Analysis, 66 (2006) pp.52-63.
Are quantum particles objects?
Simon Saunders
It is widely believed that particles in quantum mechanics are metaphysically
strange; they are not individuals (the view of Cassirer 1956), in some sense
of the term, and perhaps they are not even objects at all, a suspicion raised
by Quine(1976a, 1990). In parallel it is thought that this di¤erence, and es-
pecially the status of quantum particles as indistinguishable, accounts for the
di¤erence between classical and quantum statistics - a view with long historical
credentials.1
?Indistinguishable?here mean permutable; that states of a¤airs di¤ering only
in permutations of particles are the same - which, satisfyingly, are described by
quantum entanglements, so clearly in a way that is conceptually new. And,
indeed, distinguishable particles in quantum mechanics, for which permutations
yield distinct states, do obey classical statistics, so there is something to this
connection.
But it cannot be the whole story if, as I will argue, at least in one notable
tradition, classical particle descriptions may also be permutable (so classical
particles may also be counted as indistinguishable); and if, in that same tradi-
tion, albeit with certain exceptions, quantum particles are bona ?de objects.
1
I will follow Quine in a number of respects, ?rst, with respect to the formal,
metaphysically thin notion of objecthood encapsulated in the use of singular
terms, identity, and quanti?cation theory; second (Quine 1970), in the appli-
cation of this apparatus in a ?rst-order language L, and preferably one with
only a ?nite non-logical alphabet; and third (Quine 1960), in the use of a weak
version of the Principle of Identity of Indiscernibles (PII). Applying the latter
requires a listing of the allowable predicates (the non-logical vocabulary of L);
for our present purposes this should be dictated on theoretical and experimental
grounds, grounds internal to the physics - for example, that only predications
of measurable properties and relations should be allowed. Our minimal, logical
question is then: whether indistinguishable quantum particles are L-discernible
by their measurable properties and relations.
But as a criterion for membership in L, measurability may be somewhat too
restrictive; it threatens to settle our question, negatively, solely on the basis that
quantum particles are unobservable. Better is a condition that is both precise
and more general, namely that only predicates invariant under the symmetries
of the theory qualify. This condition implicitly or explicitly underlines a good
many recent debates in the philosophy of physics over symmetry principles,
1It has been called the ?received view?by French and Rickles (2003).
1
and in important cases (for a number of space-time symmetries) it is physically
uncontroversial.2
In our case the symmetry is the permutation group. Our criterion, then,
is that L-predicates should be invariant under permutations (I shall then call
them symmetric or symmetrized). Whatever the metaphysical questions that
accompany the idea of a ?loss of identity?in quantum mechanics (for indistin-
guishable particles), its sole mathematical signature is permutation symmetry,
the syntactical expression of which (in terms of a regimented formal language)
is surely that predicates be symmetrized. So if it is true that in the words of
an early contributor to quantum statistics ?the conception of atoms as parti-
cles losing their identity cannot be introduced into the classical theory without
contradiction?(Stern 1949: 535) - and if the di￠ culty does not concern the
details of the classical theory but its basic concepts - one would expect it to
show up in our elementary framework of a ?nitary language restricted to to-
tally symmetrized predicates. Bach (1997) indeed takes it as self-evident that
a description of particles having de?nite coordinates can only be permutation
invariant in so far as it is incomplete (specifying only the statistical properties
of a particle ensemble, not the microscopic details).
But is there really any such conceptual impediment? If not - and from
what follows it seems not - the case for metaphysical novelty following on from
particle indistinguishability in quantum mechanics remains unmotivated. Add
to this the argument (Huggett 1999) that classical statistics is every bit as
consistent with permutation symmetry as is quantum statistics, and the claim
that indistinguishability explains quantum statistics looks threadbare indeed.
2
Let F be an nary predicate; the symmetrized language LS that we envisage
must be such that if F 2LS, then in any valuation Fx1:::xn can be replaced by
Fx(1):::x(n) without change of truth value, for any permutation of f1;:::;ng:
We imagine this as our procedure: we start from some language L, based
in part on other physical theories, which lacks permutation symmetry; and
we examine the e¤ects of implementing it, de?ning totally symmetrized LS-
predicates as complex predicates in L (e.g.
_
Fx(1):::x(n), call Fdis; of course
there are plenty of other constructions too - we shall spell out a general one
shortly). Clearly L S L.
How limited is LS? The answer depends on L. Of particular interest is the
case when L has no names, so that singular reference is by means of bound
quanti?cation only (paraphrasing contexts involving names by Russellian de?-
nite descriptions). In fact, restricted to descriptions of particle distributions in
space - as used in specifying the coordinates of particles (or initial or ?nal data
more generally) - it would seem that inde?nite descriptions are enough to be
going on with, of the form ?particles of such-and-such a kind have such and such
properties and relations?. It is not at all clear that in giving such descriptions
2For a review see (Saunders 2003b).
2
one must single out one arrangement of particles, and no permutation of it; for
one has the logical equivalence:
9x1:::9xnFx1:::xn 9x1:::9xnFdisx1:::xn: (1)
It is unlikely that LS and L can di¤er very much for uses like this: what more
does one want to say in L, after listing all the relations among the mentioned
objects, in the form of a purely existential statement? - other that is than which
object (value of variable), for each k, is ak (an illegitimate question, if L has no
names).
In fact when L is devoid of names use of LS involves no restriction at all,
at least in the case of sentences with only ?nite models of a ?xed cardinality
(sayN). Given any suchLsentenceT, one can construct a logically equivalent
LS-sentence TS, true in all and only the same models.
The claim is su￠ ciently surprising to warrant at least a sketch of the proof.
We may suppose, with no loss of generality, that T is given in prenex normal
form (all the quanti?ers, say n in number, to the left). Now construct a sen-
tence T1 in a language L+, which is L supplemented by N names a1;::;aN,
by replacing each rightmost quanti?er and the complex predicate that follows
in T by a disjunction (in the case of 9) or conjunction (in the case of 8) of
formulas in each of which xn is replaced by a name (yielding
N_
k=1
Fx1:::xn1ak
or
N^
k=1
Fx1:::xn1ak respectively). ). Repeat, removing each innermost quanti-
?er, obtaining at each step a complex predicate completely symmetric in the N
names (ensured only because T has no names); on the ?nal step one obtains a
sentence T2, in which the xks do not occur. Now replace every occurrence of ak
by xk to obtain a totally symmetrized Nary L-predicate, which is therefore in
LS . Prefacing withN existential quanti?ers (and conjoining with a cardinality-
?xing sentence) one obtains the sentence TS; by construction it has the same
truth conditions as T:
The two languages, insofar as they are used to describe a ?nite collection of
objects, are in this sense strictly equivalent; under this condition, symmetriza-
tion makes no di¤erence to truth values of sentences. I suggest this is evidence
enough that indistinguishability in itself indicates nothing metaphysically unto-
ward, or otherwise strange.
Why might one have thought any otherwise? But the constraint is certainly
prohibitive applied to ordinary language; take the predicate ?... is in the kitchen,
not....?, as in:
(i) Bob is in the kitchen, not Alice.
To symmetrize and say instead:
(ii) Bob is in the kitchen and not Alice, or Alice is in the kitchen and not Bob
3
doesn?t tell us what we want to know. But in a language su￠ ciently rich in
predicates to replace ?Bob?and ?Alice?by de?nite descriptions, the situation is
quite di¤erent. We then have a sentence like:
(iii) There is someone Bob-shaped who is in the kitchen, and someone Alice-
shaped who is not
(where ?Bob-shaped?etc. is shorthand for some purely geometric, anatomical
description). Symmetrizing, in LS we say instead:
(iv) There is x1 and x2, where x1 is Bob-shaped and in the kitchen and x2 is
Alice-shaped and not, or x2 is Bob-shaped and in the kitchen and x1 is
Alice-shaped and not.
Unlike the passage from (i) to (ii), there is no di¤erence between (iii) and (iv);
they are an instance of the equivalence (1)). One might of course reintroduce
the question of which of x1 and x2 is which (say, which of two persons, speci?ed
independent of their appearance), but that is only to invite further de?nite
descriptions, whereupon we will be back to the same equivalence.
3
This argument would all by itself settle the matter, were it not for the worry
that the objects that we end up with - the values of x1 and x2; that only
contingently have the bodies or personalities (or what have you) that they do -
are themselves rather strange. It may be they are just as problematic, when it
comes to questions of identity, as quantum particles.
We should face this challenge head on. The account of identity that follows
applies to any ?rst-order language L without equality, for any ?nite non-logical
alphabet, whether or not symmetrized.
Any such L e¤ectively comes with identity, a point that Quine has often
emphasized. We get for free the de?ned sign:
s = t =
def
^
all primitive L-predicates
Fs$Ft (2)
(here s and t are L-terms, occupying the same predicate position in F). Un-
packing this schematic de?nition, and temporarily introducing the notation Fnk
for the k-th nary predicate symbol of L, we obtain on universal generalization
over free variables not in s, t:
s = t =
def
^
n
^
k
n^
j=1
8x1:::8xn(Fnk x1::xj1sxj+1::xn $Fnk x1::xj1txj+1::xn):
(3)
The RHS is of the form ^
88:::8(Fs$Ft) (4)
4
and not:
^
(88:::8(Fs) $ 88:::8(Ft)) (5)
the point that so often goes unstated.3 By construction the schemes variously
written as (2),(3),(4) (but not (5)) imply the usual axiom scheme for identity:
s = s; s = t! (s$ t)
(where can be replaced by any L-predicate, primitive or otherwise); moreover
any other scheme with implies the latter yields an equality sign coextensive with
the one de?ned, so identity as given by (2), (3), (4) is essentially unique.
But isn?t this just the familiar PII? - yes, or it should be familiar. In fact
it has received surprisingly little attention, despite its endorsement by Quine.
A correct formal classi?cation of L-discernibles, according to this scheme, was
only given quite recently.4 And Quine made no applications of the principle to
physically problematic cases (nor, so far as I know, has anyone since).
Quine?s amended classi?cation is (I follow his earlier terminology for the ?rst
two cases):
Two objects are
absolutely discernible inLif there is anL-formula in one free variable that
applies to one of them only
relatively discernible inLif there is anL-formula in two free variables that
applies to them in one order only
weakly discernible in L if they satisfy an irre?exive L-formula in two free
variables.
As stated each category contains the one before it (here I follow Quine), but
they are exhaustive: values of variables not even weakly discernible are counted
(in L) as the same.
The interesting cases are mere relative or weak discernibility. For example,
let the only non-logical symbol in L be an irre?exive and symmetric dyadic F;
then from the de?nition (3):
x = y $ 8z((Fxz $Fyz):
On any valuation in whichFxy is true, 8z(Fxz $Fyz) is false (asFxy^:Fyy
is true); it follows that x 6= y. Thus, to take Black?s famous example of two
spheres of iron, positioned in an otherwise empty universe, one mile apart in
space; they are weakly discerned by the symmetric and irre?exive relation ?one
mile apart in space?; but they are neither absolutely nor relatively discernible.5
3It is discussed at length by Quine (1976b).
4By Quine in 1976, amending, without comment, the classi?cation he gave in Word and
Object in 1960.
5For other physical examples, see Saunders (2003a) and, in mathematics, Ladyman (2005).
5
What now of names? Let L contain names, and the categories of relative
and weak discernibility seem to be obliterated. For named objects, if discernible
at all, are absolutely discernible.6 However, everything turns on the proviso, if
discernible; whether or not the names do in fact name di¤erent objects will still
depend, just as before, on predicates alone. In this sense, then, named objects
can still be classi?ed as absolutely, relatively, or only weakly discernible, just as
before.
4
We are ?nally ready to answer our question: Are permutable particles dis-
cernible? The answer depends, evidently, on LS, and speci?cally its non-logical
vocabulary. In quantum mechanics the state of a particle is speci?ed by a vector
'', up to a complex multiplicative constant, or phase, in the state-space of the
system (Hilbert space) . AnNparticle state is a sequence of 1-particle vectors,
or - this the essential di¤erence from classical theory - a sum of such sequences.
A state of a collection of quantum particles, if the particles are indistinguishable,
must be invariant under the permutation group. Among these are expressions
of the form (for a 3particle state):
const:('' +'' + ''+ '' + '' + '') (6)
where ''; ; are 1particle vectors. Pretty evidently, it does not specify which
particle is in which state - there is no such determinate rule here. It is like the
symmetrized triadic ?the ?rst particle is in the state ?const. ''?the second in
the state ?const. ?, the third in the state ?const. ?, or the ?rst particle is in
the state ?const. ''?, the second in the state ?const. ?, the third in the state
?const. ?, or ....?(evidently sequence positions are here functioning as names).
Such states (permuting and summing) are also called symmetrized; the particles
described by them are bosons.
But what are we to make of the allegedly 3-particle (and manifestly sym-
metrized) state ?const. ''''''?? Evidently, that there are three particles each in
exactly the same 1-particle state, and therefore exactly alike in every respect.
They are surely not absolutely discernible, hence, since relative discernibles re-
quire non-symmetric predicates, they are at most weakly discernible. But are
they even that? What (physical, invariant) relation does a boson enter into with
a boson in exactly the same state, supposed to be a complete description, that
it does not enter into with itself?
If the answer is none (as it appears), or none that can be sanctioned by the
physical theory, then either the PII, or the objectual status of quantum particles,
is in question. Were that the end of the story, either way our total system
would be in trouble. In fact there is another possibility - another prescription
under which the state is invariant under permutations: vectors may instead be
antisymmetric, changing sign on any odd number of interchanges of particles
6If discernible at all, then they are at least weakly discernible; so there is a totally symmetric
and totally irre?exive Nary predicate F such that the sentence Fa1:::aN is true. But then
Fa1::ak1xak+1:::aN absolutely discerns ak.
6
(the state itself - the vector up to phase - remaining unchanged). The particles
of antisymmetrized states are fermions. In place of (6) we have
const:('' '' + '' '' + '' ''):
Evidently antisymmetric states cannot assign two di¤erent particles to exactly
the same 1-particle state; antisymmetrizing ?''''''?produces the zero. The prob-
lem we encountered with bosons does not arise.
Antisymmetrization ensures Pauli?s exclusion principle (the principle that
fermions cannot have all their quantum numbers in common). The latter was
indeed early on considered, by Weyl among others, to vindicate the PII,7 but
the suggestion was squashed by Margenau in 1944 (and seems to have been
hardly advocated since). Margenau came up with a new argument to show
that fermions are indiscernible, namely, that all the 1-particle expectation values
(which may be taken as exhausting the 1-place predications) of any fermion in
an antisymmetrized state must be the same. This was thought to show that the
PII cannot apply.8
But discernibility does not require absolute discernibility; and if one con-
siders the remaining candidates, relative or weak discernibility, it seems that
Margenau was wrong and Weyl was right all along. For even in a situation of
maximal symmetry, for example in the singlet state of spin
= const. (''"''# ''#''") (7)
the two particles are still weakly discernible. Here ''"''# correspond to the two
opposite possible values (parallel or antiparallel) of the spin of the particle along
a given direction ". Here any direction can be chosen, without change of the
state - it is in this sense that (7) is a specially symmetric state, invariant under
rotations as well as permutations - still the two particles satisfy the symmetric
but irre?exive predicate ?... has opposite " component of spin to:::?.9
Why was this simple observation missed? The answer, presumably, is that
it would then seem that the particles must each have a de?nite and opposite
value for the " component of spin, implying some kind of hidden-variable
interpretation of quantum mechanics (contentious in itself, for entirely unrelated
reasons). But this is to fall back on our old habit of turning discernment on
the basis of relations into discernment by di¤erences in properties (?relational
properties?); it is to miss the logical categories of relative and weak discernibility.
Consider again Black?s two iron spheres, each exactly alike, but one mile
apart in space. They are weakly discerned by the irre?exive relation ?...one
mile apart from...?, but - on pain of begging the question against relationism
- it does not follow, because the spheres bear spatial relations to each other,
that they each have a particular position in space. Neither, if two lines are
weakly discerned by the irre?exive relation ?at right-angles?, does it follow that
each line has a particular direction in space. Two particles can have opposite "
7It was called the ?Leibniz-Pauli?principle by Weyl (1949: 247).
8Similar arguments have since been given by French and Redhead (1988) and Dieks (1990).
9For more formal details, see Saunders (2003a, 2006).
7
component of spin (they are anticorrelated as regards spin in the" direction)
without each having a particular value for the "-component of spin.
On the strength of this we can see, I think, the truth of the general case:
so long as the state of an Nfermion collective is antisymmetrized, there will
be some totally irre?exive and symmetric N-ary predicate that they satisfy.
Fermions are therefore invariably weakly discernible.
Not only are fermions secured; so too, concerning the atomic constituents of
ordinary matter, are bosons. For all but one of the stable bosons are composites
of fermions (the exception is the photon). In all these cases, the bosonic wave-
function (with its symmetrization properties) is an incomplete description, and
at a level of ?ner detail - irrelevant, to be sure, to the statistical properties of
a gas of such composites - we have a collection of weakly discernible particles.
By reference to the internal structure of atoms, if nothing else, we are assured
that atoms will be at least weakly discernible.
The only cases in which the status of quantum particles as objects is seriously
in question are therefore elementary bosons - bosons (supposedly) with no in-
ternal fermionic structure. The examples in physics (according to the Standard
Model) of truly elementary bosons are photons and the other gauge bosons (the
W and Z particles and gluons) and the conjectured (but yet to be observed)
Higgs boson. But in these cases there is a ready alternative to hand for object
position in sentences: the mode of the corresponding quantum ?eld. We went
wrong in thinking the excitation numbers of the mode, because di¤ering by
integers, represented a count of things; the real things are the modes.10
5
The answer to Quine?s question - Are quantum particles objects? - is
therefore: Yes, except for the elementary bosons.
Similar conclusions follow in the classical case. If indistinguishable, and
permutations are symmetries, we should speak of them using only symmetrized
predicates. If impenetrable they will be at least weakly discerned by the irre?ex-
ive relation ?...non-zero distance from...?; but even if one relaxes this assumption,
and allows classical particles to occupy the same points of space, they may still
be (relatively) discerned by their relative velocities. Problems only arise if rela-
tive distances and velocities are zero, in which case, if no more re?ned description
is available, they will remain structureless and forever combined, and we would
do better to say there is only a single particle present (with proportionately
greater mass). This, a classical counterpart to elementary bosons, makes the
similarities in the status of particles in classical and quantum mechanics only
the closer.
What of the more metaphysical question, of whether quantum particles are
individuals? But here it is not clear what more is required of an object if it is
to count as an individual: perhaps that it is not permutable, or that it is always
absolutely discernible, or discernible by intrinsic (state-independent) properties
and relations alone. But in all cases, one is no closer to an explanation, in
10A suggestion ?rst made by Erwin Schr?dinger in 1924.
8
logical terms, of the di¤erence between classical and quantum statistics, for
none of these distinctions cut along lines that demarcate the two.
The facts about statistics are these:11
Distinguishable classical particles obey classical statistics.
Indistinguishable classical particles obey classical statistics.
Distinguishable quantum particles obey classical statistics.
Indistinguishable quantumparticles obeyquantumstatistics(Bose-Einstein
or Fermi-Dirac statistics12).
Distinguishable particles in physics we may take to be absolutely discernible,
and in all cases they obey classical statistics; but indistinguishable particles,
particles ensured only to be weakly discernible, may or may not obey classical
or quantum statistics. No more does the discernible/indiscernible distinction
line up with the classical/quantum divide; it only serves to distinguish between
certain classical and quantum particles, on the one hand, and the elementary
bosons (and their classical analogues) on the other. And ?nally, names do not
capture the distinction; given the restriction to totally symmetrized predicates,
the presence or absence of names is irrelevant.
Is there some other dimension along which one might mark out a distinctive
status for indistinguishable quantum particles? Perhaps - say, in whether or not
quantum particles are re-identi?able over time (as argued by Feynman (1965)).
But this takes us away from permutation symmetry per se, and there are many
classical objects (shadows, droplets of water, patches of colour) that likewise
may not be identi?able over time. In the weakly interacting case, taking the 1-
particle states that enter into a symmetrized or antisymmetrized state as objects
instead, one may or may not have things reidenti?able over time,13 and yet the
statistics remain the same. But the overriding objection, in the present context,
is that in considerations like these we seem to be getting away from the purely
logical notion of identity.
It seems the only remaining alternative, if indistinguishability is to have the
explanatory signi?cance normally accorded it, is to deny that permutability is
intelligible at all as a classical symmetry - that it is simply a metaphysical mis-
take, on a traditional conception of objects, to think that particles can be really
indistinguishable. One would then be left with the clean equation: permutable
if and only if quantum mechanical.
But the claim is implausible, as we are in a position to see. Finitary, cate-
gorical descriptions in LS, that are restricted to totally symmetrized predicates,
11In all cases the entropy is extensive (even for closed systems) if and only if the particles are
indistinguishable. For arguments that extensivity (and hence indistinguishability) is strictly
required for closed systems, even in classical thermodynamics, see Pniower 2006, Saunders
2006.
12There is also the possibility of parastatistics (not so far experimentally detected), involving
mixed boson and fermion transformations.
13Depending on whether or not the total state is a superposition of states of de?nite occu-
pation numbers.
9
are logically equivalent to those in L that omit only names. Descriptions of
the latter sort, whatever their philosophical inadequacies, can hardly be called
unintelligible.
In the face of this, our conclusion is rather that indistinguishability has
nothing at all to do with the quantum and classical divide, and that the reason
for quantum statistics, in the face of permutability, must be sought elsewhere.14
University of Oxford, 10, Merton St.,
Oxford OX1 4JJ
simon.saunders@philosophy.ox.ac.uk
Acknowledgements
My thanks to Guido Bacciagaluppi and Justin Pniower for helpful discussions,
and to an anonymous referee for a number of constructive criticisms.
References
Bach, A. 1997. Indistinguishable Classical Particles. Berlin: Springer-Verlag.
Cassirer, E. 1956. Determinism and Indeterminism in Modern Physics. New
Haven: Yale University Press.
Dieks, D. 1990. ?Quantum statistics, identical particles and correlations?, Syn-
these 82: 127-55.
Feynman, R. 1965. The Feynman Lectures on Physics, Vol. 3: Quantum Me-
chanics, R. Feynman, R. Leighton, and M. Sands, eds. Reading: Addison-
Wesley.
French, S., and M. Redhead 1988. ?Quantum physics and the identity of indis-
cernibles?, British Journal for the Philosophy of Science 39: 233-46.
French, S., and D. P. Rickles 2003. ?Understanding permutation symmetry?, in
K. Brading and E. Castellani, Symmetries in Physics: New Re?ections. Cam-
bridge: Cambridge University Press.
Huggett, N. 1999. ?Atomic metaphysics?, Journal of Philosophy 96: 5-24.
Ladyman, J. 2005. ?Mathematical structuralism and the identity of indis-
cernibles?, Analysis 65: 218-21.
Margenau, H. 1944. ?The exclusion principle and its philosophical importance?,
Philosophy of Science 11: 187-208.
Pniower, J. 2006. Particles, Objects, and Physics. D. Phil thesis, University of
Oxford.
Quine, W. V. 1960. Word and Object. Cambridge: Harvard University Press.
Quine, W. V, 1970. Philosophy of Logic. Cambridge: Harvard University Press.
Quine, W.V. 1976a. ?Whither physical objects??, in Essays in Memory of Imre
Lakatos, R. S. Cohen, P. Feyerabend, and M. Wartofsky, eds. Dordrecht: Reidel.
Quine, W.V. 1976b. ?Grades of discriminability?, Journal of Philosophy 73;
reprinted in 1981, Theories and Things. Cambridge: Harvard University Press.
Quine, W.V. 1990. The Pursuit of Truth. Cambridge: Harvard University
Press.
14For further discussion, see Saunders (2006).
10
Saunders, S. 2003a. ?Physics and Leibniz?s principles?, in K. Brading and E.
Castellani, Symmetries in Physics: New Re?ections. Cambridge: Cambridge
University Press.
Saunders, S. 2003b. ?Indiscernibles, general covariance, and other symmetries:
the case for non-reductive relationalism?, in Revisiting the Foundations of Rela-
tivistic Physics: Festschrift in Honour of John Stachel, A. Ashtekar, D. Howard,
J. Renn, S. Sarkar, and A. Shimony, eds. Amsterdam: Kluwer.
Saunders, S. 2006. ?On the explanation for quantum statistics?, Studies in the
History and Philosophy of Modern Physics, forthcoming.
Stern, O. 1949. ?On the term klnn! in the entropy?, Reviews of Modern Physics
21: 534-35.
Weyl, H. 1949. Philosophy of Mathematics and Natural Science. Princeton:
Princeton University Press.
11
|
|
