Bouquets). Let {XÆ } be a set of metric spaces. Choose
a distinguished point xÆ in each XÆ . The bouquet of the XÆ is the result of
gluing them together by identifying each of the points xÆ .43 Even if the XÆ
are Riemannian manifolds of the same dimension, the resulting bouquet
will typically not be Riemannian—as in the previous example. &
Example 3.4 (Graphs). An edge is a metric space isometric to a closed finite
interval of the real line. Each edge has two endpoints. A metric graph is a
42
Under the taxi-cab metric, there are length-minimizing routes from (1, 0) to (1, 0)
and to (0, 1) that initially travel along the x-axis. But Riemannian geodesics never coincide
and then diverge in this way; see e.g. Petersen, Riemannian Geometry, §5.2, lemma 7.
43
For details, see §4.2 of Burago at al., Course.
24 possible structures of space
45
For other notions of distance for Zn see Forrest, “Discrete or Continuous?”
46
For more examples in the same spirit, see Bricker, “Fabric.”
47
Warning: the fact that we are representing spatial geometries via metric spaces is not
meant to prejudge this question. For discussion of the relation between representor and
represented in this context, see §5 below.
48
A more careful treatment would cash the question out in terms of pairs of worlds sharing
some but not all of their geometric features. See Bricker, “Fabric.”
49
See e.g. Lewis Plurality, p. 62.
26 possible structures of space
50
See Bricker, “Fabric,” §VI; Butterfield, “Against Pointillisme about Geometry,” §3.3.3;
and Nerlich, “Space-Time Substantivalism,” §1.3.
51
See Maudlin, “Buckets of Water and Waves of Space,” p. 196 and Metaphysics
within Physics, §3.2; also p. 329 of Forrest, “Discrete or Continuous?”
52
See Bricker, “Fabric.”
53
See Maudlin, Metaphysics within Physics, §3.2. Nerlich may well be committed to the
same view: at one time he espoused the view that distance relations are mediated as deeply
intuitive and suggested that we cannot understand the thesis that there are direct geometric
relations between distant parts of space (The Shape of Space, §1.3); more recently he has
asserted that the nature of geometric relations does not vary from world to world (“Space-
Time Substantivalism,” §§3.3 and 3.4).
54
Such discrete spaces are discussed in Forrest, “Discrete or Continuous?” The possibility
of such spaces is endorsed by Maudlin (Metaphysics within Physics, p. 89 fn. 6). Nerlich
embraces a related view, on which the size of spatial atoms is allowed to vary and the distance
between two given atoms is given by minimizing the lengths of chains of adjacent atoms
connecting them (Shape of Space, §§9.2–9.6). For an interesting framework that supports a
unified treatment of discrete spaces with a relation of adjacency and of path metric spaces, see
Maudlin, “Time, Topology, and Physical Geometry.”
possible structures of space 27
55
See Gerla, “Pointless Metric Spaces.”
possible structures of space 29
56
A field is a set with well-behaved notions of addition and multiplication. The simplest
finite fields are sets of the form {0,1, . . . , p1} for p prime with addition and multiplication
modulo p.
30 possible structures of space
take its usual form—it will have to be generalized in some way. Some
interesting suggestions along these lines have been made.57 &
4 Plenitude of Possibilities
Anyone who employs a possible worlds framework to make sense of
modal and counterfactual discourse needs to build into this framework
some sort of principle of plenitude: there have to be enough possible
worlds to do the job. David Lewis points out that what we want, intui-
tively, is a principle like the following:
(A) “absolutely every way a world could possibly be is a way that
some world is, and every way that a part of a world could
possibly be is a way that some part of some world is.”58
As Lewis notes, that looks at first glance like it should do the job.
(B) “It seems to mean that the worlds are abundant and logical space
is somehow complete. There are no gaps in logical space, no
vacancies where a world might have been but isn’t. It seems to be
a principle of plenitude.”
But, as Lewis also notes, (A) is not in fact up to the job: the ways a world
could be just are the possible worlds of the framework (and likewise,
mutatis mutandis for parts), so (A) as it stands is trivial. Indeed, it is a matter
of controversy how the notion of plenitude of possibilities should be
understood in general.59
But one can give a partial interpretation of the principle by taking
seriously the metaphor behind the characterization of plenitude offered
in ( B) above. For in some contexts there are perfectly clear mathematical
criteria for the gappiness of a space, and one can understand the principle
57
See Archbold, “A Metric Geometry for Plane Affine Geometry over GF(2n).” For a
related approach, see Kustaanheimo, “On the Relation of Congruence in Finite Geometries”
and “A Note on a Finite Approximation of the Euclidean Plane Geometry.” For further
discussion and references, see Fisher, “Geometry According to Euclid.”
58
Plurality, p. 86; see also p. 2.
59
Lewis introduces his principle of recombination in order to give content to the principle
of plenitude; Plurality, §1.8. For an argument that this strategy is inadequate to Lewis’s needs,
see Divers and Melia, “The Analytic Limit of Genuine Modal Realism.” For other attempts
to give content to the principle of plenitude, see Bricker, “Plenitude” and Hawthorne,
“Causal Structuralism.”
possible structures of space 31
5 Distance Relations?
Suppose that one agrees that all possible spatial geometries are represented
by metric spaces. Does that commit one to any particular view about what
geometric properties and relations are instantiated at various worlds?60
Not in itself. It does not follow from the fact that the spatial geometry of
a world is represented by a metric space that the basic or fundamental
60
Here and throughout, I speak in the idiom of the ontological realism, according to
which there is a fact of the matter concerning a world’s most basic constituents (objects,
properties, relations). On alternative deflationary or anti-realist views, questions like the one
just posed will threaten to collapse—which would tend to benefit the varieties of relation-
alism that I find most attractive.
32 possible structures of space
61
Tarski, “What is Elementary Geometry?”
62
See Blumenthal, “Distance Geometries,” §i.1 and Robinson, “Binary Relations as
Primitive Notions in Elementary Geometry,” §2.
63
The locus classicus is §5 of Field, “Can We Dispense with Space-Time?” For further
discussion see Melia, “Weaseling Away the Indispensability Argument,” §7; Mundy, “On
Quantitative Relationist Theories;” and Skow, “Are Shapes Intrinsic?”
64
See the main result in Mundy, “Space-Time and Isomorphism.”
possible structures of space 33
65
i.e., given R arising in this way, one can find a distance function d on X such that R(w1,
w2, x1, x2, y1, y2, z1, z2) if and only if jd(w1, w2) d(x1, x2)j $ jd(y1, y2) d(z1, z2)j—and any
two such distance functions differ only by a scale factor. See Krantz et al., Foundations of
Measurement, vol. i, ch. 4, theorem 6.
66
Consider a metric space with just three points, x, y, and z. Label the distances a ¼ d(x, y),
b ¼ d(y, z) and c ¼ d(x, z). Assume that a $ b $ c and take c ¼ 1. There are only four ways of
fixing the values of a and b so that the resulting metric space is replete: (i) a ¼ b ¼ 1; (ii) a ¼ b ¼
2; (iii) a ¼ 2 and b ¼ 1; (iv) a ¼ 3 and b ¼ 2.
67
On (i), cf. Field, “Can We?,” §7; Mundy, “Quantitative Relationist Theories,” §3; and
Melia, “Weaseling,” §7.
34 possible structures of space
For my own part, I am tempted to say that there are worlds that differ in
this way. If we suppose that we live in a world at which space is expanding
and spherical in geometry, doesn’t it then make sense to consider two
worlds in which space is spherical but of constant volume—one in which
space is the size our space was in olden times, one in which space is the size
it is now?
But this issue is not an important one for present purposes, and it is one
on which philosophers tend to divide, since it is one for which overarching
views about the metaphysics and epistemology of modality can have
consequences. I will typically leave it to the reader to fill in qualifications
such as ‘up to a choice of scale, if you believe that makes any sense.’
II
Spatial Structure for
Relationalists
2 Conservative Relationalism
The challenge facing relationalists is to provide surrogates for the canonical
substantivalist truth conditions for attributions of geometric structure to
space. Broadly speaking, relationalist approaches to this problem fall under
two main headings. Modal approaches appeal to possible but non-actual
configurations of matter in cashing out claims about spatial structure.
5
For discussion and references, see e.g. Higginbotham, “Truth and Understanding” and
Soames, “Truth, Meaning, and Understanding.”
6
i.e., this second condition rules out one by one all of the spatial structures that are not
finite in extent.
38 spatial structure for relationalists
7
Admittedly, the connection between the official characterization of conservative ap-
proaches and this informal gloss is somewhat loose—but this should not undermine the
discussion below, the point of which is to get clear about the shortcomings of several non-
modal approaches.
8
In addition to taking the material geometry to be time-independent, both Aristotle and
Descartes took the postulation of void or empty space to involve one in serious conceptual
difficulties. See e.g. bk. iv of Aristotle’s Physics and §§ii.16 ff. of Descartes’s Principles of
Philosophy.
9
Descartes himself was famously cagey about the extent of material extension; see e.g.
Principles of Philosophy, §§I.26 f. But it is hard to see what else he could have had in mind.
spatial structure for relationalists 39
10
Note that it would be a disaster to instead demand that there be a real number a such
that for any time and for any material point at that time and for any other time and for any
material point at that second time, a is greater than the distance between the two points. For
in general cross-time comparisons of distance may be impossible (so that the condition under
consideration would fail)—and this could happen even at Aristotelian worlds that conserva-
tive relationalists will certainly want to count as being finite in extent.
42 spatial structure for relationalists
bounded interval of the real line). Call a metric space (X, d ) uni-dimensional
if any three points in X can be labelled y1 , y2 , and y3 with
d(y1 , y3 ) ¼ d(y1 , y2 ) þ d(y2 , y3 ). There is an obvious substantivalist notion
of uni-dimensionality applicable to spatial geometries—and for substan-
tivalists, if the spatial geometry of a world is both uni-dimensional and
indefinite in extent, then it is also infinite in extent. But, sadly, our
translation scheme commits us to saying that the geometry of space at w
is uni-dimensional. And this leads to disaster when combined with our
earlier conclusion that this geometry is indefinite but not infinite in extent.
Third Gambit
Consider again a one-dimensional expanding quasi-Aristotelian world.
The instantaneous configurations of matter can be identified with intervals
of the real line, with length approaching zero as t ! 0 and approaching 0
< R # 1 as t ! 1. The expansion is bounded if R < 1, boundless if R
¼ 1. In either case, it is intuitively plausible that space should have the
metric structure of an interval of the real line of length R—so that space is
infinite at such a world if and only if expansion is boundless.
The gambits just considered were unable to do justice to this intuition—
either they provided no coherent account of the structure of space or they
required us to view that structure as time-dependent. So let us go in search
of a strategy that: (i) assigns to space a determinate time-independent metric
structure that is explored by the dynamical evolution of the instantaneous
material geometries; and (ii) stays as true as possible to the conservative spirit.
We are set the following task.
Selection Problem. Given the set {Xt } of material geometries that
occur at world w, select the metric geometry X that encodes the
geometry of space at w.
Any relationalist will recognize the following as a constraint.
Embeddability. The abstract space X representing the structure of
space at a world must contain, for each instant of time t, a region
isometric to Xt .
On its own, this latter condition is of course far too weak to single out a
solution to the Selection Problem. For note that any one-point metric
space is embeddable in any (non-empty) metric space whatsoever—so
Embeddability alone puts no substantive constraints on solutions to the
spatial structure for relationalists 43
11
See Remark 2.1 below.
44 spatial structure for relationalists
has this feature. Nor, indeed, can any space satisfy both Embeddability and
Minimality for w. For let Z be a space that, for each length, contains an
isometric copy of a finite interval of real numbers of that length. If Z were
minimal, then it would be embeddable in both R and in Y. But if Z is
embeddable in R and contains copies of intervals of arbitrary length, Z
must be uni-dimensional and unbounded—but as noted above, this is
incompatible with being embeddable in Y.
This is the end of the line for the our third gambit, conceived of as
implementing the strict conservative impulse. But it is natural (and instruct-
ive) to wonder what happens if we stray a little bit from the conservative fold.
In the special case of our one-dimensional expanding quasi-Aristotelian
world, we feel sure we know what space looks like: like a real interval of
length R, where R is the finite or infinite limit of the lengths attained by
the material configurations as t ! 1. More generally, let us call a world
cumulatively expanding if Xt is isometrically embeddable in Xt0 whenever
t < t 0 . Intuitively, at a cumulatively expanding world, there is a natural
candidate to represent the structure of space: the Gromov–Hausdorff limit
of the sequence {X0 , X1 , . . . }.12
Cumulative Limit. If w is cumulatively expanding, then X should be
a Gromov–Hausdorff limit of {Xn }, when one exists.
This delivers the desired solution to the Selection Problem at expanding
quasi-Aristotelian worlds.
Now, clearly, this doesn’t get us all that far: the class of cumulatively
expanding worlds is a very special one and it is not clear how to generalize
the strategy behind Cumulative Limit to more general classes of worlds.13
12
See Appendix B below for the relevant technical notions.
13
(1) Note that outside of the class of cumulatively expanding worlds, one can find that a
sequence of metric space converges to a limit that doesn’t contain isometric copies of any of
the metric spaces in the sequence; e.g., as discussed in Appendix B below, a sequence of two-
dimensional Riemannian spaces can converge to a one-dimensional object (a finite graph). (2)
The notion of an ultralimit of a set of metric spaces generalizes the notion for the Gromov–
Hausdorff limit of a sequence (and so shares many of its flaws). An attractive feature of the
ultralimit construction is that it returns a metric space X whenever fed an arbitrary collection
{Xi}i 2 I of metric spaces (so we can feed it the full set of material geometries that occur at a
world). But don’t expect anything for free: the construction requires auxiliary input (the
choice of a non-principal ultrafilter on the index set I) upon which the output will in general
depend. For the ultralimit construction, see Gromov, Metric Structures for Riemannian and Non-
Riemannian Spaces, §3.29; or Kapovich, Hyperbolic Manifolds and Discrete Groups, ch. 9. For
further discussion and examples, see Sormani, “How Riemannian Manifolds Converge,” §3.4.
spatial structure for relationalists 45
are loath to take space to be asymmetric when the data are symmetric—so
if the instantaneous material geometries have the structure of open or
closed intervals, considerations of symmetry militate in favour of taking
space to have the structure of the full line.
The upshot is this. In order to construct a selection procedure that
applies to classes of worlds beyond the very simplest cases, one has to begin
balancing parsimony against other desirable features such as symmetry.15
That takes us very far from the conservative impulse that is our present
concern. We will see in the next chapter where exactly it leads.
Remark 2.1 (Universal Metric Spaces) Let k be a class of metric spaces.
We say that a metric space X is universal for k if it contains, for each metric
space Y 2 k, a subspace isometric to Y.
Of course, there can be no metric space universal for the class of all
metric spaces, since there is no upper bound on the cardinality of metric
spaces.16 But there are universal metric spaces corresponding to many
interesting classes of metric spaces.
Consider ‘1 , the set of all bounded sequences of real numbers equipped
with the metric:
(the right hand side is the least upper bound on the numbers jxk yk j as k
ranges over the natural numbers).17 Let (Z, dZ ) be a metric space with
countably many points, z0 , z1 , . . . Then the map ç: Z ! ‘1 defined by
15
This appears to be a general feature of the sort of problem that we are considering here.
See Remark 2.2 below.
16
For any cardinal number Œ, one can put the discrete metric on a set of cardinality Œ.
17
A sequence {xn} of real numbers is called bounded if there is an N 2 R such that xk < N
for all k.
18
See e.g. §1.22 of Heinonen, Geometric Embeddings of Metric Spaces.
spatial structure for relationalists 47
19
More generally, a Riemannian Hilbert manifold is separable if and only if it is modelled
on a Hilbert space of countably many dimensions.
20
Given a separable metric space X, find an embedding of one of its countable dense
subsets into the universal space, then extend this embedding to all of X by continuity.
21
See e.g. Heinonen, Geometric Embeddings, theorem 3.6; or Bessaga and Pełczy nski,
Selected Topics in Infinite-Dimensional Topology, §II.1.
22
More explicitly, condition (b) says: for any finite subsets {x1, . . . , xn}, {y1, . . . , yn} U
such that d(xi, xj) ¼ d(yi, yj), 1 # i, j # n, there is an isometry ç : U ! U such that yi ¼ ç(xi),
for 1 # i # n.
23
For discussion and references, see Hušek, “Urysohn Universal Space, its Development
and Hausdorff ’s Approach” and Hubička and Nešetřil, “A Finite Presentation of the Rational
Urysohn Space.”
24
See Vershik, “Random Metric Spaces and Universality.” Note that despite its many
strange properties, the Urysohn space is well behaved topologically; see Uspenskij, “The
Urysohn Universal Metric Space is Homeomorphic to a Hilbert Space.”
48 spatial structure for relationalists
25
For what follows, see e.g. Hodges, A Shorter Model Theory, §6.1.
26
In order for f to be an age, it is necessary and sufficient that it be closed under taking
substructures and that for any two members of f, there is a third that contains isomorphic
copies of each of them.
27
Intuitively speaking, f has the amalgamation property if: whenever A, B, and C are
structures in f such that B and C each include a copy of A, then f also includes the structure
that results if one glues B and C together by identifying points corresponding to the same
point in A.
spatial structure for relationalists 49
3 Modal Relationalism
Substantivalists are able to ground their attributions of geometric structure to
space in criteria that concern the patterns of distance relations instantiated by
the points of space. Conservative relationalists work hard to provide substi-
tutes for these criteria directly in terms of the distance relations instantiated
by parts of matter. But it seems that even relatively simple worlds in which
the geometry instantiated by matter changes over time provide serious
obstacles for the most straightforward varieties of conservative relationalism.
The obvious alternative is to go modal—to take over versions of the
substantivalist criteria, understood now as asserting not that a certain
pattern of relations is instantiated but that the instantiation of such a
pattern is possible relative to the world in question.30 Thus, where sub-
stantivalists take space to be infinite in extent at a world if and only if there
exists a linear, unbounded set of points of space of that world, modal
relationalists take space to be infinite in extent at a world if and only if it is
possible relative to that world for there to exist a linear, unbounded set of
material points.
28
See Kechris et al., “Fraïssé Limits, Ramsey Theory, and Topological Dynamics of
Automorphism Groups,” §§2 and 6(E).
29
Aside: in general, the set of finite subspaces of a metric space fails to determine the
metric space (the real numbers and the positive real numbers have the same finite subspaces);
but some metric spaces are characterized up to isomorphism by their finite subsets. See
Bogatyi, “Metrically Homogeneous Spaces,” propositions 1.11 and remark 1.12.
30
Traditionally, this modal path is motivated by snippets from Leibniz. The question
whether Leibniz was in fact a modal relationalist will be left for Appendix D below. (He was.)
50 spatial structure for relationalists
31
For discussion and references, see Brighouse, “Incongruent Counterparts and Modal
Relationalism,” §3; see also Butterfield, “Relationism and Possible Worlds,” §4. Note that
the view in question is a contentious one; see fn. 8 of Ch. I above.
spatial structure for relationalists 51
32
This supposition drives the classic discussion of geometric possibility, Field, “Can We
Dispense with Spacetime?” See also e.g. Earman, World Enough and Space-Time, §6.12 and
Tooley, “A Defense of Absolute Simultaneity,” p. 230.
33
Why would substantivalists require such a notion? In considering theories in which
there is a nomic dependence of geometry on the distribution of matter, one often has recourse
to test bodies—particles that obey the laws of motion of the theory but which make no
contribution to the shaping of geometry by matter. In introducing test bodies, one has a
choice: either offer a promissory note in place of results showing that solutions featuring test
bodies emerge in certain limits of honest solutions of the equations of the theory; or excuse
oneself from this hard work via recourse to a notion of geometric possibility.
34
For further discussion and references, see §3 of Ch. V below.
52 spatial structure for relationalists
1 Introduction
It seems perfectly natural that relationalists seeking to provide an account
of the notion of geometric possibility should look to well-known accounts
of nomic possibility. Nick Huggett appears to have been the first to have
made a substantive suggestion along these lines, when he proposed a
means of adapting the best-system approach to laws of nature to the
geometric context.1
This chapter is focused on Huggett’s approach, which I take to be
representative of approaches to understanding geometric modality that
are grounded and metric but unambitious.2 I first give a brief overview of
1
“The Regularity Account of Relational Spacetime.” See also ch. 7 of Huggett, True
Motion.
2
Recall that an approach to this problem is grounded if it implies that worlds whose
material configurations are geometric duplicates of one another agree concerning geometric
possibility; that an approach is metric if it implies that two material configurations are
geometric duplicates of one another if and only if they instantiate the same pattern of distance
relations; and that (roughly speaking) an approach is ambitious if it implies that for every
substantivalist world w with material configuration C, there is a relationalist world whose
material configuration is a geometric duplicate of C, and at which the facts about geometric
possibility mirror those of w.
best-system approaches 55
3
See esp. Counterfactuals, §3.3; Philosophical Papers, vol. ii, pp. x–xii; “A Subjectivist’s
Guide to Objective Chance,” pp. 121–4; “New Work for a Theory of Universals,”
pp. 41–3; and “Humean Supervenience Debugged,” pp. 231–3. Lewis’s approach was
prefigured in remarks of Mill and Ramsey.
4
“Fundamental properties are those properties I have elsewhere called ‘perfectly
natural.’ . . . They are not at all disjunctive, or determinable, or negative. They render their
instances perfectly similar in some respect. They are intrinsic; and all other intrinsic properties
supervene on them” (“Ramseyan Humility,” p. 204; see also “New Work” and pp. 60–9 of
On the Plurality of Worlds).
5
Note: the logico-mathematical resources of Lw need not be restricted to those of first-
order logic or one of its close relatives. Of course, this in effect gives Lewis’s analysis another
free parameter—it is implausible that each reasonable candidate for Lw will lead to the same set
of laws at each world (consider by way of illustration the difference in strength between
theories of arithmetic that are finitely axiomatizable in first-order languages and those that are
finitely axiomatizable in second-order languages).
56 best-system approaches
6
“Debugged,” p. 232. But cf. Carroll, Laws of Nature, pp. 50 f.
7
“Debugged,” pp. 231 f.
best-system approaches 57
one under which the facts about nomic possibility and necessity supervene
on non-modal facts.8
Of course the approach also has its problems. Below I discuss several that
also apply to the best-system account of geometry.
8
The question whether Lewis’s approach meets this higher standard is discussed in §5 of
Ch. V below.
9
The four classical geometries admit relatively attractive axiomatizations along these lines.
See Blumenthal, A Modern View of Geometry, chs. vii and viii.
58 best-system approaches
theory would include the following axiom: the pattern of distances be-
tween material points in w is instantiated by some subset of Euclidean
three-space. This axiom is very strong and can be given a relatively simple
form as an algebraic condition on the distances among six-member sets of
material points in w.10 If this axiom alone sufficed to determine an ideal
geometric theory for w, then (setting aside worries about uniqueness) we
would have what we wanted—the patterns of distance relations geomet-
rically possible according to w would be just those consistent with Euclid-
ean geometry. But in order for this theory to be an ideal geometric theory,
it would have to achieve an undominated balance of simplicity and
strength. It does not—we can considerably strengthen our theory without
appreciably diminishing its simplicity by adding as an axiom the assertion
that there are no more than a zillion points or the assertion that there are
finitely many points. By eliminating all infinite models, the addition of
either of these (very simple) axioms would represent a gigantic increase in
strength.11 But then it follows that an ideal geometric theory for w would
permit only finite relationalist worlds—so, in particular, such a theory
would not recognize as geometrically possible an unbounded, collinear
set of material points.
So the most straightforward way of extending the best-system approach
to the geometric setting is a bit of a disaster. Under a Huggett-style
approach one avoids this mess by, in effect, working with a different
notion of a geometric theory. The crucial definitions are as follows.12
10
See Blumenthal, Theory and Application of Distance Geometry, ch. iv.
11
Either addition would cut down the set of isometry classes of models from one with the
cardinality of the power set of the continuum to one with the cardinality of the continuum.
12
The framework sketched here differs from Huggett’s approach in a couple of respects.
(i) For Huggett a geometric theory includes a particular choice of embedding. We can ignore
this wrinkle for present purposes (it engages with complications that we are postponing—see
the discussion of superposability vs. congruence in Ch. IV below). (ii) For Huggett only
metric spaces that arise from Riemannian metrics can play the role of geometric theories. But
as we saw in Ch. I above, this seems unduly restrictive: (a) there are simple metric spaces that
fail to be Riemannian but which nonetheless seem to be every bit as good candidates to
represent a possible spatial geometry as your average zillion-dimensional Riemannian mani-
fold; ( b) the principle of the plenitude of possibilities seems to require us to allow as
corresponding to possible structures of space a very wide range of spaces if we allow even
two-dimensional Riemannian manifolds. Note also that while the main concern of Huggett’s
paper is with relationalist theories of motion, I here focus on the very special case of static
universes.
best-system approaches 59
13
Restricting attention to the special case of metric spaces arising from Riemannian
manifolds, Huggett suggests that “ ‘simplest’ means in order of decreasing importance, lowest
dimension, most regular, flattest, and smoothest” (“Regularity Account,” p. 54; see also True
Motion, §7.5).
14
Of course, this feat usually requires something more powerful than standard first-order
logic.
60 best-system approaches
15
See Huggett, “Regularity Account,” p. 54.
16
For overviews of problems and responses see Carroll, “Nailed to Hume’s Cross?” and
Laws of Nature, chs. 2 and 3; Loewer, “Humean Supervenience;” and van Fraassen, Laws and
Symmetry, ch. 3.
17
See e.g. Loewer, “Humean Supervenience;” Beebee, “The Non-Governing Concep-
tion of Laws of Nature;” Earman, A Primer on Determinism, §v.5; Earman and Roberts,
“Contact with the Nomic. Part I.”
best-system approaches 61
18
See e.g. Earman “In Defense of Laws,” p. 418.
19
None of the other standard philosophical accounts of laws of nature feature an empirical
component. To my mind this is one of the most interesting aspects of the best-system
account.
20
Laws and Symmetries, p. 41. For a survey of some formal approaches to the notion of
symmetry, see Pambuccian, “Simplicity.”
62 best-system approaches
action, and so on. We feel entitled to put such notions to work even in the
absence of satisfactory analyses of them.
The worry about simplicity is not so easily dismissed, however. For it is
only to the extent that we are convinced that there is broad, almost
universal, agreement in the application of the notions of cause and know-
ledge to central cases that we are happy to put these notions to work
elsewhere.
Contrast this with a case like the notion of overall quality as applied to
hockey players, thought of as a global notion that aggregates more spe-
cialized comparative notions—for present purposes, let us take the relative
specialized notions to turn on comparisons of skill and ferocity. No one
values ferocity so highly relative to skill as to claim that Dave “The
Hammer” Schultz (or even Dave “Tiger” Williams) was, all things con-
sidered, a better player than Wayne “The Great One” Gretzky. But the
question of how these two factors should be weighed is a much-disputed
one, and there is spirited disagreement among journalists and drunk people
as to whether Gretzky or Maurice “The Rocket” Richard was the better
player. And the problem is not that these players are tied relative to some
standard, intersubjective notion of quality—imagining a slightly more
ferocious Gretzky would not settle the matter. Nor, at least for the
examples chosen, is the problem that there is a lack of intersubjective
agreement about judgements of ferocity and skill. Rather, each of Gretzky
and Richard is better than the other relative to some reasonable ways of
weighing skill against ferocity, but not others. It is implausible that we here
have a case where there is a univocal intersubjective notion of overall
quality that aggregates the special notions. If seriously pressed to name the
best player, one must ask “Best according to which way of weighing skill
and ferocity?” There is no shared default standard that resolves this ques-
tion in interesting cases.
It is natural to worry that Simplicity is in the same boat. I suspect that if
directed to attend to matters mathematical and matters ontological, almost
anyone would agree that Newton’s theory of gravity was simpler than
Einstein’s. The former enjoys such a large advantage with respect to the
simplicity of its formalism that it would be perverse to claim that this is
outweighed by the fact that the Newtonian explanation of the motion of
heavy bodies requires us to advert to both spacetime and force, while in
best-system approaches 63
general relativity spacetime is the whole story.21 But if pressed to say which
is simpler, the Poisson equation-based version of Newtonian gravity or the
Newton-Cartan version, I think the only proper response is to demand
instructions as to how the advantage of the former over the latter in terms
of mathematical simplicity (only the latter requires any differential geom-
etry) is to be weighed against the ontological simplicity of the latter over
the former (again, one theory invokes both spacetime geometry and force,
the other only geometry).22
Discussion of examples of this kind is not likely to change anyone’s
mind. For while it seems to me that the burden of proof lies with those
who posit a notion of Simplicity in service of their philosophical projects,
they are only too likely to disagree. Further, partisans of Simplicity are
likely to question the intuitions appealed to in the example just given, or
to shrug off that example as involving the sort of tie that one expects will
occasionally arise. In any case, the intuitions of we philosophers are in
fact of limited interest here—intersubjectivity in the relevant sense only
requires agreement across suitably ideal people, and in the present context
ideality will presumably involve a quite high degree of mathematical
facility and sophistication.
Productive debate as to the availability of a notion of Simplicity as
applied to physical theories would require an interesting (non-toy) family
of examples about which experts showed a degree of agreement or
disagreement out of step with the expectations of sceptics or advocates.
There is good reason to doubt that one will turn up such a family of
examples in the nomic context. For serious scientific theories are difficult
to concoct and all too often one finds such great increases of mathematical
complexity from one generation of theory to the next that the question
how one ought to weigh such complexity against other factors does not
really arise.
Things looks somewhat different in the geometric case. As noted above,
in this setting we work, in the first instance, with geometric rather
than linguistic objects—but we are of course free to consider a theory
21
Objection: in fact, some experts do judge Einstein’s theory to be simpler on math-
ematical grounds (see Dafermis, “General Relativity and the Einstein Equations,” p. 483).
Reply: to the extent that such examples are taken to heart, the illustration offered is flawed—
but unless this judgement represents the consensus of the relevant experts, to take it seriously
is to undercut the case in favour of Simplicity.
22
For these theories see e.g. Friedman, Foundations of Space-Time Theories, ch. iii.
64 best-system approaches
23
I looked in a score or so obvious places. The evidence I collected is adduced below.
24
Poincaré’s judgement concerning the simplicity of Euclidean geometry is endorsed by
Schlick (Mulder and van de Velde-Schlick (eds.), Moritz Schlick, p. 168) and (in a sense) by
Reichenbach (The Philosophy of Space and Time, p. 83).
25
Poincaré, La Science et l’Hypothèse, p. 67. The standard English translation omits the
material following the second semicolon; see Greenstreet (trans.), Poincaré, p. 50.
best-system approaches 65
But Poincaré’s position has not gone unchallenged. Karl Menger, speaking
of an axiomatization of hyperbolic geometry whose two sorts of variables
range over points and lines and whose only primitive relation is a notion of
incidence, remarks that:
in Hilbert’s terminology, this is a development on the basis of axioms of alignment
alone, without the traditional axioms of order, parallelism, congruence, and
perpendicularity. In euclidean geometry, on the other hand, assumptions about
concepts other than join and meet are indispensable. This situation does not seem
to bear out a remark repeatedly made by Poincaré: that among the geometries that
might be used in describing physical space, euclidean geometry would always be
distinguished by its greater simplicity.27
26
“On the Foundations of Geometry,” p. 145.
27
Blumenthal and Menger, Studies in Geometry, p. vii; see also Menger, “The New
Foundations of Hyperbolic Geometry,” p. 495. Torretti appears to endorse Menger’s view;
Philosophy of Geometry from Riemann to Poincaré, p. 246. For further discussion, see Pambuccian,
“Axiomatizations of Hyperbolic Geometry,” pp. 334 f.
28
Non-Euclidean Geometry, p. 13. On Coxeter in general, see Roberts, King of Infinite
Space. Although in the passage quoted Coxeter stops short of proclaiming elliptic geometry to
be the simplest in a global sense, the fact that the remark quoted is the only comment about
simplicity that occurs in a survey of the classical geometries suggests that he is at least tempted
by the global claim. Consider also in this connection Clifford’s remarks on the aesthetic
contrast between, on the one hand, Euclidean geometry (with its dreary infinities), and, on
the other, spherical geometry and elliptic geometry (far more complete and interesting);
Lectures and Essays, p. 230.
66 best-system approaches
that he has treated the classical geometries both from the group-theoretic
point of view and from the axiomatic point of view, and continues
I should especially like to emphasize again this fact, in the face of intolerant
utterances which one often hears, and which are aimed at championing this or
that pet concept of the author, as absolutely the simplest and, in fact, the only
suitable one to use in the foundations of geometry.29
29
Klein, Elementary Mathematics from an Advanced Standpoint: Geometry, p. 207.
30
For Coxeter’s tastes in geometry see Logothetti, “An Interview with H. S. M. Coxeter,
the King of Geometry.”
31
The close relation between elliptic and projective geometry is emphasized by Coxeter
on pp. viii, 95, and 110 of Non-Euclidean Geometry. Elsewhere he characterizes projective
geometry as involving “a beautiful and intricate system of propositions, simpler than Euclid’s
but not too simple to be interesting” (Projective Geometry, p. 2).
best-system approaches 67
Balance
Lewis’s account of nomic possibility requires a distinguished intersubjec-
tive notion of balance between simplicity and strength. Here again there is,
to put it mildly, a paucity of evidence in favour of the existence of such a
notion.32
But the role of balance in best-system accounts also raises conceptual
issues. These have their roots in the uncontroversial observation that it is
wholly implausible that there could be a suitable notion of the balance of
simplicity and strength which was objective in the sense of being invariant
across suitably ideal rational creatures—surely beings whose innate com-
putational capacities were quite different from our own would arrive at
quite different bargains in trading off simplicity against strength. It follows
that (at best) the notion of law has a surprising species-relativity under the
best-system account: in some cases at least, the laws-for-humans at a given
world will be different from the laws-for-Vulcans at the same world.33
Lewis floats the optimistic hope that the laws of our own world might
have the special feature that each of these species-relative notions of law of
nature agree here.
If nature is kind, the best-system will be robustly best—so far ahead of its rivals that it
will come out first under any standards of simplicity and strength. We have no
guarantee that nature is kind in this way, but no evidence that it isn’t. It is a
reasonable hope. Perhaps we presuppose it in our thinking about law. I can admit
that if nature were unkind and if disagreeing rival systems were running neck-and-
neck, then lawhood might be a psychological [i.e., species-relative] matter, and
that would be very peculiar.34
32
In fact, the situation is even worse for balance than for simplicity, since it is rare for
authorities to record their judgements concerning balance.
33
Note that ‘laws-for-humans’ is rigidified—at each world it picks out the generalizations
of the best-system describing that world as measured by our actual human notions of simplicity
and balance.
34
“Debugged,” p. 233.
68 best-system approaches
40
Kant, Critique of Pure Reason, pp. A27 f./B43 f., B72, A230 f./B283, A286 f./B
342 f.
41
Thus, Lewis: “The standards of simplicity, of strength, and of balance between them are
to be those that guide us in assessing the credibility of rival hypotheses as to what the laws are.
In a way, that makes lawhood depend on us—a feature of the approach that I do not at all
welcome!” (“Subjectivist’s Guide,” p. 123). Others insist that they see nothing unwelcome
here—see e.g. §4.3 of Cohen and Callender, “A Better Best System Account of Lawhood.”
42
Unless you are satisfied with simply counting the notion of a law of nature as being
inherently indexical. For this sort of move, see Halpin, “Scientific Law: A Perspectival
best-system approaches 73
Ties
Under any way of making precise the notions of simplicity and of the
balance of simplicity and strength, there are bound to be worlds at which
uniqueness of best theories fails. There are several ways of handling such
cases. One could take the laws to be the propositions common to all of the
best-systems for a given world.43 Or one could take it to be indeterminate
whether a proposition included in some but not all best-systems is a law.44
Or one could just say that no propositions are good candidates to be
considered laws at such worlds.45 Whichever route one takes, it turns
out that there are worlds at which the nomic facts are surprisingly weak.
Is this scenario worrying? Lewis thinks not: “. . . what of it? We haven’t
the slightest reason to think the case really arises.” 46
It is important to note that if we have no reason to think that our world
is a world that supports multiple ideal theories, it is not because such
worlds would have to be strange in themselves but because we know so
little about our world and about the relevant standards of simplicity and
balance.47
In the geometric case it is not hard to imagine cases where ties arise—
cases involving worlds which are very similar indeed to worlds supporting
unique ideal theories.
Example 4.3 (Euclidean vs. Elliptic, again). Let us return to the context of
Example 4.2 above: we suppose that the desired species-relative notions of
simplicity and balance exist and restrict attention to the classical geom-
etries, and ask what the ideal geometry is for an n-point equilateral world.
Presumably, the cognitive capacities of possible species lie on something
like a continuum. So we can take ‘Vulcan’ to name a possible species for
whom Euclidean geometry is ideal for n < 6, elliptic geometry is ideal for
Account” and Roberts, “ ‘Laws of Nature’ as an Indexical Term” and The Law-Governed
Universe, §3.3.7.
43
Lewis, “New Work,” p. 41.
44
Huggett, “Regularity Account,” §3.
45
Lewis, “Debugged,” p. 233.
46
Ibid.
47
Perhaps it will be suggested that since ties will be rare under any reasonable system of
standards, and since we have a default entitlement to assume that our world is typical, it
follows that we are entitled to think that our world does not support multiple ideal theories.
Very good. Let us add to the wish list of advocates of best-system approaches that the
appropriate standards of simplicity and balance are reasonable in this sense.
74 best-system approaches
n > 6, but there is a tie for n ¼ 6—both Euclidean geometry and elliptic
geometry achieve an undominated balance of simplicity and strength. The
result is that there is no fact of the matter as to whether the space of a six-
point equilateral relationalist world is finite or infinite, relative to Vulcan
standards. This is surprising, since there is a determinate fact of this sort for
every other equilateral relationalist world and since (presumably) there is
also a determinate fact about six-point worlds for species of slightly
different cognitive capacities than Vulcans. &
Supervenience
It sounds attractive to have nomic facts supervene on facts about the
distribution of fundamental properties and relations. But it is not difficult
to cook up examples that put some pressure on this judgement.
Example 4.4 (Contra Supervenience). Suppose that our physics involves
three types of particles—the a-particle, the Æ-particle, and the a-particle—
and that our physical theory is complete except for an account of what
happens in the incredibly rare cases in which like particles collide. General
principles tell us that for each type of collision between like particles, the
nature of the interaction depends on the value of a constant of nature
associated with the particle type. For each possible type of collision, theory
tells us the constant is a complex number, either i or i. The consequences
of a collision of like particles is delightful if the value of relevant constant is
i, disastrous if its value is i. Thirsty for knowledge, we build a giant
experiment that causes two a-particles to collide—with delightful results.
Feeling lucky, we build a second device that leads to a collision of two Æ-
particles—with disastrous results. We decide not to press our luck further.
And in fact there never is a collision between two a-particles. But what if
we had built a device and caused such a collision? It is tempting to say that
there is fact of the matter as to whether the result would have been
delightful or disastrous. But this contradicts the supervenience of nomic
facts on the the pattern of basic facts—since it implies the existence of two
possible worlds just like ours in all occurrent facts, one of which shares its
laws with worlds in which collisions of a-particles are delightful, the other
of which shares its laws with worlds in which such collisions are disastrous.
&
Intuitions conflicting with supervenience are widespread—and cases
like the one above are often put forward in the course of criticism of the
best-system approaches 75
5 Summation
The best-system approaches to laws and to geometry presuppose very
strong empirical claims about the strength and intersubjectivity of various
notions of simplicity and balance. These claims strike me as rather im-
plausible—and they are in fact largely unsupported by argument. Further-
more, the approaches have some rather surprising—and to my mind,
unattractive—consequences: notably that they render law and geometry
species-relative and imply a supervenience thesis that clashes with wide-
spread intuitions.
How much of this carries over to other grounded, metric approaches to
geometric possibility? Any such approach has to involve a solution to what
was called the Selection Problem in §2 of Chapter II above: given the
metric spaces encoding the instantaneous material geometries at a world,
to construct the metric space encoding the structure of space at that world.
When we discussed this problem previously, we saw that it had no
canonical solution—as soon as one considers even the simplest worlds,
one becomes involved in trading off considerations of size versus consid-
erations of symmetry, and so on. Any selection procedure will be subject
to complaints about the supervenience of modal facts on distance relations
53
All that is required for present purposes is that the two worlds differ in their spatial
geometry.
54
If we allow finite metric spaces as ideal geometries, then it is true at both; if we restrict to
‘nice’ metric spaces, then it may be true at one of w* and w but not the other.
best-system approaches 77
1 Introduction
Substantivalists formulate truth conditions for claims about the structure of
space at a world in terms of the pattern of geometric relations instantiated
by the points of space of that world. Modal relationalists aim to rely on
facts about which patterns of material configurations are possible in some
distinctive geometric sense in making sense of such claims.
Anyone invoking a notion of possibility faces a choice: take the notion
as primitive or attempt to ground it in other sorts of facts.
Consider the case of nomic possibility. At a minimum, realists about
nomic possibility require the notion of two possible worlds sharing their
laws of nature. It is natural to think of this notion as being encoded in an
equivalence relation N on the space of possible worlds that obtains if and
only if two worlds share their laws. One can then take the laws of nature at
a world w to be those contingent propositions that hold in all of the worlds
N -related to w. Most of the extant philosophical accounts of laws of
nature can be understood as attempts to ground the relation N by
providing necessary and sufficient conditions for N (w1 , w2 ) in terms of
facts about w1 and w2 that are not simply facts about what is nomically
possible at those worlds. Under the best-system analysis, considered in
Chapter III above, the relevant conditions turn on whether certain the-
ories are paragons of theoretical virtue at both w1 and w2 . Under the
necessitarian approach, to be considered in Chapter V below, the relevant
primitivist approaches 79
1
Under a third approach, which will not be extensively discussed here, two worlds satisfy
the same laws if the properties instantiated at them stand in the same contingent relations of
necessitation to one another. The sources of this approach are Armstrong, What is a Law of
Nature?; Dretske, “Laws of Nature;” and Tooley, “The Nature of Laws.”
2
For this sort of point, see e.g. Carroll, Laws of Nature, §3.1; Fine, “Varieties of Necessity,”
§3; Lange, Laws and Lawmakers, p. 52; Maudlin, The Metaphysics within Physics, §2.3; Wood-
ward, Making Things Happen, §6.11.
80 primitivist approaches
3
The Armstrong–Dretske–Tooley approach, noted above, allows that worlds satisfying
the description of w can differ in their laws, if the universals mass and charge stand in different
relations of necessitation at different worlds. But the notion of a contingent relation of
necessitation has come in for heavy criticism. See e.g. Lewis, “New Work for a Theory of
Universals,” p. 40; and van Fraassen, Laws and Symmetry, §5.3.
4
For primitivist approaches, see e.g. Carroll, Laws of Nature; Fine, “Varieties of Necessity;”
Maudlin, Metaphysics within Physics; and Pargetter, “Laws and Modal Realism.” Lange
advocates a closely related approach under which subjunctive facts rather than modal facts
are taken as primitive; for discussion of the advantages he sees in this see Laws and Lawmakers,
pp. 139–41 and 210 n. 32. Of course, the introduction of N need not be the end of the story
for primitivists—a view counts as primitivist so long as what is added doesn’t amount to an
attempt to ground the nomic in the non-nomic. Maudlin, for instance, adds to the core view
the thesis that laws together with initial states generate histories. See Maudlin, op. cit., esp.
§§1.3 f. and 4.4; for critical discussion, see Loewer, “Time and Law.”
5
Recall that a form of modal relationalism is: grounded if it implies that worlds whose
material configurations are geometric duplicates of one another agree concerning geometric
possibility; metric if it implies that two material configurations are geometric duplicates of one
another if and only if they instantiate the same pattern of distance relations; and (roughly
speaking) ambitious if it implies that for every substantantivalist world w with material
configuration C, there is a relationalist world whose material configuration is a geometric
duplicate of C, and at which the facts about geometric possibility mirror those of w.
primitivist approaches 81
6
Two unit-ball-shaped regions of matter will of course differ in the identity of the
material points composing them and may also differ in their mass etc., but these will not
constitute a difference in geometric properties.
7
So-called sophisticated substantivalists count possibilities as relationalists do (see e.g. Butter-
field, “The Hole Truth” and Brighouse, “Spacetime and Holes”). We can make this precise as
follows. Consider Euclidean world w containing a single material particle, Fred. And consider
all of the various subjunctive conditionals of the form: “If matter and space were just as they
are except that Fred were located in a different part of space, then . . .” Now evaluate these
conditionals as counterfactuals at w. Clearly, if w is a relationalist world, each of these
counterfactuals is vacuously true—there are no worlds satisfying the shared antecedent of
these conditionals. What if w is a substantivalist world? Non-sophisticated substantivalists will
think that there exist worlds satisfying the antecedent of our counterfactuals—and that when
we use these worlds to evaluate our counterfactuals, we find that some of them are true, some
false. But sophisticated substantivalists dig in their heels at this point. They are counterpart
theorists and do not believe that anything exists in more than one world, including points of
space; there may well be in addition to w other one-particle substantivalist Euclidean
worlds—but at any such world w 0 the only reasonable counterpart for the point of space x
occupied by Fred at w is the unique point of space x 0 that is occupied at w 0 . So when they
evaluate our family of counterfactuals, sophisticated substantivalists find (as relationalists do)
that their common antecedent is false at every world—so that each of the counterfactuals in
this family is vacuously true. This approach has some strange consequences. For if we are this
strict about counterpart relations, we will think that the counterpart of x is occupied at just
about any Euclidean world at which there is matter (exceptions include: world with rotational
primitivist approaches 83
symmetry about an unoccupied point, and small perturbations of such worlds). This attributes
to x a strong and interesting modal property lacked by the other points of space at w: it is,
roughly speaking, a necessary truth concerning x that it is occupied if any point of space is.
And this seems at odds with thinking that space was Euclidean (and hence homogenous) at w
in the first place. I am indebted here to §2.2.1 of Skow, Once Upon a Spacetime and to
discussions with Oliver Pooley.
84 primitivist approaches
11
For convenience, the following discussion focuses on finite metric spaces and on
Riemannian spaces.
primitivist approaches 89
punctured plane (the Euclidean plane R2 with the origin deleted). In this
space, two points are related by an isometry if and only if they lie at the
same distance from the origin—so we have continuum-many geometric
roles, each corresponding to continuum-many points.
If we are told to select a single point in a homogeneous space there is, as
far as the qualitative facts go, only one way to proceed. But if we are told
to select a single point in an inhomogeneous space, there are, even
qualitatively speaking, multiple choices available. So if the geometry of
space is inhomogeneous at a world, we do not necessarily manage to
specify all of the qualitative facts about that world’s material configuration
by specifying only the distance relations between material points—in
particular, in the case in which we are interested in a one-point configur-
ation, the (trivial) facts about distances between points of the configuration
always fall short of determining the geometric role of this single point, and
must be supplemented by facts about distances to landmarks (distinguished
points or regions or singularities of the space).
The homogeneity of a space tells us that all of its points play the same
geometric role. In the setting of Riemannian manifolds, it also makes sense
to speak about the directions one might proceed in from a given point.
A Riemannian manifold is isotropic if all directions are equivalent at each
point.12 Isotropy too, is a necessary condition for congruence to guarantee
superposability.13 The cylinder, for instance, is a homogeneous but aniso-
tropic space: at any point, there is a big difference between the direction that
points along the axis of the cylinder and the direction perpendicular to this
axis. And if we merely know the distance between two points in the cylinder,
we do not know all of the qualitative geometric facts about the region that
they form—in particular, we do not know whether the line that these two
points determine is a circle (perpendicular to the axis of the cylinder), or
straight (parallel to the axis), or spirals around the cylinder (skew to the axis). So
there are distinct geometric roles open to a two-point set with a given distance
between the points—and the same will be true in any anisotropic space.
Further, every Riemannian manifold in which every pair of congruent
regions is superposable is a space of constant curvature: roughly speaking,
12
i.e., for any point, and for any two unit tangent vectors at that point, there is an isometry
of the space that fixes the point and maps one of the two tangent vectors to the other.
13
Note that in the Riemannian setting, isotropy is equivalent to the property that any
isometry between two-point sets can be extended to an isometry of the entire space. So
isotropy implies homogeneity for Riemannian spaces.
90 primitivist approaches
14
One might also reasonably want to handle certain sorts of information that is non-
qualitative in virtue of depending on the identities of the material points involved. For
instance, in three-point space just discussed, the ambitious relationalist could well want to
say that there is only one way in which two particles, a and b, could be located nearby one
another, but that there are two ways that these two particles could be located far away from
92 primitivist approaches
18
In particular, the material points of the full worlds instantiate the same metric geometry,
with C corresponding to the same superposability type at each full world.
94 primitivist approaches
5 Beyond Metricity
My chief concern about the primitivist proposals discussed so far is that the
combination of ambition and metricity that they embody cannot easily be
motivated.
Primitivism about geometric possibility is motivated by a conviction
that, in general, the facts about geometric possibility at a relationalist world
do not supervene on the base of ordinary geometric facts at that world.
Metricity tells us that all intrinsic geometric facts at a world are fixed by
facts about distances. What kind of justification could a fan of the proposals
discussed above give for metricity?
This question is a live one. In the (small) literature on geometric
possibility, one can find authors who take for granted that the super-
venience base of interest consists of facts about non-quantitative relations
as well as authors who take metricity for granted.19 So it is not as if
advocates of metricity can reasonably claim that the burden of proof
clearly lies with their opponents.
Nor is it reasonable to claim that the mere fact that we are taking
possible spatial structures to be representable by metric spaces establishes
metricity, any more than this fact establishes that the fundamental geo-
metric relations at substantivalist worlds are distance relations (see the
discussion of Section 5 of Chapter I above).
If metricity seems innocuous, this is presumably due to our Euclidean
upbringing. For consider the question which we should take as our super-
venience base at a given world, facts about quantitative distance relations or
facts about non-quantitative relations such as congruence and collinearity. In
the setting of the elementary geometries (Euclidean, hyperbolic, spherical),
congruent regions are always superposable—so specifying the distances
between the points making up a region suffices to specify all facts concerning
19
On this question, cf. Field, “Can We Dispense with Space-Time?” and Huggett, “The
Regularity Account of Relational Spacetime.”
96 primitivist approaches
20
Euclidean geometry can be axiomatized in terms of betweenness and congruence. But
facts about these relations fail to distinguish between typical three-point sets in Euclidean
space.
21
e.g., facts about congruence and collinearity might tell us that three points form the
vertices of a non-degenerate equilateral triangle, without allowing us to determine the side-
length of this triangle (measured using the natural unit of distance of the elliptic plane).
primitivist approaches 97
22
In particular, the material points of the full worlds instantiate the same metric geometry,
with C corresponding to the same superposability type at each full world. Note that the
material configurations of a full world determine a metric geometry, so it makes sense to speak
of isometries and superposability types here, even if the basic geometric relations at the worlds
under consideration are not distance relations.
primitivist approaches 99
23
Those who disapprove of trans-world comparisons of distance will want to do every-
thing in a slightly more complicated framework in which what is required is preservation of
ratios of distances.
100 primitivist approaches
6 Summation
Should relationalists be satisfied with an approach along the lines of those
sketched above?
For many philosophers, the sticking point will be the reliance on a
primitive modal notion—the invocation of such notions is often taken to
be objectionably obscure or profligate.24 Of course, just how much of a
sticking point this should be depends not only on how objectionable the
proffered primitive notion is, but also on how unobjectionable the alter-
natives are.
Not being myself a man of much metaphysical principle (see Appendix
A below), I have little to say about these questions. But I will note that
primitivism seems to enjoy some benefits over the best-system approach,
in not relying on implausible empirical assumptions about our standards of
simplicity etc., and in allowing for violations of supervenience that accord
with many people’s intuitions.
Note, further, that primitivists about nomic and geometric possibility
may, if they like, build into their approaches a feature that is unavailable to
best-system analysts: they are in a position to countenance the existence of
multiple worlds devoid of matter that differ from one another as to the
facts about physical or geometric possibility.
Primitivists about laws sometimes assert that there could be two worlds
that: (i) were each empty of matter; (ii) were duplicates of one another as
far as their non-material contents go; yet (iii) differed in their laws and so
differ, e.g., as to how massive point particles would move.25 Primitivists
about geometric possibility may be tempted to follow suit. General rela-
tivity encompasses a plethora of worlds empty of matter and energy that
differ from one another as to their spacetime geometry. And this provides
some motivation for modal relationalists about spatial geometry to ac-
knowledge the existence of matter-free relationalist worlds differing as to
the facts about what sorts of material configurations are geometrically
possible.26
24
See e.g. Dorr, “Finding Ordinary Objects in the World of Quantum Mechanics,” §2;
Loewer, “Time and Law;” or van Fraassen, Laws and Symmetry, ch. 4.
25
See e.g. Carroll, Laws of Nature, p. 64 fn. 4; and Fine, “Varieties of Necessity,” §3.
26
For discussion of the force of this last point, see e.g. Butterfield, “Relationism and
Possible Worlds,” §2 and Brighouse, “Incongruent Counterparts and Modal Relationism,”
§3.3.
primitivist approaches 101
27
For discussion and references, see Casati and Varzi, Parts and Places, p. 45; Lewis, Parts of
Classes, pp. 10 f.; and Sorenson, “Nothingness,” §2.
28
Indeed, proceeding along these lines is the most obvious way to avoid commitment to
the claim that it is a necessary truth that something exists. On this question, cf. Lewis, On the
Plurality of Worlds, pp. 73 f. and Bricker, “Island Universes and the Analysis of Modality,”
pp. 47 ff.
29
Relationalists who countenance the null individual and who allow the basic geometric
relations to vary from world to world will have to add another epicycle, so that some empty
worlds are structured by distance relations, others by non-quantitiative relations—even
though no relations are instantiated at any empty world.
V
Necessitarian Approaches
1 Introduction
Ambitious relationalists maintain that there are many one-particle rela-
tionalist worlds that differ as to the structure of space. To mention just the
tiniest tip of the iceberg, they recognize worlds in which space is Euclid-
ean, and worlds in which it is spherical; worlds in which space is two-
dimensional and worlds in which it is zillion-dimensional.1
Ambitious relationalists who are also grounded hold that when one-
particle worlds differ in their geometries, it is because they differ in the
geometric intrinsic properties instantiated—if two such worlds differ as to
the geometry of space, this can only be because their respective particles
are not geometric duplicates of one another.
The challenge for ambitious, grounded relationalists is to provide an
account of how this can be so. What sort of geometric intrinsic property
1
Recall that, roughly speaking, a form of modal relationalism is ambitious if it implies that
for every substantivalist world w with material configuration C, there is a relationalist world
whose material configuration is a geometric duplicate of C, and at which the facts about
geometric possibility mirror those of w. An approach is grounded if it implies that worlds whose
material configurations are geometric duplicates of one another agree concerning geometric
possibility. An approach is metric if it implies that two material configurations are geometric
duplicates of one another if and only if they instantiate the same pattern of distance relations.
necessitarian approaches 103
2
In this context, a fundamental property is supposed to be thought of in something like
Lewis’s fashion (see fn. 4 of Ch. III above): as an elite intrinsic property whose instances are
perfectly similar in some metaphyscially special respect.
104 necessitarian approaches
Since Lewis denies that an object can exist in two worlds, his official
characterization of the principle of recombination requires that a duplicate
of any object can exist with a duplicate of any other object. It of course
follows from this principle that laws are contingent:
Episodes of bread-eating are possible because actual; as are episodes of starvation.
Juxtapose duplicates of the two, on the grounds that anything can follow anything.
Here is a possible world to violate the law that bread nourishes. So likewise against
the necessity of more serious candidates for fundamental laws of nature—perhaps
with the exception of laws constraining what can coexist at a single position, for
instance the law (if such it be) that nothing is both positive and negative in charge.4
3
On the Plurality of Worlds, pp. 87 f.
4
Ibid. 91.
5
“Ramseyan Humility,” pp. 208 f. For Lewis’s notion of fundamental properties, see fn. 4
of Ch. III above.
necessitarian approaches 105
In this new setting, the point about the contingency of laws takes a new
form: “Let it be a law that every F is a G; combinatorialism generates a law
in which every F is not a G, so that this law is violated.”6 Lewis has in mind
here a scenario of the following sort. Begin with a world w in which three
fundamental physical properties, F, G, and G * , are instantiated and in
which no F is G * but in which every F is G by law. Here G and G * are
both monadic properties. And Lewis’s combinatorialism allows us to swap
the roles of any two fundamental properties or relations of the same
category.7 So there is a possible world w * that differs from w only in that
the roles of G and G* have been switched.8 At w * it is not true that every F
is a G; so the law of w that every F is a G holds only contingently.9
Consider an example. Let w be a decently complex world at which
matter takes the form of point-particles and at which the only non-
geometric fundamental properties are mass and electric charge.10 Let us
further assume that the motions of the particles of w instantiate regularities
in accord with Newton’s “laws” of motion with the forces given by
Newton’s “law” of universal gravitation and by Coulomb’s “law.”11 Let
Pm, q be the set of point-particle worlds at which the non-geometric
fundamental properties are mass and charge. How large is Pm, q ? According
to Lewis, it is very large. It includes worlds at which the particle motions
violate Newton’s laws of motion and worlds at which they comport with
these laws, but only for non-standard forces (maybe the forces go as the
inverse cube of distance, or like charges attract while opposite charges
repel). It also includes a world w * which differs from w only quiddistically:
the particle motions at w and w * are identical, but the mass and absolute
6
“Ramseyan Humility,” p. 209.
7
Examples of categories: monadic properties; n-adic relations; magnitudes (these can be
scalar-valued, tensor-valued, etc.); relational magnitudes. See ibid. 205.
8
Such a pair of worlds is said to differ quiddistically, by analogy with pairs of worlds that
differ haecceitistically in virtue of being qualitatively identical but differing as to which
individual plays which role.
9
For this sort of example—and for an account of why Lewis accepts quiddistic differ-
ences but rejects haecceitisitic differences—see Lewis, “Ramseyan Humility,” §4.
10
Here and throughout this discussion, fundamental non-geometric properties are taken
to correspond to determinable quantities (mass, charge, spin), rather than to particular values
of such quantities. There are some subtle issues in the neighbourhood concerning which roles
of Lewisian fundamental properties can be fulfilled by determinables; see Hawthorne,
“Quantity in Lewisian Metaphysics.”
11
The scare quotes emphasize that I have not said anything about the laws at w, just about
regularities.
106 necessitarian approaches
12
So expressions for the regularities at w* arise by everywhere swapping m and jqj in
expressions describing the regularities at w.
13
For this approach, see Hawthorne, “Causal Structuralism;” and Shoemaker, “Causality
and Properties” and “Causal and Metaphysical Necessity.” Under a related approach, appeal is
made to dispositions or propensities in addition to (or in place of ) causal relations; see Bigelow
et al., “The World as One of a Kind;” Bird, Nature’s Metaphysics, ch. 3; and Ellis, “Causal
Powers and Laws of Nature.”
14
See Fales, “Are Causal Laws Contingent?;” Swoyer, “The Nature of Natural Laws;”
and Tweedale, “Armstrong on Determinable and Substantial Universals.”
15
For further positions along these lines, see Kneale, Probability and Induction; and Sellars,
“Concepts as Involving Laws and Inconceivable Without Them.” For the early modern roots
of nomic necessitarianism, see Ott, Causation and Laws of Nature in Early Modern Philosophy,
§§13–22.
necessitarian approaches 107
16
For discussion of this question, see Bird, Nature’s Metaphysics, ch. 7; and Ellis, “Response
to David Armstrong,” §2. For the remainder of the present discussion of necessitarianism
about laws of nature, it should be taken for granted that in speaking of fundamental
properties, I have in mind only non-geometric fundamental properties.
108 necessitarian approaches
governed by the Newtonian laws and Coulomb’s law, then any world at
which mass were the only fundamental property would be a world
governed by the Newtonian laws. But this seems obvious only because
mass plays a special role in the Newtonian laws of motion—What would
the laws be at a world at which charge was the only property instantiated?
It is not obvious in the general case how to pass from knowledge of what
the laws are at a given world w to knowledge of what the laws are at a
world whose set of fundamental properties is a proper subset of the set of
fundamental properties of w.
The positive case for nomic necessitarianism hangs entirely on the
question of what kind of role causal profiles and nomic roles play in the
individuation of properties. Luckily, for present purposes we can set this
question aside.17 Our interest in nomic necessitarianism is limited to
understanding the general contours of the view—how it hangs together
and what sort of resources it has to reply to certain objections. I turn now
to several such issues (but one of the most serious objections is deferred
until Section 5 below).
Whence the Name? In what sense are the laws of nature necessary for
necessitarians? Here it is essential to distinguish between two species of the
genus. Strong necessitarians add to the core commitments N0–N2 the thesis
that the same fundamental properties are instantiated at every possible
world.18 For weak necessitarians, on the other hand, the fundamental
properties instantiated vary from world to world with there being some
worlds at which fundamental properties alien to our world are instantiated.19
Under any form of necessitarianism, the laws at a world w are given by a
proposition that holds at just those worlds that share their fundamental
properties with w. Under strong necessitarianism, the fundamental prop-
erties at each world are just the fundamental properties of the actual
17
Note, though, that many critics of necessitarianism grant that properties are individu-
ated in part by nomic roles. See e.g. Fine, “The Varieties of Necessity,” §2; Loewer, “Humean
Supervenience,” §VI; Lowe, The Four-Category Ontology, ch. 10; and Roberts, The Law-
Governed Universe, pp. 59 and 74.
18
See e.g. Bird, Nature’s Metaphysics; and Shoemaker, “Causality and Properties” and
“Causal and Metaphysical Necessity.”
19
Of course, there is some logical space between strong necessitarianism and weak
necessitarianism: in principle, one could deny that there are any worlds featuring alien
properties while allowing that there are some worlds at which only some of our world’s
fundamental properties are instantiated.
necessitarian approaches 109
20
However, the laws come out as metaphysically necessary under some closely related
views. See Remark 2.2 below.
21
For the accusation that the Armstrong–Dretske–Tooley approach is “second-order
Humean,” see Swoyer, “Nature of Natural Laws,” pp. 210 f.
110 necessitarian approaches
22
There is a delicate issue here: What if the law sentence implies that there are distinct
properties that fill structurally identical causal roles? For discussion, see pt. 3 of Hawthorne,
“Causal Structuralism.”
23
For an approach of this kind, see Ellis, “Causal Powers.”
24
For worries concerning theories that recognize a plethora of vacuous laws, see Earman,
“Laws of Nature,” §2; Fine, “Varieties of Necessity,” §2; and Roberts, Law-Governed Universe,
§3.3.1.
25
For non-necessitarian approaches that treat laws in this way, see Vallentyne, “Explicating
Lawhood” and Maudlin, The Metaphysics within Physics, ch. 1.
necessitarian approaches 111
26
For an especially thorough treatment along these lines, see Shoemaker, “Causal and
Metaphysical Necessity.”
112 necessitarian approaches
which bodies possess not mass but some alien property, schmass.27 Similarly,
when it comes to scenario (2) either we imagine only one world, or we
imagine two that in fact differ in their fundamental properties.
None of this is very palatable. But presumably it is a small enough price
to pay if one has been driven towards nomic necessitarianism by general
views about the individuation of properties.
27
Of course, the third of these option is not open to strong necessitarians.
28
Indeed, the claim that their own approach to nomic possibility enjoys decisive epi-
stemological advantages over its rivals is made by advocates of most of the major approaches to
laws of nature. See e.g. Armstrong, What is a Law of Nature?; Dretske, “Laws of Nature;”
Earman and Roberts, “Contact with the Nomic. Part II;” Foster, “Induction, Explanation,
and Natural Necessity;” Kneale, Probability and Induction; Roberts, Law-Governed Universe;
Shoemaker, “Causality and Properties;” and Tooley, “The Nature of Laws.” For further
discussion, see Loewer, “Laws and Induction.”
29
It takes a rather different form under the best-system approach; see van Fraassen, Laws
and Symmetry, §3.5. For a sophisticated version of the argument sketched here, see Earman
and Roberts, “Contact with the Nomic. Part II” and ch. 4 of Roberts, Law-Governed Universe.
necessitarian approaches 113
strong regularities that hold accidentally rather than by law. Consider the
mass-and-charge point-particle world w that we have been using as an
example. Suppose that the laws at w are what one would expect:
Newton’s laws of motion with forces given by Newton’s law of univer-
sal gravitation and by Coulomb’s law. According to necessitarians, these
will be the laws at every world at which mass and charge are the
fundamental properties. Now consider a world w * of this kind in
which the particles and their attributes are carefully arranged so that no
particle ever accelerates. Universal non-acceleration is a regularity that
holds accidentally at w * .
What reason do we have to believe that the regularities at our world
that we are inclined to view as laws hold in virtue of facts about our
fundamental properties rather than by accident? For all we have said so far,
there may exist fundamental properties m* and q* from the same categories
as mass and charge whose natures underwrite very weak laws—so weak
that the class of worlds Pm* , q* is in effect as capacious as Lewis takes Pm, q to
be. How could we ever determine that the fundamental properties at our
world are m and q (with their strong laws) rather than m* and q* (with their
dismayingly weak laws)?
Even strong necessitarians face a version of this challenge. Let it be granted
that the same fundamental properties are instantiated at every possible world
and that it follows that the same laws hold at each world. Still, we can
ask what warrant we have for believing that the properties in question
underwrite strong laws (putting substantive constraints on the pattern of
instantiation of properties) rather than weak ones (involving only such
constraints as follow from the logical type of the fundamental properties).
One might be tempted at this point to argue as follows:
The problem with this argument is that it applies to all regularities, even
those too complex and gerrymandered to be considered candidates for
lawhood by physics—there are ever so many such at our world, highly
complicated regularities that we would never dream of taking to be laws.
So the methods of physics leave us with laws much weaker than this line of
thought would suggest. What is needed here is a rationale for taking our
world to have moderately strong laws.
3 Compatibility Properties
Before turning to the geometric case, it will be helpful to isolate an
important feature of nomic necessitarianism: roughly and loosely speaking,
it implies that some intrinsic physical properties encode modal information.
The notion of an intrinsic property is intuitively clear—to say that a
property is intrinsic is to say that whether an object possesses that property
depends only on the object itself, and not on the world external to the
object. There are a number of competing analyses of the notion of
intrinsicality.30 Many of these appear to be serviceable—they agree with
one another and with common intuitions in many cases. But none is
generally taken to be a complete success. I mention two of the more
promising—though nothing that follows depends on the sort of questions
of detail over which competing analyses disagree.
i. P is intrinsic1 if whether a thing instantiates P or not-P cannot be
changed by adding something to its containing world.31
ii. An object is lonely if there is no object at its world wholly distinct
from it, otherwise it is accompanied. A property P is intrinsic2 if neither
P nor not-P imply either loneliness or accompaniment (i.e., there is
a possible world at which an accompanied object has P, a world at
which an accompanied object has not-P, a world at which a lonely
object has P, and a world at which a lonely object has not-P).32
30
For discussion and references, see Weatherson, “Intrinsic vs. Extrinsic Properties.”
31
For an analysis of this sort, see Yablo, “Intrinsicness;” for an alternative that turns on
subtraction rather than addition, see Vallentyne, “Intrinsic Properties Defined.” These
proposals require that objects exist at more than one possible world. For discussion of how
Lewis’s framework can be extended to allow trans-world individuals, see McDaniel, “Modal
Realism with Overlap” and Parsons, “Is Everything a World?”
32
This is a simplified version of the proposal of Langton and Lewis, “Defining ‘Intrinsic’,”
suitable for those who don’t have to muck around with worries about disjunctive properties.
necessitarian approaches 115
As usual, we say that objects are duplicates if they share all of their
qualitative intrinsic properties.33
We say that object x has the property of compatibility with property P if
there is a duplicate of x that exists in a world at which P is instantiated.
Informally: x is compatible with P if the property of being just like x is
coinstantiated with the property P at some world.
Being such that some property is instantiated at your world is in general
one of your extrinsic properties. But being compatible with the instanti-
ation of a given property is an intrinsic property. For let x be a possible
object and consider the class d consisting of all the duplicates of x at
various worlds. For any property P, either some member of d has P or no
member of d has P. In the first case, every duplicate of x is compatible
with P; in the second case, no duplicate of x is compatible with P. Either
way, duplicates never differ concerning whether they are compatible with
P—which is to say that compatibility with P is an intrinsic property.
This much follows just from the definition of compatibility. What sort
of compatibility properties actual things have will of course depend on the
extent of the space of possible worlds.
Let Bucephalus be an actual horse and Porky an actual pig.34 Then, of
course, Bucephalus is compatible with porcinity and Porky is compatible
with equinity. But the question of their compatibility with the absence of
these properties is more subtle.
Consider first how things look under lavish views of the extent of
logical space, under which something like Lewis’s principle of recombin-
ation obtains. Equinity in Bucephalus is one thing, porcinity in Porky
another. So there will be a possible world containing just a duplicate of
Bucephalus and another containing just a duplicate of Porky. It follows
Their official proposal is: a property P which is neither a disjunctive property nor the negation of
a disjunctive property is basic intrinsic if it satisfies the condition in the text; the intrinsic properties
are those that never differ between objects sharing all of their basic intrinsic properties.
33
This is the most common notion of duplication. (1) Lewis sometimes characterizes
duplicates in terms of perfectly natural/fundamental properties rather than qualitative intrinsic
properties (Plurality, 61). The difference is inessential for present purposes, since Lewis takes
perfectly natural/fundamental properties to be intrinsic and to provide a supervenience base
for the qualitative intrinsic properties (see Langton and Lewis, “Defining ‘Intrinsic’,” §IX and
Lewis, “Ramseyan Humility,” §2). (2) For worries about the standard notions of intrinsicness
and duplication, see Bader, “Towards a Hyperintensional Theory of Intrinsicality.” On
Bader’s account, the argument of the paragraph after next would not go through.
34
Here and below, being an animal of a certain kind requires only having atoms arranged
in a certain fashion (in particular, nothing about origin or history is involved).
116 necessitarian approaches
35
Note under this scenario, equinity would not be intrinsic2 (because it would imply
accompaniment). But it would presumably be intrinsic1 (since whether the atoms of an object
are arranged horse-wise is, presumably, something that cannot be changed by adding things to
an object’s world).
36
Of course, this does not work for extrinsic properties: Loki is not compatible with the
property of living in a godless world.
necessitarian approaches 117
Postulate III. Within the set of full relationalist worlds, all geometric
facts supervene on metric facts. That is: regions R1 and R2 of full
worlds w1 and w2 are geometric duplicates if and only if there is an
isometry f : w1 ! w2 that maps R1 to R2 .
Postulate IV. For every region R in a full relationalist world, there is
a relationalist world whose material configuration is a geometric
duplicate of R.
Postulate V. Every region of a relationalist world is a duplicate of a
region of a full relationalist world.
Postulate I tells us that once we know the facts about distances and
compatibility at a relationalist world, we know everything about the
geometric properties and relations instantiated there.39 Postulate II is a
sort of principle of plenitude, telling us that for every way space could be
we have a full relationalist world where space has that structure.40 Postu-
late III tells us that regions in full worlds are duplicates of one another if
and only if: (i) their worlds instantiate the same metric geometry; and (ii)
the regions determine the same superposability class in this shared metric
geometry. Postulates IV and V govern relations of duplication between
regions at full and non-full worlds. Of course, geometric duplicates share
all of their compatibility properties as well as instantiating the same pattern
of distance relations.
Geometric duplication is an equivalence relation. So if the material
configuration at a world w is a duplicate of regions R1 and R2 in full
worlds w1 and w2 , then R1 and R2 are duplicates of each other. So w1 and
w2 instantiate the same metric geometry and R1 and R2 correspond to the
same superposability type in this geometry. So it makes sense to speak of
the metric geometry and superposability type of w.
It is natural to say that a given pattern of distance relations (or distance
relations plus compatibility properties) is geometrically possible relative to
w if it is instantiated in the metric geometry of a full world that contains a
duplicate of the material configuration of w. Similarly, it is natural to say
that a given pattern of distance relations (or distance relations plus com-
39
Note that Postulate I does not say anything about which properties or relations are
basic—in particular, it does not say that distance relations are somehow more basic than, say,
non-quantitative relations such as betweenness and congruence.
40
So what we have here is more closely analogous to weak nomic necessitarianism than to
strong nomic necessitarianism.
necessitarian approaches 119
41
It is also possible to capture the sort of non-qualitative but relationalistically acceptable
information discussed in fn. 14 of Ch. IV above. This would require keeping track not just of
which regions are duplicates of one another, but also of duplications (the trans-world relations
that match up corresponding parts of duplicate regions).
120 necessitarian approaches
Euclidean plane as a proper part. The particle at w2 has (i) but not (ii) while
the particle at w3 has (ii) but not (i).42 &
42
What about empty worlds? By Postulate IV above the material configuration at an
empty world would have to be a duplicate of a material configuration at a full world. This
would appear to be possible only if we count the null individual as a region (see § 6 of Ch. IV
above for discussion of this sort of ploy).
necessitarian approaches 121
43
See e.g. Schaffer, “Causation and Laws of Nature.”
122 necessitarian approaches
44
Perhaps geometric necessitarianism is worse off in this respect than is nomic necessitarian-
ism. In the nomic case, restrictions on recombination have independent motivation in certain
accounts of properties. But in the geometric case the restrictions on recombination are more
or less posited for the sake of grounding geometric possibility in facts about intrinsic properties.
45
According to some views, all properties have both modal and categorical aspects;
according to others, all properties are dispositional and hence modal. For representative
examples of these two strands of thought, see Heil, “Dispositions” and Bird, Nature’s Metaphys-
ics. For another view in the same neighbourhood, see Cartwright, The Dappled World.
46
To get a feeling for the difficulty of delineating nomically-involved facts from others,
see: Carroll, Laws of Nature, §1.1; Earman and Roberts, “Contact with the Nomic. Part I,”
§3.4; Skow, “Earman and Roberts on Empiricism about Laws;” and Roberts, “Reply to
Skow.”
necessitarian approaches 123
47
See e.g. Earman and Roberts, “Contact with the Nomic. Part I,” pp. 11 f. and Loewer,
“Humean Supervenience,” §vi.
48
For this sort of principle, see e.g. McDaniel, “Modal Realism with Overlap,” §ii.
49
For views of this kind, see e.g. Quine, “Necessary Truth;” and van Fraassen, The
Scientific Image, §6.5, and Laws and Symmetry, p. 354 n. 1.
50
See the epigraph to this chapter. This passage is quoted approvingly by van Fraassen,
Introduction to the Philosophy of Time and Space, §vi.6.c.
51
See e.g. Monton and van Fraassen, “Constructive Empiricism and Modal Nominalism,”
§4; and Quine, “Natural Kinds,” “Necessary Truth,” and Word and Object, §46.
124 necessitarian approaches
52
A third neo-Kantian strand of thought can be found in van Fraassen. It is sketched
in broad outline in §§iii.4 and vi.6.c of Introduction and in §6.5 of Scientific Image. But, as
becomes clear in §2 of the Postscript to Introduction, the programme is radically incomplete—
one important stumbling block being the treatment of geometric relations.
53
See fn. 7 above.
necessitarian approaches 125
54
Quine (“Reply to Parsons,” pp. 397 f.) and van Fraassen (Laws and Symmetry, pts. i and
ii) are enemies of laws of nature. But for Quine our attributions of nomic necessity are
traceable to “what passes for an explanatory trait or the promise of it” (“Necessary Truth,”
p. 76). On van Fraassen’s approach, explanations are answers to why-questions, in which
answers those propositions that hold in every model of our current theory play a special role:
“The only genuine empiricist course . . . is to deny that explaining something consists in
showing why it had to be the way it is—tout court. We must say instead that . . . the criteria
for what is a good answer are context-dependent. To be specific, I think that the important
contextual factor here is the background of accepted scientific theory. This must play the
role . . . played in Aristotle’s account by objective necessities in nature” (“A Re-Examination
of Aristotle’s Philosophy of Science,” p. 43; for elements of van Fraassen’s implementation of
this program, see ch. 5 of Scientific Image and pts. ii and iii of Laws and Symmetry).
126 necessitarian approaches
55
See e.g. Langton and Lewis, “Defining ‘Intrinsic’,” §v.
56
This venerable idea is developed in a contemporary form in Fine, “Essence and
Modality” and “Senses of Essence.”
necessitarian approaches 127
The Structure of Space. Let us begin with the question whether the
necessitarians’ distance relations and compatibility properties must be
viewed as modally-tainted in virtue of encoding information about the
structure of space.
It will be helpful to consider the corresponding question regarding the
sort of best-system approach to geometric modality considered in Chapter
III above. The best-system approach is naturally regarded as being the
approach with the strongest claim to being super-grounded. Let us ask
exactly what commitments advocates of the best-system approach would
have to take on board in order to secure this claim.
Intuitively speaking, within the best-system approach, at any world
some claims about the pattern of distance relations instantiated hold of
metaphysical necessity, some of (mere) geometric necessity, while some
are contingent. Consider, for instance, a world w with the spatial structure
of the Euclidean plane that happens to contain four points corresponding
to the vertices of a square. That the distance relations at w satisfy the metric
space axioms is a matter of metaphysical necessity. That the distance
relations between any four points of w have a vanishing Cayley–Menger
determinant (see p. 12 above) is a matter of (mere) geometric necessity.
That w includes three points corresponding to the vertices of a right-
angled triangle is a contingent matter.
Under the standard best-system approach, the metric space axioms have
a very different status from other regularities. As in the nomic case, if we
want to construct a super-grounded best-system account, we must either
drop this feature or find some way of marking the special status of the
metric space axioms that can be safely combined with anti-realism about
modality. A natural way to implement the latter strategy is to take the
130 necessitarian approaches
62
By way of motivation, consider the way that (weak) nomic necessitarians insist that the
exact same fundamental property cannot figure at worlds with different laws, even it would
be natural to give the same mathematical representation of the pattern of instantiation of
fundamental properties at the two worlds.
necessitarian approaches 131
63
A metric space is isometrically embeddable in the Euclidean plane if and only if the
distance relations instantiated by every five-point set of the given metric spaces satisfy a
certain algebraic condition. See Blumenthal, Theory and Applications of Distance Geometry,
§§38–43.
64
Indeed, as has been mentioned before, some authors take it for granted that relationalists
should take non-quantitative relations such as collinearity as the basis for a modal construction
of quantitative relations; see Field, “Can We Dispense with Space-Time?”
132 necessitarian approaches
6 Summation
Under any ambitious approach to geometric possibility, the facts about
geometric possibility at a world w with material configuration C corres-
pond to the facts about geometric possibility at a full world w * whose
material configuration includes a region C * just like C.
The precise relation between C and C * depends on the details of the
account. Under the primitivist approach developed in Section 4 of Chapter
IV, C and C * were composed of the same material points, standing in the
same distance relations. Under the variant primitivist approach of Remark
5.2 of Chapter IV, C stood in a primitive trans-world relation of super-
65
A good place to begin thinking about this topic would be with the spaces and properties
considered in Brehm, “The Shape Invariant of Triangles and Trigonometry in Two-Point
Homogeneous Spaces” and in Brehm and Et-Taoui, “Congruence Criteria for Finite Subsets
of Complex Projective and Complex Hyperbolic Spaces” and “Congruence Criteria for
Finite Subsets of Quaternionic Elliptic and Quaternionic Hyperbolic Spaces.”
necessitarian approaches 133
the intrinsic geometric facts at any world to be rich enough to encode full
information about geometric possibility at that world. I suspect however,
that some readers will feel that the groundedness of the necessitarian
approach is not so much a positive credential as a sleight of hand.
As will be clear enough from the foregoing, I think that unambitious
approaches are the least compelling of the options discussed. And while I
do not think that there is at present that much to choose between the
primitivist and necessitarian approaches, I also find the latter more intri-
guing—in part because it suggests interesting questions concerning what
sort of fundamental properties need to be invoked at each world in order
to determine the superposability type of each region (these questions are
closely related to the question whether the base of properties upon which
facts about geometric possibility supervene under the necessitarian ap-
proach is itself non-modal).
Much of the traditional interest of the substantival-relational debate
derives from its close connection with questions about the relativity of
motion and the proper form for dynamical theories in classical physics. In
the classical setting, it is natural for relationalists to take claims concerning
the state of motion of a body to be (explicitly or implicitly) claims about
the way it changes its position relative to some reference bodies. But for a
substantivalist another option is available: a body is at absolute rest if the
part of space that it occupies does not change; it is in uniform motion if
the amount of space it crosses varies linearly with time; etc. For Newton,
the availability of a notion of absolute acceleration was an invaluable
advantage—it provided the foundations for his laws of motion. At the
same time, the availability of an absolute notion of velocity was something
of an embarrassment: within Newtonian physics the absolute velocity of a
system is a real but empirically inaccessible quantity.
How much of Newton’s physics can be reformulated in terms accept-
able to relationalists? The natural starting point is to see what can be done if
one takes the relative distances and relative velocities between material
particles as one’s dynamical variables. The well-known answer is that in
terms of these variables one can write down a theory that permits exactly
the same relative motions of bodies as does Newton’s theory under the
assumption that the total angular momentum of the system vanishes.
There are various strategies that relationalists can adopt if they are
interested in capturing all of the relationalistically respectable content of
conclusion 137
66
e.g., they can rely on differential equations involving relative accelerations as well as
relative distances and velocities or they can introduce primitive variables to play the role that
total angular momentum about the centre of mass plays in the standard formulation.
138 conclusion
Orthodoxy
Let us begin with an example of the sort that provides motivation for the
orthodox approach. Consider a world w whose inhabitants set out to
determine the spatial structure of their world. Their inquiry splits naturally
into two stages.
In the first stage, the physicists of w perform a series of measurements of
distances and angles. Each set of measurements cuts down the family of
mathematical geometries (metric spaces) that are candidates to represent the
spatial geometry of w. But at any stage only finitely many measurements
have been made and infinitely many candidates remain. Nonetheless, even-
tually the physicists of w find that among the remaining candidates one
enjoys an enormous advantage in simplicity over its rivals. At this point they
close their investigation, concluding (provisionally) that the spatial structure
1
Some philosophical theses in this neighbourhood are in effect highly speculative empir-
ical conjectures—e.g., various conceptual analyses and some claims concerning the proper
interpretation of physical theories.
2
On this point, see Jauernig, “Must Empiricism be a Stance, and Could it Be One?” and
Dorr, “Review of Every Thing Must Go.”
APPENDIX A 141
3
Readers who see only pseudo-questions here should substitute an example more to their
liking.
4
“On What There Is,” p. 190. Of course, there is ample reason to doubt that Quine took
ontological theses as seriously as do many contemporary ontologists: “Our scientific theory
can indeed go wrong, and precisely in the familiar way: through failure of predicted
observation. But what if, happily and unbeknownst, we have achieved a theory that is
conformable to every possible observation, past and future? In what sense could the world
then be said to deviate from what the theory claims? Clearly in none, even if we can
somehow make sense of the phrase ‘every possible observation.’ Our overall scientific theory
demands of the world only that it be so structured as to assure the sequences of stimulation
142 APPENDIX A
Further, Quine insists that the notion of evidence owes its intelligibility to
its ties to everyday use and (especially) to its ties to scientific use—and he
thinks that to question integrity of this notion of evidence is to fall into a
peculiarly philosophical sort of error.8
Something like Quine’s method is followed throughout a wide
swath of contemporary philosophy of physics and analytic metaphysics.
that our theory gives us to expect. More concrete demands are empty . . .” (“Things and
Their Place in Theories,” pp. 246 ff.).
5
See Salmon’s scandalized report, Reality and Rationality, p. 91 n. 12. Quine sometimes
mentions theoretical virtues other than simplicity that can count in favour of hypotheses, but
these further virtues then turn out to be reducible to or trumped by considerations of
simplicity. See e.g. Quine, Word and Object, §§5 f.; and “Posits and Reality,” §I.
6
See “The Scope and Language of Science,” §II; “Posits and Reality,” §III; and “On
Simple Theories of a Complex World.”
7
“Scope and Language,” p. 198. 8
De Caelo, §§I f.
APPENDIX A 143
Although there are very few thoroughgoing Quineans about these days,
many seem to follow Quine in:
(i) Taking ontological inquiry to be (roughly) methodologically
homogeneous with scientific inquiry, in that both depend to a
very large extent on simplicity considerations.
(ii) Holding that one cannot question whether our preference for
simpler accounts in philosophical contexts is grounded in a desire
to know the truth (rather than, say, in aesthetic or pragmatic
considerations) without falling into some sort of sceptical doubt
about the pronouncements of the empirical sciences.
In what follows, I aim to undermine the faith of the orthodox on both of
these points.
Questioning Homogeneity
The hypothetico-deductive method tells us that a hypothesis H is confirmed
by a collection of data E if the members of E can be derived from H
(supplemented, if necessary, by plausible auxiliary hypotheses). Consider a
very simple application: our problem is to determine the curve y ¼ f(x) that
describes the relation between two variables of interest, x and y. So a hypo-
thesis is a guess about what this true curve is. We idealize and assume that our
measurements are perfectly accurate. So a data set consists of finitely many
points in the x-y plane that lie on the true curve: (x1 , f (x1 )), (x2 , f (x2 )), . . . ,
(xn , f (xn )). The hypothetico-deductive method tells us that a given curve is
confirmed by a data set if the points in the data set lie on that curve.
On its own, the hypothetico-deductive method is hopeless as a char-
acterization of the scientific method. This is clear already in our curve-
fitting case: there will be infinitely many curves consistent with any given
data set, of which we regard very few as corresponding to hypotheses
substantially confirmed by the data—but the hypothetico-deductive
method draws no distinction among the curves consistent with the data.
But notice that (in at least some cases) people have pretty strong (and, in
some cases, fairly intersubjective) intuitions about which of the curves
consistent with a given data set correspond to the simplest hypotheses—
and they tend to regard the simplest hypotheses as enjoying a high degree
of confirmation relative to the others. For this reason, it would be a giant
improvement in our account of the scientific method to move from a
144 APPENDIX A
9
In this category falls the dominant Bayesian approach, in its standard subjectivist variant;
see e.g. Howson and Urbach, Scientific Reasoning, §11.k. Some objectivist Bayesians seek to
build a bias in favour of simple hypotheses into the notion of rationality; for critical discussion,
see e.g. Kelly, “Ockham’s Razor, Hume’s Problem, Ellsberg’s Paradox, Dilation, and Opti-
mal Truth Conduciveness,” §3 f.
10
In this category one finds e.g the accounts of Glymour, Theory and Evidence and Forster
and Sober, “How to Tell when Simpler, More Unified, or Less Ad Hoc Theories will Provide
More Accurate Predictions.” For a delineation of the obstacles to carrying the latter of these
accounts over to the ontological context, see Sober, “Parsimony and Predictive Equivalence.”
11
One can find such limits drawn (in the course of defences of scientific realism) by e.g.
Earman and Roberts, “Contact with the Nomic. Part II,” §6; Glymour, Theory and Evidence,
pp. 354 ff.; and Roush, Tracking the Truth, ch. 6.
APPENDIX A 145
12
For an account of this kind, see Harman, “The Inference to the Best Explanation.”
Note, however, that most advocates of the importance of inference to the best explanation
assign it a much more modest role. Lipton, for instance, sees a preference for explanatory
hypotheses as a heuristic by which computationally limited beings can achieve approximate
obedience to the dictates of Bayesianism; see “Is Explanation a Guide to Inference?” and
Inference to the Best Explanation, ch. 7.
146 APPENDIX A
13
There are of course competing tendencies in our thought. Here I aim only to identify
one important respect in which our preference for simpler hypotheses is (part of ) a good
method when applied to scientific questions but not when applied to ontological questions.
APPENDIX A 147
14
Reichenbach argued that our inductive methods are asymptotically reliable and that this
observation provides a sort of solution to Hume’s problem of induction; see e.g. Experience
and Prediction, §§39 ff.
15
i.e., for any 0 < x < 1 and any > 0, there is an N 2 N such that j f (x) fk(x)j < for
k > N.
16
(1) Note that if a problem admits an asymptotically reliable method, then it admits many
of them. From a certain lofty perspective, all of these are on a par—so while it is rational for us
to adhere to some sort of principle of simplicity, it would not be irrational to follow some
other suitable method. But presumably if one takes into account facts about human psych-
ology (computational constraints) and sociology (the need for coordinated action), the field of
live options becomes quite narrow. (2) Note also that for certain problems, simplicity-based
methods enjoy a certain sort of optimality; see Kelly, “Ockham’s Razor.” (3) Note, finally,
that Bayesian agents exhibit a form of asymptotic reliability under certain circumstances; for
discussion and references, see §§6.3–6.6 and 9.4–9.6 of Earman, Bayes or Bust?
148 APPENDIX A
sequence of tosses and conjecture both: (i) that the coin that your mother
gave you has a chance p of coming up heads on any given toss; and (ii) that
your mother’s favourite coin has a chance p of coming up heads on any
given toss. It certainly seems like your method is a good one for deter-
mining the bias of the coin (it leads asymptotically to the truth with
probability one). Whether it is also a good method for determining the
bias of your mother’s favourite coin is a more controversial question.17
But there should not be any controversy concerning the method of
ontology. Consider a world w at which physicists have determined that the
spatial geometry is Euclidean and that the laws of nature are compatible
with either a substantival or a relational understanding of spatial ontology.
Philosophers at w will be interested in the question whether the spatial
ontology of w is in fact substantival or relational. But their method is
patently insensitive to the truth about this matter: they will look for the
simplest total theory (¼physics þ ontology) compatible with their experi-
mental evidence, then take this theory to be true. But what considerations
they weigh and what answer they arrive at will not depend in any way on
what the spatial ontology of their world in fact is—the same articles would
be published by philosophers situated at a world with the other spatial
ontology. In cases of this sort, the standard method of ontology does not
render its practitioners sensitive to which of the hypotheses under consid-
eration is in fact true. As far as sensitivity to the truth goes, the philosophers
of w might as well be pursuing a method that tells them to posit that w has
substantival ontology if and only if Goldbach’s conjecture is true—for, like
Goldbach’s conjecture, the claim that the simplest total theory compatible
with the physics of w is substantival is necessary if true and hence does not
depend in any way on whether w is in fact substantival.
To be led to the truth by a given method always requires an element of
luck—one needs to be at a world at which one of the hypotheses under
consideration obtains. But once this luck is in place, no further luck is
required if one is employing a (fully or asymptotically) reliable method—if
17
On the pro side, see Sober, Reconstructing the Past, §§5.3–5.5. On the con side, note that
there are competing methods that seem equally good but which will sometimes lead to
asymptotic disagreement with yours—e.g. the method that has you roll a ten-sided die before
beginning your coin flips in order to determine whether you are going to take your mother to
have been lying or telling the truth when she told you that the coin was her favourite. If both
methods count as rational, then a certain sort of objectivity fails—there are disagreements
between rational agents that cannot be resolved no matter how much evidence is taken into
account (if not in this example, then in structurally similar ones).
APPENDIX A 149
one plugs away long enough one will latch on to the truth. Not so with
the methods considered in the preceding paragraph—to reach the truth via
such a method requires one to be lucky through and through.
To the extent that a preference for simpler hypotheses is a feature of the
scientific method, this preference can be traced back to our desire for true
theories (because the method as a whole is such that in application to
scientific questions, it can be expected to lead us to the truth—if we stick
with it long enough and if the truth lies in the set of hypotheses we are
investigating). But in many philosophical contexts, no such justification
can be given for our preference for simpler hypotheses—so I think that we
ought to admit that this preference is traceable to aesthetic, pragmatic, and
psychological grounds rather than epistemic ones.
Appendix B
Limits of Sequences of Metric
Spaces
(the set of points in X that are within r units of distance of a point in A).
Definition B.1 (Hausdorff Distance). Let (Z, dZ ) be a metric space with
subsets A, B Z. The Hausdorff distance between A and B in Z is
(Recall that the infimum (‘inf ’) of a set of real numbers is the greatest lower
bound for that set; correlatively, the supremum (‘sup’) of a set of real
numbers is the least upper bound for that set).
Definition B.2 (Gromov–Hausdorff Distance). Let X and Y be metric
spaces. We call r > 0 admissible if there is a metric space Z with a subspace
X 0 isometric to X and a subspace Y 0 isometric to Y, such that
1
The material that follows is largely drawn from Burago et al., A Course in Metric Geometry,
esp. ch. 7 and §8.1. For generalization to the Lorentzian case, see Bombelli and Noldus, “The
Moduli Space of Isometry Classes of Globally Hyperbolic Spacetimes.”
appendix b 151
dHZ (X, Y ) < r. The Gromov–Hausdorff distance between X and Y, dGH (X, Y ),
is the infimum of the set of admissible r.
Definition B.3 (Gromov–Hausdorff Convergence). A sequence {Xn } of
metric spaces converges in the Gromov–Hausdorff sense to a metric space X if
limn!1 dGH (X, Xn ) ¼ 0.
So dGH takes values in [0,1]. The diameter of a metric space X is defined as
Compact Metric Spaces. But in one important special case the Gro-
mov–Hausdorff distance is as well-behaved as one could desire.
A metric space (X, d ) is totally bounded if for every > 0 one can find
finitely many x1 , . . . , xn 2 X such that every x 2 X is within distance
of one of the xi .3 A metric space is compact if it is complete and totally
bounded. Roughly and intuitively speaking, the compact path metric
2
Recall that Y is dense in X if for any point in X any metric ball at that point includes a
point of Y. Think of the rational numbers as a subset of the real numbers.
3
Total boundedness implies boundedness. The converse is not true in general (think of
the result of gluing together infinitely many copies of [0,1] by their initial points) but it does
hold for complete Riemannian manifolds; see e.g. Lang, Fundamentals of Differential Geometry,
corollary viii.6.7.
152 appendix b
spaces are those that are gap-free and finite in extent—spheres, tori, finite
graphs, etc.4
The Gromov–Hausdorff distance between two compact metric spaces
vanishes if and only if the spaces are isometric and (as noted above) the
distance between bounded metric spaces is always finite. So the Gromov–
Hausdorff distance makes the space of isometry classes of compact metric
spaces into a metric space.5
Whenever we have a metric space, we can ask whether it is complete—
whether any Cauchy sequence of points in the space has a limit in the
space. The space of (isometry classes of ) compact metric spaces is complete
relative to the Gromov–Hausdorff metric.6 The subspace of (isometry
classes of ) compact path metric spaces is likewise complete.7
But the space of (isometry classes of ) compact Riemannian manifolds is
not complete. To see this, note first that Gromov-Hausdorff convergence
does not preserve dimension: some sequences of cylinders have as their
limits line segments; some sequences of tori have as their limits circles; etc.8
The same sort of considerations show that if we consider a sequence of
Riemannian manifolds, each of which consists of a pair of spheres of a
given size connected by a cylindrical neck of a given length but with the
radius of the neck shrinking to zero as the sequence progresses, the
Gromov–Hausdorff limit of the sequence will be a pair of spheres con-
nected by a line segment. Here we have a sequence of Riemannian
manifolds whose limit is not a Riemannian manifold (because some parts
of the space are one-dimensional, others two-dimensional). In fact, every
compact path metric space arises as the limit of a Cauchy sequence of two-
dimensional compact Riemannian manifolds.9
4
Any finite metric space is also compact.
5
This space has the cardinality of the continuum; see remark 7.2.5 of Burago et al., Course.
6
See Gromov, Metric Structures for Riemannian and Non-Riemannian Spaces, §3:11: 12 þ or
Petersen, Riemannian Geometry, proposition 42.
7
Every convergent sequence of path metric spaces converges to a path metric space;
Burago et al., Course, theorem 7.5.1. So the space of compact path metric spaces is a closed
subspace of the complete space of compact metric spaces—and a closed subspace of a
complete space is complete.
8
Use the following useful fact: if f: X ! Y is a surjective map (so that every y 2 Y is the
image of some x 2 X under f ) then dGH (X, Y )# 12 disf , where the distortion of f: X ! Y is:
disf : ¼ sup(jdX (x1 , x2 ) dY ( f (x1 ), f (x2 ))j: x1 , x2 2 X}:
9
Any compact metric space X can be approximated arbitrarily well (in the Gromov–
Hausdorff metric) by finite metric spaces: if X is compact, then it is totally bounded; so for
appendix b 153
Recall that associated with any metric space (X, d ) is a metric space
(X, d), the completion of X, that can be thought of as the result of adding just
enough points to X to ensure that all Cauchy sequences in X have limits.10
The completion of a metric space is complete. Isometric metric spaces
have isometric completions. And any complete metric space is isometric to
its completion. The space of compact Riemannian manifolds, the space of
two-dimensional compact Riemannian manidolds, and the space of finite
graphs are each incomplete relative to the Gromov–Hausdorff metric. And
each has the same completion: the space of all compact path metric spaces.
If the principle of plenitude of possibilities is understood as requiring
that the space of metric geometries that represent possible spatial geom-
etries must itself be complete, then we must conclude that if every two-
dimensional compact Riemannian manifold corresponds to a possible
geometry of space, then so does every compact path metric space. If we
have all of the compact two-dimensional Riemannian manifolds and
everything that can be viewed as a limit of them, then we have all compact
path metric spaces.
Remark B.1 (Alternative Notions of Convergence.). The space of (isom-
etry classes of ) compact Riemannian manifolds, like any large space, can
be equipped with many distinct notions of convergence. Why think that
Gromov–Hausdorff convergence is the one relevant to the principle of
plenitude of possibilities? Why not rely instead on a notion according to
which the limit of any convergent sequence of Riemannian manifolds is
always a well-behaved space?
Gromov–Hausdorff convergence is indeed only one of several widely
used notions of convergence applicable to sequences of Riemannian
each > 0 one can find a finite set X ¼ {x1, . . . , xn} X such that every x 2 X is within of
one of an element of X; equipping the X with the metrics they inherit from X gives us a
family of finite metric spaces that converge to X in the Gromov–Hausdorff sense. If X is also a
path metric space, it can be approximated by finite graphs—in order to find a sequence
of finite graphs that converge to X, one just has to introduce edges that join the points in each
of the finite spaces X in a cunning fashion; one finds the desired family of Riemannian
surfaces by, in effect, embedding the resulting graphs in R3, then thickening their edges up
into hollow tubes. See Burago et al., Course, example 7.4.9, proposition 7.5.5., and exercises
7.5.6 and 7.5.15; and Cassorla, “Approximating Compact Inner Metric Spaces by Surfaces.”
10
Let X̂ be the set of Cauchy sequences of points in X. For any {xn}, {yn} 2 X̂, define
is the space that results when one identifies sequences in X̂
d̂({xn}, {yn} ¼ limk ! 1d(xk, yk). X
that are at zero d̂-distance from one another. d̂ induces a metric on X . The map that sends a
point x 2 X to the constant sequence (i.e., sequence with xk ¼ x for all k) gives us a natural
way of identifying X with a subset of X .
154 appendix b
manifolds.11 These alternative notions are strictly stronger than the notion
of Gromov–Hausdorff convergence—i.e., they consider divergent many
of the sequences that have Gromov–Hausdorff limits.12 Indeed, they do
not allow dimension or topology to change as one passes from a sequence of
Riemannian manifolds to its limit.13 This may sound attractive: it means
that if we cash out the principle of the plenitude of possibilities in terms of
one of these notions rather than in terms of Gromov–Hausdorff conver-
gence, including the Riemannian manifolds among the possible geometries
of space will not immediately force us to also include finite graphs and other
non-smooth spaces, since relative to these notions of convergence the limit
of a sequence of manifolds is always a manifold. But this comes at a very
high price: according to these notions it is impossible to approximate a
Riemannian manifold of a given dimension by a sequence of manifolds
of another dimension. It follows that if we were to cash out the principle of
the plenitude of possibilities in terms of one of these strong notions
of convergence, then the principle would no longer be incompatible
with taking the possible geometries of space to be given by all Riemannian
manifolds except those of dimension seventeen—but I take it that this is just
the sort of exclusion that should count as a gap in logical space. &
Remark B.2 (The Weyl Tile Argument). Some might be tempted to object
that something is already wrong as soon as we speak of a finite set of points
as approximating a continuous Riemannian manifold (as in footnote 9
above). There is a famous line of thought known as the Weyl tile argument
that aims to show that the Euclidean plane cannot be approximated by
discrete spaces of a certain sort.14 Consider the lattice of points in R2 with
integer coordinates and count distances by counting the numbers of nodes
one must pass through to get from one point to the other under the
supposition that one is only allowed to move horizontally and vertically.
11
See e.g. the discussions of uniform convergence and Lipschitz convergence in Burago
et al., Course, ch. 7; and the discussion of Cm,Æ-convergence in Petersen, Riemannian Geometry,
§10.3.2. See also Sormani, “How Riemannian Manifolds Converge.”
12
See Burago et al., Course, p. 260 and Petersen, Riemannian Geometry, 309.
13
i.e., if {Xk} is a sequence of Riemannian manifolds that converges in one of these senses
to a space X, then there is some N such that the Xn are homeomorphic to X for all n > N.
14
See Weyl, Philosophy of Mathematics and Natural Science, p. 43. The kernel of the
argument can be found already in Maimonides (Guide for the Perplexed, §i.73). For discussion
and references, see McDaniel, “Discrete Space and Distance” and Forrest, “Is Space-Time
Discrete or Continuous?”
appendix b 155
Pythagoras’ theorem is violated, and the size of the violation remains the
same even if the interval between nodes is shrunk. The same consider-
ations show that no Riemannian geometry can be locally approximated by
discrete spaces of this kind. What this shows, of course, is that one should
use discrete spaces of a quite different kind if one wants to approximate
locally or infinitesimally Euclidean geometries.15 &
The Non-Compact Case. For non-compact X, one can still consider the
space of all (isometry classes of ) metric spaces within finite Gromov–
Hausdorff distance of X. At the heuristic level, at least, this space is a
well-behaved complete metric space.16
And in this setting one still has the notion of Gromov-Hausdorff
convergence of sequences of metric spaces. One can of course find
examples of sequences of Riemannian manifolds that converge to non-
Riemannian metric spaces.17 For example, let A be the subset of the
Euclidean plane E2 consisting of the coordinate axes; recall that for any
r > 0, the r-penumbra of A in E2 , Ur (A), is the set of points in the plane
within distance r of a point in A; the sequence {U1n (A)} converges to A in
the Gromov–Hausdorff sense.18
However, in general the notion of Gromov–Hausdorff convergence
seems too restrictive in the non-compact setting. For example, it seems
natural to think that a sequence of open intervals like {( n, n)} should have
a non-compact limit—but as noted above, a sequence of bounded spaces can
never converge in the Gromov–Hausdorff sense to an unbounded space.
There is a widely used notion of convergence for non-compact metric
spaces which handles such examples well. Roughly and intuitively speak-
ing, the idea is that X is a limit of {Xn } in this new sense if balls of arbitrary
radius around some given point in X are approximated by balls in the Xn .
Definition B.4 (Pointed Metric Space). A pointed metric space is a pair (X, a)
where X is a metric space and a 2 X. The point a is called the distinguished
point of (X, a).
15
See Forrest, “Discrete or Continuous?”
16
See Gromov, Metric Structures, §3:11 12 þ . For a more cautious treatment, see Eichhorn,
Global Analysis on Open Manifolds, §5.1. Complications arise here due to the fact that if X and
Y are complete but non-compact, one can have dGH(X, Y ) ¼ 0 even when X and Y are non-
isometric (thanks here to Peter Petersen).
17
But a complete space that is a limit of a convergent sequence of path metric spaces is
always a path metric space.
18
This follows from the result quoted in fn. 8 above.
156 appendix b
19
For the notion of the distortion of a map, see fn. 8 above.
20
See Burago et al., Course, theorem 8.1.7. A metric space is boundedly compact if all of its
closed bounded subsets are compact.
21
See Gromov, Metric Structures, pp. 85 f. A metric space X is locally compact if sufficiently
small open balls have compact closures. For path metric spaces, local compactness is equiva-
lent to being boundedly compact; see Plaut, “Metric Spaces of Curvature $ k,” theorem 8.
Appendix C
Some Background to the
Absolute-Relational Debate
1
The canonical sources for this tale are Stein, “Newtonian Space-Time” and “Some
Philosophical Prehistory of General Relativity;” and Earman, World Enough and Space–Time.
2
For an interesting attempt to resolve this tension, see Jauernig, “Leibniz on Motion and
the Equivalence of Hypotheses.”
158 APPENDIX C
bodies may simultaneously make room for one another, though there is no
interval separate and apart from the bodies that are in movement. And this is
plain even in the rotation of continuous things, as in that of liquids.”7
Aristotle also provides positive arguments against the possibility of a
void—not only is there no space empty of matter within the cosmos, but
the cosmos itself is not to be thought of as immersed in a larger void space.
(1) In De Caelo, void is characterized as “that in which the presence of
body, though not actual, is possible . . .”8 But it is not possible for there to
be matter beyond the cosmos: such matter could not be there naturally, for
the natural place of earth, water, air, fire, and the heavenly material is
within the cosmos; nor could it have gotten there by violence, for in that
case it would have to be located in the natural place of some other
matter—and there is none such. So an extra-cosmic void is impossible.9
The same argument is supposed to show that there can be no other cosmoi
located outside of our own.
(2) In the Physics, Aristotle tells us that the partisans of the void regard
“it as a sort of place or vessel which is supposed to be ‘full’ when it holds
the bulk which it is capable of containing, ‘void’ when it is deprived of
that—as if ‘void’ and ‘full’ and ‘place’ denoted the same thing, though the
essence of the three is different.”10 Aristotle offers a series of objections to
the void in Book IV, chapter 8, showing that a body immersed in a void
would be both motionless and move with an infinite velocity, etc. These
arguments turn upon the details of the Aristotelian account of natural
place, motion through resisting media, etc., and they exercised a consid-
erable influence on medieval discussions of the possibility and nature of
motion in a void.11 To these we can add the following remark that occurs
in the preamble to Aristotle’s discussion of place: “place cannot be body;
for if it were there would be two bodies in the same place.”12 Many
of Aristotle’s medieval successors saw here a powerful consideration against
the possibility of the void. For if the void is conceived of as a sort of
7
Physics, IV.7 214a28–32. Translation of McKeon (ed.), Aristotle.
8
Physics, I.9 279a14–15. Translation of McKeon (ed.), Aristotle.
9
There is some reason to think that that Aristotle here assumes that something is possible
only if it happens at some time or other; see Hahm, The Origins of Stoic Cosmology, p. 103
esp. fn. 32.
10
Physics, IV.6 213a15–20. Translation of McKeon (ed.), Aristotle.
11
See Grant, Much Ado About Nothing, ch. 3.
12
Physics, IV.1 209a6–7; translation of McKeon (ed.), Aristotle. See also IV.8 216a34–b10.
160 APPENDIX C
13
For the medieval influence of this argument, see Grant, Much Ado, pp. 32 ff.
14
See Sorabji, Matter, Space and Motion, ch. 11.
APPENDIX C 161
15
On this, see Sorabji, Matter Spaceam Motica pp. 199–201.
16
Grant, “Place and Space in Medieval Physical Thought,” p. 154.
17
See Grant, “The Medieval Doctrine of Place,” §2 and Much Ado, p. 125.
18
See p. 69. This work is an eccentrically augmented free translation of a work by
Gassendi—and it played a pivotal role in making available in English Gassendi’s attempts to
Christianize and modernize atomism. Newton is known to have read this work carefully as an
undergraduate; see Westfall “The Foundations of Newton’s Philosophy of Nature,” p. 172
esp. fn. 5.
19
Physics, III.1 200b33–4. Translation of McKeon (ed.), Aristotle.
20
Ibid. VIII.6 260a27–8.
21
Ibid. IV.4 212a15–19.
162 APPENDIX C
cosmos, what happens if you extend your staff (or spear, or sword, . . . )
beyond the edge? If there is something there to prevent its extension, then
you are not yet at the edge—there is further matter. On the other hand, if
you are successful, then there must be receptive void. Repeating the
argument whenever a new putative boundary is reached shows that
there is infinite extension—of either matter or void.25
Lucretius also gives two detailed arguments in favour of the void, defined
as “intangible empty space.”26 The first rests upon the traditional atomist
contention that motion would be impossible in a plenum.27 The second,
cleaned up and amplified, proceeds thus: suppose two bodies in contact
along a surface move away from one another; then air must fill the space
between the surfaces initially in contact; but if it moves with only finite
velocity, there will be void immediately after the separation of the bodies.28
So far, the arguments given allow us to think of the void of the atomists
either as being something like the space of modal relationalists or as being
something like Newton’s absolute space. But following the arguments just
discussed, Lucretius remarks that:
If there were no place and space, which we call void,
Bodies could not be situated anywhere
And they would totally lack the power of movement,
As I explained a little while ago.29
Now, Lucretius has earlier told us that bodies could not move if there were
no void. But that they would be situated nowhere appears to be a new
25
De Rerum Natura, I.968–83. For discussion of origins of this argument and of Aristotel-
ian responses in antiquity, see Sorabji, Matter, Space and Motion, pp. 125–8. For Scholastic
responses, see Grant, Much Ado, pp. 106–8. The argument also appears in More and Gassendi;
see Koyré, From the Closed World to the Infinite Universe, p. 123 and Grant, op. cit., 389 n. 168.
Here is another popular atomist argument: that which is limited must be limited by some-
thing. For this latter argument, see Epicurus, Letter to Herodotus, §41 and Lucretius, De Rerum
Natura, I.957–65. For discussion, see Sorabji, op. cit., 136–8.
26
De Rerum Natura, I.334. Translation of Fowler and Fowler (eds.), Lucretius.
27
Ibid. I.335–45 and I.370–83. See also Epicurus, Letter to Herodotus, §40. Note Aristotle’s
response, Physics, IV.7 214a28–32. See Charleton, Physiologia, p. 19. Note: the undergraduate
Newton appears to have accepted this argument; Westfall, “The Foundations of Newton’s
Philosophy of Nature,” p. 174.
28
De Rerum Natura, I.384–9. See Grant, Much Ado, §4.E for Scholastic responses to this
sort of challenge.
29
Ibid. I.426–9. Translation of Fowler and Fowler (eds.), Lucretius. See also Epicurus, Letter
to Herodotus, §40: “And if there did not exist that which we call void and space and intangible
nature, bodies would have no place to be in or move through, as they obviously do move;”
translation of Inwood and Gerson (eds.), Epicurus, p. 6.
164 APPENDIX C
The most obvious way for us to make sense of this is to refer the motion of
atoms to the parts of the void, conceived of as retaining their identity and
relations to one another over time. For if in the swerveless atomist universe
we look at the relations just between the atoms, we find them utterly
static—and we would have no reason to maintain that the atoms were
falling down like drops of rain rather than sitting motionless.
THE STOICS. The Stoics, while accepting a spherical and void-free cosmos,
explicitly located it within an infinite void.34 Now, the Stoics more or less
30
One might worry about this interpretation, on the grounds that the atomists spoke of
void as non-being. But then, they also seem to have thought that the existence of void shows
that the non-existent is just as real as the existent. For discussion and references, see Barnes,
The Presocratic Philosophers, §XIX(b).
31
For discussion of pre-Epicurean atomism, see chs. 9 and 10 of Furley, The Greek
Cosmologists, vol. 1.
32
According to ancient authorities, Epicurus also held this view; see Inwood and Gerson
(eds.), Epicurus, p. 47.
33
De Rerum Natura, II.222–5; translation of Fowler and Fowler (eds.), Lucretius.
34
For discussion and references, see e.g. Hahm Origins, ch. IV and Sambursky, Physics of the
Stoics, ch. IV. For deflationary readings of the infinitude of the Stoic void, see Todd,
APPENDIX C 165
“Cleomedes and the Stoic Concept of the Void” and Inwood “Chrysippus on Extension and
the Void.” Posidonius appears to have held a heterodox view, according to which the extra-
cosmic void was only just large enough to hold the cosmos at its time of maximum expansion;
see Algra, “Posidonius’ Conception of the Extra-Cosmic Void.”
35
Bowen and Todd (eds.), Cleomedes’ Lectures, p. 24. For further discussion and references,
see Sorabji, Matter, Space and Motion, p. 129 and Hahm, Origins, p. 106. Note that Cleomedes
continues: “But if anyone claims that a conflagration does not occur, such a claim would not
confute the existence of the void. For even if we merely conceived of the substance [of the
cosmos] expanding, that is, being further extended (granted that there is no possible obstacle
to such extension), then this very thing into which it would be conceived as entering in its
extension would be void.” Here it is important that the Stoics employed a relatively liberal
notion of possibility. See Hahm, Origins, p. 103 for references and discussion.
36
See Sorabji, Matter, Space and Motion, pp. 126 ff. and Hahm, Origins, p. 106.
37
Bowen and Todd (eds.), Cleomedes’ Lectures, p. 24. (1) Cleomedes himself denied that
the cosmos was in fact in motion (see ibid. 26)—but this would seem to be perfectly
consistent with taking the possibility of such motion to establish the infinitude of the void,
in analogy with the way that the mere possibility of conflagration establishes the existence of
the void (see fn. 35 above). (2) Achilles the Grammarian records the following Stoic
166 APPENDIX C
argument: “If the cosmos were moving down in an infinite void, rain would not overtake the
earth. But it does. Therefore the cosmos does not move but stands still;” quoted at pp. 109 f.
of Hahm, Origins.
38
Chrysippus, on the other hand, seems to have denied, on broadly relationalist grounds,
that the hypothesis of cosmic motion was a sensible one. For discussion and references, see
Hahm, Origins, p. 122.
39
Grant, “Place and Space,” p. 154.
40
Grant, Much Ado, p. 180.
APPENDIX C 167
41
This was not, however, viewed as problematic by early Christians; see Sambursky, The
Concept of Place in Late Neoplatonism, pp. 14–17.
42
See Grant, “Medieval and Seventeenth-Century Conceptions of Infinite Void Space
Beyond the Cosmos” and chs. 1 and 2 of Much Ado.
43
Grant, Much Ado, pp. 106 f.
44
Ibid. 137. See also Sorabji Matter, Space and Motion, p. 129.
45
These are propositions 1, 2, 4, 79, 81, 97, 149, 178, 203, 205, and 66 in the numbering
and translation found in Lerner and Mahdi (eds.), Medieval Political Philosophy, selection 18.
168 APPENDIX C
46
For discussion and references, see Grant, “The Condemnation of 1277, God’s Absolute
Power, and Physical Thought in the Late Middle Ages” and Lindberg, The Beginnings of
Western Science, pp. 233–44.
47
For helpful discussion, see Koyré, From the Closed World and Grant, Much Ado, chs. 7 and 8.
APPENDIX C 169
48
For discussion, see Lolordo, Pierre Gassendi and the Birth of Early Modern Philosophy,
pp. 106–8. For references to others who rejected the substance-attribute dichotomy in the
case of space, see Grant, Much Ado, pp. 187, 199, 204, 217, 240, and 392 nn. 182 and 185.
49
Charleton, Physiologia, p. 69.
50
For references and discussion, see Grant, Much Ado, p. 389 n. 168.
51
See Charleton, Physiologia, p. 11 and Brush (ed.), Gassendi, p. 387. For discussion, see
Lolordo, Gassendi, pp. 109 ff.
52
Brush (ed.), Gassendi, 388; Charleton, Physiologia, 67 f. Gassendi also says that God faces
a choice in deciding where in space to create the world; see the passage quoted on p. 110 of
Lolordo, Gassendi.
53
Leibniz is of course no exception. See e.g. the passages at Loemker (ed.), Leibniz,
pp. 577 and 668.
170 APPENDIX C
related to one another in many ways. I would like to emphasize just one
aspect by claiming that the transition from Aristotelian cosmology to the
new cosmologies of the seventeenth century undermined the most
straightforward route to interpreting curves in Euclidean space as repre-
senting the motions of bodies; and that the competing accounts of the
nature of space, the nature of motion, and the relation between the two
that one finds in Descartes, Newton, and Leibniz can be viewed as aspects
of the process of recognition and resolution of this problem.
From Galileo onwards, the new mechanics was based on one form or
another of the principle of inertia, according to which bodies free from
interference naturally tend to trace out a certain sort of curve in space. The
interpretation of curves in Euclidean space as representing the motion of
bodies is unproblematic in contexts in which the motion of all bodies can
be understood as motions relative to a natural reference body. For then
(to speak anachronistically) one can regard the curves as describing motion
in the space picked out by coordinate axes attached to the reference body.
The location of a moving body relative to the fixed body is determined at
each moment of time by the parameterization of the geometric curve
associated with the moving body.
In the mainstream cosmological tradition deriving from Aristotle and
Ptolemy, the Earth is at rest at the centre of a finite series of rotating
material spheres which exhaust the contents of the universe. In this
context the Earth provides a geometrically privileged, fixed body—the
natural reference body to which the complicated trajectories of Ptolemaic
astronomy can be referred.
For Copernicus and Kepler, the cosmos is still spherical, and both the
central sun and the outer surface which encloses the fixed stars are
immobile, and are suitable to serve as reference bodies.54 According to
Copernicus, the stars are fixed to the surface of the outermost sphere;
according to Kepler they are scattered throughout a shell within the
outermost sphere, with the shell enclosing a void in which the solar system
is located. Copernicus is quite explicit: “the first and supreme of all is the
sphere of the fixed stars which contains everything and itself and which,
therefore, is at rest. Indeed, it is the place of the world to which are
referred the motion and the position of all other stars.”55
54
On this point, see Koyré, From the Closed World, pp. 29–34 and 76–87.
55
Quoted at Koyré, From the Closed World, p. 33. Of course here the planets, including the
earth, are numbered among the “other stars.”
APPENDIX C 171
56
Drake (ed.), Galileo, p. 327. Galileo, while content to grant for the sake of argument that
the cosmos is spherical in shape, makes a point of noting that there is little evidence that the
material universe is finite in extent; ibid. 319 f. And indeed, he appears to have been
genuinely undecided on—and quite likely, not especially interested in—questions concern-
ing the finitude, infinitude, or deployment of the stars; see Koyré, From the Closed World,
pp. 95–9.
57
Newton’s treatment of Descartes’s analysis of motion can be found in De Gravitatione;
see esp. the passage at pp. 19–21 of Janiak (ed.), Newton. Spinoza appears to make a similar
point in corollary 3 to proposition 22 in pt. 2 of The Principles of Cartesian Philosophy.
172 APPENDIX C
1
For texts and commentary, see e.g. Lamarra, “Leibniz on Locke on Infinity.”
2
§4 of Leibniz’s third letter to Clarke. Translation of Ariew (ed.), Leibniz and Clarke.
For similar language, see §41 of Leibniz’s fourth letter and §§47, 104, and 106 of his fifth
letter.
174 appendix d
space, like time, is a certain order, namely (in the case of space) that of coexisting,
which includes not only actual things but also possibles. It follows that it is
something indefinite, like every continuum whose parts are not actual but can be
taken at will, just like the parts or fractions of a unity.3
Now, such remarks on their own do not establish that Leibniz is a modal
relationalist. After all, Leibniz, like Descartes, holds that matter forms a
plenum with the structure of Euclidean three-space. So we could under-
stand his remarks about possibilia not as intended to suggest that possibilia
be employed to probe the structure of empty parts of space, but as
intended merely to draw our attention to the fact that there are many
ways that Euclidean space can be filled with matter.4
I believe, however, that a pretty good case can be made for the
orthodox reading of Leibniz as being a modal rather than a conservative
relationalist. My case comes in two parts. In the first I argue that there are
pretty conclusive reasons for denying that Leibniz is a conservative rela-
tionalist. In the second I argue that the texts strongly suggest that he is
indeed a modal relationalist. I conclude by considering some worries one
might have about these arguments.
In another passage, Leibniz indicates his reasons for denying that the poles
inside a void sphere would touch.
If there were a vacuum in space (for instance, if a sphere were empty inside), one
could establish its size. . . . It follows from this that we can refute someone who says
that if there is a vacuum between two bodies then they touch, since two opposite
poles within an empty sphere cannot touch—geometry forbids it.7
7
Ibid. §ii.xv.11. Translation of Remnant and Bennett (eds.), Leibniz.
8
In fact, is not clear that Leibniz endorses this line of thought (see the discussion of Worry
I below). But note that Leibniz states unambiguously that a vacuum is possible in his letter to
Johann Bernoulli of 13 Jan. 1699; pp. 170 f. in Ariew and Garber (eds.), Leibniz.
9
§30 of Leibniz’s fifth letter to Clarke (see also §73 of the same letter). Translation of
Ariew (ed.), Leibniz and Clarke.
10
Difficulties arise when one considers worlds more exotic than those Leibniz likely had
in mind. See the discussion of §2 of Ch. II above.
11
Russell, A Critical Exposition of the Philosophy of Leibniz, §§71 and 74. The passages that
Russell had in mind are cited in fnn. 3 and 15 of the present discussion. For a contrasting
reading of Leibniz, see Rescher, “The Plurality of Space-Time Frameworks,” esp. §9.
176 appendix d
that these orders—space and time, that is—relate not only to what actually
is but also to anything that could be put in its place . . .”12 In light of the
discussion above, it is natural to read this as telling us that the fact that space
is Euclidean in structure at our world determines what sort of configur-
ations of matter are possible. Thus, there is some possible world whose
matter forms a finite spherical cosmos of Aristotelian type—but no world
whose matter forms a Klein bottle, or any other configuration that could
not be embedded in a Euclidean space of three dimensions. But what
should we think about the structure of space at an Aristotelian world—is
space infinite there (with the Euclidean geometry of our own world
providing the order of possibilities) or is it finite there (with the limited
extent of matter at that world determining that relative to that world only
finite cosmoi are possible)? Leibniz’s stance is, I think, unequivocal. He tells
us that “time and space indicate possibilities beyond any that might be
supposed to be actual. Time and space are of the nature of eternal truths,
which equally concern the possible and the actual.”13 And for Leibniz the
eternal truths are of course genuinely necessary.14 So it would appear that,
for Leibniz, from the fact that the structure of space at our world is
Euclidean, it follows that every world has Euclidean spatial geometry.
And from that it follows in turn that Leibniz was not a conservative
relationalist, since he is committed to taking space to be infinite even at
worlds of Aristotelian structure.
It would seem that the only way to evade this conclusion would be
to show that distinct notions of possibility are in play in the passages in
which Leibniz allows that worlds of finite material extent are possible
and in the passages in which he seems to imply that space and time have a
fixed structure across possible worlds—perhaps the infinitude of matter is
functioning as a tacit presupposition in the latter sort of passage. But this
suggestion will not work. Consider an analogy that Leibniz develops in his
discussion in “On the Ultimate Origination of Things” for the optimiza-
tion problem that God faces in creating a world:
in this context, time, place, or in a word, the receptivity or capacity of the world
can be taken for the cost or the plot of ground on which the most pleasing building
12
From “Reply to the Thoughts on the System of Preestablished Harmony Contained in
the Second Edition of Mr Bayle’s Critical Dictionary, Article Rosarius.” Translation of
Loemker (ed.), Leibniz, p. 583.
13
New Essays, §ii.xiv.26. Translation of Remnant and Bennett (eds.), Leibniz.
14
For discussion and references, see e.g. Adams, Leibniz, §7.1.
appendix d 177
It seems that God is to consider the possible worlds that result from variant
ways of filling in space and time with matter—with space and time
themselves possessing their structure independently of their material con-
tents. From this it follows that “there would be as much as there possibly
can be, given the capacity of time and space (that is, the capacity of the
order of possible existence); in a word, it is just like tiles laid down so as to
contain as many as possible in a given area.”16 It seems clear here that space
has the same structure at worlds in which matter is sparse as it does at
worlds in which it forms a plenum with the structure of Euclidean space—
and that the infinitude of matter is a consequence, rather than a presup-
position, of the thesis that the structure of space is invariant across worlds.
That puts an end to the interpretation of Leibniz as a conservative
relationalist: there would appear to be no evading the conclusion that he
countenances possible worlds in which space is Euclidean even though the
extent of matter is permanently limited to some fixed finite size.
can be conceived in time indicates, along with that in space, that time and
space pertain as much to possibles as to existents.”18 Why does the possible
existence of void space indicate that space pertains to merely possible
existents as well as to actual ones? Why think of space as a relationship
among possible existents as if they were actual? Here it seems that Leibniz
has motivations for speaking of possibilia in the same breath as space which
far outstrip the tame observation that Euclidean space may be filled by
matter in many ways. These motivations are most explicit in a passage in
which he is commenting on Locke’s insistence that we should distinguish
extension from material extension:
there is no need to postulate two extensions, one abstract (for space) and the
other concrete (for body). For the concrete one is as it is only by virtue of the
abstract one . . . In fact time and space are only kinds of order; and an empty place
within one of these orders (called ‘vacuum’ in the case of space), if it occurred,
would indicate the mere possibility of the missing item and how it relates to the
actual.19
This seems to suggest that in worlds in which matter does not fill all of
Euclidean space, there is nonetheless some sense in which the complete
pattern of Euclidean space exists and makes possible the pattern of exten-
sion instantiated by matter—and that the gap between the full Euclidean
pattern and the pattern materially instantiated somehow directs us towards
possible ways of filling out material extension so that it would instantiate
the full Euclidean pattern of spatial relations.
There is no knockdown argument here. But it does seem to me that in
these passages we have modal relationalism all but made explicit.
Worries
The discussion above invites a number of worries. I discuss five such: three
that concern the thesis that Leibniz recognizes the possibility of void space;
two that concern the thesis that Leibniz takes the structure of space and
time to be the same at every possible world. In each case, the question is
whether the textual support adduced in favour of these theses above is
18
§ii.xiv.25. Translation of Remnant and Bennett (eds.), Leibniz.
19
§ii.iv.5. Translation of Remnant and Bennett (eds.), Leibniz. The suggestion that the
merely ideal in some sense governs the real can also be found in Leibniz’s letter to Varignon of
2 Feb. 1702; see Loemker (ed.), Leibniz, p. 544.
appendix d 179
20
Leibniz’s other favourite argument against the void is driven by the principle of
sufficient reason. To the extent that that principle is taken to be necessary, the considerations
engaged below arise for that argument as well.
21
For discussion and references, see Garber, “Leibniz: Physics and Philosophy,” §4.1.
22
Translation of Ariew (ed.), Leibniz and Clarke. A slightly more detailed version of the
argument can be found in the passage customarily printed as a postscript to Leibniz’s fourth
letter to Clarke; ibid. pp. 27 f.
23
For discussion and references see Adams, Leibniz, ch. 1.
180 appendix d
I do not say that matter and space are the same thing. I only say that there is no
space where there is no matter and that space in itself is not an absolute reality.
Space and matter differ as time and motion. However, these things, though
different, are inseparable.24
24
§62 of Leibniz’s fifth letter to Clarke. Translation of Ariew (ed.), Leibniz and Clarke.
25
§63 of Leibniz’s fifth letter to Clarke; translation, ibid.
26
Theodicy, §6 of the appendix “Observations on the Book Concerning ‘The Origin of
Evil’ Recently Published in London;” p. 410 in Farrer (ed.), Leibniz.
27
See §41 of his fourth letter to Clarke and §106 of his fifth letter. See also his letter to
Bourget of 2 July, 1716; Robinet (ed.), Correspondance, p. 118.
appendix d 181
But caution is required here: it is far from obvious that in speaking of void
space as imaginary in his letters to Clarke Leibniz meant to indicate that it
was impossible. (i) As Leibniz indicates in one of the passages under
discussion, ‘imaginary space’ was a Scholastic term of art—one which
had, and was known to have, a dizzying array of established meanings by
the time Leibniz was writing.31 For example, for some prominent authors,
such as Suárez, the distinction between real and imaginary space was the
distinction between space occupied by body and empty space capable of
being occupied by body.32 (ii) Leibniz himself sometimes uses ‘imaginary’
28
§7 of Leibniz’s fourth letter to Clarke. Translation of Ariew (ed.), Leibniz and Clarke.
29
§33 of Leibniz’s fifth letter to Clarke; translation, ibid. See also Leibniz’s letters to
Rémond of 27 March 1716 (Robinet (ed.), Correspondance, pp. 61 f.) and to Des Bosses of 29
May 1716 (Ariew and Garber (eds.), Leibniz, pp. 201–6).
30
Theodicy, §365. Translation of Farrer (ed.), Leibniz, p. 343.
31
See Grant, Much Ado About Nothing, chs. 6 and 7, esp. pp. 120 f.
32
Grant, Much Ado, §7.2. Bayle appears to have taken this for the standard use: see remark
G of the article on Leucippus in the Historical and Critical Dictionary. Gassendi likewise follows
this use—although that does not stop him from going on to speak of imaginary space as a
chimera (see the passages quoted on pp. 110 and 121 f. in Lolordo, Pierre Gassendi and the Birth
of Early Modern Philosophy). Further, Leibniz himself appears to have followed this use in notes
written in 1676 (Parkinson (ed.), De Summa Rerum, p. 77). I mention all of this in order
182 appendix d
to drive home the point that one should not jump to conclusions about the sense of
‘imaginary space.’ It is of course clear that Leibniz himself was not following this use in the
passages cited in fnn. 28 and 29 above—cf. esp. §29 of his fifth letter to Clarke.
33
See “On the Method of Distinguishing Real from Imaginary Phenomena.” In Loemker
(ed.), Leibniz, pp. 363–6.
34
Letter to Queen Sophie Charlotte of Prussia; Ariew and Garber (eds.), Leibniz, pp. 186–92.
For discussion and references concerning this theme, see McRae, “The Theory of Knowledge,”
pp. 178–86. Note that Leibniz also speaks of space as imaginary in his memorandum on
Copernicanism and relativity of motion (see Ariew and Garber (eds.), Leibniz, p. 91). It seems
plausible that this use has its roots in Leibniz’s account of mathematics.
35
All material quoted in this paragraph is from §ii.xv.11 and follows the translation of
Remnant and Bennett (eds.), Leibniz.
36
(1) How should we understand the claim that geometry forbids us from taking the poles
of the empty sphere to touch? (2) If Leibniz is allowing, as he seems to be, that there could be
worlds that differ only as to the empirically inaccessible length of a certain period of
changelessness, how can this be reconciled with the sort of verificationist sentiment that he
gives vent to in his correspondence with Clarke (see esp. §52 of Leibniz’s fifth letter)?
appendix d 183
Applying the same standards as were used above in arguing that Leibniz
is committed to the possibility of void space, it appears that we must
conclude that Leibniz here commits himself to the possibility of worlds
in which space has only a single dimension. And of course if this is
accepted then we must abandon the thesis that Leibniz takes the structure
of space to be the same across possible worlds.
A couple of possible responses suggest themselves. One option is to read
Leibniz’s remarks concerning the situation in which space is a line and
bodies are immobile as referring to a world in which the matter distribu-
tion is a one-dimensional continuum while space, as usual, has the struc-
ture of Euclidean three-space.
Another possibility is to note that Leibniz is perfectly capable in the
course of technical discussions of getting carried away and saying things
that he ought not to—things that are inconsistent with principles that he
holds dear.37 So perhaps here we have a case in which he says something
carelessly which deeper consideration would lead him to retract—the
remark in question is, after all, a fairly casual one.
This suggestion can be bolstered by considering a passage in which
Leibniz very conveniently addresses the question why our world is three-
dimensional. The context is provided by a puzzle raised by Bayle—why
should matter have three dimensions rather than, say, two or four?38
Leibniz notes that Bayle himself appears to expect that the answer should
lie in God’s will. Leibniz denies this and makes much of his denial. On
his view, that matter should have three dimensions follows not from
considerations of what sort of world is best but rather from a “geometrical
and blind necessity.”39
This provides very strong evidence for the verdict that on Leibniz’s
considered view, the dimension of space is invariant across possible worlds.
Unfortunately, the case is not quite conclusive—there are at least two sorts
of worry that one might consider grounds for appeal.
(1) Consider the reason that Leibniz offers for taking matter to be
necessarily three-dimensional:
37
Example: At one point he suggests that if there were only one material point in the
world, its trajectory through space would be a straight line; “Reply to the Thoughts on the
System of Preestablished Harmony Contained in the Second Edition of Mr Bayle’s Critical
Dictionary, Article Rosarius.” See Loemker (ed.), Leibniz, p. 577.
38
Theodicy, §351.
39
Translation of Farrer (ed.), Leibniz, p. 336.
184 appendix d
the ternary number is determined for it not by the reason of the best, but by a
geometrical necessity, because the geometricians have been able to prove that there
are only three straight lines perpendicular to one another which can intersect at one
and the same point.40
1
See Birkhoff, “Metric Foundations of Geometry. I.”
2
See Remark B.3 of Appendix B above. Recall that a metric space X is locally compact if
sufficiently small open metric balls have compact closures. For path metric spaces, local
compactness is equivalent to the compactness of all closed bounded subsets; see Plaut, “Metric
Spaces of Curvature $ k,” theorem 8.
188 appendix e
this upper bound. Being told that there is a one-particle world with the
spatial geometry of the punctured plane leaves open infinitely many quali-
tatively distinct possibilities—parameterized by how far the occupied point
is from the “missing point of space.” (Of course, if we had begun instead by
removing an asymmetric subset from the plane, then every remaining point
would have played a distinct geometric role.) &
3
Indeed, generic Riemannian manifolds have no non-trivial symmetries. For references
and discussion, see Blair, “Spaces of Metrics and Curvature Functionals,” §§1.1 f.
4
For Riemannian manifolds, two-point homogeneity is equivalent to isotropy (the
condition that for any point and any two unit tangent vectors at that point, there be an
isometry that fixes the point and whose tangent map sends the first of the given tangent
vectors to the second). See Wolf, Spaces of Constant Curvature, lemma 8.12.1.
appendix e 189
It is not hard to find spaces that are homogeneous but not two-point
homogeneous.
Example E.5 (The Cylinder). Take the vertical strip of the Euclidean plane
bounded by the lines x ¼ 1 and x ¼ 1 and glue the edges together by
identifying points on the boundary via (y, 1) $ (y, 1). The result is a
cylinder, which we will picture as being embedded in Euclidean three-
space and as having a vertical axis of symmetry. So the horizontal sections
are circles. This space is homogeneous: rotations about the axis and
translations along the axis both count as symmetries, and by composing
these we can map any point to any other point. But it is anisotropic and
hence not two-point homogeneous: let x be any point and y and z both be
separated from x by a unit of distance with x and y lying on a horizontal
line and x and z lying on a vertical line; then no isometry can fix x while
mapping y to z, for any such isometry would have to induce an isometric
mapping from a circle (the geodesic joining x and y) to an infinite line (the
geodesic joining x to z). If we have a two-particle world with cylindrical
spatial geometry, then in order to fix the totality of qualitative geometric
facts about matter we have to be told not only how far apart the particles
are, but also what direction they lie in from one another. &
r 3 k
2r þ ...
3
(where the ellipsis indicates higher-order terms in r).9 We call k(x, Õ) the
sectional curvature of Õ. We can define spaces of constant curvature as those
in which k(x, Õ) doesn’t depend on either x or Õ.10 It turns out that any
Riemannian manifold M in which for each x 2 M, k(x, Õ) is independent
of Õ is in fact a space of constant curvature.11
Recall next that if f : M ! M is a smooth map from a manifold to itself
that fixes a given x 2 M, then f induces a linear map f* on the tangent space
at x (think of the way that a rotation that fixes a given point in Euclidean
space nonetheless acts non-trivially on the space of directions at that point).
Since the sectional curvature associated with a tangent plane is definable in
terms of the metric structure of the manifold, if f : M ! M is an isometry that
fixes x 2 M, then f* must map any tangent plane at x to a tangent plane with
the same sectional curvature. Putting this together with the last point of the
preceding paragraph, we see that in order to show that a Riemannian
manifold (M, g) is a space of constant curvature, it suffices to show that for
each point x and each pair of tangent planes Õ1 and Õ2 at x, there is an
isometry f of (M, g) such that: (a) f fixes x; and (b) f* maps Õ1 to Õ2.
Let us now turn to the case of interest. Let (X, d) be a labile Riemannian
manifold and let x be any point in X and Õ1 any tangent plane at x. We can
choose > 0 so that any point x0 within a ball of radius 6 of x is connected
to x by a unique geodesic segment ‘(x, x0 ) that remains within the ball.12
9
See e.g. Bishop and Goldberg, Tensor Analysis on Manifolds, §5.14.
10
This condition is equivalent to being a connected Riemannian locally isometric to one
of the elementary geometries; see e.g. Wolf, Spaces of Constant Curvature, §2.4.
11
See e.g. Wolf, Spaces of Constant Curvature, corollary 2.2.7.
12
It suffices that 6 be less than the injectivity radius of X at x; see e.g. §§5.5 and 5.9 of
Petersen, Riemannian Geometry.
192 appendix e
14
See Brehm, “The Shape Invariant of Triangles and Trigonometry in Two-Point
Homogeneous Spaces.” The same holds true for the other two-point homogenous spaces
that are not spaces of constant curvature.
15
See Brehm and Et-Taoui, “Congruence Criteria for Finite Subsets of Complex Pro-
jective and Complex Hyperbolic Spaces,” propositions 1 and 2. For results concerning the
quaternionic case, see Brehm and Et-Taoui, “Congruence Criteria for Finite Subsets of
Quaternionic Elliptic and Quaternionic Hyperbolic Spaces.”
16
For this example, see Seidel, “Discrete Non-Euclidean Geometry,” p. 877.
194 appendix e
that (x1 , y1 , z1 ) and (x2 , y2 , z2 ) are two triples of points that satisfy the given
constraints. Note that for each triple of points, (xi , yi , zi ), there are tangent
vectors vi and wi at xi with lengths a and b (respectively) that make an angle
Ł, and such that one can reach yi or zi by travelling for a unit of time along
the geodesic at xi with tangent vector vi or wi . In order to show that
(x1 , y1 , z1 ) and (x2 , y2 , z2 ) are superposable, it suffices to show that there is
an isometry of the elliptic plane that maps x1 to x2 and whose tangent map
sends v1 to v2 and w1 to w2 . Since the elliptic plane is homogeneous, we
can without loss of generality take x1 ¼ x2 . So our question is whether
given two pairs of tangent vectors at x1 , such that the corresponding
members of each pair have equal length and such that the angles formed
by the members of each pair are equal, there exists an isometry of the
elliptic plane that fixes x1 and whose tangent map sends each element of
the first pair onto the corresponding member of the second pair. At this
point, it is helpful to switch to a different way of thinking of the elliptic
plane: as the upper half of the unit sphere, with antipodal points on the
equator identified. Since the elliptic plane is homogeneous, we can with-
out loss of generality take x1 to be the North Pole. Isometries of the elliptic
plane that fix x1 are given by rotations of the hemisphere about the
North–South axis and by reflections in vertical planes that pass through
the North Pole.21 By employing a map of the first sort, then one of the
second, we can map v1 to v2 and w1 to w2 .
Interestingly, it is almost true that the superposability type of a region in
the elliptic plane is specified by specifying the congruence type of each of
its three-point subregions.22 &
21
See Busemann and Kelly, Projective Geometry, pp. 214 f.
22
See Blumenthal, Theory and Applications of Distance Geometry, §§81, 92, 93, and 97.
23
For the results mentioned in this paragraph, see Blumenthal, “Congruence and Super-
posability in Elliptic Space” or Theory and Applications, ch. x.
24
So there is a sense in which the problem is a global one: locally, of course, the elliptic
space looks like a spherical space (which is fully homogeneous); for certain ‘large’ configur-
ations, one can exploit the global structure of an elliptic space to find congruent but non-
superposable configurations.
196 appendix e
25
How can a path metric space fail to be locally compact? e.g., by having a lot of holes
inconveniently arranged or by being infinite-dimensional. If one deletes from the R2 a
sequence of points that converges to (but does not include) the origin, then the resulting
space is still a path metric space but fails to be locally compact—because no closed metric ball
around the origin is complete. Any Hilbert space h is a path metric space; h is locally
compact if and only if finite-dimensional (Abraham et al., Manifolds, Tensor Analysis, and
Applications, proposition 2.1.11).
26
For discussion and references, see Busemann, Recent Synthetic Differential Geometry, §19;
and Freudenthal, “Lie Groups in the Foundations of Geometry,” §§2.19 ff.
appendix e 197
27
Ovchinnikov, “Homogeneity Properties of Some ‘1 -Spaces,” theorem 2.2.
28
Birkhoff gives this example, and mentions that the Urysohn space has the same feature;
“Metric Foundations,” §6.
29
Birkhoff asserts that lability is weaker than full homogeneity, but gives no examples;
“Metric Foundations,” §6. I have been unable to find any in the literature, so I leave this as a
challenge to the reader.
This page intentionally left blank
References
Abraham, R., J. Marsden, and T. Ratiu, Manifolds, Tensor Analysis, and Applications
(2nd edn.). Springer–Verlag: 1988.
Adams, R., Leibniz: Determinist, Theist, Idealist. Oxford University Press: 1994.
—— “Presumption and the Necessary Existence of God.” Noûs 22: 1988, 19–32.
Algra, K., “Posidonius’ Conception of the Extra-Cosmic Void: The Evidence and
the Arguments.” Mnemosyne 46: 1993, 473–505.
Archbold, J., “A Metric for Plane Affine Geometry Over GF(2n ).” Mathematika 7:
1960, 145–8.
Ariew, R. (ed.), G. W. Leibniz and Samuel Clarke: Correspondence. Hackett: 2000.
—— and D. Garber (eds.), G. W. Leibniz: Philosophical Essays. Hackett: 1989.
Armstrong, D., What is a Law of Nature? Cambridge University Press: 1983.
Bader, R., “Towards a Hyperintensional Theory of Intrinsicality.” Unpublished
manuscript: 2010.
Barnes, J., The Presocratic Philosophers (rev. edn.). Routledge: 1982.
Beebee, H., “The Non-Governing Conception of Laws of Nature.” Philosophy and
Phenomenological Research 61: 2000, 571–94. Reprinted in J. Carroll (ed.),
Readings on Laws of Nature. University of Pittsburgh Press: 2004, 250–76.
Benedetti, R. and C. Petronio, Lectures on Hyperbolic Geometry. Springer–Verlag:
1992.
Berger, M., “Une caractérisation purement métrique des variétés riemanniennes à
courbure constante.” In P. Butzer and F. Fehér (eds.), E. B. Christoffel: The
Influence of His Work on Mathematics and the Physical Sciences. Birkhäuser: 1981,
480–92.
—— Riemannian Geometry During the Second Half of the Twentieth Century. American
Mathematical Society: 2000.
Bessaga, C. and A. Pełczyński, Selected Topics in Infinite-Dimensional Topology.
Pańtstwowe Wydawnictwo Naukowe: 1975.
Bigelow, J., B. Ellis, and C. Lierse, “The World as One of a Kind: Natural
Necessity and Laws of Nature.” British Journal for the Philosophy of Science 43:
1992, 371–88. Reprinted in J. Carroll (ed.), Readings on Laws of Nature.
University of Pittsburgh Press: 2004, 141–60.
Bird, A., Nature’s Metaphysics: Laws and Properties. Oxford University Press: 2007.
Birkhoff, G., “Metric Foundations of Geometry. I.” Transactions of the American
Mathematical Society 55: 1944, 465–92.
Bishop, R. and S. Goldberg, Tensor Analysis on Manifolds. Dover: 1980.
200 references
Burnyeat, M., “The Sceptic in His Place and Time.” In R. Rorty, J. B. Schnee-
wind, and Q. Skinner (eds.), Philosophy in History. Cambridge University Press:
1984, 225–54. Reprinted in M. Burnyeat and M. Frede (eds.), The Original
Sceptics: A Controversy. Hackett: 1997, 92–126.
Busemann, H., The Geometry of Geodesics. Academic Press: 1955.
—— Recent Synthetic Differential Geometry. Springer–Verlag: 1970.
—— “Timelike Spaces.” Dissertationes Mathematicae 53: 1967, 5–50.
—— and P. Kelly, Projective Geometry and Projective Metrics. Academic Press: 1953.
Butterfield, J., “Against Pointillisme about Geometry.” In F. Stadler and M. Stöltzner
(eds.), Time and History: Proceedings of the 28th. International Ludwig Wittgenstein
Symposium, Kirchberg am Wechsel, Austria 2005. Ontos Verlag: 2006, 181–222.
—— “The Hole Truth.” British Journal for the Philosophy of Science 40: 1989, 1–28.
—— “Relationism and Possible Worlds.” British Journal for the Philosophy of Science
35: 1984, 101–13.
Callender, C., “Answers in Search of a Question: ‘Proofs’ of the Tri-Dimensionality
of Space.” Studies in History and Philosophy of Modern Physics 36: 2005, 113–36.
Carroll, J., Laws of Nature. Cambridge University Press: 1994.
—— “Nailed to Hume’s Cross?” In T. Sider, J. Hawthorne, and D. Zimmerman
(eds.), Contemporary Debates in Metaphysics. Blackwell: 2008.
Cartwright, N., The Dappled World: A Study of the Boundaries of Science. Cambridge
University Press: 1999.
Casati, R. and A. Varzi, Parts and Places: The Structures of Spatial Representation. MIT
Press: 1999.
Cassorla, M., “Approximating Compact Inner Metric Spaces by Surfaces.” Indiana
University Mathematics Journal 41: 1992, 505–13.
Charleton, W., Physiologia Epicuro-Gassendo-Charltoniana. Facsimile edn. Johnson
Reprint Corporation: 1966.
Clifford, W., Lectures and Essays (2nd edn.). MacMillan: 1886.
Cohen, J. and C. Callender, “A Better Best System Account of Lawhood.”
Philosophical Studies 145: 2009, 1–34.
Coxeter, H. S. M., Non-Euclidean Geometry (2nd edn.). University of Toronto
Press: 1947.
—— Projective Geometry. Blaisdell: 1964.
Dafermis, M., “General Relativity and the Einstein Equations.” In T. Gowers,
J. Barrow–Green, and I. Leader (eds.), The Princeton Companion to Mathematics.
Princeton University Press: 2008, 483–93.
Divers, J. and J. Melia, “The Analytic Limit of Genuine Modal Realism.” Mind
111: 2002, 15–36.
Dorr, C., “Finding Ordinary Objects in the World of Quantum Mechanics.”
Unpublished manuscript: 2006.
202 references
Grant, E., “The Condemnation of 1277, God’s Absolute Power, and Physical
Thought in the Late Middle Ages.” Viator 10: 1979, 211–49.
—— “Medieval and Seventeenth-Century Conceptions of Infinite Void Space
Beyond the Cosmos.” Isis 60: 1969, 39–60.
—— “The Medieval Doctrine of Place: Some Fundamental Problems and
Solutions.” In A. Maier and A. Paravincini Bagliani (eds.), Studi sul XIV secolo in
memoria de Anneliese Maier. Edizioni di Storia e Letteratura: 1981, 57–79.
—— Much Ado About Nothing: Theories of Space and Vacuum from the Middle Ages to
the Scientific Revolution. Cambridge University Press: 1981.
—— “Place and Space in Medieval Physical Thought.” In P. Machamer and
R. Turnbull (eds.), Motion and Time, Space and Matter. Ohio State University
Press, 1976, 137–67.
Greenstreet, W. (trans.), Henri Poincaré: Science and Hypothesis. Dover: 1952.
Gromov, M., Metric Structures for Riemannian and Non-Riemannian Spaces. Birkhäuser:
2001.
Grove, K., Riemannian Geometry: A Metric Entrance. University of Aarhus: 1999.
Hahm, D., The Origins of Stoic Cosmology. Ohio State University Press: 1977.
Halpin, J., “Scientific Law: A Perspectival Account.” Erkenntnis 58: 2003, 137–68.
Harman, G., “The Inference to the Best Explanation.” Philosophical Review 74:
1965, 88–95.
Hawthorne, J., “Causal Structuralism.” Philosophical Perspectives 15: 2001, 361–78.
Reprinted in J. Hawthorne, Metaphysical Essays. Oxford University Press: 2006,
211–27.
—— “Quantity in Lewisian Metaphysics.” In J. Hawthorne, Metaphysical Essays.
Oxford University Press: 2006, 229–37.
Heil, J., “Dispositions.” Synthese 144: 2005, 343–56.
Heinonen, J., Geometric Embeddings of Metric Spaces. University of Jyväskylä Depart-
ment of Mathematics and Statistics: 2003.
Henkin, L., P. Suppes, and A. Tarski (eds.), The Axiomatic Method: With Special
Reference to Geometry. North–Holland: 1959.
Higginbotham, J., “Truth and Understanding.” Philosophical Studies 65: 1992, 3–16.
Hodges, W., A Shorter Model Theory. Cambridge University Press: 1997.
Howson, C. and P. Urbach, Scientific Reasoning: The Bayesian Approach (2nd edn.).
Open Court: 1989.
Hubička, J. and J. Nešetřil, “A Finite Presentation of the Rational Urysohn
Space.” Topology and its Applications 155: 2008, 1483–92.
Huggett, N., “The Regularity Account of Relational Spacetime.” Mind 115: 2006,
41–73.
—— True Motion. Unpublished manuscript: 2009.
Hušek, M., “Urysohn Universal Space, its Development and Hausdorff ’s
Approach.” Topology and its Applications 155: 2008, 1493–1501.
references 205
Lerner, R. and M. Mahdi (eds.), Medieval Political Philosophy: A Source Book. Cornell
University Press: 1963.
Lewis, D., Counterfactuals. Blackwell: 1973.
—— “Humean Supervenience Debugged.” Mind 103: 1994, 473–90. Reprinted
in D. Lewis, Papers in Metaphysics and Epistemology. Cambridge University Press:
1999, 224–7.
—— “New Work for a Theory of Universals.” Australasian Journal of Philosophy 61:
1983, 343–77. Reprinted in D. Lewis, Papers in Metaphysics and Epistemology.
Cambridge University Press: 1999, 8–55.
—— On the Plurality of Worlds. Blackwell: 1986.
—— Parts of Classes. Blackwell: 1991.
—— Philosophical Papers, vol. ii. Oxford University Press: 1986.
—— “Ramseyan Humility.” In D. Braddon–Mitchell and R. Nola (eds.),
Conceptual Analysis and Philosophical Naturalism. MIT Press: 2009, 203–22.
—— “A Subjectivist’s Guide to Objective Chance.” In R. Jeffrey (ed.), Studies in
Inductive Logic and Probability, vol. ii. University of California Press: 1980,
263–94. Reprinted with additional postscripts in D. Lewis, Philosophical Papers,
vol. ii. Oxford University Press: 1986, 83–132.
—— “Tensing the Copula.” Mind 111: 2002, 1–13.
Lindberg, D., The Beginnings of Western Science. University of Chicago Press: 1992.
Lipton, P., Inference to the Best Explanation (2nd edn.). Routledge: 2004.
—— “Is Explanation a Guide to Inference? A Reply to Wesley C. Salmon.” In
G. Hon and S. Rakover (eds.), Explanation: Theoretical Approaches and Applica-
tions. Kluwer: 2001, 93–120.
Loemker, L. (ed.), Gottfried Wilhelm Leibniz: Philosophical Papers and Letters
(2nd edn.). D. Reidel: 1969.
Loewer, B., “Humean Supervenience.” Philosophical Topics 24: 1996, 101–27.
Reprinted in J. Carroll (ed.), Readings on Laws of Nature. University of Pittsburgh
Press: 2004, 176–206.
—— “Laws and Induction.” Unpublished manuscript: 2000.
—— “Time and Law.” Unpublished manuscript: 2004.
Logothetti, D., “An Interview with H. S. M. Coxeter, the King of Geometry.”
The Two-Year College Mathematics Journal 11: 1980, 2–19.
Lolordo, A., Pierre Gassendi and the Birth of Early Modern Philosophy. Cambridge
University Press: 2007.
Look, B. and D. Rutherford (eds.), The Leibniz–Des Bosses Correspondence. Yale
University Press: 2007.
Lowe, E., The Four-Category Ontology: A Metaphysical Foundation for Natural Science.
Oxford University Press: 2006.
McDaniel, K., “Distance and Discrete Space.” Synthese 155: 2007, 157–62.
references 207
McDaniel, K., “Modal Realism with Overlap.” Australasian Journal of Philosophy 82:
2004, 137–52. Reprinted in F. Jackson and G. Priest (eds.), Lewisian Themes: The
Philosophy of David K. Lewis. Oxford University Press: 2004, 140–55.
McKeon, R. (ed.), The Basic Works of Aristotle. Random House: 1941.
McRae, R., “The Theory of Knowledge.” In N. Jolley (ed.), The Cambridge
Companion to Leibniz. Cambridge University Press: 1995, 176–98.
Maudlin, T., “Buckets of Water and Waves of Space: Why Space-Time is Probably
a Substance.” Philosophy of Science 60: 1993, 183–203.
—— The Metaphysics within Physics. Oxford University Press: 2007.
—— “Time, Topology, and Physical Geometry.” Proceedings of the Aristotelian
Society suppl. vol. 84: 2010, 63–78.
Melia, J., “Weaseling Away the Indispensability Argument.” Mind 109: 2000, 455–79.
Menger, K., “The New Foundations of Hyperbolic Geometry.” In J. Butcher (ed.),
A Spectrum of Mathematics: Essays Presented to H. G. Forder. Auckland University
Press: 1971, 86–97. Reprinted in B. Schwiezer, A. Sklar, K. Sigmund, P. Gruber,
E. Hlawka, L. Reich, and L. Schmetterer (eds.), Karl Menger: Selecta Mathematica,
vol. i. Springer: 2002, 495–506.
Monton, B. and B. van Fraassen, “Constructive Empiricism and Modal Nominalism.”
British Journal for the Philosophy of Science 54: 2003, 405–22.
Morison, B., On Location: Aristotle’s Concept of Place. Oxford University Press: 2002.
Mulder, H. and B. van de Velde-Schlick (eds.), Moritz Schlick: Philosophical Papers,
vol. i. Kluwer: 1979.
Mundy, B., “On Quantitative Relationist Theories.” Philosophy of Science 56: 1989,
582–600.
—— “Space-Time and Isomorphism.” In A. Fine, M. Forbes, and L. Wessels
(eds.), PSA 1990: Proceedings of the Biennial Meeting of the Philosophy of Science
Association, vol. 1. Philosophy of Science Association: 1992, 515–27.
Nerlich, G., The Shape of Space (2nd edn.). Cambridge University Press: 1994.
—— “Space-Time Substantivalism.” In M. Loux and D. Zimmerman (eds.), The
Oxford Handbook of Metaphysics. Oxford University Press: 2003, 281–314.
Nolan, D., “Recombination Unbound.” Philosophical Studies 84: 1996, 239–62.
Ott, W., Causation and Laws of Nature in Early Modern Philosophy. Oxford University
Press: 2009.
Ovchinnikov, S., “Homogeneity Properties of Some ‘1 -Spaces.” Discrete and
Computational Geometry 35: 2006, 301–10.
Pambuccian, V., “Axiomatizations of Hyperbolic Geometry: A Comparison Based
on Language and Quantifier Type Complexity.” Synthese 133: 2002, 331–41.
Erratum: “Correction to ‘Axiomatizations of Hyperbolic Geometry.” Synthese
145: 2005, 497.
—— “Simplicity.” Notre Dame Journal of Formal Logic 29: 1988, 396–411.
Pargetter, R., “Laws and Modal Realism.” Philosophical Studies 46: 1984, 335–47.
208 references
Skow, B., “Earman and Roberts on Empiricism about Laws.” Philosophy and
Phenomenological Research 75: 2007, 158–62.
—— Once Upon a Spacetime. Unpublished NYU doctoral dissertation: 2005.
Soames, S., “Truth, Meaning, and Understanding.” Philosophical Studies 65: 1992,
17–35.
Sober, E., “Parsimony and Predictive Equivalence.” Erkenntnis 44: 1996, 167–97.
—— Reconstructing the Past: Parsimony, Evolution, and Inference. MIT Press: 1988.
Sorabji, R., Matter, Space and Motion: Theories in Antiquity and Their Sequel.
Duckworth: 1988.
Sorenson, R., “Nothingness.” Stanford Encyclopedia of Philosophy: 2009 (online
publication).
Sormani, C., “How Riemannian Manifolds Converge: A Survey.” Unpublished
manuscript, 2010.
Stein, H., “Newtonian Space-Time.” Texas Philosophical Quarterly 10: 1967. 174–200.
Reprinted in R. Palter (ed.), The Annus Mirabilis of Isaac Newton. MIT Press: 1970,
253–84.
—— “Some Philosophical Prehistory of General Relativity.” In J. Earman,
C. Glymour, and J. Stachel (eds.), Foundations of Space-Time Theories. University
of Minnesota Press: 1977, 3–49.
Swoyer, C., “The Nature of Natural Laws.” Australasian Journal of Philosophy 60:
1982, 203–23.
Tarski, A., “What is Elementary Geometry?” In L. Henkin, P. Suppes, and
A. Tarski (eds.), The Axiomatic Method: With Special Reference to Geometry.
North–Holland: 1959, 16–29.
Todd, R., “Cleomedes and the Stoic Concept of the Void.” Apeiron 16: 1982,
129–36.
Tooley, M., “A Defense of Absolute Simultaneity.” In W. Craig and Q. Smith
(eds.), Einstein, Relativity, and Absolute Simultaneity. Routledge: 2008, 229–43.
—— “The Nature of Laws.” Canadian Journal of Philosophy 7: 1977, 667–98.
Reprinted in J. Carroll (ed.), Readings on Laws of Nature. University of Pittsburgh
Press: 2004, 38–70.
Torretti, R., Philosophy of Geometry from Riemann to Poincaré. D. Reidel: 1984.
Tweedale, M., “Armstrong on Determinable and Substantial Universals.” In
R. Bogdan (ed.), D. M. Armstrong. D. Reidel: 1984, 171–89.
Uspenskij, V., “The Urysohn Universal Metric Space is Homeomorphic to a
Hilbert Space.” Topology and its Applications 139: 2004, 145–9.
Vallentyne, P., “Explicating Lawhood.” Philosophy of Science 55: 1988, 598–613.
—— “Intrinsic Properties Defined.” Philosophical Studies 88: 1988, 209–19.
van Fraassen, B., Introduction to the Philosophy of Time and Space (2nd edn.). Columbia
University Press: 1985.
—— Laws and Symmetry. Oxford University Press: 1989.
references 211
Cauchy sequences 12, 14, 28, 31, 152, 153 modal relationalism 52
Cayley–Menger determinants 12, 129 necessitarianism 117–21
Charleton, W. 161, 163 n27, 169, non-quantitative relations 95–6
169 n49, 169 n51, 169 n52 substantivalism 41, 84, 85
Chrysippus 166 n38 summary 137
classical geometries: Divers, J. 30 n59
and distance 15–16 Dorr, C. 92 n16, 100 n24, 140 n2
and simplicity 64–6, 70 Drake, S. 171 n56
Cleomedes 165, 166 Dretske, F. 79 n1, 112 n28;
Clifford, W. 65 n28 see also Armstrong–Dretske–Tooley
Cohen, J. 72 n41 approach
collinearity 36, 41, 90, 95–7, 131 duplicate objects 51, 115–21
compact metric spaces 151–5, 196–7
compatibility properties 114–17, Earman, J. 18 n24, 51 n32, 60 n17,
119–21, 129, 132 61 n18, 75 n49, 75 n51, 110 n24,
complete metric spaces 14, 47, 153, 155 112 n28, 112 n29, 122 n46,
complex projective planes 189–90, 123 n47, 144 n11, 147 n16,
192–3 157 n1, 174 n4
Condemnation of 1277 167–8 edges, defined 23–4
congruence 10, 11, 32, 87–91, 97, 99, Eichhorn, J. 22 n40, 155 n16
186–97 Einstein, A. 35
conservative relationalism 37–49, 50, elementary geometries 16, 196
52 n35, 174–7 elliptic geometry:
constant curvature, spaces of 18–19, generally 16
88–9, 190–2 and balance 69–71, 73–4
continuum, space as 40 best-system approaches 65
convergence 12, 14, 23, 24, 150–6 and constant curvature 19
Copernicus 170 and dimension 36 n4
countable metric spaces 46–9 elliptic planes 193–6
Coxeter, H. S. M. 65, 66 and necessitarianism 131
crumpled spheres 88, 188 and primitivist approaches 90
cumulative limit 44 Ellis, B. 53, 106 n13, 107 n16,
curves: 110 n23
spaces of constant curvature 18–19, embeddability 42–4, 45, 47, 75, 131
88–9, 190–2 empty space 38 n8; see also void space
spaces of vanishing curvature 19 empty worlds 82, 100–1, 120 n42
cylinders 17, 89, 92, 120, 190; Epicurus 163 n25, 163 n27, 164 n32
see also twisted cylinders essential properties 126–7
Et-Taoui, B. 132, 193 n15
Dafermis, M. 63 n21 Euclidean geometry:
Descartes, R. 2, 38, 157–8, 169, generally 8–9, 12, 15
171, 174 and balance 69–70, 73–4
dimension 36; see also uni-dimensionality and constant curvature 18–19
discrete metric spaces 24, 69 and dimension 36 n4
distance relations: and distance 10–14
best-system approaches 57–8 generalizations of Euclidean
compatibility properties 131 geometry 10–30
distance and Euclidean and necessitarianism 119–21
geometry 10–14 and simplicity 64–5, 66
distances from discrete fields 29–30 and substantivalism 82
inhomogeneous spaces 88–9 vs taxi-cab metrics 23
metric spaces 31–4 extrinsic properties 115–16
index 215
one-particle worlds 5, 52–3, 79, 82, 102, Quine, W. V. 102, 123 n49, 123 n51,
119–21 125 n54, 139, 141
ontology 2, 139–49
Ott, W. 106 n15 random metric spaces 47
Ovchinnikov, S. 197 n27 ratios, of distance 11 n11
realism, defined 1–4
Pambuccian, V. 61 n20, 65 n27 recombination, principle of 104, 115,
paradoxes of Aristotelian motion 117, 122 n44, 124–5
160–1, 169 regions:
Pargetter, R. 80 n4 in metric spaces 28–9
parsimony 45–6 and necessitarianism 117–21
Parsons, J. 114 n31 primitivist approaches 84, 87
particle physics: Reichenbach, H. 147 n14
general principles 68, 74 relationalism:
one-particle worlds 5, 52–3, 79, 82, and Aristotle’s cosmos 18
102, 119–21 and necessitarianism 117–21
path metric spaces 22–4, 187 n2 and primitivist approaches 80–1
paths, defined 16 and Riemannian manifolds 20
Pełczynski, A. 47 n21 and spatial structures 35–53
Petersen, P. 21 n36, 23 n42, 152 n6, summary 134
154 n11, 154 n12, 155 n16, vs substantivalism 1–4
191 n11, 192 n13 see also modal relationalism
Petronio, C. 19 n30 reliability, of methods 146–8
planes: Remnant, P. 174 n6, 175 n7, 176 n13,
complex projective planes 189–90, 177 n17, 178 n18, 178 n19, 182 n35
192–3 Riemannian manifolds:
elliptic geometry 193–6 generally 17 n23, 19–24, 191
flat spaces 17 convergence 152–6
punctured planes 89, 94, 120, 187–8 and gappiness 31
taxi-cab metrics 23 and homogeneous spaces 89–90
Plaut, C. 156 n21, 187 n2 and Huggett 58 n12
plenitude, principle of 30–1, 153 and lability 187–96
Poincaré, H. 64–5, 66, 70 Riemannian metric tensors 21
pointed metric spaces 155–6 separability 47
pointless metric spaces 28–9 and total boundedness 151 n3
Pooley, O. 82 n7 Roberts, J. 60 n17, 72 n42, 75 n49,
Posidonius 164 n34 108 n17, 110 n24, 112 n28,
positive constant curvature, spaces of 18 112 n29, 122 n46, 123 n47, 144 n11
primitivist approaches 78–101, 135–6 Roberts, S. 65 n28
projective geometry 11, 66 Robinet, A. 180 n27
218 index