ARTICLE IN PRESS

Studies in History and Philosophy of

Modern Physics 35 (2004) 221–239

Intrinsically mixed states: an appreciation

Laura Ruetsche
Department of Philosophy, University of Pittsburgh, 1001 Cathedral of Learning, Pittsburgh,
PA 15260, USA
Received 10 January 2003; received in revised form 31 January 2004; accepted 6 February 2004

Abstract

An ‘‘intrinsically mixed’’ state is a mixed state of a system that is (in a sense to be

elaborated) ‘orthogonal’ to every pure state of that system. Although the presence of such
states in the quantum theories of infinite systems is well known to those who work with such
theories, intrinsically mixed states are virtually unheralded in the philosophical literature. Rob
Clifton was thoroughly familiar with intrinsically mixed states. I aim here to introduce them to
a wider audience—and to encourage that audience to cultivate their acquaintance by
suggesting that intrinsically mixed states undermine assumptions framing standard discussions
of the quantum measurement problem.
r 2004 Elsevier Ltd. All rights reserved.

Keywords: Mixed state; Algebraic quantum theory; Infinite systems

1. Introduction

Between 1988 and 2002, Rob Clifton and his many grateful collaborators
published roughly 50 articles. Through the mid-1990s, the majority of these
addressed—with ingenuity, daring, and rigor—the foundations of what I will call
‘‘ordinary quantum mechanics’’ (QM). They advanced a tradition one might
describe (briefly and with imperfect justice) as follows: assume from the outset that
the set of physical observables pertaining to a quantum system form the self-adjoint
part of the set BðHÞ of bounded operators on some separable Hilbert space H: For
a system in a quantum state, pure or mixed, specify the observables in BðHÞ that
may be simultaneously assigned determinate values subject to ‘‘natural’’ constraints,

E-mail address: [email protected] (L. Ruetsche).

1355-2198/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.shpsb.2004.02.002
 ARTICLE IN PRESS
222 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

such as locality and noncontextuality. See to it that your scheme for determinate
value assignment navigates safely between the Scylla of No Go results and the
Charybdis of the measurement problem. To this tradition, Clifton contributed
natural histories of the ‘‘natural’’ constraints comprising Scylla (e.g., Clifton, 1989),
evermore efficient No Go results (e.g., Clifton, 1992), insightful analyses and
refinements of the scheme of determinate value assignment known as the modal
interpretation (e.g., Clifton & Bell, 1995), and (with Jeff Bub) a ‘‘Go’’ theorem in the
form of a general characterization of maximally ambitious schemes of determinate
value assignments safe from Scylla (Bub & Clifton, 1996).
Clifton’s later work pioneered another tradition: the foundations of what I will
call QMN ; whose subject matter is (paradigmatically) the quantum theories of
systems possessing an infinite number of degrees of freedom (e.g., quantum field
theory (QFT), quantum statistical mechanics (QSM) in its thermodynamic limit).
For such systems the algebra of observables may not be (isomorphic to) some
BðHÞ:1 In papers such as ‘‘Generic Bell correlations between arbitrary local algebras
in quantum field theory’’ (Halvorson & Clifton, 2000) and ‘‘The modal interpreta-
tion of quantum field theory’’ (Clifton, 2000) Clifton extended results and
problematics familiar from the setting of ordinary QM to the setting of QMN : In
the course of these labors, Clifton and collaborators recognized and publicized
features of QMN that are, with respect to expectations conditioned by traditional
work on the foundations of ordinary QM, novel and surprising.
One such feature is the possibility of a mixed state on a system that is, in a sense,
orthogonal to every pure state of that system. More precisely, such a state is a state
on a C  -algebra that is disjoint from all pure states on the algebra, including the
pure states of which it is a convex combination. Clifton and Halvorson (2001) label
such states intrinsically mixed. Intrinsically mixed states are (or so I will suggest) a
conundrum for interpretations of mixtures (including ensemble interpretations and
ignorance interpretations) that take mixture weights to offer a probability
distribution over pure states. Intrinsically mixed states also frustrate versions of
the modal interpretation that use pure states in the spectral resolution of a system’s
density operator to identify the possible value states of that system. And intrinsically
mixed states challenge standard collapse approaches to the measurement problem.
Itself assigning (by way of the eigenvector–eigenvalue link) no non-trivial observable
a determinate value, a physically interesting intrinsically mixed state also fails to
assign probabilities for collapse to any pure state that does.
This essay aims to offer an introduction2 to intrinsically mixed states, and to make
a case for their interpretational consequences. It is organized as follows. Sketching a
framework for quantum theories broad enough to accommodate the variety of
quantum theories discussed thereafter, Section 2 introduces the notion of an
1
Thus a system with a single degree of freedom can fall under the rubric of QMN as I understand it, if
significant questions arise about how to represent that system’s observables. Consider, for instance,
Halvorson (2004), which explicates for such a system representations of the Canonical Commutation
.
Relations unitarily inequivalent to the Schrodinger representation.
2
For those who need no introduction, see (in addition to Clifton and collaborators), van Aken (1985)
and Emch (1972, pp. 134–135).
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 223

intrinsically mixed state. Section 3 describes the instantiation of this framework in

ordinary QM, where no states are intrinsically mixed—a regularity taken for granted
in prominent accounts of mixed state statistics and the measurement process. Section
4 explains why this regularity breaks down for a significant class of states treated by
QMN : In that setting, intrinsically mixed states are not only possible but also
physical. Section 5 outlines consequences for philosophers’ accounts of probability
and measurement in QM.

2. A general framework for quantum theories

2.1. C  -algebras and operator topologies

A quantum theory deploys an algebra of physical magnitudes (i.e. observables),

whose structure is characteristic of the quantum system the theory treats. All of the
quantum observable algebras discussed here are instances of a general type of
structure, called a C  algebra (see Brattelli and Robinson, 1987, Section 2.1). An
algebra over the field C of complex numbers is just a set A of elements ðA; B; yÞ on
which are defined operations (binary addition, [not necessarily commutative] binary
multiplication) with nice closure properties. A  -algebra is in addition closed with
respect to an involution :3 A C  -algebra is a  -algebra equipped with a norm,4 and
is complete in the topology induced by that norm.
One example of a C  -algebra familiar from ordinary QM is the set BðHÞ of
bounded operators on a separable Hilbert space H: The adjoint operation on BðHÞ
provides the involution ; and the Hilbert space operator norm5 provides the C  -
algebraic norm. In familiar formulations of ordinary QM, observables are identified
with self-adjoint elements of such an algebra, i.e. elements A such that A ¼ A:
Notice that BðHÞ has structure in excess of its C  -algebraic structure—it has for
instance an inner product and a vector norm. While every BðHÞ is a C  -algebra, the
converse is not the case: a C  -algebra need not have the full structure enjoyed by a
set of bounded operators on a separable Hilbert space.
The availability of different operator topologies furnishes some insight into
how one obtains algebras of quantum observables that are not isomorphic to
some BðHÞ: An operator topology is a criterion for when a sequence of
operators An converges to an operator A:6 Topologies matter for the constitution
of algebras of observables, because such algebras are typically closed (or complete),
in the sense of containing the limit points of any Cauchy sequence of operators
they contain. Algebras of quantum observables, for instance, are standardly
generated by starting with a set of symmetric Hilbert space operators satisfying
3
Satisfying: ðA Þ ¼ A; ðA þ BÞ ¼ A þ B ; ðlAÞ ¼ l% A and ðABÞ ¼ B A for all A; BAA and all
complex l (where the overbar denotes the complex conjugate).
4
Satisfying jjA Ajj ¼ jjAjj2 and jjABjjpjjAjj jjBjj for all A; BAA: pffiffiffiffiffiffiffiffiffiffiffiffiffi
5
That is, the norm jjAjj :¼ supjvSAH jAjvSj=jjvSj; where jjvSj :¼ /vjvS is the vector norm on H:
6
A mathematical nicety: for the weak and strong topologies, closure is defined in terms of nets of
generalized sequences. See Kadison and Ringrose (1983, pp. 113–116, 304–309) for details.
 ARTICLE IN PRESS
224 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

the Canonical [Anti] Commutation Relations (CCRs [CARs]) for the system at issue,
constructing arbitrary polynomials, then closing. Which observables are admitted by
closure to the algebra can depend on which topology—which criterion of
convergence—is in play.
Brattelli and Robinson (1987, Section 2.4.1) provides a rigorous introduction to
operator topologies. For present purposes, what matters is that the norm (a.k.a.
uniform) topology, whose criterion of convergence uses the Hilbert space operator
norm, can be distinguished from the strong and weak topologies, whose criteria of
convergence use the vector norm and the inner product, respectively. As the names
imply, strong convergence implies weak convergence, but not vice versa; norm
convergence implies strong convergence, but not vice versa. Sequences of operators
can therefore converge strongly, or weakly, without converging in the norm
topology.

2.2. Abstract algebras and their representations

A quantum system whose canonical observables obey commutation relations is

described by its Weyl algebra, constructed by taking linear combinations of
operators satisfying (an exponentiated version of) the CCRs, then closing in the
norm topology. The Weyl algebra is unique in the sense that no matter what Hilbert
space realization of the CCRs for the system one starts with, one obtains the same
(up to a structure preserving one-to-one map called a  -isomorphism7) Weyl algebra.
Thus independent of its Hilbert space antecedents, the Weyl algebra W for a system
can be regarded as an abstract C  -algebra, the structure shared by all its Hilbert
space realizations. For systems whose canonical observables obey CARs, one can
likewise obtain from those CARs Hilbert-space realization-independent abstract
C  -algebras of observables.
Abstract C  -algebras admit Hilbert space representations. A representation of a
C -algebra A is a  -morphism p : A-BðHÞ from the abstract algebra into the

concrete algebra BðHÞ of bounded linear operators on a Hilbert space H: This
representation map p may be into rather than onto BðHÞ—that is, pðAÞ may be a
proper subset of BðHÞ: The distinction between irreducible and reducible
representations will be central to what follows. A representation ðp; HÞ of A is
irreducible just in case the only closed subspaces of H that are invariant under pðAÞ
are f0g and H: Otherwise, it is reducible.
The representation p : A-BðHÞ of a C  -algebra A is closed in the norm
topology (Brattelli & Robinson, 1987, Proposition 2.3.1). Thus starting with pðAÞ
and closing in the weaker topologies H makes available could eventuate in sets of
observables richer than those contained in pðAÞ proper. The next subsection
concerns algebras obtained by following a recipe of this sort.
7
A  -morphism from a  -algebra A to a  -algebra D is a map p such that pðlA þ mBÞ ¼ lpðAÞ þ
mpðBÞ; pðABÞ ¼ pðAÞpðBÞ; and pðA Þ ¼ pðAÞ for all A; BAA and all l; mAC: A  -isomorphism is a
 -morphism that is one-to-one and onto. Because I discuss only  -algebras here, I will often abbreviate
‘ -isomorphism’ to ‘isomorphism.’
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 225

2.3. von Neumann algebras

A von Neumann algebra is a  -algebra of bounded operators that is strong

operator-closed in its action on some Hilbert space. So, to obtain a von Neumann
algebra, start with some self-adjoint algebra of bounded Hilbert space
operators—pðAÞ; say—and close in the strong operator topology. According
to von Neumann’s celebrated double commutant theorem, the strong and weak
closures of a  -algebra D of operators on a Hilbert space coincide—and coincide as
well with D’s double commutant. (Given an algebra D of bounded operators on a
Hilbert space H; its commutant D0 is the set of all bounded operators on H that
commute with every element of D: D’s double commutant D00 is then D0 ’s
commutant.) Thus pðAÞ00 is a von Neumann algebra, known as the von Neumann
algebra affiliated with the representation p: Closed in a weaker topology than the
norm topology in which the C  -algebra pðAÞ is closed, pðAÞ00 can contain bounded
Hilbert space operators that do not make it into pðAÞ: Projection operators provide
an important example. A C  -algebra of bounded Hilbert space operators need only
contain the projections 0 and I (the additive and multiplicative identities,
respectively, for the algebraic operations), whereas von Neumann algebras are rich
with projections.
That old standby, the set BðHÞ of bounded operators on a separable Hilbert space
H; is a von Neumann algebra. But, as we shall see, not every von Neumann algebra
is isomorphic to some BðHÞ: In particular, where p and j are representations of the
same C  -algebra A; the affiliated von Neumann algebras pðAÞ00 and jðAÞ00 need not
be isomorphic. Isomorphic von Neumann algebras pðAÞ00 and jðAÞ00 —and by
extension, the representations p and j with which they are affiliated—are said to be
quasi-equivalent. (That is, representations p1 and p2 of A are quasi-equivalent if and
only if there is a  -isomorphism a from p1 ðAÞ00 to p2 ðAÞ00 such that a½p1 ðAÞ ¼ p2 ðAÞ
for all AAA:) A special case of quasi-equivalence is unitarily equivalence: unitarily
equivalent representations are quasi-equivalent ones between which the  -iso-
morphism is implemented unitarily. (That is, representations p1 and p2 of A are
unitarily equivalent if and only if there is an one-to-one invertible norm-preserving
linear map U : H1 -H2 such that Up1 ðAÞU 1 ¼ p2 ðAÞ for all AAA:)

2.4. Algebraic states

In algebraic quantum theory, a state o on a C  -algebra A is a linear map o :

A-C such that oðIÞ ¼ 1 and oðA AÞX0 for all AAA: For an observable
AAA; oðAÞ is understood as its expectation value. Thus the first clause of the
definition of algebraic state imposes a normalization requirement, and the second
clause imposes a positivity requirement.
A fundamental link between algebraic states and Hilbert space representations is
forged by the Gel’fand–Naimark–Segal (GNS) theorem.

Theorem (GNS). For any state o on a C  -algebra A there is a representation ðpo ; Ho Þ

of A and a cyclic vector jOo SAHo (i.e. fpo ðAÞjOo Sg is dense in Ho ) such that
 ARTICLE IN PRESS
226 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

oðAÞ ¼ /Oo jpo ðAÞjOo S for all AAA: This GNS representation is, moreover, the
unique, up to unitary equivalence, cyclic representation ðpo ; Ho ; jOo SÞ of A such that
oðAÞ ¼ /Oo jpo ðAÞjOo S for all AAA:

That is to say, to any abstract algebraic state there corresponds a unique concrete
Hilbert space realization—indeed a concrete Hilbert space representation as a vector
state!
The set of algebraic states is convex, allowing the pure/mixed distinction
to be made in the usual manner: a state o on a C  -algebra A is mixed if and
only if it can be written as a non-trivial convex combination of states;
i.e. oð Þ ¼ lo1 ð Þ þ ð1 lÞo2 ð Þ for 0olo1 and o1 ao2 : Otherwise, o is a pure
state. The irreducibility of a state’s GNS representation provides another criterion of
purity: an algebraic state o is pure if and only if its GNS representation is
irreducible.

2.5. Quasi-equivalence and disjointness of states

As we will see shortly, when A is an observable algebra of ordinary QM, every

algebraic state on A can be expressed in terms of the GNS representation of every
other algebraic state on A: In QMN ; this need not be the case. In that context, it is
worthwhile to ask, of an algebraic state o; which other algebraic states can be
expressed in terms of o’s GNS representation po :
Of course, in an uninteresting sense, any algebraic state can be so expressed,
because any algebraic state offers a normed positive linear functional over elements
of po ðAÞ: The interesting question is: which such states admit natural and well-
behaved extensions to the von Neumann algebra po ðAÞ00 affiliated with o’s GNS
representation? It is intuitively plausible that states on A with natural and well-
behaved extensions to po ðAÞ00 will be states with (roughly speaking) nice continuity
properties in the topologies used to construct po ðAÞ00 from po ðAÞ: States exhibiting
these continuity properties with respect to a von Neumann algebra R are called
normal states of that algebra. A state o on a von Neumann algebra R acting on a
Hilbert space H is normal if and only if there is a density operator W (i.e. a positive
trace class operator with trace (1) on H such that oðAÞ ¼ TrðWAÞ for all AAR
(Brattelli & Robinson, 1987, Theorem 2.4.21). Normal states on a von Neumann
algebra R are those that define, by way of this trace prescription, countably additive
probability measures over projections of that algebra (where o : R-C is countably
additive justPin case, ifPEk ; k ¼ 1; 2; y; are mutually orthogonal projections acting on
H; then oð Ek Þ ¼ oðEk ÞÞ:
By extension, a state j on an abstract C  -algebra A is termed normal with respect
to a representation p of that algebra (p-normal, for short) just in case j is
implemented by a density operator W on Hp : Although this W need not in general
belong to the von Neumann algebra pðAÞ00 ; it will define a normal state on that
algebra, and define it uniquely: if j on A is p -normal, it admits a unique canonical
extension j * to po ðAÞ00 : Conversely, if j fails to be p-normal, it has no canonical
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 227

extension to po ðAÞ00 ; and fails to assign well-behaved (in the sense of countably
additive) probabilities to the projections po ðAÞ encompasses.8
The folium of an algebraic state o is the set of algebraic states that can be
expressed as density matrices on o’s GNS representation. This set of po -normal
states is designated FðoÞ: States o and j are quasi-equivalent if and only if their folia
are the same. The nomenclature is apt because FðoÞ ¼ FðjÞ just in case po and pj
are quasi-equivalent. The coincidence of the folia of quasi-equivalent states has a
simple explanation: von Neumann algebras affiliated with quasi-equivalent states are
 -isomorphic; thus a state is normal on one such von Neumann algebra if and only if
it is normal on the other. That is to say, the density operator states of quasi-
equivalent representations coincide. (Unitary equivalence is the special case where
the vector states coincide as well.)
At the other extreme from quasi-equivalent algebraic states lie disjoint ones. o and
j are disjoint if and only if FðoÞ-FðjÞ ¼ |: no po -normal state is pj -normal and
vice versa. In general, a pair of algebraic states can fail to be either quasi-equivalent
or disjoint since their folia can overlap without coinciding. But the sorts of QMN
states that feature in what follows are factor states. (A representation p of a C  -
algebra A is a factor representation if and only if the affiliated von Neumann algebra
Rp :¼ pðAÞ00 is factorial, i.e. the center ZðRp Þ :¼ Rp -R0p of Rp consists of multiples
of the identity operator.9 By extension a state is said to be a factor state if and only if
it is GNS representation is a factor. The set of factor states includes all pure algebraic
states.) And any two factor states are either quasi-equivalent or disjoint.
I am at last in a position to say what an intrinsically mixed state is. An intrinsically
mixed state is a state o on a C  -algebra A that is disjoint from every pure state on A:
The next section chronicles the absence of such states from ordinary QM, and the
manner in which standard interpretations of measurement and quantum probability
exploit that absence.

3. Ordinary QM

Many of the distinctions just drawn (between an abstract C  -algebra A; say, its
representation pðAÞ on a concrete Hilbert space, and the von Neumann algebra
pðAÞ00 that is the weak closure of that representation) do not get much play in the
standard philosophy of QM literature, because they do not matter there. This is
because that literature typically concerns what I have called ‘‘ordinary QM,’’ the sort
of quantum theory whose observable algebra is  -isomorphic to BðHÞ for some
separable H: In the jargon explained below in Section 4.1, the observables of
ordinary QM are the self-adjoint part of a Type I factor von Neumann algebra.
8
In general, normality is equivalent to complete additivity. But for our purposes, complete reduces to
countable additivity, because the Type III factor von Neumann algebras we discuss admit cyclic and
separating vectors, and so are countably decomposable.
9
Factors are so called because, if Rp -R0p contains only multiples of the identity, the complete algebra
BðHp Þ of bounded operators on Hp is equivalent to ðRp ,R0p Þ00 : Thus we can think of BðHp Þ as
generated by factor subalgebras Rp and R0p :
 ARTICLE IN PRESS
228 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

Philosophers’ presentation of the measurement problem, of the Bell Inequalities and

other No Go results, of their favored interpretations of QM, and of the axioms of the
theory itself, are almost invariably cast in terms of Type I factors.
They nevertheless have broad significance. The Stone-von Neumann theorem
shows that, if p is a strongly continuous representation of a Weyl algebra W for
finitely many degrees of freedom, then pðWÞ is isomorphic to the Schrodinger .
representation’s BðHÞ: Ordinary QM thus includes all quantizations of classical
theories whose phase spaces are Rn :
The notion of state familiar from ordinary QM is that of a normed, positive linear
functional o : BðHÞ-C that is countably additive. This is just the algebraic notion
of state, with the algebra in question taken to be BðHÞ and the requirement of
countable additivity imposed. (Because an abstract C  -algebra need contain no
projections other than 0 and 1, in the general setting, the countable additivity
requirement can be toothless.) States of ordinary QM are just the density operators
acting on the Hilbert space H on which BðHÞ is defined (provided that Hilbert
space is of dimension greater than 2). These normal states form a convex set, whose
extremal elements or pure states are identical to one-dimensional projection
operators. A mixed normal state corresponds to a non-extremal density operator
that admits a (possibly degenerate) spectral resolution in terms of one-dimensional
projection
P Poperators fEi g with eigenvalues different
P from 1: if W 2 aW ; then W ¼
i li Ei ¼ i TrðWEi ÞEi ; where 0pli o1 and i li ¼ 1: These eigenvalues li are just
the weights with which the pure states, generated by the projections Ei ; appear in the
resolution
P of the mixed state, generated by W ; into extremal elements: oW ð Þ ¼
i l i o E i
ð Þ:
Although all this is utterly familiar, I would like to belabor a noteworthy aspect of
it. In ordinary QM, the weights by which pure states Ei are convexly combined to
form a mixed state W are at the same time probabilities W assigns, via the trace
prescription, those pure states. This can happen because in ordinary QM a state o is
well-defined on an observable algebra BðHÞ that contains elements (one-
dimensional projections) that, on the one hand, correspond to pure states, and, on
the other, are assigned quantum probabilities—indeed, Born Rule probabilities—
by o:
This circumstance has framed a number of interpretations of mixed states. Many
of these interpretations foster the following hope: strange and inexplicable as the
probabilities assigned by quantum states are, when those probabilities take the form
of mixture weights, understood as the probabilities a mixed state assigns pure ones,
they can (at least some of them some of the time) be understood as relative
frequencies, or as relatively unmysterious epistemic probabilities. The hope is that
quantum probabilities in the form of mixture weights, so understood, afford an
account of empirical quantum statistics. If this hope is realized then there is no need
to plumb the mysteries of quantum probability in general in order to make sense of
QM’s empirical success. I sketch some of these interpretations here, in order to
expose a presupposition they share.
On the ensemble interpretation,
P a mixed state o; corresponding to a non-extremal
density operator W ¼ li Ei ; can be used to describe an ensemble of systems, the
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 229

mixture weights li giving the proportion of the ensemble having state Ei : Admitting,
as they do on this model, an interpretation in terms of relative frequencies, the
probabilities are unmysterious—even though they are Born Rule probabilities
assigned by a quantum state.
On non-ensemble interpretations, a mixed state
P o; corresponding in ordinary QM
to a non-extremal density operator W ¼ li Ei ; can be used to describe an
individual system. Two prominent traditions10 in the philosophy of QM interpret a
mixed state describing an individual composite system involved in a measurement.
For simplicity, suppose that the measurement perfectly correlates eigenvectors
fjoi Sg of the object observable OABðH1 Þ with eigenvectors fjpi Sg of the pointer
observable PABðH2 Þ: If the object system starts out in a non-trivial superposition of
O eigenstates and the measurement
P unfolds unitarily, the composite system winds up
in the entangled state jCS ¼ Pci joi S#jpi S; which implies for the object system the
reduced density matrix W1 ¼ jci j2 Ejoi S (where Ejoi S is the projection operator for
the subspace
P spanned by joi S), and for the pointer system the reduced density matrix
W2 ¼ jci j2 Ejpi S (where Ejpi S is the projection operator for the subspace spanned by
jpi S).
On the collapse version of quantum mechanics, the post-measurement P composite
state is not the entangled pure state jCS but the mixed state W 0 ¼ jci j2 Ejoi S#jpi S ;
which also implies for the object [apparatus] system the mixed state W1 ½W2 :
According to collapse theories, the actual post-measurement state of the composite
system is one of the factorized states joi S#jpi S; the actual post-measurement state
of the object [pointer] system is one of the eigenstates joi S½jpi S ; by the eigenvector/
eigenvalue link, this is a state in which the object [pointer] observable has a
determinate value; and the mixture weights jci j2 reflect our ignorance of what the
actual quantum state is, that is, of which eigenvector was the terminus of collapse.
These mixture weights, understood as transition probabilities, fund an account of
empirical quantum statistics. They are Born Rule probabilities admitting an
ignorance interpretation.11
In no-collapse theories of measurement, it is not the case that the actual post-
measurement state of the object [pointer] system is actually one of the eigenstates
joi S ½joi S : However, on modal interpretations of QM, supposing W2 to be non-
degenerate, every observable sharing its spectral resolution has a determinate post-
measurement value. Because eigenprojections of the pointer observable provide the
spectral resolution of W2 ; the pointer observable is determinate on the apparatus
system—even though the composite state is the uncollapsed superposition of pointer
eigenstates jCS: The mixture weights by which P’s eigenprojections are convexly
combined to form W2 give a probability distribution over P’s possible values. As

10
I am leaving the old-fashioned ignorance interpretation of mixtures—according to which a system
described by a mixed state W in fact occupies one of the pure states W assigns (via the trace prescription) a
non-zero probability—out of my catalog of prominent traditions. This interpretation has been justly
maligned (see, for instance, Hughes (1989, pp. 143–145) for a rundown). The interpretations I do catalog
purport at least to solve the problem of the non-unique decomposibility of mixtures.
11
Here we should mention Cartwright (1983, especially 179ff.), who argues that all quantum
probabilities are transition (collapse) probabilities.
 ARTICLE IN PRESS
230 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

with the collapse version, the Born Rule probabilities manifested in the laboratory
are understood in terms of mixture weights, interpreted as epistemic probabilities.
All the interpretations of mixed states just catalogued, as well as the accounts of
empirical quantum statistics they underwrite, require of a mixed state o that it assigns
well-behaved probabilities to pure states. These pure states give the actual states of
individual systems in the ensemble on the ensemble interpretation; candidates for the
actual post-measurement quantum state of an individual system on the collapse
interpretation; and candidate devices for identifying the actual post-measurement value
state of an individual system on the modal interpretation. The probabilities a mixed
state assigns these pure states constitute the empirical content of quantum mechanics.
Fortunately for these interpretations of mixed states and measurement processes,
in ordinary QM it is always the case that a mixed state offers a well-defined
probability distribution over pure states. This is a consequence of the fact that in
ordinary QM, no state is intrinsically mixed. For simplicity’s sake, confine attention
to theories of ordinary QM in the scope of the Stone–von Neumann theorem. Every
irreducible representation of the Weyl algebra W (a C  algebra) for such a theory is
.
unitarily equivalent to the Schrodinger representation P of W: Thus when j is a pure
state on W; its GNS representation pj ; as irreducible, is unitarily equivalent (and so
quasi-equivalent) to P: Ordinary QM states are density matrix states on the BðHÞ of
.
the Schrodinger representation; that is, they are the P-normal states. But P-normal
states and the pj -normal states are the same. So no state of ordinary QM is disjoint
from any pure state of ordinary QM. No state is intrinsically mixed.
Because no state of ordinary QM is intrinsically mixed, every state of ordinary
QM can be expressed as a density matrix on an observable algebra BðHÞ:
Expressible as a density matrix on an algebra of observables containing elements—
the one-dimensional projections—corresponding to pure states on that algebra, a
state of ordinary QM assigns, via the trace prescription, a countably additive
probability distribution over the pure states of the system it describes. The
availability of such a probability assignment is underwritten by the home truth of
ordinary QM, that no state is intrinsically mixed. The next section explains why this
home truth breaks down in QMN ; Section 5 addresses the interpretive ramifications.

4. Intrinsically mixed states in QMN

Intrinsically mixed states arise in QMN because in QMN algebras of observables

need not be isomorphic to BðHÞ for some separable H: This section provides a very
informal introduction to some of these algebras, catalogs their myriad physical
applications, and explicates the claim that their presence in QMN implies that
intrinsically mixed states occur there as well.

4.1. Typing von Neumann algebras: finite and infinite projections

Consider two projections from a von Neumann algebra BðHÞ acting on a finite-
dimensional Hilbert space. Each projection is associated with a closed subspace of
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 231

that Hilbert space. Two projections E and F are equivalent—written EBF —just in
case their associated subspaces are isometrically embeddable one into the other.
When E projects onto a subspace of the space onto which F projects—in other
words, when E’s range is a subspace of F ’s (written EpF )—E is said to be a
subprojection of F : (Equivalent criteria are that FE ¼ EF ¼ E and that
jEjvSjpjF jvSj for all jvSAH:) The subprojection relation imposes a partial order,
in the form of the relation weaker than (written "), on projections in a von
Neumann algebra. E is weaker than F if and only if E is equivalent to a
subprojection of F : Because " is a partial order, E"F and F "E together imply
that EBF :
Now, by obvious analogy with Cantor’s definition of infinity, a projection E is
called infinite if and only if there’s some projection E0 whose range is a proper subset
of E’s (written E0 oE), such that EBE0 :
A final term of art: a non-zero projection E in a von Neumann algebra R is
Abelian if and only if the von Neumann algebra ERE (in which E serves as the
identity), acting on the Hilbert space EH; is Abelian.12 In ordinary QM, the Abelian
projections are the one-dimensional ones. (Such projections are trivially Abelian:
every operator on a one-dimensional subspace is a function of the identity operator
for that subspace.) Indeed, a projection E is Abelian if and only if E is minimal, in
the sense that E’s only subprojections are 0 and E itself. It follows that Abelian
projections are finite. In ordinary QM, Abelian/minimal projections correspond to
pure states.
Now we are ready for a gross typology of von Neumann algebras. (Each type has
subtypes, details of which need not concern us.13) The typology applies to von
Neumann algebras which are factors. The weaker than relation " imposes a total
order on projections in a factor (see Kadison & Ringrose, 1986, Proposition 6.26).
Type I: Type I factors have Abelian projections, which are therefore also minimal
and finite. The algebras BðHÞ of bounded operators on a separable Hilbert space—
these include most observable algebras familiar from discussions of non-relativistic
quantum mechanics—are Type I, and each Type I factor is isomorphic to some
BðHÞ:
Type II: Type II factors have no Abelian projections, and therefore no minimal
projections. They do have finite projections.
Type III: Type III factors have no (non-zero) finite projections (and so neither
minimal nor Abelian projections). All their projections are infinite and therefore (cf.
Kadison & Ringrose 1986, Corollary 6.3.5) equivalent.
In light of the connection in familiar settings between minimal projections and
pure states, the failure of Types II and III factors to have minimal projections
demands further investigation—particularly so because such factors crop up all over
QMN :

12
An Abelian algebra is commutative: for any A; B belonging to an Abelian algebra, AB ¼ BA:
13
They can be found, along with a rigorous presentation of the material presented in this subsection, in
Kadison and Ringrose (1983, Chapter 5; 1986, Chapter 6).
 ARTICLE IN PRESS
232 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

4.2. When purity is not normal

Let R be a Type II or III factor algebra. To see why R has no pure normal state,
suppose, for contradiction, that j is a pure normal state on R: It follows from j’s
purity that the GNS representation pj : R-BðHj Þ of R associated with j is
irreducible. Because pj ðRÞ is irreducible, pj ðRÞ00 is unitarily equivalent to BðHj Þ
(Sakai, 1971, Theorem 1.21.9). Because j is a normal state on R; pj ðRÞ is unitarily
equivalent to R (Brattelli & Robinson, 1987, Theorem 2.4.24). Because R is a von
Neumann algebra, R00 ¼ R: It follows that R is unitarily equivalent to BðHj Þ: But
this is impossible: BðHj Þ is a Type I factor; R is a Type II or III factor; factors of
different types are not even quasi-equivalent. It follows that j cannot be a pure
normal state.
Were Types II and III factors mere mathematical oddities without physical
application, their failure to have pure normal states would (arguably) be
irrelevant to the interpretation of physics. But Types II and III factors are
not mere mathematical oddities. Although Type II factors occur in physical
applications (e.g. the Bardeen–Cooper–Schrieffer (BCS) model of superconducti-
vity at temperature T ¼ N is associated with a Type II factor (Emch, 1972, 140)),
Type III factors are more prevalent in QMN : For an example from QFT, let
o0 be the Minkowski vacuum state for Klein–Gordon field and let AðOÞ be the
C  -algebra associated with an open bounded region O of Minkowski
spacetime. Then o0 jAðOÞ (the state obtained from restricting o0 ; a state on the
global algebra associated with the whole of Minkowski spacetime, to its
subalgebra AðOÞ) is a mixed state. po0 jAðOÞ ðAðOÞÞ00 is a Type III factor. (This
follows from the Reeh–Schleider theorem. See Araki (1964).) Indeed, the
von Neumann algebra pojAðOÞ ðAðOÞÞ00 is a Type III factor where o is any
global state of bounded energy. What’s more, the von Neumann algebras
po0 jAðRÞ ðAðRÞÞ00 associated with unbounded regions R with non-empty spacelike
complements—e.g., Rindler wedges—of Minkowski spacetime are Type III factors
(Stormer, 1967).
Type III factor states (i.e., states o on a C  -algebra A such that po ðAÞ00 is a
Type III factor) are also the rule in the thermodynamic limit of QSM. There,
the notion of an equilibrium state is furnished by the Kubo, Martin, and
Schwinger (KMS) condition (see Emch, 1972, Chapter 10): KMS states are the
natural QMN counterpart of ordinary QM’s Gibbs equilibrium states. And KMS
states at finite temperatures in the thermodynamic limit of QSM typically
correspond to Type III factors for a wide variety of physically interesting systems:
Bose and Fermi gases, the Einstein crystal, the BCS model (see Emch, 1972, pp. 139–
140). The exceptions are KMS states at temperatures at which phase transitions
occur (if there are any for the systems in question); then KMS states are direct sums/
integrals of Type III factors.
These examples amply illustrate that the interpretation of QMN must confront
Types II and III factor states. The next subsection argues that such states are
intrinsically mixed, and cannot be understood to assign tractable probabilities to
pure states of the systems they represent.
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 233

4.3. Intrinsically mixed states

Consider a state o on a C  -algebra A such that the GNS representation po is

factorial and po ðAÞ00 is either Type II or Type III. o is disjoint from every pure state
on A—that is, o is intrinsically mixed. To see why, suppose that j is a pure state on
A: It follows that j’s GNS representation pj is irreducible, and that pj ðAÞ00 is a Type
I factor. Both factor states, o and j are either quasi-equivalent or disjoint. If o and
j were quasi-equivalent, so too would be their affiliated von Neumann algebras,
pj ðAÞ00 and po ðAÞ00 : But this cannot be, because the former is a Type I factor and the
latter a Type II or III factor. Thus o is disjoint from every pure state on A: o is
intrinsically mixed.
No state in ordinary QM is intrinsically mixed. All states there are density matrix
states on BðHÞ; which assign, via the trace prescription, a probability distribution
over one-dimensional projections corresponding to pure states of the system they
represent. Intrinsically mixed o on A makes no analogous probability assignment to
pure states on A: Acting directly on A; o assigns expectation values to its elements.
But none of these are projections to which such an expectation value assignment
could be understood as a quantum probability o assigns the corresponding pure
state. The cyclic vector jxo S implementing o in o’s GNS representation assigns
countably additive probabilities to projections in po ðAÞ00 : But po ðAÞ00 contains no
minimal projections, no elements corresponding to pure states on po ðAÞ00 or on A:
So, albeit tractable, this probability assignment does not extend to pure states. And
while there are pure states j on A such that pj ðAÞ00 does contain projections
corresponding to pure states on A; o cannot be taken to assign tractable
probabilities to these. Disjoint from pure j; o fails to be pj -normal. o’s action
on pj ðAÞ [oðpj ðAÞÞ :¼ oðAÞ for all AAA] has no canonical extension to pj ðAÞ00 ; any
extension of o from pj ðAÞ to pj ðAÞ00 fails to be countably additive on projections in
the latter—the very elements that correspond to pure states on A: An intrinsically
mixed o cannot be understood as a statistical mixture of pure states, at least not if
the statistics are supplied by tractable quantum probabilities o assigns to elements it
acts on. Ordinary QM countenances only states whose mixture weights are also
readily understood as quantum probabilities for pure states; QMN contains as well
intrinsically mixed states, whose mixture weights resist this understanding.

4.4. What intrinsically mixed states are not like

This subsection considers, and rejects, several analogies meant to suggest that
intrinsically mixed states are not as alien as the foregoing suggests.
The point of departure for the first analogy is that, even if o on A is intrinsically
mixed, it is implemented in its GNS representation by a vector jxo S: To those
conditioned by ordinary QM to identify vector states and pure states, this might
seem to be a puzzle, whose solution might seem to lie in po ’s reducibility. The
solution proposes to understand that reducibility on the model of superselection
rules. These arise when there is a non-trivial observable O that commutes with every
other observable in a von Neumann algebra R of observables. O’s eigenspaces, a.k.a.
 ARTICLE IN PRESS
234 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

its superselection sectors, are the invariant (under R) subspaces of the Hilbert space
on which R acts: a representation afforded by R is reducible. Because there is no
interference between superselection sectors, in the presence of superselection
rules, a vector state that is a superposition (e.g., ajs1 S þ bjs2 S) of components
from different superselection sectors is indistinguishable from a mixture ðjaj2 Ejs1 S þ
jbj2 Ejs2 S Þ of those components.14 By analogy, one might say that although the GNS
representation of a mixed state o represents o as a vector jOo S; the vector state
behaves like a mixed state because it lives in a reducible representation, which is like
living in the presence of superselection rules.
But this analogy is misleading. If po ðAÞ00 is a Type II or Type III factor, po ðAÞ’s
reducibility does not reflect a superselection rule. A non-trivial superselection
observable in po ðAÞ00 must commute with every other element of po ðAÞ00 : But then
the intersection between that von Neumann algebra and its commutant would be
non-trivial, which is incompatible with po ðAÞ00 ’s status as a factor. Furthermore,
superselection rules do not transmogrify every vector state into a mixed state.
Vectors lying wholly within superselection sectors still define pure states. But if
po ðAÞ00 is a Type II or Type III factor, then no vector in Ho defines a pure state,
because every such vector defines a normal state, and no normal state on po ðAÞ00 is
pure.
The second attempt at demystifying intrinsically mixed states invokes the notion
of subrepresentation. If p is reducible, there is a non-zero subspace K of Hp
invariant under the action of p: By restricting p’s action to this invariant subspace,
one obtains another representation of A : sK p : AAA-pðAÞjK: sp >
K
is a subrepre-
>
sentation of p: K is invariant under the action of p if K is. sK p ; defined by
restricting p’s action to K> ; is also a subrepresentation. If fKa g is a pairwise
orthogonal family of closed subspaces of Hp such that (i) each Ka is invariant under
o ; and (ii) 3a Ka ¼ Hp ; then p is a direct sum of its subrepresentations sp : p ¼
Ka
p
P K>
a "sp : Hence, a reducible p is a direct sum of subrepresentations (e.g., sp
Ka
and
sKp ). A fact about factor representations is that they are quasi-equivalent to each of
their subrepresentations.
Given the associations, on the one hand, between reducible representations and
mixed states and, on the other, between irreducible representations and pure states,
one might be tempted to suppose that the decomposition of a reducible
representation into a direct sum of subrepresentations mirrors the decomposition
of a mixed state into a convex combination of pure ones. That is, one might be
tempted to suppose that if o is mixed, its reducible representation may be
decomposed into a direct sum of irreducible representations, which correspond to
the pure states of which o is a convex combination. Such a decomposition promises
to re-establish the contact between mixed states and pure ones that Types II and III
factor states apparently break.

14
To take a different expectation value in the superposition and the mixture, an observable B must be
such that for iaj; /si jBjsj Sa0: But if O is a superselection observable, that is impossible, since Bjsj S
remains in the jth eigenspace of O; and so must be orthogonal to jsi S:
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 235

QMN dashes this hope. For in QMN ; subrepresentations of a mixed state’s

reducible GNS representation can be reducible all the way down. When reducible
po ðAÞ generates a Type II or III factor algebra po ðAÞ00 ; its subrepresentations will be
reducible all the way down. This is because as a factor representation, po is quasi-
equivalent to each of its subrepresentations sK p : Because quasi-equivalence
a

preserves type, if po is a Type II or III factor representation, so must be each

subrepresentation sK p : But then sp
a Ka
is not irreducible. Iterating the reasoning,
Ka
neither are any of sp ’s subrepresentations, nor any of their subrepresentations.
Reducible all the way down, po is not hoarding (and thereby facilitating a tractable
probability assignment to) any pure states on A in the form of irreducible
subrepresentations.
A third attempt to demystify intrinsically mixed states draws an analogy between
intrinsically mixed states and states unproblematically familiar from classical
statistical mechanics. In QMN ; an intrinsically mixed state o on a C  -algebra fails
to assign well-behaved quantum mechanical probabilities to pure states on A: In
classical statistical mechanics, a system associated with phase space Rn is assigned
statistical state rðxÞ; understood as a distribution on Rn that is continuous in the
Lebesgue measure. A pure state of such a system corresponds to a point pARn :
Where wðxÞ is the characteristic
R function for p the statistical state r assigns a pure
state p probability 0: Rn rðxÞwðxÞ dx ¼ 0: Assigning probability 0 to every pure state
of the system it represents, the classical state r is (in a sense) ‘orthogonal’ to every
pure state; in this respect r resembles an intrinsically mixed state. What is more, just
as pure states on A are disjoint from an intrinsically mixed state on that algebra,
pure classical statistical states—in the sense of distributions over Rn concentrated at
one of its points—are not even in r’s state space, because they fail to be Lebesgue-
measurable. Still, few of us are inclined be puzzled by a classical statistical state such
as r; to deny that it can be understood to assign probabilities to pure states, or to
doubt that classical systems described by r really do occupy one of the pure states
available to them. Why then be baffled by intrinsically mixed states?
Against the analogy, I would suggest that a classical statistical state r’s failure to
assign a non-zero probability to any pure state is a measure 0 problem. Each pure
state (point of Rn ) is assigned probability 0 by a Lebesgue-continuous probability
measure, because each pure state is a set of Lebesgue measure 0 in Rn : Classical
statistical states (probability densities) rðxÞ are nevertheless well-behaved probability
measures over Rn : Because there is an antecedent understanding of points of Rn as
pure states, this means that classical statistical states offer well-behaved probability
measures over the pure states of the systems they describe. In algebraic QMN ; by
contrast, intrinsically mixed states are puzzling not because they assign all pure states
probability 0 but because pure states are not among the von Neumann algebras they
assign well-behaved probability measures—continuous or otherwise—to begin with.
But perhaps the analogy can be pressed.15 Just as ordinary QM’s notion of
state can be generalized, so too can classical statistical mechanics’ notion—to

15
Thanks to Jos Uffink for suggesting this line of thought, and for making it clear that much more needs
to be said to this topic than is said here.
 ARTICLE IN PRESS
236 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

a normed positive linear functional over a C  -algebra C of classical observables.

As a classical observable algebra, C will be Abelian; its states in this more
general sense will include ones dispersion-free in all observables, that is, pure
states. Let j be such a pure state. Because j is pure, its GNS representation
will be irreducible. But any irreducible representation of an Abelian C  -algebra
maps elements of that algebra to bounded operators acting on a one-dimensional
Hilbert space (see Brattelli & Robinson, 1987, Corollary 2.3.21). The GNS
representation pj of C is spanned by the cyclic vector jxj S implementing j; no
mixed state—no state other than j!—is pj -normal. This is analogous to the
disjointness that obtains in QMN between intrinsically mixed o on a C  -algebra A
and every pure state of A: Yet, once again, no one is tempted to complain that we do
not understand classical statistical states, or the probability measures they offer over
pure states, on that account.
I would still, however, resist the analogy. That every irreducible representation
of an Abelian algebra is one-dimensional means that in no case can a classical
statistical state be understood to assign probabilities to pure states through
the medium of a density matrix defined on an irreducible representation
of the observable algebra. But in ordinary QM, that is exactly how the
probabilities mixed states assign pure ones are understood: an observable
algebra BðHÞ furnishes an irreducible representation of itself; mixed states
are density matrices in BðHÞ; via the trace prescription they assign probabilities
to projections in BðHÞ; that is, to pure states. The puzzle of intrinsically
mixed states in QMN is that there is a way to understand the quantum
probabilities ordinary mixed states assign pure ones that does not extend to
intrinsically mixed states. That way is completely unavailable in the classical
statistical setting. As I understand the puzzle of intrinsically mixed states, it
cannot be resolved by appeal to the classical statistical setting because it cannot
be posed there.
Notice that on this view of the puzzle, the question is not whether an arbitrary
state o on a C -algebra A; intrinsically mixed or otherwise, can be decomposed,
continuously or otherwise, in terms of A’s pure state space S:
Z
o¼ r dmðrÞ:
S

The question is whether the dmðrÞ can be understood in terms of quantum

probabilities assigned A’s pure states by the action of the state o on some algebra
containing representatives of those pure states.

5. Interpretive consequences

I conclude by briefly sketching the ways philosophers’ interpretations of mixture

weights and/or the measurement process fail to get a grip on intrinsically mixed
states.
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 237

5.1. Ignorance/ensemble interpretations

On the old-fashioned ignorance interpretation of mixed states in ordinary

QM, a system is assigned a mixed state to reflect our ignorance of which pure
state the system in fact occupies, and so the probabilities the mixed state
assigns those pure states are epistemic. For an intrinsically mixed state o on a
C  -algebra A such that po ðAÞ00 is a Type II or Type III factor, the ignorance
interpretation appears to be a non-starter. There are no normal pure states
on po ðAÞ00 : The C  -algebra A does admit pure states, but because these
reside in folia disjoint from o’s, o fails to assign them tractable probabilities.
In this sense, it is impossible for a system in an intrinsically mixed state to
occupy, with some well-defined probability, a pure one. Associating an intrinsically
mixed state with an ensemble of systems, each of which is supposed to occupy
a pure state, does not help matters. For the intrinsically mixed state still fails to
assign tractable probabilities to any of the pure states those individual systems are
supposed to occupy—so that it is utterly unclear what pure states an ensemble of
systems described by an intrinsically mixed state are supposed to occupy, and in
what proportions.

5.2. The modal interpretation

According to a significant subset of the family of modal interpretations of

ordinary QM, if a density operator W describes the state of a system, the observables
determinate on that system are those that share W ’s spectral resolution. Assuming
that W is faithful (i.e. such that TrðWAÞ ¼ 0-A ¼ 0 for all positive A), one can
ignore complications that arise from W ’s failure to be faithful, as well as from
degeneracies in W ’s spectrum. (See the collections of Healey & Hellman (1998) or
Dieks & Vermaas (1998) for details.) Under these simplifying assumptions, each
eigenprojection Ei in W ’s spectral resolution is one dimensional, and corresponds to
a pure state. According to the modal interpretation, these pure states Ei serve as
‘‘bookkeeping devices’’ (van Fraassen, 1991) for possible value states of a system
described by W : A value state is just a catalog of determinate observable values: a
system with value state Ei has determinate values for all observables of which Ei is an
eigenprojection; those values are the associated eigenvalues; the probability that a
system described by W has value state Ei is TrðWEi Þ:16 As van Fraassen observes, on
the modal interpretation ‘‘it is as if the ignorance interpretation of mixtures were
correct’’ (1991, p. 282).
The challenge posed by QMN is this: when o is intrinsically mixed, it is
unclear what it would be for an ignorance interpretation of o to be (even as if)
correct. The standard modal account of determinate measurement outcomes
16
It should be noted that in van Fraassen’s (1991) version of the modal interpretation, outside of
measurement contexts, each pure state in the range of W (i.e. each vector jcS such that W jcSa0) is a
possible value state for a system described by W ; but W does not offer a probability distribution over
these possibilities.
 ARTICLE IN PRESS
238 L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239

and the capacity of their statistics to conform to Born Rule probability

assignments is bankrupt.17

5.3. Collapse interpretations

Intrinsically mixed states are no kinder to collapse interpretations. By ‘collapse

interpretation,’ I mean an interpretation which cleaves both to the eigenvector/
eigenvalue link (according to which non-degenerate observables are determinate only
on systems in pure states) and to the postulate of measurement collapse (according
to which a system’s state collapses during measurement to an eigenstate of the
measured observable, and the Born Rule gives a probability distribution over these
collapses). The former principle implies of an intrinsically mixed state that no non-
degenerate observables are determinate on the system it represents. The latter
principle implies of an intrinsically mixed state that it cannot undergo a
measurement transition to a state on which some observables are determinate,
because an intrinsically mixed state fails to assign a Born Rule probability to any
such collapse. Thus on the collapse interpretation, not only do systems in
intrinsically mixed states lack determinate observables, not even the miracle of
measurement collapse can rescue them from this predicament.

5.4. Conclusion

This article suggests that the positions developed to respond to the measurement
problem in ordinary QM, and to make sense of quantum probabilities, do not
usefully apply to systems in intrinsically mixed states. Insofar as intrinsically mixed
states are of central physical importance in QMN ; this suggests that there is a
significant class of quantum systems for which the usual puzzles of ordinary QM are
either even more profoundly difficult than we thought, or somehow ill-formulated.

Acknowledgements

This essay emerged from work undertaken jointly with John Earman. It should
not be inferred from this that he endorses any of its contents. I am also obliged to
Gordon Belot, Dennis Dieks, Hans Halvorson, Jos Uffink, and two anonymous
referees for valuable comments and correspondence.

References

Araki, H. (1964). Type III von Neumann algebra associated with free field. Progress of Theoretical
Physics, 32, 956–965.

17
In light of these difficulties, Dieks (2000) and Clifton (2000) have (in different ways) reformulated the
interpretation. Clifton’s strategy is to do so in a manner that the interpretation might apply to von
Neumann algebras, no matter what their type. For a sketch of Clifton’s retrenchment and a critical
assessment of its prospects, see Earman and Ruetsche (2004).
 ARTICLE IN PRESS
L. Ruetsche / Studies in History and Philosophy of Modern Physics 35 (2004) 221–239 239

Brattelli, O., & Robinson, D. W. (1987). Operator algebras and quantum statistical mechanics I (2nd ed.).
Berlin: Springer.
Bub, J., & Clifton, R. (1996). A uniqueness theorem for ‘no collapse’ interpretations of quantum
mechanics. Studies in the History and Philosophy of Modern Physics, 27, 181–219.
Cartwright, N. (1983). How the laws of physics lie. Oxford: Clarendon Press.
Clifton, R. (1989). Determinism, realism, and Stapp’s 1985 proof of non-locality. Foundations of Physics
Letters, 2, 347–359.
Clifton, R. (1992). Hardy’s nonlocality theorem for N spin-12 particles. Physics Letters A, 168, 100–102.
Clifton, R. (2000). The modal interpretation of algebraic quantum field theory. Physics Letters A, 271,
167–177.
Clifton, R., & Bell, J. (1995). QuasiBoolean algebras and definite properties in quantum mechanics.
International Journal of Theoretical Physics, 34, 2409–2421.
Clifton, R., & Halvorson, H. (2001). Entanglement and open systems in algebraic quantum field theory.
Studies in the History and Philosophy of Modern Physics, 32, 1–31.
Dieks, D. (2000). Consistent histories and relativistic invariance in the modal interpretation of quantum
mechanics. Physics Letters A, 265, 317–325.
Dieks, D., & Vermaas, P. (Eds.) (1998). The modal interpretation of quantum mechanics. Dordrecht:
Kluwer.
Earman, J., & Ruetsche, L. (2004). The modal interpretation of quantum field theory and its discontents
(unpublished manuscript).
Emch, G. (1972). Algebraic methods in statistical mechanics and quantum field theory. New York: Wiley.
Halvorson, H. (2004). Complementarity of representations in quantum mechanics. Studies in the History
and Philosophy of Modern Physics, 35.
Halvorson, H., & Clifton, R. (2000). Generic Bell correlations between arbitrary local algebras in quantum
field theory. Journal of Mathematical Physics, 41, 1711–1717.
Healey, R., & Hellman, G. (Eds.) (1998). Quantum measurement: Beyond paradox. Minneapolis:
University of Minnesota Press.
Hughes, R. I. G. (1989). The structure and interpretation of quantum mechanics. Cambridge, MA: Harvard
University Press.
Kadison, R. V., & Ringrose, J. R. (1983). Fundamentals of the theory of operator algebras, Vol. 1. New
York: Academic Press.
Kadison, R. V., & Ringrose, J. R. (1986). Fundamentals of the theory of operator algebras, Vol. 2. New
York: Academic Press.
Sakai, S. (1971). C  -algebras and W  -algebras. New York: Springer.
Stormer, E. (1967). Types of von Neumann algebras associated with extremal invariant states.
Communications in Mathematical Physics, 6, 194–204.
van Aken, J. (1985). Analysis of quantum probability theory I. Journal of Philosophical Logic, 14, 267–296.
van Fraassen, B. C. (1991). Quantum mechanics: An empiricist view. Oxford: Oxford University Press.