Int J Theor Phys (2012) 51:1600–1620

DOI 10.1007/s10773-011-1037-y

Convex Quantum Logic

Federico Holik · Cesar Massri · Nicolás Ciancaglini

Received: 12 September 2011 / Accepted: 22 November 2011 / Published online: 9 December 2011
© Springer Science+Business Media, LLC 2011

Abstract In this work we study the convex set of quantum states from a quantum logical
point of view. We consider an algebraic structure based on the convex subsets of this set.
The relationship of this algebraic structure with the lattice of propositions of quantum logic
is shown. This new structure is suitable for the study of compound systems and shows new
differences between quantum and classical mechanics. These differences are linked to the
nontrivial correlations which appear when quantum systems interact. They are reflected in
the new propositional structure, and do not have a classical analogue. This approach is also
suitable for an algebraic characterization of entanglement and it provides a new entangle-
ment criteria.

Keywords Entanglement · Quantum logic · Convex sets

1 Introduction

There exists a widely discussed distinction between proper and improper mixtures [1, 2].
In this work we adopt this distinction as a starting point. From this point of view, improper
mixtures do not admit an ignorance interpretation, as is the case for proper mixtures (for
more discussion on this subject see also [3, 4] and [5]). One of the most important conse-
quences of this fact is that improper mixtures have to be considered as states on their own
right, besides pure states.
Furthermore, the study of the convex set of quantum states plays a central role in decoher-
ence [6–8] and quantum information processing [9], and this convex set is formed mainly
by mixed states. For a Hilbert space of finite dimension N , pure states form a 2(N − 1)-
dimensional subset of the (N 2 − 2)-dimensional boundary of the convex set of states. Thus,

F. Holik () · N. Ciancaglini

Instituto de Física de La Plata (IFLP), Buenos Aires, Argentina
e-mail: [email protected]

C. Massri
Departamento de Matemática, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires,
Pabellón I, Ciudad Universitaria, Buenos Aires, Argentina
Int J Theor Phys (2012) 51:1600–1620 1601

regions of the convex set of states are the important objects to investigate, rather than the
lattice of projections.
The standard quantum logical approach to quantum mechanics (QM) [10–13], takes the
lattice of projections of the Hilbert space of the system to be the lattice of propositions (see
Sect. 2 for a brief review). In this framework, as in classical mechanics (CM), the state of
the system is in direct correspondence with the conjunction of all its actual properties. This
conjunction yields a pure state or a ray in the Hilbert space, which corresponds to an atom of
the lattice of projections. But when we consider the system S formed by subsystems S1 and
S2 , the following problem appears (see for example [14]). If S is in an entangled (pure) state,
the sates of the subsystems become improper mixtures, and so, it is no longer true that the
conjunction of all actual properties yields the true state of the subsystem. On the contrary,
the conjunction yields a non atomic proposition of the von Newmann lattice of projections
(LvN ), i.e., a subspace of dimension greater than one [15], and leaves the state undetermined.
Thus, this procedure does not give the real physical state of the system any more. We analyze
this problem in detail in Sect. 3, where we review and extend the discussion posed in [16],
and impose general conditions for the structures that we want to construct.
The usual way to incorporate improper mixtures (or more generally, states) in the QL
approach, is as measures over LvN (or equivalently, measures over the propositions of the
abstract lattice). So, in the study of cases which may involve improper mixtures, we have to
jump to a different level than that of the propositions (projections). The physical propositions
belong to LvN , while mixtures belong to the set of measures over LvN . We claim that this
split is at the heart of the problem posed in [14], and complementarily, in Sect. 3 of this
article. We will review this in detail in Sects. 2 and 3.
In Sect. 3 we give a list of conditions for the structures that we are looking for, in order
to grant that they solve the posed problems. We work out a structure which incorporates
improper mixtures in the same status as pure states (alike LvN ), i.e., in which all states of
the system are atoms of the lattice. A first approach in this direction was done in [16], where
an extension of the von Newmann lattice of projections was built. In that article it was shown
that it is possible to construct a lattice theoretical framework which incorporates improper
mixtures as atoms, denoted by L here. We briefly reproduce (without proof), some of the
results of [16] in Sect. 4.
In this work we extend the construction of [16] to a larger structure LC , formed by all
convex subsets of C . This is developed in Sects. 5 and 6. As desired, this extension solves
the problem possed in Sect. 3. We think that this approach is suitable in order to consider
decoherence or entangled systems from a quantum logical and algebraic point of view. Fur-
thermore, taking the convex set of states as an starting point could be of interest if we take
into account that there exists a formulation of QM in terms of convex sets (see [17, 18] and
[19]). This is an independent formulation of QM and has the advantage that it can include
models of theories which cannot be represented by Hilbert spaces, as is the case of non
linear generalizations of quantum mechanics.
Using LC , we can construct projection functions from the lattice of the whole system
to the lattices of the subsystems which satisfy, in turn, to be compatible with the physical
description. A similar construction can be made for L (see Sect. 4). Alike the von Newmann
case, where these projection functions do not exist, the projections defined in L and LC
satisfy this condition. They are also canonical in the sense that they are constructed using
partial traces, in accordance with the quantum formalism.
The approach presented here shows (as well as the one presented in [16]), the radical
difference between quantum mechanics and classical mechanics when two systems interact,
a difference which is not properly expressed in the orthodox QL approach. When deal-
ing with classical systems, no enlargement of the lattice of propositions is needed even in
1602 Int J Theor Phys (2012) 51:1600–1620

the presence of interactions. The phase space is sufficient in order to describe all relevant
physics about the subsystems. But the existence of non-trivial correlations in quantum me-
chanics forces an enlargement of the state space of pure states to the convex set C , and so,
the enlargement of LvN to a structure as L or LC (or any other structure which satisfies
conditions listed in Sect. 3).
In Sect. 6.3 we study the maps between LC , LC1 and LC2 (the lattices of S and its subsys-
tems), and show that our construction allows for an algebraic characterization of entangle-
ment, showing a new feature which is not so explicit in the standard QL approach. Finally,
in Sect. 7 we expose our conclusions.

2 The Convex Set of States and Improper Mixtures

Let us review first the quantum logical approach to the description of physical systems
(see for example [11]). In the standard QL approach, properties (or propositions) of a
quantum system are in correspondence with closed subspaces of a Hilbert space H. The
set of subspaces P (H) with the partial order defined by set inclusion ⊆, intersection of
subspaces ∩ as the lattice meet, closed linear spam of subspaces ⊕ as the lattice join
and orthocomplementation ⊥ as lattice complement, gives rise to an orthomodular lattice
LvN (H) = P (H), ∩, ⊕, ¬, 0, 1 where 0 is the empty set ∅ and 1 is the total space H. This
is the Hilbert lattice, named QL by Birkhoff and von Neumann. We will refer to this lattice
as LvN , the ‘von Neumann lattice’ (or simply L(H)).
Mixed states represented by density operators had a secondary role in the classical trea-
tise by von Newmann because they did not add new conceptual features to pure states. In
fact, in his book, mixtures meant “statistical mixtures” of pure states [20, p. 328], which are
known in the literature as “proper mixtures” [1, Chap. 6]. They usually represent the states
of realistic physical systems whose preparation is not well described by pure states. In the
standard formulation of QL, mixtures (as well as pure states) are included as measures over
the lattice of projections [21], that is, a state s is a function:
s : L(H) −→ [0; 1] (2.1)
such that:
1. s(0) = 0 (0 is the null subspace).  
2. For any pairwise orthogonal family of projections Pj , s( j Pj ) = j s(Pj ).
But while pure states can be put in a bijective correspondence to the atoms of L(H),
this is not the case for mixtures. We review in Sect. 3 how this difference leads to problems
when compound systems are considered. We must pay attention to improper mixtures [1, 2]
because we have to deal with them in each (non trivial) case in which a part of the system is
considered.
For a classical system with phase space , the lattice of propositions is defined as the
set of subsets of  (P ()), endowed with set intersection as conjunction “∧”, set union as
disjunction “∨” and set complement as negation “¬”. We will call this lattice LCM . The
points (p, q) ∈  are in a bijective correspondence with the states of the system. Statistical
mixtures are represented as measurable functions:
σ :  −→ [0; 1] (2.2)

such that  σ (p, q)d 3 pd 3 q = 1.
Int J Theor Phys (2012) 51:1600–1620 1603

For quantum compound systems S1 and S2 , given the Hilbert state spaces H1 and H2 as
representatives of two systems, the pure states of the compound system are given by rays in
the tensor product space H = H1 ⊗ H2 . It is not true that any pure state of the compound
system factorizes after the interaction in pure states of the subsystems. This situation is
very different from that of classical mechanics, where for state spaces 1 and 2 , we assign
 = 1 × 2 for the compound system.
Let us now briefly review the relationship between the states of the joint system and the
states of the subsystems in the quantum mechanical case. Let us focus for simplicity on the
case of two systems, S1 and S2 . If {|xi(1) } and {|xi(2) } are the corresponding orthonormal
basis of H1 and H1 respectively, then the set {|xi(1)  ⊗ |xj(2) } forms an orthonormal basis for
H1 ⊗ H2 . A general (pure) state of the composite system can be written as:
ρ = |ψψ| (2.3)

where |ψ = (1) (2)
i,j αij |xi  ⊗ |xj . And if M represents an observable, its mean value M
is given by:
tr(ρM) = M (2.4)
If observables of the form O1 ⊗ 12 and 11 ⊗ O2 (with 11 and 12 the identity operators
over H1 and H2 respectively) are considered, then partial state operators ρ1 and ρ2 can be
defined for systems S1 and S2 . The relation between ρ, ρ1 and ρ2 is given by:
ρ1 = tr2 (ρ), ρ2 = tr1 (ρ) (2.5)
where tri stands for the partial trace over the i degrees of freedom. It can be shown that:
tr1 (ρ1 O1 ⊗ 12 ) = O1  (2.6)
and that a similar equation holds for S2 . Operators of the form O1 ⊗ 12 and 11 ⊗ O2 represent
magnitudes related to S1 and S2 respectively. When S is in a product state |ϕ1  ⊗ |ϕ2 , the
mean value of the product operator O1 ⊗ O2 will yield:
tr(|ϕ1  ⊗ |ϕ2 ϕ1 | ⊗ ϕ2 |O1 ⊗ O2 ) = O1 O2  (2.7)
reproducing statistical independence.
Mixtures are represented by positive, Hermitian and trace one operators, (also called
‘density matrices’). The set of all density matrixes forms a convex set (of states), which we
will denote by C . Remember that the physical observables are represented by elements of A,
the R-vector space of Hermitian operators acting on H:

Definition 2.1 A := {A ∈ B(H) | A = A† }.

Definition 2.2 C := {ρ ∈ A | tr(ρ) = 1, ρ ≥ 0}, where B(H) stands for the algebra of
bounded operators in H. The set of pure states satisfies:
P := {ρ ∈ C | ρ 2 = ρ} (2.8)

This set is in correspondence with the rays of H by the association:

F : CP(H) −→ C | [|ψ] −→ |ψψ| (2.9)
where CP(H) is the projective space of H, and [|ψ] is the class defined by the vector |ψ
(|ϕ ∼ |ψ ←→ |ϕ = λ|ψ, λ = 0). C is a convex set inside the hyperplane {ρ ∈ A | tr(ρ) =
1} formed by the intersection of this hyperplane with the cone of positive matrixes.
1604 Int J Theor Phys (2012) 51:1600–1620

Separable states are defined [22] as those states of C which can be written as a convex
combination of product states:

ρSep = λij ρi(1) ⊗ ρj(2) (2.10)
i,j

where ρi(1)∈ C1 and ρj(2) ∈ C2 , i,j λij = 1 and λij ≥ 0. So, the set S (H) of separable states
is defined as:

Definition 2.3 S (H) := {ρ ∈ C | ρ is separable}.

As said above, it is a remarkable fact that there are many states in C which are non
separable. If the state is non-separable, it is said to be entangled [23]. The estimation of the
volume of S (H) is of great interest (see—among others—[9, 24] and [25]).
In classical mechanics mixtures are not of a fundamental nature. They represent an state
of ignorance of the observer, because we know in principle that the system is in a given state
s of phase space. On he other hand, in quantum mechanics, we must take into account the
difference between proper and improper mixtures. A proper mixture can be considered as a
density matrix plus a piece of classical information, which encodes classical probabilities for
preparations of ensembles of pure states. This extra piece of classical information may have
its source on imperfections of the preparation procedure, or could be produced deliberately,
but the important fact is that these probabilities could be determined—at least—in principle,
as is the case of mixtures in classical mechanics. But for the case of improper mixtures, this
information does not exist in the world.
What is one of the main implications of the fact that improper mixtures do not admit an
ignorance interpretation? For the standard formulation of QM we have at hand what it is
usually called “the superposition principle”:

Principle 2.4 Superposition Principle. If |ψ1  and |ψ1  are physical states, then α|ψ1  +
β|ψ1  (|α|2 + |β|2 = 1) will be a physical state too.

Are there other operations which allows us to form new states up from two given states?
If we accept that improper mixtures are states of a fundamental nature as much as pure states
do, then, the fact that we can create new physical states mixing two given states, could be
thought as a principle which stands besides the superposition principle:

Principle 2.5 Mixing Principle. If ρ and ρ  are physical states, then αρ + βρ  (α + β = 1,

α, β ≥ 0) will be a physical state too.

Mixing principle is not contained directly in the superposition principle. Mixing principle
appears as a consequence of the axiom which states that to a compound system corresponds
the tensor product of Hilbert spaces. It expresses the fact that improper mixtures are physical
states. We will not consider proper mixtures in this work, we only concentrate in physical
states.
There is a remarkable physical consequence of all this (which we think is not properly
emphasized in the literature). While for pure states there always exist “true propositions”
[13], i.e., propositions for which a test will yield the answer “yes” with certainty (and a
similar situation for “false propositions”), the situation is radically different for improper
mixtures. If we accept that improper mixtures are states of a fundamental nature as well as
pure states, then we must face the fact that there exist states for which no “true propositions”
Int J Theor Phys (2012) 51:1600–1620 1605

LCM = LCM1 × LCM2

 
π1   π2
 
 


 


LCM1 LCM2
Fig. 1 In the classical case, we can map a state of the system to the states of the subsystems using the
set-theoretical projections π1 and π2

Lv N
 
?   ?
 
 


 


L v N1 L v N2
Fig. 2 We cannot apply partial traces in order to go down from Lv N to Lv N1 , and Lv N2 . How do we
complete the “?” symbols?

exist (discarding the trivial proposition represented by the Hilbert space itself). This is the
case for example for the maximum uncertainty state (finite dimension), ρ = N1 1.

3 The Limits of LvN

In the standard QL approach there is a bijective correspondence between the atoms of the
lattice LvN and pure states. Each pure state s, which is represented by a ray [|ψ], is an
atom of the lattice of projections LvN . The relationship between the atoms of the lattice and
actual properties p of the system (represented by closed subspaces of H) is given by:
{s} = ∧{p ∈ LvN | p is actual} (3.1)
and a similar relation holds in the classical case. But there appears a subtle problem with
(3.1) when compound systems are considered. Suppose that S1 and S2 are subsystems of
a larger system S in a pure entangled state |ψ. What happens if we want to determine
the states of its subsystems using (3.1)? This problem is studied in [14]. If we make the
conjunction of all actual properties for, say S1 , we will no longer obtain an atom of LvN1 .
Instead of it, we will obtain a property which corresponds to a subspace of dimension greater
than one (see Theorem 18 of [15]). But this property does no longer corresponds to the actual
state of the subsystem. The state of the subsystem is not a pure state, but an improper mixture
given by the partial trace tr2 (|ψψ|). So, it is not possible to obtain the actual physical state
of S1 using the properties of LvN1 and (3.1).
Consider Figs. 1 and 2. Alike the classical case, in general, we will not be able to map
states from LvN into states of LvN i (i = 1, 2) using partial traces (which are the physical
maps that we should use). This is because the states of the subsystems are represented by im-
proper mixtures which are not projections, and thus, they do not belong to LvN i (i = 1, 2).
How can we complete the “?” symbols of Fig. 2? There is no way to do that at the lat-
tices level (when we use von Newmann lattices). We should have to jump into the level of
measures over LvN1 . But in this work we want to avoid this possibility.
1606 Int J Theor Phys (2012) 51:1600–1620

As said above, statistical mixtures of CM do not have a fundamental nature; on the con-
trary, they represent a state of knowledge of the observer, a loss of information. Alike clas-
sical mixtures, improper mixtures are of a fundamental nature; according with the orthodox
interpretation of QM, there is no other information available about the state of the subsys-
tem. As well as pure states (and points of classical phase space), they represent pieces of
information that are maximal and logically complete. They cannot be consistently extended
and they decide any property or the result of any experiment on the given subsystem. They
determine the physics of the subsystem. Due to this, we want to consider improper mixtures
as states in the same level to that of pure states, i.e., we want that they belong to the same
propositional structure.
The strategy that we follow in this work is to search for structures which contain improper
mixtures in such a way that they have an equal treatment as the one given to pure states. As
we will see, this is possible, and such structures can be defined in a natural way, extending
(in a sense explained below) LvN and in a way which is compatible with the physics of
compound quantum systems.
We want to avoid the fact that actual properties of the propositional system do not deter-
mine the state of the system, understood as the state of affairs which determines its physics.
We think that every reasonable notion of physical state in a propositional system should
satisfy (3.1).
As remarked above, there are improper mixtures for which all yes-no tests are uncertain.
But it is important to remark that this does not imply that the system has no testable prop-
erties at all. Making quantum state tomographies we can determine the state of the system.
These kind of “tests” however, are of a very different nature than that of the yes-no exper-
iments. But the only thing that we care about is that of the reality of physical process and
our capability of experimentally test this reality. We search for structures which reflect this
physics in a direct way.
There are other reasons for considering structures which contains improper mixtures in
a same status as that of pure states. There are a lot of studies of interest which concentrate
on mixtures. For example, this is the case in quantum decoherence, quantum information
processing, or the independent generalizations of quantum mechanics which emphasize the
convex nature of mechanics (not necessarily equivalent to “Hilbertian” QM). The set of in-
terest in these studies is C instead of the lattice of projections. So it seems to be adequate to
study structures which include improper mixtures as well as pure states in a same level of
“discourse”. Such structures could provide a natural framework in which we study founda-
tional issues related to these topics.
Let us see examples of physical situations which could be captured by propositional
structures based on C . Suppose that we have a system S1 in a given state ρ1 (which can be
an improper mixture). If we consider its environment S2 , then we may state the proposition
“the state of affairs is such that S1 is in state ρ1 ”. We note that when we look the things
from the point of view of the total system S = System + Environment, a convex subset of C
(the convex set of states of S) corresponds to this proposition. This is so because S can be
in any state ρ such that tr2 (ρ) = ρ1 , and this corresponds to the convex set tr−1 2 ({ρ1 }) (see
Sect. 6.4). Similarly, we obtain the convex set tr−12 ({ρ 1 }) ∩ tr−1
1 ({ρ 2 }) for the proposition S1 is
in state ρ1 and S2 is in state ρ2 . This propositions represent the ignorance that we have about
the actual state of the whole system. A propositional structure which includes propositions
of this kind could be useful, and more natural for the study of quantum information.
It is important to notice that propositions such as the one represented by tr−1 2 ({ρ1 }) above
cannot be tested by yes–no experiments in general. Notwithstanding, they represent actual
states of affairs, and they can certainly be tested making measures on correlations, quantum
tomographies, etc.
Int J Theor Phys (2012) 51:1600–1620 1607

As another example, consider the von Newmann entropy S(ρ) = − tr(ρ ln(ρ)). It has the
following property of concavity [23]

Proposition 3.1 If ρ = αρ1 + (1 − α)ρ2 , 0 ≤ α ≤ 1, we have

S(ρ) ≥ αS(ρ1 ) + (1 − α)S(ρ2 ) (3.2)

Now consider the proposition “the entropy of the system is greater than S0 ”. To such
a proposition—which has a very definite physical meaning—there corresponds a convex
subset of C . This is so, because if we consider the set
S≥S0 = {ρ ∈ C | S(ρ) ≥ S0 } (3.3)
and if ρ1 , ρ2 ∈ S≥S0 , then any convex combination ρ = αρ1 + (1 − α)ρ2 —due to the concav-
ity property—will also belong to S≥S0 . This example shows that there are propositions with
a very clear physical meaning which correspond to subsets of C instead of subspaces of the
Hilbert space.
We summarize below the desired properties for the lattices that we are searching for, in
order to solve the problems posed in this section:
1. All physical states are included as atoms of the new lattice. Atoms and physical states
are in one to one correspondence.
2. A state of the system will be the conjunction of all the actual properties (i.e. elements
of the structure). This means that actual properties determine univocally the state of the
system.
3. There exist projection functions which map all states (atoms) of the structure correspond-
ing to the whole system S, to the corresponding states (atoms) of its subsystems S1 and
S2 . This assignation rule must be compatible with the physics of the problem.
4. LvN is set theoretically included in the new structure, in order to preserve physical prop-
erties in the standard sense.
5. Given two propositions of the structure there must exist an operation which yields a
proposition which expresses the fact that we can form mixtures of states.
There is a trivial example which satisfies the conditions 1–4 listed above, namely, the
set of all subsets of C , which we denote P (C ). Using set intersection as conjunction and set
union as disjunction, it is a boolean lattice. If we fix an entanglement measure, consider the
proposition “the system has such amount of entanglement” or given an entropy measure,
we can say “the system has such amount of entropy” and so on. To such propositions we
can assign elements of P (C ), the set of all states which satisfy those propositions. But the
boolean “or” defined by the set union hides the fact that in quantum mechanics we can make
superpositions of states (Principle 2.4) and that we can mix states (Principle 2.5). In this
work we search for structures which satisfy condition 5. For that reason, the lattice formed
by P (C ) (from now on LB ) is not of our interest. It expresses the almost trivial fact that we
can make propositions such as “the states of C which make a given function to have such a
value” but it hides the radical differences between QM and CM.
We can define—at least—two structures which satisfy the above list. One of them which
we call L (see Sect. 4 and [16]) is in close connection with the lattice of subspaces of the
space of hermitian matrixes. In this work it plays the role of a technical step to reach LC
(Sect. 5), the lattice formed by the convex subsets of C . We show below that the study
of these structures sheds light on the study of compound quantum systems, and provide a
suitable (natural) language for them, mainly because of they sort the problems possed above.
1608 Int J Theor Phys (2012) 51:1600–1620

They show things that LvN hides, or in other words, which are not expressed clearly. For
example, given two pure states ρ1 = |ψ1 ψ1 | and ρ2 = |ψ2 ψ2 | we can apply the “or”
operation of LvN , ∨LvN , which yields the linear (closed) spam of |ψ1  and |ψ2 . But we
can also consider the “∨LC ” operation (see Sect. 5), which yields all statistical mixtures of
the form αρ1 + (1 − α)ρ2 . This operation is different from linear combination (quantum
superpositions), and is related to the—non classical—mixing of states (improper mixing).
This “mixing” operation cannot be represented in QL at the level of LvN itself, i.e., it is not
a lattice operation, but it has to be represented at the level of statistical mixtures (measures
over LvN ).
It is important to notice that it is not the aim of this work to replace the von Newmann
lattice by these new structures, but to stress its limitations for the problem of compound
systems and to define its domain of applicability. We adopt the point of view that these
constructions—including LvN —yield different complementary views of quantum systems.
In the following sections, we present L and LC .

4 The Lattice L

In this section we review (without proof) some results and definitions of [16]. Let us define
G(A) as the lattice associated to the pair (A, tr), where A is considered as an R-vector
space and tr is the usual trace operator on B(H), which induces the scalar product A, B =
tr(A · B) (dim(H) < ∞). The restriction to A of tr, makes A into an R-euclidean vector
space.
G(A) := {S ⊂ A | S is a R-subspace} (4.1)
G(A) is a modular, orthocomplemented, atomic and complete lattice (not distributive, hence
not a Boolean algebra). Let L be the induced lattice in C :
L := {S ∩ C | S ∈ G(A)} (4.2)
There are a lot of subspaces S ∈ G(A) such that S ∩ C = S  ∩ C , so for each L ∈ L we choose
as a representative the subspace with the least dimension:
min{dimR (S) | L = S ∩ C , S ∈ G(A)} (4.3)
Let [S] = L, being S ∈ G(A) an element of the class L, then

S ∩ C ⊆ S ∩ C R ⊆ S ⇒ S ∩ C ∩ C ⊆ S ∩ C R ∩ C ⊆ S ∩ C (4.4)
⇒ S ∩ C  ∩ C = S ∩ C (4.5)
So S ∩ C  and S are in the same class L. Note that S ∩ C  ⊆ S and if S is the subspace with
the least dimension, then S ∩ C  = S. Also note that the representative with least dimension
is unique, because if we choose S  such that S  ∩ C = S ∩ C , then
S = S ∩ C  = S  ∩ C  = S  (4.6)
Finally, the representative of a class L that we choose is the unique R-subspace S ⊆ A such
that
S = S ∩ C R (4.7)
We call it the good representative. It is important to remark that in the case of infinite di-
mensional Hilbert spaces we cannot define good representatives in such a way.
Int J Theor Phys (2012) 51:1600–1620 1609

Let us now define “∨”, “∧” and “¬” operations and a partial ordering relation “−→” (or
equivalently “≤”) in L as:
(S ∩ C ) ∧ (T ∩ C ) := S ∩ C  ∩ T ∩ C  ∩ C (4.8)
(S ∩ C ) ∨ (T ∩ C ) := (S ∩ C  + T ∩ C ) ∩ C (4.9)
(S ∩ C ) −→ (T ∩ C ) ⇐⇒ (S ∩ C ) ⊆ (T ∩ C ) (4.10)
⊥
¬(S ∩ C ) := S ∩ C  ∩ C (4.11)
With these operations, we have that

Proposition 4.1 L is an atomic and complete lattice. If dim(H) < ∞, L is a modular lattice.

L is not an orthocomplemented lattice, but it is easy to show that non-contradiction holds

L ∧ ¬L = 0 (4.12)
and also contraposition
L1 ≤ L2 ⇒ ¬L2 ≤ L1 (4.13)
The following proposition links atoms and states:

Proposition 4.2 There is a one to one correspondence between the states of the system and
the atoms of L.

It is well known [23] that there is a lattice isomorphism between the complemented and
complete lattice of faces of the convex set C and LvN . Due to the following proposition

Proposition 4.3 Every face of C is an element of L.

We conclude that

Corollary 4.4 The complete lattice of faces of the convex set C is a subposet of L.

The previous corollary shows that L and LvN are connected. What is the relationship
between their operations? If F1 and F2 are faces we have:
(∧) F1 , F2 ∈ LvN , then F1 ∧ F2 in LvN is the same as in L. So the inclusion LvN ⊆ L
preserves the ∧-operation.
(∨) F1 ∨L F2 ≤ F1 ∨LvN F2 and F1 ≤ F2 ⇒ F1 ∨L F2 = F1 ∨LvN F2 = F2 .
(¬) ¬L F ≤ ¬LvN F .
Given two systems with Hilbert spaces H1 and H2 , we can construct the lattices L1
and L2 . We can also construct L, the lattice associated to the product space H1 ⊗ H2 . We
define:
 : L1 × L2 −→ L | (S1 ∩ C1 , S2 ∩ C2 ) −→ S ∩ C (4.14)
where S = (S1 ∩ C1  ⊗ S2 ∩ C2 ). In terms of good representatives, ([S1 ], [S2 ]) = [S1 ⊗
S2 ]. An equivalent way to define it (in the finite dimensional case) is by saying that  is the
induced morphism in the quotient lattices of the tensor map
G(A1 ) × G(A2 ) → G(A1 ⊗R A2 ) ∼ = G(A) (4.15)
We can prove the following:
1610 Int J Theor Phys (2012) 51:1600–1620

L
 
τ1    τ2
 
 ψ τ 
  

 π2 

L1 π1 L1 × L2  L2
Fig. 3 The different maps between L1 , L2 , L1 × L2 , and L. π1 and π2 are the canonical projections

Proposition 4.5 Fixing [U ] ∈ L2 then L1 is isomorphic (as complete lattice) to L1 × [U ] ⊆

L. The same is true for L2 and an arbitrary element of L1 .

Given L1 ∈ L1 and L2 ∈ L2 , we can define the following convex tensor product:

 
 L2 := {
Definition 4.6 L1 ⊗ λij ρi1 ⊗ ρj2 | ρi1 ∈ L1 , ρj2 ∈ L2 , λij = 1 and λij ≥ 0}.

This product is formed by all possible convex combinations of tensor products of el-
ements of L1 and elements of L2 , and it is again a convex set. Let us compute C1 ⊗ C2 .
Remember that C1 = [A1 ] ∈ L1 and C2 = [A2 ] ∈ L2 :
  

C 1 ⊗ C2 = λij ρi ⊗ ρj | ρi ∈ C1 , ρj ∈ C2 ,
1 2 1 2
λij = 1 and λij ≥ 0 (4.16)

So, if S (H) is the set of all separable states, we have by definition:

 C2
S (H) = C1 ⊗ (4.17)
If the whole system is in a state ρ, using partial traces we can define states for the sub-
systems ρ1 = tr2 (ρ) and a similar definition for ρ2 . Then, we can consider the maps:
tri : C −→ Cj | ρ −→ tri (ρ) (4.18)
from which we can construct the induced projections:
τi : L −→ Li | S ∩ C −→ tri (S ∩ C ) ∩ Ci (4.19)
In terms of good representatives τi ([S]) = [tri (S)]. Then we can define the product map
τ : L −→ L1 × L2 | L −→ (τ1 (L), τ2 (L)) (4.20)
The maps defined in this section are shown in Fig. 3.

5 The Lattice of Convex Subsets

The elements of L are formed by intersections between closed subspaces and C . Given that
closed subspaces are closed sets and so is C , they are also convex subsets of C . We could
go on further and consider all convex subsets of C . On the other hand (because of linearity),
partial trace operators preserve convexity and so they will map propositions of the system
into propositions of the subsystem, as desired.
Another motivation for a further extension comes from the following analogy. If the
propositions of classical mechanics are the subsets of  (classical phase space), why cannot
we consider the convex subsets of the convex set of states? It seems, after all, that convexity
Int J Theor Phys (2012) 51:1600–1620 1611

is an important feature of quantum mechanics [17, 18], and [19]. And as will be seen below,
this “convexification” of the lattice, allows for an algebraic characterization of entanglement.
Let us begin by considering the set of all convex subsets of C :

Definition 5.1 LC := {C ⊆ C | Cis a convex subset of C }.

In order to give LC a lattice structure, we introduce the following operations (where

conv(A) stands for convex hull of a given set A):

Definition 5.2 For all C, C1 , C2 ∈ LC

∧ C1 ∧ C2 := C1 ∩ C2 ,
∨ C1 ∨ C2 := conv(C1 , C2 ). It is again a convex set, and it is included in C (using con-
vexity),
¬ ¬C := C ⊥ ∩ C ,
−→ C1 −→ C2 := C1 ⊆ C2 .

With the operations of Definition 5.2, it is apparent that (LC ; −→) is a poset. If we set
∅ = 0 and C = 1, then, (LC ; −→; 0; ∅ = 0) will be a bounded poset.

Proposition 5.3 (LC ; −→; ∧; ∨) satisfies

(a) C1 ∧ C1 = C1
(b) C1 ∧ C2 = C2 ∧ C1
(c) C1 ∨ C2 = C2 ∨ C1
(d) C1 ∧ (C2 ∧ C3 ) = (C1 ∧ C2 ) ∧ C3
(e) C1 ∨ (C2 ∨ C3 ) = (C1 ∨ C2 ) ∨ C3
(f) C1 ∧ (C1 ∨ C2 ) = C1
(g) C1 ∨ (C1 ∧ C2 ) = C1 .

Proof C1 ∧ C1 = C1 ∩ C1 = C1 , so we have (a). (b), (c) and (d) are equally trivial. In order
to prove (e) we have that
C1 ∨ (C2 ∨ C3 ) = conv(C1 , conv(C2 , C3 )) (5.1)
Given that conv(C2 , C3 ) ⊆ conv(C1 , conv(C2 , C3 )), then,
C1 , C2 , C3 ⊆ conv(C1 , conv(C2 , C3 )) (5.2)
Using the above equation and convexity of conv(C1 , conv(C2 , C3 )), we have that
conv(C1 , C2 ) ⊆ conv(C1 , conv(C2 , C3 )) (5.3)
and so, using convexity,
conv(conv(C1 , C2 ), C3 ) ⊆ conv(C1 , conv(C2 , C3 )) (5.4)
A similar argument implies the converse inclusion, and so we conclude that

(C1 ∨ C2 ) ∨ C3 = conv(conv(C1 , C2 ), C3 )
= conv(C1 , conv(C2 , C3 )) = C1 ∨ (C2 ∨ C3 ) (5.5)
In order to prove (f), we have C1 ∧ (C1 ∨ C2 ) = C1 ∩ conv(C1 , C2 ). As C1 ∩ conv(C1 , C2 ) ⊆
C1 and C1 ⊆ conv(C1 , C2 ), we have C1 = C1 ∩ conv(C1 , C2 ), and so (f) is true. Let us
1612 Int J Theor Phys (2012) 51:1600–1620

finally check (g). C1 ∨ (C1 ∧ C2 ) = conv(C1 , C1 ∩ C2 ). This implies that C1 , C1 ∩ C2 ⊆

conv(C1 , C1 ∩ C2 ). As C1 is convex, we have conv(C1 , C1 ∩ C2 ) ⊆ C1 , and so we have (g). 

Regarding the “¬” operation, if C1 ⊆ C2 , then C2⊥ ⊆ C1⊥ . So C2⊥ ∩ C ⊆ C1⊥ ∩ C , and
hence
C1 −→ C2 ⇒ ¬C2 −→ ¬C1 (5.6)
⊥
Given that C ∩ (C ∩ C ) = ∅, we also have:
C ∧ (¬C) = 0 (5.7)
and so, contraposition and non contradiction hold. But if we take the proposition C = { N1 1},
then an easy calculation yields ¬C = 0. And then, ¬(¬C) = 1, and thus ¬(¬C) = C in
general. Double negation does not hold, thus, LC is not an ortholattice.
LC is a lattice which includes all convex subsets of the quantum space of states. It in-
cludes L, and so, all quantum states (including all improper mixtures) as propositions. It is
also in strong analogy with classical physics, where the lattice of propositions is formed by
all measurable subsets of phase space (the space of states).

5.1 The Relationship Between LvN , L and LC

Proposition 5.4 LvN ⊆ L ⊆ LC as posets.

Proof We have already seen that LvN ⊆ L as sets. Moreover it is easy to see that if F1 ≤ F2
in LvN then F1 ≤ F2 in L. This is so because both orders are set theoretical inclusions.
Similarly, if L1 , L2 ∈ L, because intersection of convex sets yields a convex set (and closed
subspaces are convex sets also), L1 , L2 ∈ LC , then we obtain set theoretical inclusion. And,
again, because of both orders are set theoretical inclusions, we obtain that they are included
as posets. 

Regarding the ∨ operation, let us compare ∨LvN , ∨L and ∨LC . If L1 , L2 ∈ L, then they
are convex sets and so, L1 , L2 ∈ LC . Then we can compute
L1 ∨LC L2 = conv(L1 , L2 ) (5.8)
On the other hand (if S1 and S2 are good representatives for L1 and L2 ), then:
L1 ∨L L2 = (S1 ∩ C  + S2 ∩ C ) ∩ C (5.9)
The direct sum of the subspaces S1 ∩ C  and S2 ∩ C  contains as a particular case all convex
combinations of elements of L1 and L2 . So we can conclude
L1 ∨LC L2 ≤ L1 ∨L L2 (5.10)
As faces of C can be considered as elements of LC because they are convex, if F1 and F2
are faces, we can also state
F1 ∨LC F2 ≤ F1 ∨L F2 ≤ F1 ∨LvN F2 (5.11)
Intersection of convex sets is the same as intersection of elements of L and so we have
L1 ∧LC L2 = L1 ∧L L2 (5.12)
and similarly
F1 ∧ LvN F2 = F1 ∧LC F2 = F1 ∧L F2 (5.13)
Int J Theor Phys (2012) 51:1600–1620 1613

What is the relationship between ¬LC and ¬L ? Suppose that L1 ∈ L, then they are convex
sets also, and so L1 ∈ LC . Then we can compute ¬LC L1 . We obtain:
¬ L C L1 = L⊥
1 ∩C (5.14)
On the other hand, if L1 = S ∩ C , with S a good representative
¬L L1 = S ∩ C ⊥ ∩ C (5.15)
As L1 ⊆ S ∩ C , then S ∩ C ⊥ ⊆ L⊥
1 , and so

¬ L L1 ≤ ¬ L C L1 (5.16)

5.2 Interactions in QM and CM Compared

The origin of the extension of LvN becomes clear if wee make a comparison between classi-
cal and quantum compound systems. For a single classical system its properties are faithfully
represented by the subsets of its phase space. When another classical system is added and
the compound system is considered, no enrichment of the state space of the former system
is needed in order to describe its properties, even in the presence of interactions. No matter
which the interactions may be, the cartesian product of phase spaces is sufficient for the
description of the compound system.
The situation is quite different in quantum mechanics. This is so because, if we add a
new quantum system to a previously isolated one, pure states are no longer faithful in order
to describe subsystems. Interactions produce non trivial correlations, which are reflected
in the presence of entangled states (and violation of Bell inequalities). Thus, we have to
consider the information about the non trivial correlations that each subsystem has with other
subsystems—for example, a system with the environment. The existence of this additional
information implies that we must add new elements to the propositional structure of the
system.

6 The Relationship Between LC and the Tensor Product of Hilbert Spaces

In this section we study the relationship between the lattice LC of a system S composed
of subsystems S1 and S2 , and the lattices of its subsystems, LC1 and LC2 respectively. As in
[16], we do this by making the physical interpretation of maps which can be defined between
them.

6.1 Separable States (Going Up)

Let us define:

Definition 6.1 Given C1 ⊆ C1 and C2 ⊆ C2

C1 ⊗ C2 := {ρ1 ⊗ ρ2 | ρ1 ∈ C1 , ρ2 ∈ C2 } (6.1)

Then, we define the map:

Definition 6.2

: LC1 × LC2 −→ LC
(C1 , C2 ) −→ conv(C1 ⊗ C2 )
1614 Int J Theor Phys (2012) 51:1600–1620

In the rest of this work will use the following proposition (see for example [26]):

Proposition 6.3 Let S be a subset of a linear space L. Then x ∈ conv(S) iff x is contained
in a finite dimensional simplex  whose vertices belong to S.

From (4.17) and Definition 4.6 it should be clear that (C1 , C2 ) = S (H). Definition 4.6
also implies that for all C1 ⊆ C1 and C2 ⊆ C2 :
 C2
(C1 , C2 ) = C1 ⊗ (6.2)

Proposition 6.4 Let ρ = ρ1 ⊗ ρ2 , with ρ1 ∈ C1 and ρ2 ∈ C2 . Then {ρ} = ({ρ1 }, {ρ2 }) with
{ρ1 } ∈ LC1 , {ρ2 } ∈ LC2 and {ρ} ∈ C .

Proof We already know that the atoms are elements of the lattices. Thus,
({ρ1 }, {ρ2 }) = conv({ρ1 ⊗ ρ2 }) = {ρ1 ⊗ ρ2 } = {ρ} (6.3)


Proposition 6.5 Let ρ ∈ S (H), the set of separable states. Then, there exist C ∈ LC , C1 ∈
LC1 and C2 ∈ LC2 such that ρ ∈ C and C = (C1 , C2 ).
 
Proof If ρ ∈ S (H), then ρ = ij λij ρi1 ⊗ ρj2 , with ij λij = 1 and λij ≥ 0. Consider the
convex sets:
C1 = conv({ρ11 , ρ21 , . . . , ρk1 }), C2 = conv({ρ12 , ρ22 , . . . , ρl2 }) (6.4)
Then we define:
C := (C1 , C2 ) = conv(C1 ⊗ C2 )
Clearly, the set {ρi1 ⊗ ρj2 } ⊆ C1 ⊗ C2 , and then ρ ∈ C. 

6.2 Projections onto LC1 and LC2 (Going Down)

Let us now study the projections onto LC1 and LC2 . From a physical point of view, it is of
interest to study the partial trace operators. If the whole system is in a state ρ, using partial
traces we can define states for the subsystems ρ1 = tr2 (ρ) and a similar definition for ρ2 .
Then, we can consider the maps:
tri : C −→ Cj | ρ −→ tri (ρ) (6.5)
from which we can construct the induced projections:
τi : LC −→ LCi | C −→ tri (C) (6.6)
Then we can define the product map
τ : LC −→ LC1 × LC2 | C −→ (τ1 (C), τ2 (C)) (6.7)
We use the same notation for τ and τi (though they are different functions) as in [16] and
Sect. 4, and this should not introduce any difficulty. We can prove the following about the
image of τi .

Proposition 6.6 The maps τi preserve the convex structure, i.e., they map convex sets into
convex sets.
Int J Theor Phys (2012) 51:1600–1620 1615

Proof Let C ⊆ C be a convex set. Let C1 be the image of C under τ2 (a similar argument
holds for τ1 ). Let us show that C1 is convex. Let ρ1 and ρ1 be elements of C1 . Consider
σ1 = αρ1 + (1 − α)ρ1 , with 0 ≤ α ≤ 1. Then, there exists ρ, ρ  ∈ C such that:
σ1 = α tr2 (ρ) + (1 − α) tr2 (ρ  ) = tr2 (αρ + (1 − α)ρ  ) (6.8)
where we have used the linearity of trace. Because of convexity of C, σ := αρ + (1 − α)ρ  ∈
C, and so, σ1 = tr2 (σ ) ∈ C1 . 

Proposition 6.7 The functions τi are surjective and preserve the ∨-operation. They are not
injective.

Proof Take the convex set C1 ∈ LC1 . Choose an arbitrary element of C2 , say ρ2 . Now con-
sider the following element of LC
C = C1 ⊗ ρ 2 (6.9)
C is convex, and so belongs to LC , because if ρ ⊗ ρ2 , σ ⊗ ρ2 ∈ C, then any convex combi-
nation αρ ⊗ ρ2 + (1 − α)σ ⊗ ρ2 = (αρ + (1 − α)σ ) ⊗ ρ2 ∈ C (where we have used convexity
of C1 ). It is clear that τ1 (C) = C1 , because if ρ1 ∈ C1 , then tr(ρ1 ⊗ ρ2 ) = ρ1 . So, τ1 is surjec-
tive. On the other hand, the arbitrariness of ρ2 implies that it is not injective. An analogous
argument follows for τ2 .
Let us see that τi preserves the ∨-operation. Let C and C  be convex subsets of C .
We must compute tr2 (C ∨ C  )) = tr2 (conv(C, C  )). We must show that this is the same as
conv(tr2 (C), tr2 (C  )). Take x ∈ conv(tr2 (C), tr2 (C  )). Then x = α tr2 (ρ) + (1 − α) tr2 (ρ  ),
with ρ ∈ C, ρ  ∈ C  and 0 ≤ α ≤ 1. Using linearity of trace, x = tr2 (αρ + (1 − α)ρ  ).
αρ + (1 − α)ρ  ∈ conv(C, C  ), and so, x ∈ tr2 (conv(C, C  )). Hence we have
conv(tr2 (C), tr2 (C  )) ⊆ tr2 (conv(C, C  )) (6.10)
In other to prove the other inclusion, take x ∈ tr2 (conv(C, C  )). Then,
x = tr2 (αρ + (1 − α)ρ  ) = α tr2 (ρ) + (1 − α) tr2 (ρ  ) (6.11)
   
with ρ ∈ C1 and ρ ∈ C . On the other hand, tr2 (ρ) ∈ tr2 (C) and tr2 (ρ ) ∈ tr2 (C ). This
proves that:
tr2 (conv(C, C  )) ⊆ conv(tr2 (C), tr2 (C  )) 

Let us now consider the ∧-operation. If x ∈ τi (C ∧ C  ) = τi (C ∩ C  ) then x = τi (ρ) with

ρ ∈ C ∩ C  . But if ρ ∈ C, then x = τi (ρ) ∈ tri (C). As ρ ∈ C  also, a similar argument shows
that x = τi (ρ) ∈ tri (C  ). Then x ∈ τi (C) ∩ τi (C  ). And so:
τi (C ∩ C  ) ⊆ τi (C) ∩ τi (C  ) (6.12)
which is the same as:
τi (C ∧ C  ) ≤ τi (C) ∧ τi (C  ) (6.13)
But these sets are not equal in general, as the following example shows. Take {ρ1 ⊗ ρ2 } ∈ L
and {ρ1 ⊗ ρ2 } ∈ L, with ρ  = ρ. It is clear that {ρ1 ⊗ ρ2 } ∧ {ρ1 ⊗ ρ2 } = 0 and so, τ1 ({ρ1 ⊗
ρ2 } ∧ {ρ1 ⊗ ρ2 }) = 0. On the other hand, τ1 ({ρ1 ⊗ ρ2 }) = {ρ1 } = τ1 ({ρ1 ⊗ ρ2 }), and so,
τ1 ({ρ1 ⊗ ρ2 }) ∧ τ1 ({ρ1 ⊗ ρ2 }) = {ρ1 }. A similar fact holds for the ¬-operation.
The last result is in strong analogy with what happens in L, where lack of injectivity
of the τi may be physically interpreted in the fact that the whole system has much more
1616 Int J Theor Phys (2012) 51:1600–1620

LC
 
τ1    τ2
 τ 
 


  
π1 π2 


LC1 LC1 × LC2 LC2
Fig. 4 The different maps between LC1 , LC2 , LC1 × LC2 , and LC

information than that of its parts. It is again useful to make a comparison with the classical
case in order to illustrate what is happening. The same as in classical mechanics, we have
atoms in L which are tensor products of atoms of L1 and L2 . But in contrast to classical
mechanics, entangled states originate atoms of L which cannot be expressed in such a way,
and thus, the fiber of the projection τi is much bigger than that of its classical counterpart.
It is again an important result that the projection function τ cannot be defined properly
within the frame of the traditional approaches of QL because there is no place for improper
mixtures in those formalisms. But in the formalism presented here they are included as el-
ements of the lattices, and so we can define the projections from the lattice of the whole
system to the lattices of the subsystems. This enables a more natural approach when com-
pound systems are considered from a quantum logical point of view.

6.3 An Algebraic Characterization for Entanglement

We shown that it is possible to extend LvN in order to deal with statistical mixtures and that
L and LC are possible extensions. It would be interesting to search for a characterization of
entanglement within this framework. Let us see first what happens with the functions ◦ τ
and τ ◦ . We have:

Proposition 6.8 τ ◦ = I d.

Proof

τ1 ( (C1 , C2 )) = τ1 (conv(C1 ⊗ C2 )) = tr2 (conv(C1 ⊗ C2 )) = C1

τ2 ( (C1 , C2 )) = τ2 (conv(C1 ⊗ C2 )) = tr1 (conv(C1 ⊗ C2 )) = C2
Then τ ( (C1 , C2 )) = (C1 , C2 ). 

Again, as in [16], if we take into account physical considerations, ◦ τ is not the identity
function. This is because when we take partial traces, we face the risk of losing information
which will not be recovered when we make products of states. So we obtain the same slogan
as before [16]: “going down and then going up is not the same as going up and then going
down”. We show these maps in Fig. 4. How is this related to entanglement? If we restrict
◦ τ to the set of product states, then it reduces to the identity function, for if ρ = ρ1 ⊗ ρ2 ,
then:
◦ τ ({ρ}) = {ρ} (6.14)
On the other hand, it should be clear that if ρ is an entangled state
◦ τ ({ρ}) = {ρ} (6.15)
Int J Theor Phys (2012) 51:1600–1620 1617

because ◦ τ ({ρ}) = {Tr2 (ρ) ⊗ Tr1 (ρ)} = {ρ} for any entangled state. This property points
in the direction of an arrow characterization of entanglement. There are mixed states which
are not product states, and so, entangled states are not the only ones who satisfy (6.15). What
is the condition satisfied for a general mixed state? The following proposition summarizes
all of this.

Proposition 6.9 If ρ is a separable state, then there exists a convex set (indeed, a politope),
Sρ ⊆ S (H) such that ρ ∈ Sρ and ◦τ (Sρ ) = Sρ . More generally, for a convex set C ⊆ S (H),
then there exists a convex set SC ⊆ S (H) such that ◦ τ (SC ) = SC . For a product state, we
can choose Sρ = {ρ}. Any proposition C ∈ LC which has at lest one non-separable state,
satisfies that there is no convex set S such that C ⊆ S and ◦ τ (S) = S.

Proof We have already seen above that if ρ is a product state, then ◦ τ ({ρ}) = {ρ}, and
so S = {ρ}. If ρ is a general separable state, then there exists ρ1k ∈ C1 , ρ2k ∈ C1 and αk ≥ 0,
N ρ N
k=1 αk = 1 such that ρ = k=1 αk ρ1k ⊗ ρ2k . Now consider the convex set (a simplex)

N 
N
M = σ ∈C|σ = λi,j ρ1i ⊗ ρ2j , λi,j ≥ 0, λi,j = 1 (6.16)
i,j =1 i,j =1

It is formed by all convex combinations of products of the elements which appear in the
decomposition of ρ. It should be clear that ρ ∈ M. Let us compute ◦ τ (M). τ (M) =
(τ1 (M); τ2 (M)). An element of τ1 (M) is of the form (for σ ∈ M):

N 

N  N
tr1 (σ ) = λi,j ρ1i = μi ρ1i (6.17)
i=1 j =1 i=1
N
with μi = j =1 λi,j . In an analogous way we show that an element of τ2 (M) has the form
N N N N
j =1 νj ρ1i with νi = i=1 λi,j . Note that j =1 μj = j =1 νj = 1. In order to compute
(τ1 (M); τ2 (M)) we must form the convex hull of the set:
 N

τ1 (M) ⊗ τ2 (M) = {σ1 ⊗ σ2 |σ1 ∈ τ1 (M), σ2 ∈ τ2 (M)} = μi νj ρ1i ⊗ ρ2j (6.18)
i,j =1

then we conclude that

 

N
◦ τ (M) = conv μi νj ρ1i ⊗ ρ2j (6.19)
i,j =1

Let us now prove that ◦ τ (M) = M. If σ ∈ ◦ τ (M), then, looking  at (6.19), it is

apparent that σ belongs to M. On the other hand, if σ ∈ M, then σ = N i,j =1 λi,j ρ1i ⊗ ρ2j
(convex combination). It is important to notice that ◦ τ (M) is a convex set, because trace
operators preserve convexity, and is a convex hull. On the other hand ◦ τ ({ρ1i ⊗ ρ2j }) =
{ρ1i ⊗ ρ2j } and by definition of τ1 (M) ⊗ τ2 (M), we have that {ρ1i ⊗ ρ2j } ∈ ◦ τ (M) for
all i, j . And so, by convexity of ◦ τ (M), σ ∈ ◦ τ (M). This concludes the proof that
◦ τ (M) = M (and it is apparent that M is a politope). Then M is the desired Sρ ⊆ S (H).
If C ⊆ S (H), then all ρ ∈ C are separable. S (H) is by definition, a convex set. Let us
see that it is invariant under ◦ τ . First of all, we know that S (H) is formed by all possible
convex combinations of products of the form ρ1 ⊗ ρ2 , with ρ1 ∈ C1 and ρ2 ∈ C2 . But for each
one of these tensor products, ◦ τ ({ρ1 ⊗ ρ2 }) = {ρ1 ⊗ ρ2 }, and it is easy to see that they
1618 Int J Theor Phys (2012) 51:1600–1620

belong to ◦ τ (S (H)). This is a convex set, thus all convex combinations of them belong
to it. So we can conclude that
◦ τ (S (H)) = S (H) (6.20)
Now, consider C ∈ LC such that there exists ρ ∈ C, being ρ nonseparable. ◦ τ (S) ⊆
S (H) for all S ∈ LC . Then, it could never happen that there exists S ∈ LC such that C ⊆ S
and ◦ τ (S) = S. 

From the last proposition, we conclude that there is a property which the convex subsets
of separable states satisfy, and convex subsets which include non-separable states do not.
This motivates the following definition.

Definition 6.10 If C ∈ LC , we will say that it is a separable proposition if there exists

SC ∈ LC such that ◦ τ (SC ) = SC and C ⊆ SC . Otherwise, we will say that it is a non-
separable or entangled proposition.

Another conclusion of Proposition 6.8 is that a density matrix ρ is separable iff there
exists a convex set Cρ such that ρ ∈ Cρ and ◦ τ (Sρ ) = Sρ . Thus, Proposition 6.9 also
provides an entanglement criteria. For a complete study of this fact from a different approach
see [27].

6.4 The Inverse τ -Map

In Sect. 6.2 we defined the function τ = (τ1 , τ2 ). Now we show that using the inverse map
τ −1 = (τ1−1 , τ2−1 ) we obtain lattice morphisms. It is easy to show that τi−1 maps any propo-
sition from Ci into a proposition of C . This is because the pre-image of a convex set under
these functions is again a convex set. If C1 is a proposition of C and if τ1 (ρ), τ1 (ρ  ) ∈ C1 , it
is clear that any convex combination of ρ and ρ  will belong to τ1−1 (C1 ), because the partial
trace is linear and C1 is convex.

Proposition 6.11 For all X ∈ LC X ⊆ τ1−1 (τ1 (X)) and for all Y ∈ LC1 , τ1 (τ1−1 (Y )). For all
C ⊆ C we have C ⊆ τ1−1 (C1 ) ∧ τ2−1 (C2 ).

Proof Let X ∈ LC . Then, if x ∈ X it follows that τ1 (x) ∈ τ1 (X) and so, X ⊆ τ1−1 (τ1 (X)). If
Y ∈ LC1 and z ∈ τ1 (τ1−1 (Y )). Then by definition of τ1−1 (Y ), it follows that z ∈ Y .
Let C ∈ LC . Now τ1 (C) = C1 ∈ LC1 and τ2 (C) = C2 ∈ LC2 . Then, it is apparent that
C ⊆ τ1−1 (C1 ) and C ⊆ τ2−1 (C2 ). And so C ⊆ τ1−1 (C1 ) ∧ τ2−1 (C2 ). 

Proposition 6.12 For all a, b ∈ LC1 τ1−1 (a ∧ b) = τ1−1 (a) ∧ τ1−1 (b), τ1−1 (a ∨ b) = τ1−1 (a) ∨
τ1−1 (b). Furthermore, τ1−1 is an injective function and if a, b ∈ LC1 and a ⊆ b, then τ1−1 (a) ⊆
τ1−1 (b). If ρ = ρ  then τ1−1 (ρ) ∧ τ1−1 (ρ  ) = 0.

Proof Consider the sets τ1−1 (a ∧ b) and τ1−1 (a) ∧ τ1−1 (b). If x ∈ τ1−1 (a ∧ b), then τ1 (x) ∈
a ∧ b ⊆ a, and we obtain also τ1 (x) ∈ a ∧ b ⊆ b. This means that x ∈ τ1−1 (a) and x ∈ τ1−1 (b).
So we have τ1−1 (a ∧ b) ⊆ τ1−1 (a) ∧ τ1−1 (b). On the other hand, if x ∈ τ1−1 (a) ∧ τ1−1 (b), then
x ∈ τ1−1 (a) and x ∈ τ1−1 (b). This means that τ1 (x) ∈ a and τ1 (x) ∈ b, and so, τ1 (x) ∈ a ∧ b.
This means that x ∈ τ1−1 (a ∧ b). This concludes the proof that τ1−1 (a ∧ b) = τ1−1 (a) ∧ τ1−1 (b).
If x ∈ τ1−1 (a) ∨ τ1−1 (b) then x = αρ + βρ  , with τ1 (ρ) ∈ a and τ1 (ρ  ) ∈ b. So τ1 (x) =
ατ1 (ρ) + βτ1 (ρ  ) ∈ a ∨ b. This means that x ∈ τ1−1 (a ∨ b), and we have τ1−1 (a ∨ b) ⊇
Int J Theor Phys (2012) 51:1600–1620 1619

τ1−1 (a) ∨ τ1−1 (b). Now, let x ∈ τ1−1 (a ∨ b). Then, τ1 (x) ∈ a ∨ b. This means that τ1 (x) =
αρ + βρ  (convex combination), with ρ ∈ a and ρ  ∈ b. There exist σ ∈ τ1−1 (a) and σ  ∈
τ1−1 (b) such that τ1 (σ ) = ρ and τ1 (σ  ) = ρ  . Then τ1 (x) = ατ1 (σ ) + βτ1 (σ  ). τ1 () is a
linear function so, the last equality implies τ1 (x − (ασ + βσ  )) = 0. Then, there exists
ς ∈ Ker(τ1 ()) such that x = ασ + βσ  + ς . If β = 0, then α = 1 (convex combination),
and then, x = σ ∈ τ1−1 (a), and in that case x ∈ τ1−1 (a) ∨ τ1−1 (b). If β = 0, we can put
x = ασ + β(σ  + β1 ς ). τ1 ((σ  + β1 ς )) = τ1 (σ  ) + 0 ∈ b, and so σ  + β1 ς ∈ τ1−1 (b). This
proves that x ∈ τ1−1 (a) ∨ τ1−1 (b), and thus τ1−1 (a ∨ b) ⊆ τ1−1 (a) ∨ τ1−1 (b).
Let a and b be two propositions such that a = b. Suppose that τ1−1 (a) = τ1−1 (b). If a = b,
there exists ρa ∈ a such that ρa ∈ / b. It is clear that τ1−1 (ρa ) ⊆ τ1−1 (a) = τ1−1 (b) and then,
there exists ρ ∈ τ1 (b) such that τ1 (ρ) = ρa . But by definition of τ1−1 (b), we would have
−1

that ρa ∈ b, a contradiction. Thus, we have τ1−1 (a) = τ1−1 (b).

If a ⊆ b, suppose that x ∈ τ1−1 (a). Then τ1 (x) ∈ b, and so x ∈ τ1−1 (b) also. If x ∈ τ1−1 (ρ),
x ∈ τ1−1 (ρ  ) and ρ = ρ  , then ρ = τ1 (x) = ρ  , a contradiction. 

7 The Difference with Other Approaches and Conclusions

The problem of compound quantum systems has been widely studied from different ap-
proaches. An important difference of our approach is that it treats improper mixtures in a
different way. In CM the fundamental description is given by subsets (propositions) of the
phase space. Statistical mixtures are not fundamental; they appear as a limitation in the ca-
pability of knowledge of the observer. It is for that reason that they are expressed as proper
mixtures, and so in the orthodox logical approach they appear in different levels: the propo-
sitions belong to the lattice LCM , while the mixtures are measures over this lattice. Pure
states are in one to one correspondence with the atoms of the lattice, and so they are also
included as elements of the lattice, but mixtures are in a different level.
The situation turns radically different in QM if we accept that improper mixtures do not
admit an ignorance interpretation. If S1 is in a state represented by an improper mixture ρ,
there is no more physical information available for the observer. ρ represents the actual state
for S1 and we cannot get more information, not because of our experimental limitations, but
because of that information does not exist. Thus, they should be represented at the same
level as that of pure states, because they are maximal pieces of information.
On the other hand, influenced by historical reasons, the orthodox QL approach, still re-
tains the analogy with CM and considers improper mixtures in a different level than that
of the propositions. Thus, the orthodox QL approach presents some difficulties when com-
pound quantum systems are involved. We studied these problems in Sect. 3 and we gave a
list of conditions on the structures that we are looking for in order to solve these difficulties.
The structures presented in this work and in [16], LC and L, do not have these problems,
but on the contrary, they incorporate these quantum mechanical features explicitly. This is
so because they satisfy the conditions listed in Sect. 3. While each state (pure or mixed)
induces a measure in the lattice of projections, this has nothing to do with the identification
of these measures with classical mixtures. Indeed, any pure state is not dispersion free also
and so induces a measure over LvN . This measure has a radical different nature from that
of classical measures; in our approach, improper mixtures are in the same status than pure
states and induce measures over LvN as well as pure states, but measures and states are not
identified. This situation is very different from that of CM, in which the measures induced
by pure states are trivial.
1620 Int J Theor Phys (2012) 51:1600–1620

Our approach—specially LC —presents itself as a natural logical and algebraic language

for the study of topics which involve compound quantum systems such as quantum informa-
tion processing and decoherence, which concentrate on the study of C instead of the lattice
of projections. In particular, we can map states of the compound system into states of its
subsystems at the lattice level, while this cannot be done in the standard QL approach. Fur-
thermore, LC and L capture the physics behind the fact that we can mix states according to
the “mixing principle” of Sect. 2.
As discussed in Sect. 5.2, our construction shows a new radical difference with classical
mechanics, namely, that of the enlargement of the propositional structure when interactions
are involved, a difference which is not clear in the standard QL approach.
Moreover, as we showed in Sect. 6.3, LC sheds new light into algebraic properties of
quantum entanglement via the study of the natural arrows defined between the lattice of the
system and its subsystems. The study of these arrows reveals itself as adequate for the of
algebraic characterization of entanglement. Indeed, it provides an entanglement criteria.

Acknowledgements This work was partially supported by the following grants: PIP No 6461/05
(CONICET). We wish to thank G. Domenech for careful reading and discussions.

References

1. D’Espagnat, D.: Conceptual Foundations of Quantum Mechanics. Benjaming, Reading (1976)

2. Mittelstaedt, P.: The Interpretation of Quantum Mechanics and the Measurement Process. Cambridge
Univ. Press, Cambridge (1998)
3. Kirkpatrik, K.A.: arXiv:quant-ph/0109146, 21 Oct. 2001
4. D’Espagnat, D.: arXiv:quant-ph/0111081, 14 Nov. 2001
5. Masillo, F., Scolarici, G., Sozzo, S.: arXiv:0901.0795 [quant-ph], 7 Jan. 2009
6. Schlosshauer, M.: Decoherence and the Quantum-to-Classical Transition. Springer, New York (2007)
7. Castagnino, M., Fortin, S.: Int. J. Theor. Phys. 50(7), 2259–2267 (2011)
8. Castagnino, M., et al.: Class. Quantum Gravity 25, 154002 (2008)
9. Horodecki, M., Horodecki, P., Horodecki, R.: In: Alber, G., et al. (eds.) Quantum Information. Springer
Tracts in Modern Physics, vol. 173, p. 151. Springer, Berlin (2001)
10. Birkhoff, G., von Neumann, J.: Ann. Math. 37, 823–843 (1936)
11. Dalla Chiara, M.L., Giuntini, R., Greechie, R.: Reasoning in Quantum Theory. Kluwer Academic, Dor-
drecht (2004)
12. Jauch, J.M.: Foundations of Quantum Mechanics. Addison-Wesley, Cambridge (1968)
13. Piron, C.: Foundations of Quantum Physics. Addison-Wesley, Cambridge (1976)
14. Aerts, D.: Int. J. Theor. Phys. 39, 483–496 (2000)
15. Aerts, D.: J. Math. Phys. 25, 1434–1441 (1984)
16. Domenech, G., Holik, F., Massri, C.: J. Math. Phys. 51, 052108 (2010)
17. Mielnik, B.: Commun. Math. Phys. 9, 55–80 (1968)
18. Mielnik, B.: Commun. Math. Phys. 15, 1–46 (1969)
19. Mielnik, B.: Commun. Math. Phys. 37, 221–256 (1974)
20. von Neumann, J.: Mathematical Foundations of Quantum Mechanics, 12th edn. Princeton University
Press, Princeton (1996)
21. Rédei, M.: Quantum Logic in Algebraic Approach. Kluwer Academic, Dordrecht (1998)
22. Werner, R.: Phys. Rev. A 40, 4277–4281 (1989)
23. Bengtsson, I., Życzkowski, K.: Geometry of Quantum States: An Introduction to Quantum Entangle-
ment. Cambridge University Press, Cambridge (2006)
24. Życzkowski, K., Horodecki, P., Sanpera, A., Lewenstein, M.: Phys. Rev. A 58, 883 (1998)
25. Aubrun, G., Szarek, S.: Phys. Rev. A 73, 022109 (2006)
26. Valentine, F.: Convex Sets. McGraw-Hill, New York (1964)
27. Holik, F., Plastino, A.: Convex polytopes and quantum separability. Phys. Rev. A (to be published)