On (Some) Explanations in Physics1

James Owen Weatherall2

Logic and Philosophy of Science
University of California, Irvine

Abstract

I offer one possible explanation of why inertial and gravitational mass are equal in Newto-
nian gravitation. I then argue that this is an example of a kind of explanation that is not
captured by standard philosophical accounts of scientific explanation. Moreover, this form
of explanation is particularly important, at least in physics, because demands for this kind
of explanation are used to motivate and shape research into the next generation of physical
theories. I suggest that explanations of the sort I describe reveal something important about
one way in which physical theories can be related diachronically.

“What do we mean here by ‘explanation’ ? ... This whole issue, which perhaps lies between
nature and sociology, seems to be a bit vague. Quite possibly, an attempt to make the
word explanation more precise may do more harm to the field [of physics] than good.”

-Robert Geroch (1978, pg. 63)

1 Introduction

Consider the following questions, any of which might be heard in the halls of a physics

department.

1. Our best theory of particle physics predicts that in very high energy experiments, which
probe the smallest distance scales, the electromagnetic, weak, and strong forces should
have approximately the same strength. But at these same distance scales, gravitation
is many orders of magnitude weaker. Why is gravity so much weaker than any of the
other forces?

1I am indebted to Kyle Stanford, Dave Baker, John Manchak, Erik Curiel, and David Malament for
helpful remarks on earlier drafts of this paper. Thank you, also, to the Southern California Philosophy of
Physics group, in particular Jeff Barrett, Craig Callender, and Tarun Menon for a spirited and productive
discussion of the material presented here. I am also grateful to participants in the UC Irvine WIP seminar,
especially Bennett McNulty, Sam Fletcher, and Cailin O’Connor, for their comments. Finally, this paper
benefited from the helpful remarks of two anonymous referees.

2weatherj@uci.edu

1



2. The Standard Model of particle physics makes predictions that have been confirmed
to 15 significant digits (Odom et al., 2006). But the Standard Model’s predictions rely
on 19 parameters that are “put in by hand” to agree with experiment; in order for the
Standard Model to make accurate predictions at all, these parameters must be finely
tuned. Why do these Standard Model parameters take the values they do, and is there
a sense in which they are determined by some underlying mathematical or physical
principle?

3. In Newtonian physics,3 inertial mass (the value m that appears in F = ma) always has
the same value as gravitational mass (the coupling to the gravitational field, i.e., the
value m that appears in UG = mϕG, where ϕG is the gravitational potential and UG is
the potential energy of a particle with mass m), even though in principle the theory
distinguishes these masses. This equivalence is empirical: it was first established (in
slightly different terms) by Galileo; at the end of the 19th century, it was tested with
very high precision by Loránd Eötvös. Yet the correspondence seems highly suggestive.
Why are inertial and gravitational mass equal in Newtonian physics?

I need not multiply examples. Each of these is a why question asked in a particular scientific

context (physics). As such, I take it that they are calls for scientific explanation.4 Indeed,

they are why questions of a particularly important sort: these are the kind of questions that

physicists often use to motivate their research projects. Questions 1 and 2 are open and

form the basis of several major contemporary research programs5 in high energy particle

physics and quantum gravity. Question 3, meanwhile, has been settled, or at least, we now

have the theoretical machinery available to provide one sort of answer to it. I claim that the

answer one can now give in response to question 3 is an example of one kind of explanation

that would satisfy the physicists who ask questions 1 and 2. It may not be the only kind of

explanation that physicists would ultimately deem satisfactory, though I think it is an ideal

of the sort of explanation physicists have in mind when they ask questions 1 and 2. My

central goal in this paper will be to examine just what kind of explanation it is.

3I will use the expression “Newtonian theory” interchangeably with “Newtonian physics.” In both cases
I mean Newtonian dynamics plus gravitation.

4Perhaps not all calls for explanation take the form of why questions, and perhaps not all why questions
call for explanations. But I claim these why questions do call for explanations.

5I do not mean “research program” in a technical philosophical sense. I just mean that string theorists,
loop quantum gravity theorists, and many particle phenomenologists (in the physicist’s sense of phenomenol-
ogy) take these questions to be central to their research.

2



Before proceeding, however, I should give some context to the present discussion. Over

the last 20 years, since Salmon (1989) proposed a detente between the causal and unifica-

tionist accounts of explanation, the idea that some sort of pluralist account of explanation is

necessary to capture the full variety of explanatory phenomena has gained considerable sup-

port. Salmon’s own line was that the causal and unificationist accounts are not inconsistent.

Instead, he thought that they offer different kinds of understanding, corresponding to the

different kinds of explanation. On his view both the causal and unificationist accounts are

correct: the two accounts together offer a full account of explanation on which unificationist

explanations are “top-down” and causal explanations are “bottom-up.” Any given event or

phenomenon can be explained in both ways. But this form of pluralism is still too limiting.

It now seems that some explanations do not fit neatly into either account (see, for instance,

Batterman (2002)), and moreover, that some phenomena that are easily explained using one

kind of explanation do not, as Salmon suggests, have explanations of the other sort.6

More recently, Godfrey-Smith (2003) (following Kuhn (1977)) has suggested a different

kind of pluralism, in which what counts as a good explanation can vary depending on sci-

entific context. Explanation in biology need not be the same as explanation in physics, and

explanation in either field in the early 21st century need not be the same as explanation

was in, say, the 17th century. On this view, it is a mistake (as Geroch suggests above) to

attempt to characterize scientific explanation in advance: what counts as a good explanation

in science is evolving along with the sciences themselves. I am very sympathetic to this view.

But I take it that two kinds of project remain, even after a pluralistic, contexualist account

has been accepted. The first kind of project is to identify the working parts of contextu-

alist pluralism. Certainly, scientific context determines the array of explanations that are

available. But in many scientific contexts, such as contemporary physics, it seems that more

than one form of explanation is common. Moreover, as Salmon suggests, many phenomena

6I take Fisher (1930)’s explanation of sex ratio in humans as an example of this latter sort. (See also
Skyrms (1996, Ch. 1), where Fisher’s work is put in perspective.) A causal explanation can explain why any
individual turned out to have a particular gender, but it cannot explain why the ratio must be what it is.

3



may have explanations of radically different types. What makes a particular explanation an

appropriate one in a given instance, as an answer to a particular why question?7

The second kind of project, meanwhile, is to identify interesting explanations used in

various contexts and attempt to understand their epistemic and scientific virtues. The

current paper is an example of this second kind of project. I should say, however, that

once one has accepted some form of pluralism, merely taxonomizing the possible forms

of explanation is not fruitful. Nevertheless, some examples of explanation stand out as

particularly important for the practice of science. With this in mind, let me show my hand.

I will presently argue that the explanation I describe in the next section of this paper is not

well treated by the standard philosophical accounts of scientific explanation. This argument

will form the main thrust of the paper, but by itself the thesis need not be remarkable.

Instead, I take the principal moral of the present discussion to be that the explanation I

will describe, taken in conjunction with the many demands for such explanations that seem

to arise in the practice of contemporary physics (as witnessed by the two other questions

I mention above), reveals something important about one way in which physical theories

can be related diachronically. I will suggest that the expectation that certain cross-theoretic

explanatory demands will be met by future theories is an essential part of inquiry in physics

that is sometimes obscured by the traditional accounts of scientific explanation.

From here I will proceed as follows. I will start by clarifying what question 3 is asking. I

will then sketch what I take the answer to be.8 This question has been answered informally

in a variety of ways since General Relativity (GR) first appeared. The answer I will present

here is certainly in the spirit of these standard responses, though it precisifies a number of

details about the relationship between mass in GR and Newtonian physics that are usually

left vague. To my knowledge, the form of the answer I will present here is original and

may be of independent interest. After presenting the explanation I have in mind, I will

7I would suggest that the problem of identifying appropriate explanations is essentially pragmatic, but
will defer discussion of this point to future work.

8The discussion in the body of the paper is precise, but informal. The technical details of the explanation
are included in an appendix.

4



turn to the question of whether the explanation I offer here can be understood within the

rubrics of various well established accounts of explanation. I will conclude that it cannot.

In the remainder of the paper, I will try to articulate how the present explanation works,

highlighting its distinctive features and arguing for the moral suggested above.

2 Why are inertial and gravitational mass equal in Newtonian gravitation?

As I have said, inertial mass and gravitational mass are conceptually distinct in Newtonian

physics (I will distinguish them here by writing mI for inertial mass and mG for gravitational

mass). Indeed, one would expect them to be unrelated to one another. Inertial mass is

a constant of proportionality in the fundamental dynamical principles of the theory. It

appears in Newton’s second law, which states that F = mIa; momentum is defined as

p = mIv; kinetic energy is T = 1/2mIv
2. One can think of inertial mass as a measure

of a body’s tendency to accelerate under the influence of an impressed force. Inertial mass

is closely related to inertial motion, which enters Newtonian theory via Newton’s first law.

The first law states that a body undergoing uniform rectilinear motion will not deviate from

that motion unless acted on by an external force; inertial mass, then, determines a body’s

tendency to deviate from uniform rectilinear motion when acted on by an external force,

whether gravitational or otherwise.

Gravitational mass, meanwhile, determines the strength of the gravitational force that

a body experiences in a gravitational field.9 If a test body10 with gravitational mass mG is

placed in a gravitational potential ϕG, then the body will have gravitational potential energy

UG = mGϕG and will experience a force FG = −mG∇ϕG. Gravitational mass can be thought

of as gravitational charge, in analogy with classical electric charge. Indeed, the fundamental

9A distinction is sometimes made between “passive gravitational mass” and “active gravitational mass”.
Passive gravitational mass is what I have described above; active gravitational mass, meanwhile, can be
understood as a measure of the strength of the gravitational field produced by a body. The status of active
gravitational mass in geometrized gravitational theories is of some independent interest, but in the present
paper I am exclusively interested in passive gravitational mass. Whenever I use the expression “gravitational
mass,” the reader should assume I mean “passive gravitational mass.”

10By test body, I mean a body that is assumed not to contribute to the gravitational field itself. In other
words, when considering test bodies one neglects the “backreaction” of a body’s own gravitational field.

5



force equations have exactly the same structure. A test charge q in an electric potential ϕE

will have electrical potential energy UE = qϕE and will experience a force of FE = −q∇ϕE .

The parallel with electric force is particularly salient here. Suppose one wants to know

the acceleration exhibited by a test particle of charge q and inertial mass mI in an electric

potential ϕE . Combining Newton’s second law with the force law for a test particle in an

electromagnetic field, one finds that

a = −
q

mI
∇ϕE.

In other words, the acceleration depends on the ratio of the charge to the inertial mass of the

body, both of which are freely varying, independent objects. One can find in nature bodies

with many different values for the ratio q/mI. Meanwhile, if one performs the identical cal-

culation to determine the acceleration due to gravity (given a fixed gravitational potential),

one likewise finds,

a = −
mG
mI
∇ϕG. (2.1)

Again, the acceleration depends on the ratio of two values: the gravitational and inertial

masses. Given the structural similarities between the gravitational and electric cases, one

should expect to go out into the world and find bodies with a wide array of different values for

the ratio mG/mI. After all, (a) how much a body will tend to deviate from rectilinear motion

given an external force and (b) the strength of that external force should be independent

quantities. But when we start looking into how bodies behave in a gravitational potential,

we find something quite different. Given any body at all, the ratio mG/mI always takes the

same value: choosing the natural units, we always find that mG/mI = 1.

Given this background, the explanandum can be stated as follows: all evidence suggests

that, given any body, the gravitational and inertial masses of that body are equal, despite

the fact that Newtonian theory gives no reason to expect these two masses to be related. We

want to understand why this is. In some ways it is an unusual why question (at least with

6



respect to standard accounts of explanation), so I want to spend some time up front focusing

on its distinctive features. First off, it is a question about a general observational feature of

the world, but it is expressed in the terms of a specific physical theory. In other words, the

question takes the Newtonian concepts of gravitational and inertial mass for granted. One

can express the observational fact without reference to the Newtonian theory—Galileo first

described the phenomenon that all bodies fall at the same rate, irrespective of mass, before

Newton was born—but when one does so, the question does not arise. One might, perhaps,

wonder about the apparent universality of free fall in other contexts or even quite generally:

after all, Galileo’s results conflicted with the Aristotelian expectation, so one might well have

demanded an explanation for Galileo’s observations in the context of Aristotelian physics,

too. But without the background conceptual machinery of Newtonian physics, the question

is different. I am interested in a specific question about the world, expressed within the

Newtonian framework. It seems to me that questions 1 and 2 are similar in this regard,

mutatis mutandis.

This first feature suggests a second feature. Although the question is posed within the

Newtonian framework, and cannot be quite the same question if posed in other contexts,

it cannot be answered within the Newtonian framework. As we have seen, the Newtonian

concepts of gravitational and inertial mass are distinct. That the two masses always take the

same value for any given body is contingent, though suggestive, within Newtonian physics.

But this means that any appropriate answer will have to go beyond Newtonian physics. So

we have a question posed in one theory that can only be adequately answered by appealing

to another, presumably more general or fundamental, theory.

But what form could such an answer take? Einstein (1920) claimed that the observed

equivalence between gravitational and inertial mass was an important factor in his devel-

opment of GR, and that it can be taken as evidence in favor of the “general postulate of

relativity.” The principle that the two mass concepts should be identified, or equivalently

the idea that free fall does not depend on mass, is often called the weak equivalence principle

7



and continues to play a central role in some presentations of GR (see, for instance, Weinberg

(1972, ch. 1)). From this point of view, the observed equivalence is explained by asserting

that in a supervening theory (GR), no distinction is made between the two masses. Inertial

and gravitational masses are simply the same. But there is something strange, and poten-

tially misleading, about this answer to the original question. The reason for the difficulty is

that a more accurate account of the situation in GR, using only concepts native to GR itself,

would be to say that there is only inertial mass.11 It is not that gravitational mass is explic-

itly identified with inertial mass, but rather than gravitational mass has been stricken from

the theory altogether. There is no gravitational potential in GR, nor is there a gravitational

force, and so there is no parameter relating gravitational force to a background gravitational

field.12 When one attempts to answer the original question by appealing to some supposed

equivalence between the two kinds of mass in GR, one mixes terms from two theories in a

way that is dubious and confusing.

These considerations suggest another response to our original question. Given that grav-

itational mass does not make sense in GR, one might say that the question turns out to be

an error. We used to think that gravitation was a force (one might say), and that a body’s

gravitational mass determined the magnitude of the gravitational force exerted on the body

in a gravitational field. But now we know that gravitation is not a force at all and so ques-

tions about gravitational mass do not make any sense. This type of response is intended

to dissolve the question, by directing the questioner to a textbook on GR. But I claim that

this kind of response is unsatisfying. First, the question was asked in a specific framework;

to say that that framework is no longer widely accepted is irrelevant. Second, even if one

accepts that GR supercedes Newtonian physics, and one accepts moreover that gravitational

mass does not exist in GR, an important question remains. In Newtonian physics, which

everyone accepts as a predictively accurate theory in some regimes, there are two different

11And even inertial mass is problematic except in the case of a test body, unless a spacetime satisfies very
strong symmetry conditions.

12This point is put clearly, for instance, by Sachs (1976), though it has not always been recognized by the
physics community, as Weinberg (1972) attests.

8



kinds of mass. As such, I can point to two kinds of roles that one might expect mass to

play. Supposing that GR is correct that gravitational mass does not make sense, why is the

world so well-described, at least in certain regimes, by a theory that supports two concepts

of mass?13

I think this discussion helps to clarify both what the question really is and what kind

of answer would be appropriate. We have two fixed points to navigate between. First,

the question is such that it cannot be answered by the Newtonian theory. New physics is

required. But second, it is a question that needs to be answered in the terms in which it

was asked, i.e., within the Newtonian framework. As we have seen, using Newtonian terms

within the framework of GR leads to inconsistencies and serious confusion, while using

concepts native to GR at best allows a dissolutive response, rather than an answer, to the

question. With these two points in mind, one might rephrase the question once again, as

follows. Given that we now believe GR to have superseded Newtonian gravitation as our

best theory of large-scale dynamics and gravitation,14 why are gravitational and inertial mass

equal in Newtonian theory?15 This phrasing presupposes that even though GR has virtues

that Newtonian gravitation lacks, there are still regimes in which Newtonian gravitation

provides a satisfactory characterization of nature.

To answer the question, one needs to show, in detail, how Newtonian theory relates to

13Curiously, one can already imagine giving the same response to question 1: why should we compare
gravitation to the other three forces? After all, we already know that there is no such thing as gravitational
force! But this answer would be equally unsatisfying in that context, for the same reason: gravitation is
conceived of as a force in modern particle physics, which is at least part of the difficulty in making quantum
field theory and GR mesh. I think this serves to underscore the curious character of the questions above:
they are all expressed in the language of one theory, but one fully expects that the answer will come from a
theory in which the terms of the question may not make any sense.

14Some readers might balk at this point and argue that GR cannot provide any kind of explanations
whatsoever if it is not true. I do not share the intuition that only true theories have explanatory power. But
I do not think this issue is relevant here. Everything I say in what follows can be recast in conditional form:
explanations of the form “If GR were true, then...” are perfectly sufficient for a discussion of what kinds of
explanation are possible in science.

15Rephrasing the question in this way is only possible when one can point to the superseding theory. In
the cases of questions 1 and 2, no superseding theory is known. These questions might be rephrased in terms
of a future possible theory, or as a statement of a certain kind of research objective. One is looking for a new
theory T that can tell us why, given that theory T supersedes the Standard Model (say), the parameters in
the Standard Model take the particular values that they are observed to take.

9



GR. One needs to show why, if GR is true, Newtonian theory is such an effective theory

in some contexts. One way of doing this would be to show that Newtonian theory can be

reached from GR in an appropriate limit that captures the circumstances in which Newtonian

theory seems so effective.16 This observation suggests yet another refinement of the original

question. What we are really interested in is the following: Given that we now believe GR

to have superseded Newtonian theory, then why, insofar as Newtonian theory is a limiting

case of GR, are inertial and gravitational mass equal in Newtonian theory? I maintain

that this reformulation captures of the spirit in which the original question was asked, and

moreover, we finally have a question that is precise enough to answer. It is now clear that

the answer would involve trying to show that if Newtonian theory is taken as the limit of

GR in the appropriate way, then Newtonian theory does support two concepts of mass, and

moreover, that for any body these two masses must be equal. In other words, one shows

that gravitational mass arises in some way in the limiting process, and that the result must

be equal to inertial mass.17

It turns out that it is possible to perform this procedure exactly as described. (See ap-

16Many philosophers have questioned when and whether it is possible to show that an old theory reduces
to a new theory (in the philosopher’s parlance) or a new theory reduces to an old theory (in the language
of working physicists). Nagel (1961, 1970), for instance, treated intertheoretic reduction as explanation (in
the deductive-nomological sense) in a way that bears a rough family resemblance to what I am thinking
of; Nickles (1973), meanwhile, argues that often explanatory reductions are not possible at all (at least in
the DN sense). Curiously, Nickles identifies Newtonian physics and relativity theory as a prime example
of a reduction relationship that is not an explanation in the DN sense, but rather a collection of rough
intertheoretic relations. It seems to me that if any intertheoretic relationship deserves to be called deductive,
it is the relation between Newtonian theory and GR. But I do not intend to enter a debate on intertheoretic
reduction here. Instead, I want to distinguish identifying reduction and explanation (as Nagel does) from a
more ambiguous demand that a new theory explain, at least in some sense, why our old theories succeeded.
One way of cashing this requirement out is to say that a new theory cannot make predictions that are
inconsistent with the experimental successes of previous theories (in which case the new theory will at least
reveal regularities captured by the old theory).

17At the very least, the answer sketched would be responsive to the question when reformulated as relative
to GR. It seems to me that, given GR and the relationship between GR and Newtonian physics, the correct
way of thinking about Newtonian physics is through the lens of GR. But not all philosophers agree: there
is another way of thinking according to which Newtonian physics is an entirely unrelated (and false) theory
of spacetime and gravitation. All I can say in response is that I agree that if one understands Newtonian
physics in this way, then the explanation I offer is not satisfying. But I disagree that this is the most natural
way of understanding Newtonian physics. In particular, it seems to me that the relationship that one does
get between Newtonian physics and GR by understanding the former as a limiting case of the latter is highly
desirable from the point of view of practicing physics (and philosophy of physics).

10



pendix.) There is a precise sense in which Newtonian theory arises as a limit of GR (Künzle,

1976; Ehlers, 1981; Malament, 1986a).18 It involves a two step process. One begins by con-

sidering a one-parameter family of relativistic spacetimes, parametrized by some variable

λ. λ, at the present level of discussion, can be taken to reflect the inverse of the “speed of

light”19 in each of the spacetimes. By constructing this family of spacetimes carefully, one

can consider the limit that the spacetimes approach as λ approaches 0 (corresponding to

taking the speed of light to infinity). The result is a degenerate “classical” spacetime with

many of the characteristic features of Newtonian physics: space is always flat and Euclidean;

there is a unique sense of space at a time and absolute simultaneity holds; the spacetime is

Galilean relativistic, which means that (1) measurements of elapsed time and the distance

between simultaneous events will be the same for all observers, irrespective of their mo-

tion and (2) there is no absolute standard of rest. But we have not yet recovered Newtonian

physics. Rather, we have reached an intermediate point between GR and Newtonian physics,

18This footnote is in some sense a continuation of footnote 17; it also addresses a possible worry about
so-called “counterfactual” explanations. There is a way of thinking about Newtonian gravitation in light of
GR according to which describing the sense in which Newtonian theory is a limit of GR is tantamount to
describing the experimental conditions under which it is appropriate to use Newtonian laws for purposes of
approximation. This is not what I have in mind. When I describe Newtonian theory as a limiting case of
GR, I mean that given a fixed background manifold, (a) there is a precise sense in which classical spacetime
structure on that manifold can be understood as the limit of a sequence of relativistic spacetime structures,
and (b) when one takes this limit (with appropriate background assumptions), one recovers the correct
Newtonian laws in toto on the resulting classical spacetime. (See appendix.) There is at least one place in
which the two ways of thinking make contact, however. If, rather than considering experimental settings
in which particular Newtonian laws are appropriate approximations, one instead considers the experimental
circumstances under which one would be unable to identify failures of absolute simultaneity (perhaps because
the apparatus is too coarse grained to detect the finite speed of light), then it is appropriate (from an
experimental point of view) to assume that spacetime has a classical, Galilean structure. And if one were
to make such an assumption, it would then be appropriate to take the limit I have described to derive
the relevant approximate laws—hence “motivating” the explanatory power of the limit. I also take it that
describing the limit in this way should moderate worries that the limit “as c goes to infinity” cannot be
explanatory because it is counterfactual, since one can understand the limit as describing the case where for
all we can tell c is infinite (as was the situation in 1687).

19There is an abuse of language, here. Really, λ parametrizes something about the metrical structure of
the spacetime, specifically how wide the light cones are at a point. But the widths of the light cones indicates
how null vectors, which are the possible tangent vectors for the wordlines traversed by light, relate to the
timelike and spacelike vectors in the spacetime. Hence wider lightcones indicate “faster” light.

11



which I will call geometrized Newtonian theory.20,21 In geometrized Newtonian theory, grav-

itation is still geometrical: the geometry of spacetime is curved, with curvature determined

by the distribution of matter in the universe, and gravitational effects are manifestations of

the resulting geometry. Importantly, since gravitation is geometrical rather than a force be-

tween bodies, gravitational mass does not make any more sense in the context of geometrized

Newtonian theory than in GR.

To see where gravitational mass comes from, we need to take the second step in the

limiting process. This step makes use of a theorem due to Andrzej Trautman (Malament,

2010, Prop. 4.2.5). Trautman’s theorem tells us that, given a classical spacetime of the

sort found in geometrized Newtonian theory, satisfying certain conditions, one can find22 (1)

another spacetime that is flat,23 and (2) a scalar field ϕG that satisfies Poisson’s equation

(the relationship between the distribution of matter in the universe and the gravitational

potential in Newtonian gravitation) and which is such that for any free (inertial) massive

test point particle,

a = −∇ϕG. (2.2)

Thus, under certain circumstances, we can recover a flat spacetime and a gravitational

potential ϕG that has just the relations to both the distribution of matter in the universe

and the dynamics of a particle that we would expect from Newtonian physics, and moreover,

in this flat spacetime, the particle trajectories determined relative to the gravitational field

agree with the particle trajectories as determined by the geometrized theory in the initial

curved spacetime. More loosely, one finds a flat spacetime and gravitational field that “makes

20Geometrized Newtonian theory is sometimes called Newton-Cartan theory. It was first developed during
a lecture course by Élie Cartan in the early 1920s, as an attempt to understand how Newtonian gravitation
related to GR (Cartan, 1923, 1924). (See also Friedrichs (1927).)

21It may be helpful to emphasize that geometrized Newtonian theory is not a model within GR that is
somehow suggestive of classical physics. It is an independent theory with the empirical content of Newtonian
physics. Indeed, a classical spacetime as described here is not and could not be a relativistic spacetime
because it does not have an appropriate metric structure.

22See the appendix for details of this claim, and for a precise statement of the conditions a (geometrized)
classical spacetime needs to satisfy in order to recover standard Newtonian theory, which are not trivial.

23A classical spacetime as considered by geometrized Newtonian theory is generally curved, though space
is always flat. In standard Newtonian physics, spacetime taken as a whole has to be flat.

12



the same predictions” as the geometrized theory. We have now recovered full-blown, standard

Newtonian physics as a limit of GR.

I want to draw attention to an important feature of Eq. (2.2). It is a derived relation

between the acceleration of a particle and the gradient of the gravitational potential. If we

use F = mIa, Eq. (2.2) implies that the force on a massive test point particle arising from

the gravitational potential ϕG is

F = mIa = −mI∇ϕG. (2.3)

Eq. (2.3) tells us directly that the coupling to the gravitational field in Newtonian physics is

given by the inertial mass. Thus the reason that gravitational and inertial mass are always

equal is that gravitational mass simply is inertial mass. And thus we have an answer to the

original question.

3 What have I just done?

Now that I have offered an answer to one of my questions, I can ask what kind of explanation

I have given. It seems clear without further argument that this explanation is not a causal

explanation, in any of the senses of causal explanation that have been articulated over

the last few decades.24 It likewise does not fit into any of the earlier statistical accounts of

explanation, such as the statistical relevance account of explanation or the inducto-statistical

account.25 In the remainder of this section, I will focus on two other prominent accounts

of explanation that, at least prima facie, have a better chance of capturing the kind of

explanation I gave in the previous section: the deductive-nomological (DN) account and the

unificationist account.26

24Some prominent examples are of course Salmon (1984), but also Cartwright (1983) and Woodward (2003).
Strevens (2004, 2008) also offers a kind of causal account of explanation, though his “kairetic account” also
includes some of the desirable features of the unificationist account.

25For more on either of these, see (Salmon, 1989).
26Batterman (2002) also discusses a form of explanation that is not well-treated by the causal and unifi-

cationist accounts. He dubs it “asymptotic explanation.” I think that he has correctly identified a form of
explanation that the received accounts miss; however, I want to emphasize that the present example is differ-

13



3.1 The Deductive-Nomological Account

The DN account of explanation, originally proposed by Hempel and Oppenheim (1948), was

long the received view of explanation. On this account, an explanation is a (logical, first-

order) argument by which the thing to be explained, the explanandum, is deduced from a

set of true premises, the explanans. It is taken to be necessary that the explanans include

at least one law of nature; generally, it will also include particulars such as initial conditions

or boundary conditions. The thing to be explained is a proposition. The intuition is that

an explanation is a demonstration of law-like expectability. To explain a proposition is to

show that one could expect it to be true, given the laws of nature and some given set of

circumstances.27

Does the explanation I give in section 2 fit the DN mold? It certainly has many of the

central features of a DN explanation. The explanation consists of an argument by which the

explanandum is derived. The explanans is perhaps a bit broad: all of the central principles of

GR must be included in order to set up the limiting process necessary to recover geometrized

Newtonian theory. But among these principles are several law-like propositions. Finally,

there is a strong sense in which the argument’s explanatory power comes from the fact that

the result is derived from these central principles of GR, which means that the explanation

is nomological in an important way.

But not all is well. For one, the technical details of the explanation I describe above make

use of powerful mathematical tools. It is not at all clear that first-order logic has the resources

to express the full mathematical argument properly. A second worry concerns the language

ent from the examples Batterman offers. Asymptotic explanation involves explanations of “universality”—
properties that families of systems have at a given distance scale, irrespective of their microscopic details.
The explanation of such universal features involves the renormalization group, which is a method for moving
between different levels of description of a physical system. The form of explanation described here relies
on a very different kind of limiting procedure and concerns relations between theories, not between distance
scales. It may well turn out that the answers to questions 1 and 2 described at the beginning of this paper
will turn out to look more like Batterman’s examples. (This is especially likely for question 2.) But what is
important is that there are cases in which the kind of explanation I am describing is decidedly not asymptotic
explanation.

27See, as ever, Salmon (1989). I am glossing over many difficulties concerning what might count as a law
of nature and what kind of deduction is necessary.

14



in which the deduction is to be performed. If one were to begin with a language consisting

only of terms readily interpreted in GR, then gravitational mass would not appear and so the

proposition “inertial and gravitational mass are (always) equal” could not even be expressed.

Conversely, one might begin with an augmented language that includes gravitational mass

terms and then endeavor to show that gravitational mass is always equal to inertial mass via

a deduction in this augmented language. But it is difficult to see how to interpret such an

augmented language. In any case GR by itself would not provide a model for the language,

which means that even if the deduction were to succeed, it would not follow that one had

derived the explanandum by appeal to GR alone. Either way it seems that an essential part

of the explanation cannot be captured within the DN framework.

So the present explanation bears some family resemblance to DN explanations, though

it does not have just the form described by Hempel and Oppenheim (because it requires the

full resources of mathematical physics). More troubling, the DN account does not naturally

permit the final (important) step of the explanation. All that said, if one were committed

to the claim that all explanations must fit the DN model, it might be possible to extend

the model to include the present example. Alternatively, if disinclined to extend the DN

model, one might bite the bullet and claim that the example does not have any explanatory

power. I do not want to make either of these moves, however. Insofar as I already accept a

pluralistic view about explanation, I do not see any virtues in a procrustean reading of the

present explanation as an example of DN explanation.

There is another consideration here. Many writers have argued, I think successfully, that

Hempel and Oppenheim offer neither necessary nor sufficient conditions for an argument to

be explanatory (Godfrey-Smith, 2003; Salmon, 1989; Kitcher, 1989). In other words, even

if the present explanation does fit the DN model, it does not follow that it is explanatory

because it fits that model. One still needs to give an account of what makes a given argument

explanatory. A prominent attempt to describe what additionally may be required to make

an argument explanatory is given by the unificationist account of explanation, which I will

15



turn to presently.

3.2 The Unificationist Account

Kitcher (1989) provides the authoritative manifesto on the unificationist account of expla-

nation, so I will focus on the version of the account given there.28 The basic idea of the

unificationist account is that science aims to explain phenomena by showing how a phe-

nomenon fits into a unified systematization of our beliefs. Kitcher makes this idea precise

in the following way. Suppose that K is the set of all statements endorsed by the scien-

tific community. An acceptable explanation is a member of the explanatory store over K,

denoted E(K). How can we characterize the members of E(K)? First, we say that E(K)

is a set of arguments by which some members of K are derived from other members of K.

But in general there will be many such sets of arguments. To pick one, Kitcher introduces

general argument patterns, ordered triples consisting of (1) a schematic argument, which

are sequences of sentences with key terms replaced by dummy variables (i.e. sequences of

schematic sentences ); (2) sets of sets of filling instructions corresponding to each schematic

argument, where a given set of filling instructions tells you how to fill in the dummy variables

of a schematic sentence; and (3) a classification of the schematic argument, which tells you

which of the sentences in the argument are supposed to count as premises and which are

conclusions. One general argument pattern is more stringent than another if its classification

and the structure of its schematic sentences together make it more difficult to instantiate.

Given this machinery, Kitcher says that E(K) consists of the minimal set of maximally strin-

gent general argument patterns from which a maximal number of conclusions can be drawn.

More roughly, one wants to find the smallest subset of K from which the other members of

K can be derived, using the fewest possible stringent argument patterns.

As on the DN account, an explanation on the unificationist account is an argument.29 But

28For more on the history of the unificationist account, see Kitcher (1989) and Salmon (1989), as well as
references therein.

29Depending on how strictly one construes the constraints on what counts as an argument for the unifica-
tionist account, the two worries expressed with regard to the DN model may carry over. I will proceed by

16



the unificationist account offers an additional set of constraints on the kinds of arguments

that count as explanatory, above and beyond the conditions of the DN account. It is not

sufficient (or necessary) that a given conclusion be derived from a set of premises including a

law, as in the DN model; now, an argument is explanatory if it is a member of the explanatory

store of a set of statements endorsed by the scientific community.30 To understand whether

the explanation in section 2 is a unificationist explanation, then, we want to determine

whether we should expect it to be in the explanatory store over K. There are several

reasons to think that it should not be.

First, it is not clear that “gravitational mass and inertial mass are equal” is a member of

K—which would mean that it is not even a candidate explanandum. Kitcher assumes that K

is deductively closed and consistent. Thus, if the statements of GR are members of K, then

“there is no such thing as gravitational mass” is also a member of K, ruling out “gravitational

mass and inertial mass are equal”. Indeed, on Kitcher’s account of reduction, one would

reduce Newtonian gravitation to GR by showing that the general argument patterns of

Newtonian gravitation can be recovered from and extended as general argument patterns of

GR. But then one would not expect to be able to explain features of the world that can only

be expressed in the Newtonian framework, because the very goal of the reduction would be

to move one’s explanatory arguments out of the Newtonian framework and into GR.

Another worry comes from the opposite direction. Even if K were construed in such

a way that the explanandum were in K, it is still not clear that the explanation I have

given would be included in the explanatory store. This is because adding an additional

argument of the sort I have given makes the explanatory store larger without explaining any

new phenomena, since the argument patterns of GR are sufficient to explain the motion of

bodies. Thus adding such explanations threatens to violate the minimality condition

assuming that the unificationist account allows some flexibility about what counts as a permissible argument.
30On Kitcher’s account, one would add an “only if” to this last sentence. But if we adopt a pluralist view

of explanation on which some, but not necessarily all, explanations are unificationist explanations then we
want to understand the condition of membership in E(K), as defined by Kitcher, to be only a sufficient
condition.

17



A third worry is more general. It is difficult to see what kind of argument pattern is being

executed in the present example. The thing to be explained is a very general observational

feature of the world. For this reason it is essentially singular. One might be able to schematize

the argument by reconstruing it as an argument concerning the inertial and gravitational

masses for particular bodies, and then include a general argument pattern by which one

shows that for any given body, the inertial and gravitational masses are identical. But it is

not clear that such explanations answer the original question. Really we want to know why

inertial and gravitational mass are always the same, not why they happen to take the same

value in any variety of situations.

These points suggest that the explanation I have given is an awkward fit with the uni-

ficationist account of explanation, just as with the DN account. Once again, this does not

mean that the unificationist account cannot be adapted to fit the explanation I have given.

I think it probably can. Indeed, it is hard to prove that a given explanation, if successful, is

not a member of the maximally unified set of arguments over a set of beliefs. My point is

rather that to make the present explanation fit with the details of Kitcher’s account, some

adaptation of the unificationist position is likely necessary. More importantly, even if one

can find a modified unificationist account that would fit more naturally with the present

explanation, it is not clear that it would do justice to the explanation I have given. In other

words, it does not seem that the reason that the explanation I have given is explanatory

has anything to do with the fact that it is an instantiation of a general, stringent, and uni-

fied argument pattern. Indeed, I have suggested that it is not an instantiation of a general

argument pattern at all—it is a singular explanation of a general feature of the world, as

expressed in the language of Newtonian theory.

If anything, attempting to adapt the unificationist picture to include explanations of the

present sort threatens to gloss over the important features of the explanation. I would rather

say that very many explanations in science have the character that Kitcher describes: they

consist of arguments that are explanatory by virtue of how they fit some phenomenon into

18



a systematized body of knowledge. I find Kitcher’s analysis of his own examples convincing,

and I think that a broad class of explanations in physics fit well with the unificationist picture

of explanation. Modifying the unificationist account, which gets so many interesting cases

right, seems counterproductive. What we have here is simply a different kind of explanation.

4 Let a thousand flowers bloom

So far, I have argued that the explanation given in section 2 is neither a causal explanation

nor any kind of statistical explanation. It is, at least roughly, a deductive argument, though

it cannot easily be made to fit the strict logical structure demanded by the DN model.

Likewise, the explanation seems to be a poor fit with the unificationist model, at least as

described by Kitcher (1989). It should be clear from my remarks in section 1 that I do not

take these accounts’ inability to deal with the present explanation as an argument against

the accounts. I have simply presented a scientific explanation of a different sort (and there

are many possible sorts).

The remaining work is to identify some of the features of the present explanation and

to try to characterize what makes it explanatory. I have identified the most important

features along the way, but it seems worthwhile to tie them together now. As we have

seen, the present explanation is an argument (broadly construed), and the explanandum is a

feature of the world (observed in some regimes) as represented in the language of Newtonian

physics. The explanation involves a translation between GR and Newtonian gravitation

that is accomplished via the two step limiting process described above. This translation

is necessary because the question is asked in the language of one theory (indeed, it only

makes sense in that theory), but it is of a form that a fortiori cannot be answered without

appealing to physics that goes beyond the theory within which it is asked. The explanatory

demand is to show how, given some superseding theory, a general fact as expressed within

one theory is really necessary or to be expected, at least within the regime in which the old

theory is successful.

19



What is doing the explanatory work here? One way of viewing the explanation is as a

derivation of the explanandum as a general constraint on the models of Newtonian gravitation

that can be understood as good approximations of a relativistic world. In other words, the

explanation consists in showing that if the world is to be understood as a model of GR,

then any model of Newtonian gravitation that successfully approximates the world, even

in limited regimes, must satisfy the condition that mI = mG. In other words, insofar as

we accept GR, and insofar there are contexts in which one can understand gravitation as a

force, at least approximately, the coupling of a body to the gravitational field is given by the

body’s inertial mass. This work is done by presenting the details of the relationship between

the two theories.

At the very beginning of this paper, I gave two other examples of why questions that

I claimed would be well answered by explanations of the sort I have given here. Those

questions, and myriad others that one often encounters in physics, call for explanations

of general features of the world, at least within some particular regime. They, too, are

expressed within the framework and language of a particular theory. And yet, there is good

reason to believe that in answering these questions, we will exhibit important shortcomings

of the theories in which the questions are asked, much like GR reveals the ways in which

Newtonian gravitation is unsatisfactory. Since they are open questions, I cannot say with

certainty what their answers will be. But I do maintain that they are asked in the spirit

of the question that I focus on in the body of the paper, and moreover, they would be

well-answered by explanations that involve showing how the present theory arises via some

limiting procedure from a new, superceding theory. For this reason, I think that physicists

are regularly making explanatory demands on future theories that call for explanations ill-

treated by the philosophical literature on scientific explanation.

Now that I have explored the details of the kind of explanation I have in mind, let me

return to what I described in the introduction as the principal moral of this discussion. I

have argued that the essential piece of the explanation I have given, the part that does the

20



explanatory work, comes out in the details of the relationship between the theory in which the

question is asked and a new theory that supercedes it. I have also argued that explanatory

demands in the spirit of the explanation I have given, demands that would be met by

explanations of the sort described here, are an important part of research in physics. Indeed,

the question that I focused on in this paper, concerning the relationship between inertial

and gravitational mass, played such a role in several research projects at the beginning of

the 20th century: Lorand Eötvös and Einstein both took the unexplained equality of inertial

and gravitational mass to be of utmost importance in the study of gravitation. I think these

two points, taken together, reveal something important about the structure of inquiry in

physics.

First, in asking a certain kind of question, physicists implicitly acknowledge that current

theory is incomplete—indeed, the particular questions they ask can be taken as attempts

to identify where they expect current theory to break down. But this acknowledgement is

counterbalanced by an expectation that the framework and concepts of the present theory

offer sufficiently good descriptions of the world that one should expect the questions to make

sense, and moreover to have answers, even after a new theory has been discovered. This is

not to say that the new theory will look anything like the old theory. But one nonetheless

expects the conceptual framework of current theories to have sufficient stability that the

present questions will be recoverable from the new theory, just as one was able to recover

a statement about gravitational mass in an appropriate limit from a theory that does not

support a notion of gravitational mass.

It seems to me that the prevalence of the sort of explanatory demands just described poses

a significant problem for a certain class of views, often associated with Kuhn, concerning

the incommensurability of scientific theories. If theory change involves demands for and

offers of explanations of the sort I have described, then not only is translation between

distinct theories with overlapping domains possible, it is an essential part of how physicists

understand their own work. But I will not develop this idea further. Suffice it to say that

21



explanations, and particularly demands for explanation, play many roles in the creation and

evolution of physical theories. Some of these explanations will be of a kind with the one I

have treated here. And as Geroch suggests in the above quoted passage, to say any more

would be a mistake.

A Technical Details of Answer to Question 3

In the body of the paper I offer an answer to the question, “Why are inertial and gravitational

mass equal in Newtonian physics?” Here I offer the technical details of that argument, pulling

together certain strands of the literature that are not usually treated together systematically.

I will provide only a brief, formal review of geometrized and covariant standard Newtonian

theory; I will mostly take familiarity with GR for granted. For a pedagogical treatment

of these subjects, including proofs of the theorems stated here and an explanation of the

“abstract index notation” I will use throughout, see Malament (2010). For more details

on the sense in which geometrized Newtonian theory arises as a limit from GR, see Künzle

(1976), Ehlers (1981), and Malament (1986a).

A.1 Preliminary definitions

We begin by defining the geometrical structures we will work with. A model of GR can be

defined as follows.

Definition A.1 A relativistic spacetime is an ordered pair (M,gab), where M is a smooth,
31

connected, four-dimensional manifold and gab is a smooth, non-degenerate semi-Riemannian
metric on M with Lorentz signature (+,−,−,−).

In a relativistic spacetime, the metric defines a lightcone structure at every point as follows.

Given any point p and any vector ξa in the tangent space Mp, we say that ξ
a is timelike if

gabξ
aξb > 0, spacelike if gabξ

aξb < 0, and null if gabξ
aξb = 0. The length of any vector ξa at

a point is given by ||ξa|| = |gabξaξb|1/2. A (smooth) curve is timelike (resp. spacelike or null)
31Here and in what follows, it should be assumed that we are limiting attention to smooth (i.e. infinitely

differentiable) curves, fields, manifolds, etc., whether stated explicitly or not.

22



if its tangent vector is at every point of the curve. There is a unique torsion-free derivative

operator, ∇, satisfying the metric compatibility condition ∇agbc = 0; ∇, in turn, determines

the Riemann curvature tensor Rabcd.

Massive point particles are represented by their worldlines, which are smooth future-

directed timelike curves parametrized by arc-length. With every point particle, there is an

associated four-momentum, Pa, defined at every point of the particle’s worldline, whose

length is the (inertial) rest mass. For a point particle with non-zero mass mI,
32 we can

write Pa = mIξ
a, where ξa is the tangent vector field to the particle’s worldline (called the

particle’s four-velocity ). The acceleration of a particle’s wordline, ξn∇nξa, is determined

by the relation Fa = mIξ
n∇nξa, where Fa represents the external forces acting on the

particle. In the absence of forces, massive test point particles traverse timelike geodesics.

More generally, we can associate with any matter field a smooth symmetric field Tab, called

the energy-momentum tensor. Tab can be thought to encode the four-momentum density of

the matter field as determined by any future-directed timelike observer at a point: for all

points p ∈ M and all unit, future-directed timelike vectors at p, ξa, the four-momentum of

a matter field at p as determined by ξa is Pa = Tabξ
b. The curvature of spacetime is related

to the energy-momentum tensor by Einstein’s equation,

Rab = 8π(Tab −
1

2
Tgab), (A.1)

where T = Taa and Rab = R
n
abn is the Ricci tensor.

We can now proceed to define a parallel structure for classical theories.

Definition A.2 A classical spacetime is an ordered quadruple (M,tab,h
ab,∇), where M is

a smooth, connected, four-dimensional manifold; tab is a smooth symmetric field on M of
signature (1, 0, 0, 0); hab is a smooth symmetric field on M of signature (0, 1, 1, 1); and ∇

32Since keeping track of the distinction between inertial and gravitational mass is important for the ultimate
moral of the present discussion, I will label masses as inertial even in the context of GR and geometrized
Newtonian gravitation, were strictly speaking there can be no ambiguity. I will use capitalized calligraphic
symbols for subscripts indicating labels to distinguish them from subscripts indicating index (i.e. tensor)
structure.

23



is a derivative operator on M compatible with tab and h
ab, i.e. it satisfies ∇atbc = 0 and

∇ahbc = 0. We additionally require that tab and hab are orthogonal, i.e. tabhbc = 0.

Note that “signature,” here, has been extended to cover the degenerate case. We can see

immediately from the signatures of tab and h
ab that neither is invertible. Hence in general

neither tab nor h
ab can be used to raise and lower indices.

tab can be thought of as a temporal metric on M in the sense that given any vector ξ
a

in the tangent space at a point, p, ||ξa|| = (tabξaξb)1/2 is the temporal length of ξa at that

point. If the temporal length of ξa is positive, ξa is timelike; otherwise, it is spacelike. At

any point, it is possible to find a covector ta, unique up to a sign, such that tab = tatb. If

there is a continuous, globally defined vector field ta such that at every point, tab = tatb, then

the spacetime is temporally orientable (we will encode the assumption that a spacetime is

temporally oriented by replacing tab with ta in our definitions of classical spacetimes). h
ab,

meanwhile, can be thought of as a spatial metric. However, since there is no way to lower

the indices of hab, we cannot calculate the spatial length of a vector directly. Instead, we rely

on the fact that if ξa is a spacelike vector (as defined above), then there exists a covector

σa such that ξ
a = habσb. The spatial length of ξ

a can then be defined as (habσaσb)
1/2. (This

spatial length is independent of the choice of σa; if ξ
a is not a spacelike vector, then there

is no way to assign it a spatial length.) Note, too, that it is possible to define the Riemann

curvature tensor Rabcd and the Ricci tensor Rab with respect to ∇ as in GR (or rather, as

in differential geometry generally). Flatness (Rabcd = 0) carries over intact from GR; we say

a classical spacetime is spatially flat if Rabcd = Ranmqh
bnhcmhdq = 0. It turns out that this

latter condition is equivalent to Rab = hanhbmRnm = 0.
33

We describe matter in close analogy with GR. Massive point particles are again repre-

sented by their worldlines, which are smooth future-directed timelike curves parameterized

by elapsed time. For a point particle with (inertial) mass mI, we can always define a smooth

unit vector field ξa tangent to its worldline, again called the four-velocity, such that we can

33See Malament (2010, Prop. 4.15).

24



define a four-momentum field Pa = mIξ
a. The mass of the particle is now given by the

temporal length of its four-momentum. In similar analogy to the relativistic case, we can as-

sociate with any matter field a smooth symmetric field Tab, now called the mass-momentum

tensor. Tab once again encodes the four-momentum density of the matter field as determined

by a future directed timelike observer at a point, but in this case all observers agree on the

four-momentum density at p: Pa = tbT
ab. Contracting once more with tb yields the mass

density, ρ = tatbT
ab.

In the present covariant, four-dimensional language, standard Newtonian theory can

be expressed as follows. Let (M,ta,h
ab,∇) be a classical spacetime. We require that ∇

is flat (i.e. Rabcd = 0). We begin by considering the dynamics of a test point particle

with inertial mass mI and four-velocity ξ
a. As in GR, the force on a particle is related to

the acceleration of its worldine by Fa = mIξ
a∇aξb. In the absence of external forces, a

massive test point particle undergoes geodesic motion. If the total mass-momentum content

of spacetime is described by Tab, we require that the conservation condition holds, i.e. at

every point ∇aTab = 0. To add gravitation to the theory, we can represent the gravitational

potential as a smooth scalar field ϕ on M. ϕ is required to satisfy Poisson’s equation,

∇a∇aϕ = 4πρ (where ∇a is shorthand for hab∇b). Gravitation is considered a force; in

general, the gravitational force on a point particle is moderated by its gravitational mass,

according to FaG = −mG∇
aϕ.

In geometrized Newtonian theory we again begin with a classical spacetime (M,ta,h
ab,∇),

but now we allow ∇ to be curved. The dynamics of a point particle with inertial mass mI

and four-velocity ξa are again given by Fa = mIξ
n∇nξa; likewise, free massive test point

particles undergo geodesic motion. However, the geodesics are now determined relative to

∇, which is not necessarily flat. The conservation condition is again expected to hold. Grav-

itational interactions are now seen to be the result of the curvature of spacetime, which in

25



turn is determined by a geometrized form of Poisson’s equation,

Rab = 4πρtatb. (A.2)

The geometrized Poisson’s equation forces spacetime to be spatially flat, because if Poisson’s

equation holds, then Rab = 4πρhanhbmtntm = 0 by the orthogonality condition on the

metrics.

A.2 Relations between the theories

We are particularly interested in the relationship between these three theories. Several

results are available. First, it is always possible to “geometrize” a gravitational field on

a flat classical spacetime—that is, we can always move from the covariant formulation of

standard Newtonian gravitation to geometrized Newtonian gravitation, via a result due to

Andrzej Trautman (1965).

Proposition A.3 (Trautman Geometrization Lemma.) (Slightly modified from Mala-

ment, 2010, Prop. 4.2.1.) Let (M,ta,h
ab,

f

∇) be a flat classical spacetime. Let ϕ and ρ

be smooth scalar fields on M satisfying Poisson’s equation,
f

∇a
f

∇aϕ = 4πρ. Finally, let
g

∇ = (
f

∇,Cabc),
34 with Cabc = −tbtc

f

∇aϕ. Then (M,ta,hab,
g

∇) is a classical spacetime;
g

∇ is
the unique derivative operator on M such that given any timelike curve with tangent vector
field ξa,

ξn
g

∇nξa = 0 ⇔ ξn
f

∇nξa = −
f

∇aϕ; (G)

and the Riemann curvature tensor relative to
g

∇,
g

R abcd, satisfies

g

Rab = 4πρtatb (CC1)
g

Rab
c
d =

g

Rcd
a
b (CC2)

g

Rabcd = 0. (CC3)

34This notation is explained in Malament (2010, Prop. 1.7.3). Briefly, if ∇ is a derivative operator on M,
then any other derivative operator on M is determined relative to ∇ by a smooth symmetric (in the lower
indices) tensor field, Cabc, and so specifying the C

a
bc field and ∇ is sufficient to uniquely determine a new

derivative operator.

26



Trautmann showed that it is also possible to go in the other direction. That is, given a

curved classical spacetime, it is possible to recover a flat classical spacetime and a gravita-

tional field, ϕ—so long as the curvature conditions (CC1)-(CC3) are met.

Proposition A.4 (Trautman Recovery Theorem.) (Slightly modified from Malament,

2010, Prop. 4.2.5.) Let (M,ta,h
ab,

g

∇) be a classical spacetime that satisfies eqs. (CC1)-
(CC3) for some smooth scalar field ρ. Then, at least locally on M, there exists a smooth

scalar field ϕ and a flat derivative operator on M,
f

∇, such that (M,ta,hab,
f

∇) is a classical

spacetime; (G) holds of any timelike curve; and ϕ and
f

∇ together satisfy Poisson’s equation,
f

∇a
f

∇aϕ = 4πρ.

It is worth pointing out that the pair (
f

∇,ϕ) is not unique. It is also worth pointing out

that whenever we begin with standard Newtonian theory and move to geometrized Newto-

nian theory, it is always possible to move back to the standard theory, because Prop. A.3

guarantees that the curvature conditions (CC1)-(CC3) are satisfied.

We can now ask how either of these classical theories relate to GR. The answer is that

geometrized Newtonian theory arises as a limiting case of GR, for a properly constructed

limit. (For full details of this limiting procedure, see Malament (1986a, Sec. 5).) Intuitively,

we will begin with a relativistic spacetime, and then allow the lightcone structure at every

point to open in such a way that, in the limit, the lightcones at every point become degen-

erate. Since the lightcone structure in a sense determines the speed of light, allowing the

lightcones to widen in this fashion captures a sense in which one might allow the speed of

light to go to infinity.

The limit can be constructed as follows. Consider a manifold M admitting a one-

parameter family of nondegenerate Lorentz metrics gab(λ) (where λ ranges over some interval

(0,k) ⊆ R) that satisfy two conditions:

(Lim1) limλ→0 gab(λ) = tatb for some non-vanishing closed field ta;
35

(Lim2) limλ→0 λg
ab(λ) = −hab for some field hab of signature (0, 1, 1, 1).

35If ta is a non-vanishing closed field, the product tab = tatb automatically has signature (1, 0, 0, 0).

27



For any λ ∈ (0,k), we can associate with gab(λ) the unique covariant derivative operator

compatible with gab(λ),
λ

∇, as well as the Ricci curvature tensor associated with
λ

∇,
λ

Rab.

Thus the one-parameter family of metrics generates a one-parameter family of compatible

derivative operators and curvature tensors. Suppose further that, for any λ ∈ (0,k), we can

define a smooth symmetric field Tab(λ) that together with gab(λ) and its associated Ricci

tensor satisfy:

(Lim3) For all λ ∈ (0,k),
λ

Rab = 8π
(
Tab(λ) − 12gab(λ)T(λ)

)
, where T(λ) = Tab(λ)g

ab(λ); and

(Lim4) limλ→0 T
ab(λ) = Tab for some smooth symmetric field Tab on M.

When these conditions hold, it is possible to prove the following result.

Proposition A.5 (Classical Limit of GR.) (Adapted from Malament, 1986a, Props. on
Limits 1 & 2) Fix a smooth, connected, four-dimensional manifold M and assume λ is a real-
valued variable taking all values on an interval (0,k). Suppose that for each λ on an interval
(0,k), there exist smooth symmetric fields gab(λ) and Tab(λ) on M such that (M,gab(λ)) is a
relativistic spacetime and for each λ, gab(λ) and Tab(λ) collectively satisfy conditions (Lim1)-

(Lim4). Then there exists a derivative operator ∇a on M such that limλ→0
λ

∇a = ∇a, 36 and
for which (M,ta,h

ab,∇a) is a classical spacetime satisfying Rab
c
d = R

c
d
a
b. Moreover, there

exists a smooth field ρ on M such that limλ→0 Tab(λ) = ρtatb, which satisfies Rab = 4πρtatb.

Prop. A.5 gives the precise sense in which geometrized Newtonian gravitation is a limiting

case of GR.

A.3 Gravitational Mass in Newtonian Theory

We have now done sufficient groundwork to offer a technically precise formulation of the

explanation given in the body of the paper. The argument was that by beginning with GR

and then moving in the limit to standard Newtonian theory, one finds that a massive point

36What does it mean for a sequence of derivative operators to converge? Suppose that ∇̃a is a fixed

auxiliary derivative operator on M. Then for each
λ

∇a, there is a smooth symmetric field Cabc(λ) such that
λ

∇a = (∇̃a,Cabc(λ)). Now suppose that there is another derivative operator on M, ∇a = (∇̃a,C
a
bc). We can

say that limλ→0
λ

∇a = ∇a if limλ→0 Cabc(λ) = C
a
bc.

28



particle’s coupling to the gravitational field is given by its inertial mass. This limit proceeds

in two steps. First, using Prop. A.5, one shows that geometrized Newtonian gravitation is

a limiting case of GR. Prop. A.4, meanwhile, shows that when three curvature conditions,

(CC1)-(CC3), are satisfied, we can recover (covariant) standard Newtonian theory from

geometrized Newtonian theory. It is in the course of executing this two step process that

one is forced to associate gravitational and inertial mass.

There is an important subtlety here. To connect Props. A.4 and A.5 and show that we

can recover standard Newtonian theory as a limit from GR, we need to show that the clas-

sical spacetime we reach in the limit from GR in fact meets the three curvature conditions

necessary to recover the standard theory. Prop. A.5 gives that two of the curvature con-

ditions, (CC1) and (CC2), are satisfied automatically. But what about (CC3), Rabcd = 0?

In general, Rabcd need not vanish in a classical spacetime reached in the limit from GR. It

turns out that there is a more general recovery theorem, due to Hans-Peter Künzle (1976)

and Jürgen Ehlers (1981), that holds when Rabcd 6= 0. But the theory that you recover

in this case is not standard Newtonian gravitation—it is a non-geometrized generalization

of standard Newtonian gravitational theory in which the gravitational potential field is re-

placed by a vector field and there is an additional contribution to the force law for a particle

arising from a kind of universal rotation. The third curvature condition is sufficient to guar-

antee that this rotational contribution vanishes and that the gravitational vector field can

be written as the covariant derivative of a scalar potential.

For present purposes, establishing when Rabcd = 0 holds is unnecessary. The important

point is that condition (CC3) is both a necessary and sufficient condition for recovery of

standard Newtonian physics via Prop. A.4. This is not a problem for the explanation, per

se, since insofar as we can recover standard Newtonian theory at all, we can do so only from

families of relativistic spacetimes that converge to classical spacetimes satisfying (CC3).

I can now state the precise claim, the proof of which amount to the formal explanation.

29



Proposition A.6 Let (M,ta,h
ab,

f

∇) be a flat classical spacetime and let ϕ be a gravitational
field defined on that spacetime as in standard Newtonian gravitation. Suppose further that

(M,ta,h
ab,

f

∇) and ϕ arise via the two-step limiting process just described (which is only
possible if the intermediate curved classical spacetime satisfies (CC3)). Consider a mas-
sive point particle with inertial mass mI traversing a timelike curve in M, γ, with tangent
vector field ξa, under the influence of only gravitational force. Then the gravitational force
experienced by the massive point particle is

FaG = mIξ
n
f

∇nξa = −mI
f

∇aϕ. (A.3)

In other words, the particle’s gravitational mass is equal to its inertial mass.

Proof. By assumption, (M,ta,h
ab,

f

∇) and ϕ arise via the two-step limiting process described

above. Thus there exists a (curved) classical spacetime (M,ta,h
ab,

g

∇) satisfying (CC3) from

which (M,ta,h
ab,

f

∇) can be recovered. Since the particle experiences no non-gravitational

force, we know from the geodesic principle of geometrized Newtonian gravitation that γ must

be a geodesic relative to
g

∇. Meanwhile, by Prop. A.4, we know that if γ is a geodesic relative

to
g

∇, then ξn
f

∇nξa = −
f

∇aϕ. Thus we have the acceleration of the particle’s worldline, which

we can plug into Fa = mIξ
n
f

∇nξa to find the gravitational force on the particle. We see

that FaG = mIξ
n
f

∇nξa = −mI
f

∇aϕ, as required. It follows that the particle’s coupling to the

gravitational field is given by its inertial mass. �

References

Batterman, Robert (2002). The Devil in the Details. New York: Oxford University Press.

Cartan, Elie (1923). “Sur les variétés à connexion affine, et la théorie de la relativité
généralisée (première partie).” Annales scientifiques de l’École Normale Supérieure, 40,
325–412.

Cartan, Elie (1924). “Sur les variétés à connexion affine, et la théorie de la relativité
généralisée (première partie) (suite).” Annales scientifiques de l’École Normale Supérieure,
41, 1–25.

Cartwright, Nancy (1983). How the Laws of Physics Lie. New York: Oxford University
Press.

Ehlers, Jürgen (1981). “Über den Newtonschen Grenzwert der Einsteinschen Gravitations-
theorie.” Grundlagen Probleme der Modernen Physik, Ed. Jürgen Nitsch, Joachim Pfarr,
and Ernst-Walther Stachow. Zurich: Bibliographisches Institut.

30



Einstein, Albert (1920). Relativity: The Special and General Theory. Ed. Robert W. Lawson.
New York: Henry Holt and Co.

Fisher, Ronald Aylmer (1930). The genetical theory of natural selection. Oxford: Clarendon
Press.

Friedrichs, Kurt O. (1927). “Eine Invariante Formulierun des Newtonschen Gravitationsge-
setzes und der Grenzüberganges vom Einsteinschen zum Newtonschen Gesetz.” Mathe-
matische Annalen, 98, 566–575.

Geroch, Robert (1978). General Relativity from A to B. Chicago: University of Chicago
Press.

Godfrey-Smith, Peter (2003). Theory and Reality. Chicago: University of Chicago Press.

Hempel, Carl and Paul Oppenheim (1948). “Studies in the Logic of Explanation.” Philosophy
of Science, 15, 135–175.

Kitcher, Philip (1989). “Explanatory Unification and the Causal Structure of the World.”
Scientific Explanation, Ed. Philip Kitcher and Wesley Salmon. Minneapolis: University
of Minnesota Press.

Kuhn, Thomas S. (1977). “Concepts of Cause in the Development of Physics.” The Essential
Tension. Chicago: The University of Chicago Press. 21–30.

Künzle, H. P. (1976). “Covariant Newtonian Limit of Lorentz Space-Times.” General
Relativity and Gravitation, 7 (5), 445–457.

Malament, David (1986a). “Newtonian Gravity, Limits, and the Geometry of Space.” From
Quarks to Quasars, Ed. Robert Colodny. Pittsburgh: University of Pittsburgh Press.

Malament, David (1986b). “Gravity and Spatial Geometry.” Logic, Methodology and Philos-
ophy of Science, Ed. Ruth Barcan Marcus, George Dorn, and Paul Weingartner. Volume
VII. New York: Elsevier Science Publishers. 405–411.

Malament, David B. (2010). “Relativistic and Newtonian Spacetime Struc-
ture.” Most recent version available at http://www.lps.uci.edu/home/fac-
staff/faculty/malament/FndsofGR/.

Nagel, Ernest (1961). The Structure of Science. New York: Harcourt, Brace & World.

Nagel, Ernest (1970). “Issues in the Logic of Reductive Explanations.” Mind, Science, and
History, Ed. Howard E. Kiefer and Milton K. Munitz. Albany, NY: State University of
New York Press.

Nickles, Thomas (1973). “Two Concepts of Intertheoretic Reduction.” Journal of Philosophy,
70 (7), 181–201.

31



Odom, B., D. Hanneke, B. D’Urso, and G. Gabrielse (2006). “New Measurement of the
Electron Magnetic Moment Using a One-Electron Quantum Cyclotron.” Physical Review
Letters, 97 (3), 030801.

Sachs, Mendel (1976). “On the Logical Status of Equivalence Principles in General Relativity
Theory.” British Journal for the Philosophy of Science, 27 (3), 225–229.

Salmon, Wesley (1984). Scientific Explanation and the Causal Structure of the World.
Princeton: Princeton University Press.

Salmon, Wesley (1989). “Four Decades of Scientific Explanation.” Scientific Explanation,
Ed. Philip Kitcher and Wesley Salmon. Minneapolis: University of Minnesota Press.

Skyrms, Brian (1996). Evolution of the Social Contract. New York: Cambridge University
Press.

Strevens, Michael (2004). “The Causal and Unification Approaches to Explanation Unified–
Causally.” Nous, 38 (1), 154–176.

Strevens, Michael (2008). Depth: An Account of Scientific Explanation. Cambridge, MA:
Harvard University Press.

Trautman, Andrzej (1965). “Foundations and Current Problem of General Relativity.”
Lectures on General Relativity, Ed. Stanley Deser and Kenneth W. Ford. Englewood
Cliffs, NJ: Prentice-Hall.

Weinberg, Steven (1972). Gravitation and Cosmology. New York: Wiley.

Woodward, James (2003). Making Things Happen: A Theory of Causal Explanation. Oxford:
Oxford University Press.

32