Cartwright, N., How The Laws of Physics Lie 1983 PDF

HOW THE LAWS
OF PHYSICS LIE
NANCY CARTWRIGHT
CLARENDON PRESS· OXFORD

OXFORD UNIVERSITY PRESS . NEW YORK
Oxford University Press, Walton Street, Oxford OX2 60p
Oxford New York

Athens Auckland Bangkok Bombay
Calcutta Cape Town Dar es Salaam Delhi
Florence Hong Kong Istanbul Karachi
Kuala Lumpur Madras Madrid Melbourne
!~
I
Mexico City Nairobi Paris Singapore
Taipei Tokyo Toronto
:rr and associated companies

Berlin Ibadan
in
1r4-
!
o
N) Oxford is a trade mark of Oxford University Press
·21 Published in the United States by

Oxford University Press lnc., New York
c Nancy Cartwright 1983
First published 1983
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.
Within the UK, exceptions are allowed in respect 'of any fair dealing for the
purpose of research or priuate study, or criticism or reuieui, as permitted
under the Copyright, Designs and Patents Act, 1988, or in the case of
reprographic reproduction in accordance with the terms of the licences
issued by the Copyright Licensing Agency. Enquiries concerning
reproduction outside these terms and in other countries should be
sent to the Rights Department, Oxford University Press,
at the address above
This book is sold subject to the condition that it shall not, by way
of trade or otherwise, be lent, re-sold, hired out or otherwise circulated
without the publisher's prior consent in any form of binding or cover
other than that in which it is published and without a similar condition
including this condition being imposed on the subsequent purchaser
"
British Library Cataloguing in Publication Data
Cartwright, Nancy
. i How the laws of physics lie.
1. Physics-Philosophy
I. Title
530'.01 QC6
ISBN 0-19-824704-4
j Library of Congress Cataloging in Publication Data

; I r-=t;.i,~t;; Data available
9 10 8
Printed in Great Britain

on acid-free paper by
Biddies Ltd
Guildford and King's Lynn
To Marie and Claude
Acknowledgements
The author acknowledges permission to use previously pub-

lished material in this volume. The provenance of the essays
is as follows:
Essay 1. In part, from 'Causal Laws and Effective Strategies',
Nous, vol. 13 (1979). ©Noils 1979; reproduced by permis-
sion of Nous.
In part, new.
Essay 2. From 'Truth Doesn't Explain Much', American
Philosophical Quarterly, vol. 17 (1980).
Essay 3. In part, from 'Do the Laws of Physics State the
Facts', Pacific Philosophical Quarterly, vol. I (1980).
In part, from 'How do We Apply Science', PSA 1974
Proceedings, ed. Robert Cohen etal. (Reidel, 1974).
In part, new.
Essay 4. From 'The Reality of Causes in a World of Instru-
mental Laws', PSA 1980 Proceedings, ed. P. Asquith and
R. Giere (Philosophy of Science Association, 1980).
Essay 5. From 'When Explanation Leads to Inference', Philo-
sophical Topics, special issue on Realism (forthcoming).
Essay 6. In part, from 'How Approximation Takes Us Away
from Theory and Towards the Truth' by Nancy Cartwright
and Jon J. Nordby, Pacific Philosophical Quarterly (forth-
coming).
Essay 7. From 'Fitting Facts to Equations', Philosophical
Grounds of Rationality: Intentions, Categories, and Ends
(essays dedicated to Paul Grice), ed. Richard Grandy and
Richard Warner (Oxford University Press, forthcoming).
Essay 8. New.
Essay 9. In part, from 'How the Measurement Problem is an
Artefact of Mathematics', Space, Time, and Causality, ed.
Richard Swinburne (Reidel, forthcoming).
In part, from Studies in the Foundations of Quantum
Mechanics, ed. P. Suppes (Philosophy of Science Associa-
tion, 1980).
In part, new.
Contents
Introduction
Essay 1 Causal Laws and Effective Strategies 21
Essay 2 The Truth Doesn't Explain Much 44
Essay 3 Do the Laws of Physics State the Facts? 54
Essay 4 The Reality of Causes in a World of

Instrumental Laws 74
Essay 5 When Explanation Leads to Inference 87
Essay 6 For Phenomenological Laws 100
Essay 7 Fitting Facts to Equations 128
Essay 8 The Simulacrum Account of Explanation 143
Essay 9 How the Measurement Problem is an

Artefact of the Mathematics 163
Author Index 217
Subject Index 219

Introduction
Philosophers distinguish phenomenological from theoretical

laws. Phenomenological laws are about appearances; theoreti-
cal ones are about the reality behind the appearances. The
distinction is rooted in epistemology. Phenomenological
laws are about things which we can at least in principle
observe directly, whereas theoretical laws can be known
only by indirect inference. Normally for philosophers 'phe-
nomenological' and 'theoretical' mark the distinction
between the observable and the unobservable.
Physicists also use the terms 'theoretical' and 'phenomeno-
logical'. But their usage makes a different distinction.
Physicists contrast 'phenomenological' with 'fundamental'.
For example, Pergamon Press's Encyclopaedic Dictionary
of Physics says, 'A phenomenological theory relates ob-
served phenomena by postulating certain equations but
does not enquire too deeply into their fundamental signifi-
cance.'l
The dictionary mentions observed phenomena. But do
not be misled. These phenomenological equations are not
about direct observables that contrast with the theoretical
entities of the philosopher. For look where this definition
occurs-under the heading 'Superconductivity and super-
fluidity, phenomenological theories of. Or notice the
theoretical entities and processes mentioned in the contents
of a book like Phenomenology of Particles at High Energies
(proceedings of the 14th Scottish Universities Summer
School in Physics): (l) Introduction to Hadronic Inter-
actions at High Energies. (2) Topics in Particle Physics with
Colliding Proton Beams. (3) Phenomenology of Inclusive
Reactions. (4) Multihadron Production at High Energies:
Phenomenology and Theory."
I The Encyclopaedic Dictionary of Physics (Oxford: Pergamon Press, 1964),

p.108.
2 P. 1. Crawford and R. Jennings, Phenomenology of Particles at High Ener-
gies (London: Academic Press, 1974).
2 INTRODUCTION
Francis Everitt, a distinguished experimental physicist and

biographer of James Clerk Maxwell, picks Airy's law of
Faraday's magneto-optical effect as a characteristic phe-
nomenological law.! In a paper with Ian Hacking, he reports,
'Faraday had no mathematical theory of the effect, but in
1846 George Biddell Airy (1801-92), the English Astronomer
Royal, pointed out that it could be represented analytically
in the wave theory of light by adding to the wave equations,
which contain second derivatives of the displacement with
respect to time, other ad hoc terms, either first or third
derivatives of the displacement." Everitt and Hacking con-
trast Airy's law with other levels of theoretical statement
-'physical models based on mechanical hypotheses, ...
formal analysis within electromagnetic theory based on
symmetry arguments', and finally, 'a physical explanation
in terms of electron theory' given by Lorentz, which is
'essentially the theory we accept today'.
Everitt distinguishes Airy's phenomenological law from
the later theoretical treatment of Lorentz, not because
Lorentz employs the unobservable electron, but rather
because the electron theory explains the magneto-optical
effect and Airy's does not. Phenomenological laws describe
what happens. They describe what happens in superfluids
or meson-nucleon scattering as well as the more readily
observed changes in Faraday's dense borosilicate glass, where
magnetic fields rotate the plane of polarization of light. For
the physicist, unlike the philosopher, the distinction between
theoretical and phenomenological has nothing to do with
what is observable and what is unobservable. Instead the
terms separate laws which are fundamental and explanatory
from those that merely describe.
The divide between theoretical and phenomenological
commonly separates realists from anti-realists. I argue in these
essays for a kind of anti-realism, and typically it is an anti-
realism that accepts the phenomenological and rejects the theo-
retical. But it is not theory versus observation that I reject.
Rather it is the theoretical as opposed to the phenomenological.
3 In conversation, July 1981.
4 C. W. F. Everitt and Ian Hacking, 'Theory or Experiment: Which Comes
First?' American Scientist, forthcoming.
INTRODUCTION 3
In modern physics, and I think in other exact sciences as

well, phenomenological laws are meant to describe, and they
often succeed reasonably well. But fundamental equations
are meant to explain, and paradoxically enough the cost of
explanatory power is descriptive adequacy. Really powerful
explanatory laws of the sort found in theoretical physics
do not state the truth.
I begin from the assumption that we have an immense
number of very highly confirmed phenomenological laws.
Spectra-physics Incorporated continuously runs a quarter
of a million dollars' worth of lasers to death to test their
performance characteristics. Nothing could be better con-
firmation than that. But how do the fundamental laws of
quantum mechanics, which are supposed to explain the
detailed behaviour of lasers, get their confirmation? Only
indirectly, by their ability to give true accounts of lasers, or
of benzene rings, or of electron diffraction patterns. I will
argue that the accounts they give are generally not true,
patently not true by the same practical standards that admit
an indefinite number of commonplace phenomenological
laws. We have detailed expertise for testing the claim of
physics about what happens in concrete situations. When
we look to the real implications of our fundamental laws,
they do not meet these ordinary standards. Realists are
inclined to believe that if theoretical laws are false and
inaccurate, then phenomenological laws are more so. I urge
just the reverse. When it comes to the test, fundamental
laws are far worse off than the phenomenological laws
they are supposed to explain.
The essays collected in this volume may be grouped around
three different but interrelated arguments for this paradoxical
conclusion.
(I) The manifest explanatory power of fundamental laws
does not argue for their truth.
(2) In fact the way they are used in explanation argues for
their falsehood. We explain by ceteris paribus laws, by
composition of causes, and by approximations that improve
on what the fundamental laws dictate. In all of these cases
the fundamental laws patently do not get the facts right.
(3) The appearance of truth comes from a bad model of
4 INTRODUCTION
explanation, a model that ties laws directly to reality. As an

alternative to the conventional picture I propose a simula-
crum account of explanation. The route from theory to
reality is from theory to model, and then from model to
phenomenological law. The phenomenological laws are
indeed true of the objects in reality-or might be; but the
fundamental laws are true only of objects in the model.
1. AGAINST INFERENCE TO BEST EXPLANATION
I will argue that the falsehood of fundamental laws is a

consequence of their great explanatory power. This is the
exact opposite of what is assumed by a well-known and
widely discussed argument form-inference to the best
explanation. The basic idea of this argument is: if a hypo-
thesis explains a sufficiently wide variety of phenomena
well enough, we can infer that the hypothesis is true. Ad-
vocates of this argument form may disagree about what
counts as well enough, or how much variety is necessary.
But they all think that explanatory power, far from being at
odds with truth, leads us to it. My first line of argument in
these essays denies that explanation is a guide to truth.
Numerous traditional philosophical positions bar infer-
ences to best explanations. Scepticism, idealism, and posi-
tivism are examples. But the most powerful argument I know
is found in Pierre Duhem's Aim and Structure of Physical
Theory ,5 reformulated in a particularly pointed way by
Bas van Fraassen in his recent book The Scientific Image. 6
Van Fraassen asks, what has explanatory power to do with
truth? He offers more a challenge than an argument: show
exactly what about the explanatory relationship tends to
guarantee that if x explains y and y is true, then x should be
true as well. This challenge has an answer in the case of
causal explanation, but only in the case of causal explanation.
That is my thesis in 'When Explanation Leads to Inference'.
Suppose we describe the concrete causal process by which
a phenomenon is brought about. That kind of explanation
5 Pierre Duhem, The Aim and Structure of Physical Theory, trans. Philip P.
Wiener (New York: Atheneum, 1962).
• Bas van Fraassen, The Scientific Image (Oxford: Clarendon Press, 1980).
INTRODUCTION 5
succeeds only if the process described actually occurs. To

the extent that we find the causal explanation acceptable,
we must believe in the causes described.
For example, consider the radiometer, invented by William
Crookes in 1853. It is a little windmill whose vanes, black
on one side, white on the other, are enclosed in an evacuated
glass bowl. When light falls on the radiometer, the vanes
rotate. At first it was assumed that light pressure causes the
vanes to go round. Soon it was realized that the pressure of
light would not be nearly great enough. It was then agreed
that the rotation is due to the action of the gas molecules
left inside the evacuated bowl. Crookes had tried to produce
a vacuum in his radiometer. Obviously if we accept the
agreed explanation, we infer that Crookes's vacuum was
imperfect; the explanation demands the presence of mole-
cules in the jar.
There were two rival hypotheses about what the molecules
did. Both ideas are still defended by different camps today.
A first proposal was that the vanes are pushed around by
pressure of the molecules bouncing more energetically from
the black side than the white. But in 1879 James Clerk
Maxwell, using the kinetic theory of gases, argued that the
forces in the gas would be the same in all directions, and
so could not push the vanes. Instead differential heating in
the gas produces tangential stresses, which cause slippage
of the gas over the surface. As the gas flows around the edge,
it pulls the vanes with it. In his biography of Maxwell,
Francis Everitt urges the superiority of Maxwell's account
over the more widely accepted alternative." His confidence
in Maxwell's causal story is reflected in his ontological
views.· His opponents think the tangential stresses are neglig-
ible. But unlike them, Everitt believes that if he builds a
raidometer big enough he will be able to measure the flow
of gas around the edge of the vanes.
The molecules in Crookes's radiometer are invisible, and
the tangential stresses are not the kinds of things one would
have expected to see in the first place. Yet, like Everitt, I
believe in both. I believe in them because I accept Maxwell's
7 C. W. F. Everitt, James Clerk Maxwell-Physicist and Natural Philosopher
(New York: Charles Scribner's Sons, 1975). See Chapter 9.
6 INTRODUCTION
causal account of why the vanes move around. In producing

this account, Maxwell deploys certain fundamental laws,
such as Boltzmann's equation and the equation of con-
tinuity, which I do not believe in. But one can reject theo-
retical laws without rejecting theoretical entities. In the
case of Maxwell's molecules and the tangential stresses in
the radiometer, there is an answer to van Fraassen's question:
we have a satisfactory causal account, and so we have good
reason to believe in the entities, processes, and properties in
question.
Causal reasoning provides good grounds for our beliefs in
theoretical entities. Given our general knowledge about what
kinds of conditions and happenings are possible in the cir-
cumstances, we reason backwards from the detailed structure
of the effects to exactly what characteristics the causes must
have in order to bring them about. I have sometimes sum-
marized my view about explanation this way: no inference
to best explanation; only inference to most likely cause. But
that is right only if we are very careful about what makes a
cause 'likely'. We must have reason to think that this cause,
and no other, is the only practical possibility, and it should
take a good deal of critical experience to convince us of this.
We make our best causal inferences in very special situa-
tions-situations where our general view of the world makes
us insist that a known phenomenon has a cause; where the
cause we cite is the kind of thing that could bring about the
effect and there is an appropriate process connecting the
cause and the effect; and where the likelihood of other
causes is ruled out. This is why controlled experiments are
so important in finding out about entities and processes
which we cannot observe. Seldom outside of the controlled
conditions of an experiment are we in a situation where a
cause can legitimately be inferred.
Again the radiometer illustrates. Maxwell is at odds with
the standard account. To resolve the debate, Maxwell's
defender, Everitt, proposes not further theoretical analysis,
but rather an experiment. He wants to build an enormous
radiometer, where he can control the partial vacuum and its
viscosity, vary the coefficient of friction on the vanes,
vary their widths, take into account winds in the jar, and
INTRODUCTION 7
finally determine whether the tangential stresses are really
the major cause of rotation.
The dispute about normal and tangential stresses highlights
a nice point about observation. Philosophical debate has
focused on entities. Instrumentalists, who want to believe
only in what they can see, get trapped in footling debates:
do we really 'see' through a microscope? An electron micro-
scope? A phase-interference light microscope? Even with the
naked eye, do we not in any case see only effects? But
many of the things that are realities for physics are not
things to be seen. They are non-visual features-the spin
of the electron, the stress between the gas surface, the
rigidity of the rod. Observation-seeing with the naked
eye-is not the test of existence here. Experiment is. Experi-
ments are made to isolate true causes and to eliminate false
starts. That is what is right about Mill's 'methods'.
Where can such an idea make a difference? I think these
are just the kinds of considerations that need to be brought
to current philosophical debates about quantum electro-
dynamics. Nobody denies the enormous organizing and
predictive power of this theory, especially since the develop-
ment in the last few years of gauge field theories, which
unify weak and electromagnetic phenomena. Many think
that quantum electrodynamics is the most impressive theory
we have ever had in these respects. But as elementary particle
physicist J ames Cushing remarks,
When one looks at the succession of blatantly ad hoc moves made in
QFT [quantum field theory] (negative-energy sea of electrons, dis-
carding of infinite self energies and vacuum polarizations, local gauge
invariance, forcing renormalization in gauge theories, spontaneous
symmetry breaking, permanently confined quarks, color, just as ex-
amples) and of the picture which emerges of the 'vacuum' (aether?),
as seething with particle-antiparticle pairs of every description and
as responsible for breaking symmetries initially present, one can ask
whether or not nature is seriously supposed to be like that. 8
Does the success of quantum field theory argue for the
existence of negative energy electrons, permanently confined
quarks, and a vacuum 'seething with particle-antiparticle
8 James Cushing, 'Models and Methodologies in Current Theoretical High-
Energy Physics', Synthese 50 (1982), p. 78.
8 INTRODUCTION
pairs of every description'? Debate among philosophers has

tended to focus on the coherence of the theory, or on the
true extent of is successes." I think we should instead focus
on the causal roles which the theory gives to these strange
objects: exactly how are they supposed to bring about the
effects which are attributed to them, and exactly how good
is our evidence that they do so? The general success of the
theory at producing accurate predictions, or at unifying what
before had been disparate, is no help here. We can believe
in the unexpected entities of quantum electrodynamics if
we can give them concrete causal roles; and the rationality
of that belief will depend on what experimental evidence
supports the exact details of those causal claims.
Although I claim that a successful causal explanation
gives good reason to believe in the theoretical entities and
theoretical properties it postulates, I have repeatedly said
that I do not believe in theoretical laws. But do not properties
and laws go hand-in-hand? Bas van Fraassen asks: are not
'inferences to causes after all merely inferences to the truth
of propositions describing general characteristics of ... the
things the propositions are about'j"? The answer to van
Fraassen's question is, undoubtedly, yes. But the proposi-
tions to which we commit ourselves when we accept a causal
explanation are highly detailed causal principles and concrete
phenomenological laws, specific to the situation at hand, not
the abstract equations of a fundamental theory. Maxwell
says that the vanes are dragged around by gas sliding over the
edge. They are not pushed by light pressure or the normal
force of the gas on the surface. The acceptability of his
account depends on a host of general claims about what
happens in radiometers.
Here is one phenomenological law-in this case a causal
principle-which Maxwell uses:
[The] velocity (with which the gas slides over the surface) and the cor-
responding tangential stress are affected by inequalities of temperature
9 See the January 1982 issue of Synthese, which is devoted to this topic,
and further references therein.
10 In correspondence in June 1981.
INTRODUCTION 9
at the surface of the solid, which give rise to a force tending to make
the gas slide along the surface from colder to hotter places. 11
Here is another, this one critical to his argument that the
vanes are not pushed by pressure normal to the surface:
When the flow of heat is steady, these forces (the total forces acting
in all directions) are in equilibrium.P
Maxwell's explanation of exactly how the motion in radio-
meters takes place will not be right unless these principles
are true. But these are not fundamental laws. Maxwell
sets his particular causal story into the framework of the
developing kinetic theory of gases. It is useful to contrast
the two specific laws quoted, about what happens in radio-
meters, with two fundamental equations from this basic
theory which Maxwell uses. In his derivation, he employs
both Boltzmann's equation
dfl + ~1 dfl + 171dfl + ~1 dfl + X dfl + Y dfl + Z dfl +

dt dx dy dz d~1 d171 d~1
+ fffd~2d172d~2 fbdb fdcf>VCfI!2 - f~f~) = 0,

(1)
and the general equation of continuity
add
-[Qp] + -[Q(u + ~- U)] + -[Q(v + 17- V] +
at dx dy
d 0
+ dz[Q(w+~-W)]=POtQ· (2)
These are general, abstract equations; they are not about

any particular happenings in any particular circumstances.
The contrast is like that between moral principles, as Aris-
totle sees them in the Nicomachean Ethics, Book II,
Chapter 7, 'Among statements about conduct those which
are general apply more widely, but those which are particular
are more genuine.'
1l James Clerk Maxwell, 'On Stresses in Rarified Gases Arising from Inequali-
ties of Temperature', The Scientific Papers of James Clerk Maxwell ed. W. D.

Niven (New York: Dover Publishers, 1965), p. 703.
12 Ibid., p. 684.
10 INTRODUCTION
Explanatory power is no guarantee of truth, unless van

Fraassen's challenge can be met. I argue that, in the very
special case of causal explanation, the challenge is met. In
causal explanations truth is essential to explanatory success.
But it is only the truth of low-level causal principles and
concrete phenomenological laws. Is there no further account
that secures the truth of abstract laws as well; no story of
explanation that shows that abstract laws must be true if
they are to explain? There are two models of theoretical
explanation that could do so. I discuss them in Essay 6.
Both have serious flaws. The other essays in this volume
that argue against inference to the best explanation are
'When Explanation Leads to Inference', and 'The Reality
of Causes in a World of Instrumental Laws'.
Since I make such heavy use of the notion of causal prin-
ciples-a notion which empiricists will be wary of-one
earlier paper is included. 'Causal Laws and Effective Strategies'
argues that causal laws are quite as objective as the more
Humean laws of association. One standard point of view
maintains that causal laws are required to account for ex-
planatory asymmetries. If only laws of association are ad-
mitted, the length of the shadow can as well explain the
height of the flagpole as the reverse. In The Scientific Image
van Fraassen argues persuasively that these asymmetries
are not genuine. I think he is mistaken. But his case is power-
ful, and it might persuade us to give up certain explanatory
strategies. But we will not so easily be persuaded to give up
our strategies for action, which are essential to practical
life. Perhaps there is no fact of the matter about what ex-
plains what. But there is no doubt that spraying swamps is
an effective way to stop the spread of malaria, whereas
burning the blankets of malarial patients is not. The essay
on causation argues that laws of association are insufficient
to account for the facts about effective strategies. Causal
laws are required as well. Besides defending this central
philosophical thesis the first essay re-introduces Simpson's
paradox to the philosophical literature and it finds in Simp-
son's paradox the underlying source of a variety of different
counterexamples that philosophers have proposed against
probabilistic models of causation.
INTRODUCTION 11
1.1. Composition of causes

Explaining in physics involves two quite different kinds of
activities. First, when we explain a phenomenon, we state
its causes. We try to provide detailed accounts of exactly
how the phenomenon is produced. Second, we fit the phe-
nomenon into a broad theoretical framework which brings
together, under one set of fundamental equations, a wide
array of different kinds of phenomena. Both kinds of ex-
planations use what philosophers have called laws of nature,
but as we have seen in the case of the radiometer, the laws
for the two kinds of explanation do not look at all alike.
The causal story uses highly specific phenomenological
laws which tell what happens in concrete situations. But
the theoretical laws, like the equation of continuity and
Boltzmann's equation, are thoroughly abstract formulae
which describe no particular circumstances.
The standard covering-law account tries to fit both kinds of
explanation into the same mould. But the function of the
laws is different in the two cases, and so too, I have argued,
are their claims to truth. The difference is more than philo-
sophical. We find it in scientific practice (cf. 'The Reality
of Causes in a World of Instrumental Laws'). In physics
it is usual to give alternative theoretical treatments of the
same phenomenon. We construct different models for differ-
ent purposes, with different equations to describe them.
Which is the right model, which the 'true' set of equations?
The question is a mistake. One model brings out some aspects
of the phenomenon; a different model brings out others.
Some equations give a rougher estimate for a quantity of
interest, but are easier to solve. No single model serves all
purposes best.
Causal explanation is different. We do not tell first one
causal story then another, according to our convenience.
Maxwell's explanation involving tangential stresses in the
radiometer is incompatible with the earlier light pressure
account, and it is incompatible with the more standard
hypothesis involving normal pressures. If one of these is
adopted, the others are rejected. Alternative causal stories
compete in physics in a way in which theoretical treatments
12 INTRODUCTION
do not. Causal stories are treated as if they are true or false,

but which theoretical laws 'govern' the phenomenon is a
matter of convenience.
Perhaps laws in physics are not deployed in explanations
as if they are true. But I claim something stronger: if the
evidence is taken seriously, they must be judged false. Why
do I urge this far stronger claim? One reason is the tension
between causal explanation and theoretical explanation.
Physics aims to give both, but the needs of the two are- at
odds with one another. One of the important tasks of a
causal explanation is to show how various causes combine
to produce the phenomenon under study. Theoretical laws
are essential in calculating just what each cause contributes.
But they cannot do this if they are literally true; for they
must ignore the action of laws from other theories to do
the job.
The third essay in this volume asks, 'Do the Laws of
Physics State the Facts?' I answer no. When different kinds
of causes compose, we want to explain what happens in the
intersection of different domains. But the laws we use are
designed only to tell truly what happens in each domain
separately.P This is also the main theme of the second essay.
Realists are inclined to invoke the unity of nature in
reply: the true explanation in cases where causes combine
comes from a 'super' law which unifies the separate domains.
I am dubious about the existence of these unifying laws.
Later I will also claim that I do not believe there are enough
bridge laws of a certain sort. My reasons in both cases are the
same. I think we should believe only in laws for which we
have evidence. Maxwell showed that electromagnetism and
light could be treated together under the same theoretical
umbrella by producing Maxwell's theory, which gives mar-
vellously successful accounts of both. The brilliant applied
mathematician and cosmologist, Stephen Hawking, entitled
his inaugural lecture for the Plume an Professorship at Cam-
bridge, 'Is the End in Sight for Theoretical Physics?' He has
an immense and excited confidence that he and his colleagues
13 There is a very nice paper by Geoffrey Joseph, 'The Many Sciences and
the One World', Journal of Philosophy 77 (1980), pp. 773-90, that argues this
same point.
INTRODUCTION 13
are on the verge of writing down the right equations to unify

the basic forces in nature. We should agree that the end of
theoretical physics is in view only when it is clear they have
done so.
A second reason why I do not believe in these unified laws
is methodological. It runs throughout the essays collected
here. In metaphysics we try to give general models of nature.
We portray it as simple or complex, law-governed or chancy,
unified or diverse. What grounds do we have for our choices?
A priori intuitions and abstract arguments are not good
enough. We best see what nature is like when we look at our
knowledge of it. If our best-supported theories now are
probabilistic, we should not insist on determinism. If Russell
was right that physics does not employ causes, we should
agree with Hume, at least about the basic material phenomena
studied by physics. Unity of science is a case in point. How
unified is our knowledge? Look at any catalogue for a
science or engineering school. The curriculum is divided
into tiny, separate subjects that irk the interdisciplinist.
Our knowledge of nature, nature as we best see it, is highly
compartmentalized. Why think nature itself is unified?
So far I have concentrated on the composition of causes.
But the problems raised by the composition of causes are just
a special case. Even if we do not cross domains or study
causes which fall under different basic laws, still the use of
fundamental laws argues for their falsehood. If the funda-
mental laws are true, they should give a correct account of
what happens when they are applied in specific circumstances.
But they do not. If we follow out their consequences, we
generally find that the fundamental laws go wrong; they are
put right by the judicious corrections of the applied physicist
or the research engineer.
'For Phenomenological Laws' argues this point. Much of
this essay is taken from a joint paper with Jon Nordby!"
and it begins from a view we both share: there are no rigorous
solutions for real life problems. Approximations and adjust-
ments are required whenever theory treats reality. As an
14 Jon Nordby and Nancy Cartwright, 'How Approximation Takes Us Away
from Theory and Towards the Truth' (Pacific Lutheran University and Stanford
University: unpublished manuscript).
14 INTRODUCTION
example of the extent of the shortfall let us look at the

introduction to an advanced text, Perturbation Methods
in Fluid Mechanics by Milton Van Dyke.
Because of this basic non-linearity, exact solutions are rare in any
branch of fluid mechanics ... So great is the need that a solution is
loosely termed 'exact' even when an ordinary differential equation
must be integrated numerically. Lighthill (1948) has given a more
or less exhaustive list of such solutions for inviscid compressible flowY
Van Dyke lists seven cases, then continues,
Again, from Schlichting (1968) one can construct a partial list for in-
compressible viscous flow: 16
The second list gives seven more examples. But even these
fourteen examples do not provide a rigorous tie between
fundamental theory and practical circumstance. Van Dyke
concludes:
It is typical of these self-similar flows that they involve idealized
geometries far from most shapes of practical interest. To proceed
further one must usually approximate. I?
Realistically-oriented philosophers are inclined to think

that approximations raise no problems in principle. The
'true' solution is the rigorous solution, and departures from
it are required only because the mathematics is too difficult
or too cumbersome. Nordby calls approximations that
really aim to estimate the rigorous results ab vera approxi-
mations." Ab vero approximations seem to suit the realist's
case well, but 'For Phenomenological Laws' argues that
even these do not provide positive evidence for the truth
of fundamental laws. Worse for the realist is the widespread
use of ad verum approximation. Here the approximation goes
in the opposite direction. The steps in the derivation move
away from the rigorous consequences of the starting laws,
correcting and improving them, in order to arrive finally
15 Milton Van Dyke, Perturbation Methods in Fluid Mechanics (Standford:
Parabolic Press, 1975), p. 1.
16 Ibid., p. 1. .
17 Ibid., p, 2.
18 Jon Nordby, 'Two Kinds of Approximation in the Application of Science'
(Pacific Lutheran University: unpublished manuscript).
INTRODUCTION 15
at an accurate description of the phenomena. Two kinds of

illustrations are given in 'For Phenomenological Laws',
both taken from the joint paper with Nordby. Both under-
mine the realist's use of inference to the best explanation:
the application of laws to reality by a series of ad verum
approximations argues for their falsehood, not their truth.
Approximations enter when we go from theory to practice.
Consider the reverse direction, not 'theory exit' but 'theory
entry'. In theory entry, we begin with a factual description,
and look to see how it can be brought under a fundamental
law or equation. The canonical method is via a bridge prin-
ciple. But that proposal rests on a too-simple view of how
explanations work. To get from a detailed factual knowledge
of a situation to an equation, we must prepare the descrip-
tion of the situation to meet the mathematical needs of the
theory. Generally the result will no longer be a true descrip-
tion. In 'Fitting Facts to Equations' I give some simple
examples of the kinds of descriptions for which we have
equations in quantum mechanics. Look there to see how
different these are from the kinds of descriptions we would
give if we wanted an accurate report of the facts.
Contrary to the conventional account, which relies solely
on bridge principles, I think theory entry proceeds in two
stages. We start with an unprepared description which gives
as accurate a report as possible of the situation. The first
stage converts this into a prepared description. At the second
stage the prepared description is matched to a mathematical
representation from the theory. Ideally the prepared descrip-
tion should be true to the unprepared. But the two activities
pull in opposite directions, and a description that is adequate
to the facts will seldom have the right mathematical structure.
All this is argued in Essay 7. What I call there 'preparing
a description' is exactly what we do when we produce a
model for a phenomenon, and the two-stage view of theory
entry in that paper lays the groundwork for the account in
Essay 8 which places models at the core of explanation.
1.2 An alternative to the covering-law model of explanation

The third line of argument offers an alternative to the covering-
law model of explanation. Although I do think that we can
16 INTRODUCTION
give causal explanations of isolated events, I shall discuss

here only explanations for kinds of events that recur in a
regular way, events that can be described by phenomeno-
logical laws. I am thinking of the kinds of explanation that
are offered in highly mathematical theories.
I said in the last section that there are two quite different
kinds of things we do when we explain a phenomenon in
physics. First, we describe its causes. Second, we fit the
phenomenon into a theoretical frame. Ever since the earliest
expositions of the covering-law model, it has been objected
that its account of the first kind of explanation is inadequate.
I have been strongly influenced by the criticism due to
Michael Scriven'? and Alan Donagan.i? but many others
make similar points as well. At present Wesley Salmon'" is
developing an alternative account of this kind of explanation,
which focuses on singular causal processes and on causal
interactions. Here I consider only the second kind of explana-
tory activity. How do we fit a phenomenon into a general
theoretical framework?
Prima facie, the covering-law model seems ideally suited
to answer: we fit a phenomenon into a theory by showing
how various phenomenological laws which are true of it
derive from the theory's basic laws and equations. This way
of speaking already differentiates me from the covering-law
theorist. I do not talk about explaining a feature of a phe-
nomenon by deriving a description of that feature; but rather
of treating a phenomenon by deriving a variety of phenomeno-
logical laws about it. But this is not the primary difference.
The 'covering' of 'covering-law model' is a powerful metaphor.
It teaches not only that phenomenological laws can be
derived from fundamental laws, but also that the funda-
mental laws are laws that govern the phenomena. They are
laws that cover the phenomena, perhaps under a more
general or abstract description, perhaps in virtue of some
19 See Michael Scriven, 'Causes Connections and Conditions in History' in
William H. Dray (ed.) Philosophical Analysis and History (New York: Harper &
Row, 1966).
20 See Alan Donagan, 'Explanations in History' in P. Gardiner (ed.), Theories
of History (Glencoe, Illinois: The Free Press, 1959).
21 See Wesley Salmon's new book tentatively titled Scientific Explanation and A
Causal Structure of the World (Princeton: Princeton University Press, forthcoming).
INTRODUCTION 17
hidden micro-structural features; but still the fundamental
laws apply to the phenomena and describe how they occur.
I propose instead a 'simulacrum' account. That is not
a word we use any more, but one of its dictionary defini-
tions captures exactly what I mean. According to the second
entry in the Oxford English Dictionary, a simulacrum is
'something having merely the form or appearance of a
certain thing, without possessing its substance or proper
qualities'. On the simulacrum account, to explain a phe-
nomenon is to construct a model which fits the phenomenon
into a theory. The fundamental laws of the theory are
true of the objects in the model, and they are used to derive
a specific account of how these objects behave. But the
objects of the model have only 'the form or appearance
of things' and, in a very strong sense, not their 'substance
or proper qualities'.
The covering-law account supposes that there is, in prin-
ciple, one 'right' explanation for each phenomenon. The
simulacrum account denies this. The success of an explana-
tory model depends on how well the derived laws approxi-
mate the phenomenological laws and the specific causal
principles which are true of the objects modelled. There
are always more phenomenological laws to be had, and
they can be approximated in better and in different ways.
There is no single explanation which is the right one, even
in the limit, or relative to the information at hand. Theo-
retical explanation is, by its very nature, redundant. This
is one of the endemic features of explanation in physics
which the deductive-nomological (D-N) account misses,
albeit with the plea that this annoying feature will no longer
be present when the end of physics is achieved.
'Fitting Facts to Equations' is full of examples of the
kinds of models we use in explaining things in physics. I
offer the simulacrum account in the next essay as an alter-
native which provides a better description of actual explana-
tory practice than do conventional covering-law accounts.
It obviously serves my attack on fundamental laws. The
'simulacrum' aspect is intended seriously: generally the
prepared and unprepared descriptions cannot be made to
match. But only the prepared descriptions tall under the
18 INTRODUCTION
basic laws. The lesson for the truth of fundamental laws

is clear: fundamental laws do not govern objects in reality;
they govern only objects in models.
2. WHAT DIFFERENCE DOES IT MAKE?
Debate between realists and non-realists has been going

on for a long time. Does the outcome have any practical
consequence? I think it does. The last essay in this volume
provides one example. I have argued that many abstract
concepts in physics play merely an organizing role and do
not seem to represent genuine properties. Unitarity has
the earmarks of being just such a concept in quantum
mechanics.
Unitarity is a property of operators. Those operators
that are unitary represent motions that are indeterministic.
Unitarity plays no causal role in the theory; nothing else
about its use argues for interpreting it as a real property
either. Yet there is a tendency to think of it not only as
a mathematical characteristic of operators, but also as a
genuine property of the situations represented by the opera-
tors. This, I claim, is the source of the notorious measure-
ment problem in quantum mechanics. Unitarity marks no
real property in quantum theory, and if we do not suppose
that it must do so we have no interesting philosophical
problem about measurement.
Essay 9 is concerned exclusively with the measurement
problem, and not with the other conceptual difficulty in
quantum mechanics that philosophers commonly discuss,
the Einstein-Podolsky-Rosen paradox. These two problems
tend to sit at opposite ends of a balance: philosophical
treatments that offer hope for solving one usually fare
badly with the other. This is certainly the case with the
programme I propose. If it should work, it would at best
eliminate the measurement problem, which I believe to be a
pseudo-problem. But it would have nothing to say regarding
the very perplexing facts about locality brought out by the
EPR paradox.
INTRODUCTION 19
3. CONCLUSION
The picture of science that I present in these essays lacks the
purity of positivism. It is a jumble of unobservable entities,
causal processes, and phenomenological laws. But it shares
one deep positivist conviction: there is no better reality
besides the reality we have to hand. In the second sentence
of this introduction I characterized the distinction between
phenomenological and theoretical laws: phenomenological
laws are about appearances; theoretical ones about the reality
behind the appearances. That is the distinction I reject.
Richard Feynmann talks about explaining in physics as
fitting phenomena into 'the patterns of nature'. But where
are the patterns? Things happen in nature. Often they happen
in regular ways when the circumstances are similar; the
same kinds of causal processes recur; there are analogies
between what happens in some situations and what happens
in others. As Duhem suggests, what happens may even be
organized into natural kinds in a way that makes prediction
easy for us (see the last section of 'When Explanation Leads
to Inference'). But there is only what happens, and what
we say about it. Nature tends to a wild profusion, which
our thinking does not wholly confine.
The metaphysical picture that underlies these essays is
an Aristotelean belief in the richness and variety of the
concrete and particular. Things are made to look the same
only when we fail to examine them too closely. Pierre
Duhem distinguished two kinds of thinkers: the deep but
narrow minds of the French, and the broad but shallow
minds of the English. The French mind sees things in an
elegant, unified way. It takes Newton's three laws of motion
and turns them into the beautiful, abstract mathematics of
Lagrangian mechanics. The English mind, says Duhem, is
an exact contrast. It engineers bits of gears, and pulleys, and
keeps the strings from tangling up. It holds a thousand
different details all at once, without imposing much abstract
order or organization. The difference between the realist
and me is almost theologica1. The realist thinks that the
creator of the universe worked like a French mathematician.
But I think that God has the untidy mind of the English.
20 INTRODUCTION
GUIDE FOR THE READER
The last essay on quantum mechanics shows how anti-

realism can be put to work. The first essay defends causes.
The principal arguments for theoretical entities and against
theoretical laws are in the middle essays. Although the
essays argue in favour of theoretical entities and against
theoretical laws, the main emphasis is on the latter theme.
This book is a complement, I think, to the fine discussions
of representation, experimentation, and creation of phe-
nomena in Ian Hacking's Representing and Intervening. 22
Hacking provides a wealth of examples which show how
new entities are admitted to physics. Essay 6 here has some
detailed examples and many equations that may not be of
interest to the reader with pure philosophic concerns; but
the general point of the examples can be gleaned by reading
the introductory sections. Although Essay 9 is about quan-
tum mechanics, it is not technical and readers without
expert knowledge will be able to follow the argument.
22 Ian Hacking, Representing and Intervening (Cambridge: Cambridge Uni-

versity Press, forthcoming).
Essay 1
Causal Laws and Effective Strategies
O. INTRODUCTION
There are at least two kinds of laws of nature: laws of associa-

tion and causal laws. Laws of association are the familiar
laws with which philosophers usually deal. These laws tell
how often two qualities or quantities are co-associated. They
may be either deterministic-the association is universal-or
probabilistic. The equations of physics are a good example:
whenever the force on a classical particle of mass m is I the
acceleration is 11m. Laws of association may be time indexed,
as in the probabilistic laws of Mendelian genetics, but,
apart from the asymmetries imposed by time indexing,
these laws are causally neutral. They tell how often two
qualities co-occur; but they provide no account of what
makes things happen.
Causal laws, by contrast, have the word 'cause'-or
some causal surrogate-right in them. Smoking causes lung
cancer; perspiration attracts wood ticks; or, for an example
from physics, force causes change in motion: to quote
Einstein and Infeld, 'The action of an external force changes
the velocity ... such a force either increases or decreases
the velocity according to whether it acts in the direction
of motion or in the opposite direction.'!
Bertrand Russell argued that laws of association are all
the laws there are, and that causal principles cannot be
derived from the causally symmetric laws of association.?
I shall here argue in support of Russell's second claim, but
against the first. Causal principles cannot be reduced to
laws of association; but they cannot be done away with.
The argument in support of causal laws relies on some
facts about strategies. They are illustrated in a letter which
I Albert Einstein and Leopold Infeld, The Evolution of Physics (Cambridge:

Cambridge University Press, 1971), p. 9.
2 Bertrand Russell, 'On the Notion of Cause', Proceedings of the Aristotelian
Society 13 (1912-13), pp. 1-26.

22 CAUSAL LAWS AND EFFECTIVE STRATEGIES
I recently received from TIAA-CREF, a company that

provides insurance for college teachers. The letter begins:
It simply wouldn't be true to say,

'Nancy L. D. Cartwright . . . if you own a TIAA life insurance
policy you'll live longer.'
But it is a fact, nonetheless, that persons insured by TIAA do enjoy
longer lifetimes, on the average, than persons insured by commercial
insurance companies that serve the general public.
I will take as a starting point for my argument facts like

those reported by the TIAA letter: it wouldn't be true that
buying a TIAA policy would be an effective strategy for
lengthening one's life. TIAA may, of course, be mistaken;
it could after all be true. What is important is that their
claim is, as they suppose, the kind of claim which is either
true or false. There is a pre-utility sense of goodness of
strategy; and what is and what is not a good strategy in this
pre-utility sense is an objective fact. Consider a second
example. Building the canal in Nicaragua, the French dis-
. covered that spraying oil on the swamps is a good strategy
for stopping the spread of malaria, whereas burying con-
taminated blankets is useless. What they discovered was
true, independent of their theories, of their desire to control
malaria, or of the cost of doing so.
The reason for beginning with some uncontroversial
examples of effective and ineffective strategies is this: I
claim causal laws cannot be done away with, for they are
needed to ground the distinction between effective strategies
and ineffective ones. If indeed, it isn't true that buying a
TIAA policy is an effective way to lengthen one's life, but
stopping smoking is, the difference between the two depends
on the causal laws of our universe, and on nothing weaker.
This will be argued in Part 2. Part I endorses the first of
Russell's claims, that causal laws cannot be reduced to laws
of association.
1. STATISTICAL ANALYSES OF CAUSATION
I will abbreviate the causal law, 'C causes E' by C C+ E.

Notice that C and E are to be filled in by general terms,
CAUSAL LAWS AND EFFECTIVE STRATEGIES 23
and not names of particulars; for example, 'Force causes

motion' or 'Aspirin relieves headache'. The generic law 'C
causes E' is not to be understood as a universally quantified
law about particulars, even about particular causal facts. It
is generically true that aspirin relieves headache even though
some particular aspirins fail to do so. I will try to explain
what causal laws assert by giving an account of how causal
laws relate on the one hand to statistical laws, and on the
other to generic truths about strategies. The first task is not
straightforward; although causal laws are intimately con-
nected with statistical laws, they cannot be reduced to them.
A primary reason for believing that causal laws cannot be
reduced to probabilistic laws is broadly inductive: no attempts
so far have been successful. The most notable attempts
recently are by the philosophers Patrick Suppes" and Wesley
Salmon" and, in the social sciences, by a group of sociologists
and econometricians working on causal models, of whom
Herbert Simon and Hubert Blalock" are good examples.
It is not just that these attempts fail, but rather why they
fail that is significant. The reason is this. As Suppes urges,
a cause ought to increase the frequency of its effect. But
this fact may not show up in the probabilities if other causes
are at work. Background correlations between the pur-
ported cause and other causal factors may conceal the in-
crease in probability which would otherwise appear. A
simple example will illustrate.
It is generally supposed that smoking causes heart disease
(S 4 H). Thus we may expect that the probability of heart
disease on smoking is greater than otherwise. (We can
write. this as either Prob(HjS) > Prob(H), or Prob(HjS) >
Prob(Hh S), for the two are equivalent.) This expectation is
mistaken. Even if it is true that smoking causes heart disease,
the expected increase in probability will not appear if smok-
ing is correlated with a sufficiently strong preventative, say
3 See Patrick Suppes, A Probabilistic Theory of Causality (Amsterdam:
North-Holland Publishing Co., 1970).
4 See Wesley Salmon, 'Statistical Explanation' in Salmon, Wesley (ed.),
Statistical Explanation and Statistical Relevance (Pittsburgh: University of
Pittsburgh Press, 1971).
5 See H. M. Blalock, Jr., Causal Models in the Social Sciences (Chicago:
Aldine-Atherton, 1971).
exercising. (Leaving aside some niceties, we can render

'Exercise prevents heart disease' as X 4 I H.) To see why
this is so, imagine that exercising is more effective at pre-
venting heart disease than smoking at causing it. Then in
any population where smoking and exercising are highly
enough correlated," it can be true that Prob(HIS) = Prob(H),
or even Prob(HIS) < Prob(H). For the population of smokers
also contains a good many exercisers, and when the two
are in combination, the exercising tends to dominate.
It is possible to get the increase in conditional probability
to reappear. The decrease arises from looking at probabilities
that average over both exercisers and non-exercisers. Even
though in the general population it seems better to smoke
than not, in the population consisting entirely of exercisers,
it is worse to smoke. This is also true in the population of
non-exercisers. The expected increase in probability occurs
not in the general population but in both sub-populations.
This example depends on a fact about probabilities known
as Simpson's paradox," or sometimes as the Cohen-Nagel-
Simpson paradox, because it is presented as an exercise
in Morris Cohen's and Ernest Nagel's text, An Introduction
to Logic and Scientific Method." Nagel suspects that he
learned about it from G. Yule's An Introduction to the
Theory of Statistics (1904), which is one of the earliest
textbooks written on statistics; and indeed it is dicussed
at length there. The fact is this: any association-Prob(AIB)
= Prob(A); Prob(AIB) > Prob(A); Prob(A/B) < Prob(A)-
between two variables which holds in a given population
can be reversed in the sub-populations by finding a third
variable which is correlated with both.
In the smoking-heart disease example, the third factor
is a preventative factor for the effect in question. This is
just one possibility. Wesley Salmon? has proposed different
examples to show that a cause need not increase the prob-
ability of its effect. His examples also turn on Simpson's
6Throughout, 'A and B are correlated' will mean Prob(A/B) *- Prob(A).
7E. H. Simpson, 'The Interpretation of Interaction in Contingency Tables',
Journal of the Royal Statistical Society, Ser. B. 13 (1951), pp. 238-41.
• See Morris R. Cohen and Ernest Nagel, An Introduction to Logic and
Scientific Method (New York: Harcourt, Brace and Co., 1934).
9_SeeWesley Salmon, op. cit.,
paradox, except that in his cases the cause is correlated,

not with the presence of a negative factor, but with the
absence of an even more positive one.
Salmon considers two pieces of radioactive material,
uranium 238 and polonium 214. We are to draw at random
one material or the other, and place it in front of a Geiger
counter for some time. The polonium has a short half-life,
so that the probability for some designated large number
of clicks is .9; for the long-lived uranium, the probability
is .1. In the situation described, where one of the two pieces
is drawn at random, the total probability for a large number
of clicks is !(.9) + !(.1) = .5. So the conditional probability
for the Geiger counter to click when the uranium is present
is less than the unconditional probability. But when the
uranium has been drawn and the Geiger counter does register
a large number of clicks, it is the uranium that causes them.
The uranium decreases the probability of its effect in this
case. But this is only because the even more effective polon-
ium is absent whenever the uranium is present.
All the counter examples I know to the claim that causes
increase the probability of their effects work in this same way.
In all cases the cause fails to increase the probability of its
effects for the same reason: in the situation described the cause
is correlated with some other causal factor which dominates in
its effects. This suggests that the condition as stated is too
simple. A cause must increase the probability of its effects;
but only in situations where such correlations are absent.
The most general situations in which a particular factor is
not correlated with any other causal factors are situations
in which all other causal factors are held fixed, that is situa-
tions that are homogeneous with respect to all other causal
factors. In the population where everyone exercises, smoking
cannot be correlated with exercising. So, too, in populations
where no-one is an exerciser. I hypothesize then that the
correct connection between causal laws and laws of associa-
tion is this:
'C causes E' if and only if C increases the probability of

E in every situation which is otherwise causally homo-
geneous with respect to E.
Carnap's notion of a state description!" can be used to

pick out the causally homogeneous situations. A complete
set of causal factors for E is the set of all C, such that either
C, c., + E or C, c., ,E. (For short C, c., ±E.) Every possible
arrangement of the factors from a set which is complete
except for C picks out a population homogeneous in all
causal factors but C. Each such arrangement is given by
one of the 2n state descriptions K; = /\ ±Cj over the set
{Cj} (i ranging from 1 to n) consisting of all alternative
causal factors. These are the only situations in which prob-
abilities tell anything about causal laws. I will refer to them
as test situations for the law C c., E.
Using this notation the connection between laws of
association and causal laws is this:
CC: C c., E iff Prob(E/CKj) > Prob(E/Kj) for all state
descriptions Kj over the set {Cj}, where {Cj} satisfies
(i) qE{Cj}=>q~±E
(ii) C Et {Cj}
(iii) VD (D c., ±E => D = C or D E { CJ)
(iv) CjE{q} =>-, (C4G).
Condition (iv) is added to ensure that the state descriptions

do not hold fixed any factors in the causal chain from
C to E. It will be discussed further in the section after next.
Obviously CC does not provide an analysis of the schema
C c., E, because exactly the same schema appears on both
sides of the equivalence. But it does impose mutual con-
straints, so that given sets of causal and associational laws
cannot be arbitrarily conjoined. CC is, I believe, the strongest
connection that can be drawn between causal laws and laws
of association.
1.1 Two Advantages for Scientific Explanation
C. G. Hempel's original account of inductive-statistical
explanation'! had two crucial features which have been
given up in later accounts, particularly in Salmon's: (1) an
10 See Rudolf Carnap, The Continuum of Inductive Methods (Chicago:
University of Chicago Press, 1952).
11 See C. G. Hempel, Aspects of Scientific Explanation (New York: Free
Press, 1965).
explanatory factor must increase the probability of the

fact to be explained; (2) what counts as a good explanation
is an objective, person-independent matter. Both of these
features seem to me to be right. If we use causal laws in
explanations, we can keep both these requirements and
still admit as good explanations just those cases that are
supposed to argue against them.
(i) Hempel insisted that an explanatory factor increase
the probability of the phenomenon it explains. This is an
entirely plausible requirement, although there is a kind of
explanation for which it is not appropriate. In one sense,
to explain a phenomenon is to locate it in a nomic pattern.
The aim is to layout all the laws relevant to the phenomenon;
and it is irrelevant to this aim whether the phenomenon has
high or low probability under these laws. Although this
seems to be the kind of explanation that Richard Jeffrey
describes in 'Statistical Explanation vs. Statistical Infer-
ence',12 it is not the kind of explanation that other of
Hempel's critics have in mind. Salmon, for instance, is
clearly concerned with causal explanation.P Even for causal
explanation Salmon thinks the explanatory factor may
decrease the probability of the factor to be explained. He
supports this with the uranium-plutonium example described
above.
What makes the uranium count as a good explanation for
the clicks in the Geiger counter, however, is not the probabil-
istic law Salmon cites (Prob( clicks/uranium) < Probeclicks»,
but rather the causal law-'U:ranium causes radioactivity'.
As required, the probability for radioactive decay increases
when the cause is present, for every test situation. There is
a higher level of radioactivity when uranium is added both
for situations in which polonium is present, and for situations
in which polonium is absent. Salmon sees the probability
12 See Richard C. Jeffrey, 'Statistical Explanation vs. Statistical Inference',

in Wesley Salmon, op, cit.
13 This is explicitly stated in Salmon's later papers (see 'Theoretical Explana-
tion'in S. Korner, Explanation (Oxford: Basil Blackwell, 1975), but it is already
clear from the treatment in 'Statistical Explanation' that Salmon is concerned
with causal explanations, otherwise there is no accounting for his efforts to
rule out 'spurious' correlations as explanatory.
decreasing because he attends to a population which is not

causally homogeneous.
Insisting on increase in probability across all test situations
not only lets in the good cases of explanation which Salmon
cites; it also rules out some bad explanations that must
be admitted by Salmon. For example, consider a case which,
so far as the law of association is concerned, is structurally
similar to Salmon's uranium example. I consider eradicating
the poison oak at the bottom of my garden by spraying it
with defoliant. The can of defoliant claims that the spray
is 90 per cent effective; that is, the probability of a plant's
dying given that it is sprayed is .9, and the probability of
its surviving is .1. Here in contrast to the uranium case
only the probable outcome, and not the improbable, is
explained by the spraying. One can explain why some plants
died by remarking that they were sprayed with a powerful
defoliant; but this will not explain why some survive.!"
The difference is in the causal laws. In the favourable
example, it is true both that uranium causes high levels
of radioactivity and that uranium causes low levels of radio-
activity. This is borne out in the laws of association. Holding
fixed other causal factors for a given level of decay, either
high or low, it is more probable that that level will be reached
if uranium is added than not. This is not so in the unfavour-
able case. It is true that spraying with defoliant causes death
in plants, but it is not true that spraying also causes survival.
Holding fixed other causes of death, spraying with my
defoliant will increase the probability of a plant's dying;
but holding fixed other causes of survival, spraying with
that defoliant will decrease, not increase, the chances of
a plant's surviving.
(ii) All these explanations are explanations by appeal to
causal laws. Accounts, like Hempel's or Salmon's or Suppes's,
which instead explain by appeal to laws of association, are
plagued by the reference class problem. All these accounts
allow that one factor explains another just in case some
privileged statistical relation obtains between them. (For
14 This example can be made exactly parallel to the uranium-polonium
case by imagining a situation in which we choose at random between this de-
foliant and a considerably weaker one that is only 10 per cent effective.
Hempel the probability of the first factor OIl the second

must be high; for Suppes it must be higher than when the
second factor is absent; Salmon merely requires that the
probabilities be different.) But whether the designated
statistical relation obtains or not depends on what reference
class one chooses to look in, or on what description one
gives to the background situation. Relative to the description
that either the uranium or the polonium is drawn at random,
the probability of a large number of clicks is lower when
the uranium is present than it is otherwise. Relative to the
description that polonium and all other radio-active sub-
stances are absent, the probability is higher.
Salmon solves this problem by choosing as the privileged
description the description assumed in the request for expla-
nation. This makes explanation a subjective matter. Whether
the uranium explains the clicks depends on what informa-
tion the questioner has to hand, or on what descriptions
are of interest to him. But the explanation that Hempel
aimed to characterize was in no way subjective. What ex-
plains what depends on the laws and facts true in our world,
and cannot be adjusted by shifting our interest or our focus.
Explanation by causal law satisfies this requirement.
Which causal laws are true and which are not is an objective
matter. Admittedly certain statistical relations must obtain;
the cause must increase the probability of its effect. But
no reference class problem arises. In how much detail should
we describe the situations in which this relation must obtain?
We must include all and only the other causally relevant
features. What interests we have, or what information we
focus on, is irrelevant.
I will not here offer a model of causal explanation, but
certain negative theses follow from my theory. Note par-
ticularly that falling under a causal law (plus the existence
of suitable initial conditions) is neither necessary nor suffic-
ient for explaining a phenomenon.
It is not sufficient because a single phenomenon may be
in the domain of various causal laws, and in many cases it
will be a legitimate question to ask, 'Which of these causal
factors actually brought about the effect on this occasion?'
This problem is not peculiar to explanation by causal law,
however. Both Hempel in his inductive-statistical model

and Salmon in the statistical relevance account sidestep the
issue by requiring that a 'full' explanation cite all the possibly
relevant factors, and not select among them.
Conversely, under the plausible assumption that singular
causal statements are transitive, falling under a causal law
is not necessary for explanation either. This results from the
fact that (as CC makes plain) causal laws are not transitive.
Hence a phenomenon may be explained by a factor to which
it is linked by a sequence of intervening steps, each step
falling under a causal law, without there being any causal
law that links the explanans itself with the phenomenon
to be explained.
1.2 Some Details and Some Difficulties
Before carrying on to Part 2, some details should be noted
and some defects admitted.
(a) Condition (tv). Condition (iv) is added to the above
characterization to avoid referring to singular causal facts.
A test situation for C 4 E is meant to pick out a (hypo-
thetical, infinite) population of individuals which are alike
in all causal factors for E, except those which on that occa-
sion are caused by C itself. The test situations should not
hold fixed factors in the causal chain from C to E. If it
did so, the probabilities in the populations where all the
necessary intermediate steps occur would be misleadingly
high; and where they do not occur, misleadingly low. Con-
dition (iv) is added to except factors caused by C itself from
the description of the test situation. Unfortunately it is too
strong. For condition (iv) excepts any factor which may
be caused by C even on those particular occasions when
the factor occurs for other reasons. Still, (iv) is the best
method I can think of for dealing with this problem, short
of introducing singular causal facts, and I let it stand for
the nonce.
(b) Interactions. One may ask, 'But might it not happen
that Prob(E/C) > Prob(E) in all causally fixed circumstances,
and still C not be a cause of E?' I do not know. I am unable
to imagine convincing examples in which it occurs; but that
is hardly an answer. But one kind of example is clearly taken
account of. That is the problem of spurious correlation

(sometimes called 'the problem of joint effects'). If two
factors £1 and £2 are both effects of a third factor e, then
it will frequently happen that the probability of the first
factor is greater when the second is present than otherwise,
over a wide variety of circumstances. Yet we do not want
to assert £1 4 £2' According to principle ee, however,
£14£2 only if Prob(£d£2) > Prob(£1) both when e ob-
tains, and also when e does not obtain. But the story that
£1 and £2 are joint effects of e provides no warrant for
expecting either of these increases.
One may have a worry in the other direction as well.
Must a cause increase the probability of its effect in every
causally fixed situation? Might it not do so in some, but not
in all? I think not. Whenever a cause fails to increase the
probability of its effect, there must be a reason. Two kinds
of reasons seem possible. The first is that the cause may be
correlated with other causal factors. This kind of reason
is taken account of. The second is that interaction may
occur. Two causal factors are interactive if in combination
they act like a single causal factor whose effects are different
from at least one of the two acting separately. For example,
ingesting an acid poison may cause death; so too may the
ingestion of an alkali poison. But ingesting both may have
no effect at all on survival.
In this case, it seems, there are three causal truths: (1)
ingesting acid without ingesting alkali causes death; (2) in-
gesting alkali without ingesting acid causes death; and (3)
ingesting both alkali and acid does not cause death. All
three of these general truths should accord with ec.
Treating interactions in this way may seem to trivialize
the analysis; anything may count as a cause. Take any factor
that behaves sporadically across variation of causal circum-
stances. May we not count it as a cause by looking at it
separately in those situations where the probability increases,
and claim it to be in interaction in any case where the prob-
ability does not increase? No. There is no guarantee that
this can always be done. For interaction is always inter-
action with some other causal factor; and it is not always
possible to find some other factor, or conjunction of factors,
which obtain just when the probability of E on the factor

at issue decreases, and which itself satisfies principle CC
relative to all other causal factors. 15 Obviously, consider-
ably more has to be said about interactions; but this fact at
least makes it reasonable to hope they can be dealt with
adequately, and that the requirement of increase in prob-
ability across all causal situations is not too strong.
(c) 0, 1 probabilities and threshold effects. Principle CC
as it stands does not allow C 4 E if there is even a single
arrangement of other factors for which the probability of
E is one, independent of whether C occurs or not. So CC
should be amended to read:
C4 E iff (Vj){ Prob(E/C.~) > Prob(E/~) or Prob(E/~)

= I = Prob(E/C.Kj)} and (3jHProb(E/Kj) =1= I}.
It is a consequence of the second conjunct that something

that occurs universally can be the consequent of no
causal laws. The alternative is to let anything count as the
cause of a universal fact.
There is also no natural way to deal with threshold
effects, if there are any. If the probability of some phe-
nomenon can be raised just so high, and no higher, the
treatment as it stands allows no genuine causes for it.
(d) Time and causation. CC makes no mention of
time. The properties may be time indexed; taking aspirins
at t causes relief at t + t:J.t,but the ordering of the indices
plays no part in the condition. Time ordering is often
introduced in statistical analyses of causation to guarantee
the requisite asymmetries. Some, for example, take in-
crease in conditional probability as their basis. But the
causal arrow is asymmetric, whereas increase in condi-
tional probability is symmetric: Prob(E/C) > Prob(E) iff
Prob(C/E) > Prob (C). This problem does not arise for
CC, because the set of alternative causal factors for E
will be different from the set of alternative causal factors
for C. I take it to be an advantage that my account leaves
open the question of backwards causation. I doubt that
we shall ever find compelling examples of it; but if there
15 See section 3.
were a case in which a later factor increased the prob-
ability of an earlier one in all test situations, it might
well be best to count it a cause.
2. PROBABILITIES IN DECISION THEORY
Standard versions of decision theory require two kinds

of information. (1) How desirable are various combina-
tions of goals and strategies and (2) how effective are
various strategies for obtaining particular goals. The first
is a question of utilities, which I will not discuss. The
second is a matter of effectiveness; it is generally rendered
as a question about probabilities. We need to know what
may roughly be characterized as 'the probability that the
goal will result if the strategy is followed.' It is customary
to measure effectiveness by the conditional probability.
Following this custom, we could define
!S is an effective strategy for G iff Prob(G/S) > Prob(G).
I have here used the volative mood marker ! introduced
by H. P. Grice.l" to be read 'let it be the case that'. I shall
refer to S as the strategy state. For example, if we want
to know whether the defoliant is effective for killing
poison oak, the relevant strategy state is 'a poison oak
plant is sprayed with defoliant'. On the above character-
ization, the defoliant is effective just in case the prob-
ability of a plant's dying, given that it has been sprayed,
is greater than the probability of its dying given that it
has not been sprayed. Under this characterization the
distinction between effective and ineffective strategies
depends entirely on what laws of association obtain.
But the conditional probability will not serve in this
way, a fact that has been urged by Allan Gibbard and
William Harper.!? Harper and Gibbard point out that the
increase in conditional probability may be spurious, and
16 H. P. Grice, 'Some Aspects of Reason', the Immanuel Kant Lectures,
Stanford University, 1977.
17 See Allan Gibbard and William Harper, 'Counterfactuals and Two
Kinds of Expected Utility'. Discussion Paper No. 194, Center for Mathe-
matical Studies in Economics and Management Science Northwestern Uni-
versity, January 1976.
that spurious correlations are no grounds for action. Their

own examples are somewhat complex because they specific-
ally address a doctrine of Richard Jeffrey's not immediately
to the point here. We can illustrate with the TIAA case
already introduced. The probability of long life given that
one has a TIAA policy is higher than otherwise. But, as the
letter says, it would be a poor strategy to buy TIAA in
order to increase one's life expectancy.
The problem of spurious correlation in decision theory
leads naturally to the introduction of counterfactuals.
We are not, the argument goes, interested in how many
people have long lives among people insured by TIAA,
but rather in the probability that one would have a long
life if one were insured with TIAA. Apt as this suggestion is,
it requires us to evaluate the probability of counterfactuals,
for which we have only the beginnings of a semantics (via
the device of measures over possible worldsj '" and no metho-
dology, much less an account of why the methodology is
suited to the semantics. How do we test claims about prob-
abilities of counterfactuals? We have no answer, much
less an answer that fits with our nascent semantics. It would
be preferable to have a measure of effectiveness that re-
quires only probabilities over events that can be tested
in the actual world in the standard ways. This is what I
shall propose.
The Gibbard and Harper example, an example of spurious
correlation due to a joint cause, is a special case of a general
problem. We saw that the conditional probability will not
serve as a mark of causation in situations where the putative
cause is correlated with other causal factors. Exactly the
same problem arises for effectiveness. For whatever reason
the correlation obtains, the conditional probability is not a
good measure of effectiveness in any populations where the
strategy state is correlated with other factors causally rele-
vant to the goal state. Increase in conditional probability
is no mark of effectiveness in situations which are causally
heterogeneous. It is necessary, therefore, to make the same
18 See William Harper, Robert Stalnaker, and Glenn Pearce (eds), University
of Western Ontario Series in Philosophy of Science (Dordrecht: D. Reidel Pub-
lishing Co., 1981).
restrictions about test situations in dealing with strategies
that we made in dealing with causes:
!S is an effective strategy for obtaining G in situation
L iffProb(G/S.Kd > Prob(G/Kd.
Here KL is the state description true in L, taken over the
complete set {Gil of causal factors for G, barring S. But
L may not fix a unique state description. For example
L may be the situation I am in when I decide whether
to smoke or not, and at the time of the decision it is not
determined whether I will be an exerciser. In that case
we should compare not the actual values Prob(G/S.KL)
and Prob(G/KL), but rather their expected values:
SC: !S is an effective strategy for obtaining G in L iff
L Prob(G/S.Kj)Prob(Kj) > L Prob(G/Kj) Prob(Kj),

j j
where j ranges over all Kj consistent with L. 19
This formula for computing the effectiveness of strategies

has several desired features: (l) it is a function of the prob-
ability measure, Prob, given by the laws of association in
the actual world; and hence calculable by standard methods
of statistical inference. (2) It reduces to the conditional
probability in cases where it ought. (3) It restores a natural
connection between causes and strategies.
(1) SC avoids probabilities over counterfactuals. Implica-
tions of the arguments presented here for constructing a
semantics for probabilities for counterfactuals will be pointed
out in section 2.2.
(2) Troubles for the conditional probability arise in
cases like the TIAA example in which there is a correlation
between the proposed strategy and (other) causal factors
for the goal in question. When such correlations are absent,
the conditional probability should serve. This follows im-
mediately: when there are no correlations between Sand
.9 I first derived this formula by reasoning about experiments. I am especially
grateful to David Lewis for pointing out that the original formula was mathe-
matically equivalent to the shorter and more intelligible one presented here.
other causal factors, Prob(Rj/S) = Prob(Rj); so the left-

hand side of SC reduces to Prob(G/S) in the situation L
and the right-hand side to Prob(G) in L.
(3) There is a natural connection between causes and
strategies that should be maintained; if one wants to obtain
a goal, it is a good (in the pre-utility sense of good) strategy
to introduce a cause for that goal. So long as one holds
both the simple view that increase in conditional probability
is a sure mark of causation and the view that conditional
probabilities are the right measure of effectiveness, the
connection is straightforward. The arguments in Part I
against the simple view of causation break this connection.
But SC re-establishes it, for it is easy to see from the com-
bination of CC and SC that if X 4 G is true, then !X will
be an effective strategy for G in any situation.
2.1. Causal Laws and Effective Strategies

Although SC joins causes and strategies, it is not this con-
nection that argues for the objectivity of sui generis causal
laws. As we have just seen, one could maintain the connec-
tion between causes and strategies, and still hope to eliminate
causal laws by using simple conditional probability to treat
both ideas. The reason causal laws are needed in character-
izing effectiveness is that they pick out the right properties
on which to condition. The Rj which are required to char-
acterize effective strategies must range over all and only the
causal factors for G.
It is easy to see, from the examples of Part I, why the K,
must include all the causal factors. If any are left out, cases
like the smoking-heart disease example may arise. If exercis-
ing is not among the factors which Rj fixes, the conditional
probability of heart disease on smoking may be less than
otherwise in Rj, and smoking will wrongly appear as an
effective strategy for preventing heart disease.
It is equally important that the Rj not include too much.
{Rj} partitions the space of possible situations. To partition
too finely is as bad as not to partition finely enough. Parti-
tioning on an irrelevancy can make a genuine cause look
irrelevant, or make an irrelevant factor look like a cause.
Earlier discussion of Simpson's paradox shows that this is
structurally possible. Any association between two factors

C and E can be reversed by finding a third factor which is
correlated in the right way with both. When the third factor
is a causal factor, the smaller classes are the right ones to use
for judging causal relations between C and E. In these,
whatever effects the third factor has on E are held fixed
in comparing the effects of C versus those of --,C. But when
the third factor is causally irrelevant to E-that is, when it
has no effects on E-there is no reason for it to be held
fixed, and holding it fixed gives wrong judgements both
about causes and about strategies.
I will illustrate from a real life case.P? The graduate school
at Berkeley was accused of discriminating against women in
their admission policies, thus raising the question 'Does being
a woman cause one to be rejected at Berkeley?' The accusa-
tion appeared to be borne out in the probabilities: the
probability of acceptance was much higher for men than for
women. Bickel, Hammel, and O'Connelf" looked at the data
-more carefully, however, and discovered that this was no
longer true if they partitioned by department. In a majority
of the eighty-five departments, the probability of admission
for women was just about the same as for men, and in some
even higher for women than for men. This is a paradigm
of Simpson's paradox. Bickel, Hammel and O'Connell ac-
counted for the 'paradoxical reversal of associations by
pointing out that women tended to apply to departments
with high rejection rates, so that department by department
women were admitted in about the same ratios as men;
but across the whole university considerably fewer women,
by proportion, are admitted.
This analysis seems to exonerate Berkeley from the charge
of discrimination. But only because of the choice of par-
titioning variable. If, by contrast, the authors had pointed
out that the associations reversed themselves when the
.0 Roger Rosenkrantz and Persi Diaconis first pointed out to me that the
feature of probabilities described here is called 'Simpson's Paradox', and the
reference for this example was supplied by Diaconis .
•1 See Peter J. Bickel, Eugene A. Hammel and J. William O'Connell, 'Sex Bias
in Graduate Admissions: Data from Berkeley', in William B. Fairley and Frederick
Mosteller, Statistics and Public Policy (Reading, Mass: Addison-Wesley, 1977).
applicants were partitioned according to their roller skating

ability that 'vould count as no defence.P Why is this so?
The difference between the two situations lies in our
antecedent causal knowledge. We know that applying to a
popular department (one with considerably more applicants
than positions) is just the kind of thing that causes rejection.
But without a good deal more detail, we are not prepared
to accept the principle that being a good roller skater causes
a person to be rejected by the Berkeley graduate school,
and we make further causal judgements accordingly. If the
increased probability for rejection among women disappears
when a causal variable is held fixed, the hypothesis of dis-
crimination in admissions is given up; but not if it disappears
only when some causally irrelevant variable is held fixed.
The Berkeley example illustrates the general point: only
partitions by causally relevant variables count in evaluating
causal laws. If changes in probability under causally irrelevant
partitions mattered, almost any true causal law could be
defeated by finding, somewhere, some third variable that
correlates in the right ways to reverse the required association
between cause and effect.
2.2. Alternative Accounts which Employ 'True
Probabilities' or Counterfactuals
One may object: once all causally relevant factors have been
fixed, there is no harm in finer partitioning by causally
irrelevant factors. Contrary to what is claimed in the remarks
about roller skating and admission rates, further partitioning
will not change the probabilities. There is a difference be-
tween true probabilities and observed relative frequencies.
Admittedly it is likely that one can always find some third,
irrelevant, variable which, on the basis of estimates from
finite data, appears to be correlated with both the cause
and effect in just the ways required for Simpson's paradox.
But we are concerned here not with finite frequencies, or
estimates from them, but rather with true probabilities.
You misread the true probabilities from the finite data, and
think that correlations exist where they do not.
22 William Kruskal discusses the problem of choosing a partition for these data
briefly in correspondence following the Bickel, Hammel, and O'Connell article
referred to in footnote 21.
For this objection to succeed, an explication is required

of the idea of a true probability, and this explication must
make plausible the claim that partitions by what are pre-
analytically regarded as non-causal factors do not result in
different probabilities. It is not enough to urge the general
point that the best estimate often differs from the true prob-
ability; there must in addition be reason to think that that
is happening in every case where too-fine partitioning seems
to generate implausible causal hypotheses. This is not an
easy task, for often the correlations one would want to
classify as 'false' are empirically indistinguishable from
others that ought to be classified 'true'. The misleading,
or 'false', correlations sometimes pass statistical tests of
any degree of stringency we are willing to accept as a general
requirement for inferring probabilities from finite data.
They will often, for example, be stable both across time
and across randomly selected samples.
To insist that these stable frequencies are not true prob-
abilities is to give away too much of the empiricist programme.
In the original this programme made two assumptions.
First, claims about probabilities are grounded only in stable
frequencies. There are notorious problems about finite versus
infinite ensembles, but at least this much is certain: what
probabilities obtain depends in no way, either epistemo-
logically or metaphysically, on what causal assumptions are
made. Secondly, causal claims can be reduced completely
to probabilistic claims, although further empirical facts
may be required to ensure the requisite asymmetries.
I attack only the second of these two assumptions. Prior
causal knowledge is needed along with probabilities to
infer new causal laws. But I see no reason here to give up
the first, and I think it would be a mistake to do so. Prob-
abilities serve many other concerns than causal reasoning
and it is best to keep the two as separate as possible. In
his Grammar of Science Karl Pearson taught that probabili-
ties should be theory free, and I agree. If one wishes never-
theless to mix causation and probability from the start then
at least the arguments I have been giving here show some of
the constraints that these 'true probabilities' must meet.
Similar remarks apply to counterfactual analyses. One
popular kind of counterfactual analysis would have it that

!S is effective strategy for G in L iff Prob(S o- G/L) >
Probf t S' ~ G/L)23
The counterfactual and the causal law approach will agree,
only if
A: Probto ~ G/X) = Prob(G/O'.Kx)
where Kx is the maximal causal description (barring 0')
consistent with X. Assuming the arguments here are right,
condition A provides an adequacy criterion for any satis-
factory semantics of counterfactuals and probabilities.
3. HOW SOME WORLDS COULD NOT

BE HUME WORLDS24
The critic of causal laws will ask, what difference do

they make? A succinct way of putting this question is to
consider for every world its corresponding Hume world-
a world just like the first in its laws of association, its
temporal relations, and even in the sequences of events
that occur in it. How does the world that has causal laws
as well differ from the corresponding Hume world? I
have already argued that the two worlds would differ
with respect to strategies.
Here I want to urge a more minor point, but one that
might go unnoticed: not all worlds could be turned into
Hume worlds by stripping away their causal laws. Given
the earlier condition relating causal laws and laws of
association, many worlds do not have related Hume
worlds. In fact no world whose laws of association provide
any correlations could be turned into a Hume world. The
demonstration is trivial. Assume that a given world has no
causal laws for a particular kind of phenomenon E. The
earlier condition tells us to test for causes of E by looking
for factors that increase the probability of E in maximal
causally homogeneous sub-populations. But in the Hume
23See articles in Harper, op. cit.
2'I learned the term 'Hume World' from David Lewis. Apparently it
originated with Frank Jackson and other Australian philosophers.
world there are no causes, so every sub-population is
homogeneous in all causal factors, and the maximal
homogeneous population is the whole population. So
if there is any C such that Prob(E/C) > Prob(E), it will
be true that C causes E, and this world will not be a
Hume world after all.
Apparently the laws of association underdetermine
the causal laws. It is easy to construct examples in which
there are two properties, P and Q, which could be used
to partition a population. Under the partition into P
and ,P, C increases the conditional probability of E
in both sub-populations; but under the partition into
Q and, Q, Prob(E/C) = Prob(E). So relative to the assump-
tion that P causes E, but Q does not, 'C causes E' is true.
It is false relative to the assumption that Q 4 E, and
P 9+ E. This suggests that, for a given set of laws of
association, any set of causal laws will do. Once some
causal laws have been settled, others will automatically
follow, but any starting point is as good as any other.
This suggestion is mistaken. Sometimes the causal laws
are underdetermined by the laws of association, but not
always. Some laws of association are compatible with only
one set of causal laws. In general laws of association do
not entail causal laws: but in particular cases they can.
Here is an example.
Consider a world whose laws of association cover three
properties, A, B, and C; and assume that the following
are implied by the laws of association:
(I) Prob(C/A) >Prob(C)
(2}Prob(C/B & A) > Prob(C/A); Prob(C/B & ,A) >
Prob(ChA)
(3) Prob(C/B) = Prob(C)
In this world, B 4 C. The probabilities might for instance
be those given in Chart 1. From just the probabilistic
facts (1), (2), and (3), it is possible to infer that both A
and B are causally relevant to C. Assume B 4 ±C. Then
. by (1), A 4 C, since the entire population is causally
homogeneous (barring A) with respect to C and hence
counts as a test population for A's effects on C. But if
20
.=c
iA
iA
80
80
A
A 20
12 8 42 38
8 18
CHART 1
A 4 C, then by (2), B 4 ±C. Therefore B 4 ±c. But

from (3) this is not possible unless A is also relevant, either
positively or negatively, to C. In the particular example
pictured in the chart, A and B are both positively relevant
to C.
This kind of example may provide solace to the Humean.
Often Humeans reject causal laws because they have no
independent access to them. They suppose themselves able
to determine the laws of association, but they imagine that
they never have the initial causal information to begin to
apply condition C. If they are lucky, this initial knowledge
may not be necessary. Perhaps they live in a world that is
not a Hume world; it may nevertheless be a world where
causal laws can be inferred just from laws of association.
4. CONCLUSION
The quantity Prob(E/C.~), which appears in both the causal

condition of Part 1 and in the measure of effectiveness from
Part 2, is called by statisticians the partial conditional prob-
ability of E on C, holding ~ fixed; and it is used in ways
similar to the ways I have used it here. It forms the foundation
for regression analyses of causation and it is applied by both

Suppes and Salmon to treat the problem of joint effects. In
decision theory the formula SC is structurally identical to
one proposed by Brian Skyrms in his deft solution to New-
comb's paradox; and elaborated further in his book Causal
Necessity.P What is especially significant about the partial
conditional probabilities which appear here is the fact that
these hold fixed all and only causal factors.
The choice of partition, {KjJ, is the critical feature of the
measure of effectiveness proposed in S'C. This is both (a)
what makes the formula work in cases where the simple
conditional probability fails; and (b) what makes it neces-
sary to admit causal laws if you wish to sort good strategies
from bad. The way you partition is crucial. In general you
get different results from SC if you partition in different
ways. Consider two different partitions for the same space,
K1, ••. , Kn- and II, ... Is, which cross-grain each other-
the Ki are mutually disjoint and exhaustive, and so are the
1;. Then it is easy to produce a measure over the field (±C,
±C, ±Ki, ±I;) such that
n n
L Prob(C/C.Kj)Prob(Kj) L
=1= Prob(C/C.I;)Prob(I;).
;=1 ;=1
What partition is employed is thus essential to whether
a strategy appears effective or not. The right partition-the
one that judges strategies to be effective or ineffective in
accord with what is objectively true-is determined by
what the causal laws are. Partitions by other factors will
give other results; and, if you do not admit causal laws,
there is no general procedure for picking out the right factors.
The objectivity of strategies requires the objectivity of
causal laws.
25 Brian Skyrms, Causal Necessity (New Haven: Yale University Press, 1980).
Essay 2
The Truth Doesn't Explain Much
O. INTRODUCTION
Scientific theories must tell us both what is true in nature,

and how we are to explain it. I shall argue that these are
entirely different functions and should be kept distinct.
Usually the two are conflated. The second is commonly
seen as a by-product of the first. Scientific theories are
thought to explain by dint of the descriptions they give of
reality. Once the job of describing is done, science can
shut down. That is all there is to do. To describe nature-
to tell its laws, the values of its fundamental constants, its
mass distributions-is ipso facto to lay down how we are
to explain it.
This is a mistake, I shall argue; a mistake that is fostered
by the covering-law model of explanation. The covering-law
model supposes that all we need to know are the laws of
nature-and a little logic, perhaps a little probability
theory-and then we know which factors can explain which
others. For example, in the simplest deductive-nomological
version, 1 the covering-law model says that one factor explains
another just in case the occurrence of the second can be de-
duced from the occurrence of the first given the laws of nature.
But the D-N model is just an example. In the sense which
is relevant to my claims here, most models of explanation
offered recently in the philosophy of science are covering-law
models. This includes not only Hempel's own inductive statis-
tical model," but also Patrick Suppes's probabilistic model
of causation.' Wesley Salmon's statistical relevance model,"
1 See C. G. Hempel, 'Scientific Explanation', in C. G. Hempel (ed.), Aspects

of Scientific Explanation (New York: Free Press, 1965).
2 See C. G. Hempel, 'Scientific Explanation', ibid.
3 See Patrick Suppes, A Probabilistic Theory of Causality (Amsterdam:
North-Holland Publishing Co., 1970).
• See Wesley Salmon, 'Statistical Explanation', in Wesley Salmon (ed.),
THE TRUTH DOESN'T EXPLAIN MUCH 45
and even Bengt Hanson's contextualistic model. 5 All these
accounts rely on the laws of nature, and just the laws of
nature, to pick out which factors we can use in explanation.
A good deal of criticism has been aimed at Hempel's
original covering-law models. Much of the criticism objects
that these models let in too much. On Hempel's account
it seems we can explain Henry's failure to get pregnant by
his taking birth control pills, and we can explain the storm
by the falling barometer. My objection is quite the opposite.
Covering-law models let in too little. With a covering-law
model we can explain hardly anything, even the things of
which we are most proud-like the role of DNA in the
inheritance of genetic characteristics, or the formation of
rainbows when sunlight is refracted through raindrops.
We cannot explain these phenomena with a covering-law
model, I shall argue, because we do not have laws that
cover them. Covering laws are scarce.
Many phenomena which have perfectly good scientific
explanations are not covered by any laws. No true laws,
that is. They are at best covered by ceteris paribus general-
izations-generalizations that hold only under special
conditions, usually ideal conditions. The literal translation
is 'other things being equal'; but it would be more apt to
read 'ceteris paribus' as 'other things being right.'
Sometimes we act as if this does not matter. We have
in the back of our minds an 'understudy' picture of ceteris
paribus laws: ceteris paribus laws are real laws; they can
stand in when the laws we would like to see are not avail-
able and they can perform all the same functions, only
not quite so well. But this will not do. Ceteris paribus
generalizations, read literally without the 'ceteris paribus'
modifier, are false. They are not only false, but held by
us to be false; and there is no ground in the covering-law
picture for false laws to explain anything. On the other
hand, with the modifier the ceteris paribus generaliza-
tions may be true, but they cover only those few cases
where the conditions are right. For most cases, either
we have a law that purports to cover, but cannot explain
5 See Bengt Hanson, 'Explanations-Of What?' (mimeograph, Stanford
University, 1974).
46 THE TRUTH DOESN'T EXPLAIN MUCH
because it is acknowledged to be false, or we have a law

that does not cover. Either way, it is bad for the covering-
law picture.
1. CETERIS PARIBUS LAWS
When I first started talking about the scarcity of covering

laws, I tried to summarize my view by saying 'There are no
exceptionless generalizations'. Then a friend asked, 'How
about "All men are mortal"?' She was right. I had been focus-
ing too much on the equations of physics. A more plausible
claim would have been that there are no exceptionless
quantitative laws in physics. Indeed not only are there no
exceptionless laws, but in fact our best candidates are known
to fail. This is something like the Popperian thesis that
every theory is born refuted. Every theory we have pro-
posed in physics, even at the time when it was most firmly
entrenched, was known to be deficient in specific and de-
tailed ways. I think this is also true for every precise quanti-
tative law within a physics theory.
But this is not the point I had wanted to make. Some
laws are treated, at least for the time being, as if they were
exceptionless, whereas others are not, even though they
remain 'on the books'. Snell's law (about the angle of in-
cidence and the angle of refraction for a ray of light) is a
good example of this latter kind. In the optics text I use
for reference (Miles V. Klein, OptiCS),6 it first appears on
page 21, and without qualification:
Snell's Law: At an interface between dielectric media, there is (also)

a refracted ray in the second medium, lying in the plane of incidence,
making an angle ()t with the normal, and obeying Snell's law:
sin ()/sin ()t = nZ/nl
where VI and Vz are the velocities of propagation in the two media,
and nl = (CjVI), nz = (cjvz) are the indices of refraction.
6 Miles V. Klein, Optics (New York: John Wiley and Sons, 1970), p. 21,
italics added. 8 is the angle of incidence.
It is only some 500 pages later, when the law is derived

from the 'full electromagnetic theory of light', that we
learn that Snell's law as stated on page 21 is true only for
media whose optical properties are isotropic. (In anisotropic
media, 'there will generally be two transmitted waves'.)
So what is deemed true is not really Snell's law as stated
on page 21, but rather a refinement of Snell's law:
Refined Snell's Law: For any two media which are optically isotropic,
at an interface between dielectrics there is a refracted ray in the second
medium, lying in the plane of incidence, making an angle 0 t with the
normal, such that;
sin 0 /sin 0t = nZ/nl'
The Snell's law of page 21 in Klein's book is an example
of a ceteris paribus law, a law that holds only in special
circumstances-in this case when the media are both iso-
tropic. Klein's statement on page 21 is clearly not to be
taken literally. Charitably, we are inclined to put the modi-
fier 'ceteris paribus' in front to hedge it. But what does this
ceteris paribus modifier do? With an eye to statistical versions
of the covering law model (Hempel's I-S picture, or Salmon's
statistical relevance model, or Suppes's probabilistic model
of causation) we may suppose that the unrefined Snell's
law is not intended to be a universal law, as literally stated,
but rather some kind of statistical law. The obvious candidate
is a crude statistical law: for the most part, at an interface
between dielectric media there is a refracted ray ... But
this will not do. For most media are optically anisotropic,
and in an anisotropic medium there are two rays. I think
there are no more satisfactory alternatives. If ceteris paribus
laws are to be true laws, there are no statistical laws with
which they can generally be identified.
2. WHEN LAWS ARE SCARCE

Why do we keep Snell's law on the books when we both
know it to be false and have a more accurate refinement
available? There are obvious pedagogic reasons. But are
there serious scientific ones? I think there are, and these
reasons have to do with the task of explaining. Specifying

which factors are explanatorily relevant to which others is
a job done by science over and above the job of laying out
the laws of nature. Once the laws of nature are known,
we still have to decide what kinds of factors can be cited
in explanation.
One thing that ceteris paribus laws do is to express our
explanatory commitments. They tell what kinds of explana-
tions are permitted. We know from the refined Snell's law
that in any isotropic medium, the angle of refraction can
be explained by the angle of incidence, according to the
equation sin O/sin 0t = n2/nl' To leave the unrefined Snell's
law on the books is to signal that the same kind of explana-
tion can be given even for some anisotropic media. The
pattern of explanation derived from the ideal situation
is employed even where the conditions are less than ideal;
and we assume that we can understand what happens in
nearly isotropic media by rehearsing how light rays behave
in pure isotropic cases.
This assumption is a delicate one. It fits far better with the
simulacrum account of explanation that I will urge in Essay 8
than it does with any covering-law model. For the moment I
intend only to point out that it is an assumption, and an
assumption which (prior to the 'full electromagnetic theory')
goes well beyond our knowledge of the facts of nature. We
know that in isotropic media, the angle of refraction is due
to the angle of incidence under the equation sinO/sin 0t =
n2/nl' We decide to explain the angles for the two refracted
rays in anisotropic media in the same manner. We may have
good reasons for the decision; in this case if the media are
nearly isotropic, the two rays will be very close together, and
close to the angle predicted by Snell's law; or we believe in
continuity of physical processes . But still this decision is not
"forced by our knowledge of the laws of nature.
Obviously this decision could not be taken if we also
had on the books a second refinement of Snell's law, im-
plying that in any anisotropic media the angles are quite
different from those given by Snell's law. But laws are
scarce, and often we have no law at all about what happens
in conditions that are less than ideal.
Covering-law theorists will tell a different story about

the use of ceteris paribus laws in explanation. From their
point of view, ceteris paribus explanations are elliptical for
genuine covering law explanations from true laws which we
do not yet know. When we use a ceteris paribus 'law' which
we know to be false, the covering-law theorist supposes us
to be making a bet about what form the true law takes. For
example, to retain Snell's unqualified law would be to bet
that the (at the time unknown) law for anisotropic media
will entail values 'close enough' to those derived from the
original Snell law.
I have two difficulties with this story. The first arises
from an extreme metaphysical possibility, in which I in fact
believe. Covering-law theorists tend to think that nature
is well-regulated; in the extreme, that there is a law to cover
every case. I do not. I imagine that natural objects are much
like people in societies. Their behaviour is constrained by
some specific laws and by a handful of general principles,
but it is not determined in detail, even statistically. What
happens on most occasions is dictated by no law at all.
This is not a metaphysical picture that I urge. My claim is
that this picture is as plausible as the alternative. God may
have written just a few laws and grown tired. We do not
know whether we are in a tidy universe or an untidy one.
Whichever universe we are in, the ordinary commonplace
activity of giving explanations ought to make sense.
The second difficulty for the ellipsis version of the covering-
law account is more pedestrian. Elliptical explanations are
not explanations: they are at best assurances that explana-
tions are to be had. The law that is supposed to appear
in the complete, correct D-N explanation is not a law we
have in our theory, not a law that we can state, let alone test.
There may be covering-law explanations in these cases.
But those explanations are not our explanations; and those
unknown laws cannot be our grounds for saying of a nearly
isotropic medium, 'sin 0t ::::::
k(n2/nt) because sin 0 = k'.
What then are our grounds? I assert only what they are
not: they are not the laws of nature. The laws of nature
that we know at any time are not enough to tell us what
kinds of explanations can be given at that time. That requires
a decision; and it is just this decision that covering-law

theorists make when they wager about the existence of
unknown laws. We may believe in these unknown laws, but
we do so on no ordinary grounds: they have not been tested,
nor are they derived from a higher level theory. Our grounds
for believing in them are only as good as our reasons for
adopting the corresponding explanatory strategy, and no
better.
3. WHEN LAWS CONFLICT
I have been maintaining that there are not enough covering

laws to go around. Why? The view depends on the picture
of science that I mentioned earlier. Science is broken into
various distinct domains: hydrodynamics, genetics, laser
theory, ... We have many detailed and sophisticated theories
about what happens within the various domains. But we
have little theory about what happens in the intersection
of domains.
Diagramatically, we have laws like
ceteris paribus, (x) (S(x) 41(x))
and
ceteris paribus, (x) (A(x) 4,I(x)).
For example, (ceteris paribus) adding salt to water decreases
the cooking time of potatoes; taking the water to higher
altitudes increases it. Refining, if we speak more carefully
we might say instead, 'Adding salt to water while keeping
the altitude constant decreases the cooking time; whereas
increasing the altitude while keeping the saline content
fixed increases it'; or
(x) (S(x) & ,A(x) 41(x))
and
(x) (A (x) & ,Sex) 4 ,lex)).
But neither of these tells what happens when we both add

salt to the water and move to higher altitudes.
Here we think that probably there is a precise answer
about what would happen, even though it is not part of

our common folk wisdom. But this is not always the case.
I discuss this in detail in the next essay. Most real life cases
involve some combination of causes; and general laws that
describe what happens in these complex cases are not always
available. Although both quantum theory and relativity
are highly developed, detailed, and sophisticated, there is
no satisfactory theory of relativistic quantum mechanics.
A more detailed example from transport theory is given in
the next essay. The general lesson is this: where theories
intersect, laws are usually hard to come by.
4. WHEN EX PLANA nONS CAN BE GIVEN ANYWAY

SO far, I have only argued half the case. I have argued that
covering laws are scarce, and that ceteris paribus laws are
no true laws. It remains to argue that, nevertheless, ceteris
paribus laws have a fundamental explanatory role. But
this is easy, for most of our explanations are explanations
from ceteris paribus laws.
Let me illustrate with a humdrum example. Last year
I planted camellias in my garden. I know that camellias like
rich soil, so I planted them in composted manure. On the
other hand, the manure was still warm, and I also know
that camellia roots cannot take high temperatures. So I did
not know what to expect. But when many of my camellias
died, despite otherwise perfect care, I knew what went
wrong. The camellias died because they were planted in
hot soil.
This is surely the right explanation to give. Of course,
I cannot be absolutely certain that this explanation is the
correct one. Some other factor may have been responsible,
nitrogen deficiency or some genetic defect in the plants,
a factor that I did not notice, or may not even have known
to be relevant. But this uncertainty is not peculiar to cases
of explanation. It is just the uncertainty that besets all of
our judgements about matters of fact. We must allow for
oversight ; still, since I made a reasonable effort to eliminate
other menaces to my camellias, we may have some con-
fidence that this is the right explanation.
SO we have an explanation for the death of my camellias.

But it is not an explanation from any true covering law.
There is no law that says that camellias just like mine, planted
in soil which is both hot and rich, die. To the contrary, they
do not all die. Some thrive; and probably those that do, do
so because of the richness of the soil they are planted in.
We may insist that there must be some differentiating factor
which brings the case under a covering law: in soil which
is rich and hot, camellias of one kind die; those of another
thrive. I will not deny that there may be such a covering law.
I merely repeat that our ability to give this humdrum explana-
tion precedes our knowledge of that law. On the Day of
Judgment, when all laws are known, these may suffice to
explain all phenomena. But in the meantime we do give
explanations; and it is the job of science to tell us what
kinds of explanations are admissible.
In fact I want to urge a stronger thesis. If, as is possible,
the world is not a tidy deterministic system, this job of
telling how we are to explain will be a job which is still
left when the descriptive task of science is complete. Imagine
for example (what I suppose actually to be the case) that
the facts about camellias are irreducibly statistical. Then it
is possible to know all the general nomological facts about
camellias which there are to know-for example, that 62 per
cent of all camellias in just the circumstances of my camellias
die, and 38 per cent survive." But one would not thereby
know how to explain what happened in my garden. You
would still have to look to the Sunset Garden Book to learn
that the heat of the soil explains the perishing, and the
richness explains the plants that thrive.
5. CONCLUSION
Most scientific explanations use ceteris paribus laws. These

laws, read literally as descriptive statements, are false, not
only false but deemed false even in the context of use. This
7 Various writers, especially Suppes (footnote 3) and Salmon (footnote 4),

have urged that knowledge of more sophisticated statistical facts will suffice
to determine what factors can be used in explanation. I do not believe that this
claim can be carried out, as I have argued in Essay 1.
is no surprise: we want laws that unify; but what happens

may well be varied and diverse. We are lucky that we can
organize phenomena at all. There is no reason to think
that the principles that best organize will be true, nor that
the principles that are true will organize much.
Essay 3
Do the Laws of Physics State the Facts?
O. INTRODUCTION
There is a view about laws of nature that is so deeply en-
trenched that it does not even have a name of its own. It
is the view that laws of nature describe facts about reality.
If we think that the facts described by a law obtain, or
at least that the facts that obtain are sufficiently like those
described in the law, we count the law true, or true-for-the-
nonce, until further facts are discovered. I propose to call
this doctrine the facticity view of laws. (The name is due to
John Perry.)
It is customary to take the fundamental explanatory laws
of physics as the ideal. Maxwell's equations, or Schroedinger's,
or the equations of general relativity, are paradigms, para-
digms upon which all other laws-laws of chemistry, bi-
ology, thermodynamics, or particle physics-are to be
modelled. But this assumption confutes the facticity view
of laws. For the fundamental laws of physics do not describe
true facts about reality. Rendered as descriptions of facts,
they are false; amended to be true, they lose their funda-
mental, explanatory force.
To understand this claim, it will help to contrast biology
with physics. J. J. C. Smart argues that biology has no
genuine laws of its own.! It resembles engineering. Any
general claim about a complex system, such as a radio or
a living organism, will be likely to have exceptions. The
generalizations of biology, or engineering's rules of thumb,
are not true laws because they are not exceptionless. Many
(though not Smart himself) take this to mean that biology
is a second-rate science. If this is good reasoning, it must be
physics that is the second-rate science. Not only do the
laws of physics have exceptions; unlike biological laws, they
are not even true for the most part, or approximately true.
1 See J. J. C. Smart, Philosophy and Scientific Realism (London: Routledge
and Keegan Paul, 1963).
DO THE LAWS OF PHYSICS STATE THE FACTS? 55
The view of laws with which I begin-'Laws of nature

describe facts about reality'-is a pedestrian view that, I
imagine, any scientific realist will hold. It supposes that laws
of nature tell how objects of various kinds behave: how
they behave some of the time, or all of the time, or even
(if we want to prefix a necessity operator) how they must
behave. What is critical is that they talk about objects-
real concrete things that exist here in our material world,
things like quarks, or mice, or genes; and they tell us what
these objects do.
Biological laws provide good examples. For instance, here
is a generalization taken from a Stanford text on chordates:
The gymnotoids [American knife fish] are slender fish with enormously
long anal fins, which suggest the blade of a knife of which the head
is a handle. They often swim slowly with the body straight by un-
dulating this fin. They [presumably 'always' or 'for the most part']
are found in Central and South America ... Unlike the characins
they ['usually'?] hide by day under river banks or among roots, or
even bury themselves in sand, emerging only at night."
The fundamental laws of physics, by contrast, do not tell

what the objects in their domain do. If we try to think of
them in this way, they are simply false, not only false but
deemed false by the very theory that maintains them. But
if physics' .basic, explanatory laws do not describe how
things behave, what do they do? Once we have given up
facticity, I do not know what to say. Richard Feynman, in
The Character of Physical Law, offers an idea, a metaphor.
Feynman tells us 'There is ... a rhythm and a pattern be-
tween the phenomena of nature which is not apparent to
the eye, but only to the eye of analysis; and it is these
rhythms and patterns which we call Physical Laws . . .'3
Most philosophers will want to know a lot more about how
these rhythms and patterns function. But at least Feynman
does not claim that the laws he studies describe the facts.
I say that the laws of physics do not provide true descriptions
2 R. McNeill Alexander, The Chordates (Cambridge: Cambridge University
Press, 1975), p. 179.
3 Richard Feynman, The Character of Physical Law (Cambridge, Mass: MIT
Press, 1967), p. 13.
56 DO THE LAWS OF PHYSICS STATE THE FACTS?
of reality. This sounds like an anti-realist doctrine. Indeed it

is, but to describe the claim in this way may be misleading.
For anti-realist views in the philosophy of science are tradi-
tionally of two kinds. Bas van Fraassen" is a modern advocate
of one of these versions of anti-realism; Hilary Putnam" of
the other. Van Fraassen is a sophisticated instrumentalist.
He worries about the existence of unobservable entities,
or rather, about the soundness of our grounds for believing
in them; and he worries about the evidence which is supposed
to support our theoretical claims about how these entities
behave. But I have no quarrel with theoretical entities; and
for the moment I am not concerned with how we know
what they do. What is troubling me here is that our explana-
tory laws do not tell us what they do. It is in fact part of
their explanatory role not to tell.
Hilary Putnam in his version of internal realism also
maintains that the laws of physics do not represent facts
about reality. But this is because nothing-not even the
most commonplace claim about the cookies which are
burning in the oven-represents facts about· reality. If
anything did, Putnam would probably think that the basic
equations of modern physics did best. This is the claim that
I reject. I think we can allow that all sorts of statements
represent facts of nature, including the generalizations one
learns in biology or engineering. It is just the fundamental
explanatory laws that do not truly represent. Putnam is
worried about meaning and reference and how we are trapped
in the circle of words. I am worried about truth and explana-
tion, and how one excludes the other.
1. EXPLANATION BY COMPOSITION OF CAUSES,

AND THE TRADE-OFF OF TRUTH AND
EXPLANATORY POWER
Let me begin with a law of physics everyone knows-the

law of universal gravitation. This is the law that Feynman
4See Bas van Fraassen, The Scientific Image (Oxford: Clarendon Press, 1980).
5See Hilary Putnam, Meaning and the Moral Sciences (London: Routledge
and Kegan Paul, 1978) and 'Models and Reality', Journal of Symbolic Logic,
forthcoming.
uses for illustration; he endorses the view that this law is

'the greatest generalization achieved by the human mind'. 6
Law of Gravitation: F = Gmm'[r',

In words, Feynman tells us:
The Law of Gravitation is that two bodies exert a force between each
other which varies inversely as the square of the distance between
them, and varies directly as the product of their masses.'
Does this law truly describe how bodies behave?

Assuredly not. Feynman himself gives one reason why.
'Electricity also exerts forces inversely as the square of the
distance, this time between charges .. .'8 It is not true that
for any two bodies the force between them is given by the
law of gravitation. Some bodies are charged bodies, and
the force between them is not Gmm'lr", Rather it is some
resultant of this force with the electric force to which
Feynman refers.
For bodies which are both massive and charged, the law of
universal gravitation and Coulomb's law (the law that gives
the force between two charges) interact to determine the
final force. But neither law by itself truly describes how the
bodies behave. No charged objects will behave just as the
law of universal gravitation says; and any massive objects
will constitute a counterexample to Coulomb's law. These
two laws are not true; worse, they are not even approxi-
mately true. In the interaction between the electrons and the
protons of an atom, for example, the Coulomb effect swamps
the gravitational one, and the force that actually occurs
is very different from that described by the law of gravity.
There is an obvious rejoinder: I have not given a complete
statement of these two laws, only a shorthand version. The
Feynman version has an implicit ceteris paribus modifier
in front, which I have suppressed. Speaking more carefully,
the law of universal gravitational is something like this:
6 Feynman, op, cit., p. 14.

7 Ibid., p. 14.
8 Ibid., p. 30.
If there are no forces other than gravitational forces at work, then

two bodies exert a force between each other which varies inversely
as the square of the distance between them, and varies directly as the
product of their masses.
I will allow that this law is a true law, or at least one that is
held true within a given theory. But it is not a very useful
law. One of the chief jobs of the law of gravity is to help
explain the forces that objects experience in various complex
circumstances. This law can explain in only very simple, or
ideal, circumstances. It can account for why the force is as
it is when just gravity is at work; but it is of no help for
cases in which both gravity and electricity matter. Once the
ceteris paribus modifier has been attached, the law of gravity
is irrelevant to the more complex and interesting situations.
This unhappy feature is characteristic of explanatory laws.
I said that the fundamental laws of physics do not represent
the facts, whereas biological laws and principles of engineer-
ing do. This statement is both too strong and too weak.
Some laws of physics do represent facts, and some laws of
biology-particularly the explanatory laws-do not. The
failure of facticity does not have so much to do with the
nature of physics, but rather with the nature of explanation.
We think that nature is governed by a small number of simple,
fundamental laws. The world is full of complex and varied
phenomena, but these are not fundamental. They arise from
the interplay of more simple processes obeying the basic
laws of nature. (Later essays will argue that even simple
isolated processes do not in general behave in the uniform
manner dictated by fundamental laws.)
This picture of how nature operates to produce the subtle
and complicated effects we see around us is reflected in the
explanations that we give: we explain complex phenomena
by reducing them to their more simple components. This is
not the only kind of explanation we give, but it is an import-
ant and central kind. I shall use the language of John Stuart
Mill, and call this explanation by composition of causes. 9
It is characteristic of explanations by composition of
9 John Stuart Mill, A System of Logic (New York: Harper and Brothers,
1893). See Book lll, Chapter VI.
causes that the laws they employ fail to satisfy the require-
ment of facticity. The force of these explanations comes
from the presumption that the explanatory laws 'act' in
combination just as they would 'act' separately. It is critical,
then, that the laws cited have the same form, in or out of
combination. But this is impossible if the laws are to describe
the actual behaviour of objects. The actual behaviour is the
resultant of simple laws in combination. The effect that
occurs is not an effect dictated by anyone of the laws
separately. In order to be true in the composite case, the
law must describe one effect (the effect that actually happens);
but to be explanatory, it must describe another. There is
a trade-off here between truth and explanatory power.
2. HOW VECTOR ADDITION INTRODUCES

CAUSAL POWERS
Our example, where gravity and electricity mix, is an example

of the composition of forces. We know that forces add
vectorially. Does vector addition not provide a simple and
obvious answer to my worries? When gravity and electricity
are both at work, two forces are produced,' one in accord
with Coulomb's law, the other according to the law of
universal gravitation. Each law is accurate. Both the gravita-
tional and the electric force are produced as described;
the two forces then add together vectorially to yield the
total 'resultant' force.
The vector addition story is, I admit, a nice one. But it
is just a metaphor. We add forces (or the numbers that
represent forces) when we do calculations. Nature does not
'add' forces. For the 'component' forces are not there,
in any but a metaphorical sense, to be added; and the laws
that say they are there must also be given a metaphorical
reading. Let me explain in more detail.
The vector addition story supposes that Feynman has
left something out in his version of the law of gravitation.
In the way that he writes it, it sounds as if the law describes
the resultant force exerted between two bodies, rather
than a component force-the force which is produced
between the two bodies in virtue of their 'gravitational
masses (or, for short, the force due to gravity). A better

way to state the law would be
Two bodies produce a force between each other (the force due to
gravity) which varies inversely as the square of the distance between
them, and varies directly as the product of their masses.
Similarly, for Coulomb's law

Two charged bodies produce a force between each other (the force due
to electricity) which also varies inversely as the square of the distance
between them, and varies directly as the product of their masses.
These laws, I claim, do not satisfy the facticity require-

ment. They -appear, on the face of it, to describe what
bodies do: in the one case, the two bodies produce a force
of size Gmm' /r2; in the other, they produce a force of size
qq'[r". But this cannot literally be so. For the force of size
Gmm'[r? and the force of size qq'[r', are not real, occurrent
forces. In interaction a single force occurs-the force we call
the 'resultant'-and this force is neither the force due to
gravity nor the electric force. On the vector addition story,
the gravitational and the electric force are both produced, yet
neither exists.
Mill would deny this. He thinks that in cases of the com-
position of causes, each separate effect does exist-it exists
as part of the resultant effect, just as the left half of a table
exists as part of the whole table. Mill's paradigm for com-
position of causes is mechanics. He says:
In this important class of cases of causation, one cause never, properly
speaking, defeats or frustrates another; both have their full effect. If
a body is propelled in two directions by two forces, one tending to
drive it to the north, and the other to the east, it is caused to move in
a given time exactly as far in both directions as the two forces would
separately have carried it ... 10
Mill's claim is unlikely. Events may have temporal parts,

but not parts of the kind Mill describes. When a body has
moved along a path due north-east, it has travelled neither
due north nor due east. The first half of the motion can be
a part of the total motion; but no pure north motion can be
10 Ibid., Bk. III, Cli. VL
a part of a motion that always heads northeast. (We learn

this from Judith Jarvis Thomson's Acts and Other Events.)
The lesson is even clearer if the example is changed a little:
a body is pulled equally in opposite directions. It does not
budge, but in Mill's picture it has been caused to move
both several feet to the left and several feet to the right.
I realize, however, that intutions are strongly divided on
these cases; so in the next section I will present an example
for which there is no possibility of seeing the separate effects
of the composed causes as part of the effect which actually
occurs.
It is implausible to take the force due to gravity and the
force due to electricity literally as parts of the actually
occurring force. Is there no way to make sense of the story
about vector addition? I think there is, but it involves giving
up the facticity view of laws. We can preserve the truth of
Coulomb's law and the law of gravitation by making them
about something other than the facts: the laws can describe
the causal powers that' bodies have.
Hume taught that 'the distinction, which we often make
betwixt power and the exercise of it, is ... without founda-
tion';'! It is just Hume's illicit distinction that we need here:
the law of gravitation claims that two bodies have the power
to produce a force of size Gmm' /r2• But they do not always
succeed in the exercise of it. What they actually produce
depends on what other powers are at work, and on what
compromise is fmally achieved among them. This may be
the way we do sometimes imagine the composition of causes.
But if so, the laws we use talk not about what bodies do,
but about the powers they possess.
The' introduction of causal powers will not be seen as a
very productive starting point in our current era of moderate
empiricism. Without doubt, we do sometimes think in terms
of causal powers, so it would be foolish to maintain that the
facticity view must be correct and the use of causal powers
a total mistake .. But facticity cannot be given up easily. We
need an account of what laws are, an account that connects
them, on the one hand, with standard scientific methods
11 David Hume, A Treatise of Human Nature, ed, L. A. Selby Bigge (Oxford:
Clarendon Press, 1978), p. 311.
for confirming laws, and on the other, with the use they are
put to for prediction, construction, and explanation. If laws
of nature are presumed to describe the facts, then there
are familiar, detailed philosophic stories to be told about
why a sample of facts is relevant to their confirmation,
and how they help provide knowledge and understanding
of what happens in nature. Any alternative account of what
laws of nature do and what they say must serve at least as
well; and no story I know about causal powers makes a
very good start.
3. THE FORCE DUE TO GRAVITY
It is worth considering further the force due to gravity and

the force due to electricity, since this solution is frequently
urged by defenders of facticity. It is one of a class of sugges-
tions that tries to keep the separate causal laws in something
much like their original form, and simultaneously to defend
their facticity by postulating some intermediate effect which
they produce, such as a force due to gravity, a gravitational
potential, or a field.
Lewis Creary has given the most detailed proposal of this
sort that I know. Creary claims that there are two quite
different kinds of laws that are employed in explanations
where causes compose-laws of causal influence and laws
of causal action. Laws of causal influence, such as Coulomb's
law and the law of gravity, 'tell us what forces or other
causal influences operate in various circumstances', whereas
laws of causal action 'tell us what the results are of such
causal influences, acting either singly or in various com bina-
tions'.12 In the case of composition of forces, the law of
interaction is a vector addition law, and vector addition
laws 'permit explanations of an especially satisfying sort'
because the analysis 'not only identifies the different com-
ponent causal influences at work, but also quantifies their
relative importance'i P Creary also describes less satisfying
kinds of composition, including reinforcement, interference,
12 Lewis Creary, 'Causal Explanation anp the Reality of Natural Component
Forces,' Pacific Philosophical Quarterly 62(1981), p. 153.
13 Ibid., p, 153.
and predomination. On Creary's account, Coulomb's law and
the law of gravity come out true because they correctly
describe what influences are produced-here, the force due
to gravity and the force due to electricity. The vector addi-
tion law then combines the separate influences to predict
what motions will occur.
This seems to me to be a plausible account of how a lot
of causal explanation is structured. But as a defence of the
truth of fundamental laws, it has two important drawbacks.
First, in many cases there are no general laws of interaction.
Dynamics, with its vector addition law, is quite special in
this respect. This is not to say that there are no truths about
how this specific kind of cause combines with that, but
rather that theories can seldom specify a procedure that
works from one case to another. Without that, the collection
of fundamental laws loses the generality of application
which Creary's proposal hoped to secure. The classical study
of irreversible processes provides a good example of a highly
successful theory that has this failing. Flow processes like
diffusion, heat transfer, or electric current ought to be
studied by the transport equations of statistical mechanics.
But usually, the model for the distribution functions and
the details of the transport equations are too complex:
the method is unusable. A colleague of mine in engineering
estimates that 90 per cent of all engineering systems cannot
be treated by the currently available methods of statistical
mechanics. 'We analyze them by whatever means seem
appropriate for the problem at hand,' he adds.!"
In practice engineers handle irreversible processes with
old fashioned phenomenological laws describing the flow
(or flux) of the quantity under study. Most of thse laws
have been known for quite a long time. For example there
is Fick's law, dating from 1855, which relates the diffusion
velocity of a component in a mixture to the gradient of its
density (1m = - Dac/ax). Equally simple laws describe
other processes: Fourier's law for heat flow, Newton's law
for sheering force (momentum flux) and Ohm's law for
electric current. Each of these is a linear differential equation
.4 Kline's studies of methods of approximation appear in S. J. Kline, Simili-
tude and Approximation Theory (New York: McGraw-Hill, 1969), p. 140.
in t (e.g. the Jm in Fick's law cited above is dm/dt), giving

the time rate of change of the desired quantity (in the case
of Fick's law, the mass). Hence a solution at one time com-
pletely determines the quantity at any other time. Given
that the quantity can be controlled at some point in a process,
these equations should be perfect for determining the future
evolution of the process. They are not.
The trouble is that each equation is a ceteris paribus law.
It describes the flux only so long as just one kind of cause
is operating. More realistic examples set different forces at
play simultaneously. In a mixture of liquids, for example,
if both the temperatures and the concentrations are non-
uniform, there may be a flow of liquid due not only to the
concentration gradients but also to the temperature gradients.
This is called the Soret effect.
The situation is this. For the several fluxes J we have laws
of the form,
Jq = fn(cxq)
Each of these is appropriate only when its cx is the only
relevant variable. For cross-effects we require laws of
the form.
Jq = gn(cxm, ... ,cxq).

This case is structurally just like the simple causal examples
that I discussed in the last essay. We would like to have laws
that combine different processes. But we have such laws only
in a few special cases, like the Soret effect. For the Soret
effect we assume simple linear additivity in our law of action,
and obtain a final cross-effect law by adding a thermal diffu-
sion factor into Fick's law. But this law of causal action
is highly specific to the situation and will not work for com-
bining arbitrary influences studied by transport theory.
Are there any principles to be followed in modifying to

allow for cross-effects? There is only one systematic account
of cross-effect modification for flow processes. It originated
with Onsager in 1931, but was not developed until the
1950s. Onsager theory defines force-flux pairs, and prescribes
a method for writing cross-effect equations involving different
forces. As C. A. Truesdell describes it, 'Onsagerism claims
to unify and correlate much existing knowledge of irrevers-
ible processes'i!" Unfortunately it does not succeed. Truesdell
continues:
As far as concerns heat conduction, viscosity, and diffusion ... this is

not so. Not only does Onsagerism not apply to any of these phenomena
without a Procrustean force-fit, but even in the generous interpretation
of its sectaries it does not yield as much reduction for the theory of
viscosity as was known a century earlier and follows from fundamental
principles .... 16
Truesdell claims that the principles used in Onsager theory are

vacuous. The principles must sometimes be applied in one
way, sometimes in another, in an ad hoc fashion demanded
by each new situation. The prescription for constructing
laws, for example, depends on the proper choice of con-
jugate flux-force pairs. Onsager theory offers a general
principle for making this choice, but if the principle were
followed literally, we would not make the proper choice in
even the most simple situations. In practice on any given
occasion the choice is left to the physicist's imagination.
It seems that after its first glimmer of generality the Onsager
approach turns out to be a collection of ad hoc techniques.
I have illustrated with statistical mechanics; but this is
not a special case. In fact classical mechanics may well be
the only discipline where a general law of action is always
available. This limits the usefulness of Creary's idea. Creary's
scheme, if it works, buys facticity, but it is of little benefit
to realists who believe that the phenomena of nature flow
from a small number of abstract, fundamental laws. The
fundamental laws will be severely limited in scope. Where
15 C. A. Truesdell, Rational Thermodynamics (New York: McGraw-Hili,
1969), p. 140.
16 Ibid., p. 140.
the laws of action go case by case and do not fit a general

scheme, basic laws of influence, like Coulomb's law and the
law of gravity, may give true accounts of the influences that
are produced; but the work of describing what the influences
do, and what behaviour results, will be done by the variety
of complex and ill-organized laws of action: Fick's law with
correction factors, and the like. This fits better with my
picture of a nature best described by a vast array of phe-
nomenological laws tailored to specific situations, than
with one governed in an orderly way from first principles.
The causal influences themselves are the second big draw-
back to Creary's approach. Consider our original example.
Creary changes it somewhat from the way I originally set
it up. I had presumed that the aim was to explain the size
and direction of a resultant force. Creary supposes that it
is not a resultant force but a consequent motion which is
to be explained. This allows him to deny the reality of the
resultant force. We are both agreed that there cannot be
three forces present-two components and a resultant.
But I had assumed the resultant, whereas Creary urges the
existence of the components.
The shift in the example is necessary for Creary. His
scheme works by interposing an intermediate factor-the
causal influence-between the cause and what initially
looked to be the effect. In the dynamic example the re-
structuring is plausible. Creary may well be right about
resultant and component forces, But I do not think this
will work as a general strategy, for it proliferates influences
in every case. Take any arbitrary example of the composition
of causes: two laws, where each accurately dictates what
will happen when it operates in isolation, say 'C causes E'
and C' causes E"; but where C and C' in combination pro-
duce some different effect, E". If we do not want to assume
that all three effects-E, E', E"-occur (as we would if we
thought that E and E' were parts of E"), then on Creary's
proposal we must postulate some further occurrences,
F and F', as the proper effects of our two laws, effects
that get combined by a law of action to yield E" at the end.
In some concrete cases the strategy will work, but in general
I see no reason to think that these intermediate influences
can always be found. I am not opposed to them because of

any general objection to theoretical entities, but rather
because I think every new theoretical entity which is ad-
mitted should be grounded in experimentation, which
shows up its causal structure in detail. Creary's influences
seem to me just to be shadow occurrences which stand
in for the effects we would like to see but cannot in
fact find,
4. A REAL EXAMPLE OF THE COMPOSITION

OF CAUSES
The ground state of the carbon atom has five distinct energy
levels (see Figure 3.1). Physics texts commonly treat this
phenomenon sequentially, in three stages. I shall follow the
discussion of Albert Messiah in Volume II of Quantum
MechanicsP In the first stage, the ground state energy is
calculated by a central field approximation; and the single
line (a) is derived. For some purposes, it is accurate to
assume that only this level occurs. But some problems
's 'so
'D2
1s22s22p2
3P2
3p,
3
po
(a) (b) (e)
FIG. 3.1. The levels of the ground state of the carbon atom; (a) in the
central field approximation (Vi = V 2 = 0); (b) neglecting spin-orbit
coupling (V2 == 0); (c) including spin-orbit' coupling. (Source: Messiah,
Quantum Mechanics.)
11 Albert Messiah, Quantum Mechanics (Amsterdam: North-Holland, 1961),

p.703.
require a more accurate description. This can be provided by

noticing that the central field approximation takes account
only of the average value of the electrostatic repulsion of the
inner shell electrons on the two outer electrons. This defect
is remedied at the second stage by considering the effects
of a term which is equal to the difference between the
exact Coulomb interaction and the average potential used
in stage one. This corrective potential 'splits' the single
line (a) into three lines depicted in (b).
But the treatment is inaccurate because it neglects spin
effects. Each electron has a spin, or internal angular mo-
mentum, and the spin of the electron couples with its orbital
angular momentum to create an additional potential. The
additional potential arises because the spinning electron
has an intrinsic magnetic moment, and 'an electron moving
in [an electrostatic] potential "sees" a magnetic field',l8
About the results of this potential Messiah tells us, 'Only
the 3P state is affected by the spin-orbit energy term; it
gets split into three levels: 3PO, 3Pl and 3P2',l9 Hence the
five levels pictured in (c).
The philosophic perplexities stand out most at the last
stage. The five levels are due to a combination of a Coulomb
potential, and a potential created by spin-orbit coupling
'splits' the lowest of these again into three. That is the
explanation of the five levels. But how can we state the
laws that it uses?
For the Coulomb effect we might try
Whenever a Coulomb potential is like that in the carbon atom, the
three energy levels pictured in (b) occur.
(The real law will of course replace 'like that in the carbon
atom' by a mathematical description of the Coulomb po-
tential in carbon; and similarly for 'the three energy levels
pictured in (b )'.) The carbon atom itself provides a counter-
example to this law. It has a Coulomb potential of the right
kind; yet the five levels of (c) occur, not the three levels
of (b).
18 Ibid" p. 552.
19 Ibid" p. 552.
We might, in analogy with the vector addition treatment

of composite forces, try instead
The energy levels produced by a Coulomb potential like that in the
carbon atom are the three levels pictured in (b).
But (as with the forces 'produced by gravity' in our earlier

example) the levels that are supposed to be produced by the
Coulomb potential are levels that do not occur. In actuality
five levels occur, and they do not include the three levels
of (b). In particular, as we can see from Messiah's diagram,
the lowest of the three levels-the 3P-is not identical
with any of the five. In the case of the composition of
motions, Mill tried to see the 'component' effects as parts
of the actual effect. But that certainly will not work here.
The 3p level in (b) may be 'split' and hence 'give rise to',
the 3PO, 3Pt> and 3P2 levels in (c); but it is certainly not a
part of any of these levels.
It is hard to state a true factual claim about the effects
of the Coulomb potential in the carbon atom. But quantum
theory does guarantee that a certain counter/actual is true;
the Coulomb potential, if it were the only potential at work,
would produce the three levels in (b). Clearly this counter-
factual bears on our explanation. But we have no model of
explanation that shows how. The covering-law model shows
how statements of fact are relevant to explaining a phe-
nomenon. But how is a truth about energy levels, which
would occur in quite different circumstances, relevant to
the levels which do occur in these? We think the counter-
factual is important; but we have no account of how it
works.
5. COMPOSITION OF CAUSES VERSUS

EXPLANA TION BY COVERING LA W
The composition of causes is not the only method of ex-

planation which can be employed. There are other methods,
and some of these are compatible with the facticity view
of laws. Standard covering-law explanations are a prime
example.
Sometimes these other kinds of explanation are available
even when we give an explanation which tells what the

component causes of a phenomenon are. For example,
in the case of Coulomb's law and the law of gravity, we
know how to write down a more complex law (a law with
a more complex antecedent) which says exactly what hap-
pens when a system has both mass and charge. Mill thinks
that such 'super' laws are always available for mechanical
phenomena. In fact he thinks, 'This explains why mechanics
is a deductive or demonstrative science, and chemistry
is not'.20
I want to make three remarks about these super laws
and the covering explanations they provide. The first is
familiar from the last essay: super laws are not always avail-
able. Secondly, even when they are available, they often do
not explain much. Thirdly, and most importantly, even when
other good explanations are to hand, if we fail to describe
the component processes that go together to produce a phe-
nomenon, we lose a central and important part of our under-
standing of what makes things happen.
(I) There are a good number of complex scientific phenomena

which we are quite proud to be able to explain. As I
urged in the last essay, for many of these explanations,
super covering laws are not available to us. Sometimes
we have every reason to believe that a super law exists.
In other cases we have no good empirical reason to sup-
pose even this much. Nevertheless, after we have seen
what occurs in a specific case, we are often able to
understand how various causes contributed to bring it
about. We do explain, even without knowing the super
laws. We need a philosophical account of explanations
which covers this very common scientific practice, and
which shows why these explanations are good ones.
(2) Sometimes super laws, even when they are available to
cover a case, may not be very explanatory. This is an
old complaint against the covering-law model of explana-
tion: 'Why does the quail in the garden -bob its head up
and down in that funny way whenever it walks?' ...
'Because they all do.' In the example of spin-orbit coupling
20 Mill, op, cit., p, 267.
it does not explain the five energy levels that appear in

a particular experiment to say 'All carbon atoms have
five energy levels'.
(3) Often, of course, a covering law for the complex case
will be explanatory. This is especially true when the ante-
cedent of the law does not just piece together the par-
ticular circumstances that obtain on the occasion in
question, but instead gives a more abstract description
which fits with a general body of theory. In the case of
spin-orbit coupling, Stephen Norman remarks that
quantum mechanics provides general theorems about
symmetry groups, and Hamiltonians, and degeneracies,
from which we could expect to derive, covering-law
style, the energy levels of carbon from the appropriate
abstract characterization of its Hamiltonian, and the
symmetries it exhibits.
Indeed we can do this; and if we do not do it, we will
fail to see that the pattern of levels in carbon is a particular
case of a general phenomenon which reflects a deep fact
about the effects of symmetries in nature. On the other
hand, to do only this misses the detailed causal story of
how the splitting of spectral lines by the removal of sym-
metry manages to get worked out in each particular case.
This two-faced character is a widespread feature of ex-
planation. Even if there is a single set of super laws which
unifies all the complex phenomena one studies in physics,
our current picture may yet provide the ground for these
laws: what the unified laws dictate should happen, happens
because of the combined action of laws from separate do-
mains, like the law of gravity and Coulomb's law. Without
these laws, we would miss an essential portion of the ex-
planatory story. Explanation by subsumption under super,
unified covering laws would be no replacement for the
composition of causes. It would be a complement. To
understand how the consequences of the unified laws are
brought about would require separate operation of the
law of gravity, Coulomb's law, and so forth; and the
failure of facticity for these contributory laws would
still have to be faced.
6. CONCLUSION
There is a simple, straightforward view of laws of nature

which is suggested by scientific realism, the facticity view:
laws of nature describe how physical systems behave. This
is by far the commonest view, and a sensible one; but it
does not work. It does not fit explanatory laws, like the
fundamental laws of physics. Some other view is needed if
we are to account for the use of laws in explanation; and
I do not see any obvious candidate that is consistent with
the realist's reasonable demand that laws describe reality
and state facts that might well be true. There is, I have
argued, a trade-off between factual content and explana-
tory power. We explain certain complex phenomena as
the result of the interplay of simple, causal laws. But what
do these laws say? To play the role in explanation we de-
mand of them, these laws must have the same form when
they act together as when they act singly. In the simplest
case, the consequences that the laws prescribe must be
exactly the same in interaction, as the consequences that
would obtain if the law were operating alone. But then,
what the law states cannot literally be true, for the con-
sequences that would occur if it acted alone are not the
consequences that actually occur when it acts in com-
bination.
If we state the fundamental laws as laws about what
happens when only a single cause is at work, then we can
suppose the law to provide a true description. The problem
arises when we try to take that law and use it to explain
the very different things which happen when several causes
are at work. This is the point of 'The Truth Doesn't Explain
Much'. There is no difficulty in writing down laws which
we suppose to be true: 'I[ there are no charges, no nuclear
forces, . . . then the force between two masses of size m
and m' separated by a distance r is Gmm'lri,' We count
this law true-what it says will happen, does happen-or
at least happens to within a good approximation. But this
law does not explain much. It is irrelevant to cases where
there are electric or nuclear forces at work. The laws of
physics, I concluded, to the extent that they are true, do
not explain much. We could know all the true laws of nature,
and still not know how to explain composite cases. Ex-
planation must rely on something other than law.
But this view is absurd. There are not two vehicles for
explanation: laws for the rare occasions when causes occur
separately; and another secret, nameless device for when
they occur in combination. Explanations work in the same
way whether one cause is at work, or many. 'Truth Doesn't
Explain' raises perplexities about explanation by composition
of causes; and it concludes that explanation is a very peculiar
scientific activity, which commonly does not make use of
laws of nature. But scientific explanations do use laws. It
is the laws themselves that are peculiar. The lesson to be
learned is that the laws that explain by composition of
causes fail to satisfy the facticity requirement. If the laws
of physics are to explain how phenomena are brought about,
they cannot state the facts.
Essay 4
The Reality of Causes

in a World of Instrumental Laws
o. INTRODUCTION
Empiricists are notoriously suspicious of causes. They have

not been equally wary of laws. Hume set the tradition when
he replaced causal facts with facts about generalizations.
Modern empiricists do the same. But nowadays Hume's
generalizations are the laws and equations of high level
scientific theories. On current accounts, there may be some
question about where the laws of our fundamental theories
get their necessity; but it is no question that these laws
are the core of modern science. Bertrand Russell is well
known for this view:
The law of gravitation will illustrate what occurs in any exact science ...
Certain differential equations can be found, which hold at every in-
stant for every particle of the system . . . But there is nothing that
could be properly called 'cause' and nothing that could be properly
called 'effect' in such a system.'
For Russell, causes 'though useful to daily life and in the

infancy of a science, tend to be displaced by quite different
laws as soon as a science is successful'.
It is convenient that Russell talks about physics, and that
the laws he praises are its fundamental equations-Hamilton's
equations or Schroedinger's, or the equations of general
relativity. That is what I want to discuss too. But I hold
just the reverse of Russell's view. I am in favour of causes
and opposed to laws. I think that, given the way modern
theories of mathematical physics work, it makes sense only
to believe their causal claims and not their explanatory laws.
1 Bertrand Russell, 'On the Notion of Cause', Mysticism and Logic (London:
Allen & Unwin, 1917), p. 194.

THE REALITY OF CAUSES 75
1. EXPLAINING BY CAUSES
Following Bromberger, Scriven, and others, we know that

there are various things one can be doing in explaining.
Two are of importance here: in explaining a phenomenon
one can cite the causes of that phenomenon; or one can set
the phenomenon in a general theoretical framework. The
framework of modern physics is mathematical, and good
explanations will generally allow us to make quite precise
calculations about the phenomena we explain. Rene Thorn
remarks the difference between these two kinds of explana-
tion, though he thinks that only the causes really explain:
'DesCartes with his vortices, his hooked atoms, and the
like explained everything and calculated nothing; Newton,
with the inverse square of gravitation, calculated everything
and explained nothing'i?
Unlike Thorn, I am happy to call both explanation, so long
as we do not illicitly attribute to theoretical explanation
features that apply only to causal explanation. There is a
tradition, since the time of Aristotle, of deliberately con-
flating the two. But I shall argue that they function quite
differently in modern physics. If we accept Descartes's
causal story as adequate, we must count his claims about
hooked atoms and vortices true. But we do not use Newton's
inverse square law as if it were either true or false.
One powerful argument speaks against my claim and for
the truth of explanatory laws-the argument from coin-
cidence. Those who take laws seriously tend to subscribe
to what Gilbert Harman has called inference to the best
explanation. They assume that the fact that a law explains
provides evidence that the law is true. The more diverse
the phenomena that it explains, the more likely it is to be
true. It would be an absurd coincidence if a wide variety
of different kinds of phenomena were all explained by a
particular law, and yet were not in reality consequent from
the law. Thus the argument from coincidence supports a
good many of the inferences we make to best explanations.
The method of inference to the best explanation is subject
2 Rene Thorn, Structural Stability and Morphogenesis, trans. C. H. Wadding-
ton (Reading, Mass.: W. A. Benjamin, 1972), p. 5.
76 THE REALITY OF CAUSES
to an important constraint, however-the requirement of

non-redundancy. We can infer the truth of an explanation
only if there are no alternatives that account in an equally
satisfactory way for the phenomena. In physics nowadays,
I shall argue, an acceptable causal story is supposed to satisfy
this requirement. But exactly the opposite is the case with
the specific equations and models that make up our theo-
retical explanations. There is redundancy of theoretical
treatment, but not of causal account.
There is, I think, a simple reason for this: causes make their
effects happen. We begin with a phenomenon which, relative
to our other general beliefs, we think would not occur
unless something peculiar brought it about. In physics we
often mark this belief by labelling the phenomena as
effects-the' Sorbet effect, the Zeeman effect, the Hall
effect. An effect needs something to bring it about, and
the peculiar features of the effect depend on the particular
nature of the cause, so that-in so far as we think we have
got it right-we are entitled to infer the character of the
cause from the character of the effect.
But equations do not bring about the phenomenological laws
we derive from them (even if the phenomenological laws are
themselves equations). Nor are they used in physics as if they
did. The specific equations we use to treat particular phe-
nomena provide a way of casting the phenomena into the gen-
eral framework of the theory. Thus we are able to treat a variety
of disparate phenomena in a similar way, and to make use of
the theory to make quite precise calculations. For both of these
purposes it is an advantage to multiply theoretical treatments.
Pierre Duhem used the redundancy requirement as an
argument against scientific realism, and recently Hilary
Putman uses it as an argument against realism in general.
Both propose that, in principle, for any explanation of
any amount of data there will always be an equally satis-
factory alternative. The history of science suggests that
this claim may be right: we constantly construct better
explanations to replace those of the past. But such argu-
ments are irrelevant here; they do not distinguish between
causal claims and theoretical accounts. Both are likely to
be replaced by better accounts in the future.
Here I am not concerned with alternatives that are at best

available only in principle, but rather with the practical
availability of alternatives within theories we actually have
to hand. For this discussion, I want to take the point of
view that Putnam calls 'internal realism'; to consider actual
physical theories which we are willing to account as accept-
able, even if only for the time being, and to ask, 'Relative
to that theory, which of its explanatory claims are we to
deem true?' My answer is that causal claims are to be deemed
true, but to count the basic explanatory laws as true is to
fail to take seriously how physics succeeds in giving ex-
planations.
I will use two examples to show this. The first-quantum
damping and its associated line broadening-is a phe-
nomenon whose understanding is critical to the theory of
lasers. Here we have a single causal story, but a fruitful
multiplication of successful theoretical accounts. This con-
trasts with the unacceptable multiplication of causal stories
in the second example.
There is one question we should consider before looking
at the examples, a question pressed by two colleagues in
philosophy of science, Dan Hausman and Robert Ennis.
How are we to distinguish the explanatory laws, which I
argue are not to be taken literally, from the causal claims
and more pedestrian statements of fact, which are? The
short answer is that there is no way. A typical way of treat-
ing a problem like this is to find some independent criterion
-ideally syntactical, but more realistically semantical-
which will divide the claims of a theory into two parts.
Then it is argued that claims of one kind are to be taken
literally, whereas those of the other kind function in some
different way.
This is not what I have in mind. I think of a physics theory
as providing an explanatory scheme into which phenomena
of interest can be fitted. I agree with Duhem here. The
scheme simplifies and organizes the phenomena so that we
can treat similarly happenings that are phenomenologically
different, and differently ones that are phenomenologically
the same. It is part of the nature of this organizing activity
that it cannot be done very well if we stick too closely to
stating what is true. Some claims of the theory must be

literally descriptive (I think the claims about the mass and
charge of the electron are a good example) if the theory
is to be brought to bear on the phenomena; but I suspect
that there is no general independent way of characterizing
which these will be. What is important to realize is that if
the theory is to have considerable explanatory power, most
of its fundamental claims will not state truths, and that
this will in general include the bulk of our most highly
prized laws and equations.
2. EXAMPLES: QUANTUM DAMPING

In radiative damping, atoms de-excite, giving off photons
whose frequencies depend on the energy levels of the atom.
We know by experiment that the emission line observed
in a spectroscope for a radiating atom is not infinitely sharp,
but rather has a finite linewidth: that is, there is a spread
of frequencies in the light emitted. What causes this natural
linewidth? Here is the standard answer which physicists give,
quoted from a good textbook on quantum radiation theory
by William Louisell:
There are many interactions which may broaden an atomic line, but
the most fundamental one is the reaction of the radiation field on the
atom. That is, when an atom decays spontaneously from an excited
state radiatively, it emits a quantum of energy into the radiation
field. This radiation may be reabsorbed by the atom. The reaction
of the field on the atom gives the atom a line width and causes the
original level to be shifted as we show. This is the source of the natural
line width and the Lamb shift. 3
Following his mathematical treatment of the radiative

decay, Louisell continues:
We see that the atom is continually emitting and reabsorbing quanta of

radiation. The energy level shift does not require energy to be con-
served while the damping requires energy conservation. Thus damping
3 William H. Louisell, Quantum-Statistical Properties of Radiation (New

York: John Wiley & Sons, 1973), p. 285.
is brought about by the emission and absorption of real photons
while the photons emitted and absorbed which contribute to the
energy shift are called virtual photons."
This account is universally agreed upon. Damping, and its

associated line broadening, are brought about by the emis-
sion and absorption of real photons.
Here we have a causal story; but not a mathematical
treatment. We have not yet set line broadening into the
general mathematical framework of quantum mechanics. Q
r-
There are many ways to do this. One of the Springer Tracts 'J
by G. S. Agarwal" summarizes the basic treatments whicb, t:
are offered. He lists six different approaches in his tabl?"
of contents: (1) Weisskopf-Wigner method; (2) Heitler-Ma. :
method; (3) Goldberger-Watson method; (4) Quantum.Y
Statistical method: master equations; (5) Langevin equations .
corresponding to the master equation and a c-number re-.
presentation; and (6) neoclassical theory of spontaneous
emission.
Before going on to discuss these six approaches, I will
give one other example. The theory of damping forms the
core of current quantum treatments of lasers. Figure 4.1
is a diagram from a summary article by H. Haken on 'the'
quantum theory of the laser. 6 We see that the situation I
described for damping theory is even worse here. There
are so many different treatments that Haken provides a
'family tree' to set straight their connections. Looking at
the situation Haken himself describes it as a case of 'theory
overkill'. Laser theory is an extreme case, but I think there
is no .doubt that this kind of redundancy of treatment,
which Haken and Agarwal picture, is common throughout
physics.
Agarwal describes six treatments of line broadening. All
six provide precise and accurate calculations for the shape
4 Ibid., p. 289.
5 See G. S. Agarwal, Quantum-Statistical Theories of Spontaneous Emission
and their Relation to Other Approaches (Berlin: Springer-Verlag, 1974).
6 II. Haken, 'The Semiclassical and Quantum Theory of the Laser', in S. M.
Kay and A. Maitland (008), Quantum Optics (London: Academic Press, 1970),
p.244.
Results which can be checked by experiments: Photon distribution

correlation functions for phase and higher order correlations
amplitude (intensity) fluctuations.
approx.
1
direct approx.
solution solution ~
"C
o
;;
E
..E
c
o
.,
o
C.
0.
<I:
I
----.,
I
I I
I I
I I
I I
I I
I I neglect of higher
order derivatives
I I
_I L_
--------1
Quantum mechan. general. I
Langevin equations Fokker-Planck equation I
field + atoms field + atoms
I
I
I
I
This box I
contains the I
II exact equations JI
*A: adiabatic elimination of atomic variables

t E : exact elimination of atomic variables
t L + Q: linearization and quantum mechanical quasilinearization
FIG. 4.1. Family tree of the quantum theory of the laser. (Source:
Haken, 'The Semiclassical and Quantum Theory of the Laser")
and width of the broadened line. How do they differ? All of

the approaches employ the basic format of quantum mech-
anics. Each writes down a Schroedinger equation; but it is
a different equation in each different treatment. (Actually
among the six treatments there are really just three different
equations.) The view that I am attacking takes theoretical

explanations to provide, as best they can, statements of
objective laws. On this view the six approaches that Agarwal
lists compete with one another; they offer different laws
for exactly the same phenomena.
But this is not Agarwal's attitude. Different approaches
are useful for different purposes; they complement rather
than compete. The Langevin and Master equations of (4)
and (5), for instance, have forms borrowed from statistical
mechanics. They were introudced in part because the de-
velopment of lasers created an interest in photon correlation
experiments. Clearly, if we have statistical questions, it is
a good idea to start with the kind of equations from which
we know how to get statistical answers.
Let us consider an objection to the point of view I have
been urging. We all know that physicists write down the
kinds of equations they know how to solve; if they cannot
use one approximation, they try another; and when they
find a technique that works, they apply it in any place they
can. These are commonplace observations that remind us
of the pragmatic attitude of physicists. Perhaps, contrary
to my argument, the multiplication of theoretical treat-
ments says more about this pragmatic orientation than it
does about how explanatory laws ought to be viewed. I
disagree. I think that it does speak about laws, and in par-
ticular shows how laws differ from causes. We do not have
the same pragmatic tolerance of causal alternatives. We do
not use first one causal story in explanation, then another,
depending on the ease of calculation, or whatever.
The case of the radiometer illustrates. The radiometer
was introduced by William Crookes in 1873, but it is still
not clear what makes it work. Recall from the Introduction
to these essays that there are three plausible. theories. The
first attributes the motion of the vanes to light pressure.
This explanation is now universally rejected. As M. Goldman
remarks in 'The Radiometer Revisited',
A simple calculation shows that on a typical British summer day,

when the sky is a uniform grey (equally luminous all over) the torque
from the black and silver faces exactly balance, so that for a perfect
radiometer [i.e., a radiometer with a perfect vacuum] no motion

would be possible."
Two explanations still contend. The first is the more

standard, textbook account, which is supported by Gold-
man's calculations. It supposes that the motion is caused
by the perpendicular pressure of the gas in the perfect
vacuum against the vanes. But as we have seen, on Maxwell's
account the motion must be due to the tangential stress
created by the gas slipping around the edge of the vanes.
There is a sense in which Maxwell and Goldman may both
be right: the motion may be caused by a combination of
tangential and perpendicular stress. But this is not what
they claim. Each claims that the factor he cites is the single
significant factor in bringing about the motion, and only one
or the other of these claims can be accepted. This situation
clearly constrasts with Agarwal's different theoretical treat-
ments. In so far as we are interested in giving a causal ex-
planation of the motion, we must settle on one account or
the other. We cannot use first one account, then the other,
according to our convenience.
I know of this example through Francis Everitt, who
thinks of building an experiment that would resolve the
question. I mention Everitt's experiment again because it
argues for the difference in objectivity which I urge between
theoretical laws and causal claims. It reminds us that unlike
theoretical accounts, which can be justified only by an
inference to the best explanation, causal accounts have an
independent test of their truth: we can perform controlled
experiments to find out if our causal stories are right or
wrong. Experiments of these kinds in fact play an important
role in an example from which Wesley Salmon defends
inferences to the best explanation.
3. THE ARGUMENT FROM COINCIDENCE
In a recent paper" Salmon considers Jean Perrin's arguments

for the existence of atoms and for the truth of Avagadro's
7 M. Goldman, 'The Radiometer Revisited', Physics Education 13 (1978),
p.428. 8 See footnote 21, Introduction.
hypothesis that there are a fixed number of molecules in
any gram mole of a fluid. Perrin performed meticulous
experiments on Brownian motion in colloids from which
he was able to calculate Avagadro's number quite precisely.
His 1913 tract, in which he summarizes his experiments
and recounts the evidence for the existence of atoms, helped
sway the community of physicists in favour of these hy-
potheses. Besides Brownian motion, Perrin lists thirteen
quite different physical situations which yield a deter-
mination of Avogadro's number. So much evidence of such
a variety of kinds all pointing to the same value must surely
convince us, urges Perrin, that atoms exist and that Avogadro's
hypothesis is true.
For many, Perrin's reasoning is a paradigm of inference
to the best explanation; and it shows the soundness of that
method. I think this misdiagnoses the structure of the argu-
ment. Perrin does not make an inference to the best explana-
tion, where explanation includes anything from theoretical
laws to a detailed description of how the explanandum was
brought about. He makes rather a more restricted inference
-an inference to the most probable cause.
A well-designed experiment is constructed to allow us to
infer the character of the cause from the character of its
more readily observable effects. Prior to Perrin, chemists
focused their attention on the size and velocities of the
suspended particles. But this study was unrewarding; the
measurements were difficult and the results did not signify
much. Perrin instead studied the height distribution of the
Brownian granules at equilibrium. From his results, with a
relatively simple model for the collision interactions, he
was able to calculate Avogadro's number. Perrin was a bril-
liant experimenter. It was part of his genius that he was
able to find quite specific effects which were peculiarly
sensitive to the exact character of the causes he wanted to
study. Given his model, the fact that the carrier fluids had
just 6 X 1023 atoms for every mole made precise and cal-
culable differences to the distribution he observed.
The role of the model is important. It brings out exactly
what part coincidence plays in the structure of Perrin's
argument. Our reasoning from the character of the effect
to the character of the cause is always against a background

of other knowledge. We aim to find out about a cause with a
particular structure. What effects appear as a result of that
structure will be highly sensitive to the exact nature of the
causal processes which connect the two. If we are mistaken
about the processes that link cause and effect in our experi-
ment, what we observe may not result in the way we think
from the cause under study. Our results may be a mere
artefact of the experiment, and our conclusions will be
worthless.
Perrin explicitly has this worry about the first of the
thirteen phenomena he cites: the viscosity of gases, which
yields a value for Avogadro's number via Van der Waal's
equation and the kinetic theory of gases. In his Atoms
he writes that 'the probable error, for all these numbers
is roughly 30 per cent, owing to the approximations made
in the calculations that lead to the Clausius-Maxwell and
Van der Waal's equations.' He continues: 'The Kinetic
Theory justly excites our admiration. [But] it fails to carry
complete conviction, because of the many hypotheses it
involves.' (I take it he means 'unsubstantiated hypotheses'.)
What sets Perrin's worries to rest? He tell us himself in the
next sentence: 'If by entirely independent routes we are
led to the same values for the molecular magnitudes, we
shall certainly find our faith in the theory considerably
strengthened. '9
Here is where coincidence enters. We have thirteen phe-
nomena from which we can calculate Avogadro's number.
Anyone of these phenomena-if we were sure enough
about the details of how the atomic behavour gives rise
to it-would be good enough to convince us that Avogadro
is right. Frequently we are not sure enough; we want further
assurance that we are observing genuine results and not
experimental artefacts. This is the case with Perrin. He lacks
confidence in some of the models on which his calculations
are based. But he can appeal to coincidence. Would it not
be a coincidence if each of the observations was an artefact,
and yet all agreed so closely about Avogadro's number?
9 Jean Perrin, Atoms, trans. D. LL Hammick (New York: D. Van Nostrand
Co., 1916), p. 82.
The convergence of results provides reason for thinking

that the various models used in Perrin's diverse calculations
were each good enough. It thus reassures us that those
models can legitimately be used to infer the nature of the
cause from the character of the effects.
In each of Perrin's thirteen cases we infer a concrete
cause from a concrete effect. We are entitled to do so be-
cause we assume that causes make effects occur in just
the way that they do, via specific, concrete causal processes.
The structure of the cause physically determines the struc-
ture of the effect. Coincidence enters Perrin's argument,
but not in a way that supports inference to the best explana-
tion in general. There is no connection analogous to causal
propagation between theoretical laws and the phenomeno-
logical generalizations which they bring together and explain.
Explanatory laws summarize phenomenological laws; they
do not make them true. Coincidence will not help with
laws. We have no ground for inferring from any phenomeno-
logical law that an explanatory law must be just so; multi-
plying cases cannot help.
I mentioned that Gilbert Harman introduced the expression
'inference to the best explanation'. Harman uses two ex-
amples in his original paper.'? The first is the example that
we have just been discussing: coming to believe in atoms.
The second is a common and important kind of example
from everyday life: inferring that the butler did it. Notice
that these are both cases in which we infer facts about
concrete causes: they are not inferences to the laws of
some general explanatory scheme. Like Perrin's argument,
these do not vindicate a general method for inferring the
truth of explanatory laws. What they illustrate is a far more
restrictive kind of inference: inference to the best cause.
4. CONCLUSION
Perrin did not make an inference to the best explana-
tion, only an inference to the most probable cause. This
is typical of modern physics. 'Competing' theoretical
10 G. H. Harman, 'Inference to the Best Explanation', Philosophical Review
74 (1965), pp. 88-95.
treatments-treatments that write down different laws

for the same phenomena-are encouraged in physics, but
only a single causal story is allowed. Although philosophers
generally believe in laws and deny causes, explanatory
practice in physics is just the reverse.
Essay 5
When Explanation Leads to Inference
O. INTRODUCTION
When can we infer the best explanation? This question

divides scientific realists on the one hand, from operation-
alists, instrumentalists, positivists, and constructive empiri-
cists on the other. There must obviously be certain provisions
to ensure that 'the best' is good enough. But once these are
understood, the realist answer to the question is 'always';
the anti-realist, 'never'. The realist asks, 'How could some-
thing explain if it was not true?' The anti-realist thinks this
question exposes a mistaken view about what we do in
explaining. Explanations (at least the high level explanations
of theoretical science which are the practical focus of the
debate) organize, briefly and efficiently, the unwieldy,
and perhaps unleamable, mass of highly detailed knowledge
that we have of the phenomena. But organizing power has
nothing to do with truth.
I am going to discuss two anti-realists, Bas van Fraassen
and Pierre Duhem. Van Fraassen's book, The Scientific
Image,l provides a powerful and elegant defence of a brand
of anti-realism which he calls 'constructive empiricism'.
Duhem's views are laid out in his classic work of 1906,
The Aim and Structure of Physical Theory.? According to
van Fraassen the constructive empiricist maintains:
Science aims to give us theories which are empirically adequate; and
acceptance of a theory involves as belief only that it is empirically
adequate. (Basically, a theory is empirically adequate 'exactly if what
it says about the observable things and events in this world is true'i)"
Van Fraassen presents the difference between the realist and

the constructive empiricist as one of attitude. Both may
1 Bas C. van Fraassen, The Scientific Image (Oxford: Clarendon Press, 1980).
2 See Pierre Duhem, The Aim and Structure of Physical Theory, trans. Philip
P. Wiener (New York: Atheneum, 1962).
3 van Fraassen, op, cit., p. 12.
88 WHEN EXPLANATION LEADS TO INFERENCE
explain by showing how the phenomena at hand can be

derived from certain fundamental princples. But the two
kinds of philosopher have opposing attitudes to the principles.
The realist believes that they are true and genuinely give
rise to the phenomena; the constructive empiricist believes
merely that the principles are sufficient to derive the phe-
nomena.
The realist, says van Fraassen, is making a mistake. When
a theory succeeds in saving the phenomena, the scientific
realist is ready to infer that its laws are true (or near true,
or true for the nonce) and that its entities exist. Van Fraassen
holds that a theory's success at saving the phenomena gives
reason to believe just that: that it saves the phenomena,
and nothing more. That the theory is true is a gratuitous
additional assumption.
This is the core of Duhem's view as well. Duhem has no
quarrel with phenomenological laws, which can be con-
firmed by inductive methods. What he opposes are theo-
retical laws, whose only ground is their ability to explain.
Like van Fraassen, Duhem rejects theoretical laws because
he does not countenance inference to the best explanation.
Neither van Fraassen nor Duhem are opposed to ampliative
inference in general. They make a specific and concrete
attack on a particular kind of inference which they see as
invalid-inference to the best explanation-and thereby
on the scientific realism to which it gives rise.
This is the real interest of their view. They have a specific
objection to one mode of reasoning and one class of scientific
conclusions. They do not argue from a sweeping sceptical
position that starts with the weakness of our senses and
the poverty of our capacities and concludes that no one can
ever know anything. Nor do they argue from a theory of
meaning that counts theoretical talk as devoid of truth
value along with all of our claims to morality, causality,
and religion. Finally, they are not transcendental idealists
like Kant. Nor (to use Ian Hacking's apt label) are they
transcendental nominalists like Hilary Putnam, who argues
that since thought can never connect with reality, our know-
ledge can achieve at best an internal coherence. Duhem
and van Fraassen make a distinction within the field of
WHEN EXPLANATION LEADS TO INFERENCE 89
scientific knowledge, while scepticism, positivism, and the

transcendentalisms are global doctrines about the whole
domain of science. Duhem and van Fraassen allow that
many inferences are sound, but not inferences to pure
theory that are justified only in terms of explanation.
Their arguments are persuasive. But 1 think that van
Fraassen and Duhem eliminate more than they should.
It is apparent from earlier essays that 1 share their anti-
realism about theoretical laws. On the other hand, 1 believe
in theoretical entities, and that is my main topic in this essay.
Arguments against inference to the best explanation do
not work against the explanations that theoretical entities
provide. These are causal explanations, and inference from
effect to cause is legitimate. 1 will have nothing new to
say about the structure of these inferences. 1 aim only to
show that we can be realists about theoretical entities on
van Fraassen and Duhem's very own grounds.
1. V AN FRAASSEN'S ATTACK
Van Fraassen asks, 'Why should 1 believe in theoretical

entities?' There is a canonical answer: there are no genuine
regularities at the phenomenological level. It is only among
theoretical entities that science finds true regularities. Once
we have laid these out, we have a powerful explanatory
scheme. The exceptionless laws that we postulate at the
theoretical level can explain not only why the phenomena
are as regular as they are, but why we see the exceptions
we do. Van Fraassen grants this. But, he asks, what reason
do we have for inferring from the fact that a bundle of
principles save the phenomena to the fact that they are
true? We need some reason, some good reason, though
certainly not a conclusive reason. Many arguments wear
their validity on their sleeve: 'I think. Therefore, 1 exist.'
But not, 'P explains Q. Q is true. Therefore P is true.'
This argument, and Duhem's as well, assumes that truth
is an external characteristic of explanation; i.e., that some-
thing could satisfy all the other criteria for being an ex-
planation and yet fail to be true. This is the way we are
often taught to think of Ptolemaic astronomy. It might
well form a completely satisfactory explanatory scheme, yet

that does not settle the question of its truth. This, for in-
stance, is what the medieval Piccolomini, one of Duhem's
heroes, says of Ptolemy and his successors:
for these astronomers it was amply sufficient that their constructs

save the appearances, that they allow for the reckoning of the move-
ments of the heavenly bodies, their arrangements, and their place.
Whether or not things really are as they envisage them-that question
they leave to the philosophers of nature."
As we saw in the last essay, Duhem's own argument against

inference to the best explanation is the argument from
redundancy: for any given set of phenomena, in principle
there will always be more than one equally satisfactory
explanation, and some of these explanations will be in-
compatible. 'Since not all of them can be true, it is clear
that truth is independent of satisfactoriness for explanation.
Sometimes Duhem's argument is read epistemologically
instead. He is taken to make a point, not about what the
criteria are, but rather about our ability to see if they obtain.
On the epistemological reading Duhem maintains merely
that there will always be different laws which appear equally
true, so far as we can ever tell, and yet are incompatible.
This, I think, is a mistaken reading. For it is a general
feature of our knowledge and does not show what is peculiar
to the inference-to-best-explanation which Duhem attacks.
Duhem is not, for example, opposed to phenomenological
laws, which arise by inductive generalization. It is a familiar
fact that it is possible to construct different inductive rules
which give rise to different generalizations from the same
evidence. Here too there will always be more than one
incompatible law which appears equally true so far as we
can tell. These kinds of problems with inductive inference
were known to Duhem. But he did not dwell on them. His
concern was not with epistemological issues of this sort,
but rather with the relationship between truth and ex-
planation.
I said that Duhem and van Fraassen take truth to be an
4 Pierre Duhem, To Save the Phenomena, trans. Edmund Doland and Chanenah
Maschler (Chicago: University of Chicago Press, 1969), p. 82.
external characteristic to explanation. Here is an analogy.

I ask you to tell me an interesting story, and you do so.
I may add that the story should be true. But if I do so,
that is a new, additional requirement like the one which
Piccolomini's natural philosophers make. They will call
something a genuine explanation only if it does all else
it should and is in addition true. Here is another analogy.
Van Fraassen and Duhem challenge us to tell what is special
about the explanatory relation. Why does the truth of
the second relatum guarantee the truth of the first? Two-
placed relations do not in general have that characteristic.
Consider another two-placed relation: -- is the paper at
the 1980 Western Washington Conference that immediately
preceded --. In that year my paper immediately preceded
Richard Wollheim's; Wollheim's paper may well have been
true. But that does not make mine true.
2. THE CASE FOR THEORETICAL ENTITIES
Van Fraassen and Duhem argue that explanation has truth

going along with it only as an extra ingredient. But causal
explanations have truth built into them. When I infer from
an effect to a cause, I am asking what made the effect occur,
what brought it about. No explanation of that sort explains
at all unless it does present a cause; and in accepting such
an explanation, I am accepting not only that it explains
in the sense of organizing and making plain, but also that
it presents me with a cause. My newly planted lemon tree
is sick, the leaves yellowing and dropping off. I finally
explain this by saying that water has accumulated in the base
of the planter: the water is the cause of the disease. I drill
a hole in the base of the oak barrel where the lemon tree
lives, and foul water flows out. That was the cause. Before
I had drilled the hole, I could still give the explanation
and to give that explanation was to present the supposed
cause, the water. There must be such water for the explana-
tion to be correct. An explanation of an effect by a cause
has an existential component, not just an optional extra
ingredient.
Likewise when I explain the change in rate in fall of a
a light droplet in an electric field, by asserting that there are

positrons or electrons on the ball, I am inferring from effect
to cause, and the explanation has no sense at all without the
direct implication that there are electrons or positrons on
the ball. Here there is no drilling a hole to let the electrons
gush out before our eyes. But there is the generation of
other effects: if the ball is negatively charged, I spray it
with a positron emitter and thereby change the rate of fall
of the ball: positrons from the emitter wipe out the electrons
on the ball. What I invoke in completing such an explana-
tion are not fundamental laws of nature, but rather properties
of electrons and positrons, and highly complex, highly specific
claims about just what behaviour they lead to in just this
situation. I infer to the best explanation, but only in a
derivative way: I infer to the most probable cause, and
that cause is a specific item, what we call a theoretical
entity. But note that the electron is not an entity of any
particular theory. In a related context van Fraassen asks
if it is the Bohr electron, the Rutherford electron, the Lorenz
electron or what. The answer is, it is the electron, about
which we have a large number of incomplete and sometimes
conflicting theories.
Indeed I should use an example of van Fraassen's here
to show how we differ. In a cloud chamber we see certain
tracks which he says have roughly the same physical ex-
planation as the vapour trail from a jet in the sky. In each
case I may explain the trail by stating some laws. But what
about the entities? I say that the most probable cause of the
track in the chamber is a particle, and as I find out more, I
can even tell you with some specificity what kind of particle.
That, argues van Fraassen, is still quite different from the
jet in the sky. For, there, van Fraassen says, look at the
speck just ahead of the trail, or here, use this powerful
glass to spy it out. There is no such spying out when we get
to the cloud chamber. I agree to that premise but not to
the conclusion. In explaining the track by the particle, I am
saying that the particle causes the track, and that explanation,
or inference to the most probable cause, has no sense unless
one is asserting that the particle in motion brings about,
causes, makes, produces, that very track. The particle in
the cloud chamber is just one example. Our belief in theo-

retical entities is generally founded on inferences from
concrete effects to concrete causes. Here there is an answer
to the van Fraassen-Duhem question. What is special about
explanation by theoretical entity is that it is causal explana-
tion, and existence is an internal characteristic of causal
claims. There is nothing similar for theoretical laws.
Van Fraassen does not believe in causes. He takes the
whole causal rubric to be a fiction. That is irrelevant here.
Someone who does not believe in causes will not give causal
explanations. One may have doubts about some particular
causal claims, or, like van Fraassen, about the whole enter-
prise of giving causal explanations. These doubts bear only
on how satisfactory you should count a causal explanation.
They do not bear on what kind of inferences you can make
once you have accepted that explanation.
We can see this point by contrasting causal explanation
with the explanation of one law by another, or with the
'preceding paper' relation I mentioned above. We need to
sort the special van Fraassen-Duhern challenge we have
been discussing from more general epistemological worries
that make us question (as perhaps we always should) whether
we really do have a good explanation. So let us introduce a
fiction. God may tell you that Wollheim's paper is after
mine, and that his paper is true. You have no do buts about
either of those propositions. This signifies nothing about
the truth of my paper. Similarly, God tells you that Schroe-
dinger's equation provides a completely satisfactory deriva-
tion of the phenomenological law of radioactive decay. You
have no doubt that the derivation is correct. But you still
have no reason to believe in Schroedinger's equation. On
the other hand, if God tells you that the rotting of the
roots is the cause of the yellowing of the leaves, or that the
ionization produced by the negative charge explains the track
in the cloud chamber, then you do have reason, conclusive
reason, to believe that there is water in the tub and that there
is an electron in the chamber.
3. AN OBJECTION
I argue that inferences to the most likely cause have a differ-

ent logical force than inferences to the best explanation.
Larry Laudan raises a serious objection: 'It seems to me that
your distinctions are plausible only because you insist
(apparently arbitrarily) on countenancing a pragmatic view of
theoretical laws and a non-pragmatic view of causal talk."
In order to explain why I think this distinction is not arbi-
trary, I will layout two very familiar views of explanation,
one that underlies the deductive-nomological (D-N) model,"
and the second, the view of Duhem. Van Fraassen challenges
the realist to give an account of explanation that shows
why the success of the explanation, coupled with the truth
of the explanandum, argues for the truth of the explanans.
I said there was an answer to this question in the case of
causal inference. Similarly, I think there is an answer in
the D-N account. If the D-N model is a correct account of
what explanation is like, I agree that my distinction is arbi-
trary; but this is not so if Duhem is right.
If we could imagine that our explanatory laws made the
phenomenological laws true, that would meet van Fraassen's
challenge. But there is another, more plausible account
that would do just as well. I discuss this account in great
detail in the next essay. Adolf Grunbaum gives a brief sketch
of the view:
It is crucial to realize that while (a more comprehensive law) G entails
(a less comprehensive law) L logically, thereby providing an explanation
of L, G is not the 'cause' of L. More specifically, laws are explained
not by showing the regularities they affirm to be products of the
operation of causes but rather by recognizing their truth to be special
cases of more comprehensive truths."
For any specific situation the fundamental laws are sup-
posed to make the same claims as the more concrete phe-
nomenological laws that they explain. This is borne out
5In correspondence, dated 15 September 1981.
For a description of the deductive-nomological model of explanation,
6
see C. G. Hempel, Philosophy of Natural Science (Englewood Cliffs, N.J.:

Prentice-Hall, 1966).
7 Adolf Griinbaum, 'Science and Ideology', The Scientific Monthly (July
1954), pp, 13-19, italics in the original.
WHEN EXPLANA nON LEADS TO INFERENCE 95
by the fact that the phenomenological laws can be deduced

from the fundamental laws, once a description of the situa-
tion is supplied. If the phenomenological laws have got it right,
then so too do the fundamental, at least in that situation.
There is still an inductive problem: are the fundamental laws
making the right generalization across situations? But at least
we see why the success of the explanation requires the truth
of the explanans. To explain a phenomenological law is to
restate it, but in a sufficiently abstract and general way that
states a variety of other phenomenological laws as well. Ex-
planatory laws are true statements of what happens; but unlike
phenomenological laws, they are economical ways to say a lot.
This may seem straightforward. What else could explana-
tion be? But contrast Duhem. Duhem believes that phe-
nomena in nature fall roughly into natural kinds. The realist
looks for something that unifies the members of the natural
kind, something they all have in common; but Duhem
denies that there is anything. There is nothing more than
the rough facts of nature that sometimes some things behave
like others, and what happens to one is a clue to what the
others will do. Explanations provide a scheme that allows
us to make use of these clues. Light and electricity behave
in similar ways, but the procedures for drawing the analogies
are intricate and difficult. It is easier for us to postulate
the electromagnetic field and Maxwell's four laws, to see
both light and electricity as a manifestation of one single
underlying feature. There is no such feature, but if we are
careful, we are better off to work with these fictional unifiers
than to try to comprehend the vast array of analogies and
disanalogies directly. The explanatory schemes we posit
work as well as they do, even to producing novel productions,
because phenomena do roughly fall into natural kinds. But
in fact the phenomena are genuinely different. They only
resemble each other some times in some ways, and the
D-N attempt to produce one true description for all the
members of the same class must inevitably fail. We can-
not expect to find an explanatory law that will describe
two phenomena that are in fact different, and yet will
be true of both. What we can require of explanation is a
scheme that allows us to exploit what similarities there are.
These are very cursory descriptions of the two views.

But it is enough to see that the two embody quite different
conceptions of explanation. Nor is it just a matter of choos-
ing which to pursue, since they are joined to distinct meta-
physical pictures. In practice the two conceptions meet;
for in real life explanations, failure of deductivity is the
norm. Duhem predicts this. But proponents of the D-N
model can account for the practical facts as well. They
attribute the failure of deductivity not to the lack of unity
in nature, but to the failings of each particular theory we
have to hand.
The difference between the two conceptions with respect
to van Fraassen's challenge may be obscured by this practical
convergence. We sometimes mistakenly assume that individual
explanations, on either account, will look the same. Van
Fraassen himself seems to suppose this; for he requires that
the empirical substructures provided by a theory should be
isomorphic to the true structures of the phenomena. But
Duhem says that there can be at best a rough match. If
Duhem is right, there will be no wealth of truly deductive
explanations no matter how well developed a scientific
discipline we look to.
Duhem sides with the thinkers who say 'A physical theory
is an abstract system whose aim is to summarize and to
classify logically a group of experimental laws without
aiming to explain these laws', where 'to explain (explicate,
explicate) is to strip reality 'of the appearances covering
it like a veil, in order to see the bare reality itself." In an
effort to remain metaphysically neutral, we might take an
account of explanation which is more general than either
Duhem's or the D-N story: to explain a collection of phe-
nomenological laws is to give a physical theory of them, a
physical theory in Duhem's sense, one that summarizes the
laws and logically classifies them; only now we remain neutral
as to whether we are also called upon to explain in the
deeper sense of stripping away appearances. This is the
general kind of account I have been supposing throughout
this essay.
There is no doubt that we can explain in this sense.
8 Pierre Duhem, The Aim and Structure of Physical Theory, op. cit., p. 7.
Physical theories abound, and we do not have to look to the

future completion of science to argue that they are fairly
successful at summarizing and organizing; that is what they
patently do now. But this minimal, and non-question-begging,
sense of explanation does not meet van Fraassen's challenge.
There is nothing about successful organization that requires
truth. The stripped down characterization will not do. We
need the full paraphernalia of the D-N account to get the
necessary connection between truth and explanation. But
going beyond the stripped down view to the full metaphysics
involved in a D-N account is just the issue in question.
There is still more to Laudan's criticism. Laudan himself
has written a beautiful piece against inference to the best
explanation." The crux of his argument is this: it is a poor
form of inference that repeatedly generates false conclusions.
He remarks on case after case in the history of science where
we now know our best explanations were false. Laudan
argues that this problem plagues theoretical laws and theo-
retical entities equally. Of my view he says,
What I want to know, is what epistemic difference there is between the
evidence we can have for a theoretical law (which you admit to be
non-robust) and the evidence we can have for a theoretical entity-
such that we are warranted in concluding that, say electrons and
protons exist, but that we are not entitled to conclude that theoretical
laws are probably true. It seems to me that the two are probably on
an equal footing epistemically.
Laudan's favourite example is the electromagnetic aether,
which 'had all sorts of independent sources of support for
it collected over a century and a half'. He asks, 'Did the
enviable successes of one- and two-fluid theories of electricity
show that there really was an electrical fluid?,lO
I have two remarks, the first very brief. Although the
electromagnetic aether is one striking example, I think these
cases are much rarer than Laundan does. So we have a
historical dispute. The second remark bears on the first. I
have been arguing that we must be committed to the exist-
ence of the cause if we are to accept a given causal account.
9 Larry Laudan, 'A Confutation of Convergent Realism'. Philosophy of Science
48 (March 1981), pp. 19-49.
10 In correspondence referred to in footnote 5.
The same is not true for counting a theoretical explanation

good. The two claims get intertwined when we address
the nontrivial and difficult question, when do we have
reasonable grounds for counting a causal account accept-
able? The fact that the causal hypotheses are part of a
generally satisfactory explanatory theory is not enough,
since success at organizing, predicting, and classifying is
never an argument for truth. Here, as I have been stressing,
the idea of direct experimental testing is crucial. Consider
the example of the laser company, Spectra Physics, men-
tioned in the Introduction to these essays. Engineers at
Spectra Physics construct their lasers with the aid of the
quantum theory of radiation, non-linear optics, and the
like; and they calculate their performance characteristics.
But that will not satisfy their customers. To guarantee
that they will get the effects they claim, they use up a
quarter of a million dollars' worth of lasers every few months
in test runs.
I think there is no general theory, other than Mill's
methods, for what we are doing in experimental testing;
we manipulate the cause and look to see if the effects change
in the appropriate manner. For specific causal claims there
are different detailed methodologies. Ian Hacking, in 'Experi-
mentation and Scientific Realism', gives a long example of
the use of Stanford's Peggy II to test for parity violations
in weak neutral currents. There he makes a striking claim:
The experimentalist does not believe in electrons because, in the words
retrieved from medieval science by Duhem, they 'save the phenomena'.
On the contrary, we believe in them because we use them to create
new phenomena, such as the phenomenon of parity violation in weak
neutral current interactions. 11
I agree with Hacking that when we can manipulate our theo-

retical entities in fine and detailed ways to intervene in other
processes, then we have the best evidence possible for our
claims about what they can and cannot do; and theoretical
entities that have been warranted by well-tested causal claims
like that are seldom discarded in the progress of science.
11 See Ian Hacking, 'Experimentation and Scientific Realism', Philosophical
Topics (forthcoming).
4. CONCLUSION
I believe in theoretical entities. But not in theoretical laws.

Often when I have tried to explain my views on theoretical
laws, I have met with a standard realist response: 'How
could a law explain if it weren't true?' Van Fraassen and
Duhem teach us to retort, 'How could it explain if it were
true?' What is it about explanation that guarantees truth?
I think there is no plausible answer to this question when
one law explains another. But when we reason about theo-
retical entities the situation is different. The reasoning is
causal, and to accept the explanation is to admit the cause.
There is water in the barrel of my lemon tree, or I have
no explanation for its ailment, and if there are no electrons
in the cloud chamber, I do not know why the tracks are there.
Essay 6
For Phenomenological Laws
O. INTRODUCTION
A long tradition distinguishes fundamental from phenomeno-
logical laws, and favours the fundamental. Fundamental laws
are true in themselves; phenomenological laws hold only on
account of more fundamental ones. This view embodies an
extreme realism about the fundamental laws of basic explana-
tory theories. Not only are they true (or would be if we had
the right ones), but they are, in a sense, more true than the
phenomenological laws that they explain. I urge just the
reverse. I do so not because the fundamental laws are about
unobservable entities and processes, but rather because of
the nature of theoretical explanation itself. As I have often
urged in earlier essays, like Pierre Duhem, I think that the
basic laws and equations of our fundamental theories organ-
ize and classify our knowledge in an elegant and efficient
manner, a manner that allows us to make very precise calcula-
tions and predictions. The great explanatory and predictive
powers of our theories lies in their fundamental laws. Never-
theless the content of our scientific knowledge is expressed
in the phenomenological laws.
Suppose that some fundamental laws are used to explain
a phenomenological law. The ultra-realist thinks that the
phenomenological law is true because of the more funda-
mental laws. One elementary account of this is that the
fundamental laws make the phenomenological laws true.
The truth of the phenomenological laws derives from the
truth of the fundamental laws in a quite literal sense-
something like a causal relation exists between them. This is
the view of the seventeenth-century mechanical philosophy
of Robert Boyle and Robert Hooke. When God wrote the
Book of Nature, he inscribed the fundamental laws of
mechanics and he laid down the initial distribution of matter
in the universe. Whatever phenomenological laws would be
true fell out as a consequence. But this is not only the view of
FOR PHENOMENOLOGICAL LAWS 101
the seventeenth-century mechanical philosophy. It is a view

that lies at the heart of a lot of current-day philosophy of
science-particularly certain kinds of reductionism-and
I think it is in part responsible for the widespread appeal of
the deductive-nomological model of explanation, though
certainly it is not a view that the original proponents of the
D-N model, such as Hempel, and Grunbaum, and Nagel,
would ever 'have considered. I used to hold something like
this view myself, and I used it in my classes to help students
adopt the D-N model. I tried to explain the view with two
stories of creation.
Imagine that God is about to write the Book of Nature
with Saint Peter as his assistant. He might proceed in the way
that the mechanical philosophy supposed. He himself decided
what the fundamental laws of mechanics were to be and how
matter was to be distributed in space. Then he left to Saint
Peter the laborious but unimaginative task of calculating
what phenomenological laws would evolve in such a universe.
This is a story that gives content to the reductionist view that
the laws of mechanics are fundamental and all the rest are
epi-phenomenal.
On the other hand, God may have had a special concern
for what regularities would obtain in nature. There were to
be no distinctions among laws: God himself would dictate
each and everyone of them-not only the laws of mechanics,
but also the laws of chemical bonding, of cell physiology, of
small group interactions, and so on. In this second story Saint
Peter's task is far more demanding. To Saint Peter was left
the difficult and delicate job of finding some possible
arrangement of matter at the start that would allow all the
different laws to work together throughout history without
inconsistency. On this account all the laws are true at once,
and none are more fundamental than the rest.
The different roles of God and Saint Peter are essential
here: they make sense of the idea that, among a whole collec-
tion of laws everyone of which is supposed to be true, some
are more basic or more true than others. For the seventeenth-
century mechanical philosophy, God and the Book of Nature
were legitimate devices for thinking of laws and the relations
among them. But for most of us nowadays these stories are
102 FOR PHENOMENOLOGICAL LAWS
mere metaphors. For a long time I used the metaphors, and

hunted for some non-metaphorical analyses. I now think that
it cannot be done. Without God and the Book of Nature
there is no sense to be made of the idea that one law derives
from another in nature, that the fundamental laws are basic
and that the others hold literally 'on account of the funda-
mental ones.
Here the D-N model of explanation might seem to help.
In order to explain and defend our reductionist views, we
look for some quasi-causal relations among laws in nature.
When we fail to find any reasonable way to characterize these
relations in nature, we transfer our attention to language.
The deductive relations that are supposed to hold between
the laws of a scientific explanation act as a formal-mode
stand-in for the causal relations we fail to find in the material
world. But the D-N model itself is no argument for realism
once we have stripped away the questionable metaphysics.
So long as we think that deductive relations among statements
of law mirror the order of responsibility among the laws
themselves, we can see why explanatory success should argue
for the truth of the explaining laws. Without the metaphysics,
the fact that a handful of elegant equations can organize a
lot of complex information about a host of phenomenological
laws is no argument for the truth of those equations. As
I urged in the last essay, we need some story about what the
connection between the fundamental equations and the more
complicated laws is supposed to be. There we saw that Adolf
Grunbaum has outlined such a story. His outline I think
coincides with the views of many contemporary realists.
Grunbaum's view eschews metaphysics and should be accept-
able to any modern empiricist. Recall that GrUnbaum says:
It is crucial to realize that while (a more comprehensive law) G entails

(a less comprehensive law) L logically, thereby providing an explanation
of L, G is not the 'cause' of L. More specifically, laws are explained not
by showing the regularities they affirm to be products of the operation
of causes but rather by recognizing their truth to be special cases of
more comprehensive truths. 1
I A. Grunbaum, 'Science and Ideology', The Scientific Monthly (July 1954),

p. 14, italics in original.
I call this kind of account of the relationship between

fundamental and phenomenological laws a generic-specific
account. It holds that in any particular set of circumstances
the fundamental explanatory laws and the phenomenological
laws that they explain both make the same claims. Phe-
nomenological laws are what the fundamental laws amount
to in the circumstances at hand. But the fundamental laws
are superior because they state the facts in a more general
way so as to make claims about a variety of different circum-
stances as well.
The generic-specific account is nicely supported by the
deductive-nomological model of explanation: when funda-
mental laws explain a phenomenological law, the phenomeno-
logical law is deduced from the more fundamental in
conjunction with a description of the circumstances in which
the phenomenological law obtains. The deduction shows
just what claims the fundamental laws make in the circum-
stances described.
But explanations are seldom in fact deductive, so the
generic-specific account gains little support from actual
explanatory practice. Wesley Salmon? and Richard Jeffrey,"
and now many others, have argued persuasively that explana-
tions are not arguments. But their views seem to bear most
directly on the explanations of single events; and many philo-
sophers still expect that the kind of explanations that we are
concerned with here, where one law is derived from others
more fundamental, will still follow the D-N form. One reason
that the D-N account often seems adequate for these cases
is that it starts looking at explanations only after a lot of
scientific work has already been done. It ignores the fact
that explanations in physics generally begin with a model.
The calculation of the small signals properties of amplifiers,
which I discuss in the next section, is an example." We first
2 See Wesley Salmon, 'Statistical Explanation', in Wesley Salmon (ed.),

3 See R. C. Jeffrey, 'Statistical Explanation vs, Statistical Inference', in
Wesley Salmon, op. cit.
4 This example is taken from my joint paper with Jon Nordby, 'How Approxi-
mations Take Us Away from Theory and Towards the Truth' (unpublished manu-
script: Stanford University and Pacific Lutheran University).
decide which model to use-perhaps the T-model, perhaps

the hybrid-a model. Only then can we write down the equa-
tions with which we will begin our derivation.
Which model is the right one? Each has certain advantages
and disadvantages. The T-model approach to calculating the
midband properties of the CE stage is direct and simple, but
if we need to know how the CE stage changes when the bias
conditions change, we would need to know how all the para-
meters in the transistor circuit vary with bias. It is incredibly
difficult to produce these results in a T-model. The T-model
also lacks generality, for it requires a new analysis for each
change in configuration, whereas the hybrid-s model approach
is most useful in the systematic analysis of networks. This is
generally the situation when we have to bring theory to bear
on a real physical system like an amplifier. For different
purposes, different models with different incompatible laws
are best, and there is no single model which just suits the
circumstances. The facts of the situation do not pick out one
right model to use.
I will discuss models at length in the next few essays.
Here I want to lay aside my worries about models, and think
about how derivations proceed once a model has been
chosen. Proponents of the D-N view tend to think that at
least then the generic-specific account holds good. But this
view is patently mistaken when one looks at real derivations
in physics or engineering. It is never strict deduction that
takes you from the fundamental equations at the beginning
to the phenomenological laws at the end. Instead we require
a variety of different approximations. In any field of physics
there are at most a handful of rigorous solutions, and those
usually for highly artificial situations. Engineering is worse.
Proponents of the generic-specific account are apt to think
that the use of approximations is no real objection to their
view. They have a story to tell about approximations: the
process of deriving an approximate solution parallels a D-N
explanation. One begins with some general equations which
we hold to be exact, and a description of the situation
to which they apply. Often it is difficult, if not impossible,
to solve these equations rigorously, so we rely on our descrip-
tion of the situation to suggest approximating procedures.
In these cases the approximate solution is just a stand-in.

We suppose that the rigorous solution gives the better results;
but because of the calculational difficulties, we must satisfy
ourselves with some approximation to it.
Sometimes this story is not far off. Approximations
occasionally work just like this. Consider, for example, the
equation used to determine the equivalent air speed, VE,
of a plane (where PT is total pressure, Po is ambient pressure,
PS is sea level density, and M is a Mach number):
VE ~ [2~Tp~ PJX (I + M2/~ + M'/40)r (6.1)
The value for VE determined by this equation is close to the

plane's true speed.
Approximations occur in connection with (6.1) in two
distinct but typical ways. First, the second term,
1
2
1 +M /4 +M4/40
is significant only as the speed of the plane approaches Mach
One. If M < 0.5, the second term is discarded because the
result for V E
given by [2tT
~ Po)II
will differ insignificantly from the result given by (6.1).
Given this insignificant variation, we can approximate VE by
using
(6.2)
for M < 0.5. Secondly, (6.1) is already an approximation,

and not an exact equation. The term
1
1 + M /4 + M4/40
2
has other terms in the denominator. It is a Taylor series

expansion. The next term in the expansion is M6/1600,
so we get
1 +M2/4 +M4/40 +M6/1600

in the denominator, and so on. For Mach numbers less than
one, the error that results from ignoring this third term is
less than one per cent, so we truncate and use only two
terms. .
Why does the plane travel with a velocity, V, roughly
equal to
Because of equation (6.1). In fact the plane is really travel-

ling at a speed equal to
I~ ) ] 1
PT - Po 1 2"
2 Ps X 1 +M2/4 +M4/40 + ...

But since M is less than 0.5, we do not notice the difference.
Here the derivation of the plane's speed parallels a covering
law account. We assume that equation (6.1) is a true law that
covers the situation the plane is in. Each step away from
equation (6.1) takes us a little further from the true speed.
But each step we take is justified by the facts of the case, and
if we are careful, we will not go too far wrong. The final
result will be close enough to the true answer.
This is a neat picture, but it is not all that typical. Most
cases abound with problems for the generic-specific account.
Two seem to me especially damaging: (l) practical approxi-
mations usually improve on the accuracy of our fundamental
laws. Generally the doctored results are far more accurate
than the rigorous outcomes which are strictly implied by the
laws with which we begin. (2) On the generic-specific account
the steps of the derivation are supposed to show how the
fundamental laws make the same claims as the phenomeno-
logical laws, given the facts of the situation. But seldom are
the facts enough to justify the derivation. Where approxi-
mations are called for, even a complete knowledge of the
circumstances may not provide the additional premises
necessary to deduce the phenomenological laws from the
fundamental equations that explain them. Choices must be
made which are not dictated by the facts. I have already
mentioned that this is so with the choice of models. But it is
also the case with approximation procedures: the choice is
constrained, but not dictated by the facts, and different
choices give rise to different, incompatible results. The
generic-specific account fails because the content of the
phenomenological laws we derive is not contained in the
fundamental laws which explain them.
These two problems are taken up in tum in the next two
sections. A lot of the argumentation, especially in Section 2,
is taken from a paper written jointly by Jon Nordby and me,
'How Approximations Take Us Away from Theory and
Towards the Truth'. 5 This paper also owes a debt to Nordby's
'Two Kinds of Approximations in the Practice of Science'."
1. APPROXIMATIONS THAT IMPROVE ON LAWS

On the generic-specific account, any approximation detracts
from truth. But it is hard to find examples of this at the level
where approximations connect theory with reality, and that
is where the generic-specific account must work if it is to
ensure that the fundamental laws are true in the real world.
Generally at this level approximations take us away from
theory and each step away from theory moves closer towards
the truth. I illustrate with two examples from the joint paper
with Jon Nordby.
1.1. An Amplifier Model

Consider an amplifier constructed according to Figure 6.1.
As I mentioned earlier, there are two ways to calculate the
small signal properties of this amplifier, the T-model and the
hybrid-s model. The first substitutes a circuit model for the
transistor and analyses the resulting network. The second
characterizes the transistor as a set of two-port parameters
and calculates the small signal properties of the amplifier
5 N. Cartwright and J. Nordby, 'How Approximations Take Us Away from

Theory and Towards the Truth' (unpublished manuscript: Stanford University
and Pacific Lutheran University).
6 See Jon Nordby, 'Two Kinds of Approximation in the Practice of Science'
(unpublished manuscript: Pacific Lutheran University).
FIG.6.1
------------------1
I ale
I I
I -'----I
11
I 'b I 12
Rs- I I~
I I +
I I
+
II r, t Ie IRLV2
I
I 'c I
I
L ~ I
FIG. 6.2
1-----------------\
I, I
R s -----. rb r2 I-4--12
+
+
VJr 'n
I I
~-------------------
FIG. 6.3
in terms of these parameters. The two models are shown in

Figures 6.2 and 6.3 inside the dotted area.
The application of these transistor models in specific
situations gives a first rough approximation of transistor
parameters at low frequencies. These parameters can be
theoretically estimated without having to make any measure-
ments on the actual circuit. But the theoretical estimates are
often grossly inaccurate due to specific causal features
present in the actual circuit, but missing from the models.
One can imagine handling this problem by constructing a

larger, more complex model that includes the missing causal
features. But such a model would have to be highly specific
to the circuit in question and would thus have no general
applicability.
Instead a different procedure is followed. Measurements
of the relevant parameters are made on the actual circuit
under study, and then the measured values rather than
the theoretically predicted values are used for further
calculations in the original models. To illustrate, consider
an actual amplifier, built and tested such that IE = 1 rna;
RL = 2.71115 = 2.3 k ohm; R1 = 1 k ohm. (3 = 162 and
Rs = 1 k ohm and assume rb 50 ohms. The theoretical
expectation of midband gain is
RL 2.3 kohms
IAvl = =- = 72. (6.3)
re + (rb + Rs)(l - a) 32 k ohms
The actual measured midband gain for this amplifier with an
output voltage at f = 2 kHz and source voltage at 1.8 mv and
2 kHz, is
80mv
Av, meas = = 44. (6.4)
1.8 mv
This result is not even close to what theory predicts. This
fact is explained causally by considering two features of the
situation: first, the inaccuracy of the transistor model due to
some undiagnosed combination of causal factors; and second,
a specifically diagnosed omission-the omission of equi-
valent series resistance in the bypass capacitor. The first
inaccuracy involves the value assigned to re by theory. In
theory, re = kT/qIE, which is approximately 25.9/Je. The
actual measurements indicate that the constant of propor-
tionality for this type of transistor is 30 mv, so r; = 30/IE,
not 25.9/Je.
Secondly, the series resistance is omitted from the ideal
capacitor. But real electrolytic capacitors are not ideal. There
is leakage of current in the electrolyte, and this can be
modelled by a resistance in series with the capacitor. This
series resistance is often between 1 and 10 ohms, sometimes
as high as 25 ohms. It is fairly constant at low frequency but
increases with increased frequency and also with increased

temperature. In this specific case, the series resistance, rc , I
is measured as 12 ohms.
We must now modify (6.3) to account for these features,
given our measured result in (6.4) of Av, meas. = 44:
RL
IAvl = (6.5)
re + (rb + Rs)(l - a) + rc 1
Solving this equation gives us a predicted midband gain of

47.5, which is sufficiently close to the measured midband
gain for most purposes.
Let us now look back to see that this procedure is very
different from what the generic-specific account supposes.
We start with a general abstract equation, (6.3); make some
approximations; and end with (6.5), which gives rise to de-
tailed phenomenological predictions. Thus superficially it
may look like a D-N explanation of the facts predicted.
But unlike the covering laws of D-N explanations, (6.3) as
it stands is not an equation that really describes the circuits
to which it is applied. (6.3) is refined by accounting for the
specific causal features of each individual situation to form
an equation like (6.5). (6.5) gives rise to accurate predictions,
whereas a rigorous solution to (6.3) would be dramatically
mistaken.
But one might object: isn't the circuit model, with no
resistance added in, just an idealization? And what harm is
that? I agree that the circuit model is a very good example of
one thing we typically mean by the term idealization. But
how can that help the defender of fundamental laws? Most
philosophers have made their peace with idealizations: after
all, we have been using them in mathematical physics for well
over two thousand years. Aristotle in proving that the
rainbow is no greater than a semi-circle in Meterologica
III. 5 not only treats the sun as a point, but in a blatant
falsehood puts the sun and the reflecting medium (and hence
the rainbow itself) the same distance from the observer.
Today we still make the same kinds of idealizations in our
celestial theories." Nevertheless, we have managed to discover
7 See Hilary Putnam, 'The "Corroboration" of Theories', Philosophical Papers,
Vol. 1 (Cambridge: Cambridge University Press, 1975) for a nice discussion of this.
the planet Neptune, and to keep our satellites in space.

Idealizations are no threat to the progress of science.
But what solace is this to the realist? How do idealiza-
tions save the truth of the fundamental laws? The idea
seems to be this. To call a model an idealization is to suggest
that the model is a simplification of what occurs in reality,
usually a simplification which omits some relevant features,
such as the extended mass of the planets or, in the example
of the circuit model, the resistance in the bypass capacitor.
Sometimes the omitted factors make only an insignificant
contribution to the effect under study. But that does not
seem to be essential to idealizations, especially to the idealiza-
tions that in the end are applied by engineers to study real
things. In calling something an idealization it seems not so
important that the contributions from omitted factors be
small, but that they be ones for which we know how to
correct. If the idealization is to be of use, when the time
comes to apply it to a real system we had better know how
to add back the contributions of the factors that have been
left out. In that case the use of idealizations does not seem to
counter realism: either the omitted factors do not matter
much, or in principle we know how to treat them.
In the sense I just described, the circuit model is patently
an idealization. We begin with equation (6.3), which is in-
adequate; we know the account can be improved-Nordby
and I show how. But the improvements come at the wrong
place for the defender of fundamental laws. They come from
the ground up, so-to-speak, and not from the top down. We
do not modify the treatment by deriving from our theoretical
principles a new starting equation to replace (6.3). It is clear
that we could not do so, since only part of the fault is diag-
nosed. What we do instead is to add a phenomenological
correction factor, a factor that helps produce a correct
description, but that is not dictated by fundamental law.
But could we not 'in principle' make the corrections right
at the start, and write down a more accurate equation from
the beginning? That is just the assumption I challenge. Even
if we could, why do we think that by going further and
further backwards, trying to get an equation that will be right
when all the significant factors are included, we will eventually
get something simple which looks like one of the fundamental

laws of our basic theories? Recall the discussion of cross-
effects from Essay 3. There I urged that we usually do not
have any uniform procedure for 'adding' interactions. When
we try to write down the 'more correct' equations, we get
a longer and longer list of complicated laws of different
forms, and not the handful of simple equations which could
be fundamental in a physical theory.
Generality and simplicity are the substance of explanation.
But they are also crucial to application. In engineering, one
wants laws with a reasonably wide scope, models that can be
used first in one place then another. If I am right, a law that
actually covered any specific case, without much change or
correction, would be so specific that it would not be likely to
work anywhere else. Recall Bertrand Russell's objection to
the 'same cause, same effect' principle:
The principle 'same cause, same effect,' which philosophers imagine

to be vital to science, is therefore utterly otiose. As soon as the ante-
cedents have been given sufficiently fully to enable the consequents
to be calculated with some exactitude, the antecedents have become
so complicated that it is very unlikely they will ever recur. Hence,
if this were the principle involved, science would remain utterly
sterile."
Russell's solution is to move to functional laws which state

relations between properties (rather than relations between
individuals). But the move does not work if we want to
treat real, complex situations with precision. Engineers
are comfortable with functions. Still they do not seem
able to find functional laws that allow them to calculate
consequences 'with some exactitude' and yet are not 'so
complicated that it is very unlikely they will ever recur'. To
find simple laws that we can use again and again, it looks
as if we had better settle for laws that patently need im-
provement. Following Russell, it seems that if we model
approximation on D-N explanation, engineering 'would
remain utterly sterile'.
8 Bertrand Russell, 'On the Notion of Cause with Application to the Problem
of Free Will', in H. Feigl and M. Brodbeck (eds), Readings in Philosophy of
Science (New York: Appleton-Century-Crofts, 1953), p. 392.
1.2. Exponential Decay

The second example concerns the derivation of the exponen-
tial decay law in quantum mechanics. I will describe this
derivation in detail, but the point I want to stress can be
summarized by quoting one of the best standard texts, by
Eugen Merzbacher: 'The fact remains that the exponential
decay law, for which we have so much empirical support in
radioactive processes, is not a rigorous consequence of quan-
tum mechanics but the result of somewhat delicate approxi-
mations."
The exponential decay law is a simple, probabilistically
elegant law, for which-as Merzbacher says-we have a
wealth of experimental support. Yet it cannot be derived
exactly in the quantum theory. The exponential law can only
be derived by making some significant approximation. In the
conventional treatment the rigorous solution is not pure
exponential, but includes several additional terms. These
additional terms are supposed to be small, and the difference
between the rigorous and the approximate solution will be
unobservable for any realistic time periods. The fact remains
that the data, together with any reasonable criterion of
simplicity (and some such criterion must be assumed if we
are to generalize from data to laws at all) speak for the truth
of an exponential law; but such a law cannot be derived
rigorously. Thus it seems that the approximations we make
in the derivation take us closer to, not further from, the
truth.
There are two standard treatments of exponential decay:
the Weisskopf-Wigner treatment, which was developed in
their Classic paper of 1930,10 and the more recent Markov
treatment, which sees the exponential decay of an excited
atom as a special case in the quantum theory of damping.
We will look at the more recent treatment first. Here we
consider an abstract system weakly coupled to a reservoir.
The aim is to derive a general master equation for the
9 Eugen Merzbacher, Quantum Mechanics (New York: John Wiley & Sons,
1970), pp. 484-5.
10 V. Weisskopf and E. Wigner, 'Die Rechnung der naturlichen Linienbreite
auf Grund der Diracschen Lichttheorie', Zeitschrift fiir Physik 63 (1930),
pp.54-73.
evolution of the system. This equation is similar to the evolu-

tion equations of classical statistical mechanics. For the
specific case in which we are interested, where the system is
an excited atom and the reservoir the electromagnetic field,
the master equation turns into the Pauli rate equation, which
is the analogue of the exponential law when re-excitement
may occur:
Pauli equation:
as,
- = -r.s. + I W1"kSk'
at I I
k'/"i
(Here Sj is the occupation probability of the jth state; rj,

the inverse of the lifetime; and Wjk is the transition prob-
ability from state k to state j.)
The derivation of the master equation is quite involved.
I will focus on the critical feature from my point of view-
the Markov approximation. Generally such a derivation
begins- with the standard second-order perturbation expan-
sion for the state x of the composite, system and reservoir,
which in the interaction picture looks like this:
1 t
f
I I
x(t) = x(to) +:- [V(t - to), x(to)] dt +
ih to
2 ,
I t,t /I I /I
+ ( ih )
Jo dt Jo dt [V(t - to), [V(t - to), x(to)] ].
Notice that the state of the system and reservoir at t depends

on its entire past history through the integrals on the right-
hand side of this equation. The point of the Markov approxi-
mation is to derive a differential equation for the state of the
system alone such that the change in this state at a time
depends only on the facts about the system at that time,
and not on its past history. This is typically accomplished by
two moves: (i) extending the time integrals which involve
only reservoir correlations to infinity, on the grounds that
the correlations in the reservoir are significant for only a
short period compared to the periods over which we are ob-
serving the system; and (ii) letting t - to --+ 0, on the grounds
that the periods of time considered for the system are small
compared to its lifetime. The consequence is a master equation
with the desired feature. As W. H. Louisell remarks in his

chapter on damping:
We note that the r.h.s. [right hand side] of [the master equation] no
longer contains time integrals over Set') [S is the state of the system
alone] for times earlier than the present so that the future is now
indeed determined by the present. We have assumed that the reservoir
correlation times are zero on a time scale in which the system loses
an appreciable amount of its energy ... One sometimes refers to the
Markoff approximation as a coarse-grained averaging. 11
Thus the Markov approximation gives rise to the master

equation; for an atom in interaction with the electromagnetic
field, the master equation specializes to the Pauli equation;
and the Pauli equation predicts exponential decay for the
atom. Without the Markov approximation, the decay can at
best be near exponential.
Let us now look at the Weisskopf-Wigner method, as it is
employed nowadays. We begin with the exact Schroedinger
equations for the amplitudes, but assume that the only sig-
nificant coupling is between the excited and the de-excited
states:
-de, -_" L gef

2 f t ,» .
dt exp{l(weg
-
Wf)(t -
t ' )}Ce(t , )
dt f 0
(Weg is E; - Eg/h, for E; the energy of the excited state, Eg

the energy of the de-excited; wf is the frequency of the fth
mode of the field; and gef is the coupling constant between
the excited state and the fth mode. Ce is the amplitude in the
excited state, no photons present.)
The first approximation notes that the modes of the field
available to the de-exciting atom form a near continuum. (We
will learn more about this in the next section.) So the sum
over f can be replaced by an integral, to give
de,
- =
f+oo r s 2 (w)fZl(w)dw f t ,
dt exp{i(weg - w)(t-
,
t)}cit).
,
dt --00 0
We could now pull out the terms which are slowly varying
in w, and do the W integral, giving a delta function in t:
11 William H. Louisell, Quantum Statistical Properties of Radiation (New
York: John Wiley & Sons, 1973), p. 341.
x L:oo exp{i(weg - w)(t - t')}dw
= g2(Weg)9&(Weg)1T f: l /
dt ceCt )5(t - t').
or setting 'Y== 21Tg2(Weg)9&(Weg),
dCe 'Y
-=--c (t)
dt 2 e
and finally,
ceCt) = exp(-'Y/2t).
But in moving so quickly we lose the Lamb shift-a small
displacement in the energy levels discovered by Willis Lamb
and R. C. Retherford in 1947. It will pay to do the integrals
in the opposite order. In this case, we note that ce(t) is itself
slowly varying compared to the rapid oscillations from the
exponential, and so it can be factored out of the t' integral,
and the upper limit of that integral can be extended to
infinity. Notice that the extension of the t' limit is very
similar to the Markov approximation already described, and
the rationale is similar. We get
de,
- =
J +00
g2 (w)9&(w)ce(t)exp {i(weg - w)t}dw X
dt -00
oo
X Jo exp{ -i(weg - w)t'}dt'
+OO
=
J -00 g2 (w)9&(w)ce (t)exp{i(weg - w)t} dw X
(.gil(X) = principal part of x.)
de,
dt =-
('Y2.+ i~WJCe(t).
.\
Here ~w is the Lamb shift. The second method, which
results in a Lamb shift as well as the line-broadening 'Y,is what
is usually now called 'the Weisskopf-Wigner' method.
We can try to be more formal and avoid approximation
altogether. The obvious way to proceed is to evaluate the
Laplace transform, which turns out to be
then
ce(t) = -.
1 f€+i~ exp(M)ce(-6)d-6.
21Tl €-i~
To solve this equation, the integrand must be defined on the

first and second Riemann sheets. The method is described
clearly in Goldberger and Watson's text on collision theory.P
The primary contribution will come from a simple pole
~€ such that
--=
i
M€
-'Y.
2
+ 1~W = l'Hfl
-6-->0+
"L
t weg -
get.
-
wf + 1-6
This term will give us the exponential we want:
c.(t) ~ expl-(~ iLlW)+ +

But this is not an exact solution. As we distort the contour
on the Riemann sheets, we cross other poles which we have
not yet considered. We have also neglected the integral
around the final contour itself. Goldberger and Watson cal-
culate that this last integral contributes a term proportional
to h312/t3I2Ee. They expect that the other poles will add only
12 See Marvin L. Goldberger and Kenneth M. Watson, Collision Theory (New
York: John Wiley & Sons, 1964), Chapter 8.
negligible contributions as well, so that the exact answer

will be a close approximation to the exponential law we
are seeking; a close approximation, but still only an approxi-
mation. If a pure exponential law is to be derived, we had
better take our approximations as improvements on the
initial Schroedinger equation, and not departures from
the truth.
Is there no experimental test that tells which is right? Is
decay really exponential, or is the theory correct in predict-
ing departures from the exponential law when t gets large?
There was a rash of tests of the experimental decay law in the
middle 1970s, spurred by E. T. Jaynes's 'neo-classical' theory
of the interaction of matter with the electromagnetic field,
a theory that did surprisingly well at treating what before had
been thought to be pure quantum phenomena. But these
tests were primarily concerned with J aynes's claim that decay
rates would depend on the occupation level of the initial
state. The experiments had no bearing on the question of
very long time decay behaviour. This is indeed a very difficult
question to test. Rolf Winter l''
has experimented on the
decay of Mn56 up to 34 half-lives, and D. K. Butt and A. R.
Wilson on the alpha decay of radon for over 40 half-lives.I"
But, as Winter remarks, these lengths of time, which for
ordinary purposes are quite long, are not relevant to the
differences I have been discussing, since 'for the radioactive
decay of Mn56 ••• non-exponential effects should not occur
before Toughly 200 half-lives. In this example, as with all the
usual radioactive decay materials, nothing observable should
be left long before the end of the exponential region' ,is In
short, as we read in a 1977 review article by A. Pais, 'experi-
mental situations in which such deviations playa role have
not been found to date.'!" The times before the differences
emerge are just too long.
13 See Rolf Winter, 'Large- Time Exponential Decay and "Hidden Variables" "
Physical Review 126 (1962), pp. 1152-3.
14 See D. K. Butt and A. R. Wilson, 'A Study of the Radioactive Decay Law',
Journal of Physics A: General Physics 5 (1972), pp. 1248-51.

IS Rolf Winter, op. cit., p. 1152.
16 A. Pais, 'Radioactivity's Two Early Puzzles', Reviews of Modern Physics
49 (1977), p. 936.
2. APPROXIMA nONS NOT DICTATED

BY THE FACTS
Again, I will illustrate with two examples. The examples
show how the correct approximation procedure can be un-
determined by the facts. Both are cases in which the very
same procedure, justified by exactly the same factual claims,
gives different results depending on when we apply it: the
same approximation applied at different points in the deriva-
tion yields two different incompatible predictions. I think
this is typical of derivation throughout physics; but in order
to avoid spending too much space laying out the details of
different cases, I will illustrate with two related phenomena:
(a) the Lamb shift in the excited state of a single two-level
atom; and (b) the Lamb shift in the ground state of the atom.
2.1. The Lamb Shift in the Excited State

Consider again spontaneous emission from a two-level atom.
The traditional way of treating exponential decay derives
from the classic paper of V. Weisskopf and Eugene Wigner in
1930, which I described in the previous section. In its present
form the Weisskopf-Wigner method makes three important
approximations: (l) the rotating wave approximation; (2) the
replacement of a sum by an integral over the modes of the
electromagnetic field and the factoring out of terms that vary
slowly in the frequency; and (3) factoring out a slowly-varying
term from the time integral and extending the limit on the
integral to infinity. I will discuss the first approximation
below when we come to consider the level shift in the ground
state. The second and third are familiar from the last section.
Here I want to concentrate on how they affect the Lamb
shift in the excited state.
Both approximations are justified by appealing to the
physical characteristics of the atom-field pair. The second
approximation is reasonable because the modes of the field
are supposed to form a near continuum; that is, there is a
very large number of very closely spaced modes. This allows
us to replace the sum by an integral. The integral is over a
product of the coupling constant as a function of the fre-
quency, w, and a term of the form exp(-iwt). The coupling
constant depends on the interaction potential for the atom

and the field, and it is supposed to be relatively constant in
w compared to the rapidly oscillating exponential. Hence
it can be factored outside the integral with little loss of
accuracy. The third approximation is similarly justified by
the circumstances.
What is important is that, although each procedure is
separately rationalized by appealing to facts about the atom
and the field, it makes a difference in what order they are
applied. This is just what we saw in the last section. If we
begin with the third approximation, and perform the t
integral before we use the second approximation to evaluate
the sum over the modes, we predict a Lamb shift in the
excited state. If we do the approximations and take the
integrals in the reverse order-which is essentially what
.Weisskopf and Wigner did in their original paper-we lose
the Lamb shift. The facts that we cite justify both pro-
cedures, but the facts do not tell us in what order to apply
them. There is nothing about the physical situation that
indicates which order is right other than the fact to be
derived: a Lamb shift is observed, so we had best do first (3),
then (2). Given all the facts with which we start about the
atom and the field and about their interactions, the Lamb
shift for the excited state fits the fundamental quantum
equations. But we do not derive it from them.
One may object that we are not really using the same
approximation in different orders; for, applied at different
points, the same technique does not produce the same
approximation. True, the coefficients in t and in ware slowly
varying, and factoring them out of the integrals results in
only a small error. But the exact size of the error depends on
the order in which the integrations are taken. The order that
takes first the t integral and then the w integral is clearly
preferable, because it reduces the error.
Two remarks are to be made about this objection. Both
have to do with how approximations work in practice. First,
in this case it looks practicable to try to calculate and to
compare the amounts of error introduced by the two
approximations. But often it is practically impossible to
decide which of two procedures will lead to more accurate
results. For instance, we often justify dropping terms from

an equation by showing that the coefficients of the omitted
terms are small compared to those of the terms we retain.
But as the next example will show, knowing the relative sizes
of terms in the equation is not a sure guide to the exact
effects in the solution, particularly when the approximation
is embedded in a series of other approximations. This is just
one simple case. The problem is widespread. As I argued in
Essay 4, proliferation of treatments is the norm in physics,
and very often nobody knows exactly how they compare.
When the situation becomes bad enough, whole books may
be devoted to sorting it out. Here is just one example, The
Theory of Charge Exchange by Robert Mapleton. The
primary purpose of the book is to explain approximating
methods for cross sections and probabilities for electronic
capture. But its secondary purpose is
to compare different approximate predictions with each other and with
experimentally determined values. These comparisons should enable us
to determine which approximating procedures are most successful in
predicting cross sections for different ranges . . . they also should
indicate which methods show most promise for additional improve-
ment.'?
Comparing approximations is often no easy matter.
Secondly, the objection assumes the principle 'the more
accuracy, the better'. But this is frequently not so, for a
variety of well-known reasons: the initial problem is set only
to a given level of accuracy, and any accuracy in the conclu-
sion beyond this level is spurious; or the use of certain
mathematical devices, such as complex numbers, will gener-
ate excess terms which we do not expect to have any physical
significance; and so on. The lesson is this: a finer approxima-
tion provides a better account than a rougher one only if the
rougher approximation is not good enough. In the case at
hand, the finer approximation is now seen to be preferable, not
because it produces a quantitatively slightly more accurate
treatment, but rather because it exposes a qualitatively sig-
nificant new phenomenon-the Lamb shift in the ground
state. If you look back at the equations in the last section
17 Robert Mapleton, The Theory a/Charge Exchange (New York: John Wiley
& Sons, 1972), p. 1.
it is obvious that the first-co-then t order misses an imaginary

term: the amplitude to remain in the excited state has the
form exp(-trt) rather than exp{-(tr + iw)t}. This addi-
tional imaginary term, iw, appears when the integrals are
done in the reverse order, and it is this term that represents
the Lamb shift. But what difference does this term make?
In the most immediate application, for calculating decay
probabilities, it is completely irrelevant, for the prob-
ability is obtained by multiplying together the ampli-
tude exp {-(!I' + iw)t}, and its complex conjugate
exp {-(t I' - iw)t}, in which case the imaginary part dis-
appears and we are left with the well-known probability for
exponential decay, expr-T'r).
The point is borne out historically. The first-co-then-r
order, which loses the Lamb shift, is an equivalent
approximation to the ansatz which Weisskopf and Wigner
use in their paper of 1930, and it was the absolutely conven-
tional treatment for seventeen years. To calculate the value
of the missing imaginary terms, one has to come face to face
with divergences that arise from the Dirac theory of the elec-
tron, and which are now so notorious in quantum electro-
dynamics. These problems were just pushed aside until the
remarkable experiments in 1947 by Willis Lamb and his
student R. C. Retherford, for which Lamb later won the
Nobel prize.
The Dirac theory, taking the spin of the electron into
account, predicted an exact coincidence of the 22 PI 12 and the
228112 levels. There was a suspicion that this prediction
was wrong. New microwave techniques developed during the
war showed Lamb a way to find out, using the metastable
228112 state of hydrogen. In the 1947 experiment the Lamb
shift was discovered and within a month Bethe had figured a
way to deal with the divergences. After the discovery of the
Lamb shift the original Weisskopf-Wigner method had to be
amended. Now we are careful to take the integrals in the
first-r-then-co order. But look at what Bethe himself has
to say:
By very beautiful experiments, Lamb and Retherford have shown that
the fine structure of the second quantum state of hydrogen does not
agree with the prediction of the Dirac theory. The 2s level, which
according to Dirac's theory should coincide with the 2pl/2 level, is
actually higher than the latter by an amount of about 0.033 cm -1 or
1000 megacycles ...
Schwinger and Weisskopf, and Oppenheimer have suggested that a
possible explanation might be the shift of energy levels by the inter-
action of the electron with the radiation field. This shift comes out
infinite in all existing theories, and has therefore always been ignored. 18
Or consider Lamb's comment on Bethe in his Nobel Prize
address:
A month later [after 'the fine structure deviations were definitely
established experimentally' by Lamb and Retherford], Bethe found
that quantum electrodynamics had really hidden behind its divergences
a physical content that was in very close agreement with the micro-
wave observations."
Now we attend to the imaginary terms because they have

real 'physical content ... in very close agreement with micro-
wave observations'. But until Lamb's experiments they were
just mathematical debris which represented nothing of
physical significance and were, correctly, omitted.
2.2. The Lamb Shift in the Ground State

The details of the second example are in G. S. Agarwal's
monograph on spontaneous emission.'? Recall that there are
two common methods for treating spontaneous emission.
The first is the Weisskopf-Wigner method, and the second is
via a Markov approximation, leading to a master equation or
a Langevin equation, analogous to those used in classical
statistical mechanics. As Agarwal stresses, one reason for
preferring the newer statistical approach is that it allows us to
derive the Lamb shift in the ground state, which is not pre-
dicted by the Weisskopf-Wigner method even after that
method has been amended to obtain the shift in the excited
state. But we can derive the ground state shift only if we are
careful about how we use the rotating wave approximation.
18 Hans Bethe, 'The Electromagnetic Shift of Energy Levels', Physics Review,
72 (1947), p. 339, italics added.
19 Willis E. Lamb, Jr., 1955 Nobel Prize Address, Science 123 (1956), p. 442.
20 G. S. Agarwal, Quantum Statistical Theories of Spontaneous Emission

and their Relation to Other Approaches (Berlin: Springer-Verlag, 1974). See
Chapter 10 and Appendix A.
The rotating wave approximation is used when the inter-

action between radiation and matter is weak. In weak inter-
actions, such as those that give rise to spontaneous emission,
the atoms and field can be seen as almost separate systems,
so that energy lost by the atoms will be found in the field,
and vice versa. Thus virtual transitions in which both the
atom and the field simultaneously gain or lose a quantum
of energy will have negligible effects. The rotating wave
approximation ignores these effects. When the coupling is
weak, the terms which represent virtual transitions vary as
exp{± i(w + Wk)t} (hw = Em - En' for energies levels Em
and En of the atom; Wk is a mode frequency of the field).
Energy conserving transitions vary as exp{± i(w - Wk)t}.
For optical frequencies Wk is large. Thus for ordinary times
of observation the exp{± i(w + Wk)t} terms oscillate rapidly,
and will average approximately to zero. The approximation
is called a 'rotating-wave' approximation because it retains
only terms in which the atom and field waves 'rotate together'.
The statistical treatments which give rise to the master
equation make an essential use of a Markov approximation.
In the last section I outlined one standard way to derive the
Pauli equations, which are the master equations relevant for
spontaneous emission. But according to Agarwal, there are
two ways to carry through such a derivation, and the results
are significantly different depending on where we apply the
rotating wave approximation. On the one hand, we can begin
with the full Hamiltonian for the interaction of the orbiting
electron with the electromagnetic field, and drop from this
Hamiltonian the 'counter-rotating' terms which represent
virtual transitions, to obtain a shortened approximate
Hamiltonian which Agarwal numbers (2.24). Then, following
through steps like those described above, one obtains a
version of the master equation-Agarwal's equation A.7.
Alternatively, we can use the full Hamiltonian throughout,
dropping the counter-rotating terms only in the last step.
This gives us Agarwal's equation (A. 6).
What difference do the two methods make to the Lamb
shift? Agarwal reports:
The shift of the ground state is missing from (A. 7), mainly due to
the virtual transitions which are automatically excluded from the
Hamiltonian (2.24). The master equation (A.6) obtained by making

RWA (the rotating wave approximation) on the master equation
rather than on the Hamiltonian does include the shift of the ground
state. These remarks make it clear that RWA on the original Hamilton-
ian is not the same as RWA on the master equation and that one should
make RWA on the final equations of motion."
The rotating wave approximation is justified in cases of
spontaneous emission by the weakness of the coupling
between the atom and the field. But no further features of
the interaction determine whether we should apply the
approximation to the original Hamiltonian, or whether
instead we should apply it to the master equation. Agarwal
applies it to the master equation, and he is thus able to derive
a Lamb shift in the ground state. But his derivation does not
show that the Schroedinger equation dictates a Lamb shift
for a two-level atom in weak interaction with an electro-
magnetic field. The shift is consistent with what the equation
says about weak interactions, but it does not follow from it.
This kind of situation is even more strikingly illustrated
if we try to calculate the values of the shifts for the two-
level atoms. Lamb and Retherford's experiments, for
example, measured the value of the shift for the 28 state in
hydrogen to be 1057 mega-cycles per second. We can 'derive'
a result very close to this in quantum electrodynamics
using the technique of mass renormalization for the electron.
But the derivation is notorious: the exact details, in which
infinities are subtracted from each other in just the right way
to produce a convergent result, are completely ad hoc, and
yet the quantitative results that they yield are inordinately
accurate.
The realist has a defence ready. I say that the Schroedinger
equation does not make a claim about whether there is or
is not a Lamb shift in the circumstances described. But the
realist will reply that I have not described the circumstances
as fully as possible. The rotating wave approximations
depend on the fact that the interaction between the field and
the atom is 'weak'; but if realism is correct, there is a precise
answer to the question 'How weak?' The atom-field inter-
action will have a precise quantitative representation that
21 Agarwal, op. cit., p. 116.
can in principle be written into the Schroedinger equation.

The exact solution to the equation with that term written
in will either contain a Lamb shift, or it will not, and that
is what the Schroedinger equation says about a Lamb shift
in cases of spontaneous emission.
This defence allows me to state more exactly the aims of
this essay. I do not argue against taking the fundamental laws
as true, but try only to counter the most persuasive argument
in favour of doing so. As a positive argument in favour of
realism, the realist's defence, which plumps for exact solu-
tions and rigorous derivations, gets the logic of the debate
out of order. I begin with the challenge, 'Why should we
assume that the short abstract equations, which form the
core of our fundamental theories, are true at all, even "true
enough for now"?' The realist replies 'Because the funda-
mental equations are so successful at explaining a variety of
messy and complicated phenomenological laws'. I say 'Tell
me more about explanation. What is there in the explanatory
relation which guarantees that the truth of the explanandum
argues for the truth of the explanans?' The realist's answer,
at least on the plausible generic-specific account, is that the
fundamental laws say the same thing as the phenomeno-
logical laws which are explained, but the explanatory laws
are more abstract and general. Hence the truth of the phe-
nomenological laws is good evidence for the truth of the
fundamental laws. I reply to this, 'Why should we think the
fundamental and the phenomenological laws say the same
thing about the specific situations studied?' and the realist
responds, 'J ust look at scientific practice. There you will
see that phenomenological laws are deducible from the more
fundamental laws that explain them, once a description of
the circumstances is given.'
We have just been looking in detail at cases of scientific
practice, cases which I think are fairly typical. Realists can
indeed put a gloss on these examples that brings them into
line with realist assumptions: rigorous solutions to exact equa-
tions might possibly reproduce the correct phenomenological
laws with no ambiguity 'when the right equations are found'.
But the reason for believing in this gloss is not the practice
itself, which we have been looking at, but rather the realist
metaphysics, which I began by challenging. Examples of the
sort we have considered here could at best be thought con-
sistent with realist assumptions, but they do not argue for
them.
3. CONCLUSION
Fundamental laws are supposed by many to determine what

phenomenological laws are true. If the primary argument for
this view is the practical explanatory success of the funda-
mental laws, the conclusion should be just the reverse. We
have a very large number of phenomenological laws in all
areas of applied physics and engineering that give highly
accurate, detailed descriptions of what happens in realistic
situations. In an explanatory treatment these are derived
from fundamental laws only by a long series of approxima-
tions and emendations. Almost always the emendations
improve on the dictates of the fundamental law; and even
where the fundamental laws are kept in their original form,
the steps of the derivation are frequently not dictated by the
facts. This makes serious trouble for the D-N model, the
generic-specific account, and the view that fundamental
laws are better. When it comes to describing the real world,
phenomenological laws win out.
Essay 7
Fitting Facts to Equations
O. INTRODUCTION
At H. P. Grice's seminar on metaphysics in the summer of

1975, we discussed Aristotle's categories. I argued then that
the category of quantity was empty; there were no quanti-
ties in nature-no attributes with exact numerical values
of which it could be said that they were either precisely
equal or unequal to each other. I was thinking particularly
about physics, and the idea I had was like the one I have
been defending in these essays, that the real content of
our theories in physics is in the detailed causal knowledge
they provide of concrete processes in real materials. I thought
that these causal relations would hold only between qualities
and not between quantities. Nevertheless I recognized that
real materials are composed of real atoms and molecules
with numerically specific masses, and spins, and charges;
that atoms and molecules behave in the way that they do
because of their masses, spins, and charges; and that our
theoretical analyses of the causal processes they are involved
in yield precise numerical calculations of other quantities,
such as line shapes in spectroscopy or transport coefficients
in statistical mechanics.
Why then did I want to claim that these processes were
essentially qualitative? It was because our knowledge about
them, while detailed and precise, could not be expressed in
simple quantitative equations of the kind that I studied in
theoretical physics. The distinction I wanted, it turns out,
was not that between the qualitative and the quantitative,
but rather the distinction between the tidy and simple
mathematical equations of abstract theory, and the.intricate
and messy descriptions, in either words or formulae, which
express our knowledge of what happens in real systems
made of real materials, like helium-neon lasers or turbo-jet
engines. We may use the fundamental equations of physics
to calculate precise quantitative facts about real situations,
FITTING FACTS TO EQUATIONS 129
but as I have urged in earlier essays, abstract fundamental
laws are nothing like the complicated, messy laws which
describe reality. I no longer want to urge, as I did in the
summer seminar, that there are no quantities in nature, but
rather that nature is not governed by simple quantitative
equations of the kind we write in our fundamental theories.
My basic view is that fundamental equations do not govern
objects in reality; they govern only objects in models. The
second half of this thesis grew out of another Grice seminar
on metaphysics not long after. In the second seminar we
talked about pretences, fictions, surrogates, and the like;
and Grice asked about various theoretical claims in physics,
where should we put the 'as if' operator: helium gas behaves
as if it is a collection of molecules which interact only on
collision? Or, helium gas is composed of molecules which
behave as if they interact only on collision? Or ... ?
Again, I wanted to make apparently conflicting claims.
There are well-known cases in which the 'as if' operator
should certainly go all the way in front: the radiating mole-
cules in an ammonia maser behave as if they are classical
electron oscillators. (We will see more of this in the last
essay.) How closely spaced are the oscillators in the maser
cavity? This realistic question is absurd; classical electron
oscillators are themselves a mere theoretical construct. What
goes on in a real quantum atom is remarkably like the theo-
retical prescriptions for a classical electron oscillator. The
oscillators replicate the behaviour of real atoms; but still,
as laser specialist Anthony Siegman remarked in his laser
engineering class, 'I wouldn't know where to get myself
a bagful of them'.'
The classical electron oscillators are undoubted fictions.
But, even in cases where the theoretical entities are more
robust, I still wanted to put the 'as if' operator all the way
in front. For example, a helium-neon laser behaves as if
it is a collection of three-level atoms in interaction with
a single damped mode of a quantized field, coupled to a
pumping and damping reservoir. But in doing so, I did not
want to deny that the laser cavities contain three-level
1 Anthony Siegman, 'Lasers' (Electrical Engineering 231), Stanford University,
Autumn Term 1981-2.
130 FITTING FACTS TO EQUATIONS
atoms or that a single mode of the electromagnetic field

is dominant. I wanted both to acknowledge these existential
facts and yet to locate the operator at the very beginning.
It seems now that I had conflicting views about how to
treat this kind of case because I was conflating two functions
which the operator could serve. On the one hand, putting
things to the left of the operator is a sign of our existential
commitment. A helium-neon laser is a collection of three-
level atoms . . . But putting things on the right serves a
different function. Commonly in physics what appears on
the right is just what we need to know to begin our mathe-
matical treatment. The description on the right is the kind
of description for which the theory provides an equation.
We say that a 'real quantum atom' behaves like a classical
electron oscillator; already the theory tells us what equation
is obeyed by a classical electron oscillator. Similarly, the
long description I gave above of a laser as a collection of
three-level atoms also tells us a specific equation to write
down, in this case an equation called the Fokker-Planck
equation; and there are other descriptions of gas lasers
which go with other equations. We frequently, for example,
treat the laser as a van der Pol oscillator, and then the appro-
priate equation would be the one which B. van der Pol
developed in 1920 for the triode oscillator.
Contrary to my initial assumption I now see that the two
functions of the 'as if operator are quite distinct. Giving a
description to which the theory ties an equation can be
relatively independent of expressing existential commitment.
Both treatments of the laser which I mentioned assume that
the helium-neon laser contains a large number of three-
level neon atoms mixed with a much greater number of
helium atoms, in interaction almost entirely with a single
mode of the electromagnetic field. Similarly, when an experi-
mentalist tells us of a single mode of a CW GaAs (gallium
arsenide) laser that 'below threshold the mode emits noise
like a narrow band black body source; above threshold
its noise is characteristic of a quieted amplitude stabilized
oscillator'i? he is telling us not that the make-up of the laser
2 T. A. Armstrong and A. W. Smith, 'Intensity Fluctuations in a GaAs Laser',
Physical Review Letters 14 (1965), p. 68, italics added.
has changed but rather that its intensity fluctuations follow

from different equations above and below threshold. In
these cases what goes on the right of the 'as if operator
does not depend on what we take to be real and what we
take to be fictional. Rather it depends on what description
we need to know in order to write down the equation that
starts our mathematical treatment.
The views I urged in the two seminars go hand-in-hand.
It is because the two functions of the 'as if' operator are
independent that the fundamental equations of our theories
cannot be taken to govern objects in reality. When we use
the operator to express existential commitment, we should
describe on the left everything we take to be real. From a
first, naive point of view, to serve the second function we
should just move everything from the left of the operator
to the right. To get a description from which we can write
down an equation, we should simply report what we take
to be the case.
But that is not how it works. The theory has a very limited
stock of principles for getting from descriptions to equations,
and the principles require information of a very particular
kind, structured in a very particular way. The descriptions
that go on the left-the descriptions that tell what there
is-are chosen for their descriptive adequacy. But the
'descriptions' on the right-the descriptions that give rise
to equations-must be chosen in large part for their mathe-
matical features. This is characteristic of mathematical
physics. The descriptions that best describe are generally
not the ones to which equations attach. This is the thesis
that I will develop in the remaining sections of this paper.
1. TWO STAGES OF THEORY ENTRY
Let us begin by discussing bridge principles. On what Fred

Suppe has dubbed the 'conventional view of theories', 3
championed by Hempel, Grunbaum, Nagel, and others in
the tradition of logical empiricism, the propositions of a
theory are of two kinds: internal principles and bridge
• In Frederick Suppe, The Structure of Scientific Theories (Urbana: Uni-
versity of Illinois Press, 1977).
principles. The internal principles present the content of the

theory, the laws that tell how the entities and processes of
the theory behave. The bridge principles are supposed to
tie the theory to aspects of reality more readily accessible
to us. At first the bridge principles were thought to link the
descriptions of the theory with some kind of observation
reports. But with the breakdown of the theory-observation
distinction, the bridge principles were required only to link
the theory with a vocabulary that was 'antecedently under-
stood'.
The network of internal principles and bridge principles
is supposed to secure the deductive character of scientific
explanation. To explain why lasers amplify light signals,
one starts with a description in the antecedent vocabulary
of how a laser is constructed. A bridge principle matches
this with a description couched in the language of the
quantum theory. The internal principles of quantum mech-
anics predict what should happen in situations meeting
this theoretical description, and a second bridge principle
carries the results back into a proposition describing the
observed amplification. The explanation is deductive be-
cause each of the steps is justified by a principle deemed
necessary by the theory, either a bridge principle or an
internal principle.
Recently, however, Hempel has begun to doubt that
explanations of this kind are truly deductive." The fault
lies with the bridge principles, which are in general far from
exceptionless, and hence lack the requisite necessity. A
heavy bar attracts iron filings. Is it thus magnetic? Not
necessarily: we can never be sure that we have succeeded
in ruling out all other explanations. A magnet will, with
surety, attract iron filings only if all the attendant circum-
stances are right. Bridge principles, Hempel concludes, do
not have the character of universal laws; they hold only for
the most part, or when circumstances are sufficiently ideal.
I think the situation is both much better and much worse
than Hempel pictures. If the right kinds of descriptions
are given to the phenomena under study, the theory will
• C. G. Hempel, in a paper at the conference 'The Limits of Deductivity',
University of Pittsburgh, Autumn 1979.
tell us what mathematical description to use and the prin-

ciples that make this link are as necessary and exceptionless
in the theory as the internal principles themselves. But the
'right kind of description' for assigning an equation is seldom,
if ever, a 'true description' of the phenomenon studied; and
there are few formal principles for getting from 'true descrip-
tions' to the kind of description that entails an equation. There
are just rules of thumb, good sense, and, ultimately, the re-
quirement that the equation we end up with must do the job.
Theory entry proceeds in two stages. I imagine that we
begin by writing down everything we know about the system
under study, a gross 'exaggeration, but one which will help
to make the point. This is the unprepared description-it
is the description that goes to the left of the 'as if' operator
when the operator is used to express existential commitment.
The unprepared description contains any information we
think relevant, in whatever form we have available. There is
no theory-observation distinction here. We write down
whatever information we have: we may know that the
electrons in the beam are all spin up because we have been
at pains to prepare them that way; or we may write down
the engineering specifications for the construction of the
end mirrors of a helium-neon laser; and we may also know
that the cavity is filled with three-level helium atoms. The
unprepared description may well use the language and the
concepts of the theory, but it is not constrained by any
of the mathematical needs of the theory.
At the first stage of theory entry we prepare the descrip-
tion: we present the phenomenon in a way that will bring
it into the theory. The most apparent need is to write down
a description to which the theory matches an equation. But
to solve the equations we will have to know what boundary
conditions can be used, what approximation procedures
are valid, and the like. So the prepared descriptions must
give information that specifies these as well. For example,
we may describe the walls of the laser cavity and their
surroundings as a reservoir (a system with a large number
of resonant modes). This means that the laser has no memory.
Formally, when we get to the derivation, we can make a
Markov approximation. (Recall the discussion in Essay 6.)
This first stage of theory entry is informal. There may be

better and worse attempts, and a good deal of practical
wisdom helps, but no principles of the theory tell us how
we are to prepare the description. We do not look to a
bridge principle to tell us what is the right way to take the
facts from our antecedent, unprepared description, and to
express them in a way that will meet the mathematical
needs of the theory. The check on correctness at this stage
is not how well we have represented in the theory the facts
we know outside the theory, but only how successful the
ultimate mathematical treatment will be.
This is in sharp contrast with the second stage of theory
entry, where principles of the theory look at the prepared
description and dictate equations, boundary conditions,
and approximations. Shall we treat a CW GaAs laser below
threshold as a 'narrow band black body source' rather than
the 'quieted stabilized oscillator' that models it above thres-
hold? Quantum theory does not answer. But once we have
decided to describe it as a narrow band black body source,
the principles of the theory tell what equations will govern
it. So we do have bridge principles, and the bridge principles
are no more nor less universal than any of the other prin-
ciples. But they govern only the second stage of theory
entry. At the first stage there are no theoretical principles
at all-only rules of thumb and the prospect of a good
prediction.
This of course is a highly idealized description. Theories
are always improving and expanding, and an interesting
new treatment may offer a totally new bridge principle.
But Hempel's original account was equally idealized; it
always looked at the theory as it stood after the explanation
had been adopted. I propose to think about it in the same
way. In the next section I' want to illustrate some bridge
principles, and I shall describe the two stages of theory
entry with some examples from quantum mechanics.
2. SOME MODEL BRIDGE PRINCIPLES

If we look at typical formalizations of quantum rnechanics.!
it seems that the fundamental principles do divide into
internal principles and bridge principles, as the conventional
view of theories maintains. The central internal principle
is Schroedinger's equation. The Schroedinger equation tells
how systems, subject to various forces, evolve in time.
Actually the forces are not literally mentioned in the equa-
tion since quantum mechanics is based on William Hamilton's
formulation of classical mechanics, which focuses not on
forces, but on energies. In the standard presentation, the
Schroedinger equation tells how a quantum system evolves
in time when the Hamiltonian of the system is known, where
the Hamiltonian is a mathematical representation of the
kinetic and potential energies for the system. Conservation
principles, like the conservation of energy, momentum, or
parity, may also appear as internal principles in such a
formalization. (On the other hand, they may not, despite
the fact that they are of fundamental importance, because
these principles can often be derived from other basic
principles.)
The second class of principles provide schemata for getting
into and out of the mathematical language of the theory:
states are to be represented by vectors; observable quantities
are represented by operators; and the average value of a
given quantity in a given state is represented by a certain
product involving the appropriate operator and vector. So
far, all looks good for the conventional view of theories.
But notice: one may know all of this and not know any
quantum mechanics. In a good undergraduate text these
two sets of principles are covered in one short chapter. It is
true that the Schroedinger equation tells how a quantum
system evolves subject to the Hamiltonian; but to do quantum
mechanics, one has to know how to pick the Hamiltonian.
5 See, for example, the formalization in Chapter 5, 'Development of the

Formalism of Wave Mechanics and its Interpretation', in Albert Messiah's highly
respected text, Quantum Mechanics (Amsterdam: North-Holland, 1969) or the
formal axiomatization in George W. Mackey's Mathematical Foundations of
Quantum Mechanics (New York: W. A. Benjamin, 1963). '
The principles that tell us how to do so are the real bridge

principles of quantum mechanics. These give content to
the theory, and these are what beginning students spend the
bulk of their time learning.
If the coventional view were right, students should be at
work learning bridge principles with mathematical formulae
on one end and descriptions of real things on the other. Good
textbooks for advanced undergraduates would be full of
discussions of concrete situations and of the Hamiltonians
which describe them. There might be simplifications and
idealization for pedagogical purposes; nevertheless, there
should be mention of concrete things made of the materials
of the real world. This is strikingly absent. Generally there
is no word of any material substance. Instead one learns the
bridge principles of quantum mechanics by learning a se-
quence of model Hamiltonians. I call them 'model Hamil-
tonians' because they fit only highly fictionalized objects.
Here is a list of examples. I culled it from two texts, both
called Quantum Mechanics, one by Albert Messiah" and
the other by Eugen Merzbacher.? This list covers what one
would study in just about any good senior level course on
quantum mechanics. We learn Hamiltonians for:
free particle motion, including

the free particle in one dimension,
the free particle in three dimensions,
the particle in a box;
the linear harmonic oscillator;
piecewise constant potentials, including
the square well,
the potential step,
the periodic potential,
the Coulomb potential;
'the hydrogen atom';
diatomic molecules;
central potential scattering;
• Messiah, op. cit.

, Eugen Merzbacher, Quantum Mecahnics, Second Edition (New York: John
Wiley & Sons, 1970).
and eventually, the foundation of all laser theory,

the electron in interaction with the electromagnetic field.
There is one real material mentioned in this list-hydro-
gen. In fact this case provides a striking illustration of my
point, and not a counterexample against it. The Hamiltonian
we learn here is not that for any real hydrogen atom. Real
hydrogen atoms appear in an environment, in a very cold
tank for example, or on a benzene molecule; and the effects
of the environment must be reflected in the Hamiltonian.
What we study instead is a hypothetically isolated atom.
We hope that later we will be able to piece together this
Hamiltonian with others to duplicate the circumstances of
an atom in its real situation.
But this is not the most striking omission. In his section
titled 'The Hydrogen Atom', Messiah proposes a particular
Hamiltonian and uses it to provide a solution for the energy
spectrum of hydrogen. He says:
This spectrum is just the one predicted by the Old Quantum Theory;
its excellent agreement with the experimental spectrum was already
pointed out. To be more precise, the theory correctly accounts for
the position of the spectral lines but not for their fine structure. Its
essential shortcoming is to be a non-relativistic theory . . . [Also] the
Schroedinger theory does not take the electron spin into account.P
These are critical omissions. The discovery and the account

of the fine structure of hydrogen were significant events
in quantum mechanics for the reasons Messiah mentions.
Fine structure teaches important lessons both about relativity
and about the intrinsic spin of the electron.
The passage quoted above appears about three quarters
of the way through Volume I. About the same distance into
Volume II, Messiah again has a section called 'The Hydrogen
Atom'. There he uses the relativistic theory of the Dirac
electron. Even the second treatment is not true to the real
hydrogen atom. The reasons are familiar from our discussion
of the Lamb shift in Essay 6. Here is what Messiah him-
self says:
8 Messiah,op. cit., p. 419.

The experimental results on the fine structure of the hydrogen atom

and hydrogen-like atoms (notably He+) are in broad agreement with
these predictions.
However, the agreement is not perfect. The largest discrepancy is
observed in the fine structure of the n = 2 levels of the hydrogen
atom. In the non-relativistic approximation, the three levels 2S112,
2P1/2, and 2P312 are equal. In the Dirac theory, the levels 2s1I2, and
2P112 are still equal, while the 2P312 level is slightly lower (the separa-
tion is of the order of IO-4eV). The level distance from 2P3/2 to
2P1I2 agrees with the theory but the level 2S1/2 is lower than the level
2PII2' and the distance from 2s1l2 to 2P1/2 is equal to about a tenth
of the distance from 2P312 to 2P1I2' This effect is known as the Lamb
shift. To explain it, we need a rigorous treatment of the interaction
between the electron, the proton and the quantized electromagnetic
field; in the Dirac theory one retains only the Coulomb potential
which is the main term in that interaction; the Lamb shift represents
'radiative corrections' to this approximation."
We know from our earlier discussion that the treatment of
these 'radiative corrections' for the hydrogen spectrum is
no simple matter.
The last sentence of Messiah's remark is telling. The
two sections are both titled 'The Hydrogen Atom' but in
neither are we given a Hamiltonian for real hydrogen atoms,
even if we abstract from the environment. Instead, we are
taught how to write the Coulomb potential between an
electron and a proton, in the first case non-relativistically
and in the second, relativistically. Messiah says so himself:
'The simplest system of two bodies with a Coulomb inter-
action is the hydrogen atom'i '? 'The hydrogen atom' on our
list is just a name for a two-body system where only the
Coulomb force is relevant. Even if the system stood alone
in the universe, we could not strip away the spin from the
electron. Even less could we eliminate the electromagnetic
field, for it gives rise to the Lamb shift even when no photons
are present. This two-body system, which we call 'the hydro-
gen atom', is a mere mental construct.
Messiah's is of course an elementary text, intended for
seniors or for beginning graduate students. Perhaps we are
looking at versions of the theory that are too elementary?
9 Ibid., pp. 932-3. 10 Ibid., p. 412.

Do not more sophisticated treatments-journal articles,
research reports, and the like-provide a wealth of different,
more involved bridge principles that link the theory to more
realistic descriptions? I am going to argue in the next chapter
that the answer to this question is no. There are some more
complicated bridge principles; and of course the theory is
always growing, adding both to its internal principles and its
bridge principles. But at heart the theory works by piecing
together in original ways a small number of familiar prin-
ciples, adding corrections where necessary. This is how it
should work. The aim is to cover a wide variety of different
phenomena with a small number of principles, and that
includes the bridge principles as well as the internal principles.
It is no theory that needs a new Hamiltonian for each new
physical circumstance. The explanatory power of quantum
theory comes from its ability to deploy a small number of
well-understood Hamiltonians to cover a wide range of
cases. But this explanatory power has its price. If we limit
the number of Hamiltonians, that is going to constrain our
abilities to represent situations realistically. This is why our
prepared descriptions lie.
I will take up these remarks about bridge principles again
in the next chapter. Here I want to proceed in a different
way. I claim that in general we will have to distort the true
picture of what happens if we want to fit it into the highly
constrained structures of our mathematical theories. I think
there is a nice analogy that can help us see why this is so.
That is the topic of the next section.
3. PHYSICS AS THEATRE
I will present first an analogy and then an example. We begin

with Thucydides' views on how to write history:
XXII. As to the speeches that were made by different men, either
when they were about to begin the war or when they were already
engaged therein, it has been difficult to recall with strict accuracy
the words actually spoken, both for me as regards that which I my-
self heard, and for those who from various other sources have brought
me reports. Therefore the speeches are given in the language in which,
as it seemed to me, the several speakers would express, on the subjects
under consideration, the sentiments most befitting the occasion, though

at the same time I have adhered as closely as possible to the general
sense of what was actually said.'!
Imagine that we want to stage a given historical episode. We
are primarily interested in teaching a moral about the motives
and behaviour of the participants. But we would also like
the drama to be as realistic as possible. In general we will
not be able simply to 'rerun' the episode over again, but this
time on the stage. The original episode would have to have
a remarkable unity of time and space to make that possible.
There are plenty of other constraints as well. These will
force us to make first one distortion, then another to
compensate. Here is a trivial example. Imagine that two of
the participants had a secret conversation in the corner of
the room. If the actors whisper together, the audience will
not be able to hear them. So the other characters must be
moved off the stage, and then back on again. But in reality
everyone stayed in the same place throughout. In these
cases we are in the position of Thucydides. We cannot
replicate what the characters actually said and did. Nor is
it essential that we do so. We need only adhere 'as closely
as possible to the general sense of what was actually said'.
Physics is like that. It is important that the models we
construct allow us to draw the right conclusions about the
behaviour of the phenomena and their causes. But it is not
essential that the models accurately describe everything
that actually happens; and in general it will not be possible
for them to do so, and for much the same reasons. The
requirements of the theory constrain what can be literally
represented. This does not mean that the right lessons cannot
be drawn. Adjustments are made where literal correctness
does not matter very much in order to get the correct effects
where we want them; and very often, as in the staging ex-
ample, one distortion is put right by another. That is why it
often seems misleading to say that a particular aspect of a
model is false to reality: given the other constraints that is
just the way to restore the representation.
Here is a very simple example of how the operation of
11 Thucydides, The Peloponnesian War, VoL 1, trans. Charles Forster Smith
(New York: G. P. Putnam's Sons, 1923), p. 39.
constraints can cause us to set down a false description in
physics. In quantum mechanics free particles are represented
by plane waves-functions that look like sines or cosines,
stretching to infinity in both directions. This is the repre-
sentation that is dictated by the Schroedinger equation,
given the conventional Hamiltonian for a free particle. So
far there need be nothing wrong with a wave like that. But
quantum mechanics has another constraint as well: the
square of the wave at a point is supposed to represent the
probability that the particle is located at that point. So the
integral of the square over all space must equal one. But that
is impossible if the wave, like a sine or cosine, goes all the
way to infinity.
There are two common solutions to this problem. One is
to use a Dirac delta function. These functions are a great
help to physics, and generalized function theory now ex-
plains how they work. But they side-step rather than solve
the problem. Using the delta function is really to give up
the requirement that the probabilities themselves integrate
to one. Merzbacher, for instance, says 'Since normalization
of J'lt*'lt to unity is out the question for infinite plane
waves, we must decide on an alternative normalization
for these functions. A convenient tool in the discussion
of such wave functions is the delta function'. 12 I have thus
always preferred the second solution.
This solution is called 'box normalization'. In the model
we assume that the particle is in avery, very large box,
and that the wave disappears entirely at the edges of this
box. To get the wave to go to zero we must assume that
the potential there-very, very far away from anything
we are interested in-is infinite. Here is what Merzbacher
says in defence of this assumption:
The eigenfunctions are not quadratically integrable over all space.
It is therefore impossible to speak of absolute probabilities and of
expectation values for physical quantities in such a state. One way
of avoiding this predicament would be to recognize the fact that
physically no particle is ever absolutely free and that there is inevit-
ably some confinement, the enclosure being for instance the wall
12 Merzbacher, op. cit., p. 82, italics added.

of an accelerator tube or of the laboratory. V [the potential] rises to

infinity at the boundaries of the enclosure and does then not have the
same value everywhere, the eigenfunctions are no longer infinite plane
waves, and the eigenvalue spectrum is discrete rather than continuous.P
Here is a clear distortion of the truth. The walls may interact
with the particle and have some effect on it, but they cer-
tainly do not produce an infinite potential.
I think Merzbacher intends us to think of the situation
this way. The walls and environment do contain the particle;
and in fact the probability is one that the particle will be
found in some finite region. The way to get this effect in
the model is to set the potential at the walls to infinity. Of
course this is not a true description of the potentials that
are actually produced by the walls and the environment.
But it is not exactly false either. It is just the way to achieve
the results in the model that the walls and environment
are supposed to achieve in reality. The infinite potential is
a good piece of staging.
IS Ibid., p. 82.
Essay 8
The Simulacrum Account of Explanation
O. INTRODUCTION
We saw in the last chapter that the bridge principles in a
theory like quantum mechanics are few in number, and
they deal primarily with highly fictionalized descriptions.
Why should this be so? Some work of T. S. Kuhn suggests
an answer. In his paper, 'A Function for Measurement in
the Physical Sciences', and in other papers with the word
'function' in the title, Kuhn tries something like functional
explanations of scientific practice. Anthropologists find
a people with a peculiar custom. The people themselves give
several or maybe no reasons for their custom. But the anthro-
pologist conjectures that the custom remains among these
people not only for their avowed reasons, but also because
other customs or ecological conditions make it very difficult
for their society to survive without it. This then is the 'func-
tion' of the custom in question, even though it is not practised
with any conscious awareness of that function. Naturally
all functional explanations have a dubious logic, but they
do often bring out instructive aspects of the custom in
question.
Now let us ask what function might be served by having
relatively few bridge principles to hand when we are engaged
in constructing models of phenomena. Kuhn concludes his
paper on measurement by saying he believes 'that the nine-
teenth century mathematization of physical science produced
vastly refined professional criteria for problem selection and
that it simultaneously very much increased the effectiveness
of professional verification procedures';' I think that some-
thing similar is to be said about having a rather small number
of bridge principles. The phenomena to be described are
endlessly complex. In order to pursue any collective research,
a group must be able to delimit the kinds of models that
1 T. S. Kuhn, 'A Function for Measurement in the Physical Sciences', in T. S.
Kuhn, The Essential Tension (Chicago: University of Chicago Press, 1977), p. 220.
144 THE SIMULACRUM ACCOUNT OF EXPLANATION
are even contenders. If there were endlessly many possible

ways for a particular research community to hook up phe-
nomena with intellectual constructions, model building
would be entirely chaotic, and there would be no consensus
of shared problems on which to work.
The limitation on bridge principles provides a consensus
within which to formulate theoretical explanations and
allows for relatively few free parameters in the construction
of models. This in turn provides sharp criteria for problem
selection. Naturally there may be substantial change in the
structure of bridge principles if nothing works; but we hang
on to them while we can. It is precisely the existence of
relatively few bridge principles that makes possible the
construction, evaluation, and elimination of models. This
fact appears also to have highly anti-realist side effects. As
we have seen, it strongly increases the likelihood that there
will be literally incompatible models that all fit the facts
so far as the bridge principles can discriminate.
This is just a sketch of a Kuhnian account, but one which
I believe is worth pursuing. Borrowing a term from the
historians of science, it might be called an 'external' account
of why bridge principles need to be limited in number.
But it fits nicely with a parallel 'internal' account, one
that holds that the limitation on bridge principles is crucial
to the explanatory power of the theory. I will argue for
this internal account in section I of this essay. In section 2
I propose a model of explanation that allows for the paucity
of bridge principles and makes plain the role of fictionalized
descriptions.
2. BRIDGE PRINCIPLES AND 'REALISTIC' MODELS
A good theory aims to cover a wide variety of phenomena

with as few principles as possible. That includes bridge
principles. It is a poor theory that requires a new Hamil-
tonian for each new physical circumstance. The great ex-
planatory power of quantum mechanics comes from its
ability to deploy a small number of well-understood Hamil-
tonians to cover a broad range of cases, and not from its
ability to match each situation one-to-one with a new
THE SIMULACRUM ACCOUNT OF EXPLANATION 145
mathematical representation. That way of proceeding would
be crazy.
This is an obvious fact about what theories must be like
if they are to be manageable at all. But it has the anti-realist
consequences that we have seen. Why have realists not been
more troubled by this fact? The answer, I think, is that
many realists suppose that nature conspires to limit the
number of bridge principles. Only a few bridge principles
are needed because only a small number of basic interactions
exist in nature. An ideal theory will represent each of the
basic interactions; new cases will not require new bridge
principles because the representation for complex situations
can be constructed from the representations for the basic
components.
I think that this is a radically mistaken point of view.
First, it is a model of a physics we do not have. That is a
familiar point by now. Much worse, it is a model of a physics
we do not want. The piecing-together procedure would be
unbearably complex. It goes in exactly the wrong direction.
The beauty and strength of contemporary physics lies in
its ability to give simple treatments with simple models,
where at least the behaviour in the model can be under-
stood and the equations can not only be written down
but can even be solved in approximation. The harmonic
oscillator model is a case in point. It is used repeatedly in
quantum mechanics, even when it is difficult to figure out
exactly what is supposed to be oscillating: the hydrogen
atom is pictured as an oscillating electron; the electro-
magnetic field as a collection of quantized oscillators; the
laser" as a van der Pol oscillator; and so forth. The same
description deployed again and again gives explanatory
power.
It is best to illustrate with a concrete case. In the last essay
we looked to elementary texts for examples of bridge prin-
ciples. Here I will present a more sophisticated example:
a quantum theoretical account of a laser. Recall from Essay 4
that there are a variety of methods for treating lasers in
quantum mechanics. One is the quantum statistical approach
in which a master equation (or a Langevin equation) is de-
rived for the system. This kind of approach is familiar from
our discussion in Essay 6 of the Markov approximation

for radiating atoms, so this is a good example to choose.
There is a thorough development of this method in William
Louisell's Quantum Statistical Properties of Radiation?
The treatment there is most suitable for a gas laser, such
as the early helium-neon laser. Louisell proposes what he
calls a 'block diagram'. (See Figure 8.1.) He imagines that
the laser consists of three-level atoms in interaction with
a quantized electromagnetic field. Before the quantum
statistical approach, treatments of lasers were generally
semi-classical: the atoms were quantized but the field was
not. Louisell also explicitly includes the interaction of
both the atoms and the field with a damping reservoir.
These two features are important for they allow the deriva-
tion of correlations among the emitted photons which are
difficult to duplicate in the earlier semi-classical approaches.
Mode of
radiation Reservoir I
field
Atom-Field
interaction
N three- Pumping I
level
atoms Damping I
FIG. 8.1. Block diagram of laser modeL (Source: Louisell, Quantum

Statistical Properties of Radiation.)
In Essay 6 I talked briefly about idealizations that are

hard to eliminate at the theoretical level. Here is a good
illustration. Louisell supposes that the atoms are uniformly
distributed, N per unit volume, and that they do not interact
with each other: 'They are coupled to each other only
through their atom-field interaction'." In reality the atoms
do interact, though this does not have much effect on the
performance of the laser. The missing effects can sometimes
2 w. H. Louisell, Quantum Statistical Properties of Radiation (New York:
John Wiley & Sons, 1973), Ch. 9.
3 Ibid., p. 469.
be corrected for, but this is done piece-meal when the theory

is applied and not by adding terms to the fundamental
Hamiltonian given in the theoretical treatment.
Louisell's equation for the system represented by his
block diagram consists of three parts. I will write it here
just for reference:
asat = ~ [S,
ih
u; + W] +[as)
\at F
+ (as) .
at A
He tells us of this equation that 'the first terms describe the

causal behavior. The second term describes the interaction of
the field mode with its reservoir ... The last term describes
the interaction of the atoms with their pumping and damping
reservoirs'." This equation is deceptively simple because it
is still just a schema. We do not yet know how W, caSjat)F,
and so forth are to be represented for the block laser. This
is where bridge principles enter. A page and a half later
when these variables have been filled in, this simple-seeming
equation will take twelve lines of text for Louisell to write.
The first term is supposed to represent the 'causal be-
haviour', in contrast with the last two terms. Another com-
mon way of expressing this contrast would be to say: the
atom-field interaction is represented realistically, but the
terms for the reservoir interactions are just phenomenological.
Louisell's method of expression is better because it is nar-
rower. Physicists use 'realistic' in a variety of senses. One
common sense contrasts 'realistic' with 'idealized'. This
sense concerns the relation between the model and the
situation depicted in the model: how well do the prepared
and the unprepared descriptions match? We have seen that
Louisell's treatment of the atoms is highly idealized. So too
are other aspects of his block diagram. In this sense, Louisell's
model for the 'causal behaviour' is not very realistic.
There is another, different way in which physicists use
the word 'realistic'. I will illustrate with three examples.
The first example comes from the laser engineering course
I referred to in the last eassy. 5 After several lectures on
4 Ibid., p. 470.
S Anthony Siegman, 'Lasers' (Electrical Engineering 231), Stanford Uni-
versity, Autumn term 1981-2.
classical electron oscillators, Professor Anthony Siegman

announced that he was ready to talk about the lasing medium
in a real laser. I thought he was going to teach about ruby
rods-that ruby is chromium doped sapphire, that the
3+ chromium ions are interspersed randomly at low den-
sities throughout the sapphire lattice, and that an ordinary
electric discharge is used to excite the chromium ions and
bring about a population inversion. Instead he began, 'Con-
sider a collection of two-level atoms.' In a sense he started
to talk about real lasers-a laser medium is indeed com-
posed of quantized systems like atoms, and not of the
fictitious electron oscillators. But in another sense he did
not: two-level atoms are but a crude stand-in for the intri-
cate and variegated structure of real laser materials.
The second example comes from a conversation with
my colleague Francis Everitt, an experimental physicist
whom I have mentioned before in connection with his
historical work on James Clerk Maxwell. In the last essay
we saw that a laser can be treated by van der Pol's equation:
in a variety of ways the laser will behave like a triode oscil-
lator in a d.c. circuit. In talking with Everitt I contrasted
this description with Louisell's. Louisell mentions real com-
ponents of the laser, like the atoms and the field. I took
Louisell's to be the more realistic description. Everitt agreed.
But he added, 'The reservoir is still only a model.' In Louisell's
diagram the damping reservoir represents the walls of the
cavity and the room in which it is housed. The three-level
atoms represent the lasing medium. In what sense is the
reservoir, unlike the atoms, just a model?
The third example gives an explicit clue. In the text
Quantum Optics John Klauder and E. C. G. Sudarshan
report, 'A number of authors have treated idealized inter-
acting systems as models for lasers'. 6 Louisell is an example.
Although highly idealized, the Louisell model is still realistic
in a way in which the models of Klauder and Sudarshan
are not. They themselves describe their models as 'phe-
nomenological'. What do they mean? They say that their
models are phenomenological because the models 'work
6 J. Klauder and E. C. G. Sudarshan, Quantum Optics (New York: Benjamin,
1968), p. 234.
directly on the state . . . as a function of time and do not

derive it as a solution to a Hamiltonian'." Recall that the
Hamiltonian goes into the Schroedinger equation and deter-
mines the time evolution of the state. It represents the
energies which guide the behaviour of the system. Sudarshan
and Klauder aim to get the right state; but they write down
this state directly, post hoc, with an eye to the behaviour
it is supposed to predict. They do not write down a Schroe-
dinger equation and derive the state as a solution to it; and
thus they do not show what energies produce the state.
Their treatment is unrealistic from the point of view of the
explanatory theory. It gives a theoretical description of the
behaviour, but nothing in the model shows what gives rise
to this behaviour.
Look back now to the damping reservoir, and recall our
discussion of atomic radiation in Essay 6. The effect of a
damping reservoir is to bring about an irreversible change in
the system which couples to it. Information which goes
into the reservoir gets lost there, and the memory of the
system is erased. The reservoir is a way of representing the
walls of the cavity and of the surrounding environment.
But it is like the proverbial black box. It generates the
effects that the walls are supposed to have, but there is no
representation of the method by which the walls produce
these effects. No description is given of how the walls are
made up, or of what gives rise to the formal characteristics
that reservoirs must have to bring about damping. This
constrasts with Siegman's treatment of the lasing medium.
Two-level atoms are not very much like chromium ions in
a ruby laser. But they do give rise to equations in accord
with established explanatory principles and not in an ad
hoc way.
The two senses of 'realistic' act at different levels. The
first bears on the relation between the model and the world.
The model is realistic if it presents an accurate picture of
the situation modelled: it describes the real constituents of
the system-the substances and fields that make it up- and
ascribes to them characteristics and relations that actually
obtain. The second sense bears on the relation between the
7 Ibid., p. 226.
model and the mathematics. A fundamental theory must

supply a criterion for what is to count as explanatory.
Relative to that criterion the model is realistic if it explains
the mathematical representation.
The two senses of realistic are nicely illustrated in Louisell's
treatment. We have already seen that Louisell's model is
only quasi-realistic in the first sense. It describes the signifi-
cant components, but the features it ascribes to them are a
caricature of those in reality. The model is both realistic
and unrealistic in the second sense as well. The first term
in Louisell's equation represents the potential arising from
the atom-field interaction which he sets down in the model.
That is what he means by saying that it represents the 'causal
behaviour'. The reservoir terms are different. They give rise
to the right solutions but no concrete mechanisms are sup-
plied in the model to explain them.
The two ways in which a model may be unrealistic are
related. Louisell's modelling of the reservoir is unrealistic
in the first sense as well as in the second, in part because
he does not intend to use the detailed structure of the
reservoir to generate his equation. But that is not all there
is to it. We say in Essay 6 that if the reservoir is really to
do its job in getting the atoms to decay, the time correlations
there must be exactly zero. This is an assumption that
Louisell makes; but it is highly unrealistic in the first sense.
This case is just like the infinite potentials in the last section
of the last essay. The conventional Schroedinger theory
cannot be fitted exactly to the situation, so we deal with
the problem by distorting the situation. But we put the
distortion as far away from the system of immediate con-
cern as possible. If we are interested in the atoms only,
we can put the distortion in the description of the field,
assigning it an infinite number of degrees of freedom. But
if we want to study the field as well, the infinite degrees of
freedom or. the zero time correlations are put into the walls
of the cavity and the surrounding environment. And so on.
We learn an important lesson about bridge principles
from these considerations. A treatment that is more realistic
in the second sense will employ more bridge principles. The
quantum statistical approach is a highly sophisticated method
for predicting fine details about photon statistics in laser

light. Even in an advanced treatment like this, a large part
of the work is done by phenomenological terms which
minimize the number of bridge principles needed. For
example, the phenomenological terms that Louisell employs
are from his general theory of damped systems and can be
employed again and again independent of how the damping
is brought about.
The first term of Louisell's equation also illustrates this
point about bridge principles. In the last essay I raised the
worry that the bridge principles I discussed there were too
elementary to be instructive. Louisell's equation shows that
this is not so. Only the first term is a genuine Hamiltonian
term matched by a conventional bridge principle with a
description of the potential. What Hamiltonian is it? It is
just the Hamiltonian for the interaction of an atom with
a radiation field, which appeared on our list in Essay 7 and
which was developed in a classic paper by Enrico Fermi
in 1932. The only change that Louisell makes is to sum
the Hamiltonian over all the atoms in the cavity. This bears
out my general claim about bridge principles. The success
of the quantum statistical treatment does not depend on
using novel principles that are highly involved, but rather
in using some well-known and well-understood principles
in a novel way.
2. THE SIMULACRUM ACCOUNT OF

EXPLANA TION
The conventional D-N account supposes that we have ex-
plained a phenomenon when we have shown how it follows
from a more fundamental law. This requires that the treat-
ments we give for phenomena in physics must certainly be
realistic in the first sense, and preferably in the second as
well, if they are to serve as explanations. I propose an alter-
native to the D-N model that brings the philosophic account
closer to explanatory practices in physics as I have pictured
them. It is based on Duhem's view of explanation, which I
sketched in Essay 5, and follows immediately from the
discussion of the last section.
The primary aim of this book is to argue against the

facticity of fundamental laws. As we saw in the very first
essay, one of the chief arguments that realists use in favour
of their facticity is their broad explanatory and predictive
success. I have been arguing here that the vast majority of
successful treatments in physics are not realistic. They are
not realistic in the first sense of picturing the phenomena
in an accurate way; and even in the second sense, too much
realism may be a stop to explanatory power, since the use
of 'phenomenological' terms rather than a more detailed
'causal' construction may allow us more readily to deploy
known solutions with understood characteristics and thereby
to extend the scope of our theory.
If what I say is correct, it calls for a new account of
explanation. Recall the discussion of Essay 6. To explain
a phenomenon is to find a model that fits it into the basic
framework of the theory and that thus allows us to derive
analogues for the messy and complicated phenomenological
laws which are true of it. The models serve a variety of
purposes, and individual models are to be judged according
to how well they serve the purpose at hand. In each case
we aim to 'see' the phenomenon through the mathematical
framework of the theory, but for different problems there
are different emphases. We may wish to calculate a particular
quantity with great accuracy,' or to establish its precise
functional relationship to another. We may wish instead
to replicate a broader range of behaviour, but with less
accuracy. One important thing we sometimes want to do
is to layout the causal processes which bring the phenomena
about, and for this purpose it is best to use a model that
treats the causally relevant factors as realistically as possible,
in both senses of 'realistic'. But this may well preclude
treating other factors realistically. We should not be misled
into thinking that the most realistic model will serve all
purposes best.
In order to stress this 'anti-realistic' aspect of models, I
call my view of explanation a 'simulacrum' account. The
second definition of 'simulacrum' in the Oxford English
Dictionary says that a simulacrum is 'something having
merely the form or appearance of a certain thing, without
without possessing its substance or proper qualities'i'' This is

just what I have been urging that models in phsyics are like.
Is a helium-neon laser really a van der Pol oscillator? Well,
it is really a mix of helium and neon atoms, in about the
ratio nine to one, enclosed in a cavity with smooth walls
and reflecting mirrors at both ends, and hooked up to· a
device to pump the neon atoms into their excited state.
It is not literally a triode oscillator in a d.c. circuit. If we
treat it with van der Pol's equation for a triode oscillator,
we will be able to replicate a good deal of its behaviour
above threshold, and that is our aim. The success of the
model depends on how much and how precisely it can
replicate what goes on.
A model is a work of fiction. Some properties ascribed
to objects in the model will be genuine properties of the
objects modelled, but others will be merely properties of
convenience. The term 'properties of convenience' was
suggested by H. P. Grice, and it is apt. Some of the proper-
ties and relations in a model will be real properties, in the
sense that other objects in other situations might genuinely
have them. But they are introduced into this model as a
convenience, to bring the objects modelled into the range
of the mathematical theory.
Not all properties of convenience will be real ones. There
are the obvious idealizations of physics-infinite potentials,
zero time correlations, perfectly rigid rods, and frictionless
planes. But it would be a mistake to think entirely in terms
of idealizations-of properties which we conceive as limiting
cases, to which we can approach closer and closer in reality.
For some properties are not even approached in reality.
They are pure fictions.
I would want to argue that the probability distributions
of classical statistical mechanics are an example. This is a
very serious claim, and I only sketch my view here as an
illustration. The distributions are essential to the theory-
they are what the equations of the theory govern-and
the theory itself is extremely powerful, for example in the
detailed treatment of fluid flow. Moreover, in some simple
special cases the idea of the probability distribution can be
8 The Oxford English Dictionary (Oxford: Oxford University Press, 1933).
operationalized; and the tests support the distributions

ascribed by the theory. 9
Nevertheless, I do not think these distributions are real.
Statistical mechanics works in a massive number of highly
differentiated and highly complex situations. In the vast
majority of these it is incredible to think that there is a true
probability distribution for that situation; and proofs that,
for certain purposes, one distribution is as good as another,
do not go any way to making it plausible that there is one
at all. It is better, I think, to see these distributions as fic-
tions, fictions that have a powerful organizing role in any
case and that will not mislead us too much even should we
take them to be real in the simple cases.
We can illustrate with Maxwell's treatment of the radio-
meter, described in the introduction to this book. Maxwell
begins with Boltzmann's equation (equation I, Introduction),
which governs the evolution of the velocity distribution
of the gas molecules. (This distribution gives the probability,
for every possible combination of values for v, w, x, ... ,
that the first molecule has velocity v; the second, velocity w;
the third velocity x; etc.) Maxwell writes down one of the
many functions which solve Boltzmann's equation and he
claims that this function is the distribution for 'a medium
in which there are inequalities of temperature and velocity'
and in which the viscosity varies 'as the first power of the
absolute temperature'. 10
I claim that the medium which Maxwell describes is only
a model. It is not the medium which exists in any of the
radiometers we find in the toy department of Woolworth's.
The radiometers on the shelves in Woolworth's do not have
delicate well-tuned features. They cost $2.29. They have a
host of causally relevant characteristics besides the two
critical ones Maxwell mentions, and they differ in these
characteristics from one to another. Some have sizeable
convection currents; in others the currents are negligible;
probably the co-efficients of friction between vanes and
9 See, for example, T. K. Roberts and A. R. Miller, Heat and Thermodynamics
(London: Blackie & Son, 1960).
10 J ames Maxwell, 'On Stresses in Rarified Gases arising from Inequalities of
Temperature', The Scientific Papers of James Clark Maxwell, ed. W. D. Niven
(New York: Dover Publications, 1965), pp. 691, 692.
gases differ; as do the conduction rates, the densities of the

enclosed gases, and the make-up of the gas itself.
We may be inclined to think that this does not matter
much. Maxwell has made a harmless idealization: the effects
of the other factors are small, and the true distribution
in each Woolworth radiometer is close enough for the pur-
poses at hand to the one Maxwell proposes. A simulacrum
account is unnecessary, the standard covering-law story
will do. But this is not so. For on the covering-law theory,
if Maxwell's treatment is to explain the rotation in a Wool-
worth radiometer, the radiometer must have a specific
distribution function and that function must be nomo-
logically linked to the conditions that obtain. But Maxwell's
theory records no such laws. The conditions in these radio-
meters are indefinitely varied and indefinitely complex, so
that a multitude of highly complicated unknown laws must
be assumed to save Maxwell's explanation. I think these laws
are a complete fiction. We cannot write them down. We
certainly cannot construct experiments to test them. Only
the covering-law model of explanation argues for their
existence.
Recall Hempel's worries about bridge principles, which I
discussed in the last essay. Hempel was concerned that bridge
principles do not have the proper exceptionless character
to ensure deductive connections between explanans and
explanandum. Hempel illustrated with magnets and iron
filings. But Maxwell's radiometer is as good an example.
Not all radiometers that meet Maxwell's two descriptions
have the distribution function Maxwell writes down; most
have many other relevant features besides. This will probably
continue to be true no matter how many further corrections
we add. In general, as Hempel concluded, the bridge law
between the medium of a radiometer and a proposed dis-
tribution can hold only ceteris paribus.
This, however, is a difficult position for a covering-law
theorist to take. As I argued early on, in the second essay,
a law that holds only in restricted circumstances can explain
only in those circumstances. The bulk of the radiometers
in Woolworth's are untouched by Maxwell's explanation.
The idealization story with which we began supposes that
each Woolworth radiometer has some distribution function

true of it and that the distribution functions in question
are sufficiently close to Maxwell's. In this case Maxwell's
explanation for the ideal medium serves as a proxy to the
proper explanations for each of the real radiometers. But
these last are no explanations at all on the covering-law
view unless the Book of Nature is taken up with volumes
and volumes of bridge laws.
I say there are no such bridge laws, or, more cautiously,
we have no reason to assume them. But without the bridge
laws, the distribution functions have no explanatory power.
Thus our chief motivation for believing in them vanishes,
and rightly so. The distribution functions play primarily
an organizing role. They cannot be seen; they cause nothing;
and like many other properties of convenience, we have no
idea how to apply them outside the controlled conditions
of the laboratory, where real life mimics explanatory models.
What is the distribution function for the molecules in this
room? Or the value of the electric field vector in the region
just at the tip of my pencil? These questions are queer.
They are queer because they are questions with no answers.
They ask about properties that only objects in models
have, and not real objects in real places.
I think we are often misled by a piece of backwards
reasoning here. Sometimes for a given model, it is possible
to contrive (or to find) a real situation in which the principal
features relevant to the phenomenology are just the features
mentioned in the model, and no others. Low density helium,
for example, is an almost ideal gas from the point of view
of the billiard ball model of statistical mechanics. In these
cases, we are inclined to think of the model as an exact
replica of reality, and to attribute to the objects modelled
not only the genuine properties of the model, but also the
properties of convenience. By continuity, we then argue,
the properties of convenience must apply to more complex
cases as well. But this is just backwards. With a good many
abstract theoretical properties, we have no grounds for
assigning them to complex, realistic cases. By continuity,
they do not apply to the ideal cases either.
Returning to models, it may help to recall a disagreement
between Mary Hesse!' and Wilfrid Sellars. 12 Hesse's paradigm

is the billiard ball model for the kinetic theory of gases.
She thinks that the objects in the model (the billiard balls)
and the objects modelled (the molecules of gas) share some
properties and fail to share others; and she talks in terms
of the positive, negative, and neutral analogies between the
model and the objects modelled. Sellars disagrees. His atten-
tion is one level up. What is important for Sellars is not the
sharing of properties, but the sharing of relationships among
properties. I take it that our laser example would suit Sellars
well. The helium-neon laser and a real triode oscillator
need have no properties in common. What is relevant is that
the properties each has behave in similar ways, so that
both can be treated by the same van der Pol equation.
I share Sellars's stress on the relations among properties,
for the point of the kind of models I am interested in is to
bring the phenomenon under the equations of the theory.
But Sellars and I are opposed over realism. He sees that
phenomenological laws are hard to get right. If we want
regular behaviour, the description of the circumstances
must grow more and more complicated, the laws have less
and less generality, and our statements of them will never
be exceptionless. Fundamental laws, by contrast, are simple,
general, and without exception. Hence for Sellars they are
the basic truths of nature.
In opposition to Sellars, I have been arguing that their
generality and exceptionlessness is mere appearance, appear-
ance which arises from focusing too much on the second
stage of theory entry. The fundamental equations may be
true of the objects in the model, but that is because the
models are constructed that way. To use the language I
introduced in the last essay, when we present a model of
a phenomenon, we prepare the description of the phe-
nomenon in just the right way to make a law apply to it.
The problem for realism is the first stage of theory entry.
If the models matched up one-to-one, or at least roughly
11 Mary Hesse, Models and Analogies in Science (Notre Dame: University

of Notre Dame Press, 1966).
12 Wilfrid Sellars, Philosophical Perspectives (Reseda, California: Ridgeview
Press, 1977), Ch. VIV, 'Scientific Realism or Irenic Instrumentalism'.
SO, with the situations we study, the laws which govern the
model could be presumed to apply to the real situations as
well. But models are almost never realistic in the first sense;
and I have been arguing, that is crucial to how physics works.
Different incompatible models are used for different pur-
poses; this adds, rather than detracts, from the power of the
theory. We have had many examples already but let me
quote one more text describing the variety of treatments
available for lasers:
A number of authors have treated idealized interacting systems as
models for lasers. Extensive studies have been carried out by Lax,
Scully and Lamb, Haken, Sauerman, and others. Soluble models have
been examined by Schwab le and Therring. Several simplified dynamical
models for devices of various sorts are given in the last chapter of
l..ouisell's book.!"
And so on.
There has been a lot of interest in models among philo-
sophers of science lately. It will help to compare the use I
make of models with other accounts. First, Redhead and
Cushing: both Michael Redhead 14 and James Cushing."
have recently done very nice investigations of models in
mathematical physics, particularly in quantum mechanics
and in quantum field theory. Both are primarily concerned
not with Hesse's analogical models, but with what Redhead
calls 'theoretical models', or incomplete theories (Cushing's
modelj-e-guinea-pig or tinker toy models). Although, like
me, Cushing explicitly says that models serve to 'embed'
an account of the phenomena into a mathematical theory,
he and Redhead concentrate on a special kind of model-a
theory which is admittedly incomplete or inaccurate. I am
concerned with a more general sense of the word 'model'.
I think that a model-a specially prepared, usually fictional
description of the system under study-is employed when-
ever a mathematical theory is applied to reality, and I use
the word 'model' deliberately to suggest the failure of exact
IS Klauder and Sudarshan, op, cit., pp. 234-5.

14 Michael Redhead, 'Models in Physics', British Journal for the Philosophy
of Science 31 (1980), pp. 145-63.
I' James Cushing, 'Models and Methodologies in Current Theoretical High-
Energy Physics', Synthese 50 (1982), pp. 5-101.
correspondence which simulacra share with both Hesse's

analogical models and with Redhead and Cushing's theo-
retical models.
Secondly, the semantical view of theories: on the simu-
lacrum account, models are essential to theory. Without
them there is just abstract mathematical structure, formulae
with holes in them, bearing no relation to reality. Schroe-
dinger's equation, even coupled with principles which tell
what Hamiltonians to use for square-well potentials, two-
body Coulomb interactions, and the like, does not constitute
a theory of anything. To have a theory of the ruby laser, or
of bonding in a benzene molecule, one must have models
for those phenomena which tie them to descriptions in the
mathematical theory. In short, on the simulacrum account
the model is the theory of the phenomenon. This sounds
very much like the semantic view of theories, developed
by Suppes!" and Sneed 17 and van Fraassen.l" But the em-
phasis is quite different. At this stage I think the formal
set-theoretic apparatus would obscure rather than clarify
my central points. It is easiest to see this by contrasting
the points I want to make with the use to which van Fraassen
puts the semantic view in The Scientific Image. 19
Van Fraassen holds that we are only entitled to believe
in what we can observe, and that we must remain agnostic
about theoretical claims which we cannot confirm by obser-
vation. This leads him to require that only the observable
substructure of models permitted by the laws of a theory
should map onto the structure of the situations modelled.
Only that part of a theory which is supposed to represent
observable facts, and not the parts that represent theoretical
facts, need be an accurate representation of how things
really are.
Van Fraassen's book takes a firm stand against realism.
Sellars, I have mentioned, is a profound realist. But they
16 Patrick Suppes, 'Set Theoretic Structures in Science', mimeographed
(Stanford: Stanford University Press, 1967).
17 Joseph Sneed, The Logical Structure of Mathematical Physics (Dordrecht:
Reidel, 1971).
18 Bas van Fraassen, The Scientific Image (Oxford: Clarendon Press, 1980).
See also 'On the Extension of Beth's Semantic Theories', Philosophy of Science
37 (1970), pp. 325-39. 19 Ibid.
have in common a surprising respect for theory. Both expect

that theories will get the facts right about the observable
phenomena around us. For van Fraassen, the theoretical
claims of a good theory need not match reality, but the
claims about observables should. In a good theory, the
observable substructure prescribed by the theory should
match the structure of reality. This is not how I see good
theories ·working. The observational consequences of the
theory may be a rough match to what we suppose to be
true, but they are generally not the best we can do. If we
aim for descriptive adequacy, and do not care about the
tidy organization of phenomena, we can write better phe-
nomenological laws than those a theory can produce. This is
what I have tried to show, beginning with 'Truth Doesn't
Explain Much' and ending with the prepared, but inaccurate
descriptions discussed in the last essay.
There is also a second important difference with van
Fraassen that does not fit readily into the semantic formalism.
I have talked about observational substructures in order to
contrast my views with van Fraassen's. But unlike van
Fraassen, I am not concerned exclusively with what can be
observed. I believe in theoretical entities and in causal pro-
cesses as well. The admission of theoretical entities makes
my view much closer to Sellars than it earlier sounded. All
sorts of unobservable things are at work in the world, and
even if we want to predict only observable outcomes, we will
still have to look to their unobservable causes to get the
right answers.
I want to focus on the details of what actually happens
in concrete situations, whether these situations involve
theoretical entities or not, and how these differ from what
would happen if eves the best of our fundamental laws
played out their consequences rigorously. In fact, the simu-
lacrum account makes the stronger claim: it usually does
not make sense to talk of the fundamental laws of nature
playing out their consequences in reality. For the kind of
antecedent situations that fall under the fundamental laws
are generally the fictional situations of a model, prepared
for the needs of the theory, and not the blousy situations
of reality. I do not mean that there could never be situations
to which the fundamental laws apply. That is only precluded

if the theory employs properties or arrangements which are
pure fictions, as I think classical statistical mechanics does.
One may occur by accident, or, more likely, we may be able
to construct one in a very carefully controlled experiment,
but nature is generally not obliging enough to supply them
freely.
Let me repeat a point I have made often before. If we are
going to argue from the success of theory to the truth of
theoretical laws, we had better have a large number and a
wide variety of cases. A handful of careful experiments will
not do; what leads to conviction is the widespread applica-
tion of theory, the application to lasers, and to transistors,
and to tens of thousands of other real devices. Realists
need these examples, application after application, to make
their case. But these examples do not have the right structure
to support the realist thesis. For the laws do not literally
apply to them.
The simulacrum account is not a formal account. It says
that we layout a model, and within the model we 'derive'
various laws which match more or less well with bits of
phenomenological behaviour. But even inside the model,
derivation is not what the D-N account would have it be,
and I do not have any clear alternative. This is partly
because I do not know how to treat causality. The best
theoretical treatments get right a significant number of phe-
nomenological laws. But they must also tell the right causal
stories. Frequently a model which is ideal for one activity
is ill-suited to the other, and often, once the causal principles
are understood from a simple model, they are just imported
into more complex models which cover a wider variety of
behaviour. For example, Richard Feynman, when he deals
with light refraction in Volume II of his famous Berkeley
lectures, says:
We want now to discuss the phenomenon of the refraction of light ...
by dense materials. In chapter 31 of Volume I we discussed a theory of
the index of refraction, but because of our limited mathematical
abilities at that time, we had to restrict ourselves to finding the index
only for materials of low density, like gases. The physical principles
that produced the index were, however, made clear ... Now, however,
we will find that it is very easy to treat the problem by the use of differ-
ential equations. This method obscures the physical origin of the index
(as coming from the re-radiated waves interfering with the original
waves), but it makes the theory for dense materials much simpler.i"
But what is it for a theoretical treatment to 'tell' a causal
story? How does Feynman's study of light in Volume I
'make clear' the physical principles that produce refraction?
I do not have an answer. I can tell you what Feynman
does in Volume I, and it will be obvious that he succeeds in
extracting a causal account from his model for low density
materials. But I do not have a philosophic theory about
how it is done. The emphasis on getting the causal story
right is new for philosophers of science; and our old theories
of explanation are not well-adapted to the job. We need a
theory of explanation which shows the relationship between
causal processes and the fundamental laws we use to study
them, and neither my simulacrum account nor the tradi-
tional covering-law account are of much help.
Causal stories are not the only problem. Even if we want
to derive only pure Humean facts of association, the D-N
account will not do. We have seen two ways in which it
fails in earlier chapters. First, from Essay 6, the fundamental
laws which start a theoretical treatment frequently get
corrected during the course of the derivation. Secondly,
many treatments piece together laws from different theories
and from different domains, in a way that also fails to be
deductive. This is the theme of Essay 3.
These are problems for any of our existing theories of expla-
nation, and nothing I have said about simulacra helps to solve
them. Simulacra do a different job. In general, nature does
not prepare situations to fit the kinds of mathematical theories
we hanker for. We construct both the theories and the objects
to which they apply, then match them piecemeal onto real
situations, deriving-sometimes with great precision-a bit
of what happens, but generally not getting all the facts straight
at once. The fundamental laws do not govern reality. What they
govern has only the appearance of reality and the appearance
is far tidier and more readily regimented than reality itself.
20 Richard Feynman, The Feynman Lecture on Physics, vol. II (Reading,
Mass: Addison-Wesley, 1964), p. 32.1.
Essay 9
How the Measurement Problem

is an Artefact of the Mathematics
O. INTRODUCTION
Von Neumann's classic work of 1932 set the measurement
problem in quantum mechanics.' There are two kinds of
evolution in the quantum theory, von Neumann said. The
first kind is governed by Schroedinger's equation. It is
continuous and deterministic. The second, called reduction
of the wave packet, is discontinuous and indeterministic.
It is described by von Neumann's projection postulate.
The terminology arises this way: classical systems have
well-defined values for both position and momentum.
Quantum systems, on the other hand, may have no well-
defined value for either. In this case they are said to be
in a superposition of momentum states, or of position
states. Consider position. An electron will frequently behave
not like a particle, but like a wave. It will seem to be spread
out in space. But when a measurement is made we always
find it at a single location. The measurement, we say, reduces
the wave packet; the electron is projected from a wave-like
state to a particle-like state.
But what is special about measurement? Most of the time
systems are governed by the Schroedinger equation. Reduc-
tions occur when and only when a measurement is made.
How then do measurements differ from other interactions
between systems? It turns out to be very difficult to find
any difference that singles out measurements uniquely.
Von Neumann postulated that measurements are special
because they involve a conscious observer. Eugene Wigner
concurred f and I think this is the only solution that succeeds
in picking measurements from among all other interactions.
1 John von Neumann, Mathematical Foundations of Quantum Mechanics

(Princeton: Princeton University Press, 1955). First published in German, 1932.
2 E. Wigner, 'Remarks on the Mind-Body Question', item 98 in I. J. Good
(00.), The Scientist Speculates (London: Heinemann, 1961).
164 THE MEASUREMENT PROBLEM
It is clearly not a very satisfactory solution. The measurement

problem is one of the long-standing philosophical difficulties
that troubles the quantum theory.
I will argue here that the measurement problem is not a
real problem. There is nothing special about measurement.
Reductions of the wave packet occur in a wide variety of
other circumstances as well, indeterministically and on their
own, without the need for any conscious intervention.
Section I argues that reductions occur whenever a system
is prepared to be in a given microscopic state. Section 2 urges
that reductions occur in other transition processes as well,
notably in scattering and in decay. There is good reason
for attending to these transition processes. On the conven-
tional interpretation, which takes position probabilities as
primary, quantum propositions have a peculiar logic or a
peculiar probability structure or both. But transition pro-
cesses, where reductions of the wave packet occur, have
both a standard logic and a standard probability. They
provide a non-problematic interpretation for the theory.
The proposal to develop an interpretation for quantum
mechanics based on transition probabilities seems to me
exactly right. But it stumbles against a new variation of
the measurement problem. Two kinds of evolution are
postulated. Reductions of the wave packet are no longer
confined to measurements, but when do they occur? If
there are two different kinds of change, there must be
some feature which dictates which situations will be governed
by Schroedinger's law and which by the projection postulate.
The very discipline that best treats transitions suggests an
answer. Quantum statistical mechanics provides detailed
treatments of a variety of situations where reduction seems
most plausibly to occur. This theory offers not two equations,
but one. In general formulations of quantum statistical
mechanics, Schroedinger's evolution and reduction of the
wave packet appear as special cases of a single law of motion
which governs all systems equally. Developments of this
theory offer hope, I urge in Section 4, for eliminating the
measurement problem and its variations entirely. The two
evolutions are not different in nature; their difference is an
artefact of the conventional notation.
THE MEASUREMENT PROBLEM 165
1. IN DEFENCE OF REDUCTION OF THE

WAVE PACKET
In 1975 I wrote 'Superposition and Macroscopic Observation'.
Here is the problem that I set in that paper:
Macroscopic states, it appears, do not superpose. Macroscopic bodies

seem to possess sharp values for all observable quantities simultan-
eously.
But in at least one well-known situation-that of measurement-
quantum mechanics predicts a superposition. It is customary to try to
reconcile macroscopic reality and quantum mechanics by reducing
the superposition to a mixture. This is a program that von Neumann
commenced in 1932 and to which Wigner, Groenewold, and others
have contributed. Von Neumann carried out his reduction by treating
measurement as a special and unique case that is not subject to the
standard laws of quantum theory. Subsequent work has confirmed
that a normal Schroedinger evolution cannot produce the required
mixture. This is not, however, so unhappy a conclusion as is usually
made out. Quantum mechanics requires a superposition: the philo-
sophical problem is not to replace it by a mixture, but rather to explain
why we mistakenly believe that a mixture is called for. 3
These are the words of a committed realist: the theory

says that a superposition occurs; so, if we are to count the
theory a good one, we had best assume it does so.
Today I hold no such view. A theory is lucky if it gets
some of the results right some of the time. To insist that a
really good theory would do better is to assume a simplicity
and uniformity of nature that is belied by our best attempts
to deal with it. In quantum mechanics in particular I think
there" is no hope of bringing the phenomena all under the
single rule of the one Schroedinger equation. Reduction
of the wave packet occurs as well, and in no systematic or
uniform way. The idea that all quantum evolutions can be
cast into the same abstract form is a delusion. To under-
stand why, it is important to see how the realist programme
that I defended in 1975 fails.
Quantum systems, we know, are not supposed to have
3 N. Cartwright, 'Superposition and Macroscopic Observation', in Patrick C.
Suppes (ed.), Logic and Probability in Quantum Mechanics (Dordrecht: D.
Reidel Publishing Co.), pp. 221-34.
I
H<,C/C~C/Br
II
c
I
c
H/ """C~ <,Br
(b) 1
FIG 9.1.0rthodibromobenzene (Source:Feynman, Lectures on Physics.)
well-defined values for all of their variables simultaneously.

A system with a precise momentum will have no position.
It will behave like a wave, and be distributed across space.
This is critical for the explanations that quantum mech-
anics gives for a variety of phenomena. The famous case
of the bonding energy in benzene is a good example. This
energy is much lower than would be expected if there were
double bonds where they should be between the carbon
atoms. The case is easier to see if we look at orthodibromo-
benzene-that is, benzene with two bromines substituted
for two hydrogens. There are two possible structures (see
Figure 9.1).
In fact all the forms in nature are the same. Do they have
a single bond separating the bromines, as in 9.1a, or a double
bond, as in 9.1b? Neither. The energy is in between that
predicted from one bond and that predicted from two. It
is as if there is a bond and a half. Linus Pauling said that each
molecule resonates between the first arrangement and the
second, and that is not far off the quantum answer. Accord-
ing to quantum mechanics the molecule is in a superposition
of the two configurations. That means that it is neither
in configuration 9.1a entirely, nor in 9.1b entirely, but in
both simultaneously. Where then are the electrons that
form the bonds? On a classical particle picture they must
be located at one point or another. Quantum mechanics
says that, like waves, they are smeared across the alternative
locations.
This behaviour is not too disturbing in electrons. After
all, they are very small and we do not have very refined
intuitions about them. But it is not a correct account for

macroscopic objects, whose positions, velocities, and such
are narrowly fixed. How can we prevent smearing in macro-
scopic objects? First, we notice that smearing across one
variable, like position, occurs only when an incompatible
quantity, like momentum, is precisely defined. This suggests
that all macroscopic observables are compatible. We can
arrange this in the following way: admittedly, the real
quantities that describe macroscopic systems will not all
be compatible. But we do not actually observe the real
value of a quantity at a time. Instead what we see is a long-
time average over these values-long compared to relaxa-
tion times in the objects themselves. By a coarse-grained
averaging, we can construct new quantities that are all
compatible. We then claim that these new quantities, and
not the originals, are the macroscopic observables that
concern us.
But this is not enough. The construction ensures that
it is possible for macroscopic objects to be in states with
well-defined values for all macroscopic observables. It is
even possible that they would always evolve from one such
state into another if left to themselves. But interactions
with microscopic objects bring them into superpositions.
That is what happens in a measurement according to the
Schroedinger equation. The electron starts out in a super-
position, with the apparatus in its ground state. Together
the composite of the two finishes after the measurement in
a superposition, where the pointer of the apparatus has no
well-defined position but is distributed across the dial.
This is why von Neumann postulated that measurement
is special. In the end measurement interactions are not
governed by the Schroedinger equation. After the measure-
ment has ceased, a new kind of change occurs. The super-
posed state of the apparatus-plus-object reduces to one of
the components of the superposition. This kind of change
is called reduction of the wave packet, and the principle
that governs it is the projection postulate. The process is
indeterministic: nothing decides which of the final states
will occur, although the probabilities for each state are ante-
cedently determined. In an ensemble of similar measurement
interactions, there will be a mix of final states. So we some-

times say that a reduction of the wave packet takes super-
positions into mixtures. But it is important to remember
that this is a description entirely at the level of the ensemble.
Individually, reduction of the wave packet takes super-
positions into components of the superposition.
There are two obvious tests to see if von Neumann is
right, and both tell in favour of reduction of the wave packet.
The first looks at individuals. If the pointer is indeed in a
superposition after a measurement, it should somehow be
distributed across space. But in fact we always observe
the pointer at a definite place. The second looks at en-
sembles. Mixtures and superpositions generate different
statistical predictions about the future behaviour of the
ensemble. Collections of macroscopic systems always behave,
statistically, as if they were in mixtures. In 'Superposition
and Macroscopic Observation' I defend the superposition
against both of these difficulties.
The first defence consists in 'breaking the eigenvalue-
eigenvector link'. The discussion so far has assumed that the
electron has no position value because it is in a superposition
of position states. We have thus been adopting the principle:
S has a value for a given observable (an eigenvalue) if and only
if S is in the corresponding state (the eigenstate). To deny the
inference in the direction from state to value would require
serious revision of our understanding of quantum theory,
which teaches that the probability in a given eigenstate for a
system to exhibit the corresponding eigenvalue is one. But
there is no harm in breaking the inference in the other
direction. One must be careful not to run foul of various
no-hid den-variable proofs, like those of Kochen and Specker"
and of J. S. Bell.f But there are a variety of satisfactory ways
of assigning values to systems in superpositions. 6
Let us turn to the second difficulty. Why are physicists
4 S. Kochen and E. P. Specker, 'The Problem of Hidden Variables in Quantum

Mechanics', Journal of Mathematics and Mechanics 17 (1967), pp. 59-87.
5 J. S. Bell, 'On the Problem of Hidden Variables in Quantum Mechanics',
Reviews of Modern Physics 38 (1966), pp.447-52.
6 See the discussion in 'Superposition and Macroscopic Observation' and
the references given there.
not much concerned with the measurement problem in
practice? Here is a common answer: macroscopic objects
have a very large number of degrees of freedom with ran-
domly distributed phases. The effect of averaging over all
of these degrees of freedom is to eliminate the interference
terms that are characteristic of superpositions. In a very
large system with uncorrelated phases the superposition
will look exactly like a mixture. This argument is just like
the one described in Essay 6 for deriving the exponential
decay law. The atom is in interaction with the electro-
magnetic field. If it interacted with only one mode, or a
handful of modes, it would oscillate back and forth as if it
were in a superposition of excited and de-excited states.
(There is a very nice discussion of this in P. C. W. Davies,
The Physics of Time Asymmetry.f In fact it is in inter-
action with a 'quasi-continuum' of modes, and averaging
over them, as we do in the Weisskopf-Wigner approximation,
eliminates the terms that represent the interference. An
ensemble of such atoms evolving over time will look exactly
as if it is in a mixture of excited and de-excited states, and
not in a superposition.
There have been various attempts to apply these ideas
more rigorously to measurement situations. The most de-
tailed attempt that I know is in the work of Danieri, Loinger,
and Prosperi. 8 This is what I describe in 'Superposition and
Macroscopic Observation'. Daneri, Loinger, and Prosperi
propose an abstract model of a measurement situation, and
consider what happens in this model as the macroscopic
apparatus comes to equilibrium after the interaction. They
conclude that the superposition that the Schroedinger
equation predicts will be statistically indistinguishable
from the mixture predicted by the projection postulate. This
is not to say that the superposition and the mixture are the
same. There is no way under the Schroedinger equation
7 P. C. W. Davies, The Physics of Time Asymmetry (Berkeley and Los

Angeles: University of California Press, 1974, 1977), section 6.1.
8 A. Daneri, A. Loinger, and G. M. Prosperi, 'Quantum Theory of Measure-
ment and Ergodicity Conditions', Nuclear Physics 33 (1962), pp. 297-319;
'Further Remarks on the Relation Between Statistical Mecahnics and Quantum
Theory of Measurement', Il Nuovo Cimento 44B (1966), pp. 119-28.
for the system to end up in a genuine mixture. The claim of

Daneri, Loinger, and Prosperi is weaker. Superpositions
and mixtures make different statistical predictions. But in
this case, the two will agree on predictions about all macro-
scopic observables. The superposition and the mixture
are different, but the difference is not one we can test
directly with our macroscopic instruments.
Formally, there is an exact analogue between measure-
ment in the treatment of Daneri, Loinger, and Prosperi
and the case of exponential decay. In deriving the exponen-
tial decay law, we do not produce, in approximation, a
mixture as the final state. Rather we show that the super-
position is indistinguishable from a mixture with respect
to a certain limited set of test variables.
Jeffrey Bub9 and Hilary Putnam '? have both attacked the
Daneri, Loinger, and Prosperi programme. They do not
object to the details of the measurement model. Their
objection is more fundamental. Even if the model is satis-
factory, it does not solve the problem. A superposition is
still a superposition, even if it dresses up as a mixture. We
want to see a mixture at the conclusion of the measurement.
It is no credit to the theory that the superposition which it
predicts looks very much like a mixture in a variety of tests.
I now agree that a correct account will produce a mixture
for the final state. But in 1975 I thought that Bub and
Putnam had the situation backwards. Theirs is exactly the
wrong way for a realist to look at the work of Daneri,
Loinger, and Prosperi. The theory predicts a superposition,
and we should assume that the superposition occurs. The
indistinguishability proof serves to account for why we are
not led into much difficulty by the mistaken assumption
that a mixture is produced.
That was the argument I gave in 1975 in defence of the
Schroedinger equation. Now I think that we cannot avoid
admitting the projection postulate, and for a very simple
reason: we need reduction of the wave packet to account
9 J. Bub, 'The Daneri-Loinger-Prosperi Quantum Theory of Measurement',

Il Nuovo Cimento 57B (1968), pp. 503-20.
10 H. Putnam, 'A Philosopher Looks at Quantum Mechanics', in R. G. Colodny
(ed.), Beyond the Edge of Certainty (Englewood Cliffs: Prentice-Hall, 1965).
for the behaviour of individual systems over time. Macro-

scopic systems have orderly histories. A pointer that at one
instant rests in a given place is not miraculously somewhere
else the next. That is just what reduction of the wave packet
guarantees. After the measurement the pointer is projected
onto an eigenstate of position. Its behaviour will thereafter
be governed by that state-and that means it will behave
over time exactly as we expect.
There is no way to get this simple result from a super-
position. Following the lines of the first defence above, we
can assign a location to the pointer even though it is in a
superposition. An instant later, we can again assign it a
position. But nothing guarantees that the second position
will be the time-evolved of the first. The Daneri, Loinger, and
Prosperi proof, if it works, shows that the statistical dis-
tributions in a collection of pointers will be exactly those
predicted by the mixture at any time we choose to look.
But it does not prove that the individuals in the collection
will evolve correctly across time. Individuals can change
their values over time as erratically as they please, so long
as the statistics on the whole collection are preserved.
This picture of objects jumping about is obviously a
crazy picture. Of course individual systems do not behave
like that. But the assumption that they do not is just the
assumption that the wave packet is reduced. After measure-
ment each and every individual system behaves as if it really
is in one or another of the components and not as if it is
in a superposition. There are in the literature a few treat-
ments for some specific situations that attempt to replicate
this behaviour theoretically but there is no good argument
that this can be achieved in general. One strategy would be
to patch up the realist account by adding constraints: in-
dividual values are to be assigned in just the right way to
guarantee that they evolve appropriately in time. But if the
superposition is never again to play a role, and in fact the
systems are henceforth to behave just as if they were in
the component eigenstates, to do this is just to admit reduc-
tion of the wave packet without saying so. On the other hand
what role can the superposition play? None, if macroscopic
objects have both well-defined values and continuous histories.
So far we have been concentrating on macroscopic objects

and how to keep them out of superpositions. In fact we need
to worry about microscopic objects as well. Quantum
systems, we have seen, are often in superpositions. The
electron is located neither on one atom nor on another in
the benzene molecule. In the next section we will see another
example. When an electron passes a diffraction grating, it
passes through neither one or another of the openings in the
grating. More like a wave, it seems to go through all at once.
How then do we get micro systems out of superpositions
and into pure states when we need them? To study the
internal structure of protons we bombard them at high
velocity with electrons. For this we need a beam, narrowly
focused, of very energetic electrons; that is, we need electrons
with a large and well-defined momentum. The Stanford
Linear Accelerator (SLAC) is supposed to supply this need.
But how does it do so?
Here is a simple model of a linear accelerator (a 'drift tube
linac')-see Figure 9.2. The drift tube linac consists of two
major parts: an injector and a series of drift tubes hooked
up to an alternating voltage. The injector is a source of well-
collimated electrons-that is, electrons which have a narrow
spread in the direction of their momentum. We might get
them this way: first, boil a large number of electrons off a hot
wire. Then accelerate them with an electrostatic field-for
example in a capacitor. Bend them with a magnet and screen
all but a small range of angles. The resulting beam can be
focused using a solenoid or quadrature magnets before
entering the drift tube configuration for the major part of
the particle's acceleration.
Arrows represent field lines at one instant of time
FIG 9.2. Model of a linear accelerator

Inside the drift tubes the electric field is always zero.

In the gaps it alternates with the generator frequency. Con-
sider now a particle of charge e that crosses the first gap
at a time when the accelerating field is at its maximum.
The length L of the next tube is chosen so that the particle
arrives at the next gap when the field has changed sign.
So again the particle experiences the maximum accelerating
voltage, having already gained an energy 2eVo. To do this,
L must be equal to t vT where v is particle velocity and T
the period of oscillation. Thus L increases up to the limiting
value: for v ~ c, L ~! cT.
The electron in the linac is like a ball rolling down a
series of tubes. When it reaches the end of each tube, it
drops into the tube below, gaining energy. While the ball
is rolling down the second tube, the whole tube is raised
to the height of the first. At the end the ball drops again.
While it traverses the tube the ball maintains its energy.
At every drop it gains more. After a two-mile acceleration
at SLAC, the electrons will be travelling almost at the speed
of light.
Imagine now that we want to bring this whole process
under the Schroedinger equation. That will be very com-
plicated, even if we use a simple diagrammatic model.
Quantum systems are represented by vectors in a Hilbert
space. Each system in the process is represented in its own
Hilbert space-the emitting atoms, the electrostatic field,
the magnetic field, the absorbing screen, and the fields in
each of the separate drift tubes. When two systems interact,
their joint state is represented in the product of the Hilbert
spaces for the components. As we saw in the discussion of
measurement, the interaction brings the composite system
into a superposition in the product space for the two. When
a third system is added, the state will be a superposition in
the product of three spaces. Adding a fourth gives us a
superposition in a four-product space, and so on. At the end
of the accelerator we have one grand superposition for the
electrons plus capacitor plus magnet plus field-in-the-first-
tube plus field-in-the-second, and so forth.
The two accounts of what happens are incompatible. We
want an electron travelling in a specified direction with a
well-defined energy. But the Schroedinger equation predicts

a superposition for an enormous composite, in which the
electron has neither a specific direction nor a specific energy.
Naively, we might assume that the problem is solved early
in the process. A variety of directions are represented when
the electrons come off the wire or out of the capacitor, but
the unwanted angles are screened out at the absorbing plate.
If the formal Schroedinger treatment is followed, this solu-
tion will not do. The absorbing screen is supposed to produce
a state with the momentum spread across only a very small
solid angle. On the Schroedinger formalism there is no way
for the absorbing screen to do this job. Interaction with
the screen produces another superposition in a yet bigger
space. If we are to get our beam at the end of the accelerator,
Schroedinger evolution must give out. Reduction of the
wave packet must occur somewhere in the preparation
process.
This kind of situation occurs all the time. In our labora-
tories we prepare thousands of different states by hundreds
of different methods every day. In each case a wave packet
is reduced. Measurements, then, are not the only place to
look for a failure of the Schroedinger equation. Any success-
ful preparation will do.
2. WHY TRANSITION PROBABILITIES

ARE FUNDAMENTAL
Position probabilities play a privileged role in the interpreta-

tion of quantum mechanics. Consider a typical text. In the
section titled 'The Interpretation of 'l' and the Conservation
of Probability' Eugen Merzbacher tells us, about the wave-
function, 'l':
Before going into mathematical details it would, however, seem wise
to attempt to say precisely what 'l' is. We are in the paradoxical situa-
tion of having obtained an equation [the Schroedinger equation] which
we believe is satisfied by this quantity, but of having so far given only
a deliberately vague interpretation of its physical significance. We
have regarded the wave function as 'a measure of the probability'
of finding the particle at time t at the position r. How can this state-
ment be made precise?
'It itself obviously cannot be a probability. All hopes we might have
entertained in that direction vanished when 'It became a complex
function, since probabilities are real and positive. In the face of this
dilemma the next best guess is that the probability is proportional to
l'lt 12, the square of the amplitude of the wave function ...
Of course, we were careless when we used the phrase 'probability
of finding the particle at position r.' Actually, all we can speak of
is the probability that the particle is in a volume element d3r which
contains the point r. Hence, we now try the interpretation that
l'lt(r, t)12d3r is proportional to the probability that upon a measure-
ment of its position the particle will be found in the given volume
element. The probability of finding the particle in some finite region
of space is proportional to the integral of 'It*'lt over this region.'!
So I'¥(r) P is a probability. That is widely acknowledged.

But exactly what is 1'¥(r)12d3r the probability of? What is
the event space that these quantum mechanical probabilities
range over? There is no trouble-free answer. Merzbacher
makes the conventional proposal, that 'I'l'(r, t)12d3r is pro-
portional to the probability that upon a measurement of its
position the particle will be found in the given volume
element.' But why must we refer to measurement? We
should first rehearse the difficulties that beset the far simpler
proposal that I'¥(r)12d 3r represents the probability that the
particle is located in the region around r, measured or not.
This answer supposes that the events which quantum
probabilities describe are the real locations of quantum
objects. It has a hard time dealing with situations where
interference phenomena are significant-for instance,
the bonding of atoms in benzene molecule, or the wave-
like patterns which are observed in a diffraction experiment.
The two-slit experiment is the canonical example. A
beam of electrons passes through a screen with two holes
and falls onto a photographic plate. Imagine that, at the
time of passing the screen, each electron is located either
at slit I (r = Sl) or at slit 2 (r = S2)' Then we may reason
as follows. An electron lands at a point y on the screen if
and only if it lands at y and either passes through slit one
or slit two, i.e.
11 E. Merzbacher, Quantum Mechanics, Second Edition (Englewood Cliffs:
Prentiee-Hall1970), p. 36.
So
It follows that
C3: Prob(y) = Prob{ (y & Sl) V (y & S2)}
and, since Sl and S2are disjoint

C4: Prob(y) = Prob(y & Sl) + Prob(y & S2)
= Prob(y/sl)Prob(sl) + Prob(y/s2)Prob(s2)'
Now do the quantum mechanical calculation. If the source
is equidistant from both slit I and slit 2, the probability of
passing through slit I = probability of passing through
slit 2 = t, and the state of the electrons at the screen will
be 'I!(y) = l/y2¢1(y) + l/y2¢2(y), where ¢1(Y) is the
state of an electron beam which has passed entirely through
slit 1; ¢2(Y), a beam passing entirely through slit 2. So,
using the quantum mechanical rule
Q: Prob(y) = IljI(y) 12= II/y2¢1(y) + l/y2¢2(y)12 =
(y) 12 + t 1¢2(y) 12
t k/>l + t¢l (y )¢t(y) + t¢I*(y)¢2(y)'
But identifying
Q2: t = Probts.),
and similarly for S2, we see that the classical calculation C
for the probability oflanding aty, and the quantum mechanical
calculation Q do not give the same results. They differ
by the interference terms t¢l(Y)¢i'(Y) + t¢r(y)¢2(y). The
C calculation is the one that supposes that the electrons have
a definite location at the screen and its consequences are
not borne out in experiment.
There are various ways to avoid the conclusion C4, well-
rehearsed by now. The first insists that quantum mechanical
propositions have a funny logic. In particular they do not
obey the distributive law. This blocks the argument at the
move between C1 and ~. This solution was for a while
persuasively urged by Hiliary Putnarn.P Putnam has now
given it up because of various reflections on the meaning
and use of the connective or; but I am most impressed by
the objections of Michael Gardner!" and Peter Gibbins!"
that the usual quantum logics do not in fact give the right
results for the two-slit experiment.
The second well-known place to block the C derivation
is between the third and the fourth steps. The first attempt
maintained that quantum propositions have a peculiar logic;
this attempt insists that they have a peculiar probability
structure. It is a well-known fact about quantum mechanics
that the theory does not define joint probabilities for non-
commuting (i.e. incompatible) quantities. Position and
momentum are the conventional examples. But the different
positions of a single system at two different times are also
incompatible quantities, and so their joint probabilities are
also not defined. (The incompatibility of 'r at t' and 'r at t"
is often proved on the way to deriving the theorem of Ehren-
fest that, in the mean, quantum mechanical processes obey
the classical equations of motion.) So Prob(y & S1) and
Prob(y & S2) from C4 do not exist. If Prob(y) is to be cal-
culated it must be calculated in some different way-
notably, as in the Q derivation.
What does it mean in a derivation like this to deny a
joint probability? After all, by the time of C3 we have gone so
far as to admit that each electron, individually, has a well-
defined position at both the screen and at the plate. What
goes wrong when we try to assign a probability to this con-
junctive event? Operationally, the failure of joint distributions
must show up like this. We may imagine charting finite
histograms for the joint values of 'r at t' and 'r at t", but
the histograms forever bounce about and never converge to
a single curve. There is by hypothesis some joint frequency
12 H. Putnam, 'Is Logic Empirical?' in R. Cohen and M. Wartofsky (eds),

Boston Studies in the Philosophy of Science, 5 (Dordrecht: D.. Reidel, 1968).
13 Michael R. Gardner, 'Is Quantum Logic Really Logic?' Philosophy of
Science 38 (1971), pp. 508-29.
14 P. Gibbins, forthcoming in Foundations of Physics.
in every finite collection, but these frequencies do not

approach any limit as the collections grow large. This may
not seem surprising in a completely chaotic universe, but it
is very surprising here where the marginal probabilities are
perfectly well defined: in the very same collections, look
at the frequencies for 'r at t' or 'r' at t' separately. These
frequencies can always be obtained by summing over the
joint frequencies: e.g., in any collection the frequency
(y) = frequency (y & Sl) + frequency (y & S2)' As the col-
lections grow larger, the sum approaches a limit, but neither
of the terms in the sum do. This is peculiar probabilistic
behaviour for a physical system. (Though since, as I re-
marked in the last essay, I am suspicious even of the corres-
ponding probabilities in classical statistical mechanics, my
own view is that it is considerably less peculiar than quantum
logic.)
Before proceeding, it is important to notice that both
attempts to begin with C1 and to block the inference to
C4 depend on incompatibility. It is the incompatibility of
location-when-passing-the-screen and location-at-the-plate that
stops the inference from C1 to C2. In conventional quantum
logics, the distributive law holds if all the propositions are
compatible. Similarly, the theory will always give a joint
distribution for any two quantities that commute.
We began with the question, what are quantum prob-
abilities probabilities of? Practitioners of quantum theory
have been reluctant to adopt either non-standard logics
or non-standard probabilities. They have rejected the first
proposal altogether. Quantum probabilities are not prob-
abilities that the system is 'located at r', but rather, as
Merzbacher says, that it 'will be found at r in a position
measurement'. This answer is no more trouble-free than
the first. It supposes that the electron passes through neither
one slit nor the other when we are not looking. When we do
look, there, suddenly, it is, either at the top slit or at the
bottom. What is special about looking that causes objects
to be places where they would not have been otherwise?
This is just another version of the notorious measurement
problem, which we first discussed in the last section.
I find neither of these two conventional answers very
satisfactory, and I propose a more radical alternative. I want
to eliminate position probabilities altogether, and along
with them the probabilities for all the classic dynamic quanti-
ties. The only real probabilities in quantum mechanics, I
maintain, are transition probabilities. In some circumstances
a quantum system will make a transition from one state to
another. Indeterministically and irreversibly, without the
intervention of any external observer, a system can change
its state: the quantum number of the new state will be differ-
ent and a quantum of some conserved quantity-energy,
or momentum, or angular momentum, or possibly even
strangeness-will be emitted or absorbed. When such a
situation occurs, the probabilities for these transitions can
be computed; it is these probabilities that serve to interpret
quantum mechanics.
I shall illustrate with a couple of examples: the first,
exponential decay; and the second, the scattering of a moving
particle by a stationary target. Whether in an atom when
an outer shell electron changes orbit and a photon is given
off, or in a nucleus resulting in 0', {3, or 'Y radiation, decay
is the classic indeterministic process. Consider a collection
of non-interacting atoms, each in an excited state. The
number of atoms that continue to be in the excited state
decreases exponentially in time. One by one the atoms decay,
and when an atom decays, something concrete happens-a
photon is given off in a specific direction, the energy of the
electromagnetic field is increased, and its polarization is
affected. Which atom will decay, or when, is completely
indeterministic. But the probability to decay is fixed, and
this transition probability is the probability on which I
want to focus.
The second example is from scattering theory-a theory
of critical importance in high energy physics, where funda-
mental particles are studied almost entirely through their
behaviour in collisions. Consider a beam which traverses a
very long tube of very small diameter and then collides
with a target of massive particles and is scattered. If we
ring the target with detectors, we will find that many of the
incoming particles miss the target and continue in the for-
ward direction. But a number of others will be scattered
through a variety of angles. If the scattering is elastic, the

incoming particles will change only their direction; in in-
elastic scattering the moving particles will exchange both
momentum direction and energy with the target. The transi-
tion probabilities in this case are the probabilities for a
particle whose momentum was initially in the forward
direction to be travelling after scattering in another specific
direction, and with a well-defined energy.
The transition probabilities that occur in scattering have a
venerable history: they are the first to have been introduced
into quantum mechanics. Max Born's preliminary paper of
1926 is generally agreed to provide the original proposal for
interpreting the wave function probabilistically. The abstract
of that paper reads: 'By means of an investigation of col-
lision processes the viewpoint is developed that quantum
mechanics in Schrodinger's form can be used to describe
not only stationary states but also quantum jumps. 'IS Born
treats a collision between a moving electron and an atom.
In the middle of the paper he says
If one now wishes to understand this result in corpuscular terms,
then only one interpretation is possible: <l>n,.m(a<,13,'Y) defines the prob-
ability! that the electron coming from the z-direction will be projected
in the direction defined by (a, ~, 'Y) (and with phase change D), where
its energy I has increased by a quantum hV~m at the expense of the
atomic energy.l"
The footnote is famous:
1. Remark added in proof corrections: More precise analysis shows
that the probability is proportional to the square of <l>nTm(a<,f3,'Y)' 17
So probability entered quantum mecahnics not as a position
probability or a momentum probability, but rather as the
transition probability 'that the electron coming from the
z-direction will be projected in the direction defined by
(a. (3, 'Y)'.
I am urging that we give up position probabilities,
15 M. Born, 'Zur Quantenmechanik der Stossvorgange', Zeitschrift fur Physik
37 (1926), p. 863. Quotations taken from an English translation prepared by
Martin Curd for a seminar in the Department of History and Philosophy of
Science at the University of Pittsburgh, Spring 1978.
16 Ibid., pp. 865-6.
17 Ibid., p. 865, italics added.

momentum probabilities, and the like, and concentrate

instead on transition probabilities. The advantage of transi-
tion probabilities is that they have a classical structure, and
the event space over which they range has an ordinary
Boolean logic. To understand why, we need for a moment
to look at the formal treatment of transitions. Transitions
occur in situations where the total Hamiltonian H is naturally
decomposed into two parts-Ho, the 'free' Hamiltonian in
the situation, and V, a disturbing potential: H = Ho + V.
The system is assumed to begin in an eigenstate of Ho.
Supposing, as is the case in examples of transitions, that
Ho does not commute with V and hence does not commute
with H, the eigenvalue of Ho will not be conserved. Formally,
the initial eigenstate of Ho will evolve into a new state
which will be a superposition of Ho eigenstates: I/J(t) =
~Cj(t)<A where Ho = ~ai I<A><<A I. The transition probabilities
are given by the k:-Jt) 12• In the old language, they are the
probabilities for a system initially in a given <Po to 'be' or
'be found' later, at t, in some one or another of the eigen-
states <A. The probabilities and the event space they range
over are classical just because we look at no incompatible
observables. It is as if we were dealing with the possessed
values for Ho. All observables we can generate from Ho will
be mutually compatible, and the logic of compatible pro-
positions is classical. So too is the probability.
The primary consideration that has made philosophers
favour quantum logic is the drive towards realism (though
Alan Stairs!" lays out other, more enduring grounds). They
want to find some way to ensure that a quantum system will
possess values for all the classic dynamic quantities-posi-
tion, momentum, energy, and the like. But this motivation
is ill founded. If we want to know what properties are real
in a theory, we need to look at what properties playa causal
role in the stories the theory tells about the world. This is
a point I have been urging throughout this book, and it
makes a crucial difference here. Judged by this criterion,
the classical dynamic quantities fare badly. The static values
18 A. Stairs, 'Quantum Logic and Interpretation', talk give at the Annual
Meeting of the Society for Exact Philosophy, Tempe, Arizona, 26 February 1982
(unpublished).
of dynamic variables have no effects; it is only when systems

exchange energy or momentum, or some other conserved
quantity, that anything happens in quantum mechanics.
For example, knowing the position of a bound electron tells
us nothing about is future behaviour, and its being located
wherever it is brings about nothing. But if the electron
jumps orbits-that is, if the atom makes an energy transi-
tion-this event has effects that show up in spectroscopic
lines, in the dissolution of chemical bonds, in the formation
of ions, and so on. This is true even for measurements of
position. The detector does not respond to the mere presence
of the system but is activated only when an exchange of
energy occurs between the two. We need to consider an
example that shows how transitions play a causal role in
quantum mechanics which the conventional dynamic quanti-
ties do not. This seems a good one to use.
Henry Margenau '? has long urged that all quantum mea-
surements are ultimately measures of position. But position
measurements themselves are basically records of scattering
interactions. Position measurements occur when the particle
whose position is to be measured scatters from a detector.
The scattering is inelastic: energy is not conserved in the
particle and the detector is activated by the energy that the
particle gives up as it scatters. In the common position-
measuring devices-cloud chambers, scintillation counters,
and photographic plates-the relevant scattering interaction
is the same. The particle scatters from a target in the detector,
and the energy that is given up by the particle causes the
detector to ionize. The devices differ in how they collect
or register the ions; but in each case the presence of the
particle is registered only if the appropriate ionization inter-
action occurs. So, at best, the probability of a count in the
detecting device is equal to the probability of the designated
ionization interaction.
Matters may be much worse. General background radiation
may produce ionization when no particle is present. Con-
versely, the ion-collecting procedure may be inefficient and
19 H. Margenau, 'Critical Points in Modem Physical Theory', Philosophy of
Science 4 (1937), pp. 352-6; 'Philosophical Problems Concerning the Meaning of
Measurement in Physics', Philosophy of Science 25 (1958), pp. 23 ff.
ions that are produced by the scattered particle may fail

to register. Photographic emulsions are highly efficient
in this sense; efficiencies in some cases are greater than
98 per cent. But other devices are not so good. In principle
it is possible to correct for these effects in calculating prob-
abilities. For simplicity I shall consider only devices that
are ideally efficient-that is, I shall assume that all and
only ions that are produced by the scattered particle are
collected and counted.
I have urged that a real detector cannot respond to the
mere presence of a particle, but will react only if the particle
transfers energy to it. If the amplification process is maxim-
ally efficient so that the counter registers just in case the
appropriate energy is transferred, then a particle will register
in the detector if and only if the appropriate energy inter-
change occurs. This could create a serious consistency prob-
lem for the theory: a particle will be found at r in a position
measurement just in case a detector located at r undergoes
some specific energy interaction with the particle. As we
have been discussing, the probability of the first event is
supposed to be I W(r)!2d3r. But the probability of the second
is calculated in a quite different way, by the use of the
methods of scattering theory. Quantum mechanics is con-
sistent only if the two probabilities are equal, or approxi-
mately so. Otherwise Iw(r) 12 will not give the probability
that the system will be found at r in a real physical mea-
surement.
In fact this is stronger than necessary, for we are not
interested in the absolute values of the probabilities, but
merely in their relative values. Consider, for example, the
photographic plate, which is the device best suited for
determining densities of particles at fairly well-defined
positions. In a photographic plate we are not concerned that
the number of spots on the plate should record the actual
number of particles, but rather that the pattern on the plate
should reflect the distribution of particles. This requires
establishing not an equality, but merely a proportionality:
Probability of ionization in a target at r Iw(r) 12
(9.1)
Probability of ionization in a target at r' IW(r')12·
Plate
-----~~~---__+---------z
FIG 9.3. Origin located at source of wave
F or a general justification, (9.1) should be true for any

state function 1/1 for which the problem is well defined. For
simplicity I will treat a two-dimensional example, in which
a row of detectors is arrayed in a line perpendicular to the
z axis. We may think of the detectors as the active elements
in a photographic plate. In this case 1/I(r, t) should be arbi-
trary in shape, except that at t = 0 it will be located entirely
to the left of the plate (see Figure 9.3). It is easiest to estab-
lish (9.1) Hot in its immediate form, but rather by inverting
to get
Probability of ionization in a target at r
2
11/I(r) 1 (9.2)
= Probability of ionization in a target at r'
2
11/I(r') 1
Thus, the aim is to show that the ratio in (9.2) is a constant

independent of r. A shift of coordinates will help. We think
of fixing the centre of the wave and varying the location of
Plate
----------==--''-------l------_z
FIG 9.4. Origin located at detector
the detector across the plate (Figure 9.3). But the same
effect is achieved by fixing the location of the detector
and varying the centre of the wave (Figure 9.4). Looked
at in this way, the consistency result can be seen as a trivial
consequence of a fundamental theorem of scattering theory.
This theorem states that the scattering cross section, both
total and differential, is a constant independent of the
shape or location of the incoming wave packet. The total
scattering cross section is essentially the ratio of the total
probability for the particle to be scattered, divided by the
probability for crossing the detector. Roughly, the theorem
assures that the probability of scattering from the detector
for a wave centred at fO, divided by the probability for
'being at' the detector, is a constant independent of fo.
This is just what is required by (9.2).
There is one difficulty. Standard textbook proofs of
this theorem do not establish the result for arbitrary waves,
but only for wave packets that have a narrow spread in
the direction of the momentum. This is not enough to
ensure consistency. In a previous paper I have calculated

(9.2) directly and shown it to hold for arbitrary initial
states.P?
I have been urging that the interpretation of quantum
mechanics should be seen entirely in terms of transition
probabilities. Where no transitions occur, 1/J must remain
uninterpreted, or have only a subjunctive interpretation:
if the system were subject to a perturbing potential of the
right sort, the probability for a transition into state-would
be --. This makes the theory considerably less picturesque.
We think in terms of where our microsystems are located,
and all our instincts lead us to treat these locations as real.
It is a difficult matter to give up these instincts; but we
know from cases like the two-slit experiment that we must,
if our talk is to be coherent.
We should consider one more example in detail, to see
exactly how much of our reasoning in quantum mechanics
relies on picturing the positions of quantum systems realistic-
ally, and what further steps we must take if we are going
to reject this picture. Dipole radiation is one of the clearest
examples where position seems to count. Recall from Essay 7
that the atoms in a laser behave very much like classical
electron oscillators. I will use the treatment in Sargent,
Scully, and Lamb to make it easy for the reader to follow,
but the basic approach is quite old. Dipole radiation is one
of the very first situations to which Schroedinger applied
his new wave mechanics. Sargent, Scully and Lamb tell us:
It is supposed that quantum electrons in atoms effectively behave

like charges subject to the forced, damped oscillation of an applied
electromagnetic field. It is not [at] all obvious that bound electrons
behave in this way. Nevertheless, the average charge distribution does
oscillate, as can be understood by the following simple argument.
We know the probability density for the electron at any time, for this
is given by the electron wave function as I/J*(r, t)l/J(r, t). Hence, the
effective charge density is
el/J*(r, t)l/J(r, t).
20 N. Cartwright, 'Measuring Position Probabilities', in P. Suppes (ed.), Studies

in the Foundations of Quantum Mechanics (East Lansing: Philosophy of Science
Association, 1980), pp. 109-18.
For example, consider a hydrogen atom initially in its ground state
with [a] spherical distribution ... Here the average electron charge
is concentrated at the center of the sphere. Application of an electric
field forces this distribution to shift with respect to the positively
charged nucleus . . . Subsequent removal of the field then causes
the charged sphere to oscillate back and forth across the nucleus
because of Coulomb attraction. This oscillating dipole acts something
like a charge on a spring."
z z z
x y x y x y
t =0 t='/.ll/I1E t=%1i/tJ.E
z z
x y x y
t=%1i/I1E t=f1/I1E
FIG 9.5. Oscillating charge distribution in a hydrogen atom (Source:

Siegman, Lasers).
Consider as a concrete example a hydrogen atom in its

ground state (the Is state, which I shall designate by Ua(r)).
If the atom is subject to an external electric field, it will
evolve into a superposition of excited and de-excited states,
21 M. Sargent, M. Scully, and W. Lamb, Laser Physics (Reading: Addison-
Wesley, 1974), p. 31.
as we have seen before. (Here I will take the excited state

to be the 2p state, with m = 0, designated by Ub (r).) So at
t the state of the hydrogen atom in an electric field is
1/I(r, t) = Ca(t)exp(- iwat)Ua(r) + Cb(t)exp(- iWbt)Ub(r).
If we chart the charge density e 11/I(r,t) 12 at intervals At =

!(h!AE) apart, we see from Figure 9.5 that the charge distri-
bution moves in space like a linear dipole, and hence has a
dipole moment.F The dipole moment for the atom is
given by
(er) = pCaCE'exp{ - i(wa - Wb)t} +

+ pCa*Cbexp{ + i(wa - Wb)t}.
The brackets, (er), on the left indicate that we are taking

the quantum mechanical expectation. As Sargent, Scully,
and Lamb say, 'The expectation value of the electric dipole
moment (er) is given by the average value of er with respect
to this probability density [the density 1/I*(r)1/I(r)]'.23 The
location of the electron is thus given a highly realistic treat-
ment here: the dipole moment at a time t is characterized
in terms of the average position of the electron at t.
Classically an oscillating dipole will radiate energy. The
quantum analogue of this dipole radiation is central to the
Lamb theory of the laser. In a laser an external field applied
to the cavity induces a dipole moment in' the atoms in the
lasing medium. The calculation is more complicated, but the
idea is the same as in the case of the hydrogen atom that
we have just looked at. Summing each of the microscopic
dipole moments gives the macroscopic moment, or polariza-
tion, of the medium. This in turn acts as a source in Maxwell's
equations. A condition of self-consistency then requires that
the assumed field must equal the reaction field. Setting
the two equal, we can derive a full set of equations to de-
scribe oscillation in a laser.
I have two comments about the realistic use of electron
22. Figure 9.5 is taken from an unpublished draft of a forthcoming second

edition of A. E. Siegman, An Introduction to Lasers and Masers, where it appears
as figure 2-16 on p. 2-49.
23 Op. cit. footnote 20, p. 3.
position in the Lamb theory. The first is fairly instrumenta-
list in tone. The principal equations we derive in this kind
of treatment are rate equations either for the photons or for
the atoms. These are equations for the time rate of change
of the number of photons or for the occupation numbers
of the various atomic levels: that is, they are equations for
transition probabilities. As with scattering or simple decay,
these probabilities are completely classical. The equations
are just the ones that should result if the atoms genuinely
make transitions from one state to another. Louisell, for
instance, in his discussion of Lamb's treatment derives an
equation for the occupation number Na, of the excited state:
L.8.3.21: dNa/dt = Ra(t) - faNa(t) +

+ i/h[Pn(t)E*(t) - P;(t)E(t)].
He tells us:
The physical meaning of these equations should be quite clear. Equa-

tion (8.3.21) gives the net rate of change at which atoms are entering
and leaving state la). The Ra term gives the rate at which atoms are
being 'pumped' into level la). The - raNa term represents the inco-
herent decay of atoms from level la) to lower levels. We could also
add a term + WabNb to represent incoherent transitions from Ib) to
la), but we omit this for simplicity. The r;l is the lifetime of the atom
in level la) in the absence of [a] driving field. These first terms are
incoherent since they contain no phase information ...
The last term i(PE* - P*E) represents the net induced population
change in level Ia) due to the presence of a driving field.24
What .is important to notice is that the change in the number

of excited atoms is just equal to the number of atoms going
into the excited state minus the number going out. There
are no terms reflecting interference between the excited
and the de-excited states.
Recall that in Essay 6 we showed that the exponential
decay law can be drived by a Markov approximation, which
assumes that there is no correlation among the reservoir
variables. Exactly the same device is used to eliminate the
24 w. H. Louisell, Quantum Statistical Properties of Radiation (New York:
John Wiley & Sons, 1973), p. 459.
non-classical interference terms here. For example, in de-

riving the photon rate equation, H. Haken remarks:
We shall neglect all terms with b~bA (A =1= A'). This can be justified
as follows ... The phases of the b*'s fluctuate. As long as no phase
locking occurs the phase fluctuations of the different b's are uncor-
related. The mixed terms on the right hand side of 134 [i.e, the inter-
ference terms in the photon rate equation] vanish if an average over
the phases is taken."
The classical character of the rate equations leaves us in
a peculiar position. On the one hand, the superposition of
excited and de-excited states is essential to the theoretical
account. There is no dipole moment without the super-
position-the charge density e 11/I(r) 12 does not oscillate
in space in either the excited state alone or in the de-excited
state alone. Without the dipole moments, there is no macro-
scopic polarization produced by the medium, and hence no
grounds for the self-consistency equations with which Lamb's
derivations begin. On the other hand, the classical character
of the rate equations suggests that the atoms genuinely
change their states. If so, this is just like the picture I have
sketched for simple decay cases: the atom makes a transition,
Schroedinger evolution stops and reduction of the wave
packet takes over. The very formalism that allows us to
predict that this will occur does not apply throughout the
entire process.
This leads me to suppose that the whole account is an
explanatory fiction, including the oscillating dipoles, whose
role is merely to motivate us to write down the correct
starting equation. The account is best seen as a simulacrum
explanation. The less radical response is to notice that
Lamb and others describe e 11/I(r) 12as 'the charge distribution'.
In this very simple case, we need not see it as a probability
distribution at all, but as a genuine charge density in space.
This of course will not do as a general interpretation of
1/1*1/1. (Generally 1/1 must be taken as a function in a phase
space. Where more than one electron is involved, 1/1 will be
a function of the position of all of the electrons; and in
2S H. Haken, 'The Semi-classical and Quantum Theory of the Laser', in S. M.
Kay and A. Maitland (eds), Quantum Optics (London: Academic Press, 1970),
p.226.
general 1/1* 1/1 in the phase space will not reduce to a simple
distribution in real space for the charges involved.) But for
any case where we want to claim that the dipole moment
arises from a genuine oscillation, this must be possible.
Otherwise the correct way to think about the process is
not in terms of probabilities at all, but rather this way:
the field produced by the atoms contributes to the macro-
scopic polarization. (er) is the quantity in the atoms respons-
ible for the micro-polarization it produces. We learn to
calculate this quantity by taking e 11/1(r) 12 , and this calcula-
tion is formally analogous to computing a moment or an
average. But the analogy is purely formal. e 11/1(r) 12 is a
quantity which gives rise to polarization in the field; it has
no probabilistic interpretion. This way of viewing 'expecta-
tions' is not special to this case. Whenever an expectation
is given a real physical role, it must be stripped of its prob-
abilistic interpretation. Otherwise, how could it do its job?
(er) certainly is not the average of actual possessed positions,
since electrons in general do not possess positions. The
conventional alternative would have it be the average of
positions which would be found on measurement. But that
average can hardly contribute to the macroscopic polariza-
tion in the unmeasured cavity.
In this section I have argued that two independent reasons
suggest interpreting quantum mechanics by transition prob-
abilities and not by position probabilities or probabilities for
other values of classic dynamic quantities. First, transition
events playa causal role in the theory which is not matched
by the actual positions, momenta, etc. of quantum systems.
The second is that transition probabilities are classical, and
so too are their event spaces, so there is no need for either
a special logic or a special probability. Both of these argu-
ments suppose that transitions are events which actually
occur: sometime, indeterministically, the system does indeed
change its state. This is the view that Max Born defended
throughout his life. We find a modern statement of it in the
text of David Bohm:
We conclude thatlGnl2 yields the probability that a transition has
taken place from the sth to the rnth eigenstate of Ho, since the time
t = to. Even though the Gn's change continuously at a rate determined
by Schrodinger's equation and by the boundary conditions at t = to,

the system actually undergoes a discontinuous and indivisible transi-
tion from one state to the other. The existence of this transition
could be demonstrated, for example, if the perturbing potential were
turned off a short time after t = to, while the Gn's were still very
small. If this experiment were done many times in succession, it would
be found that the system was always left in some eigenstate of Ho.
In the overwhelming majority of cases, the system would be left in
its original state, but in a number of cases, proportional to IGn 12,
the system would be left in the mth state. Thus, the perturbing poten-
tial must be thought of as causing indivisible transitions to other
eigenstates of Ho.26
But this view is not uncontroversial. To see the alternative

views, consider radioactive decay. There are two ways to
look at it, one suggested by the old quantum theory, the
other by the new quantum theory. The story told by the
old quantum theory is just the one with which most of
us are familiar and which I adopt, following Bohm. First,
radioactive decay is indeterministic; second, the activity
decreases exponentially in time; and third, it produces a
chemical change in the radioactive elements. Henri Becquerel
reported the first observations of radioactivity from uranium
in a series of three papers in 1886. Marie Curie did a syste-
matic study of uranium and thorium beginning in 1898,
for which she and Pierre Curie shared a Nobel prize with
Becquerel. But it was not until the work of Rutherford
and Soddy in 1902 that these three important facts about
radioactivity were recognized. The first and second facts
come together. The probability for the material to remain in
its excited state decreases exponentially in time, and no
external influence can affect this probability, either to
increase it or to decrease it. Rutherford and Soddy report:
It will be shown later that the radioactivity of the emanation pro-

duced by thorium compounds decays geometrically with the time
under all conditions, and is not affected by the most drastic chemical
and physical treatment. The same has been shown by one of us to
hold for the excited radioactivity produced by the thorium emanation.
26 D. Bohm, Quantum Theory (Englewood Cliffs: Prenctice-Hall, 1951).

This decays at the same rate whether on the wire on which it is origin-
ally deposited, or in the solution of hydrochloric or nitric acid. The
excited radioactivity produced by the radium emanation appears
analogous."
The third fact is the one of relevance to us. From Ruther-
ford and Soddy's introduction: 'Radioactivity is shown to be
accompanied by chemical changes in which new types of
matter are being continuously produced.F" During decay,
uranium 238 is transformed into thorium 234. When the
alpha particle is emitted, the state of the material changes.
This is exactly analogous to Einstein's treatment of atomic
decay in his derivation of the Plank law of black body
radiation (though Einstein's real feeling about this situation
seems to have been far more ambiguous). Concerning what
we now call spontaneous emission he says, 'This is a transi-
tion from the state Zm into the state Z; together with
emission of the radiation energy Em - En. This transition
takes place without any external influence. One can scarcely
avoid thinking of it as like a kind of radioactive reaction. '29
Bohr too has the same picture when he first quantizes the
energy levels of the atom. Only certain orbits are allowed
to the electrons. In radiation the electron changes from one
fixed orbit to another, emitting a quantum of light energy.
In discussing the spectrum of hydrogen Bohr says, 'During
the emission of the radiation the system may be regarded
as passing from one state to another.P? On the old-quantum-
theory picture the time of decay is undetermined. But when
decay occurs, a photon, or an alpha particle or a beta particle
is given off, and the emitting material changes its state.
Contrast this with the new-Quantum-theory story. This is
the story we read from the formalism of the developed
mathematical theory. On this story nothing happens. In
27 E. Rutherford and F. Soddy, 'The Cause and Nature of Radioactivity',

Philosophical Magazine 4 (1902), p. 387.
28 Ibid., p. 371.
29 A. Einstein, 'Strahlungs-Emission und -Absorbtion nach der Quanten-
theorie', Deutsche Physicalische Gesellschaft; Verhandlungen 18 (1916), p. 321.
Reference and translation provided by Brent Mundy, Stanford University.
30 N. Bohr, 'On the Spectrum of Hydrogen', address before Physical Society
Copenhagen, 1913, p. 11, quoted in H. A. Boorse and 1. Motz (eds), The World
of the Atom (New York: Basic Books, 1966), p. 747.
atomic decay the atom begins in its excited state and the
field has no photons in it. Over time the composite atom-
plus-field evolves continuously under the Schroedinger
equation into a superposition. In one component of the
superposition the atom is still in the excited state and there
are no photons present; in the other, the atom is de-excited
and the field contains one photon of the appropriate fre-
quency. The atom is neither in its outer orbit nor in its
inner orbit, and the photon is neither there in the field
travelling away from the atom with the speed of light, nor
absent. Over time the probability to 'be found' in the state
with an excited atom and no photons decays exponentially.
In the limit as t -+ 00, the probability goes to zero. But
only as t -+ co] On the new-quantum-theory story, never,
at any finite time, does an atom emit radiation. Contrary
to Bohr's picture, the system may never be regarded 'as
passing from one state to another'.
The situation with scattering is no better. A particle with
a fixed direction and a fixed energy bombards a target and
is scattered. The state of the scattered particle is represented
by an outgoing spherical wave (see Figure 9.6 in the Appen-
dix). After scattering the particle travels in no fixed direction;
its outgoing state is a superposition of momentum states in
all directions. We may circle the target with a ring of de-
tectors. But as we saw in discussing the problem of prepara-
tion, this is of no help. If we look at the detectors, we will
find the particle recorded at one and only one of them. We
are then cast into a superposition; each component self sees
the count at a different detector. It is no wonder that von
Neumann says that here at least reduction of the wave
packet occurs. But his reduction comes too late. Even without
the detectors, the particle must be travelling one way or
another far away from the target.
So on my view, as in the old quantum theory, reduction
of the wave packet occurs in a variety of situations, and
independent of measurement. Since I have said that super-
positions and mixtures make different statistical predictions,
this claim ought to be subject to test. But direct statistical
test will not be easy. For example, to distinguish the two in
the case of atomic decay, we would have to do a correlation
experiment on both the atom and its associated photon,

and we would have to measure some observable which
did not commute with either the energy levels of the atom
or the modes of the perturbed field. (This is laid out formally
in 'Superposition and Macroscopic Observation'.) But these
kinds of measurements are generally inaccessible to us.
That is what the work of Daneri, Loinger, and Prosperi
exploits. Still, with ingenuity, we might be able to expose
the effects of interference in some more subtle way.
P. H. Eberhard has proposed some experiments which
attempt to do SO.31 Vandana Shiva proposes a test as well. 32
Not all tests for interferences will be relevant, of course, for
reductions occur, but not all of the time. Otherwise there
would be no interference pattern on the screen in the two-
slit experiment, and the bonding energy of benzene would
be different. But there is one experiment that Eberhard
proposes which I would count as crucial. This is a test to
look for reduction of the wave packet in one of the very
cases we have been considering-in scattering. I will discuss
Eberhard's experiment in the Appendix.
3. HOW THE MEASUREMENT PROBLEM IS

AN ARTEFACT OF THE MATHEMATICS
Von Neumann claimed that reduction of the wave packet
occurs when a measurement is made. But it also occurs
when a quantum system is prepared in an eigenstate, when
one particle scatters from another, when a radioactive nucleus
disintegrates, and in a large number of other transition pro-
cesses as well. That is the lesson of the last two sections.
Reductions of the wave packet go on all of the time, in a
wide variety of circumstances. There is nothing peculiar
about measurement, and there is no special role for con-
sciousness in quantum mechanics.
This is a step forward. The measurement problem has
31 P. H. Eberhard, 'Should Unitarity Be Tested Experimentally?'. CERN

Report No. CERN 72-1 (1972, unpublished).
32 V. Shiva, 'Are Quantum Mechanical Transition Probabilities Classical?
A Critique of Cartwright's Interpretation of Quantum Theory', Synthese 44
(1980), pp. 501-8.
disappeared. But it seems that another has replaced it.

Two kinds of evolution are still postulated: Schroedinger
evolution and reduction of the wave packet. The latter
is not restricted to measurement type situations, but when
does it occur? What features determine when a system
will evolve in accord with the Schroedinger equation and
when its wave packet will be reduced? I call this problem
the characterization problem. I am going to argue that
it is no real problem: it arises because we mistakenly take
the mathematical formulation of the theory too seriously.
But first we should look at a more conventional kind of
answer.
There is one solution to the characterization problem that
is suggested by the formal Schroedinger attempts to mini-
mize interference terms: reduction of the wave packet
occurs when and only when the system in question inter-
acts with another which has a very large number of inde-
pendent degrees of freedom. Recall the derivations of
exponential decay discussed in Essay 6. In the Weisskopf-
Wigner treatment we assume that the atom couples to a
'quasi-continuum' of modes of the electromagnetic field.
If instead it coupled to only one, or a few, the probability
would not decay in time, but would oscillate back and
forth forever between the excited and de-excited states.
This is called Rabi-flopping. I mentioned before the dis-
cussion by P. C. W. Davies.P Davies's derivation shows
clearly how increasing the number of degrees of freedom
eliminates the interference terms and transforms the Rabi
oscillation into an exponential decay. Similarly, the Daneri-
Loinger-Prosperi proof that I described in Section 1 of
this chapter relies on the large number of degrees of freedom
of the measuring apparatus. This is the ground for their
assumption (corresponding to assumption A2 in 'Super-
position and Macroscopic Observation') that there will
be no correlation in time between systems originating in
different microstates of the same macro-observable. This is
exactly analogous to Haken's assumption, quoted in the last
section, that b~ and bA are uncorrelated over time, and it
plays an analogous role. For that is just the assumption
33 Davies, op. cit., in footnote 7.
that allows Haken to eliminate the interference terms for

the photons and to derive the classical rate equations.
The most general proofs of this sort I know are in de-
rivations of the quantum statistical master equation, an
equation analogous to the evolution equations of classical
statistical mechanics. The Markov treatment of radioactive
decay from Essay 6 is a special case of this kind of deriva-
tion. In deriving the master equation the quantum system
is coupled to a reservoir. In theory the two should evolve
into a superposition in the composite space; but the Markov
approximation removes the interference terms and decouples
the systems. Again the Markov approximation, which treats
the observation time as infinitely long, is justified by the
large number of independent degrees of freedom in the
reservoir, which gives rise to short correlation times there.
There are two difficulties with these sorts of proofs as
a way of solving the characterization problem. The first is
practical. It is a difficulty shared with the approach I shall
defend in the end. Reduction of the wave packet, I have
argued, takes place in a wide variety of circumstances. But
treatments of the sort I have been describing have been
developed for only a small number of cases, and, with respect
to measurement in particular, treatments like that of Daneri,
Loinger, and Prosperi are very abstract and diagrammatic.
They do not treat any real measurement processes in detail.
The second difficulty is one in principle. Even if treat-
ments like these can be extended to cover more cases, they
do not in fact solve the characterization problem. That
problem arises because we postulate two different kinds of
evolution in nature and we look for a physical characteristic
that determines when one occurs rather than the other. Un-
fortunately the characteristic we discover from these proofs
holds only in models. It is not a characteristic of real situa-
tions. To eliminate the interference the number of relevant
degrees of freedom must be infinite, or correlatively, the
correlation time zero. In reality the number of degrees of
freedom is always finite, and the correlation times are always
positive.
Conceivably one could take the opposite tack, and urge that
all real systems have infinitely many degrees of freedom.
That leaves no ground to distinguish the two kinds of evolu-

tion either. The fact that this view about real systems is
intrinsically neither more nor less plausible than the first
suggests that the concept relevant number of degrees of
freedom is one that applies only in models and not in reality.
If we are to apply it to reality I think we had best admit
that real systems always have a finite number of degrees
of freedom.
A real system may be very large-large enough to model
it as having infinitely many degrees of freedom, or zero
time correlations-but this does not solve the characteriza-
tion problem. That problem requires one to separate two
very different kinds of change, and size does not neatly
divide the world in pieces. This is a familiar objection: if
bigness matters, how big is ·big enough? Exactly when is a
system big enough for nature to think it is infinite?
Sheer size cannot solve the characterization problem as
I have laid it out. But I now think that what I have laid
out is a psuedo-problem, The characterization problem is
an artefact of the mathematics. There is no real problem
because there are not two different kinds of evolution in
quantum mechanics. There are evolutions that are correctly
described by the Schroedinger equation, and there are
evolutions that are correctly described by something like
von Neumann's projection postulate. But these are not
different kinds in any physically relevant sense. We think
they are because of the way we write the theory.
I have corne to see this by looking at theoretical frame-
works recently developed in quantum statistical mechanics.
(E. B. Davies's Quantum Theory of Open Systemst" prob-
ably represents the best abstract formalization.) The point
is simple and does not depend on details of the statistical
theory. Von Neumann said there were two kinds of evolu-
tions. He wrote down two very different looking equations.
The framework he provides is not a convenient one for
studying dissipative systems, like lasers or radiating atoms.
As we have seen, the Schroedinger equation is not well able
to handle these; neither is the simple projection postulate
34 E. B. Davies, Quantum Theory of Open Systems (New York: Academic
Press, 1976).
given by von Neumann. Quantum statistical mechanics has
developed a more abstract formalism, better suited to these
kinds of problems. This formalism writes down only one
evolution equation; the Schroedinger equation and the
projection postulate are both special cases of this single
equation.
The evolution prescribed in the quantum statistical formal-
ism is much like Schroedinger evolution, but with one
central difference: the evolution operators for quantum
statistical processes form a dynamical semi-group rather
than a dynamical group. The essential difference between
a group and a semi-group is that the semi-group lacks inverses
and can thereby give rise to motions which are irreversible.
For instance, the radiating atom will decay irreversibly
instead of oscillating forever back and forth as it would
in Rabi-flopping. Correlatively, the quantum statistical
equation for evolution looks in abstract like a Schroedinger
equation, except that the operation that governs it need
not be represented by a unitary operator. A unitary operator
is one whose adjoint equals its inverse. The effect of this is
to preserve the lengths of vectors and the angles between
them. There are other mathematical features associated
with unitarity as well, but in the end they all serve to block
reduction of the wave packet. So it is not surprising that
the more general quantum statistical framework does not
require a unitary operator for evolution.
From this new point of view there are not two kinds
of evolution described by two kinds of equation. The new
formalism writes down only one equation that governs every
system. Is this too speedy a solution to the characterization
problem? Perhaps there is only one equation, but are there
not in reality two kinds of evolution? We may agree that
Schroedinger's equation and reduction of the wave packet
are special cases of the quantum statistical law: a Schroedinger-
like equation appears when a unitary operator is employed;
a reduction of the wave packet when there is a non-unitary
operator. Does this not immediately show how to reconstruct
the characterization problem? Some situations are described
by a unitary operator and others by a non-unitary operator.
What physical difference corresponds to this?
There is a simple, immediate reply, and I think it is the

right reply, although we cannot accept it without considering
some questions about determinism. The reply is that no
physical difference need explain why a unitary operator is
used in one situation and why a non-unitary one is required
in another. Unitarity is a useful, perhaps fundamental,
characteristic of operators. That does not mean that it
represents a physical characteristic to be found in the real
world. We are misled if we take all mathematics too seriously.
Not every significant mathematical distinction marks a
physical distinction in the things represented. Unitarity is a
case in poin t.
'How', we may still ask, 'does nature know in a given
situation whether to evolve the system in a unitary or a
non-unitary way?' That is the wrong question. The right,
but trite, question is only, 'How does nature know how to
evolve the system?' Well, nature will evolve the system as
the quantum statistical equation dictates. It will look at
the forces and configurations and at the energies these give
rise to, and will do what the equation requires when those
energies obtain. Imagine that nature uses our methods of
representation. It looks at the energies, writes down the
operator that represents those energies, solves the quantum
statistical equation, and finally produces the new state that
the equation demands. Sometimes a unitary operator will
be written down, sometimes not.
I reject the question, 'How does nature know in a given
situation whether to use a unitary or non-unitary operator?'
That question presupposes that nature first looks at the
energies to see if the operator will be unitary or not, and
then looks to see which particular operator of the designated
kind must be written down. But there is no need for the
middle step. The rules that assign operators to energies need
not first choose the kind of operator before choosing the
operator. They just choose the operator.
I write as if unitarity has no physical significance at all.
It has only mathematical interest. Is this so? No, because
unitarity is just the characteristic that precludes reduction
of the wave packet; and, as we have seen, reduction of the
wave packet is indeterministic, whereas Schroedinger evolution
is deterministic. This, according to the quantum theory, is

a genuine physical difference. (Deterministic motions are
continuous; indeterministic are discontinuous.) I do not
want to deny that there is this physical difference; but
rather to deny that there must be some general fact about
the energies that accounts for it. There need be no general
characteristic true of situations in which the evolution is
deterministic, and false when the evolution is indeterministic.
The energies that obtain in a given situation determine how
the system will evolve. Once the energies are fixed, that
fixes whether the evolution is deterministic or indetermin-
istic. No further physical facts are needed.
If this interpretation is adopted, determinism becomes
for quantum mechanics a genuine but fortuitous physical
characteristic. I call it fortuitous by analogy with Aristotle's
meeting in the market place. In the Physics, Book II,
Chapter 5, Aristotle imagines a chance encounter between
two men. Each man goes to the market place for motives
of his own. They meet there by accident. The meeting is
fortuitous because the scheme of motives and capacities
explains the presence of each man separately, but it does
not explain their meeting. This does not mean that the
meeting was not a genuine physical occurrence; nor that
anything happened in the market place that could not be
predicted from the explanatory factors at hand-in this
case the motions and capacities of the individuals. It means
only that meetings as meetings have no characteristic cause
in the scheme of explanation at hand. This does not show
that there is some mistake in the scheme, or that it is in-
complete. We may be very interested in meetings, or for that
matter, determinism, or the money cycle; but this does not
guarantee that these features will have characteristic causes.
A theory is not inadequate just because it does not find
causes for all the things to which we attend. It can be faulted
only if it fails to describe causes that nature supplies.
Professor Florence Leibowitz suggests another way to
understand the place of determinism in the interpretation I
propose. She says, 'Your claim about unitarity is in effect an
assertion that indeterministic evolutions ought to be seen
as "primitive" for quantum mechanics, in the sense that
behaving according to the law of inertia is a primitive for

post-scholastic mechanics.t" The conventional von Neumann-
based view of evolution sees deterministic evolution as the
natural situation, Leibowitz suggests. Indeterministic evolu-
tion is seen as a departure from what would naturally occur,
and hence requires a cause-an interaction with a reservoir,
perhaps, or with a conscious observer. But on my under-
standing of the quantum statistical formalism, indeterministic
motions are natural too. They are not perturbations, and
hence do not require causes. This is exactly analogous to
what happens to inertial motion in the shift from Scholastic
to Newtonian mechanics. For the Scholastic, continued
motion in a straight line was a perturbation and a cause
was required for it. In the Newtonian scheme the continued
motion is natural or, as Leibowitz says, 'primitive'. Likewise
quantum mechanics does not need a physical property to
which unitarity corresponds: even if there were such a
property, Leibowitz points out, 'it would not have any
explanatory work to do'.
Throughout these essays I have urged that causality is
a clue to what properties are real. Not all predicates that
play a significant role in a theory pick out properties that
are real for that theory. Many, for example, represent only
properties in models, characteristics that allow the derivation
of sound phenomenological laws, but which themselves
play no role in the causal processes described by those laws.
Unitarity is a different kind of example.
To understand what role unitarity plays in the theory,
first look at another property that evolution operators may
have: invariance under some group of transformations. In
Schroedinger theory, the Hamiltonian describes the energies
in the situation and thereby determines the evolution opera-
tor. Whenever the Hamiltonian is invariant under a group of
transformations, there will be some constant of the motion
which exhibits degeneracies; that is, different, incompatible
states will have the same value for that quantity. Rotational
invariance is a simple example. Rotational invariance in a
Hamiltonian corresponds to spherical symmetry in the
energies and forces represented by that Hamiltonian. The
35 F. Leibowitz, private correspondence, 29 December 1981.
spherical symmetry produces degeneracies in the energy

levels, and disturbing the symmetry eliminates the degen-
eracies. If a small non-symmetric field is added, a number
of closely spaced lines will appear in the spectroscope where
before there was only one. The rotational invariance of the
Hamiltonian is a sign of a genuine physical characteristic
of the situation.
Unitarity is different. We are often interested in whether
the motion in a given kind of situation will be deterministic
or indeterministic. There is a long way to find this out:
consider the energies in the situation; write down the oper-
ator that represents them; solve the quantum statistical
equation; and look to see if the consequent change is con-
tinuous. This is cumbersome. We would like some way to
read from the operator itself whether the solutions will be
continuous or not. Unitarity is a mark we can use, and it
is one of the strengths of the particular mathematical struc-
ture we have that it gives us such a mark. Unitarity and
rotational invariance are both significant features of the
evolution operator, features that we single out for attention.
But they play very different roles. Rotational invariance
marks a genuine characteristic of the energy, a characteristic
that is postulated as the source of a variety of physical
consequences; unitarity provides an important mathematical
convenience. To demand a physical correlate of unitarity is to
misunderstand what functions it serves in the quantum theory.
I do not want to insist that unitarity does not represent
a genuine property, but rather that the failure to find such
a property is not a conceptual problem for the theory.
Larry Laudan's description of a conceptual problem fits
here. Laudan says, 'Such problems arise for a theory, T, ...
when T makes assumptions about the world that run counter
to ... prevailing metaphysical assumptions'r'" Under von
Neumann's proposal that quantum systems evolve under
two distinct laws, some feature was required to signal which
law should operate where. No physical characteristic could be
found to serve, and the theory seemed driven to metaphysically
36 L Laudan, 'A Problem Solving Approach to Scientific Progress', in I.
Hacking (ed.), Scientific Revolutions (Oxford: Oxford University Press, 1981)
p.146.
SUSpiCIOUS characteristics-fictional properties like infinite

degrees of freedom or zero correlation times, or, even worse,
interaction with conscious observers. But if the quantum
statistical formalism can be made to work, no such property
is required and the theory will not run counter to the 'pre-
vailing metaphysical assumptions' that neither sheer size
nor consciousness should matter to physics.
The i[here presents an important condition. The comments
of Jeremy Butterfield about my proposal seem to me right.
Butterfield says,
Quantum statistical mechanics has provided a general theory of evolu-
tion of quantum states (pure and mixed) that encompasses [von
Neumann's] two sorts of evolution, and many others, as special cases.
Nor is this just a mathematical umbrella. It allows us to set up detailed
models of phenomena that cannot be treated easily, if at all with the
traditional formalism's two sorts of evolutton.P?
Quantum optics is one place these detailed models have been
developed, especially in the study of lasers. But we have seen
that a broad range of cases make trouble for the traditional
Schroedinger account-scattering, for example, or any
situation in which pure states are prepared, and finally the
issue that started us off-measurement. Much later Butter-
field continues:
I do not want to pour cold water on this programme [of Cartwright's];
I find it very attractive. But I want to stress that it is a programme,
not a fait accompli. To succeed with it, we need to provide detailed
analyses of measurement situations, showing that the right mixtures
are forthcoming. We need not of course cover all measurement situa-
tions, but we need to make it plausible that the right mixtures are
generally obtained. (And here 'generally' need not mean 'universally';
it is the pervasiveness, not necessarily universality, of definite values
that needs to be explained.) Only when we have such detailed analyses
will the measurement problem be laid to rest. 38
Butterfield gives good directions for what to do next.
I recommended the book of E. B. Davies as a good source
for finding out more about the quantum statistical approach.
37 J, Butterfield, 'Reply to N. Cartwright's 'How the Measurement Problem
is an Artefact of the Mathematics', in R. Swinburne (ed.), Space, Time, and
Causality (Dordrecht: D. Reidel, forthcoming).
38 Ibid.
I should mention that Davies himself does not use the formal-
ism in the way that I urge; for he is at pains to embed the
non-unitary evolutions he studies into a Schroedinger evolu-
tion on a larger system. This goes along with his suggestion
that non-unitarity is a mark of an open system, one which
is in interaction with another. Open systems are presumably
parts of larger closed systems, and these in Davies's account
always undergo unitary change. I think this view is mistaken,
for the reasons I have been urging throughout this essay.
If the wave packet is not reduced on the larger system, it
is not in fact reduced on the smaller either. The behaviour
of the smaller system at best looks as if a reduction has
occurred, and that is not good enough to account for
measurements or for preparations.
I have been urging that, if the quantum statistical pro-
gramme can work, the measurement problem becomes a
psuedo-problem. But other, related, problems remain. These
are the problems of how to pick the right operators, unitary
or no, to represent a given physical situation. This is the
piece by piece work of everyday physics, and it is good
to have our philosophical attentions focused on it again.
This is what on-going physics is about, and it knows no
general procedure. In quantum mecahnics the correspond-
ence principle tells us to work by analogy with classical
mechanics, but the helpfulness of this suggestion soon runs
out. We carry on by using our physical intuitions, analogies
we see with other cases, specializations of more general
considerations, and so forth. Sometimes we even choose
the models we do because the functions we write down are
ones we can solve. As Merzbacher remarks about the Schroe-
dinger equation:
Quantum dynamics contains no general prescription for the con-
struction of the operator H whose existence it asserts. The Hamil-
tonian operator must be found on the basis of experience, using the
clues provided by the classical description, if one is available. Physical
insight is required to make a judicious choice of the operators to be
used in the description of the system ... and to construct the Hamil-
tonian in terms of these variables. 39
39 Merzbacher, op. cit. footnote 11, p. 336-7.

As I argued in Essays 7 and 8, not bridge principles, but

physical insights, are required to choose the right operators.
But at least the quantum statistical programme offers hope
that this mundane, though difficult, job of physics is all
that there is to the measurement problem.
APPENDIX; AN EXPERIMENT TO TEST

REDUCTION OF THE WAVE PACKET
In 1972 P. H. Eberhard considered non-unitary theories
of the kind I endorse here and proposed four types of tests
for them. I shall discuss in detail his tests involving the
optical theorem of scattering theory, since this is the case
I understand best, and it is a case that fits nicely with the
discussion earlier in this essay. Eberhard tells us, about the
theory he discusses, 'Our non-unitary theory resembles the
description of quantic systems in interaction with a measure-
ment apparatus, but no apparatus is involved in the physical
processes that our theory is applied to'."? Eberhard calls
theories that respect unitarity class-A theories. He will be
concerned with a particular class of non-unitary theories-
class-B theories. These are theories which model change in
quantum systems on what happens in a complete measure-
ment. Specifically, for an observable M = ~m 1<I>m )«(/>m I, a
B-type interaction takes the state D into D':
B-type theories are thus exactly the kind of theory I have

urged, in which transitions genuinely occur into eigenstates
{(/>m I. and the final state for the ensemble is a classical
mixture in which the (/>m do not interfere.
Eberhard tests type-B theories with the optical theorem.
He tells us,
The optical theorem is derived from the principle that the wave
scattered in the forward direction interferes with the incident wave in
such a way as to conserve probabilities. If the forward scattering
contained a mixture, i.e., non-interfering components, the test of the
40 Eberhard, op. cit. footnote 31, p. 1.
optical theorem would fail. That test involves measurements of the

differential cross section in the forward direction, including the inter-
ference region between Coulomb and strong interaction scatterings.
The results can then be compared to the measurements of the total
cross section."
Eberhard looks at elastic scattering in a 11'- p interaction
at 1.015, 1.527 and 2.004 GeV. The results agree with the
prediction of the optical theorem 'within ±3%, when aver-
aged over the three momenta'i'" This agreement is good
enough. If Eberhard's analysis is correct, class-B theories
are ruled out for scattering interactions, and if they do not
hold for scattering, they are not very plausible anywhere.
FIG 9.6. Scattering from a stationary target
The optical theorem is an obvious place to look for a test

of non-unitary evolution. Consider a typical textbook pre-
sentation of the formal theory of scattering. I use Eugen
Merzbacher's Quantum Mechanics. Merzbacher tells us
'The scattering matrix owes its central importance to the
fact that it is unitary't" and later 'From the unitary property
of the scattering matrix, we can derive an important theorem
41 P. H. Eberhard, 'Tests of Unitarity', in A. Zichichi (ed.), Progress in

Scientific Culture. The Interdisciplinary Journal of the Ettore Maiorana Centre.
Winter 1976 (Trapani, Italy: Tipografia 'Cartograf', 1977).
42 P. Eberhard; R. D. Tripp; Y. Deelais; J. Seguinot; P. Baillon; C. Bricman;
M. Ferro-Luzzi; J. M. Perrau; and T. Ypsilantis, 'A Test of the Optical Theorem',
Physics Letters 53B, no. 1 (1974), p. 12l.
43 Merzbaeher, op, cit. footnote 11, p. 501, italies in original.
for the scattering amplitudes'v'-c-the optical theorem.

Nevertheless the optical theorem does not rule out class-B
theories. The optical theorem, I will argue, holds good in
just the kind of class-B theory I have described for scattering.
In the case of elastic scattering, where the bombarding
particle neither loses nor gains energy, the asymptotic state
for large r for an outgoing particle whose initial momentum
is k, is given by
I
l/J~(r)"'"21T3/2 (exp(ik ° r) + l/r exp(ikr)!k(r)). (1)
This state is a superposition of the original momentum eigen-

state expt ik-r) and an outgoing spherical wave l/rexp(ikr)
as in Figure 9.6. The quantity !k(r) is called the scattering
amplitude. It is the imaginary part of the scattering ampli-
tude in the forward direction-Imfk(O)-which enters the
optical theorem. As we can see from Figure 9.6, in the
forward direction the original unscattered wave and the
outgoing spherical wave interfere. The interference subtracts
from the probability of the incoming wave in the forward
direction. This is as we would expect, since the beam in the
forward direction will be depleted by the particles that
strike the target and are scattered.
It is easiest to calculate the interference if we switch to
the formal theory of scattering. Equation (1) is the wave
function version, for large r, of the Lippman-Schwinger
equation:
1
,f,+ =
Ylk
.1,
Ylk
+ '"
~ E _ flo + im .1,
YIn
T.
nk »
s
Here the momentum states {l/Jn} are eigenstates of the

unperturbed Hamiltonian, Ho, and it is understood that the
limit IX ~ 0 is to be taken at the end of the calculation.
The transition matrix Tnk is proportional to the scattering
amplitude. We are interested in what proportion of the
beam will be travelling in the forward direction after scat-
tering, so we must calculate the probability !< l/JkI l/Jt>12 •
Substituting
44 Ibid., p. 505.
21Th2 A I
Tkk, = -3 fdk)
J.l.L
and using the fact that

I
lim . = 1T<l)(W)
0<-0 W + la
and that
8(Ek - Ek:) = +st« -

hk
k'),
we get
4 3
12 81T I 12
,to
I <1/Jkll/)k) = I + k3L6 A(O) + 41T
kL3 ImA(O). (2)
Now we may repeat the kind of classical argument that

we considered for the two-slit experiment. A particle travel-
ling in the forward direction after passing the target was
either scattered from the target, or passed the target un-
scattered:
K = K & (7S v S).
Since Sand 7S are disjoint events

Prob K = Prob(K/7S) Prob(7S) + Prob(K/S) Prob(S). (3)
But equation (2) shows that this classical reasoning will
not do. The first two terms in equations (2) and (3) are
identical, but, as in the two-slit experiment, the quantum
mechanical calculation differs from the classical one by
the interference terms, which are responsible in the two-
slit case for the troughs in the diffraction pattern, and in
scattering for the shadow cast by the target. We see that
the amount of interference depends on the imaginary part
of the forward scattering amplitude.
The optical theorem relates the total cross section, a, to
1m A(O):
41T
Optical Theorem: a = -lmA(O).
k
The cross section, a, measures the total probability for
scattering into any angle. Recall that Eberhard reported,

'The optical theorem is derived from the principle that the
wave scattered in the forward direction interferes with the
incident wave in such a way as to conserve probabilities'.
We can now see why this is so. The optical theorem says
that the loss in the beam in the forward direction, which
we have seen comes from the interference term, Im !k(0),
is equal to the total number of particles scattered. Inter-
ference is thus an essential part of the optical theorem.
How then can I maintain that reduction of the wave packet
after scattering is consistent with the optical theorem?
The key to the answer is that one must be careful about
what final states the system is supposed to enter when
reduction occurs. I suggested, following Bohm, that after
scattering each particle will be travelling in a specific direc-
tion, and with a specific energy. The reduction takes place
into the eigenstates of momentum. The optical theorem
precludes only a reduction into the pair scattered-unscattered.
But the momentum probabilities already contain the inter-
ference between the incoming plane wave and the scattered
spherical wave.
We can see this by looking back to the Lippman-Schwinger
equation. It follows from that equation (taking the limit as
ex -+ 0) that the amplitude for a system with initial momen-
tum k at t = - 00 to have momentum k' at t = + 00, is given by
Here I have identified this amplitude with the k, k'th element

of the scattering matrix, S, as we learn we can do from the
formal theory of scattering. The total amplitude is thus a
superposition of the amplitude from the scattered wave plus
the amplitude from the unscattered wave, so the interference
between the two is present right in the momentum ampli-
tudes. When the complex conjugate is taken, equation (2)
will result as required. It is no surprise then that the optical
theorem still holds for the kind of reduction I propose.
Formally, I imagine that after reduction the state of
particles in the beam is given by D':
D' = L I~k')( 'h' IU(- 00, + oo)l~k)(~k I

k'
x jU(- 00, + oo)l~k')( %, I
=L St'k S~k l~k')(~k,l,

k'
where Utt, t') is the normal unitary evolution operator

supplied by Schroedinger theory. (It is conventional to
take the limits as t ~ too since the times involved before
detection are very long on the microscopic scale.) Since
D ~ D' is a measurement-type interaction on the momen-
tum eigenstates, the momentum probabilities after reduc-
tion will be the same as before reduction. But the optical
theorem is a trivial consequence of the conservation of
total probability among the momentum states. Here is where
the unitarity of the scattering matrix, which Merzbacher
stresses, plays its role. Because S is unitary, the probabilities
to be in one or another of the momentum eigenstates sum
to one
not only in the unreduced state but in the reduced state as

well. This is enough to guarantee that the optical theorem
holds. The proof is simple, and I shall omit it here. (It is set
as exercise 19.5 in Merzbacher: 'Derive the optical theorem
from the conservation of probability and (19.12)', 4S where
equation (19.12) gives the amplitude for the k'th momentum
state at t.) Thus, the optical theorem is consistent with a
class-B interaction in which the wave packet is reduced into
momentum eigenstates after the particle has been scattered.
What then of Eberhard's claims? To see how to reconcile
what Eberhard says with the facts I have just noted, we
need to look more closely at the kind of class-B "theory
which Eberhard considers. Eberhard notes that a non-unitary
evolution like D ~ D' can always be written as a sum of
unitary evolutions. This gives rise to something like a hidden
45 Ibid., exercise 19.5, p. 507.
variable theory: where we see a single physical process

which appears to follow a non-unitary rule, there are in
fact a mixture of different processes each manifesting a
unitary Schroedinger evolution. Or, alternatively: take the
final pure states produced by Eberhard's set of 'component'
unitary evolutions and wind them backwards by using the
inverse of the original unitary scattering matrix. Then, the
Eberhard style hidden variable theory says that, contrary
to the normal assumption, the incoming state is not pure,
but instead a mixture of these wound-backward states.
Each state behaves exactly as the Schroedinger equation
predicts. We end with a mixture, but only because we begin
with one.
Even though he does not explicitly say so, Eberhard's
calculations take this hidden variable theory very seriously.
Eberhard's test uses two theorems from scattering theory.
The first relates the differential cross-section in the forward
direction to the scattering amplitude in that direction:
da(O) = 1!t(O)j2 dn
Using Rand J as Eberhard does to refer to the real and
imaginary parts of !teO), we get Eberhard's equation (4.2).46
E(4.2): da/dn = R2 + P.
The second theorem is the optical theorem, which Eberhard
writes as
E(4.3): J = ka/47Th
As Eberhard remarks, R can be calculated from J, or it can
be determined from the interference with Coulomb scattering.
Since a and da(O)/dn can be measured in independent
experiments, a test of the optical theorem is at hand.
Let us now look to see how Eberhard turns this into a test
of class-B evolutions using his earlier theorem that any
non-unitary evolution of B-type is equivalent to a weighted
average of unitary changes. Eberhard claims:
In a class B theory, there are pseudo-states j that correspond to weights
"}" and unitary matrices Sj. Each unitary matrix Sj corresponds to a
class A theory, therefore to a aj, to a daj/dD., to a Rj(E) and to a .Ij(E)
4. Eberhard, op, cit., note 30, p. 14.
satisfying eq. (4.1) to (4.5). The effective probability distributions
are the weighted averages of those class A predictions and the effective
cross sections a and da/dn are the weighted averages of the aj's and
of the dai/drl respectively'?
So, says Eberhard,
E(4.7): da(O)/dil = L wi dai/dil

i
and
E(4.8)a: J = L wi.!; = -k L "tai

.
=-
ka
i 41T i 41Th
E( 4.8)b: R ~ t(WIIi)
Using E(4.8), R 2 + j2 = (L WiRi)2 + (L Wi.!;)2 • But in general
E.!. (tW;R/+ tW/~140 I(tW1RJ '+ (t"lJ)'j

So, if E(4.7) is correct, E(4.2) will be violated.
It remains to substantiate that equations E(4.7) and E(4.8)
are true for class-B theoreis. E(4.7) is straightforward. Letting
W represent the total transition rate into a solid angle dil,
fromB
d
W = L L-
k'Ed!1 k" dt
(VJk,IVJk")(VJk"IUIVJk)(VJkIU1VJk")(VJk" IVJk')
L -d (VJk,IUIVJk)(VJk IUIVJk')
k'Ed!1 dt
= lim _1
~ /'->0 At
(L
k'Ed!1
(VJk' IU(t + At) IVJk)( VJk IU(t + At)IVJk')
- L (VJk-lU(t)IVJk)( VJk 1 U(t) IVJk'»)

k'Ed!1
4' Ibid., p, 14.
From Eberhard's earlier theorem that class-B evolutions are

equivalent to weighted averages of unitary evolution ~,
I
W= lim -x
.1.t-->O !:!.t
X [~( ~ w/"" I UN + M)l>h)(", IU,(t + "01",,)
- ~Wj ("" "" 11W

111(01.1,,)( )I",,»)1
I
= Iw;
i
lim -I
.1.,.....o!:!.t k'
(l/Jk,I~(t + !:!.t)ll/Jk)(l/Jkl~(t + !:!.t)Il/Jk')
~(t)Il/Jk)( l/JkI ~(t) Il/Jk')

- (l/Jk,1
I W; Wi'
But
where hk/ ul. 3 is the probability that a particle is incident on a

unit area perpendicular to the beam per unit time. Hence,
ilL 3
da = hk I w;W; = I w;da;.
So equation E( 4.7) holds.

But what of E( 4.8)? E( 4.8) is justified only if we insist
on an Eberhard-style hidden variable theory. Eberhard shows
that the non-unitary evolution B can always be mathematic-
ally decomposed as an average over unitary evolutions. The
hidden variable version of the class-B theory supposes that
this mathematical decomposition corresponds to ,a physical
reality: scattering is really a mix of physical processes,
each process governed by one of the unitary operators,
Ui, which make up the mathematical decomposition. In this
case, we have a variety of different scattering processes,
each with its own scattering amplitude, R(k'), and its own
cross-sections, ai' and E(4.8) is a reasonable constraint on

the real and imaginary parts of the scattering amplitudes.
But is this physically robust hidden variable version of
a class-B theory a reasonable one from our point of view?
No, it is not. For it does not solve the problem of preparation
that motivated our class-B theory to begin with. Scattering
interactions prepare beams in momentum eigenstates: par-
ticles which appear in a particular solid angle dS1at one time
are expected-those very same particles-to be travelling
in exactly the same solid angle later, unless they are inter-
fered with. So we look for physical processes whose end-
states are eigenstates of momentum. But the end-states of
the processes U; are nothing like that.
From Eberhard's decomposition proof there are as many
'hidden' processes as there are dimensions to the space
needed to treat the system. In the case of scattering, a
'quasi-continuum' of states is required. Each 'hidden process'
turns out to be itself a scattering-type interaction. The end-
state of the first hidden process is just the normal outgoing
spherical wave of conventional scattering theory. This con-
ventional state will have the weight lin, when n is the dimen-
sion of the space. So, too, will each of the remaining states.
The end-state of the second process will be much like the
first, except that the amplitude of the second momentum
eigenstate will be rotated by 1800• Similarly, the third pro-
cess rotates the amplitude of the third momentum eigen-
state by 1800; the fourth rotates the fourth amplitude;
and so on. In average, the effect of these rotations is to
cancel the interference among the momentum states and to
produce a final mixture whose statistical predictions are
just like those from a mixture of momentum states. But
in truth, when the hidden variable account is taken seriously,
the final state is a mixture of almost spherical waves, each
itself a superposition of momentum eigenstates; and it is
not in fact a mixture of momentum eigenstates as we had
hoped. But if we do not take the hidden variable theory
seriously and give a physical interpretation to the decom-
position into unitary processes, there is no ground for equa-
tions E(4.8)a and E(4.8)b, and the optical theorem is no
test of B-type evolutions.
The Eberhard inequality E.I. is based on equations E( 4.8)a

and E( 4.8)b, which I claim are plausible for an Eberhard-
style hidden variable theory, but which do not hold for a
class-B theory which takes each incoming particle into a
momentum eigenstate. We should now confirm this last
claim. The process which produces a mixture of momentum
eigenstates is itself composed of a mixture of processes,
each of which produces one or another of the eigenstates
of momentum as its final product. (Note: each of these
processes is non-unitary, because it shrinks the vector.
Nor can we reconstruct it as a unitary evolution by taking
the shrinking factor, as Eberhard does, as a weight with which
a unitary change into a momentum eigenstate might occur,
because the 'weights' for the various processes would depend
not on the nature of the interaction but on the structure of
the incoming state.) We need to be sure that E.!. does not
follow equally on my account as on the hidden scattering
account. But this is easy. The 'weights' here are each 1.
Each interaction scatters entirely into a single direction,
and it is entirely responsible for all the scattering that occurs
into that direction. So R(k') = !k (k') Ojk' and hence
(L wI fJ(O») 2 = (II fJ (0)6,.) 2 ~ 11.(0)1'
LI
" (!k(O)Ojk) =L
2"· w;(fJ(O)) 2.
So E.!. does not plague the view that combines reductions

into different momentum eigenstates to get the mixture
D', though it does plague the composition of hidden scatter-
ing processes, as Eberhard shows. But this latter view is not
one I would incline to, since it is not one that allows for
the preparation of momentum eigenstates in scattering
in teractions.
A uthor Index
Agarwal, G. S. 79,123,124,125 Gibbard, A. 33

AirY,G. B. 2 Gibbins, P. 177
Alexander, P. 55 Goldberger, M. 117
Aristotle 9, 110,201 Goldman, M. 81
Armstrong,T.A.130 Grice, H. P. 33, 128, 129, 153
Griinbaum, A. 94,101,102,131
Boyle, R. 100
Bell,J.S.168 Hacking, I. 2, 20, 88, 98
Becquerel, H. 192 Haken,H. 79,80,190,196,197
Bethe, H. 122, 123 Hamilton, W. 135
Bickel, P. 37 Hammel, E. 37
Blalock, H. 23 Hanson, B. 45
Bohm, D. 191, 192 Harman, G. 75, 85
Bohr, N. 193, 194 Harper, W. 38
Born, M. 180,191 Hausman, D. 77
Bromberger, S. 75 Hawking, S. 12
Butt, D. K. 118 Hempel, C. G. 26,27,28,29,30,44,
Butterfield, J. 204 45,94,101,131,132,134,155
Hesse, M. 157
Carnap, R. 26 Hooke, R. 100
Cohen, M. 24 Hume, D. 13,61,74
Creary, L. 62, 63, 65, 66, 67
Crookes, W. 5, 81 Infeld, L. 21
Curie, M. 192
CUrie,P.I92 Jackson, F. 40
Cushing, J. 7, 158 Jaynes, E. T. 118
Jeffrey,R.27,34,103
Daner~A. 169, 170,295, 196, 197 Joseph, G. 12
Davies, E. B. 198, 204, 205
Davies, P. C. W. 169,196 Kant, I. 88
Descartes, R. 75 Klauder, J. 148,149,158
Donagan, A. 16 Klein, M. 46, 47
Duhem, P. 4, 19,76,77,87,88,89, Kline, S. 63
90,91,92,93,94,96,97,159 Kochen, S. 168
Kruskal, S. 168
Eberhard, P. H. 195, 206, 207, 210, Kuhn, T. 143
211,212,213,214,215,216
Einstein, A. 21, 193 Lamb, W. 116, 122, 123, 125, 186,
Ennis, R. 77 188, 189
Everitt, C. W. F. 2,5,6, 82, 148 Laudan,L.94,97,203
Leibowitz, F. 201, 202
Faraday, M. 2 Loinger, A. 169,170,195,196,197
Fermi, E. 151 Lorentz, H. 2
Feynman, R. 19,55,56,57,59,161, Louisell, W. 78, 115,146, 147, 148,
162, 166 150,151,189
Gardner, M. 177 Mackey, G. W. 135

218 AUTHOR INDEX
Mapleton, R. 121 Scully, M. 186, 188

Margenau, H. 182 Sellars, W. 157,159, 160
Maxwell, J. 2,5,6,8,9,11,12,154, Shiva, V. 195
155, 156 Siegman, A. 129, 147, 148, 149,
Merzbacher, E. 113, 136, 141, 142, 187,188
174,175,178,207,211 Simon, H. 23
Mes~ah,A.67,68, 136, 137, 138 Simpson, E. H. 7, 10
Mill, 1. S. 7, 58,60,61,70 Skyrrns, B. 43
Smart, J. J. C. 54
Nagel, E. 24, 101, 131 Smith, A. W. 130
Newton, 1. 75 Sneed,1. 159
Nordby, J. 13, 14, 15, 103, 107, 111 Soddy, F. 192, 193
Norman, S. 71 Specker, E. P. 168
Stairs, A. 181
O'Connell, J. W. 37 Sudarshan, E. C. G. 148, 149, 158
Onsager, 1. 65 Suppe, F. 131
Suppes,P. 23, 28,29,44,52, 159
Pais, A. 118
Pauling, 1. 166 Thorn, R. 75
Pearson, K. 39 Thomson, 1. J. 61
Perrin.T. 82,83,84 Thucydides 139
Perry, J. 54 Truesdell, C. A. 65
Piccolomini 90, 91
Prosperi, G. M. 169, 170, 195, 196, Van der Pol, B. 130
197 Van Dyke, M. 14
Ptolemy, 90 Van Fraassen, B. 4, 6, 8, 10, 56, 87,
Putnam, H. 56, 57, 77, 88,110,170,177 88, 89, 90, 91, 92, 93, 94, 96,
97,99,159
Redhead, M. 158 Von Neumann, J. 163, 167, 194, 195,
Retherford, R. C. 116, 122, 125 198,203
Russell, B. 13, 21,74,112
Rutherford, E. 192, 193 Watson, K. 117
Weisskopf, V. 113, 119, 122, 163
Salmon, W. 16, 23, 24, 25,26, 27,28, Wilson, A. R. 118
29,30,44,52,82,103 Winter, R. 118
Sargent, M. 186, 188 Wollheim, R. 91
Schroedinger, E. 186
Scriven, M. 16,75 Yule, G. 24
Subject Index
Airy's Law 2 cross-effects 64, 65, 112

amplifiers 103, 104, 107-13 see also composition of causes and
approximations 3, 13 ff., 104 ff., interactions, causal
107-19, 110, 112, 113, 117,
113-127,133,134 damping 77, 78-82, 113, 149
ab vero 14 decay, exponential 27, 93, 113-18,
ad verum 14,15 122, 164, 169, 170, 179, 189,
'as if operator 129, 130,131,133 190,192 ff., 196, 197, 199
Avogadro's hypothesis 84 ff. decision theory 33-40
deductive-nomological model of ex-
Boltzmann's equation 6, 9, 11, 154 planation 17,44,94 ff., 101 rr.,
bridge laws (bridge principles) 12, 15, 110,112,127,151,161,162
131 ff., 135-9, 143, 144-51, degrees of freedom 150, 196 ff., 204
155, 156 determinism (indeterminism) 21, 52,
164,167,179,191,192,200 ff.
causal behaviour (lasers) 147, 150 diffraction 3, 172, 175
causal explanation 4, 5, 8, 10 ff., 15, dipole radiation 186 ff.
27,29,36,75-8,89,93
causal account 6, 82,97,98, 162 effectiveness 33 ff., 42, 43
causal role 8, 182, 191 eigenvalue-eigenvector link 168
causal story 5, 9, 11, 12, 76, 77, Einstein-Podolsky-Rosen paradox 18
79,81,82,86, 161, 162 experiments 6, 7, 20, 82 ff., 98, 160,
causal models 23 161
causal powers 59 experimental test of exponential law
causal principles (causal laws) 10, 118
21-44,72 explanatory power 10, 20, 34, 56-9,
causal process 4, 16, 19, 61, 84, 85, 72, 139, 144, 152
128,152,160,162,202 exponential decay, see decay, exponen-
ceteris paribus laws (ceteris paribus tial
principles) 3, 45, 46-7, 57, 58,
64, 155 facticity 54, 58,61,62,65,71 ff., 152
characterization problem 169 rr., 199 Fick's law 63, 64
classical electron oscillator 129, 130, Fokker-Planck equation 130
148, 186 forces, component and resultant 59,
cloud chamber 92, 93, 99,182 60,66,69,70
coincidence, argument from 75, 82-5 fortuitous characteristic 201
composition (combination) of causes
3, 11-15, 51, 56-9, 62, 66, generic-specific account 94, 103, 104,
69-72,73 106,107,110,126
see also interactions, causal and gravitational force (gravitational law)
cross-effects 56,57,59 ff., 66, 69 rr., 74
Coulomb force (Coulomb law) 57,
59 ff., 66, 68 ff., 138, 159,212 Hamiltonian, model 135, 136, 139,
counterfactuals 34, 35, 38-40, 69 144, 159
covering law model of explanation hidden variables 168,212,214,215
11, 15 ff., 44, 45, 48 ff., 52,69, homogeneity, causal 25,26,28,40,41
155,162 Hume world 40-2
220 SUBJECT INDEX
idealization 48, 109 ff., 134, 136, 147, open systems 198,205
148,153,155,158 optical theorem 206 ff,
inductive-statistical model of explana- order of integration 120 ff.
tion 26, 30,44,47
inference to best explanation 4, 6, partial conditional probability 42
15,75,82,83,85,87-9 partitioning 36 ff., 41, 43
inference to most likely cause 6, 85, Pauli equation 114, 115, 124
92,94 phenomenological laws 1 ff., 8,11,16,
infinite potentials 150, 153 19, 85, 88 rr., 93 ff., 100 rr.,
interactions, causal 30-2, 63 106, 160, 161
see also composition of causes phenomenological models (phenomeno-
and cross-effects logical terms) 148,151
internal principles 131, 132, 135, 139 photographic plate 182 ff.
positions of quantum systems 163,
joint effects 43 166, 167, 171, 178, 182, 183,
see also spurious cause 186,188 ff.
joint probability distribution 156, 177, position probabilities 164, 174, 179,
178,181 180,188,190,191
preparation process (problem of prep-
Lamb shift 116, 117, 137, 138 aration) 174, 194, 205, 215,
in the excited state 119-23 216
in the ground state 119, 123-7 prepared descriptions (unprepared de-
lasers 3, 79, 80, 98, 128 ff., 145; 146, scriptions) 15, 17, 133, 139,
148, 149, 152, 153, 157 ff., 147,160,162
161,186,189,198,204 probability distributions, classical 153
laws of association 10, 21, 25, 26, 28, ff.
35,40 ff., 46 probabilistic model of causation 44, 47
line broadening 77 ff. projection postulate 162, 167, 170,
198
macroscopic observable (macroscopic property of convenience 153, 156
system) 167, 168, 170 ff., 195,
196 quantum electrodynamics (quantum
Markov approximation 113 ff., 123, field theory) 7, 8
124, 133, 146, 189, 197 quantum logic 164, 177, 181
master equation 79, 81, 113, 114, quantum statistical mechanics 164,
123,125,145,197 199 ff., 204, 205
measurement problem 18, 162-5
mechanical philosophy 100, 107 Rabi-flopping 196, 199
Mill's methods 7,98 radioactivity 25,27, 28, 78, 93, 118,
mixtures 168, 170, 171, 194, 204, 192,193
206,212,215 see also decay, exponential
models, explanatory 4, 11, 12, 15, radiometer 5, 6, 8, 9, 11, 81, 82,
17, 44, 83 ff., 95, 104, 107, 154 ff.
111, 129, 140, 143, 144, 148, rate equations 189, 190
152,154,156 ff., 197, 198,202 realistic in the first sense (in the
second sense) 150, 152
Newton's laws 63 reduction of the wave packet 163, 164,
167,170,174,194 ff., 206-16,
observable (unobservable) 1, 2, 5, 6, appendix
19,56,83,100,106,159,160 redundancy 76, 79, 90
Ohm's law 63 reference class problem 28, 29
old quantum theory (new quantum reservoir 113 rr., 129, 133, 146, 148
theory) 137, 192 ff. ff., 189, 197, 202
SUBJECT INDEX 221
rotating wave approximation 119, 123 theoretical entities 6, 7, 56, 31-4,

ff. 97,99
theory-observation distinction 132,
scattering 164, 179, 180, 182, 189, 133
194, 204, 206 ff. transition process (transition prob-
Schroedinger evolution 164, 196, 200, ability) 164, 174-95, 206
205,212 true probability 38-40
semantical view of theories 159 ff. two-slit experiment 175, 186, 195,
Simpson's paradox 10, 24, 36 ff. 209
simulacrum account of explanation
4,17,48,151-62,190 unitarity 18, 199 rr., 205 ff., appendix
SLAC 172, 173 unity of science (unity of nature) 12,
Snell's law 46 ff. 13,95,96
spurious cause 31, 33, 34
see also joint effects Van der Pol's equation 148, 153, 157
statistical relevance 30,40,44, 47 vector addition 59, 60, 62, 63, 69
strategies 10, 21, 22, 33 ff., 36-8,
43 Weisskopf-Wigner treatment 99, 113,
super laws 12, 70, 71 115,117,119,122,123,196
test situation 26 ff., 30, 35 zero-time correlations 150,153, 196

Cartwright, N., How The Laws of Physics Lie 1983 PDF

Uploaded by

Copyright:

Available Formats

Cartwright, N., How The Laws of Physics Lie 1983 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cartwright, N., How The Laws of Physics Lie 1983 PDF

Uploaded by

Copyright:

Available Formats

HOW THE LAWS

CLARENDON PRESS· OXFORD

Oxford New York

:rr and associated companies

·21 Published in the United States by

c Nancy Cartwright 1983

First published 1983

All rights reserved. No part of this publication may be reproduced,

j Library of Congress Cataloging in Publication Data

Printed in Great Britain

The author acknowledges permission to use previously pub-

Essay 1 Causal Laws and Effective Strategies 21

Essay 2 The Truth Doesn't Explain Much 44

Essay 3 Do the Laws of Physics State the Facts? 54

Essay 4 The Reality of Causes in a World of

Essay 5 When Explanation Leads to Inference 87

Essay 6 For Phenomenological Laws 100

Essay 7 Fitting Facts to Equations 128

Essay 8 The Simulacrum Account of Explanation 143

Essay 9 How the Measurement Problem is an

Author Index 217

Subject Index 219

Philosophers distinguish phenomenological from theoretical

I The Encyclopaedic Dictionary of Physics (Oxford: Pergamon Press, 1964),

Francis Everitt, a distinguished experimental physicist and

In modern physics, and I think in other exact sciences as

explanation, a model that ties laws directly to reality. As an

1. AGAINST INFERENCE TO BEST EXPLANATION

I will argue that the falsehood of fundamental laws is a

succeeds only if the process described actually occurs. To

causal account of why the vanes move around. In producing

pairs of every description'? Debate among philosophers has

dfl + ~1 dfl + 171dfl + ~1 dfl + X dfl + Y dfl + Z dfl +

+ fffd~2d172d~2 fbdb fdcf>VCfI!2 - f~f~) = 0,

These are general, abstract equations; they are not about

ties of Temperature', The Scientific Papers of James Clerk Maxwell ed. W. D.

Explanatory power is no guarantee of truth, unless van

1.1. Composition of causes

do not. Causal stories are treated as if they are true or false,

are on the verge of writing down the right equations to unify

example of the extent of the shortfall let us look at the

Van Dyke lists seven cases, then continues,

Realistically-oriented philosophers are inclined to think

at an accurate description of the phenomena. Two kinds of

1.2 An alternative to the covering-law model of explanation

give causal explanations of isolated events, I shall discuss

basic laws. The lesson for the truth of fundamental laws

2. WHAT DIFFERENCE DOES IT MAKE?

Debate between realists and non-realists has been going

GUIDE FOR THE READER

The last essay on quantum mechanics shows how anti-

22 Ian Hacking, Representing and Intervening (Cambridge: Cambridge Uni-

Causal Laws and Effective Strategies

There are at least two kinds of laws of nature: laws of associa-

I Albert Einstein and Leopold Infeld, The Evolution of Physics (Cambridge:

Society 13 (1912-13), pp. 1-26.

I recently received from TIAA-CREF, a company that

It simply wouldn't be true to say,

I will take as a starting point for my argument facts like